Tidal Analysis and Predictions PDF

U.S.
DEPARTMENT OF COMMERCE
National Oceanic and Atmospheric Administration
National Ocean Service
Center for Operational Oceanographic Products and Services
Tidal Analysis and Prediction

NOAA Special Publication NOS CO-OPS 3
NOAA Special Publication NOS CO-OPS 3
Bruce B. Parker, Ph.D.
Silver Spring, Maryland

July 2007
Library of Congress Control Number: 2007925298
U.S. DEPARTMENT OF COMMERCE

Carlos M. Gutierrez , Secretary
National Oceanic and Atmospheric Administration

Conrad C. Lautenbacher, Jr.
Undersecretary of Commerce for Oceans and Atmosphere and NOAA Administrator
National Ocean Service

John H. Dunnigan, Assistant Administrator

Michael Szabados, Director
NOTICE
Mention of a commercial company or product does not constitute an endorsement

by NOAA. Use for publicity or advertising purposes of information from this
publication concerning proprietary products or the tests of such products is not
authorized.
Request for copies should be addressed to

1305 East West Highway
SSMC IV, Rm 6616
Silver Spring, Maryland 20910-3281
Phone (301) 713-2981ext. 125
http://tidesandcurrents.noaa.gov
Table of Contents
page
1. Introduction 1
1.1 Purpose of This Book 1
1.2 Relationship to Previous Tidal Analysis and Prediction Books and
Manuals 3
1.3 The Uses For Tidal Analysis and Prediction 6
1.4 Brief Historical Background 7
1.5 The Importance of Hydrodynamics 10
1.6 Organization of This Book 11
2. Theory Behind Tidal Analysis and Prediction .15

2.1 Short Introductory Overview the Big Picture 15
2.1.1 Some Definitions 15
2.1.2 The Generation of Tides 24
2.2 Astronomical Considerations 27
2.2.2 The Orbits of the Moon Around the Earth and the Earth Around the Sun 27
2.2.2 The Origins of the Largest Tidal Harmonic Constituents 31
2.2.3 Derivation of Tidal Frequencies from the Basic Astronomical Frequencies
Insights from Tidal Spectra 35
2.2.4 The Origin of Node Factors and Equilibrium Arguments 39
2.2.5 Definitions Revisited, With Relationship to Harmonic Constants 42
2.3 Hydrodynamic Considerations 48
2.3.1 Hydrodynamic Effects On Tide Ranges, Phase Lags and Tidal Current
Speeds 49
2.3.2 Nonlinear Effects of Shallow Water Overtides and Compound Tides 55
2.3.3 Nonlinear Tidal Interaction With River Flow and Storm Surge 62
2.3.4 Nonlinear Hydrodynamic Effects On Node Factors 67
2.3.5 Zero-frequency Effects From Nonlinear Tidal Interactions 69
2.3.6 Special Aspects of Tidal Currents 69
3. Methods of Tidal Analysis and Prediction .81

3.1 Introduction 81
3.2 Simple Demonstration of Extracting Tidal Constituents From a Data Time
Series 82
iii
3.3 The Synodic Period Length of Time Series Needed To Separate Two
Constituents 84
3.4 Harmonic Analysis 90
3.4.1 Introduction 90
3.4.2 The Harmonic Tide Prediction Equation 90
3.4.3 The Harmonic Tidal Current Prediction Equations 95
3.4.4 Fourier-based Harmonic Analysis 97
3.4.5 Least-squares-based Harmonic Analysis 99
3.5 Other Frequency Domain Methods 102
3.5.1 The Response Method 102
3.5.2 The Harmonic Shallow-Water Corrections Method 103
3.5.3 The Species Concordance Method 104
3.5.4 The Method for Superfine Resolution of Harmonic Constituents 105
3.5.5 Continuous Wavelet Transform Method 105
3.5.6 Harmonic Constants From a Long Series of High and Low Waters 106
3.6 Nonharmonic Comparison Methods 109
3.6.1 Overview and Short Historical Background 109
3.6.2 Types of Nonharmonic Analyses 111
3.6.3 Variations In Nonharmonically Determined Time and Height Differences 112
3.7 Analyses for determining Sa and Ssa 119
3.7.1 The Meteorological Origins of Sa and Ssa 119
3.7.2 Considerations In Whether To Use Sa and Ssa In a Prediction 120
3.7.3 Methods for Calculating Sa and Ssa 122
3.8 Tidal Filtering and Detiding 123
3.8.2 Specially Designed Tidal Filters 127
3.9 Calculating Consistently Defined Maxima and Minima 129
3.9.1 Consistently Defined High and Low Waters 129
3.9.2 Consistently Defined Maximum Floods and Ebbs and Slacks
(or Minimums) 131
3.10 Spectral Analysis and EOF Analysis 132
3.10.1 Spectral Analysis of Water Level Data Time Series 133
3.10.2 Tidal Cusps In Spectra 134
3.10.3 Spectral Analysis of Current Data Time Series Rotary Spectra 135
3.10.4 Cross-spectral Analysis 136
3.10.5 Empirical Orthogonal Function (Principal Component) Analysis 137
4. Harmonic Analysis of Water Level Data .139

4.1 Considerations In Carrying Out the Analysis 139
4.1.1 Which Tidal Constituents Should Be Solved For 139
4.1.2 Keeping or Rejecting Small Constituents 142
4.1.3 Instrument Errors and Their Effects On Analysis Results 142
4.1.4 Assessing the Potential Effects of Nontidal Influences 146
4.1.5 Use of Node Factors and Equilibrium Arguments
(or of Satellite Constituents) 148
4.2 Methods for Analyzing Short Time Series 150
iv
Table of Contents
4.2.1 Schuremans Inference and Elimination Technique 152

4.2.2 Use of a Reference Station To Improve the Results of a Short Analysis 153
4.2.3 Inference In the Frequency Domain 155
4.2.4 Other methods 156
4.3 Assessing the Quality of the Predicted Tide Series 158
4.3.2 Comparison of High and Low Waters In the Predicted Time Series
Versus In the Observed Time Series 160
4.3.3 Spectral Analysis of the Residual Time Series 161
4.3.4 Examination of the Residual Time Series For Periods With Transient
Tidal Oscillations 162
4.3.5 The Use and Misuse of Build-up Factors 162
4.3.6 Looking For Changes in Bathymetry or Shoreline 163
4.4 Producing a Reference Station for Table 1 of a Tide Table 164
4.5 Summary Overview: Steps In Harmonically Analyzing Water Level Data 165
5. Harmonic Analysis of Current Data .169

5.1 Special Problems With Current Data 169
5.2 Tidal Current Harmonic Constituent Ellipses 174
5.3 Considerations In Carrying Out the Analysis 181
5.3.1 Which Tidal Current Constituents Should Be Solved For 181
5.3.2 Selection of Orthogonal Axes 181
5.3.3 Instrument Errors and Their Effects On Analysis Results 182
5.3.4 Assessing the Potential Effects of Nontidal Influences 186
5.3.5 Whether To Include the Calculated Mean In a Tidal Current Prediction 188
5.3.6 Use of Node Factors and Equilibrium Arguments
(or of Satellite Constituents) 189
5.4 Methods for Analyzing Short Time Series 189
5.5 Assessing the Quality of the Predicted Tidal Current Series 190
5.5.2 Comparison of Maximum Floods and Ebbs and Slacks In the Predicted
Time Series Versus In the Observed Time Series 190
5.5.3. Spectral Analysis of the Residual Time Series 191
5.5.4 Examination of the Residual Time Series For Periods With Transient
Tidal Current Oscillations 192
5.5.5 Looking For Changes in Bathymetry or Shoreline 192
5.6 Producing a Reference Station for Table 1 of a Tidal Current Table 193
5.7 Summary Overview: Steps In Harmonically Analyzing Current Data 194
6. Nonharmonic Analysis of Water Level and Current Data .195

6.1 The Need For Nonharmonic Tidal Analyses 196
6.2 Hydrodynamic Considerations 196
6.3 Monthly Mean Tide Analysis 197
6.3.1 Details of the Analysis 197
6.3.2 The Limitations of the Monthly Mean Tide Analysis 202
v
6.4 Tide-by-Tide Analysis 202

6.4.2 The Benefits and Limitations of the Tide-by-Tide Analysis 202
6.5 Reversing Reduction Analysis for Tidal Currents 205
6.5.2 Advantages and Limitations of the Reversing Reduction Analysis 210
6.6 Rotary Reduction Analysis for Tidal Currents 210
6.6.1 The Benefits of Using the Rotary Analysis 210
6.6.2 The Limitations of the Rotary Analysis 212
6.6.4 Special Consideration and Choices To Be Made For the Rotary Analysis 215
6.7 Selecting a Reference Station 217
6.8 Putting a Subordinate Station In a Tidal Current Table and On a Tidal
Current Chart 218
6.9. Minimizing the Errors Due to the Variations Throughout the Month of
Nonharmonically Determined Time and Height Differences 219
6.10 Summary Overview: Steps In Analyzing Water Level or Current Data With
a Nonharmonic Comparison Method 220
7. Interpretation of Tidal Analysis Results Based On

Hydrodynamics .223
7.1 Products for showing Tidal Analysis Results 223
7.1.1 Cotidal, Corange, and Co-amplitude Charts 223
7.1.2 Cospeed and Cophase Charts 226
7.1.3 Tidal Current Ellipses 226
7.1.4 Various Types of Graphs 229
7.2 Spatial Variation in Tidal and Tidal Current Parameters In a Waterway
Due to Hydrodynamic Processes 229
7.3 The Equations of Motion as the Basis for Tidal Hydrodynamic Models 232
7.3.1 The Equation of Continuity 232
7.3.2 The Momentum Equations 233
7.3.3 Simplifying the Equations of Motion For Idealized Cases 236
7.4 Tidal Variations Along a Very Narrow Waterway 238
7.4.1 Tidal Variations In a Narrow Rectangular Waterway 238
7.4.2 Tidal Variations In a Narrowing Waterway 247
7.5. Tidal Variations Along Wider Waterways the Effect of Coriolis 251
7.6. Nonlinear Shallow-water Effects on Tides 254
7.6.1 The Nonlinear Terms In the Equations of Motion 254
7.6.2 The Generation of Compound Tides and Overtides 256
7.6.3 The Nonlinear Interaction of the Tide With River Flow 260
7.6.4 The Nonlinear Interaction of the Tide With Low-Frequency Storm Surge 262
7.6.5 A Physical Explanation of the Nonlinear Mechanisms 263
7.6.6 A Real-World Example of Nonlinear Shallow-water Effects Using a
Numerical Model 267
7.6.7 The Effect of the Nonlinear Lateral Inertial Terms 273
vi
Table of Contents
7.7. Variation of Tidal Currents With Depth 274

7.8. Tidal Dynamics of the Oceans of the World 277
8. The Use of Numerical Hydrodynamic Models For

Predicting Tides and Tidal Currents .281
8.1 Introduction 281
8.2 The Workings of a Very Simple Numerical Hydrodynamic Model 281
8.3 Advantages of Using a Numerical Hydrodynamic Model 284
8.4 Disadvantages of Using a Numerical Hydrodynamic Model 287
8.5 Proper Tidal Forcing of a Numerical Hydrodynamic Model 288
8.6 Tidal Validation of a Numerical Hydrodynamic Model 290
8.7 The Operational Use of Numerical Hydrodynamic Models 293
8.7.1 Creation of Tidal Current Charts, Forecast Atlases and Digital Tidal
Prediction Products 293
8.7.2 Tide-Based Corrections In Hydrographic Surveys 295
8.7.3 Real-time Based Nowcast/Forecast Model Prediction Systems 296
9. Products for Presenting Tide and Tidal Current Predictions .299

9.1 Introduction 299
9.2 Tide Tables 299
9.2.1 Table 1 Daily Predictions for References Stations 300
9.2.2 Table 2 Time and Height Differences for Subordinate Stations 302
9.2.3 Table 8 Tide Prediction Accuracy 304
9.2.4 Other Tables and Information in the Tide Tables 306
9.2.5 Limitations of Tide Tables 307
9.2.6 The Construction of Tide Tables 308
9.3 Tidal Current Tables 309
9.3.1 Table 1 Daily Predictions for References Stations 310
9.3.2 Table 2 Differences and Ratios for Subordinate Stations 310
9.3.3 Table 5 Rotary Tidal Currents 313
9.3.4 Other Tables and Information In the Tidal Current Tables 313
9.3.5 Current Diagrams 314
9.3.6 Limitations of Tidal Current Tables 316
9.3.7 The Construction of Tidal Current Tables 317
9.4 Tidal Current Charts 318
9.4.1 Description of Tidal Current Charts 318
9.4.2 The Advantages and Limitations of Tidal Current Charts 321
9.4.3 The Construction of the Tidal Current Charts 322
9.5 Tidal Circulation and Water Level Forecast Atlases 323
9.6 Tide and Light Diagrams a bit of history 326
9.7 Digital Tidal Prediction Products 330
Appendices .333
References .345
vii
List of Figures .359

List of Tables .366
Acknowledgments .367
Index .369
viii
1. Introduction
1.1 Purpose of This Book
The purpose of this book is to provide the reader with the knowledge required to carry out the
most accurate tidal analysis and tidal prediction possible using any set of water level or current data
that he or she may have available. The book is also intended to provide the reader with tools to
interpret the analysis results with respect to the hydrodynamics (the physics of the water movement)
of the bay or ocean from which the data were obtained, so that these results can best be used for
particular oceanographic applications. Tidal analysis and prediction involves more than simply
running a harmonic analysis program to obtain tidal harmonic constants and then putting them in
a tidal prediction program. It requires understanding both the astronomical and the hydrodynamic
aspects of the tide. Lack of such an understanding can lead to problems when performing a tidal
analysis. A few examples of such problems are very briefly mentioned below (they are explained
in more detail later in this book, and the technical terms used below are defined in Chapter 2).
It is the astronomy, namely the relative periodic motions of the earth, moon, and sun, that
determines the frequencies at which tidal energy is found. The contribution to the tide by the energy
at each tidal frequency is usually represented by a tidal harmonic constituent, for which there will
be an amplitude and a phase lag. The pairs of amplitudes and phase lags are referred to as harmonic
constants. Which of these tidal constituents can be included in a harmonic analysis depends on the
length of the data times series one has available. The longer the time series the more tidal
constituents that can be included in the analysis and the more accurate the tidal predictions will be.
Attempting to include in the analysis more tidal constituents than can be resolved with the available
length of the time series can lead to erroneous results, or even to no results at all because in such
cases numerical instability can cause the harmonic analysis program to fail ( blow up). Even when
the appropriate tidal constituents are included in a harmonic analysis, one must remember that the
energy of the tidal constituents that could not be included in the harmonic analysis (because they
were too close in frequency to other larger tidal constituents) will still affect the constituents that
were included in the analysis. As a result one may see errors, namely, differences between the tide
predictions and the actual water level data, that slowly oscillate in time due to the missing tidal
constituents. Such errors may be significant if one has analyzed only 15 days of data or even 29
days.
It is the hydrodynamics of the ocean and bay that determines how large the tide or tidal current
will be at a particular location, as well as the timing of high and low waters, maximum floods and
ebbs, and slack waters. In shallow water the hydrodynamics becomes nonlinear, distorting the tide
and adding new higher harmonic tidal constituents (overtides) and new tidal constituents within the
semidiurnal tidal band (compound tides), some with the same frequencies as some of the original
astronomically caused tidal constituents. Knowing whether an analyzed constituent is a compound
1
tidal constituent or an astronomical tidal constituent (with the same frequency) can make a
difference in the accuracy of the subsequent tide predictions, especially when making predictions
for years other than the year whose data was used for the analysis.
Shallow-water hydrodynamics also causes nonlinear interactions between the tide and nontidal
phenomena such as river flow and wind-produced changes in water level (storm surges) and
currents. For example, high river flow reduces the tide range and distorts the tide curve (modifying
the astronomical tidal constituents and adding additional higher harmonic constituents, the
overtides). And so, if water level data obtained during a time period with high river flow are
analyzed and then tide predictions are made using the harmonic constants derived from those data,
the predicted high waters will be too small throughout the rest of the year. Likewise data obtained
during strong wind events may have tides that are modified by low-frequency storm surge and thus
are not representative of the rest of the year.
In most cases water level or current data are only available at a few distinct locations in a bay
or along a coast. Often some of these locations have data times series that are not long enough to
allow a useful harmonic analysis, so oceanographers developed other ways to extract tidal
information from locations with short data time series. For decades this has been done
nonharmonically, by simply comparing the high and low waters in water level data from the short
stations (usually called subordinate stations, or secondary ports) with the high and low waters in
predictions harmonically derived from longer stations (usually called reference stations, or standard
ports). However, there can be severe limitations on how well this can work, due to the
hydrodynamics of the location where the data were obtained.
Although this book provides some rules of thumb for carrying out tidal analysis and
prediction, the intent is to go well beyond this. This book explains not just the how but also the
why, namely it provides explanations of the astronomical causes of the tide and the hydrodynamic
modifications of the tide, so the reader can determine how to maximize the accuracy of the analysis
results and predictions. This understanding is also important for interpreting the analysis results.
This book explains and illustrates all state-of-the-art tidal analysis and prediction methods
presently in use, as well as the astronomical, hydrodynamic, and statistical theories behind them.
This is not intended to be a complete textbook on tides. The emphasis here is on subjects the reader
must understand in order to carry out accurate tidal analyses and to make skillful tidal predictions.
However, in meeting this objective, the result is a reasonably complete study of the tides (with
references for subjects not covered in detail). The book provides practical operational procedures,
including considerations related to maximum analysis accuracy and maximum prediction skill.
The book is written at an introductory level, so that the reader should need little background in
tidal or oceanographic theory. With an eye toward the teaching aspects of this book, it begins with
a general overview of the subject of tides, so that the reader can first see the big picture. Then as
the material becomes more detailed, the reader will be able to understand that material within a
larger context. Since the astronomical and hydrodynamic aspects of the subject affect each other,
it was felt that such an overview should be given first, rather than simply jumping right into detailed
astronomical theory followed by detailed hydrodynamic theory. Because of this approach, there
may occasionally be some redundancy, as well as frequent references to other sections in the book.
Although this book is written at a level accessible to the nonexpert, it is also hoped that tidal experts
will still find of interest some of the topics that they may not have dealt with themselves.
2
1. Introduction
1.2 Relationship to Previous Tidal Analysis and Prediction Books and Manuals
The Center for Operational Oceanographic Products and Services (CO-OPS) in NOAAs
National Ocean Service (NOS) continues to print and provide tidal analysis and prediction manuals
and reports produced in the past (under previous organizational names such as the Coast Survey, the
Coast and Geodetic Survey, and the National Ocean Survey). In particular the Manual of Harmonic
Analysis and Prediction of Tides by Paul Schureman (first published in 1924, revised in 1940, and
reprinted with corrections in 1958) and the review/update of that manual, Computer Applications
to Tides in the National Ocean Survey, by Bernard Zetler (1982) are still useful. The technical
report A Users Guide to a Computer Program for Harmonic Analysis of Data at Tidal Frequencies
by Robert Dennis and Elmo Long (1971) describes the first computer implementation of
Schuremanss harmonic analysis method, for data time series of length 15 days or 29 days. When
Schuremans manual was first printed in 1924 (and for four decades thereafter) harmonic analyses
were done by hand using special paper keys. These keys were a series of paper templates with
holes in them, each of which was placed over sheets of paper that had long columns of hourly tidal
heights. The numbers showing through the holes were added with mechanical adding machines as
part of a very time consuming process. Once the harmonic constants were calculated, tide
predictions were then produced using large brass analog tide predicting machines with dozens of
pulleys and gears representing the tidal harmonic constituents that came out of the hand analysis (see
Figure 1.1).
These past manuals, besides being out of date in numerous respects, are also generally narrow
in their scope, concentrating only on the one rendition of the harmonic analysis method developed
by Schureman. Schuremans original work (often simply referred to as S.P. 98) is still quite
valuable from the standpoint of the elaborate astronomical equations that he worked out, and the
(sometimes limited) explanations behind them. The 1982 Zetler review/update of the Schureman
manual made a few corrections to Schuremans text (very few actually, considering the extensive
astronomical/mathematical work done by Schureman). He also put Schuremans work in the context
of important British work done by Arthur Doodson (a contemporary of Schuremans) and by David
Cartwright (a contemporary of Zetler, but who was still to contribute a great deal more after 1982).
Zetler also updated Schuremans node factor and equilibrium argument tables to the year 2025.
None of these manuals provided much insight on the best way to use the harmonic method for
producing the most accurate results. Also, there was only some mention of the importance of
hydrodynamic effects on tidal analysis and prediction.
Prior to Schuremans manual there were also books or sections of major Coast Survey reports
that dealt with tidal analysis and prediction and which in many cases provided the bases for
Schuremans work. These included William E. Ferrels Tidal Researches (1874) and Rollin A.
Harriss five part Manual of Tides (1897-1907). These reports included much original research and
analysis, and were not really instructional manuals. Harry A. Marmers The Tide (1926) was not
a manual but rather an introduction to the subject, which included an explanation of harmonic
analysis and prediction.
For some decades now, the Schureman harmonic method (based on Fourier analysis) has only
been used at CO-OPS with data series of 15 and 29 days duration. That computer program (Dennis
and Long, 1972) makes use of Schuremans particular method of inference and elimination to infer
some additional tidal constants that cannot be calculated from only 15 or 29 days of data. It also
corrects the constituents that could be calculated for the effects of those constituents that could not
calculated (see Section 4.2.1). Schuremans method of inference and elimination is based on
3
Figure 1.1. The Harris-Fischer tide predicting machine.
astronomical considerations (in the form of the so-called equilibrium theory, see Section 2.1.2), and
so does not always work well in the real world, where hydrodynamics has a major influence.
Another program has been the primary method used for tidal analysis at CO-OPS. It was
originally written in the 1960's by D. Lee Harris using a least squares harmonic analysis technique
(Harris, et al., 1963). Unfortunately there has never been more than an informal in-house manual
for that program, although the program listing is provided in the Zetler publication mentioned above.
Over the years the Schureman/Dennis&Long program and the Harris program have both been
4
1. Introduction
modified by the author of this book and by others in CO-OPS (and its predecessor organizations)
for use with tidal currents and for additional refinements. Here again the only manuals have been
in-house manuals, although they have sometimes been documented in journals and other scientific
literature (which are included in References at the end of this book). A recent CO-OPS technical
report (Zervas, 1999) serves as an updated operational manual with instructions on how to run
specific CO-OPS tidal analysis and prediction computer programs with current data.
Of course, outside of CO-OPS and its predecessor organizations other versions of harmonic
analysis have been developed, as well as other types of tidal analysis techniques, of which the
response method of Munk and Cartwright (1966) is best known. These other methods are also
described in this book. Some of the manuals, books, and papers describing these methods are
reasonably broad in scope. Probably the classic tidal analysis and prediction manual of all time is
the Admiralty Manual of Tides by Doodson and Warburg (1941), which includes much of Arthur
T. Doodsons important original work in tidal analysis and prediction. A broader coverage of the
subject of tides that included much more hydrodynamics was J.J. Dronkers Tidal Computations In
Rivers and Coastal Waters (1964), as well as Albert Defants Physical Oceanography (1961; 272
pages of volume II are devoted to tides). The first of Gabriel Godins two books on tides The
Analysis of Tides (1972), as the title indicates, concentrated on tidal analysis and prediction and was
the basis for the analysis and predictions programs used in Canada written by Mike Foreman
(Foreman, 2004). Godins second book, simply called Tides (1988) and published in Mexico, goes
well beyond analysis and prediction and includes a great deal of hydrodynamics, including nonlinear
effects. A year earlier David T. Pugh published Tides, Surges, and Mean Sea-Level (1987), which
includes a chapter on analysis and prediction as part of a broader coverage of the tidal and nontidal
effects on water level and currents. Pughs more recent book Changing Sea Levels (Pugh, 2004) has
four chapters on tides including one on analysis and prediction.
The book Tidal Hydrodynamics (Parker, 1991), edited by the author of this book, includes many
chapters on analysis and prediction by the worlds leading tidal experts at that time, as well as
chapters on the hydrodynamic influences on tides. In 1999, David E. Cartwright published Tides
A Scientific History, which although a history, provides a great deal of information and insights
on tidal analysis and prediction. Just recently published is Understanding the Tide by Steacy Hicks
(2006), a basic introduction to tides including analysis and prediction. These are all valuable and
useful books.
In addition there have been various manuals for specific analysis and prediction computer
programs, which primarily provide instructions on how to run those programs, but sometimes also
provide additional useful information related to the particular statistical technique and/or tidal theory
behind it. A good example are the manuals written by Mike Foreman (Foreman, 2004a, 2004b).
The present book is positioned somewhere between the manuals mentioned above (both inside
CO-OPS and outside) and the more theoretical texts, books, and papers also mentioned above. It
is like a manual in its how to slant, while still providing a good deal of theoretical background,
both astronomical and hydrodynamic. That theoretical background, however, is provided in the
context of helping the reader carry out better tidal analyses and produce more accurate tidal
predictions. Although the emphasis is on harmonic and nonharmonic analyses, this book also
includes other methods of tidal analysis and prediction. It includes descriptions of the statistical,
astronomical, and hydrodynamic theories on which the analysis and prediction techniques are based,
and which must be understood for appropriate interpretation of the results. It includes practical
operational step-by-step instructions on how to carry out an analysis and prediction, as well as
5
specific things to look for in the data and in the hydrodynamic situation that could affect or even
compromise the results.
This book does not provide instructions on how to run specific computer analysis and prediction
programs, because such programs generally involve data formats and particular parameters that
require certain values, and there are technical reports or in-house manuals for that purpose.
Knowing how to run a computer program is different than understanding what the program is doing,
how to choose the data properly, and how to use the results. This book also does not repeat the
elaborate mathematical derivations of the equations dealing with the tide producing forces or the
astronomical aspects of the various analysis techniques (found in Schureman, and in papers or
reports by Ferrel, Harris, Doodson, Cartwright, and many others), but references are given. This
book also may not cover some theoretical aspects of the tides included in some textbooks. The
emphasis here is on the knowledge required to carry out the most accurate tidal analysis and tidal
prediction possible for a particular set of water level or current data, and to interpret the results.
1.3 The Uses for Tidal Analysis and Prediction
There are various causes of changes in water level and currents, but for most coastal and marine
inland waterways the astronomical tide is usually the dominant cause. The tide and tidal current are
by far much more predictable than changes in water level and currents caused by changes in the
wind (usually the second most dominant cause, except during storms when it dominates the tide),
changes in atmospheric pressure, changes in river flow (which can be dominant during spring runoff
periods), or changes in density (caused by changes in temperature and salinity). Throughout the
centuries, because of the tides usual dominance and its predictability, changes in water level have
often been simply referred to as the tide even when other effects were included. Extreme changes
in water level that cause flooding during a storm are still often referred to as the storm tide, even
though the wind component (the storm surge) is often larger than the astronomical tide component.
For any situation where there has been a need to know the water level, the mariner has typically
used a tide prediction, even in place of an actual water level observation. Recently, however,
mariners have begun using real-time water level information [from systems such as CO-OPS
Physical Oceanographic Real-Time System (PORTS)] when making on-the-water decisions,
although they still rely on tide predictions for planning purposes.
There are a whole host of needs for water level information. The most frequent use was (and
probably still is) for navigation. For a deep-draft vessel especially, having sufficient water depth
under its keel is critical so that it does not run aground. Many large vessels, such as tankers, cargo
ships, and container ships, plan their departure from or arrival at a port to coincide with the time of
high water, and thus critically depend on tide predictions. Even in waterways where real-time
systems like PORTS are used to provide more accurate information about actual water level values
(which includes wind and other effects), it is still the tide prediction that is used for advanced
planning. When hurricanes or storms are projected to hit a coast and cause flooding, the height of
the tide during its landfall is critical to how much flooding there will be (and to whether the National
Weather Service will issue coastal flood warnings). For areas with large tide ranges, there will be
less damage if a storm makes landfall near the time of low water. Tide predictions are thus used for
emergency management planning. The tide often determines the type of habitats that can exist in
specific locations or the degree to which waves can erode beaches or harm a wetland. Since the
depth soundings and the shorelines depicted on all nautical charts are referenced to a tidal datum
(a vertical reference level based on the average high or low tides), tide predictions are also critical
6
1. Introduction
during the hydrographic and photogrammetric surveys that obtain the data for those charts. Even
today as hydrographic/bathymetric and shoreline surveys make use of real-time water level
observations and the global positioning system (GPS), tide predictions are still important for
planning the surveys.
Often it is information on currents that is needed, and here again for most coastal and marine
inland waterways it is the tidal current that is the dominant component, and the component that is
far more predictable than the wind-driven component or the gravitational component (caused by
salinity differences up an estuary). There are many reasons for wanting to know the currents.
Navigation again is one use, since strong currents can greatly affect the maneuvering of a ship
(especially when docking), so much so that the prediction of slack water is an important
requirement. In some narrow channels where large vessels must pass each other in opposite
directions, the right of way is given to the ship moving with the fair tide (i.e., to the ship moving
with the tidal current) because it has less maneuverability. Tidal current predictions are often used
to determine this. Currents are critical for transporting oil spills, flushing pollutants from a harbor
or estuary, transporting particular species of marine life, and stirring up and transporting sediments
(and attached pollutants). Tidal current predictions are therefore used by responders to oil spills,
environmental managers, fishermen, and ecologists, to name just a few.
Today numerical hydrodynamic computer models are used for a whole range of activities from
water quality modeling to coastal engineering to harmful algal bloom landfall forecasts. Tide
predictions are needed as a primary forcing to run such models, which are used to provide forecasts
and to produce scenarios for planning.
Ironically the most accurate tidal predictions are sometimes needed for studies where the tidal
signal must be removed from data records, for example, for global sea level studies dealing with
seasonal-to-interannual phenomena such as the El Nio Southern Oscillation (ENSO), or for studies
of basin-scale circulation using satellite altimetry. It is especially important for removing the tidal
aliasing from satellite altimetry data, which not only have large time intervals between data points
at a specific geographic location (from 3 to 20 days, depending on the satellite) but can have orbit
errors that fall in the tidal frequency bands. In these applications centimeter accuracy is required.
For tsunami warning systems, water level station observations during tsunami events are detided
using predicted tides so that the signature of a tsunami wave can be more clearly seen in the data
record.
1.4 Brief Historical Background
Only a very brief historical background on tidal analysis and prediction is included here. For
a much more complete treatment see Cartwright (1999), as well as a much earlier but also interesting
work by Harris (1897). Hicks (1967) presents a history dealing specifically with the Coast Survey,
which was renamed the Coast and Geodetic Survey (C&GS), of which CO-OPS predecessor
organizational units were a part. Zetler (1987, 1991) extends that history another 20 years, by which
time C&GS had become the National Ocean Service (NOS) and was part of NOAA. [Prior to this,
NOS was briefly known as the National Ocean Survey, when it was part of the Environmental
Science Services Administration (ESSA).]
Tidal prediction is the oldest form of ocean prediction, and is still the most accurate. The two
earliest tide tables that have so far been discovered were for the tidal bore in the Tsientang River in
China in 1056 and for London Bridge on the Thames River in England in the early 1200's. The long
history of tide prediction is a result of the long-recognized correlation between the tide and the
7
changing position and phases of the moon. That correlation was obvious even to ancient
civilizations. Centuries ago mariners invented rule-of-thumb techniques for tide prediction, which
were often treated as treasured family secrets and passed on to the next generation. Beginning in
the late 1700's fairly sophisticated nonharmonic techniques using various lunar characteristics were
used to predict the tide. For example, for a particular place, high tide might occur a certain number
of hours after the moon was directly overhead, and the highest (spring) tide might occur a certain
number of days after full moon or after new moon. In 1752 Daniel Bernoulli produced the first
practically reliable tide table in France using luni-tidal intervals (see Section 2.2.5) that took into
account the moons elliptical orbit (see Section 2.2.1). In 1832 John Lubbock produced tide tables
in England using a nonharmonic method that also included the effect of lunar declination (see
Section 2.2.1).
It was not until 1867 that the much more accurate harmonic method was developed by Sir
William Thomson (later called Lord Kelvin) (1869, 1881). This was 180 years after Sir Isaac
Newton first explained how tides are generated by the gravitational effects of the moon and sun, and
91 years after Pierre Simon Marquis de Laplace first suggested representing the tide as a series of
harmonic oscillations. The harmonic method was modified and improved by George Darwin (1883)
and Arthur Doodson (1921, 1928) in England. In the U.S., William Ferrel (1874) in the Coast and
Geodetic Survey developed harmonic analysis and prediction totally separate from the work of
Thomson and Darwin, Ferrel basing his work on the papers of Laplace as had Thomson. This work
was further improved and added to by Rollin Harris (1897-1907). About twenty years later, at the
same time that Doodson was carrying out his tidal work in Britain, Paul Schureman was also
working out the needed astronomical equations and Fourier techniques for harmonic analysis and
prediction (Schureman, 1924).
Understanding that all the energy in the tide is found at particular frequencies (mostly in the
semidiurnal and diurnal bands) these researchers used a harmonic method to analyze a time series
of water level measurements and to determine how much energy is at each of these tidal frequencies,
for the specific location of the water level measurements. [Back then such water level measurements
were made by hand by noting where the surface of the water was with respect to a long graduated
staff permanently mounted on the side of a pier or some other solid structure.] They represented the
contribution to the tide of the energy at each tidal frequency using a tidal harmonic constituent, for
which there was an amplitude and a phase lag (called an epoch), which were referred to as harmonic
constants. The amplitude is the maximum height that a tidal constituent contributes to the tide, and
the epoch is the time when this maximum constituent contribution occurs (relative to a reference
time, such as the moons transit over a local time meridian). By combining the amplitudes and
epochs of all the tidal constituents, one can predict the tide at any time on any day of any year.
In that pre-computer era Thomson came up with an ingenious way to automate tide predictions
using harmonic tidal constituents. He invented a mechanical analog tide predicting machine, a
machine made up of dozens of gears and pulleys, with a wire running over all the pulleys and then
connected to a pen touching a moving roll of paper. Each tidal constituent was represented by one
gear with a pin and yoke arrangement that turned the rotating motion of the gear into an up and
down motion, moving a pulley up and down and thus providing that constituents contribution to
the tide curve being drawn by the pen on the moving roll of paper (see Figure 1.2). The machines
built to predict the tides were made out of brass and were finely crafted. Thomsons first tide
predicting machine was built in London in 1872 by the Lg Engineering Company and summed
the contributions of the ten most important tidal constituents. In the U.S., Ferrel was not far behind,
designing a 19-constituent machine for the U.S. Coast and Geodetic Survey that was built in
8
1. Introduction
Figure 1.2. Gear and pulley system of an early analog tide

predicting machine. Each gear and pulley combination represents
one tidal harmonic constituent. The wire running over all the
pulleys sums the motions and moves a pen on a moving roll of paper
to draw the tide curve.
Washington in 1882 by Fauth & Company. Ferrels machine directly predicted high and low waters
instead of tide elevations. In 1894 in the U.S., Rollin Harris designed a second tide predicting
machine, which was completed in 1912 in the workshops of the U.S. Coast and Geodetic Survey.
It used 37 tidal constituents and predicted tide elevations or tidal currents (see Figure 1.1).
By the time that World War II broke out in 1939 many maritime nations had one of these
huge brass tide predicting machines, and accurate tide predictions for the major ports of the world
were commonplace. However, it still required a major manual effort to harmonically analyze water
level data in order to calculate the tidal harmonic constants that would be represented by the
numerous wheels on one of these machines. The records from the tide gauges used at that time were
long analog strip-chart graphs from which hourly heights were scaled off and written onto specially
designed forms, along with the times and heights of high and low waters. To perform a harmonic
analysis, a series of keys or stencils were overlaid. These stencils were sheets of paper with holes
revealing only particular observations for use in subsequent summation of a particular constituent
onto another form. This manual iterative process could take a week or more to harmonically
analyzed a years worth of data. It takes seconds on a computer today.
The brass tide predicting machines were used by numerous countries until the mid 1960s when
computers took over. Today both tidal analyses and tidal predictions are done on high-speed
computers, but the statistical techniques used do not differ that much from those earlier methods,
although there have been refinements based on modern numerical techniques.
Likewise there have been improvements in obtaining water level data. The tide staff attached
to a pier was eventually replaced by a stilling well (to damp out wave action) with a float inside,
which was attached to a wire that moved a pen on a moving roll of paper. More recently, acoustic
methods have been commonly used, where the height of the water surface relative to some reference
point above the surface can be determined by the time it takes sound to reach the water surface and
reflect back. Other techniques have also been used such as satellite altimetry, GPS receivers on
buoys, laser systems, and radar systems. (More information on water level measurement is found
9
in Section 4.1.3. Current data measurement methods are briefly discussed in Section 5.3.3.) Modern
communication techniques and the Internet have made real-time systems such as PORTS a reality.
Most important for data analysis, data from these modern systems are obtained in real time, so that
quality control procedures can spot malfunctions that might have led to weeks of bad or missing data
in self-recording instruments.
In the U.S. the first published tide predictions appeared in The American Almanac in 1830. They
were produced by the U.S. Coast Survey, the nations first scientific agency, established in 1807 by
Thomas Jefferson. These predictions consisted merely of high water predictions for Boston, New
York, and Charleston, along with time differences for 96 other locations, along with the spring range
predictions for 84 locations. (see Hicks, 1967). Low water predictions were not included until 1887.
In 1844 tide notes began to appear on nautical charts. Finally in 1867 the first official U.S. Tide
Tables were published, produced by the Tide Division of the Coast Survey, the first national
government agency to take responsibility for publishing tide predictions (in Great Britain tide tables
were produced by private companies; Cartwright, 1999, p.91).
In 1885 tide predictions in the U.S. were produced for the first time using the brass analog tide
prediction machine designed by Ferrel. This was the first time the harmonic prediction method was
used in the U.S., in this case accounting for 19 tidal constituents. (Prior to 1885, daily predictions
had used a nonharmonic method making use of lunitidal intervals.) In 1887 low waters began to be
predicted, and three years after that, tidal currents began to be predicted. In 1912 the Ferrel machine
was replaced by the Harris-Fischer tide predicting machine, which used 37 tidal constituents. In
1923 tidal current predictions were published in two volumes separate from the four volumes of tide
predictions. In 1966 electronic computers finally replaced the Harris-Fischer analog tide predicting
machine in producing the nations Tide and Tidal Current Tables, using a program written by Harris,
Pore, and Cummings (1965). Since that time numerous improvements and additions have been
made to tide and tidal current analysis and prediction in the Coast and Geodetic Survey and its later
reincarnations (as government agencies were reorganized) leading to the present Center for
Operational Oceanographic Products and Services (CO-OPS) in the National Ocean Service (NOS)
in the National Oceanic and Atmospheric Administration (NOAA) in the U.S. Department of
Commerce (DOC). These improvements and additions are mentioned at appropriate places in the
following chapters of this book.
1.5 The Importance of Hydrodynamics
The traditional harmonic tidal prediction methods mentioned in the last section ignore
hydrodynamics (that is, the physics of the water movement). They rely only on a knowledge of the
astronomically produced tidal frequencies. The forced hydrodynamic system guarantees that tidal
energy in a water level data time series (or a current data time series) will appear only at those
particular astronomical frequencies. All that is required then to make an accurate tide or tidal
current prediction at a particular location is a water level data series (or current data time series) that
is long enough so that, when it is harmonically analyzed, the most important of these lunar and solar
tidal frequencies can be resolved, and an amplitude and epoch (phase lag) can be determined for
each tidal constituent.
As a tide wave propagates in the ocean, onto the continental shelf, and up into bays and
estuaries, the hydrodynamics of these bodies of water (e.g., full and partial reflections, continuity
effects, inertial effects, and frictional damping, all of which are discussed in this book) affects the
tide range and phase lag of the tide, and the amplitude and epoch of each of the tidal constituents.
10
1. Introduction
In shallow waterways the hydrodynamics, through various nonlinear processes, also transfers tidal
energy to new tidal frequencies. However, these new frequencies can be determined, and additional
shallow-water tidal constituents can then simply be included in the harmonic analysis. The details
of how these overtides (higher harmonics of the astronomical constituents) and compound tides (new
tidal constituents resulting from the interaction of two or more astronomical constituents) are
actually generated have often been ignored. In some cases the mechanisms that produced them were
not even fully understood, and the shallow-water constituent frequencies were merely determined
by adding or subtracting astronomical frequencies. Not understanding the nonlinear hydrodynamics
that generate these constituents can cause problems in tidal prediction accuracy for years other than
the one whose data were analyzed. Improvements in tidal prediction have come about as a result
of taking hydrodynamics into account.
In recent years numerical hydrodynamic models, which include all the necessary ocean physics,
have been used to produce tide and tidal current predictions. These models have the advantage of
being able to provide predictions at hundreds and even thousands of locations. This is especially
beneficial for tidal currents, which can vary dramatically over short distances. For example, tidal
currents near the sea bottom can be very different in speed and direction than those near the water
surface. Tidal currents can also be quite different in a channel and in the shallows just a few feet
away. Numerical hydrodynamic models can also inherently handle the nonlinear interaction
between the tide and nontidal phenomena such as storm surge and river flow. However, such
models still depend on having very accurate tide predictions to force the models at their seaward
open boundary. While these hydrodynamic models can provide predictions at an almost limitless
number of locations they are still rarely as accurate as tide predictions derived from a well-done
harmonic analysis at a location with a sufficiently long data time series. However, there will be
situations where a model should be able to provide better tidal current predictions than can be
obtained from harmonically analyzing current data. The degree of accuracy required, of course,
varies considerably with the application for which the predictions are to be used, and for many
applications the accuracies from numerical models are more than sufficient.
1.6 Organization of This Book
Chapter 2 begins with a general overview of the subject of tides that summarizes in introductory
terms both its astronomical and hydrodynamic aspects. This is intended to give the reader the big
picture first, so that when the material becomes more detailed, the reader will be able to understand
that material within a larger context. Since the astronomical and hydrodynamic aspects of the tide
are interrelated, it will be helpful to understand this connection before jumping right into detailed
astronomical theory followed by detailed hydrodynamic theory. However, because of this approach,
there may occasionally be some redundancy, as well as frequent references to other sections in the
book.
This overview (including many basic definitions) is followed by a section providing an
introduction to the astronomical aspects of the tide. It describes which frequencies have tidal energy
and the specific aspects of the moons orbit around the Earth and of the Earths orbit around the sun
which are responsible for this energy. This is followed by a section dealing with the hydrodynamic
aspects of the tide. It is the hydrodynamics (determined by the dimensions of the ocean, continental
shelves, bays, and rivers) that determines how large the tide range will be, how fast the tidal currents
will flow, the time of high and low waters (and maximum floods and ebbs and slack waters), and
the type of tide (that is, whether it is semidiurnal, diurnal, or somewhere in between). This section
11
also explains how shallow water (through nonlinear processes) causes the transfer of tidal energy
to other frequencies, including higher harmonics of the original astronomical tidal frequencies and
the corresponding distortion of the tide and tidal current curve. These same nonlinear processes in
shallow water lead to interactions between the tide and various nontidal phenomena such as river
flow and storm surge. Finally, Chapter 2 ends with a section devoted to the special aspects of tidal
currents not found in the tide.
Chapter 3, before giving an overview of the various methods for tidal analysis and prediction,
explains the important effect that the length of the data time series has on the ability to resolve
nearby tidal frequencies. The longer the data time series is the more tidal energy that can be
accurately represented and the more accurate the tidal predictions will be. This chapter discusses
harmonic analysis (using either a Fourier technique or a least squares technique), the response
method, and several other frequency domain methods. It also discusses several nonharmonic
comparison analysis techniques, which are still important today because of their use in calculating
time and height differences for thousands of so-called subordinate stations (or secondary ports)
listed in tide tables, those differences being relative to a nearby (harmonically produced) reference
station. In addition, Chapter 3 covers tidal filtering (and detiding), spectral analysis, empirical
orthogonal function analysis, the importance of calculating consistently defined maxima and
minima, and the dominant meteorological energy found in the annual and semiannual tidal
constituents Sa and Ssa.
Chapter 4 describes in more detail the harmonic analysis of water level data including specific
things to look for in the data and in the hydrodynamic situation that could affect or even compromise
the results, including the potential effects of nontidal influences. It includes a discussion of methods
for analyzing short data time series, for which one could normally not obtain information about all
the required tidal frequencies. There is a discussion on assessing the quality of predicted tide series
produced with the harmonic constants that come out of the harmonic analysis. The chapter ends
with a detailed list of all the steps that should be taken to harmonically analyze water level data.
Chapter 5 is similar to Chapter 4, except that it deals in detail with the harmonic analysis of
current data. This chapter covers the same basic topics as the previous chapter, but prefaces this
discussion with an explanation of the special problems that one has to deal with in the analysis of
currents. It explains the use of tidal currents ellipses, and again ends with a detailed list of all the
steps that should be taken to harmonically analyze current data.
Chapter 6 deals in detail with four basic types of nonharmonic comparison analyses, including
how to deal with the variations of the time and height differences (or current speed ratios) that occur
throughout the month, and which can be quite significant if the subordinate station and reference
station are not similar enough in their harmonic makeup. Such variations translate into prediction
errors when these time and height differences (or current speed ratios) are used to make predictions
at a subordinate station. This chapter also ends with a detailed list of all the steps that should be
taken to nonharmonically analyze water level or current data.
Chapter 7 goes back to look again at tidal hydrodynamics, but in more detail than how it was
first presented in Chapter 2. The information in Chapter 2 was meant to provide enough
hydrodynamic background to allow one to make the best choices when statistically analyzing the
tides and tidal currents. Chapter 7 looks at the hydrodynamics more thoroughly in order to help in
the interpretation of the results that come out of this statistical tidal analysis, especially the spatial
variation of tidal parameters over a waterway. Such insights are gained through the use of simple
analytical hydrodynamic models, whose derivation from the equations of motion (which explain the
physics of the water movement) is also explained. These simple models allow one to understand
12
1. Introduction
why the tide range, times of high and low water, tidal current speeds, type of tide, and other features
of the tide change spatially in ways that ultimately depend on the length, depth, and width of the
waterway. This chapter also explains in more detail the nonlinear effects of shallow water on the
tides and tidal currents, and the nonlinear interaction between the tide and nontidal phenomena (such
as river flow and storm surge), making use of a nonlinear numerical model and a Fourier
decomposition of the equations of motion.
Chapters 3 through 6 describe statistical methods for predicting the tide and tidal current, but
Chapter 8 looks at the use of numerical hydrodynamic models for this purpose, and the situations
where such models may do a better job than the statistical methods. The chapter discusses both the
advantages and disadvantage of using such models. The chapter also explains the importance of and
the methods for proper forcing of a numerical model, as well as the importance of and methods for
tidal validation of such a model. Chapter 8 ends with a description of three important applications
where numerical hydrodynamic models are being used today.
Finally, Chapter 9 discusses in some detail the various products that have been created to present
tide and tidal current predictions. This includes, of course, tide tables and tidal current tables. It
also includes tidal current charts and the related tidal circulation and water level forecast atlases, as
well as the tide and light diagrams used for a special purpose in the past. The chapter ends with a
discussion of the advantages of harmonically produced digital tidal prediction products.
13
14
2. Theory Behind Tidal Analysis and Prediction
2.1 Short Introductory Overview the Big Picture
Before looking in depth at tidal analysis and prediction, a short overview of the entire subject
will be presented. This look at the big picture will allow the reader to put the specific discussions
to follow within a larger context to see where it fits and how it is affected by other aspects of the
tide. This overview will begin with some basic tidal terminology to make sure that the reader is
clear on the specific meanings of some key terms.
2.1.1 Some Definitions
Tides are the periodic motion of the waters of the sea caused by the changing gravitational
effects of the moon and the sun as they change position relative to the rotating Earth (which will be
explained in more detail in Section 2.1.2). The tides in the oceans are actually very long waves
hundreds or thousands of miles long. Although produced by astronomical forces, their behavior in
the oceans and connected bays (and the size of the resulting water level oscillations) is determined
by hydrodynamics, that is, by the physics of the water movement. To fully understand and predict
the tides one must understand both its astronomical forcing and the hydrodynamics of the oceans,
bays, and rivers.
The vertical rise and fall of the water surface is usually referred to as the tide, while the
accompanying horizontal movement is referred to as the tidal current (or as the tidal stream in some
countries). In its simplest form (such as at the coast of an island in the open ocean) the plot of
changing tidal height looks like a sine wave, with the maximum height reached by the water surface
called high water, while the minimum height is called low water (see Figure 2.1.). In this figure the
sine curve oscillates above and below mean sea level (MSL). The difference in height between high
water and low water is the tide range. The period of this tidal oscillation is the time from one high
water to the next high water (or likewise, from one low water to the next low water) and is typically
12.42 hours for most waterways of the world, but can be 24.84 for some areas. The tidal frequency
is the inverse of the tidal period, and so is one complete tidal cycle in 12.42 hours, or 1.932 cycles
per day (cpd) [this referring to a solar day], or 2.0 cycles per lunar day [a lunar day being 24.84
hours long, as will be seen in Section 2.2.1]. Classically, in tidal science, frequency has been called
the angular speed. The angular speed is expressed in degrees per hour (o/hour), where 360o is equal
to one complete tidal constituent cycle. One can obtain the frequency in cycles per hour (cph) by
dividing its angular speed by 360o (to make it cpd, multiply by 24). The primary tidal frequency of
1.932 cpd is equivalent to an angular speed of 28.9841o/hour. However, tidal energy occurs at many
15
Figure 2.1. A simple tide curve with some definitions of

terms. (A lunar day is approximately 24.84 hours, or two
tidal periods.)
frequencies, with each frequency being related to some aspect of the relative movements of the
Earth, moon and sun, due to the moon-Earth orbit, the orbit of the Earth around the sun, and the
rotation of the Earth on its axis.
The tide range varies from day to day throughout the month (see Figure 2.2). The most obvious
change correlates with the phases of the moon. Near the times of new moon or full moon the tides
have a larger range and are called spring tides. Near times of first quarter and third quarter the tides
have smaller ranges and are called neap tides. Springs tides occur when the Earth, moon, and sun
are all lined up, so that the tidal bulges caused by the moon and by the sun add to each other (see
Section 2.1.2). Neap tides occur when the moon and the sun work against each other.
There are other variations in tide range throughout the month which will be described in Section
2.2.1, such as due to the changing distance between the moon and Earth, or due to the changing
angle of the plane of the moon-Earth orbit relative to the Earths equatorial plane. From month to
month throughout the year there is also some variation in tide range on a smaller scale, due to the
changing angle of the plane of the Earths orbit around the sun relative to the Earths equatorial
plane, or due to certain of the monthly variations coming into and going out of sync. There are also
variations from year to year throughout the 18.6-year lunar nodal cycle which are smaller, on the
order of 5 to 10% of the average tidal range. The reasons for this variation will be discussed in
Section 2.2.1 and 2.2.4. Times of high water or low water are usually given relative to the moons
transit over a particular location or over a time meridian (time zone), which can be local or the zero
time meridian at Greenwich, UK (referred to as Greenwich meridian Time, GMT, or Universal time, UT).
The tidal current flow into a bay is called the flood current and the tidal flow out of a bay is
called the ebb current. If the bay or channel is very narrow, a plot of changing tidal current looks
like a tide curve except the sine curve is above and below a zero speed line, which serves the same
reference function as a datum does for the tide (see Figure 2.3). This reversing tidal current, like the
tide, is a one-dimensional scalar quantity. In one tidal cycle, the current goes from its maximum
positive value, maximum flood, through the zero current speed (called slack water) to the maximum
negative value, maximum ebb, and back again through slack water to maximum flood. The slacks are
usually referred to as slack before flood (SBF) and slack before ebb (SBE). [Alternative names for the
above, include flood strength, ebb strength, slack flood begins (SFB), and slack ebb begins (SEB).]
16
Figure 2.2. Four tide curves showing how the tide range varies through the month.
These curves also illustrate different types of tides: a semidiurnal tide at Boston; a
mixed tide at both San Francisco and Seattle; and a diurnal tide at Dutch Harbor.
17
Figure 2.3. A simple tidal current curve with definitions for a reversing tidal
current.
However, for wider waterways, and especially in the open ocean, the tidal current is a two-
dimensional vector property. For this rotary tidal current the direction of flow rotates around the
compass, the tip of the current vector tracing out an ellipse (or a circle in the open ocean). (See
Figure 2.4) This rotation is due to both the Earths rotation and certain geographic conditions.
In most waterways of the Earth, the rise and fall (and the flood and ebb) of the tide occur twice
a day, in which case the tide is referred to as a semidiurnal tide. In most locations, for at least half
Figure 2.4. Examples of three rotary tidal currents and one reversing tidal current.
The first two are off an ocean coast and are idealized. The last two are within a bay
and are from actual data. In each example, each vector shows the speed and
direction of flow for that hour of the tidal cycle, but for the last two examples only
the tips of the vectors are shown (and connected).
18
of every month, the two high waters on a given day are not exactly the same in height, nor are the
two low waters. The higher of each pair of high waters is referred to as higher high water (HHW)
and the lower as lower high water (LHW), and likewise for each pair of low waters, i.e., higher low
water (HLW) and lower low water (LLW). The height difference between two successive high
waters (that is between HHW and LHW) is called the high water diurnal inequality and the height
difference between two successive low waters (HLW and LLW) is called the low water diurnal
inequality.
When the diurnal inequality of two high waters (or of two low waters) is significant, then the tide
is usually referred to as a mixed tide. But even for a mixed tide the diurnal inequality can disappear
twice a month when the moon is over the equator (equatorial declination). In some waterways the
height difference between the two successive high waters or the two successive low waters becomes
so extreme that in reality there is only one high water per day and one low water per day, in which
case the tide is referred to as a diurnal tide.
When one refers to the type of tide, that means the tide is either semidiurnal, mixed, or diurnal.
(See Figure 2.2). However, one needs to be careful and very specific how one uses type of tide, for
it can be confusing. One should specify whether one is talking about the tide on a particular day at
a station, or whether one is classifying the station itself. For example, a station might be classified
as being mixed, but on a few days of the year the tide on a particular day at that station might be
diurnal. This will be discussed more in Section 2.2.5 and will include the use of harmonic tidal
constituents in trying to develop criteria for classifying stations with regard to the type of tide. The
type of tidal current may also be semidiurnal, mixed, or diurnal, but in many locations the type of
tidal current will be quite different than the type of tide because of the hydrodynamics (as will be
seen in Section 2.3.1).
The tide is not the only phenomenon in the ocean that produces variations in water level and
currents. Such variations can also be caused by changes in wind speed and direction, changes in
atmospheric pressure, changes in river discharge, and changes in water density (due to changes in
salinity and temperature). But these nontidal changes are not periodic like the astronomical tide and
are not nearly as predictable, since they are associated with changing weather (or, on slower time
scales, with changing climate). There are cases where there can be short-lived periodic currents due
to meteorological effects such as land breeze-sea breeze and inertial currents, both discussed in
Section 2.3.6. One will also see annual variations in water level and currents due to seasonal effects
on temperature and winds, which vary from year to year, as discussed in Section 3.7. Nontidal water
level changes caused by changes in the wind and barometric pressure are usually referred to as storm
surges (even when the wind speeds are below what might be considered storm level). The term sea
level is generally used for longer-period slower changes in water level. Mean sea level is the
average of water level measurements over some time period (such as a day, a month, or a year),
which averages out shorter-term oscillations like the tide.
The tide dominates our thinking about changes in water level, not only because it usually causes
the largest water level changes (except during storms), but also because it is very predictable
(especially in comparison to how well the weather can be predicted). After analyzing only a month's
worth of water level measurements from a water level gauge the tide can be predicted quite
accurately (for that location) for years into the future. This high predictability is due to the tide's
periodic nature and our very precise knowledge of its astronomical forcing. The Earth-moon orbit,
the revolution of the Earth around the sun, and the rotation of the Earth on its axis involve periodic
motions with fixed and precisely known time periods. Tidal energy is found at the frequencies
determined by these astronomical motions.
19
While it is the astronomical forcing of the tide that is the basis for the tides predictability, it is
the hydrodynamics of the tide that is responsible for the size of the tide range, the timing of high and
low waters, and the type of tide, as well as the speed and timing of the tidal current. It is the length,
width, and depth of the bay or river (and of any adjoining waterways) that control the
hydrodynamics. Because of the hydrodynamics, tide ranges, times/phase lags of high and low tides,
the tide regime, and the relationship between the tide and tidal current vary with horizontal
(geographic) distance throughout a waterway, especially in shallower water. The variation in tide
range over a waterway is usually depicted on corange charts and the variation in the phase lag of
high water is likewise depicted on cotidal charts. [See Figures 2.5 and 2.6.] As will be seen later
in this chapter and later in the book, such charts are useful in understanding the hydrodynamics of
the tide in a particular waterway.
In addition, it is because of hydrodynamics that tidal currents often vary significantly with depth
and with horizontal distance. In shallow waterways the hydrodynamics also transfers tidal energy
to new frequencies, and distorts the shape of the tide curve away from a perfect sine curve. These
Figure 2.5. A corange chart showing the geographic variation of the tide range of the largest tidal
constituent M2 (defined in Section 2.2.2) for the Strait of Juan de Fuca - Strait of Georgia, located
between the U.S. and Canada (from Parker, 1977).
20
same shallow-water processes also lead to interactions between the tide and nontidal phenomena
such as storm surge and river discharge that can temporarily modify the tide for some period of time.
The largest tide ranges occur in shallow coastal waters, in particular, at the ends of certain bays
or along coasts with very wide continental shelves. (Why shallow water increases the tidal range
will be described in Section 2.3.1.) The increase in tide range and tidal current speeds that one sees
in the shallow waters of bays, rivers and straits can go to dramatic extremes if the circumstances are
right. Tide ranges greater than 50 feet (15 meters) can occur in Minas Basin in the Bay of Fundy
and at the southern end of Ungava Bay, both in Canada. Tidal ranges greater than 40 feet (12 m)
occur at locations such as the northern end of Cook Inlet near Anchorage, Alaska, in the Magellan
Strait in Chile, in the Gulf of Cambay in India, along the Gulf of St. Malo portion of the French
coast bordering the English Channel, in the Severn River in England, and along the open coast of
southern Argentina. Tidal current speeds greater than 15 knots (7.7 m/sec) occur in the Saltstraumen
Strait (on the western coast of Norway) and in Seymour Narrows (between Vancouver Island and
Figure 2.6. A cotidal chart showing the geographic variation of the time (phase) of high water
for the largest tidal constituent M2 for the Strait of Juan de Fuca - Strait of Georgia. The time
(phase) of high water is expressed in degrees of an M2 cycle; dividing by 360o and multiplying
by 12.42 hours converts this to hours. It takes 5.2 hours (150o) for high water to progress from
the entrance to the head of the waterway. (from Parker, 1977).
21
the mainland of British Columbia, Canada). Tidal currents of 10 knots (5.1 m/sec) are found in
South Inian Pass in Southeast Alaska and in Kanmon Strait, Japan.
When measuring the water level with a water level gauge (still referred to by many as a tide
gauge, because of the usual dominance of the tide) there must be a reference level for the
measurement of height of the water surface. This reference level is called a vertical datum. Vertical
datums have traditionally come in two categories: those based on tidally derived surfaces of high
or low water, called tidal datums, and those based on the gravity field of the Earth, called
orthometric datums. More recently a third category has become very important, the so-called 3D
or ellipsoid datums realized through space-based systems such as the Global Positioning System
(GPS).
A tidal datum at a particular location is generally defined as an average height of a particular
stage of the tide. To minimize all significant tidal variations (daily, monthly and yearly), a tidal
datum such as mean high water (MHW) is defined as the average of all the high water elevations
over a 19-year National Tidal Datum Epoch period (for example,1983-2001). Nineteen years is
selected because it averages out the variations in the tide due to the 18.6-year lunar nodal cycle
(explained in Section 2.2.4). Such a long time period also averages out most meteorological effects
on water level, which could bias a tidal datum computed from a shorter length data time series.
Even after averaging over a 19-year period there are some climatic variations that may not be
eliminated, and with the slow rise in mean sea level (due to land subsidence and/or global warming)
the National Tidal Datum Epoch is often updated after a number of years. Slow vertical land
movement also shows up as a slow change in mean sea level. The sea level change seen in a water
level record is actually a relative sea level change (i.e., relative to land). It can be made an absolute
sea level change only by subtracting out any land movement (such as glacial rebound, subsidence
due to sediment compaction, or tectonic movement). Upward land movement from glacial rebound,
for example, produces a lowering in mean sea level, while sinking of the land due to sediment
compaction produces a rise in mean sea level (which is sometimes confused with the rise due to
global warming).
Just as the tide regime varies with horizontal (geographic) distance, so do tidal datum elevations,
especially in shallower waters. They usually vary more rapidly in the horizontal direction than do
orthometric or 3D/ellipsoid vertical datums.
Tidal datums provide the vertical reference for bathymetry and shoreline on NOAAs nautical
charts. There are many tidal datums (e.g., mean lower low water, mean low water, mean sea level,
mean tide level, mean high water, mean higher high water, etc.), but in the U.S. two of these are
most important mean lower low water and mean high water. Chart datum on a nautical chart in
the United States is defined as the mean lower low water (MLLW) at each location, which is the
average of all the lower low waters over some time period, usually 19 years. Depth soundings on
a nautical chart are the depths below the chart datum, and the predicted tidal heights found in Tide
Tables are the heights above the chart datum. Adding the two together gives the total water depth
at that moment in time. The legal shoreline in the U.S. is the shoreline depicted on NOAAs nautical
charts, which is the mean high water (MHW) shoreline. More specifically, each point on a MHW
shoreline represents the horizontal position of the land-water interface at the time when the water
level at that point is at a height equal to the MHW elevation value at that point (Parker, 2003).
Bridge clearances are also referenced to MHW.
These tidal datums also provide the legal definition of marine boundaries (see Figure 2.7).
MLLW, for example, is the dividing line between Federal territorial seas and State submerged lands,
and MHW is the dividing line between many State tidelands and private uplands. All tidal datums
22
at a water level gauge are referenced to the land through geodetic leveling to a number of
benchmarks, which are brass markers driven into solid rock or other permanent structures. Tidal
datums can change over decades if the land subsides (or rises due to glacial rebound) or if relative
sea level rises due to effects such as global warming.
In some other countries the lowest astronomical tide (LAT) is used as chart datum. LAT is
defined as the lowest level which can be predicted to occur under average meteorological conditions
and under any astronomical condition (Pugh, 1987). There seems to be no generally accepted
method to determine LAT, although one commonly used technique is to select the lowest tide from
many years of astronomical tide predictions (often 19 years). Another method is to make predictions
for known astronomical extremes. LAT appears to have come into favor mainly because it eliminates
negative predicted low water values that occur occasionally in tide tables when MLLW (or MLW)
is used as the chart datum. LAT has been agreed to by members of the International Hydrographic
Organization for use on international charts and as a target chart datum for all countries. Although
the U.S. chart datum is MLLW, CO-OPS also calculates LAT and HAT (highest astronomical tide)
for reference stations in the U.S. Tide Tables based on 19 years of tide predictions (usually for the
National Tidal Datum Epoch, which is presently 1983-2001).
In addition to tidal datums, there are two important geodetic vertical reference systems
orthometric datums and 3D/ellipsoid datums. It is important to know the relationship among all
three types of datums for modern application to surveying and mapping. Orthometric heights are
determined by geodetic leveling (surveying) relative to the geoid. The geoid is defined by a specific
gravitational equipotential surface which best fits (in the least squares sense) theoretical global mean
Figure 2.7. Various tidal datums and their relationship to U.S. marine boundaries.
23
sea level. The current orthometric vertical control system is the North American Vertical Datum
(NAVD88), which was established by NOSs National Geodetic Survey (NGS) in 1991. NAVD88
is based on North American-wide adjustment of leveling observations holding fixed the elevation
of mean sea level at one water level station on the east coast of Canada. This replaced the older
National Geodetic Vertical Datum of 1929 (NGVD29), whose leveling adjustments were originally
adjusted to the local mean sea level elevations determined in 1929 at several water level stations on
the east and west coasts. Due to sea level change, crustal motion, subsidence and uplift, the loss
of bench mark networks and unwieldy adjustments, NGVD29 became severely outdated and
cumbersome to use. NGVD29 has often been used incorrectly as equivalent to present day local
mean sea level, mainly because it was the vertical reference datum used on topographic maps (e.g.,
from the U.S. Geological Survey). NAVD88 should also not be used as equivalent to local mean
sea level. All GPS positioning data are referenced to one of many 3D/ellipsoid datums, which are
based on a geometric model, i.e., the ellipsoid. The ellipsoid is a two-parameter mathematically
modeled smooth theoretical surface that approximates the Earths surface without the topography,
i.e., it does not take into account irregularities such as mountains and ocean trenches that the geoid
attempts to model. Two frequently used ellipsoid datums, the North American Datum of 1983
(NAD 83) and the World Geodetic System of 1984 (WGS 84) differ by as much as two meters.
GEOID03, a geoid model developed by the National Geodetic Survey (NGS) in NOS in NOAA,
specifically relates NAD 83 ellipsoid heights to NAVD 88 orthometric heights. VDatum, a vertical
datum transformation tool developed in NOS (Parker et al, 2003), allows the transformation of
elevation data between any two vertical datums among a choice of 30 vertical datums, including
tidal datums, orthometric datums, and 3D ellipsoid datums. Use of tide prediction models in
producing VDatum will be described in Section 8.7.2.
In this section only some of the most basic terms have been covered. Others will be introduced
throughout the book. In Section 2.2.5 some of the definitions given above will revisited from the
perspective of their relationship to harmonic tidal constituents.
2.1.2 The Generation of Tides
The tides are caused by both the moon and the sun. Although the sun is about 27 million times
more massive than the moon, its effect is smaller than that of the moon since the sun is 93 million
miles away. This is because the influence of a celestial body (the moon or the sun) on the tides of
the Earth is directly proportional to the mass of that celestial body but inversely proportional to the
cube of its distance from the Earth (for derivation of the formula with this relation, see Dean, 1966;
Defant, 1961, Pugh, 1987, Godin, 1988, or Forrester, 1988).
Although the moon appears to orbit around the Earth, the Earth and moon both actually revolve
around a common point, which, because the Earth is 82 times more massive than the moon, is inside
the Earth, but not at the Earths center (see Figure 2.8). At the center of the Earth there is a balance
between gravitational attraction (trying to pull the Earth and moon together) and centrifugal force
(trying to push the Earth and moon apart as they revolve around that common point). At a location
on the Earths surface closest to the moon, the gravitational attraction of the moon is greater than
the centrifugal force of the Earth (moving around the center of the revolving Earth-moon system).
On the opposite side of the Earth, facing away from the moon, the centrifugal force is greater than
the moons gravitational attraction. In a hypothetical ocean covering the whole Earth with no
continents (see Figure 2.8) there will be two tidal bulges resulting from these imbalances of
gravitational and centrifugal forces, one facing the moon (where the gravitational force is greater
24
Figure 2.8. The Earth-Moon system (viewed from very high above the North Pole)
revolving around a common point. A hypothetical ocean is shown covering the entire
Earth (with no continents) with two tidal bulges resulting from the imbalances of
gravitational and centrifugal forces.
than the centrifugal force) and one facing away from the moon (where the centrifugal force is greater
than the gravitational force) .
However, the tide-producing forces that result from the difference between the moons
gravitational attraction and the centrifugal force of the Earths revolution around the center of the
Earth-moon system are actually quite small much smaller than the Earths own gravitational force.
Thus, at the equator on the side of the Earth facing the moon, the tide producing force (which at the
equator is vertically upward from the Earth toward the moon) is so small compared to the Earths
gravitational force that it could not cause the tidal bulge. Further north or south of the equator,
Figure 2.9. The tide generating forces (the thick black arrows) on the
Earth resulting from the difference between gravitation attraction (the
open arrows) and centrifugal force (the hatched arrows). The small thin
arrows are the horizontal components of the tide generating forces, which
tend to move the water into the two bulges.
25
however, at another point on the Earth that is not directly under the moon, the small tide producing
force is still pointing toward the moon, but is no longer perfectly vertical relative to the Earth (see
Figure 2.9). At this point, the force toward the moon can be separated in a vertical component and
a horizontal component, the latter one being tangential to the Earths surface. This horizontal
component of the tide-producing force, though small, has nothing opposing it, and so it can move
the water in the ocean. One can see from Figure 2.9 that all the horizontal components tend to move
the water toward the equator, causing a bulge centered around the point that is directly under the
moon. Similarly, on the other side of the Earth another bulge results.
One can envision the Earth rotating under these two bulges in this hypothetical ocean that covers
the entire Earth. In one complete rotation in one day there will be two high tides (when under a
bulge) and two low tides (when halfway between bulges), and thus one tidal cycle would be
completed in approximately half a day (actually 12.42 hours, for reasons to be explained below).
However, this is an extreme simplification (called the equilibrium tide) used merely to show how
the tide generating forces change as the Earth rotates. Not only are the continents left out (which
affect the tides via hydrodynamics), but the equilibrium tide assumes that the oceans respond
instantly to the tide-generating forces, which they do not. However, as will be seen later (e.g., in
Section 4.2.1), the idea of a simplified equilibrium tide has sometimes been made use of in tidal
analysis and prediction.
Now continents are added and only one of the oceans is looked at with a bay connected to it (see
Figure 2.10). The tide-generating forces are too small to cause a tide directly in a small body of
water like a bay. Only in a large ocean are the cumulative effects of the tide-generating forces
throughout the ocean large enough to produce a tide. What is actually generated is a very long wave
with a small amplitude (i.e., a small tide range, the range equaling twice the amplitude), on the order
of a foot or less. However, when this long wave reaches the reduced depths of the continental shelf,
there is a partial reflection of the wave, and the part of the wave that continues toward the coast is
Figure 2.10. The tide generating forces cause by the moon and the sun
produce a very long wave of relatively small amplitude in the ocean. When
this long wave reaches the continental shelf, then the coast, and finally
propagates up a bay, it is amplified by an amount that depends on the depth
and length of the basins.
26
increased in amplitude. At the coast another reflection further increases the amplitude of this long
wave, now reaching three feet or more along most coasts. When the wave moves up into a bay there
can be even more amplification depending on the depth, length, and width of the bay, with tidal
ranges reaching 15, 30, or even 50 feet for bays with the right dimensions. The propagation of this
tide wave is also influenced by the Coriolis force (due to the Earths rotation), which affects the tide
ranges. This will be discussed in Sections 2.3.1 and 7.5.
How large the tide range is depends on how close the natural period of free oscillation of the
basin is to the period of the tide-generating force. Much like a pendulum, which swings back and
forth with a specific period determined by its length, the water in a basin sloshes back and forth with
a specific period determined by its depth and length, which is called its natural period of oscillation.
If the natural period of the basin is the same as the period of the tide-generating force, then the
energy from the tidal forcing will be input in the same direction as the water is already moving and
the resulting tide range will be larger. This is called resonance.
The natural period of oscillation of a basin, depends on its length and its depth. The longer the
basin, the larger the period of oscillation, but the deeper the basin the smaller the period of
oscillation. The length has the greater effect. More precisely, the natural period of the basin, Tn,
(for the special case of no friction) is given by the formula
Tn = 2L
(gD)
where L is the length of the basin, D is the depth, and g is the acceleration due to gravity. (This
result ignores friction, which can have a very significant effect in shallow water, as will be seen in
Section 7.4.) The Atlantic Ocean is too wide for there to be resonance (its 19-hour natural period
being much longer than the 12.42-hour tidal period). The largest tide ranges in the world are in
shallower basins with just the right length and depth combination to have natural periods close to
the tidal period. Narrowing widths in rivers and bays also amplify the tide range due to a continuity
effect. These hydrodynamic effects will be discussed in more detail in Chapter 7.
While the hydrodynamics, as determined by the dimensions of the ocean, bays, and rivers,
determines the tide ranges and times of high and low waters, it is the astronomy that determines the
particular frequencies at which tidal energy will be found. Because the system is a forced
hydrodynamic system, the energy remains in those astronomically determined frequencies, and it
is because of this that tides are so predictable using the methods to be explained in this book. In the
next section these particular frequencies will be identified and their origins explained.
2.2 Astronomical Considerations
2.2.1 The Orbits of the Moon Around the Earth and the Earth Around the Sun
As mentioned above, because the ocean and connected bays are a forced oscillating system, the
tide will oscillate with the same frequencies as the astronomical tide-producing forces, which are
determined by the relative motions of the Earth, moon, and sun. There are many different tidal
frequencies because of the complex nature of the orbit of the moon around the Earth and of the orbit
of the Earth around the sun. Astronomers have very precisely determined all of the required
astronomical frequencies. The fact that tidal energy will always be at known frequencies allows one
27
to predict the tide at a specific location for any time in the future, as long as there are data to analyze
to determine the amplitude and epoch (phase lag) for each important tidal constituent (the amplitude
and epoch not being known ahead of time because they are determined by the hydrodynamics).
If the moon-Earth orbit and the Earth-sun orbit were both circular and were both in the plane of
the Earth's equator, there would only be two tidal frequencies. One could then predict the tide using
only two semidiurnal tidal harmonic constituents (M2 and S2, which are defined in Section 2.2.2).
However, the orbital motions are much more complicated. Both orbits are elliptical, so the distance
between the moon and Earth changes throughout the month, and the distance between the Earth and
sun changes throughout the year. Both orbital planes are also at angles relative to the Earth's
equatorial plane. Because of the angle between the moons orbit and the plane of the Earths
equator, the moon appears to an observer on Earth to move north of the equator and then south of
the equator and back north of the equator over roughly a month (actually 27.3 days). Similarly, the
sun appears north of the equator half of the year (summer in the Northern Hemisphere) and south
of the equator the other half of the year (winter in the Northern Hemisphere). To further complicate
matters, the angles between the orbital planes and the equator also slowly change periodically over
the years. All these motions modulate (that is, periodically vary the strength of) the tidal forces,
so that tidal energy is spread out among many more frequencies (in addition to M2 and S2).
The fact that the moons monthly orbit around the Earth is angled to the Earths equatorial plane
means that the moon will be over the equator only twice a month (equatorial declination). The rest
of the month the moon will appear either north or south of the equator and the two tidal bulges (on
the opposite sides of the Earth) will be asymmetric with respect to the axis of rotation (see Figure
2.11; here as in Figure 2.8, the idealized equilibrium tide is being considered, i.e., an ocean covering
the whole earth with no continents). As a result there will be a once per day inequality in the
elevation of the two high waters (and in the elevation of the two low waters) formed by the tidal
bulge at any given latitude. This asymmetric declination effect puts energy into diurnal tidal
frequencies. The strength of the resulting diurnal tide will vary in strength from zero when the moon
Figure 2.11. When the moon is at maximum declination north or south

of the equator, the tidal bulges also shift north or south, and a location
on the Earth will rotate under two different height bulges. The result
will be a diurnal tidal component. In the real world with continents,
hydrodynamics will actually determine which waterways have stronger
diurnal signals.
28
is over the equator, to a maximum value when the moon appears (to an observer on the Earth)
farthest north of the equator (northern declination) or farthest south of it (southern declination).
The fact that the Earths yearly orbit around the sun is at an angle to the plane of the Earths
equator means that the sun will be over the equator only twice a year, March 21st (vernal equinox)
and September 21st (autumnal equinox). The rest of the year the sun appears north of the equator
half of the year (summer in the Northern Hemisphere) and south of the equator the other half of the
year (winter in the Northern Hemisphere). The sun is farthest north of the equator on June 21st
(summer solstice) and farthest south of the equator December 21st (winter solstice). This also
produces diurnal tidal frequencies because the two tidal bulges caused by the sun (on the opposite
sides of the Earth) will be again asymmetric with respect to the axis of rotation (Figure 2.11 also
applies, with the moon replaced by the sun).
The fundamental astronomical periods of the motions of the Earth, moon, and sun are shown in
Table 2.1. This includes three types of day the solar day (24.000 hours), the lunar day (24.8412
hours), and the sidereal day (23.9344 hours). These differ from each other with respect
to the reference for measuring one complete rotation of the Earth. The solar day is one complete
rotation with respect to the sun. The lunar day is one complete rotation with respect to the moon,
and is longer than the solar day because the moon revolves around the Earth in the same direction
as the Earth rotates, so the Earth must rotate a little farther (and longer) to catch up with the
revolving moon. The sidereal day is one complete rotation of the Earth with respect to the vernal
equinox, which in this context means the location of the intersection of the plane of the Earths
equator and the plane of the Earths orbit around the sun (called the ecliptic). The angle between
these two planes is 23o. The sidereal day differs only slightly from a day referenced to a fixed star
in space, and the vernal equinox is usually used as the primary reference point for astronomical
discussions.
Table 2.1 also includes the tropical month (27.3216 days), which is the period of the lunar
declination. It is the time it takes the moon to revolve around the Earth with respect to the vernal
Table 2.1. The fundamental astronomical periods of the motions of the earth, moon,
and sun (reworked from Platzman, 1971). The frequency notation is referred to in Table
2.2 and in the text. (wrt = with respect to)
29
Figure 2.12. The orbits of the moon around the Earth and of the Earth around the sun, as
viewed from a distant star. See text for discussion. (Modified from Dean, 1966)
equinox. In other words, it is the time it takes the moon to go from maximum northern declination
to the next maximum northern declination. There are also other types of months, the only difference
being the reference point against which the revolution of the moon is measured. If it is measured
with respect to the sun it is called the synodic month (29.5301 days, on average), which is the month
commonly observed when one notices the changing phases of the moon. This is longer than the
tropical month, because, due to the Earths orbit around the sun, it takes a little more time for the
moon to get back to the same position with respect to the sun.
The Earth-moon orbit is elliptical (Figure 2.12), so that the distance between them varies from
perigee (the moon closest to the Earth) to apogee (the moon farthest from the Earth). The period
from perigee to apogee to the next perigee is on average 27.5546 days. The elliptical shape of the
moons orbit also slowly changes, causing the position of perigee to slowly move over an 8.847-year
period. The angle between the plane through the moon's orbit and the plane through the equator
varies over a 18.613-year period. This is referred to as lunar nodal regression because the
intersection of the moon's orbital plane with the ecliptic, called the ascending lunar node, regresses
backwards along the ecliptic over this 18.6-year period (see Figure 2.13).
Figure 2.13. The moons orbit and the suns orbit (the
ecliptic) with respect to the Earths equatorial plane. See text
for explanation of terms. (From Schureman, 1958)
30
The tropical year (365.2422 days) is the length of time for the Earth to revolve around the sun
with respect to the vernal equinox. The orbit of the Earth around the sun is also elliptical, so that
the distance between the Earth and the sun varies from perihelion (the Earth closest to the sun; also
called the solar perigee) to aphelion (the Earth farthest from the sun; also called the solar apogee).
However, this orbit also very slowly changes, so that the position of perihelion very slowly moves
over a period of 20,940 years (too long to worry about in tide predictions, but important in long-term
climate studies).
A simplified schematic of the relative motions of the Earth, moon, and sun, from the view point
of a distant star is shown in Figure 2.12. In this figure one sees the plane of the moons orbit around
the Earth inclined at an angle of 5o9' relative to a plane through the Earths equator. The plane of
the Earths orbit around the sun (the ecliptic) is inclined to the plane of the Earths equator at an
angle of 23o27'. These same two orbital planes and the Earths equatorial plane are also shown in
Figure 2.13, but from the point of view of the Earth. Since one is used to seeing the moon and sun
overhead moving across the sky and around the Earth, using Figure 2.13 it is usually easier to
visualize the effects of the orbits on the tidal bulges. In Figure 2.13, is the angle between the
plane of the Earths orbit (the ecliptic) and the plane of the Earths equator and equals 23o27'; it is
called the obliquity of the ecliptic. In the same figure, i is the angle between plane of the Earths
orbit and the plane of the moons orbit, and it is 5o9'. I is the angle between the plane of the moons
orbit and the plane of the Earths equator, and it changes with the movement of the ascending lunar
node, which moves through one cycle in 18.6 years. Thus this angle goes from approximately 28o
to 18o and back in 18.6 years and has an important effect on the tide range.
The periods for the (three types of) day, month, year, lunar perigee, lunar node, and perihelion
are listed in Table 2.1. L, and 1 through 5 in Table 2.1 are the notation for the equivalent
frequencies for these periods. As will be seen in Section 2.2.3, combinations of these frequencies
will produce the frequencies of all the tidal constituents (Table 2.2). Before examining Table 2.2
it will be instructive to see how the frequencies of some of the most important tidal constituents are
derived.
2.2.2 The Origins of the Largest Tidal Harmonic Constituents
As was mentioned in Chapter 1, the contribution to the tide by the energy at a particular
frequency is usually represented by a tidal harmonic constituent, for which there is an amplitude and
a phase lag (called an epoch). Harmonic tidal prediction involves the summing of a set of cosine
curves representing the various tidal harmonic constituents (see Section 3.4.2). The nomenclature
for most tidal harmonic constituents is to name each constituent with a capital letter followed by a
subscript indicating the approximate number of cycles per day for that constituent, for example M2,
S2, N2, K1, and O1 (all of these constituents are discussed in this section). There are also a few tidal
constituents named with Greek letters (e.g., 2, 2, 2, and 1), as well as compound tides, whose
names have more than one capital letter (e.g., 2MN2, 2MS2, MK3, MN4, etc.) indicating the primary
constituents from which the compound constituent was nonlinearly generated by shallow water.
There are also long-period tides which tend to have names with a capital letter and one or more
lower case letter (e.g., Mn, Mf, Sa, Ssa, etc.). At the end of Section 2.2.3 there is some discussion
of where this naming convention came from.
This section will explain the origin of some of the most important tidal harmonic constituents,
namely, why there is tidal energy at a particular frequency and the particular aspect of the Moons
orbit or the Earths orbit that is responsible.
31
The moon orbits around the Earth in the same approximate direction as the rotation of the Earth,
so that one lunar day (i.e. one complete rotation of the Earth with respect to the moon) is longer than
the 24-hour solar day, because, by the time the Earth has rotated the moon has revolved a small
distance. So a lunar day is 24.8412 hours long (1/L). There are two tidal high water bulges on the
Earth, so the period of the largest semidiurnal lunar harmonic constituent, M2, is half a lunar day,
or 12.4206 hours. Its frequency (1/tidal period) is 2L, which is (1/12.4206 hours=) 1.9323 cycles
per (solar) day.
The Earth turns under the sun exactly once every solar day, which leads to the main solar
semidiurnal tidal constituent, S2 , with a period of 12.0000 hours. Its frequency is 2s, which is
2.0000 cycles per day. Although the sun is approximately 27 million times more massive than the
moon, it is approximately 339 times farther from Earth than the moon, and since the tidal force is
inversely proportional to the cube of the distance, S2 is usually much smaller in size than M2. When
the moon and sun are in alignment at new and full moons their tidal forces work together to produce
larger tide ranges (called spring tides). When the moon and sun are at first and third quarters they
work against each other to produce smaller ranges (called neap tides). In Figure 2.14 in the upper
plot, two sine curves, one representing the M2 tide and one representing the S2 tide, are plotted
together. In that plot they begin in sync, with their tide-producing forces adding to each other, thus
causing the larger spring tides seen in the lower plot (where the M2 and S2 contributions are
summed). After roughly 7 days the two sine curves in the upper plot are now out of sync and their
tide-producing forces are now subtracting from each other, thus causing the smaller neap tides seen
in the lower plot. After another 7 days they are back in sync and producing spring tides once again.
This sequence could represent going from full moon to third quarter to new moon, or likewise, going
from new moon to first quarter to full moon. The cartoon representation of the relative positions of
the Earth, moon, and sun for these two alignments of the moon and sun are shown in the upper most
part of Figure 2.14.
The Earth-moon orbit is elliptical, so that the distance between them varies over a 27.5546-day
period (1/[1-3]), from perigee (the moon closest to the Earth, and so a stronger tidal force) to
apogee (the moon farthest from the Earth, and so a weaker tidal force) and back to perigee. This
periodic change in the distance from the moon to the Earth modulates the lunar tidal force, that is,
it slowly increases and then decreases the tidal force over the 27.5546-day period. Such a
modulation of M2 can be represented by combining it with another semidiurnal lunar tidal
constituent with a period of 12.6583 hours, which is called N2. [The effect of the 27-day variation
in the moons distance to the Earth is often referred to as the parallax effect. In this context parallax
is the angle between two lines from the center of the moon, one to the center of the Earth and one
to the surface of the Earth (tangent to the Earths surface). Because this angle is very small it is
approximately equal to the sine of the angle, which is equal to the ratio of the Earths radius to the
distance from the moon to the Earth. Since the parallax is a function of the distance to the moon,
the term is applied to tidal inequalities arising from this changing distance.]
The plots in Figure 2.14 can be used to illustrate the perigee-apogee effect, if, in that figure, S2
is replaced with N2, spring tide is replaced with perigean tide, and neap tide is replaced with
apogean tide. However, in using this figure it must be remembered that these are two very different
situations. The difference is that with the M2 plus S2 case, there really are two distinct effects being
added, the lunar tide and the solar tide. However, in the case of the changing distance between the
moon and Earth, that changing distance directly affects the amplitude of the lunar tide, and this
modulating amplitude of the lunar tide is being represented by adding an N2 tidal constituent to the
M2 tidal constituent. The stronger perigean tidal force is represented when M2 and N2 come into
32
Figure 2.14. The combined effect of the moon and sun varies
throughout the month. When the moon and sun are working with each
other (at Full Moon and New Moon) there are larger tide ranges (spring
tides). At First Quarter and at Last Quarter the moon and the sun work
against each other resulting in smaller tide ranges (neap tides).
phase (leading to larger tidal ranges), while the weaker apogean tidal force occurs when M2 and N2
are exactly out of phase (leading to smaller tidal ranges). However, that being said, if one carries
out a spectral analysis of a water level data series (that is, if one calculates how energy is spread out
over the entire frequency domain), one will find that to the left of the M2 spectral line (1.9323 cpd),
there is another spectral line at 1.8960 cpd, which is the frequency of another tidal constituent, N2
(see next section). Thus, this is really more than just finding a convenient tidal constituent with a
frequency, that when added to the M2 constituent, will produce the correct modulation. The perigee-
apogee modulation of M2 actually transfers energy from M2 to another frequency, N2.
There are several times a year when lunar perigee is reasonably close in time to new or full
moon, the result being much larger tidal ranges called perigean spring tides. Similarly, much
smaller tidal ranges (called apogean neap tides) result when lunar apogee occurs near first or third
33
quarter phases of the moon. With respect to the tide-producing forces S2 should be larger than N2
and for many waterways it is, but for other waterways N2 is larger than S2, this being due to the
effect of the hydrodynamics in such waterways that amplifies N2 more than S2. They are both, or
course, much smaller than M2 . (But even here there is an exception, at Tahiti, where the
hydrodynamics of the Pacific Ocean causes there to be a minimum in M2 near Tahiti, allowing S2
to be of comparable size, and therefore leading to the very unusual situation of high water occurring
at approximately the same time every day. See Marmer, 1926.)
As mentioned in the last section, the plane of the moon's orbit around the Earth is at an angle
to the plane through the Earth's equator. As the Earth rotates under the moon, there will be times
of the month when the moon is north of the equator (Northern Declination), over the equator
(Equatorial Declination), and south of the equator (Southern Declination). When the moon is north
or south of the equator, one of the tidal bulges is more north of the equator and one is more south,
so that at a particular location on the Earth there will either be only one high water per day (a diurnal
tide), or, if there are two, they will be of different heights, the difference being the diurnal inequality
(for example, see the San Francisco tide curve in Figure 2.12). The diurnal lunar tidal forces
resulting from lunar declination are represented by two diurnal tidal constituents, O1 and K1, with
periods of 25.8193 and 23.9345 hours (and frequencies of 0.9295 cpd and 1.0027 cpd). The
minimum combined effect of these two constituents occurs every 13.66 days (1/21), at the times
when the moon is over the equator. Their maximum combined effect occurs at maximum
declination. The sum of the O1 and K1 frequencies is equal to the M2 frequency, so that the time
of the diurnal high water does not change with respect to the times of the two semidiurnal high
waters. (The maximum angle between the plane of the moon's orbit and the Earth's equator varies
from 18.3o to 28.5o over a 18.6-year period, the lunar nodal cycle; see Section 2.2.4 for further
discussion.)
As mentioned in the last section, the plane of the Earth's orbit around the sun (the ecliptic) is also
at an angle to the plane through the Earth's equator. Around December 21st the sun is furthest south
of the equator (December solstice) and around June 21st it is furthest north of the equator (June
solstice), the angle between the ecliptic and the equator reaching 23.5o in each case. December
solstice marks the beginning of winter in the Northern Hemisphere and the beginning of summer in
the Southern Hemisphere, and vice versa for June solstice. Around March 21st the sun is over the
equator (vernal equinox) and again around September 21st (autumnal equinox). This movement of
the sun north and south of the equator also leads to diurnal tidal constituents, which represent this
very slow modulation, in this case P1 with a period of 24.0658 hours [1/(S-2)]), and another K1.
Thus, the K1 used for tidal prediction has both lunar and solar parts. P1 and the solar part of K1 have
a minimum combined effect every 182 days, at vernal and autumnal equinoxes, and have their
maximum combined effects at winter and summer solstices.
Although with respect to the tide producing forces, K1 and O1 should always be smaller than M2,
there are in fact many waterways where hydrodynamics amplifies the diurnal constituents and/or
reduces the semidiurnal constituents, so that K1+O1 is comparable to or larger than M2. In this case
the result is a diurnal tide, that is, one high water and one low water per day (see Sections 2.3.1 and
7.4.1).
34
2.2.3. Derivation of Tidal Frequencies from the Basic Astronomical Frequencies

Insights from Tidal Spectra
In the last section, with the exception of M2 and S2, each of the other tidal constituent frequencies
looked at represented a modulation of the lunar or solar tide due to some orbital effect. That
modulating effect could be represented by using additional tidal constituents in combination with
M2 or S2. For example, it was seen that the variation in tidal forces due to the elliptical shape of the
moons orbit (i.e., going from apogee to perigee) could be represented by adding N2 to M2. It was
also seen that the diurnal variation in tidal forces due to the changing declination of the moon (i.e.,
the moon being north of and then south of the equator) could be represented by two tidal
constituents, K1 and O1, in combination with M2.
It was also seen that the use of tidal constituents like N2, K1, and O1 is more than just a
convenience. If one carries out a spectral analysis of water level data or current data (which
calculates the energy at all frequencies), one will find spectral lines clearly above the nontidal
continuum (see Section 3.1) at the exact frequencies of N2, K1, and O1. The modulation of the M2
tidal oscillation by the variation in the moon-Earth distance (from apogee to perigee and back) has
indeed transferred energy from the M2 frequency to N2 frequency. Likewise, the modulation of the
M2 tide by the changing declination of the moon has also transferred energy to the K1 and O1
frequencies. In a water level spectrum one will also find many more spectral lines, which are
additional smaller tidal constituents also produced by some variation in the moon-Earth orbit or the
Earths orbit around the sun. Now it will be explained where those constituents come from (and
Tables 2.1 and 2.2 will be more closely looked at).
First consider a typical spectral analysis of a water level times series (for example that shown
in Figure 2.16). The longer the data time series the smaller the interval one can have as one moves
along the frequency domain from zero frequency (i.e., the mean energy over the time series length)
to the highest frequency for which energy can be calculated. The resolution frequency is simply the
reciprocal of the length of the time series. The highest frequency for which energy can be calculated
is determined by the sampling rate, t , and is known as the Nyquist frequency, defined as fN=1/(2
t). For example, for a time series with an hourly sampling rate (t=1 hour) the highest frequency
in the calculated spectrum will be one cycle in two hours, i.e., fN = 12 cycles per day (cpd). So for
looking at the tidal spectral lines at most stations, even those in very shallow water, an hour
sampling rate is usually more than adequate. However, there are a few shallow-water areas with
large tidal ranges where higher harmonics beyond 12 cpd can actually be detected. In such cases
a higher sampling rate is required. Water level gauges presently operated by CO-OPS usually have
a sampling rate of 6 minutes (which would allow one to do spectral analysis out to 5 cph or 120
cpd).
The Nyquist frequency is also called the folding frequency, because the energy at frequencies
above the Nyquist frequency affect the calculated energy at frequencies below the Nyquist frequency
they fold back in. This is know as aliasing or folding. For example, Figure 2.15 shows a sinusoid
with a frequency (f) higher than the Nyquist freqency (fN). In such a case the sampled data points
take energy from different phases of the high-frequency sinusoid, and the result looks like a much
lower-frequency sinusoid (i.e., with a sinusoid with a frequency, fN f, which is lower than the
Nyquist frequency). So if the sampling interval is too large, not only will the higher tidal
frequencies not be included in the spectrum, but that higher-frequency energy will actually be folded
back in (through the aliasing) and affect the values of the spectrum that can be calculated.
(However, one can sometimes use known aliasing to help reduce computations. See Section 3.5.2.)
35
Figure 2.15. A demonstration of aliasing. The higher-frequency sine wave

has a frequency greater than the Nyquist frequency (namely, the sampling
interval is too large), so the data points (the small boxes) appear to describe a
much lower-frequency sine wave (see text).
A spectral analysis of an hourly time series that is a year long, will show us how much energy
is at frequencies from 0 cpd to 12 cpd, at frequency intervals of approximately 0.00274 cpd. Two
tidal constituents whose frequencies are closer to each other than 0.0027 cpd will, for this case, not
be separated and will be seen in one spectral line. (Actually, in Section 3.3 it will be seen that this
rule, the so-called Rayleigh criterion, may actually be affected by the signal-to-noise ratio of the
data time series, and for cases with high signal-to-noise may be too restrictive.)
Figure 2.16 shows the tidal bands from a typical spectral analysis of water level data, in this case
from Anchorage, Alaska. The tidal spectral lines are grouped in bands, one band for each tidal
species semidiurnal and diurnal, as well as the higher harmonics (the overtides). The long-period
tides tend to be overwhelmed and hidden by low-frequency nontidal energy. The surface of the
ocean moves up and down due to other nontidal phenomena which do not have most of their energy
concentrated in a few frequencies. Thus, in the plot of a water level spectrum, the energy from
winds, atmospheric pressure, and temperature and salinity changes shows up primarily as a smooth
l
gy(
Ener e)
ogscal
Figure 2.16. A water level spectrum from near Anchorage,

Alaska, at the northern end of Cook Inlet, where the water is
shallow and the tide range large.
36
continuous curve covering a broad range of frequencies, called the continuum, ranging from wind
waves at the higher frequencies to low-frequency storm surges at the lowest frequencies. The tidal
bands (or on a fine scale the tidal lines) rise above this continuum. The continuum curve at a water
level station near the coast or in a bay is often highest near zero cpd (which represents the mean of
the data time series), with the very low frequencies near zero cpd representing very slow changes
in the height of the water surface due to the effects of wind, atmospheric pressure, density changes,
and/or river discharge.
As Munk and Cartwright (1966) pointed out, in the language of spectroscopy (i.e., the study of
spectra) each tidal species shows three orders of splitting: monthly splitting, a finer yearly
splitting, and hyperfine splitting from long-period effects such as the variation in lunar perigee and
nodal regression. This splitting is another way of saying that modulation of the lunar and solar
tide producing forces by monthly, yearly, and longer-period variations adds additional spectral lines
to the tidal spectra, each line of which can be represented by a tidal harmonic constituent. In the
previous section, in describing how some of the more important tidal constituents are produced, it
was not mentioned that each modulation actually produces two additional tidal constituents (the
second usually much smaller) and thus the term splitting. Thus, around each tidal spectral line,
a monthly variation will produce two additional spectral lines, one on each side, often referred to
as a pair of side-band lines. The longer the period of the modulating effect the closer the lines are
to the central line.
Thus, for example, the monthly modulation of M2 due to the changing distance between the
moon and the Earth (the perigee-apogee effect of the Moons elliptical orbit), produces two
additional spectral lines around the M2 spectral line, the N2 and L2 spectral lines. M2 has a frequency
of 1/(12.4206 hr) = 1.9323 cpd, and 0.0363 cpd to the left and right of this spectral line one will find
N2 [1/(12.658 hr) = 1.8960 cpd] and L2 [1/12.192 hr) = 1.9686 cpd], respectively.
Similarly, the yearly modulation of S2 due to the effects of the elliptical orbit of the Earth around
the sun (the perihelion-aphelion effect) produces two additional tidal constituents around S2, both
0.0027 cpd from S2; the frequencies of T2, S2, and R2 are respectively, 1.9973, 2.0000, and 2.0027
cpd (see Figure 2.17).
Likewise, the 18.6-year modulation of the lunar tidal forces due to the lunar nodal regression can
be represented by many additional tidal constituents, usually referred to as satellite constituents.
If one has 19 years of data, one can include these additional satellite constituents and not need to use
node factors (which will be discussed more in Section 2.2.4).
The modulating forces that produce additional semidiurnal and diurnal tidal constituents also
directly produce other smaller tidal constituents that have the same frequencies as the modulation.
For example, there is a tidal constituent Mm, the lunar monthly constituent, which has a period of
27.55 days (a frequency of 0.036 cpd) and is directly produced by variation in tidal force due to the
moons elliptical orbit. There is a tidal constituent Mf, the lunar fortnightly constituent, which has
a period of 13.66 days (a frequency of 0.073 cpd) and is directly produced by the variation in tidal
force due to the moons declination. There is a tidal constituent MSf, the lunar synodic fortnightly
constituent, which has a period of 13.77 days (a frequency of 0.068 cpd) and is also directly
produced by the variation in tidal force due to the moons declination. All of these long-term tidal
constituents are very small, and are often dominated by meteorological effects on water level and
currents that can put energy into these same frequencies. More important, however, nonlinear
shallow-water effects produce compound tides with the same frequencies as these fortnightly
andmonthly constituents, but with larger amplitudes. (See Section 2.3.2 and Table 2.4.) For
37
Figure 2.17. Tidal spectral lines for the semidiurnal, diurnal, and long-period constituents (top
panel) and with greater resolution for the diurnal constituents (middle panel) and for the
semidiurnal constituents (lower panel).
example, the nonlinear interaction of M2 and S2 produces the shallow-water constituent MS with the
same frequency as MSf.
Also dominated by meteorological forces are the even longer-period and even smaller tidal
constituents directly produced by yearly variations in the orbits. Sa and Ssa are very small tidal
constituents directly generated by the nonuniform changes in the Sun's declination and distance (the
perihelion-aphelion effect). The solar annual constituent, Sa, has a period of 365 days (a frequency
of 0.0027 cpd) and the solar semiannual constituent, Ssa, has a period of 183 days (a frequency of
0.0055 cpd). These constituents and their dominance in data time series by meteorological effects
will be discussed more in Section 3.7.
Although these and other additional tidal constituents represent the tidal energy at new
frequencies due to the modulating effects of the moons and Earths orbits, early tide researchers
found it convenient to think of each as being produced by a separate and distinct heavenly body, like
the moon and sun. This idea was first proposed by Laplace, who called it astres fictifs, or fictional
38
celestial bodies, and the idea was then used by Thomson (Lord Kelvin). These celestial bodies
could be thought of as each moving in the Earths equatorial plane with a constant angular speed
(which may be one reason why the term angular speed was used in tides to mean frequency).
The more modern name for these fictitious heavenly bodies is satellites, and as was just mentioned,
this term has come into use again when carrying out 19-year or long harmonic analyses and using
satellite constituents instead of node factors.
The naming of the actual tidal constituents themselves (at the least the larger ones studied by the
earliest researchers) appear to primarily have come from Thomson and Darwin and used the letters
K, L, M, N, O, P, Q, R, S, and T. The logic for M for moon and S for sun is obvious. The letters
L and N for the two semidiurnal constituents with periods less than and greater than M, probably
were used so that alphabetical order matched the order of increasing tidal constituent period, and
similarly for R and T around S, but the reason for choosing the letters K, O, and Q seems less
obvious. In the U.S., Ferrel developed his own harmonic development of the tide completely
independent of Thomson and Darwin (but following on the work of Laplace) and had a totally
different naming scheme which involved only numerical suffixes. However, the British naming
system found favor and by the time Schureman first published his work on harmonic analysis and
prediction in 1924 he was also using the British naming system.
When harmonically analyzing the forces involved in the moons and Earths orbits, Doodson
(1921) came up with over 400 tidal constituents (admittedly most were very small). This remained
the standard reference work for tidal analysis and prediction for fifty years, until it was finally
recalculated and updated by Cartwright and Tayler (1971) and Cartwright and Edden (1973), who
used an ephemeris of high precision for the moon and sun and the latest I.A.U. astronomical
constants.
In Doodsons and others work, the harmonic development was actually done on the tide-
generating potential. The tide potential is related to the tidal force by the formula F = LV , where
L stands for the gradient operator, which turns the scalar V into a vector F. It is mainly a
mathematical convenience, and one can think of the tide-generating potential as merely representing
the tide-producing force (see Godin, 1972).
Figure 2.17 shows some of the tidal spectral lines. For each tidal frequency there is a tidal
constituent for which the amplitudes and epochs (phase lags) are calculated during a harmonic
analysis of a data time series. Those amplitudes and epochs are then used to make tidal predictions.
Table 2.2 shows how key tidal constituents are derived from these fundamental astronomical
frequencies.
The classic Doodson number used in many tidal papers is a shorthand notation that indicates
which of the six frequencies, L, and 1 through 5 from Table 2.1, are used to produce a particular
tidal constituent (e.g., the examples shown in Table 2.2). Doodson added 5 to each number (except
the first) to keep them from being negative. Cartwright numbers are the same as the Doodson
numbers but without the added 5's. The six digits of a Cartwright number are the multiplying
coefficients in front of the six frequencies L, and 1 through 5 from Table 2.1. Cartwright
numbers for each astronomical tidal constituent are given in Tables 3.2, A.1, and A.2.
2.2.4. The Origin of Node Factors and Equilibrium Arguments
As mentioned earlier, there are some very slow variations in the tide-producing forces with
periods on the order of many years. Most important of these is the 18.6-year variation in the angle
between the plane through the moon's orbit and the plane through the Earths equator (4 in Table 2.1).
39
Table 2.2. Tidal constituents and their origin from astronomical frequencies. The speed
is the angular speed, a classical form of frequency (see text). Mo and So represent constant
lunar and solar forces. The coefficient C gives a global measure of each constituents
relative portion of the tide potential (reworked from Platzman, 1971).
This is referred to as the lunar nodal regression because the intersection of the moon's orbital plane
with the ecliptic (the Earths orbital plane around the sun), called the ascending lunar node, regresses
backwards along the ecliptic over this 18.6-year period. Also significant is the 8.85-year variation
in lunar distance due to the rotation of the longitude of the lunar perigee (3 in Table 2.1). These
longer-term variations in the tide (and especially the 18.6-year variation) are clearly evident in water
level records. The 100-year record of monthly and annual mean tide ranges at Seattle, Washington,
in Figure 2.18, provides one example. In this figure one can see five complete 18.6-year cycles in
the plot of the annual mean tide range. These cycles are also observable in the plot of the monthly
mean tide ranges, although that graph shows much greater variability, probably due to the nonlinear
interaction effects of various nontidal influences on water level.
40
These longer-term modulating effects can also be represented as tidal harmonic constituents, but
to do so one must use one or more additional satellite constituents for every lunar tidal constituent
already considered. The frequencies of these additional constituents will be extremely close in
frequency to those original constituents. (As was seen in Section 2.2.1, the spectral splitting due to
these very long-period effects puts small spectral lines very close to each tidal spectral line.)
Because these additional constituents are so close in frequency to the original constituents, one must
analyze a very long data time series, namely 18.6 years of data. When this is done, these additional
constituents are usually referred to as satellite constituents (see Zetler, et al, 1985; Amin, 1976; and
Foreman and Neufeld, 1991).
In the pre-computer era, analyzing 19 years of data was much too labor intensive, especially
since these additional constituents accounted for too little variation to be worth the large computing
effort. But oceanographers did not want to ignore their effects. The 18.6-year variation due to the
lunar nodal regression cycle can affect M2 by as much as +4 percent, and K1 by as much as +11
percent. So a method was devised that included the effect of the lunar nodal regression by including
it in a form that directly represents the modulation of each lunar tidal constituent. This method
involved using a multiplicative factor (usually called f ), one for each tidal constituent, as well as
a phase difference (u) that also varied over the 18.6 years. In Section 3.4.2 it will be seen that f and
u are explicitly included in the harmonic prediction equation.
The multiplicative factor f is called the node factor, even though this method is also used to
include the effect of the 8.85-year variation in lunar distance due to the rotation of the longitude of
the lunar perigee. Each node factor, for a specific tidal constituent, is a factor that multiplies the
tidal amplitude and fluctuates around 1.0. For M2, for example, it fluctuates from 0.963 to 1.038 and
back to 0.963 over the18.6-year period. f has usually been obtained from a table created with the
appropriate astronomical formulas, such as Table 14 in Schureman, 1958 (a portion of which
originally appeared in Harris, 1897). Zetler (1982) extended this table through the year 2025.
In the U.S. a single value of f has typically been used for each year, for each tidal constituent,
that value being for the middle of the year, it being assumed that the variation in f was slow enough
that the midyear value could be used for the whole year. However, some oceanographers prefer to
use f values for every one- or two-month period. The largest variation in f over a 18.6-year period
(for the most important tidal constituents) is found in O1 , which varies 18%, and in K1 , which
varies 11%. (There is a 40% variation in Mf, but that constituent is usually relatively small and
usually overwhelmed by meteorological energy.) The variation of f for M2 is 4%, but since in
many locations M2 is often much larger than O1 and K1, it may only be the 4% variation with which
one is most concerned. Since this is a lunar effect, there is no direct nodal effect on solar
constituents such as S2 or P1 (but, as will be seen later, there may be nonlinear hydrodynamic effects
that can lead to variations in S2 and P1 over the 18.6-year cycle ).
The phase difference associated with the 18.6-year or 8.85-year modulations is also treated in
the U.S. as a constant for each year and is included as part of the so-called equilibrium argument,
which is described in Section 3.4.2. Schuremans Table 15 provides the equilibrium argument
value for various tidal constituents for the beginnings of the years 1850 through 2000, the first
hundred years being taken from Harris, and Zetler extending it through 2025 (the u part of the
equilibrium is actually for the center of each year ). His Tables 16, 17, and 18 allow one to adapt
the equilibrium arguments (Vo+ u) in Table 15 to any month, day, and hour, but in these tables the
u portion is still considered constant for an entire year. These tables apply to the Greenwich time
meridian, but can be modified to apply to other time meridians and other longitudes (see Section
3.4.2).
41
Figure 2.18. A 100-year record of monthly (light line) and annual mean (dark line) tide
ranges at Seattle, illustrating the 18.6-year lunar nodal cycle (from Gill and Schultz, 2001).
In a traditional tidal prediction the amplitude of each tidal harmonic constituent is multiplied by
the node factor for the year that the tidal prediction is being made for. When a traditional harmonic
analysis is carried out, the resulting tidal constituent amplitudes are divided by the node factor for
the year that the data were obtained. (Or similarly, multiplied by the inverse of the node factor,
which Schureman calls the reduction factor and labels as F.) Although tables like those in
Schureman are often used, both the node factor and the equilibrium argument can be directly
included in harmonic analysis and prediction programs through use of the appropriate astronomical
equations. Of course, if a full set of satellite constituents is used, no node factors will be used, but
this is still typically not done for routine operational tidal predictions, and is primarily used only in
research activities.
Schureman (1958) and others used equilibrium theory when calculating the node factors and the
u portion of the equilibrium argument, but f and u can be affected by the hydrodynamics of the
oceans and adjoining waterways (see Section 2.3.4). However, except for the most precise analyses
and predictions, the use of f and u should be quite adequate. The use of satellite constituents, if one
has a 19-year data time series, will get around this problem, although it may take plots such as those
shown in Section 2.3.4 to understand and visualize the variation of each tidal constituent over the
19- (or longer) year period and the possible reasons for that variation.
2.2.5 Definitions Revisited, With Relationship to Harmonic Constants
In this section some of the basic definitions of various tidal quantities presented in Section 2.1.1
will again be looked at along with a some new ones. In addition, the relationship of these tidal
quantities to particular combinations of tidal harmonic constants will be looked at. Although most
42
of these tidal quantities are typically derived directly from data, namely, from tabulations of
observed high waters and low waters, in some cases they can also be calculated from the amplitudes
and phase lags resulting from a harmonic analysis (although not always with a simple formula). For
shorter data series, the harmonic constituent method has sometimes been preferred, the assumption
being that it eliminates meteorological effects, which for a short data time series could contaminate
the results. On the other hand, if the data series is too short, the harmonic constants being used may
not be accurate enough (see Section 3.3). The harmonic method does have the advantage of
providing some insight as to where such quantities come from and how they may vary with time,
but generally these quantities are still usually obtained directly from long data time series. Much
of what is presented in this section is covered in more detail in Special Publication 260 (S.P.260)
(U.S. C&GS,1952), which includes some additional terms not covered here.
Type of tide was defined in Section 2.1.1 as being either semidiurnal, mixed, or diurnal, and
examples of stations that are semidiurnal, mixed, or diurnal were shown in Figure 2.1. However,
there was a caution given about the difference between applying this phrase (as well as the terms
semidiurnal, mixed, and diurnal) to the tide on a particular day at a station versus applying this
phrase to the station itself, in the sense of classifying that station as being semidiurnal, mixed, or
diurnal. A station might be classified as being mixed, but there will be days at that station where
there will be only one HW and one LW and thus the tide on such days would be called diurnal.
There have been a variety of definitions proposed for classifying a station according to type-of-
tide categories and these definitions make use of the tidal harmonic constants at the station. Such
definitions are usually based on the (K1+O1)/M2 amplitude ratio [or the (K1+O1)/(M2+S2) ratio], and
sometimes take into consideration the difference in their epochs (phase lags), usually defined as
(M2o-K1o-O1o). When using one of these type-of-tide classification schemes, one must remember
that the chosen classification is an overall one and does not apply to every day of the month.
Stations will look mixed (that is, have two unequal high waters a day) at one time of the month but
can look diurnal (have only one high water per day) at other times of the month. The combined
effect of the K1 and O1 contributions varies throughout the month. The maximum K1+O1 combined
effect occurs once every 13.66 days, near maximum northern declination of the moon or maximum
southern declination (referred to as tropic tides). This occurs when the K1 and O1 constituent curves
are in phase and the maximum contribution is the sum of the amplitudes of K1 + O1. Likewise the
phase difference between the M2, K1, and O1 contributions also varies throughout the month. Figure
2.19 (which is reworked from S.P. 260, 1952) shows examples of tide curves for particular
combinations of changing (K1+O1)/M2 contribution ratios and changing (M2o-K1o-O1o) contribution
phase differences, which show examples of semidiurnal, mixed, and diurnal tides.
In S.P.260 it is stated that if the (K1+O1)/M2 ratio is less than 2.0, then the tide will be
semidiurnal all the time, no matter what the phase relation among M2, K1, and O1. If the contribution
ratio is greater than 4.0, then the tide will be diurnal all the time, no matter what the phase relation.
However, these statements are being technically very picky, since they consider even a slight
indication of a second low water or of a second high water as making the tide mixed. But as one can
see in Figure 2.19, the curves for a ratio of 3.0 also look diurnal. And even the curves for a ratio of
2.0 look diurnal (but with what might be called double high waters or double low waters).
With regard to classifying a station as being semidiurnal, mixed, or diurnal, there has been a
difference of opinion with regards to a standard set of tidal constituent amplitude ratio values that
could be used for defining the type of tide at a station. The CO-OPS Tides and Currents Glossary
(CO-OPS, 2000) has for some time referenced the values used by Dietrich (1967), who gives four
classifications: (1) a semidiurnal tide, which has a (K1+O1)/M2 tidal constituent amplitude ratio of
43
less than 0.25; (2) a mixed, mainly semidiurnal tide, which has a ratio from 0.25 to 1.5; (3) a mixed,
mainly diurnal tide, which has a ratio from 1.5 to 3.0; and (4) a diurnal tide, which has a ratio
greater than 3.0. Using this classification scheme, the tide curve examples shown in Figure 2.1
include semidiurnal, mixed, and diurnal tide curves. Boston has an (K1+O1)/M2 amplitude ratio of
0.19 and is semidiurnal. San Francisco (Golden Gate) has an (K1+O1)/M2 amplitude ratio of 1.03
and is mixed, mainly semdiurnal. Seattle has an (K1+O1)/M2 amplitude ratio of 1.21 and is also
mixed, mainly semidiurnal (but with a stronger diurnal signal). Dutch Harbor, Alaska, has an
(K1+O1)/M2 amplitude ratio of 2.07 and is mixed, mainly diurnal.
Dronkers (1964) references earlier ratio values given by the Coast and Geodetic Survey, which
are also given by Marmer (1926), which includes 0.5 as the upper limit for semidiurnal tides and 2.0
as the lower limit for diurnal tides, with mixed tides in between. Dronkers also states that the French
(in 1964) instead use the (K1+O1)/(M2+S2) tidal constituent ratio, and define semidiurnal as below
0.25, diurnal as above 1.25, and mixed as in between 0.25 and 1.25. Defant (1961) also uses
(K1+O1)/(M2+S2) ratios, but his categories are: semidiurnal for a ratio less than 0.25; mixed, mainly
semidiurnal for ratios between 0.25 and 1.5; mixed, mainly diurnal for ratios between 1.5 and 3.0;
and diurnal for ratios above 3.0. Parker (1977) uses Defants classification for classifying the type
of tide in his study of the tidal hydrodynamics of the Strait of Juan de Fuca - Strait of Georgia. The
Figure 2.19. Tide curves for particular combinations of (K1+O1)/M2 amplitude

ratios and phase differences (in degrees) defined as (M2o-K1o-O1o). See text for
discussion. Each tick along the horizontal axis is two hours. (Reworked from
S.P. 260, 1952. The second row is also found in Zetler (1959), but with the phase
difference defined without the .)
44
results are shown in Figure 2.20. The type of tide in this long deep waterway begins at the entrance
to the Strait of Juan de Fuca as mixed, mainly semidiurnal, then slowly changes to mixed, mainly
diurnal, and then slowly changes back to mixed, mainly semidiurnal up the entire Strait of Georgia.
The highest (K1+O1)/(M2+S2) tidal constituent amplitude ratio is 2.3 at Pedder Bay, a little southwest
of Victoria, British Columbia, Canada. The tide station at Victoria is classified as mixed, mainly
diurnal, but if one looks in the Tide Tables, the tide at Victoria with one HW and one LW on a given
day (thus being diurnal for that day) occurs 44% of the time. This percentage is an average for a
whole year, since it will vary from month to month. However, this again might be considered as a
technical fine point, since when one looks at the Tide Tables there are many days that show two
HWs and two LWs, but the LHW and HHW are close in value and the HLW between them is also
close in value, so that the tide looks diurnal with a double high water. Calling the tide diurnal on
days when this happens, the tides are actually diurnal 77% of the time at Victoria.
When the tide is mixed the order in which HHW, LHW, HLW, and LLW occur is called the
sequence of tide. Table 2.3 (reworked from S.P. 260) shows the sequence of tide for different phase
relationships among the contributions of M2, K1, and O1, this phase difference being the (M2o-K1o-
O1o). Thus, for example, the sequence of tide might be HHW, LLW, LHW, HLW or it might be
Figure 2.20. Chart illustrating the geographic variation in the (K1+O1)/(M2+S2) tidal
constituent amplitude ratio, as well as the type of tide classification for each region in the Strait
of Juan de Fuca Strait of Georgia. (From Parker, 1977)
45
LLW, HHW, HLW, LHW, or situations where the two low waters are almost equal or the two high
waters are almost equal.
When there is a mixed tide, the diurnal effect can accelerate the time of high water, that is, make
high water occur earlier than if there was no diurnal effect, or retard the time of high water, that is,
make high water occur later. The acceleration or retardation of the high and low waters are also
indicated in Table 2.3 for different phase relations.
One can also see in Figure 2.19 examples of the sequence of tide and acceleration/retardation
of HW or LW indicated in Table 2.3. For example, in this figure when the phase difference of the
M2, K1, and O1 contributions has become 45o and the contribution ratio is 0.5, one can see that the
sequence of tide is HHW, LLW, LHW, HLW, and that the HHW is accelerated while the LLW is
retarded (thus the tide rises more quickly to high water and falls more slowly to low water).
With all the various effects on the heights of high and low waters (and thus on the tide range)
and their timing, one ends up with a whole host of tidal definitions. Mean high water (MHW) is the
average of all the high waters over some time period, and similarly for mean low water (MLW). In
practice this time period is typically 19 years (the 18.6-year lunar nodal cycle rounded to the nearest
year) and is called the National Tidal Datum Epoch. The mean range of tide (Mn) is the difference
between mean MHW and MLW, in other words, the average of all the tidal ranges over some time
period. The average height of the high waters at spring tide is called mean high water springs
(MHWS) or spring high water, and similarly for mean low water springs (MLWS). The (mean)
spring range (Sg) is MHWS minus MLWS and is the average range occurring at the time of spring
tides. It is larger than the mean range, but is used only if the type of tide does not have a strong
diurnal signal. In the latter case the diurnal range is used (see below). Analogously, the (mean)
perigean range (Pn) is the average range occurring at the time of perigean tides.
When there is a significant diurnal inequality, a different set of high and low water and tidal
range definitions is more useful. Mean higher high water is the average of all the higher high waters
for some time period (again usually 18.6 years), and similarly for mean lower low water (MLLW),
mean lower high water (MLHW), and mean higher low water (MHLW). The great diurnal range,
Table 2.3. Table showing the effect of the phase difference P = (M2o-K1o-O1o) on the
sequence of tide and the acceleration (acc) or retardation (ret) of the high or low water.
P = (M2o-K1o-O1o) 1st HW of day 1st LW of day 2nd HW of day 2nd LW of day
0o HHW same LW* ret LHW same LW* acc
0o< P<90o HHW acc LLW ret LHW ret HLW acc
90o HW* acc LLW same HW* ret HLW same
90o< P<180o LHW acc LLW acc HHW ret HLW ret
180o LHW same LW* acc HHW same LW* ret
180o< P<270o LHW ret HLW acc HHW acc LLW ret
270o HW* ret HLW same HW* acc LLW same
270o< P<360o HHW ret HLW ret LHW acc LLW acc
46
or just referred to as the diurnal range, is the difference in height between MHHW and MLLW. The
(mean) tropic higher high water is the average of the higher high waters occurring at times of
maximum lunar declination, and likewise for the (mean) tropic lower low water. The great tropic
range (Gc), or just tropic range, is the difference in height between tropic higher high water and
tropic lower low water.
The timing of high waters has always been referenced to the moons transit, either over the
location of the tide gauge or over a time meridian. A lunitidal interval is the interval between the
moon's transit (upper or lower) over the local time meridian or the Greenwich meridian and the
following high or low water. (The transit is designated as upper transit when it crosses the time
meridian near the tide gauge, and as lower transit when it crosses the meridian that is 180 degrees
from the tide gauge location.) The mean high water lunitidal interval, or simply the high water
interval (HWI), is the average of all intervals between the moons transit and the following high
waters, for all phases of the moon, and similarly for mean low water lunitidal interval, or simply the
low water interval (LWI). When there is a diurnal inequality in the tide, separate intervals are
calculated for the higher high waters, lower high waters, higher low waters, and lower low waters.
These are designated respectively as higher high water interval (HHWI), lower high water interval
(LHWI), higher low water interval (HLWI), and lower low water interval (LLWI). In such cases,
and also when the tide is diurnal, it is necessary to distinguish between the upper and lower transit
of the Moon with reference to its declination. Intervals referenced to the moon's upper transit at the
time of its north declination or the lower transit at the time of south declination are usually marked
a. Intervals referenced to the moon's lower transit at the time of its north declination or to the upper
transit at the time of south declination are usually marked b. There is also a historical term that is
sometimes still seen, the establishment of the port (also called the vulgar establishment), which is
the average high water interval on days of the New Moon and Full Moon. The term high water full
and change (HWF&C) is more often used today. HWF&C is typically ten or fifteen minutes earlier
than HWI.
There is usually a delay between the exact occurrence of new moon or full moon and the
occurrence of spring tides. That delay is called the age of the tide (or sometimes the age of phase
inequality or simply the phase age) and is often on the order of a day or two, but it can be negative
(that is, spring tide in some locations, due to hydrodynamics, can come before new moon or full
moon).
There are also a variety of other ages. The age of parallax inequality is the interval between the
exact time when the moon reaches perigee and when the maximum effect (of the smaller distance
between the moon and Earth) is seen on the tide range. The age of diurnal inequality is the interval
between the exact time when the moon reaches maximum northern declination or maximum
southern declination and the maximum effect of this upon the range of tide (the tropic tide).
All of these ages can be calculated from the relevant tidal harmonic constituent epochs. The age
of tide is the time required for the M2 and S2 tidal constituents to arrive at a phase agreement. It can
therefore be obtained by dividing the difference in their epochs by the difference in their frequencies
(i.e., their angular speeds). The resulting formula for the age of the tide is thus
Age of the tide (in hours) = 0.984 (S2o - M2o)
in which the epochs can be local or Greenwich. Similarly, the age of parallax inequality is the time
required for the M2 and N2 tidal constituents to arrive at a phase agreement. Again, dividing the
47
difference in their epochs by the difference in their angular speeds, the formula for the age of
parallax inequality is thus
Age of parallax inequality (in hours) = 1.837 (M2o - N2o) .
And similarly, the age of diurnal inequality is the time required for K1 and O1 to arrive at phase
agreement, and the appropriate formula is
Age of diurnal inequality (in hours) = 0.911 (K1o - O1o)
Each of the types of tide range can also be defined according to the amplitudes of particular tidal
harmonic constants, but the formulas are not as simple as one might expect unless one is interested
in an approximate result. In S.P.260 one finds fairly elaborate formulas that require the use of a
form (Form 180) and tables in order to calculate values. Doodson and Warburg (1941, page 95) on
the other hand, provide much simpler formulas for what they term approximate values of the tidal
quantities. They give approximate values of the mean range of tide, the spring range, and the neap
range as:
Mean range = 2M2
Spring range = 2(M2+S2)
Neap range = 2(M2-S2)
These formulas assume there are not significant shallow-water overtides, M4 and M6.
S.P.260, however, states that the mean range is actually a little larger than twice the M2
amplitude because of the effects of some other semidiurnal constituents (even ignoring the effects
of M4 and M6). It uses a table (Table 4, page 50) showing the effect of S2 on the range of tide, and
another (Table 5, page 51) showing the effect of K1 and O1 on the range of tide. Zetler (1959)
provides approximate average values based on S.P.260 that are still quite simple and a small
refinement of the Doodson and Warburg values, i.e.,
Mean range (long period average) = 2.2 M2

Spring range (at new and full moon) = 2.1 (M2+S2)
Neap range (at quadrature) = 2.1 (M2S2)
Perigean range (moon closest to Earth) = 2.2 (M2+N2)
Apogean range (moon farthest from Earth) = 2.2 M2 1.7 N2
Tropic range (diurnal range at extreme declination)
for the special case of M2 << (K1+O1) = 2 (K1+O1)
2.3 Hydrodynamic Considerations
Only the oceans are large enough for the tide-generating forces to directly produce a tide of
significant size. As the tide wave generated in the ocean propagates onto the continental shelf and
into bays and estuaries (Figure 2.10), it is hydrodynamics (determined by the dimensions of the
ocean, bays, and rivers) that becomes the critical issue. It is the hydrodynamics that determines how
large the tide range will be and when the high and low waters will occur. It is the hydrodynamics
that determines how fast the tidal currents will flow and when slack waters will occur; and it is the
48
hydrodynamics that determines how significant the diurnal signal will be compared with the
semidiurnal signal.
In terms of harmonic analysis and prediction, the tidal frequencies may be determined by the
astronomy, but the values of the amplitudes and epochs result from the hydrodynamics. Full and
partial reflections of the very long tidal waves, the nearness of basin lengths or shelf widths to
resonance for particular tidal frequencies, continuity effects, frictional damping, and
advective/inertial effects all affect the tide and tidal currents. Even very long-term variations such
as that due to the 18.6-year regression of the lunar node can be affected by nonlinear
hydrodynamics, so that 18.6-year variations in a tidal constituent may not exactly match the
variation expected from astronomical considerations.
In shallow waterways the hydrodynamics also transfers tidal energy, through various nonlinear
processes, to new frequencies, creating overtides and compound tides (especially within the
semidiurnal band). These so-called shallow-water tidal constituents can be larger than many of the
astronomically generated tidal constituents. These same nonlinear processes also lead to interactions
between the tide and other nontidal phenomena such as storm surge and river discharge, which must
be considered when analyzing water level or current data for the tidal signal. To fully understand
the tide and to be able to interpret the results of a tidal analysis, one must have a basic understanding
of tidal hydrodynamics, an overview of which is provided in the following sections. More details
will be provided in Chapter 7 as determined from mathematical models.
2.3.1 Hydrodynamic Effects On Tide Ranges and Phase Lags and Tidal Current Speeds
When the very long tide wave generated in the ocean reaches the shallow water of the
continental shelf and the even shallower water of the bays and rivers, it is slowed up, amplified,
modulated, and distorted by a number of hydrodynamic mechanisms.
To understand what happens to a tide wave in a bay or river, it is helpful to first look at two
opposite extremes in idealized waves. Neither of these tide waves actually exist in a real waterway,
because they ignore friction, but they are useful in visualizing a real tide wave, which is always
some where between these two extremes.
If there is no friction and a long tide wave enters an endless river (that is, this tide wave is not
reflected back at some point), that wave will propagate up the river as a progressive wave (see
Figure 2.21). For a progressive tide wave, the crest of the wave (namely, high water) moves
progressively up the river, as does the trough of the wave (namely, low water). In such a progressive
tide wave the maximum flood current (namely, when the current is flowing the fastest up the river)
occurs at the same time as high water, and the maximum ebb current (namely, when the current is
flowing the fastest down the river) occurs at the same time as low water. Slack water, when the
current speed is zero, occurs exactly half way between high water and low water. (In a real river
with the energy dissipating effects of bottom friction, such a progressive wave would be damped,
and thus would slowly decrease in amplitude, and high water and maximum flood current would not
coincide. For now, however, this discussion will assume that there is no friction.)
If the river is of constant width, the amplitude (tide range) of this frictionless progressive tide
wave will not change as it moves up the river. However, if the width decreases as one moves up
the river, then the amplitude (tide range) will increase, because the same amount of water is being
forced through a smaller opening. If the depth of the river decreases there is a similar though less
dramatic amplifying effect (which in the real world is generally out weighed by the increased
frictional energy loss due to the shallower depths).
49
If this frictionless river is closed off at some point (e.g. by a dam) or there is a frictionless bay,
then the tide wave progressing up the river or bay will be reflected at the closed end, and will travel
back down the river or bay. This reflected wave is not observable by someone on the shore because
it is superimposed on the next incoming tide wave propagating up the bay, and it is the combination
of the two waves that is observed. The resulting combined wave is called a standing wave, because
the high and low waters do not progress up the bay (see Figure 2.22). The water surface simply
moves up and down everywhere at the same time, with the greatest tide range at the head of the bay.
With a standing tide wave, the tide range decreases as one move from the head of the bay toward
the ocean entrance, and if the bay is long enough, reaches a point where the tide range is zero (for
this idealized frictionless example), that point being called a node, and then starts increasing again
(see Figure 2.22). This node occurs at of a tidal wavelength from the head of the bay. (In a
progressive wave, high water comes a wavelength before low water, so if the high water of a
progressive wave travels a distance equal to of a tidal wavelength up the bay to the head, where
it is reflected, and then travels of a wavelength back down the bay, it will have gone a
wavelength and so coincide with low water of the next incoming progressive wave, and the two will
cancel each other out at that location, producing a zero tide range, the node.) For a standing wave,
high waters occur at the same time everywhere on one side of the node, which is the same time that
low waters occur on the other side of the node. The strongest tidal currents occur when water level
is near mean tide level, halfway between the times of low water and high water. At the times of high
water and low water there is no current flow (slack water). The water flows into the bay, stopping
the inward flow at high water, reverses direction, flows out of the bay until low water, at which time
it reverses again and starts flowing into the bay again.
One finds the largest tide ranges in bays that are exactly of a tidal wavelength long, due to
what is called resonance. When the water in the bay is forced to move up and down by the tide at
Figure 2.21. An idealized frictionless tide wave propagating up a river as a progressive

wave. High water occurs later as one moves upstream, and maximum flood occurs at the
same time as high water. The tidal wavelength is typically on the order of hundreds of miles.
(From Parker, 2004.)
50
Figure 2.22. An idealized frictionless tide in a bay as a standing wave (the water level is
shown for two opposite extremes, high water and low). High water occurs at the same time
everywhere on one side of the node (the point of zero range) and maximum flood current
occurs half way between low water and high water. With friction there is no point of zero tide
range, only a point with a minimum tide range, and the times of high water progress slightly
up the bay. (From Parker, 2004.)
the entrance, it will freely oscillate (slosh up and down and back and forth) with a natural period that
depends directly on the bays length and inversely on the square root of its depth. If the basin has
the right combination of length and depth so that the natural period is exactly the same as the tidal
period, then the oscillation inside the bay will be synchronized with the oscillation at the entrance
due to the ocean tide. In other words, the next ocean tide will be raising the water level in the bay
at the same time that it would already be rising due to its natural oscillation (stimulated by the
previous ocean tide wave), so that both are working together, thus making the tide range inside
higher. (In the real world friction keeps these resonating oscillations from being infinitely large, by
taking away some of the energy.) The tidal wavelength is determined by the depth of the bay, and
when the length of the bay equals of a tidal wavelength, then the bays natural period of
oscillation will be the same as the tidal period.
In the above discussions bottom friction was left out of the discussion, but bottom friction
greatly affects all hydrodynamics and is especially important in shallow waters. Because of bottom
friction the tide wave in real bays actually falls in between the extremes of pure progressive wave
and pure standing wave described above. This is because friction reduces the amplitude of the tide
wave as it travels. Thus, the reflected wave will always be smaller than the incoming wave,
especially near the bay entrance (since the reflected wave has traveled the longest to get all the way
back to the entrance), and the combination of the two frictionally damped progressive waves will
not be a pure standing wave. There will be no point of zero tidal range (no real node), but only an
area of minimum tidal range (a quasinode). There will be some progression of high waters and low
waters up the bay, but not as quickly as a pure progressive wave. This progression will be faster
near the entrance and slowest near the head of the bay. Maximum flood or ebb currents will not
occur exactly half way between high water and low water. A basin of a wavelength long will still
51
produce the largest possible tidal range at the head of the bay, but friction keeps that tide range much
smaller than it would be without friction.
In some bays the very high tide range at the head of the bay is due to a combination of both a
narrowing width and a near resonant situation (due to the right length and depth). The highest tide
ranges may involve several amplifications, the bay being perhaps connected to a gulf which is
perhaps connected to a wide continental shelf, with amplifications of the tide wave occurring in each
basin. This is the case with the Bay of Fundy tides, the tide wave being already amplified by the
continental shelf and the Gulf of Maine prior to entering the Bay of Fundy (see Section 7.4.2).
(Formulas representing the effects described above are derived mathematically in Section 7.4.1,
except for the width effects, which are derived in Section 7.4.2.)
If a bay is wide enough one also sees larger tide ranges on the right side of the bay (looking up
the bay) due to the Coriolis effect. The Coriolis force, a fictitious force due to observing motion
from the rotating reference frame of the Earth (see Parker, 1998), acts perpendicular to the flow of
the water, thus pushing water currents to the right in the Northern Hemisphere (and to the left in the
Southern Hemisphere) as they flow. For a pure progressive tide wave (no friction), at high water
the tidal current flows up the bay, so the tidal height will be greater on the right-hand shore than on
the left-hand shore. At low water the tidal current flows down the bay, so the tidal height will be
lower on the right-hand shore than on the left hand shore. The result is that the tide range (high
water minus low water) will be greatest on the right-hand shore (looking up the bay).
For a pure standing wave (no friction) the pattern of high water caused by the Coriolis force is
more complicated, and is shown in the upper half of Figure 2.23. This figure shows lines of
constant tidal range (corange lines), as well as lines of locations with the same time of high water
(cotidal lines), in an idealized rectangular basin for the case where the effect of bottom friction is
ignored. A single point of zero tidal range (a node) occurs in the center of the bay of a tidal
wavelength from the head of the bay. This figure comes close to representing the corange and
cotidal lines in a bay that is very deep. A more typical case, including the damping effect of bottom
friction, is shown in the bottom half of Figure 2.23. In this case the node has moved to the left
(when looking up the bay) and becomes a virtual node since it is on land. (See Section 7.5 for
derived mathematical formulas describing this effect.) One can see some similarity between the
pattern of corange lines in the lower half of Figure 2.23 and the corange lines in Figure 2.5 for the
Strait of Juan de Fuca - Strait of Georgia, (although the latter has many geographic variations not
included in the simple regular basin of Figure 2.23), and likewise for the cotidal lines in Figure 2.23
and Figure 2.6. In Figure 2.5 the pattern of a quasinode is also seen to the southwest of Victoria.
A similar pattern, including a quasinode on the western shore near Smith Point, is also apparent in
the M2 coamplitude (half the tide range) chart for Chesapeake Bay shown in Figure 2.24.
The largest tide ranges are found in bays that are close to of an M2 tidal wavelength long, such
as the Bay of Fundy-Gulf of Maine (Canada), Ungava Bay (Canada), Bristol Bay (United Kingdom),
Gulf of St. Malo (France), Cook Inlet (Alaska, US), the Gulf of Cambay (India), and the eastern end
of the Magellan Strait (Chile). Huge tide ranges are not restricted to bays. If the continental shelf
is the right combination of depth and width, a near resonant situation can also result. This is the
reason for the 40-foot tidal ranges along the coast of southern Argentina. The continental shelf there
is over 600 miles wide, and includes the Falkland Islands near the edge of the shelf (where the tide
range only reaches 6.5 feet). The distance from the Argentinean coast to the edge of the shelf is
fairly close to of a tidal wavelength for that depth of water. Essentially, that wide shelf has a
natural period of oscillation that is fairly close to the tidal period.
52
Figure 2.23. The effect of Coriolis force on the M2 tide range (corange lines) and the time
of high water (cotidal lines) for an idealized rectangular bay. The top panel shows the case
with no bottom friction and has a single point of zero tide range (the node) in the middle of
the bay. The bottom panel includes the effect of bottom friction, and there is no point of zero
tide range.
The largest tidal currents in bays tend to be near the entrances. Maximum tidal current speeds
are zero at the head of the bay (since there is no place for the water to flow). As one moves down
the bay toward the ocean the maximum flood and maximum ebb tidal current speeds increase, with
the greatest speeds occurring at the entrance, or, if the bay is long enough, at the area of smallest tide
range (the nodal area). However, if the width of the bay decreases at any point, the current speeds
will be increased in that narrow region (since the same volume of water is being forced to flow
through a smaller cross-section, it must flow faster). This can be especially dramatic if there is a
sudden decrease in width and depth. The largest tidal currents are found in narrow straits in which
the tides at either end have different ranges or times of high water (see Section 2.3.6g). Where a
strait suddenly becomes very narrow or where it bends, eddies and whirlpools can be formed as the
result of the sheltering effect of the land and the inertia of the coastal flow.
The dimensions of a basin can also determine the size of the diurnal tidal signal compared with
the usually dominant semidiurnal tidal signal (also see Section 7.4.1). A particular bay could have
a natural period of oscillation that is closer to the diurnal tidal period (approximately 24.84 hours)
than to the semidiurnal period, thus amplifying the diurnal forcing at the entrance to the bay more
than the semidiurnal signal. Depending on the size of the diurnal signal at the entrance the result
could be a mixed tide or a diurnal tide. At such locations (such as parts of the Gulf of Mexico) the
tide will be diurnal near times of maximum lunar declination, but will be mixed near times when the
moon is over the equator.
53
Figure 2.24. M2 coamplitude chart for Chesapeake Bay (in feet). (From Fisher,
1986.)
54
The wavelength, , of a tide wave in a bay depends on the depth of the water, D, and on the tidal
period, T, according to =T(gD) (if frictional effects are ignored). The shallower the bay the
shorter the wavelength. The longer the tidal period the longer the wavelength. A diurnal tidal
component has a wavelength twice as long as a semidiurnal tidal component since its period is twice
as long. When a waterway is shallow enough and long enough so that more than of a semidiurnal
wavelength fits in the waterway, there will be a nodal area with a very small semidiurnal tidal range.
This will be an area where the diurnal tide could dominate, since the diurnal tide would still be large
at the semidiurnal nodal area (the diurnal node being twice as far from the head of the bay). Thus
near the head of the waterway the tide could be semidiurnal, but near the semidiurnal nodal area the
tide could be mixed or even diurnal. This is the case near Victoria, British Columbia, at the
southeastern end of Vancouver Island (see Figure 2.20). At that location along the Strait of Georgia-
Strait of Juan de Fuca waterway, the semidiurnal tidal component decreases to a minimum, but the
diurnal component does not, and so the tide becomes diurnal, while at the northern end of the Strait
of Georgia the tide is mixed, mainly semidiurnal.
Whether due to a basin size conducive to amplifying the diurnal signal or due to the existence
of a semidiurnal nodal area (leaving the diurnal signal as the dominant one), there are numerous
areas around the world with strong diurnal tides places like Norton Sound in Alaska near the
Bering Strait, and various (but not all) locations in the Philippines, New Guinea, and the islands of
Indonesia. In southern China, at Beihai, and at Do Son, Vietnam, the diurnal signal is very
dominant, with tidal ranges that reach 15 feet and 10 feet respectively (near times of maximum
southern declination of the moon). In these locations the tide remains diurnal even when the moon
is over the equator.
2.3.2 Nonlinear Effects of Shallow Water Overtides and Compound Tides
The shallower the water depth is the more the tidal wavelength will shorten and the faster the
tidal characteristics of a waterway will change with horizontal (geographic) distance. When the
tidal wavelength is shortened to near the length of a bay or river basin, this can bring the dynamic
situation closer to resonance and increase the tide ranges. [Or, one can also look at it from the point
of view of the shallower depths increasing the natural periods of a bay or river basin to be closer to
the tidal period.]
Shallow water, however, can have other effects on the tide. It can, for example, distort the shape
of the tide wave, that is, make it very asymmetric, so that its rise and fall (and its flood and ebb) are
no longer equal (see the second curve in Figure 2.25). The tide can then no longer be described by
a simple sine wave (the first curve in Figure 2.25). In some cases such distortion leads to double
high waters or double low waters (see the third curve in Figure 2.25). The extreme case of distortion
is a tidal bore (the fourth curve in Figure 2.25),when the tide wave becomes so steep that it is
essentially breaking and it moves up a river as a turbulent wall of water.
Shallow water distorts the tide through several mechanisms that are nonlinear that is, in the
equations of motion (which will be discussed in Sections 7.3 and 7.6) each mechanism can be tied
to a specific term in which key parameters (such as elevation, , or velocity, u) multiply each other,
which leads to energy transfer. (Linear terms contain only one key parameter and their separate
effects simply add, with no interaction.)
55
The speed, C, at which a long tide wave travels depends on the depth of the water, D,
approximately as the formula C=(gD). When the depth of the water is much greater than the tidal
range, the speed of the crest of a tide wave and the speed of the trough are virtually the same, since
the tide wave itself has only a very small effect on the total water depth. However, in the shallow
water where the depth is not much greater than the tide range, the total water depth under the crest
is significantly larger than the total water depth under the trough. In this case, the crest of the wave
(high water) travels faster than the trough of the wave (low water). If the tide wave travels far
Figure 2.25. Typical tide curves (over 1 tidal

cycles) for an area with no shallow-water effect (top
panel) and for three areas with increasing degrees of
distortion caused by the shallow water. The third
panel shows a double low water. The fourth panel
shows the almost instantaneous rise in water level
due to the passage of a tidal bore. (From Parker,
1999)
56
enough, the crest begins to catch up with the trough ahead of it (which is falling behind the crest
ahead of it). Thus, high water arrives sooner than it would in deeper water, and there is a faster rise
to high water and a slower fall to low water. This can be seen in the water level curves in Figure
2.26 as one moves further up the Gironde River (in France).
In other waterways the tide wave will not be progressive since it will reflect back from a closed
end (or from a sudden width decrease) and produce a more standing wave. Most waterways are
somewhere in between progressive and standing (see Section 7.4.1). The shape of the tide curve
could perhaps look more like that shown in the second curve in Figure 2.25 (or one of the curves in
Figure 2.26). In terms of harmonic constituents, this distortion transfers energy from M2 into the
second harmonic, a constituent called M4, with half the period of M2. Combining an M2 tide curve
and an M4 tide curve, one can produce the distorted tide curves shown in Figure 2.25, with the
M4/M2 ratio increasing as one goes from top curve to bottom curve in that figure. The third curve
shows a double low water, but with a different phase relationship between M2 and M4 one could
obtain a double high water.
Another shallow-water distorting mechanism is caused by bottom friction, which can have both
asymmetric and symmetric effects. The asymmetric effect (similar to that just discussed and
represented in Figure 2.25) results because friction has a greater effect in shallow water than in deep
water (there being less water to have to slow down), and so it slows down the trough more than the
crest, contributing to the distortion of the tide wave and the generation of M4. The symmetric effect
results because energy loss due to friction is proportional to the square of the current speed. This
means that there will be much more energy loss during times of maximum flood and maximum ebb
Figure 2.26. Water level curves from the Gironde River in France showing the effect of
nonlinear shallow-water distortion. At LeVerdon (just inside the river entrance) the rise to high
water takes about an hour longer than the fall to low water, while at Bordeaux the fall takes 4
hours longer than the rise. (Reworked from George and Simon, 1984.)
57
than near times of slack water (or minimum flow). This results in the generation of another higher
harmonic, M6, with a period of one third that of M2. This effect, combined with the asymmetric
effect, can lead to double high or low waters (see third curve in Figure 2.25).
Higher harmonic tidal constituents like M4 and M6 are referred to as overtides (a term analogous
to the term overtones in acoustics). M4 is the first overtide, and the second harmonic of M2. M6 is
the second overtide and the third harmonic of M2. M8 is the third overtide and the fourth harmonic
of M2. Whereas M4 and M6 are generated by first-order nonlinear processes, M8 is generated by
second-order nonlinear processes (that is, it is generated by nonlinear effects on M6).
Friction dissipates energy from the entire tide wave and slowly wears the entire wave down.
However, if, as the tide wave propagates up the river, the rivers width is decreasing significantly,
this can keep the amplitude of the wave high in spite of the friction. Thus, the tide wave can
continue to travel up a narrowing river, getting more and more distorted in shape. A further
distortion can be caused by the river flow interacting with the tide (see below). In the extreme case
the distortion from all these effects can lead to the creation of a tidal bore (see fourth curve in Figure
2.25).
The above symmetric quadratic friction effect also causes the interaction of two tidal
constituents, such as M2 and N2. M2 and N2 go in and out of phase over a 27.6-day cycle (perigee
to apogee to perigee). In this case the greatest energy loss occurs when M2 and N2 are in phase and
producing the strongest tidal currents, and the lowest energy loss occurs 13.8 days later with M2 and
N2 are out of phase and producing the weakest tidal currents. Because energy loss is proportional
to the square of the current speed, the increased energy loss when M2 and N2 are in phase is greater
than the decreased energy loss when they are out of phase, and the result is that each constituent will
be smaller than if it existed without the other present. The reduction in N2 (due to M2) will be
greater than the reduction in M2 (due to N2), because M2 is much greater than N2. However, M2 will
be reduced by the combined interactions of all the other tidal constituents.
There is a 27.6-day modulation of this energy loss from M2 and N2 and this produces two new
compound tidal constituents called 2MN2 and 2NM2. (Similarly, the above asymmetric mechanisms
also cause interactions between constituents, producing higher frequency constituents such as MN4
from M2 and N2.) Table 2.4 lists many overtides and compound tides and shows which nonlinear
mechanisms can produce them. These shallow-water effects that distort and modulate the tide (and,
as will be seen, cause interactions with storm surge and river discharge) are called nonlinear effects
because the mechanisms that produce these effects are represented by several nonlinear terms in the
equations of motion used to model the tidal hydrodynamics. These various nonlinear mechanisms
that lead to the generation of overtides and compound tides are described in more detail in Section
7.6, with mathematical derivations and physical explanations. Table 2.4 is based on a Fourier
decomposition of the one-dimensional nonlinear hydrodynamic equations (described in Section
7.6), from which four basic nonlinear mechanisms are identified for this simple one-dimensional
case.
The two classical nonlinear shallow-water terms, M(u)/Mx in the one-dimensional continuity
equation, and uMu/Mx in the one-dimensional momentum equation (where is the water level
elevation, u is the current, and x is distance), produce asymmetric effects, that is they affect one half
of the tidal cycle differently than the other half of the cycle, and this leads to even harmonic
overtides and low-frequency compound tides. The same asymmetric results are also produced by
the effect of elevation on the frictional momentum loss per unit volume of fluid, as first pointed out
by Parker (1984).
58
M2 interacting Tidal Constituents Generated With Frequency:

with constituent
C * *M2 C* (2M2 C) (M2 + C) (2M2 + C) (4M2 C)
M2 mean * ----- M4 M6 M6
N2 MN (Mm) 2MN2 (L2) MN4 2MN6 4MN6
S2 MS (MSf) 2MS2 (2) MS4 2MS6 4MS6
K1 MK1 (O1) 2MK3 [MO3] MK3 [2MO3] 2MK5 [3MO5] 4MK7 [3MO7]
O1 MO1 (K1) 2MO3 [MK3] MO3 [2MK3] 2MO5 [3MK5] 4MO7 [3MK7]
Nonlinear M (u)/Mx M (u)/Mx
terms u*u* u*u*
responsible uMu/Mx u*u* uMu/Mx u*u* u*u*
u*u* with a u*u* with a
mean flow mean flow
symmetric or
asymmetric asymmetric symmetric asymmetric symmetric symmetric
NL mechanism
in terms of r ** r M2 + r M2 r 3M2 r 3M2 + r
* the mean can be in the form of a residual mean circulation or a shift in mean sea level
** r = *M2 C*
Table 2.4. Compound and overtide constituents generated by four nonlinear terms in the one-
dimensional equations of motion. The nonlinear terms are explained in the text. C is any tidal
constituent listed in the first column. A constituent in ( ) is an astronomical constituent with the
same frequency as the compound tidal constituent. A constituent in [ ] is another compound
constituent with the same frequency. See text, as well as Sections 7.3 and 7.6 for the
mathematical treatment behind this table. (reworked from Parker, 1991a)
This is represented by the term u*u*, which is actually the second term of the binomial expansion
of the one-dimensional friction term, u*u*/(1+0), where the elevation is in the denominator and
0 is a scaling factor (0 = tidal amplitude/depth). (See Section 7.6.) Quadratic friction, u*u*, can
also produce an asymmetric effect and these same constituents if there is a mean flow present, such
as river flow. However, without a mean flow present, quadratic friction primarily produces
symmetric effects that lead to odd harmonic overtides as well as to compound tides in the
semidiurnal band. A few of these compound tides have the same frequencies as particular
astronomical constituents, for example, 2MN2 has the same frequency as L2, and 2MS2 has the same
frequency as 2. And two compound tidal constituents produced by different nonlinear mechanisms
can also have the same frequency, such as 2MK3 and MO3. Not shown in this table, the lateral
inertial terms in the 2-dimensional or 3-dimensional momentum equations can also produce
asymmetric effects in tidal currents similar to those produced by the four terms mentioned above
(see Section 7.6.7 and Parker, 1991a).
Until the 1970's tidal analysts had simply treated these shallow-water tidal constituents as being
the sums or differences of the astronomical constituents, and paid little attention to the specific
59
nonlinear hydrodynamic mechanism that produced them. In particular the quadratic frictional
nonlinear mechanism for generating compound tides and overtides was ignored for decades after
Proudman (1923) had first proposed it. For example, it was assumed that M6 was produced from
M4 via the classic shallow-water nonlinear continuity term (actually a second-order effect) rather
than directly from M2 via quadratic friction (a first-order effect).
Quadratic friction produces many compound tidal constituents that have the same frequency as
particular astronomical constituents. The compound tidal constituent 2MN2, for example, has the
same frequency as the astronomical constituent L2, but a very different node factor. In coastal
regions and especially up into bays and estuaries, 2MN2 is much larger than L2, yet L2 was often still
used for predictions. Use of the L2 node factor for (in reality) the 2MN2 constituent can lead to the
effect of this constituent being doubled or halved in predictions for other years (than the year that
was analyzed). The compound tidal constituent 2MS2 has the same frequency as the astronomical
constituent 2, but in this case both constituents should theoretically have the same node factor.
The node factor for M6 was often calculated by cubing the node factor for M2 (for example, this
was erroneously done in Schureman, 1958), as though M6 was a third-order harmonic resulting from
the same asymmetric nonlinear phenomenon that produces M4, i.e. generating M6 from M4 in the
same way that M4 is generated from M2. However, a Fourier decomposition of the hydrodynamic
equations (see Section 7.6) shows that M6 is primarily due to the symmetric nonlinear effect of
quadratic friction and is a first-order effect, so that the node factor should equal the square of the
node factor for M2 . The effect of not understanding the origin of shallow-water constituents, and
thus of perhaps choosing the wrong node factor, can really add up for a location like Anchorage,
Figure 2.27. Tide curves for particular combinations of M4/M2 amplitude ratios and 2M2o-M4o
phase differences (in degrees). See text. Each tick on the horizontal axis is one hour.
(Reworked from S.P.260, U.S.C.&G.S., 1952.)
60
Alaska, at the northern end of Cook Inlet, where most of the 114 constituents used by CO-OPS to
predict the 30-foot range tides are shallow-water constituents.
The way in which a tide or tidal current curve is distorted by the nonlinear effects of shallow
water depends on the hydrodynamics of the particular waterway. There is a correlation between the
particular shape of the curve and the amplitudes and epochs of the overtide constituents relative to
the semidiurnal constituents. Figure 2.27 shows examples of tide curves for particular combinations
of M4/M2 amplitude ratios and phase differences between M4 and M2 (defined as 2M2o-M4o). The
effect of M4 on M2 is asymmetric, that is, it affects one half of the tidal cycle differently than the
other half. For an amplitude ratio of 0.3 and a phase difference of 0o, one sees in this figure a
steepening of high water and a flattening of the low water, the latter often referred to as a stand of
tide (at low water). If the amplitude ratio approaches 0.5, one sees double low waters. The same
applies to high waters if the phase difference is 180o, in which case for an amplitude ratio of 0.3
there will be a stand of tide at high water and a steeping of low water, and for a ratio of 0.5 there will
be double high waters. For a phase difference of 90o one sees a flattened area of the curve near the
mean tide level point of the falling tide; here the tide has a slow fall from high water to low water
and a rapid rise from low water to high water. For a phase difference of 270o there is again a
flattened area around mean tide level, but this time during the rising tide, which is now slower than
the rapid fall.
Figure 2.28. Tide curves for particular combinations of M6/M2 amplitude ratios and
3M2o M6o phase differences. Each tick mark along the horizontal axis is one hour.
(Reworked from S.P. 260, U.S.C.&G.S., 1952.)
61
Figure 2.28 shows examples of tide curves for particular combinations of M6/M2 amplitude ratios
and phase differences between M6 and M2 (defined as 3M2o-M6o). The effect of M6 on M2 is
symmetric, that is, it affects both high waters and low waters in the same way. Depending on the
phase difference this can mean flattening both high and low waters or making them both steeper.
However, if M6 is present there will always be an M4, so M6 would never act solely on M2. A double
high water, for example, is more likely caused by a combination the effects of both M4 and M6, with,
for example, M4 flattening out high water and M6 causing the minimum between two high water
peaks. M6 can affect the overall tide range. The asymmetric effect of M4 will probably not
significantly affect the overall tide range, but it will significantly shift the times of high water (often
earlier) and low water (often later).
M8 is often larger than many of the compound constituents shown in Table 2.4. This table only
goes up to seventh diurnal tidal constituents (e.g., 4MK7), because it includes only constituents
generated by a first-order nonlinear interaction between two astronomical constituents. M8 is
generated by the interaction of M6 with M2 via the asymmetric nonlinear terms. It often reaches a
significant size only because M2 is usually so much larger than the other tidal constituents.
2.3.3 Nonlinear Tidal Interaction With River Flow and Storm Surge
In a tidal river, water flow is due to both the tidal current and the river current itself (i.e., the
fresh water flowing downhill). The result of the combined tidal current and river current is a faster
and longer lasting ebb current phase and a slower shorter flood current phase. Far enough up a river,
where the river flow is faster than the strongest tidal current, the flow of water will always be
downstream. In this case, the speed of flow will oscillate, flowing the fastest downstream at the time
when maximum ebb occurs further down the river and flowing the slowest downstream at the time
when maximum flood occurs further down the river. This is a simple linear addition of the river
current to the tidal current.
However, because of the shallow water, the river flow also nonlinearly interacts with the tide
and distorts it, mainly due to the effect of bottom friction. As already mentioned, energy loss due
to friction is proportional to the square of the total current speed. During ebb, the tidal current is in
the same direction as the river current and the result is a larger combined ebb current, with
increased energy loss. During flood, the tidal current is in the opposite direction as the river current
and the result is a smaller combined current, with reduced energy loss. This not only has an
asymmetric effect that distorts the tide (causing a faster rise to high water, delaying the time of low
water, and increasing the size of M4), but it also further wears down the entire wave because the
increased energy loss during ebb is larger than the decreased energy loss during flood. In Figure
2.29 one can see the tide range shrink when the river discharge increases. (See also Sections 7.6.3
and 7.6.6c for the mathematical treatment of this effect.)
Another type of shallow-water effect causes interactions between the tide and low-frequency
storm surges (generated by the wind) that have periods longer than tidal periods. In this case, when
the water level is raised by an onshore wind, the water depth increases and changes the tidal
dynamics, usually increasing the tide range. When an offshore wind lowers the water level,
decreasing the water depth, the result is usually a decreased tidal range. As will be seen in Sections
7.6.4, 7.6.5, and 7.6.6d, it is the two asymmetric nonlinear terms that involve elevation () that have
the most significant effect (e.g., see Table 7.1 in Section 7.6.6d).
Knowing that river discharge and storm surge can modify the tide, it is important when
harmonically analyzing water level data to make sure that these data were not taken only during such
62
Figure 2.29. Water level data from the tide gauge at Trenton, NJ, during a high river
discharge period (January 20-29, 1979). The tide range is reduced when the river
discharge is high. (From Parker, 1984.)
meteorological events. For example, Figure 2.30 shows the results of a harmonic analysis of 15 days
of data taken from six NOS water level stations during a high runoff period in the Delaware River
in March 1978, compared with the results of a 15-day analysis of a low runoff period (chosen to
have similar astronomical conditions). The three upriver stations show the effects of the higher river
current speeds on the harmonic constants the M2 amplitude is reduced and the M4/M2 ratio is
increased when the river flow is greatest. Down river, where it is wider and deeper and the river
current speeds are much slower, these effects are not seen.
If a water level data time series includes a period with higher than average river runoff, and the
harmonic constants calculated from this data record are used to make a tide prediction, then the tide
will be under predicted for the rest of the year. This may be a seen by looking at the two curves in
Figure 2.31, which shows another high runoff period at the Trenton tide gauge in the Delaware
River. The solid curve shows the tidal (high-pass filtered) portion of the water level record from that
high runoff period. The dashed curve is the predicted tide for that time period if there had been no
river runoff (based on 31 harmonic constants from a 7-month period when there wasn't high runoff).
If one didn't have those harmonic constants from the 7-month data record, but only had constants
from the high runoff period, then one would produce tide predictions (for the entire year) that would
look more like the solid curve (including the distortion due to the increased M4/M2 ratio) than the
dashed curve. This river example is a fairly extreme one, heavy runoff during a freshet period after
a snow melt. Such extremes, with differences on the order of 3 to 5 feet in tidal range, are usually
limited to only a short part of each year (at least for this river).
Storm surges, however, occur throughout the year (although they are generally more frequent
and larger during the winter months). Subtidal storm surges (with periods longer than the tidal
period) are the largest wind induced changes in water level, and, like river flow, they can also interact
63
Figure 2.30. Comparison of 15-day harmonic analyses of water level from Delaware River
and Bay during a high runoff period and during a low runoff period, both periods with similar
astronomical conditions. is the local epoch, i.e., the phase lag relative to the local time
meridian (75oW). (From Parker, 1991a)
64
Figure 2.31. The tidal portion of a water level record from Trenton, NJ, on the Delaware River,
during a high river discharge period (solid line) plotted with the tide predictions for the same
time period as if there were no river discharge (dashed line). The difference in the two curves
is shown in the lower plot. The tidal portion of the water level record was produced by filtering
the water level with a Doodson tidal filter (see Section 3.8.2) and subtracting it from the water
level time series. (From Parker, 1991a)
with and modify the tide. Figure 2.32 shows an example, also from Delaware Bay. This figure shows
the tidal portion of water level records from NOS four water level stations in Delaware Bay and
from the NOS water level station at Atlantic City (outside the Bay on the Atlantic Coast) along with
the predicted tide curves based on harmonic constants from seven months of data. The bottom curve
in the figure shows the subtidal water level signal at Atlantic City. The trough of a subtidal storm
surge occurs on January 18th. Within the Bay there is not only a water level drop (not shown in
these high passed water level curves), but a tide range reduction which increases as one moves up
the Bay to Trenton. Late on the January 19th and into the 20th one sees crest conditions for the
subtidal signal, and a corresponding tide range increase up the Bay (before the increased river
discharge on the January 21st starts to have an effect).
Again, harmonic constants calculated from a time period with significant storm surge could have
errors, especially if the record is short. If the record is not short and includes several subtidal crests
and troughs, there may not be a dramatic affect on the harmonic constituents. What does happen,
as seen in the frequency domain, is that the some of the energy in a tidal spectral line is smeared to
both sides, forming a "cusp" around the line (see Section 3.10.2). This implies that some energy is
lost from the tidal lines, and that the harmonic constant amplitudes are reduced to some degree.
During periods with large river discharge or large storm surge, the tide or tidal current cannot
be predicted accurately using statistical/harmonic analysis methods. No matter how accurate the
tidal harmonic constants are (in predicting the tide during most of the year), they cannot provide
accurate predictions during strong nontidal events, because the tide has been changed by those
strong nontidal events. If the nontidal phenomena themselves cannot be accurately predicted (which
is often the case), then one cannot predict how the harmonic constants will be modified by those
nontidal phenomena. Recently a statistical method called the continuous wavelet transform method
65
Figure 2.32. The tidal portion of the water level record plotted with tide predictions for three
locations along the Delaware River and Bay and a station on the Atlantic Coast outside the Bay.
The bottom curve is the low-pass filtered water level record from Atlantic City showing a
subtidal storm surge. The upper curves show the effect of the storm surge on the tide (see text).
(From Parker, 1991a)
was developed to try to deal statistically with the effect of strong nontidal events on the tide and tidal
current. This method will be discussed in Section 3.5.5. Fully nonlinear hydrodynamic numerical
models can easily handle the nonlinear interaction between the tide and river flow or storm surge,
and they can do a fine job hindcasting or nowcasting strong nontidal events. They can also be used
to forecast the water level (and currents) during such events if reasonably accurate forecast surface
winds and other forecast meteorological parameters can be input into them.
Over the last two decades there has been much progress in providing real-time water level and
current information to the maritime community. Such real-time data is needed because tide and tidal
current predictions alone cannot give the mariner information about the often dramatic effects of
wind on water levels and currents (Parker, 1986, 1995), and so the actual water level and current
data are needed. For predictions into the future (beyond the time of the last real-time data points)
numerical hydrodynamic models are also now being used operationally by NOS (driven by tide
predictions, forecast winds, forecast atmospheric pressure, and forecast fresh water inflows) to find
ways to forecast the nontidally-induced water level changes and current and their interactions with
the tide and tidal current (see Section 8.6.3). Since such models are nonlinear, they automatically
reduce (or increase) and distort the tide as necessary.
66
2.3.4 Nonlinear Hydrodynamic Effects On Node Factors
As discussed in Section 2.2.4, the 18.6-year variation in the lunar tidal force has typically been
handled in a form that directly represents the modulation of each lunar tidal constituent using a lunar
node factor, f. The f for each lunar tidal constituent is usually regarded as constant for the period
of analysis (or prediction) and is typically obtained from astronomically-determined tables, such as
Table 14 in Schureman (1958). However, there are hydrodynamic effects on these node factors,
specifically through nonlinear interactions due to shallow water. To illustrate this one can look at
how closely the long-term variation of uncorrected tidal constituents actually matches the variation
predicted by astronomical considerations (using equilibrium tide theory).
Figure 2.33 shows the node factors calculated from harmonically analyzed water level data,
specifically 19 uncorrected one-year least squares harmonic analyses of water level data from
Philadelphia, Pennsylvania, on the Delaware River. For each tidal constituent, the 19 calculated
(and uncorrected) amplitudes were averaged, and then each individual one-year constituent
amplitude was divided by this 19-year average to obtain the calculated node factor. Also plotted in
the figure are the astronomically determined mid-year node factors taken from Table 14 of
Schureman (1958). Philadelphia is in a very dynamic location, located half way up a shallow
estuary and affected by river discharge as well as storm surges that propagate into Delaware Bay
from the wide Atlantic continental shelf.
It is apparent from the plots in Figure 2.33 that for most of the tidal constituents, the calculated
node factor curve does not match the astronomical node factor curve very well. Before looking at
these curves too closely, it should be remembered that other factors can affect tidal amplitudes over
long periods beside the 18.6-year nodal cycle. For example, if a basin depth changed over the years
due either to shoaling or dredging or land subsidence, the tide range could also change over the
years. This could affect different tidal constituents differently depending on whether the change
brings a constituent closer to or farther from resonance. Also, due to nonlinear frictional
interactions, a change that made M2 tidal currents stronger might decrease the other constituents.
River discharge could also vary over the years, which would also make the tidal range vary.
Changing frequency or size of storm surges over the years might also have a long-term effect. Thus,
there are many potential causes of the year-to-year variation of tidal constituents, along with the 18.6
year-variation. Here, however, one merely is trying to see if there is any apparent 18.6-year
variation, and if so, whether it is similar to the astronomically determined variation.
The two M2 node factor curves in Figure 2.33 are difficult to compare because of the trend in
the calculated curve (which shows an increasing M2 tidal range over the 19 years). Removing that
trend would give a minimum node factor at 1930, about 1.5 years earlier than the astronomical
minimum node factor. The N2 calculated node factor curve looks very little like the astronomical
curve, and S2 , being a solar constituent, should not show any lunar nodal variation at all (hence the
straight dashed curve), but there are obviously variations from year to year. The two M4 curves have
minima at the same year, but do not look very similar otherwise. Only the two diurnal constituents
show strong similarities between calculated and astronomical curves, with the two O1 node factor
curves showing a very good match, and the two K1 node factor curves looking very similar but with
the calculated curve lagging the astronomical curve by about a year.
There may be two reasons for the good match with the diurnal constituents. Quadratic-friction-
caused interactions among the tidal constituents have more of an effect in the semidiurnal band, but
perhaps more important may be the shape of the Delaware River and Bay estuary. The width of the
upper Delaware estuary decreases exponentially at such a rate that, if the hydrodynamic system were
67
Figure 2.33. Calculated node factors (solid line) from 19 one-year uncorrected least
squares harmonic analyses of Philadelphia water level data, plotted with the node factors
determined from astronomical (equilibrium tide) consideration (dashed line), the latter
taken from Schureman (1958). (From Parker, et al, 1999.)
frictionless, the dimensionless cutoff frequency, above which the solution is oscillatory and below
which the solution is monotonic, would fall between the diurnal and semidiurnal bands (see Section
7.4.2). With friction included the monotonic solution becomes oscillatory, but the exponential width
decrease still increases the K1 wavelength much more than the M2 wavelength, and the O1
wavelength even more. It would seem that the exponential width decrease inhibits the
hydrodynamics and thus inhibits the nonlinear interactions that would have modified the calculated
diurnal node factor curves. (See Parker, 1984)
The fact that observed node variations for the semidiurnal constituents were found to be different
than those values typically used in tide prediction (which were determined from
astronomical/equilibrium tide considerations) has been pointed out by Amin (1976) and Godin
(1986) for other waterways.
Thus, if for a particular application, one needs to obtain the highest possible accuracy in tidal
harmonic constants and/or tidal predictions, one might consider (if they have very long water level
time series) carrying out an analysis similar to the one done above for Delaware Bay above, and
using the node factor variation obtained in that way rather than from astronomically based tables like
in Schureman (1958). If one has a short station one could use the variation in node factor at a nearby
long station. However, this is only useful for self-prediction or predictions into past time periods
68
with data. For future predictions, one would have to try to extrapolate the node factor curves into
the future.
If one does have at least 19 years of data, one can also use satellite constituents (see Sections
2.2.4 and 4.1.5) in the harmonic analysis and do away with the use of node factors and the u portion
of the equilibrium arguments altogether. This will work for the future as well as the past. This will
include hydrodynamic effects on the 18.6-year variations, as well as any other long-term periodic
variation for which satellite constituents were included. The only drawback of using satellite
constituents is that it might be somewhat difficult to interpret the results. The plots such as those
shown above are easier to understand and it is easier to visualize the variation of each tidal
constituent over the 19- (or longer) year period and the possible reasons for that variation.
2.3.5 Zero-frequency Effects From Nonlinear Tidal Interaction
Tidal heights are referenced to some type of datum (see Section 2.1.1), typically the mean level
of a particular key point on the tide curve. Datums such as mean lower low water (MLW) or mean
high water (MHW) or mean tide level (MTL) are usually calculated by averaging over 19 years of
data to eliminate the lunar nodal variations . Such datums may slowly change over the years due
to a change in sea level, or due to slow vertical land movement (e.g., due to glacial rebound or
sediment compaction), the latter looking like a sea level change to the water level gauge, which is
held fast to the moving land. All the datums slowly move up and down with the long-term changes
in relative sea level.
However, in shallow-water areas there can also be a tidally induced change in "mean sea level"
due to nonlinear tidal effects. Here the "mean" results from averaging an asymmetry within a tidal
cycle caused by the nonlinear effects, rather than from a uniform shifting up or down of the water
level on which the tide propagates, but this is an effect that is included in the datum calculation. It
is therefore an effect that should be considered when assessing long-term changes in sea level in a
shallow-water area, since something as simple as dredging or shoaling can change the tidal
hydrodynamics of a waterway and then also affect the value of mean sea level (and thus all the tidal
datums). As one example, Parker (1984) used a nonlinear numerical model of the Delaware River
and Bay to show that shallow-water nonlinear effects increased the mean sea level value near
Philadelphia by an amount equivalent to 8% of the M2 amplitude.
Tidal currents do not require datums, the zero current speed serving that purpose (if there is no
mean current from a nontidal source). If there is a mean permanent current, such as due to a mean
river flow or a mean wind drift, then the tidal current oscillates about that mean flow. Or, as was
seen in Figure 2.4 (the fourth example), the entire tidal current ellipse can be shifted in the direction
of the mean flow. But here again, the mean current can also be affected by the nonlinear tidal
hydrodynamics causing an asymmetry within the tidal cycle. This can be due to not only shallow-
water effects, but even more importantly by nonlinear lateral inertia effects (that do not need shallow
water). Such tidally induced residual currents effects are discussed in Section 2.3.6e, as one of the
many phenomena that appear in tidal currents but not in the tide.
2.3.6 Special Aspects of Tidal Currents
As will be seen in Chapter 5, the same methods of tidal analysis and prediction can be used for
tidal currents, except that, since currents are vectors, the analysis is done twice, once for each of two
orthogonal components (e.g., for the major and minor axes). However, there are many more
69
difficulties with tidal current analysis and prediction than with tide analysis and prediction. Godin
(1988, 1991) describes various analysis approaches for tidal currents, and generally takes a very
pessimistic view about how accurately they can be predicted. In this book we take a much less
pessimistic view, for as long as the special aspects of tidal currents are kept in mind and accounted
for in some fashion, there is no reason why accurate tidal current predictions cannot be made for
most situations.
2.3.6a Short current data times series
In the past most current data records have been much shorter than water level data records, and
shorter time series led to fewer tidal constituents that could be resolved and thus less accurate tidal
current predictions and to the need for special methods to try to improve the accuracy of such
predictions (see Section 5.4). These shorter time series were because current measurements have
been more difficult and more expensive to acquire than water level measurements. Water level
measurements are usually made on a pier or other coastal structure, where the gauge can be easily
maintained, provided with power, connected to a real-time data acquisition system, and repaired if
there is a malfunction. Until recently, current measurements could not be made from a pier or from
land. Now with sideward-looking acoustic Doppler current profilers (ADCPs) and high frequency
(HF) radar systems, that is no longer true, although there are still some limitations in these two new
systems that make in situ current meters or bottom-mounted upward looking ADCPs still necessary.
Such systems can now be more easily maintained for long periods of times, but they are still more
expensive (in terms of operational maintenance cost) than water level gauges.
Current meters also tended to have fouling problem due to biological growth, or debris clogging
the sensor (especially when propellers or rotors were used), so that only portions of a data time
series could be used. Todays current sensors are less susceptible to fouling problems. Even
ADCPs partially covered by sand waves have been shown to work well (up to a point). (See
Section 5.3.3.)
2.3.6b Noisy current data
Current data also tends to be noisier than water level data, usually due to the effects of surface
waves (water level data are also affected by waves, but usually not to the same extent) . Wave
motion can not only show up in the current data (and have to be averaged out in some fashion), but
it can also cause various motions in the current meters themselves. This additional noise, however,
varies considerably depending on the type of instrument used to measure the currents. In past
decades current meters were hung from surface buoys that were very susceptible to the pitching and
rolling caused by waves (although some buoys were specially designed to minimize such motions).
Some of the early current meters were more susceptible to wave action, in terms of adding noise to
the data record. For example, many current meters used savonious rotors to measure the current
speed and a separate small vane to measure the current direction. The S-shaped savonious rotor
would turn no matter which way the water was flowing (unlike a propeller), so fairly high (apparent)
current speeds could be seen by a savonius rotor even during times of slack water, because the wave
action would be turning the rotor. The small size of the vane allowed it to move so easily that the
result was very noisy direction data. The savonious-rotor-small-vane combination led to serious
noise problems when there was any kind of wave action at the water surface. Vector-averaging
current meters, usually of the electromagnetic variety (but one variety had two propellers at right
70
Figure 2.34. Current flow at one moment in time through a cross section near the
Chesapeake Bay Bridge. (From Parker, 1997.)
angles to each other), did a much better job of not letting wave action influence the data. ADCPs
also break up the current vector into orthogonal components (in this case covering portions of the
water column), and also average over some time interval. Such time averaging (and spatial
averaging in the case of the ADCP) reduces higher frequency noise (including the orbital motions
of the waves). (See Section 5.3.3.)
2.3.6c Dramatic Spatial Variation in Tidal Currents Vertically
Perhaps most important, the spatial variation of tidal currents, both vertically in the water
column and horizontally (i.e., geographically), is much more complex than for tides. The tide, being
the movement of the water surface, does not have a depth variation. Tidal currents, however, vary
considerably from the surface to the bottom. Bottom friction is a major cause of this variation. The
tidal current is slowest near the bottom and faster nearer to (but usually not at) the water surface.
Usually the tidal current will turn (from flooding to ebbing, or vice versa) earlier near the bottom
than near the water surface, so there will be times when the tidal current is going in the opposite
direction at one depth than at another depth. One example of the variation in tidal current speed
across a cross section of Chesapeake Bay is shown in Figure 2.34. In this figure one sees the current
flooding (i.e., flowing up the Bay) in the deeper depths and ebbing (i.e., flowing down the Bay) near
the waters surface (except on the west side, where it is still flooding).
There are also other effects that can also make the tidal current vary vertically. Tidal currents
can be modified by baroclinic effects, that is, by density differences vertically along the water
column, due to salinity difference (in estuaries that are not well mixed) or temperature differences
(offshore in the coastal ocean), which allow the propagation of internal tide waves. Such baroclinic
effects on tidal currents are often seen in the middle of the water column but not near the bottom or
near the water surface, where frictionally caused mixing takes place. [No direct baroclinic effect is
71
usually noticeable in the tide, although in estuaries there may be indirect effects due to differences
in energy dissipation caused by stratification. In the deep ocean, surface manifestations of internal
tides have been detected (see Ray and Mitchum, 1997.)] Other types of currents, such as wind-
induced currents and river flow, also vary from the water surface to the bottom, and these currents
can nonlinearly interact with the tidal currents.
2.3.6d Dramatic Spatial Variation in Tidal Currents Horizontally
Horizontally (i.e., geographically), the tide varies quite smoothly due to the hydrodynamic
effects of such things as changing depths and widths, resonance, and Coriolis. Such variations can
often be reproduced or predicted with even simple analytical models. Although tidal currents are
also affected by the same hydrodynamics and can change in similar ways, there are other
hydrodynamic effects which can make tidal currents change dramatically in the horizontal direction
over surprisingly short distances. One example is the tidal current in a navigation channel compared
with the tidal current in the nearby shallows. The tidal current is much faster in the deeper channel
than in the shallows, and the times of slacks and of maximum floods and ebbs can be quite different
than those in the nearby shallows. One does not require a dramatic change in depth to see
Figure 2.35. The spatial variation in tidal current near the surface to the west of
Bergen Point in the Port of New York and New Jersey acquired from a towed
acoustic doppler current profiler. (From Parker, 1977.)
72
differences in the tidal currents. Even in a wide bay, changes in bathymetry will affect the tidal
currents. Not only will current speeds vary with horizontal distance (as mentioned above) but the
bathymetry will steer the current. Also, currents within a channel or constricted portion of
bathymetry will tend to be more reversing than currents in a more open and flat part of the bay,
where a more rotary tidal current will be possible.
In Figure 2.34, where vertical variation in the tidal currents were seen, there were also horizontal
variation along the width of the bay. Another example with more rapid horizontal variation in the
tidal currents is shown in Figure 2.35, which presents a synoptic view (from above) of tidal current
vectors in different locations in a small region of the Port of New York and New Jersey.
In decades past when current meters were primarily hung from surface buoys, those buoys would
swing around with the changing current direction carrying the current meter to a location where the
tidal current could be different. Such a difference could be quite significant, if, for example, there
was a long mooring line and the buoy swung from over a deep channel to over shallow water next
to the channel. Bottom mounted acoustic Doppler current profilers (ADCPs) are much more stable,
but even they can be moved by storms or strong currents, and here again the tidal currents in the new
location can be slightly (or not so slightly) different. (So with current data, one must always pay
careful attention to the exact latitude and longitude of the measurements, and note when that changes
due to movement of the current meter.)
Bottom friction is often the main reason for the horizontal variation in tidal current, due to the
changing waters depth as one moves horizontally. However, there are many other causes of
horizontal variation in tidal currents, such as that discussed in the next section.
2.3.6e Nonlinear Lateral Inertial Effects On Tidal Currents
One effect which can be even more dramatic than the frictional effect (due to changing depths)
is found where there is a bending waterway, a channel bend, a point of land sticking out into the
waterway, or some other similar geographic variation in the shoreline. Such a feature can cause the
formation of an eddy during one or both phases of the tidal current. A point sticking out into a
waterway, produces an eddy on the side of the point sheltered from the tidal current. Thus, during
the flood phase there will be an eddy on the backside of the point, and during the ebb phase there
will be an eddy on the front side of the point. This is a lateral inertial effect (see Section 7.6.7).
After the ebb phase is done, for example, the water on the backside of the point keeps moving
roughly in the same flood direction, because that location is sheltered from the opposing flood
currents by the point of land and inertia keeps the sheltered water moving.
If one harmonically analyzes current data from a location within this eddy, one will obtain a
consistent mean current, which is usually referred to as a tidally-induced residual flow (and the
process that causes it is often called tidal rectification). However, one will also see that the size of
the tidal harmonic constants is very different than those for the waters not sheltered by the point of
land, because the inherent asymmetry leads to energy being transferred to the second harmonics of
the tidal constituents. Thus, the ratio of semidiurnal-to-diurnal constituents in the tidal current will
be larger in sheltered locations than in unsheltered locations (and thus, the usually shown diurnal-to-
semidiurnal constituent ratio will be smaller). Similarly the ratio of quarter diurnal tidal current
constituents to semidiurnal tidal current constituents will also be larger, leading to distorted tidal
current curves. None of this affects the tide, and one will not see such dramatic variations in tide
constituents across the waterway.
73
Figure 2.36. Variation in (K1+O1)/M2 and M4/M2 tidal current amplitude ratios, and in tidally
induced residual currents, at three stations across the entrance to Haro Strait at a depth of 70
feet (21m) below MLLW. (From Parker, 1991a.)
This inertial effect is one of many nonlinear effects which can modify tidal currents, however,
this effect does not depend on the water depth being shallow, as most nonlinear effects do (see next
section). When looking at the 2-dimensional or 3-dimensional momentum equations (see Sections
7.3.2), it is the lateral advective/inertial terms that produce the tidally-induced residual current and
the transfer of energy to higher frequency tidal constituents. This rectification of tidal currents can
be demonstrated by a Fourier decomposition of the inertial terms (see Section 7.6.7) and as
mentioned above is a result of the inherent asymmetry in the problem.
Figure 2.36 shows a good example of these inertial effects on tidal currents, for the region where
the Strait of Juan de Fuca meets Haro Strait (between Canada and the U.S.). Both are deep
waterways, but they meet at an angle, namely the waterway bends significantly to the north near
Victoria, BC. Thus, the flood current in the Strait of Juan de Fuca flowing past Victoria cannot
make a sharp left turn because inertia keeps it moving eastward, allowing the southerly ebb current
at the western-most current station in Haro Strait to keep moving longer than at the current station
on the eastern side of the waterway. Figure 2.36 shows analysis results from three NOS current
stations across the entrance to Haro Strait (entering from the Strait of Juan de Fuca). The
74
(K1+O1)/M2 amplitude ratio in the tidal current varies dramatically, decreasing from 1.21 on the
eastern side of the waterway to 0.37 on the sheltered western side (while for the tide the (K1+O1)/M2
ratio does not change much, only from 2.25 to 2.50). The M4/M2 ratio for the tidal current also
varies significantly across the waterway, from 0.075 on the eastern side to 0.314 on the western side.
2.3.6f Nonlinear Shallow-water Effects On Tidal Currents
The shallow-water nonlinear processes that affect the tide (see Section 2.3.2) also affect the tidal
currents, and in many situations those shallow-water effects are seen to be more dramatic in the tidal
current than in the tide. When one considers distorted tidal currents one usually speaks of flood
dominance or ebb dominance. The asymmetric tidal current can have: (1) a shorter flood phase with
higher speeds and a longer ebb phase with slower speeds (called flood dominance because of the
higher flood current speeds); (2) a shorter ebb phase with higher speeds and a longer flood phase
with slower speeds (ebb dominance); or (3) the case midway between (where one of the slacks can
last for a couple of hours). Asymmetry in the tidal current is important in the transport of sediment
and pollutants. The transport of coarse sediment depends on the maximum speeds achieved, and so
might be transported up an estuary with a flood dominant situation. Fine sediment stays suspended
except near slacks, so the case with longer slacks before ebb might lead to deposition at that time.
(See Speer, et al, 1991, for more discussion on flood and ebb dominance.)
The distortion in tidal currents can be greatly enhanced by a strictly linear superposition effect,
that in fact, will not similarly enhance the distortion in the tide. Parker (1991a) shows an extreme
example in Ramshorn Creek, a small shallow channel connecting the shallow Cooper and New
Rivers, both part of the intracoastal waterway in South Carolina and both connected to the Atlantic
Ocean (Figure 2.37). Two tide waves, each distorted by shallow water, enter Ramshorn Creek at
Figure 2.37. Current speed (top panel on right) and direction (lower panel) plots of the unusual
occurrence of dominant quarter-diurnal tidal currents at an NOS current station in Ramshorn
Creek, SC. There are four cycles per day in the tidal currents (see text). (From Parker, 1991a.)
75
opposite ends and cross, superimposing their effects. The M2 flood currents for the two waves are
in opposite directions, and so the superposition of the two waves leads to a reduction in the M2 tidal
current. Each tide wave is distorted to a different degree because the two waterways leading to
Ramshorn Creek do not have exactly the same width or depth. Thus, there is a different 2M2o-M4o
phase relationship, so that when the M4 tidal constituent waves are superimposed the two waves add
together (rather than cancel each other out like the M2 constituent waves) thus increasing the M4
tidal current. For this particular situation the result is the very unusual situation of dominant
quarter-diurnal tidal currents (Figure 2.37), that is, the tidal current actually changes directions eight
times a day instead of four (i.e., there are four tidal current cycles per day instead of two) There are
other waterways in the salt marsh estuaries of South Carolina and Georgia that also exhibit the
effects of two tide waves entering from opposite ends and crossing. Most are not as extreme as in
Ramshorn Creek, but the result is very distorted tidal current curves. (This superposition effect does
not lead to a dominant M4 in the tide because the two M2 waves add positively at the crossover point,
not negatively as with the tidal current.)
2.3.6g Hydraulic Tidal Currents
There is another type of tidal current which involves the crossing of two tide waves from the
opposite ends of a waterway. In this case the waterway is a short and narrow strait, and there is
often a significant difference in tide range at each end of the strait and/or a difference in timing in
the high waters at the two ends. For example, the very fast currents in Seymour Narrows (noted in
Section 2.1.1) are due to tidal height differences between the Strait of Georgia end and the Queen
Charlotte Strait end, the tide being approximately 180o out of phase at opposite ends. Tidal currents
in such straits, which can have very high speeds, were given the name hydraulic currents. Since in
the past it was difficult to maintain a current meter where there were very fast currents (now one can
put an upward-looking ADCP successfully on the sea bottom or a sideward-looking ADCP on a
pier), another method to predict the tidal currents in a strait was devised which used the predicted
water level differences at the opposite ends of the strait.
Two often cited examples in the U.S. of hydraulic currents driven by tidal height differences at
opposite ends of a strait are Deception Pass, Washington (between Rosario Strait and Reservation
Bay), where the tidal currents reach speeds over 8 knots, and the East River (connecting New York
Harbor and Long Island Sound), where the tidal current speed reaches over 5 knots at Hell Gate,
about half way between the entrances. These two stations, and four others with hydraulic currents
(Cape Cod Canal, Chesapeake and Delaware Canal, Sergius Narrows, and Isanotski Strait) still
appear as reference stations in the U.S. Tidal Current Tables.
The daily predictions for each of these six hydraulic current stations are based on harmonic
constants that were derived from the harmonic constants at the tide stations at the opposite ends of
each of the straits (and in some cases were modified based on a small amount of actual current data).
Differences in the water level at opposite ends of the strait are partly due to any difference in the
tidal range at the two ends and partly due to any difference in the time of the high waters at the two
ends. In the method described in The Manual of Current Observations (U.S. C&GS , 1950, S.P.215,
pp78-83), the current speed is assumed to vary as the square root of the water level difference. By
the reasoning of this method, the current speed will be a maximum when the difference in water
level is greatest, and will be zero (slack current) when the water levels are the same at opposite ends.
However, this ignores friction and inertia (and other details of the actual hydrodynamics that would
occur in the strait), so there will usually be a lag in the response of the currents due to the difference
76
in the water levels at opposite ends. This lag was usually determined by obtaining a short record
of current data by whatever technique could work for a few tidal cycles in the fast currents.
This method was developed in the pre-computer era and before the numerous modern methods
which can be used today to measure fast currents in a strait, such as an upward-looking ADCP on
the sea bottom, a sideward-looking ADCP on a pier, or HF or microwave radar. It is no longer
necessary, and was only a reasonable approximation. If one is interested in it, one can look at
S.P.215.
2.3.6h Limitations On Tidal Current Predictions
Because of these various effects on tidal currents (that do not similarly affect the tide), there are
several limitations that should be remembered about tidal current predictions, which will be
discussed below.
Tidal current predictions are only accurate for the exact depth and location where the current
data were obtained. Such predictions may be usable at nearby locations (preferably at the same
depth), but only if the bathymetry does not change dramatically.
It is also very important to remember that bathymetry and shoreline can slowly change over the
years, which means that the tidal currents can also slowly change over the years (while in most cases
the tide will not similarly change over the years). Thus, for shallow-water areas (i.e., for many bays
and estuaries), new current measurements may have to be made and new tidal current harmonic
constants calculated, because the bathymetry has changed (while the harmonic constants for the tide
are still perfectly acceptable). Dramatic changes in bathymetry, such as due to dredging or rapid
shoaling or due to the dramatic movement of sediments after a large storm, can make the predictions
in the Tidal Current Tables invalid (or at least much less accurate) for any locations within the area
of the dredging or shoaling or storm movement. Even for this extreme situation the tide predictions
may not be affected at all, unless the bay or estuary is very small and the changes due to the
dredging or shoaling represent a significant portion of the total volume of the waterway. (For
example, dredging in a small bay can increase the tide range.)
As will be discussed in Section 3.6.1, since many current stations have short data time series
tidal current predictions at these stations are often produced using time differences and amplitude
ratios (found in Table 2 of the Tidal Current Tables), calculated through a nonharmonic comparison
analysis. For a nonharmonic comparison tidal analysis (and the resulting predictions) to be
reasonably accurate, it is crucial that the reference and subordinate stations have very similar
frequency domain characteristics, that is, the ratios of diurnal-to-semidiurnal constituents should be
very similar, as well as the ratios of quarterdiurnal-to-semidiurnal constituents. From what has been
seen in this section, there can be significant variation of tidal currents over a geographic region,
making it more difficult for all these short (subordinate) stations to have the same frequency
characteristics as the reference station(s) being used. For example, in the waterway shown in Figure
2.36 and other waterways nearby, there is such a dramatic variation over geographic distance in the
diurnal-to-semidiurnal ratios in the tidal currents, that it is difficult to have enough reference stations
to handle all the subordinate stations accurately.
Another bad situation arose sometimes in the past, when the lack of a long-term current data time
series led to the use of a tide station as the reference station for tidal current stations. However,
when one wishes to compare the tidal current to the tide one must remember that basic linear
hydrodynamics determines that the tidal characteristics can be very different for tidal currents and
tides at the same location. For example, the quasinode for a constituent of the tide will occur 1/4
77
of a wavelength from the head of bay (where the reflection of the tide wave takes place), where as
for a constituent of the tidal current the quasinode will occur of a wavelength from the head of
bay (see Section 7.4.1). Thus, near the M2 quasinode of the tide, where the diurnal effect is therefore
stronger and the tide may be mixed or diurnal, the tidal currents will still be semidiurnal. This
occurs in the eastern end of the Strait of Juan de Fuca near Victoria, British Columbia (see Parker,
1977).
All these various geographic and bathymetric effects can also cause the directions of the
maximum flood current and the maximum ebb current at a particular location to not be directly
opposite of each other, that is, to not be 180o apart. The direction difference between maximum
flood and maximum ebb tidal currents will depend on exactly where the current meter is positioned.
If placed in a straight section of channel, then flood and ebb directions will be 180o apart. However,
if the current meter is placed at a bend in the channel, the difference between flood and ebb
directions can be very different than 180o. When this latter situation occurs, and one harmonically
analyzes the tidal current data, there will be an apparent mean current and an artificially large
overtide, i.e., M4, that comes out of the analysis of the two orthogonal components (see Section 5.1).
Because of all these mechanisms that can cause significant variation in tidal currents over depth
and horizontal/geographic distance, this is an area of tidal prediction where (high-resolution)
numerical hydrodynamic models are most useful, and may actually do a better job than obtaining
many current records and obtaining harmonic constants for each. (See Section 8.3). Given that
slightly moving a current meter can change the harmonic constants significantly, and not knowing
the location where the current meter would be most representative of the local area, it may be better
to rely on the predicted tidal currents over the small area of a grid cell in a carefully calibrated
numerical model (that is forced at the entrance with very accurate tide predictions). That being said,
one must remember that there can be problems with how well a model might represent the currents,
especially if it does not have high enough spatial resolution. Thus, for locations with uniform
geography and bathymetry for which one has a long data time series of high quality current data, it
will usually be better to rely on harmonic analysis of the current data and the subsequent prediction
of the tidal current from the calculated harmonic constants.
There is another aspect of current data that does not apply to water level data. In current data
one is more likely to find energy at frequencies close to tidal frequencies that is not caused by the
tide. In the summer in many coastal locations one will see a land breeze-sea breeze, where the name
comes from that fact that generally the wind blows toward the land during the day (the landward
wind replacing rising heated air over the warmer land surface) and toward the sea at night (the
opposite situation), although the change in direction of flow can in many locations actually rotate
around the compass over the 24-hour period. (This is mainly due to the Coriolis effect, so rotations
of the land breeze-sea breeze are clockwise in the Northern Hemisphere, although the orientation
and shape of the coastline can often significantly affect this.) This changing wind speed and
direction has a 24-hour period and thus would show up in a harmonic analysis as S1. To notice this
effect, however, one would need a year-long time series to separate this meteorological S1 from K1.
Since in many locations the land breeze-sea breeze usually only occurs in the summer, the S1 value
would be very different in the summer than in the winter. One should remember this effect when
one is analyzing current data obtained during the summer, because it may have a noticeable effect
on the K1 value that comes out of a harmonic analysis.
Along these same lines (that is, of energy from a nontidal source occurring at a frequency near
to one of the tidal frequencies), at some latitudes one must also consider the possibility of inertial
currents. Inertial currents are caused by Coriolis effects on the wind. They are oscillatory currents
78
that rotate in a clockwise direction in the Northern Hemisphere (and counterclockwise in the
Southern Hemisphere), and they have energy at a frequency that depends on latitude. The formula
for the period of an inertial current is Ti = 11.97 / sin , where the numerator is half a sidereal day
in hours and is the latitude in degrees. While generally transient and usually lasting only a few
days, they may last long enough to affect a tidal current analysis, at least in locations where the
inertial frequency is at or close to the frequency of a particular tidal constituent. Thus, in the Arctic
Ocean (or in the Ross Sea and Weddell Sea next to Antarctica) inertial currents may affect the
calculated M2 harmonic constants, because the inertial period at 75oN is the same as the M2 period.
At latitude 30oN inertial currents will have their energy at exactly the K1 tidal frequency, and at
27.6oN the inertial energy will be at exactly the frequency of O1, and thus they might affect the
calculations for these diurnal constituents in large parts of the Atlantic or Pacific Oceans.
Thus, as has been seen, there are many aspects of tidal currents that make them more difficult
to analyze and predict accurately than the tide, but keeping these problems in mind, one should be
able to produce useful results. The next chapter will look at the various methods of time series
analysis that can be used for this purpose.
79
80
3. Methods of Tidal Analysis and Prediction
3.1 Introduction
The prediction of tides and tidal currents is unique in the world of geophysical prediction on
Earth. There is no other geophysical phenomenon where all its energy is known a priori to be found
only at very specific frequencies which as explained in the last chapter, are determined by the
relative astronomical movements of the Earth, moon, and sun in their various orbits, rotations, and
axis tiltings. For other geophysical systems, notably weather (and on a longer time scale, climate)
only a small portion of the energy is found at known frequencies primarily the daily cycle (heating
during the day and cooling at night) and the annual cycle (more heating in the summer and less in
the winter). Like the tides, these cycles are astronomical in origin, but unlike the tides, the complex
physics of weather and climate systems is so chaotic that the daily and annual cycles are only a part
of the whole story. Although tidal systems can also be very nonlinear in shallow waterways, they
are not chaotic. The nonlinearity of the tides simply transfers energy from some (astronomical) tidal
frequencies to other new tidal frequencies, which in the time domain show up as distorted tide and
tidal current curves and asymmetries between the flood phase and the ebb phase. The tides are a
periodic deterministic process, whereas the processes affecting weather are stochastic processes that
involve random and aperiodic fluctuations.
[In shallow waterways nonlinear processes also cause an interaction between the (deterministic
periodic) tide and the (stochastic random/aperiodic) weather-induced effects (such as river discharge
or storm surges). The tide can be temporarily changed by these nontidal effects (through this
nonlinear interaction; see Sections 2.3.3, 7.6.3, and 7.6.4), thus bringing a stochastic element into
the analysis and prediction of tides. When dealing with random processes one usually is concerned
with whether the process (data) is stationary, meaning that the average properties computed over
short intervals do not vary significantly from interval to interval. When the tide is nonlinearly
modified by random meteorological effects in shallow water, the tide is no longer stationary. But
a random (usually small) modification to strongly periodic deterministic process like the tide, does
not make the tide a stochastic process in the same sense as other geophysical phenomena, and the
methods for dealing with these nonlinear modifications require a different approach than might be
used with other nonstationary processes.]
Tidal data analysis and tidal prediction are a special subcategory of time series analysis (and a
further subcategory of statistics). As with any other time series of data, water level data and current
data can be looked at in the time domain and in the frequency domain. In the time domain one looks
at actual real-world measured data as well as tide and tidal current predictions the water level
heights or current speeds and directions plotted against time. In the frequency domain one looks at
plots showing how the energy found in the data time series is spread over different frequencies. As
was seen in Section 2.2.3, when one looks at a frequency domain plot resulting from the spectral
81
analysis of the time series, one again sees the uniqueness of the tides, because the tidal energy is
represented by a series of spikes at only a number of specific frequencies. If the time series analyzed
is fairly short one will see spikes at semidiurnal and diurnal frequencies (and at higher harmonics
of these frequencies if the data came from a shallow waterway), each spike trailing off from a peak
in the middle of the band. However, for a longer time series (which allows finer resolution along
the frequency axis of the spectral plot), one can see the actual tidal spectral lines for the various tidal
constituents. Between these tidal spikes is the rest of the spectrum, resulting from the assortment
of nontidal phenomena that can make the ocean move up and down (or horizontally), but these
effects do not have most of their energy concentrated in a few frequencies. In a plot of a water level
spectrum, the energy from winds, atmospheric pressure, and temperature and salinity changes shows
up primarily as a smooth continuous curve covering a broad range of frequencies, called the
continuum, ranging from wind waves at the higher frequencies to low-frequency storm surges at the
lowest frequencies (see Figure 2.16). The tidal spikes or lines rise above this continuum. The
continuum curve at a water level station near the coast or in a bay is often highest near zero cycles
per day (where zero cpd represents the mean), with the very low frequencies near zero cpd
representing very slow changes in the height of the water surface due to the effects of wind,
atmospheric pressure, density changes, and/or river discharge.
As was mentioned in Section 2.2.3, one can use spectral analysis for tidal analysis, but the results
merely give us an idea of how strong the tidal signal is, the relative importance of the diurnal and
semidiurnal bands (and thus information on the type of tide), and whether there are strong higher
harmonics due to shallow water. The spectral results are of no use for tidal prediction, for they
provide no phase information, that is, there is no information about the timing of these effects. Since
one already knows at exactly which frequencies the tidal energy will be found (because of its
astronomical origin), there are much better ways to analyze a water level record or a current record
in order to both characterize the tidal signal, and more importantly, to be able to predict the tide or
tidal current using the results of the analysis.
3.2 Simple Demonstration of Extracting Tidal Constituents From a Data Time

Series
The general methodology for extracting tidal constituents from a data time series, taking
advantage of our a priori knowledge of the frequencies where the tidal energy will be found, can
be seen with a simple demonstration. If one would like to see the contribution to the tide of, for
example, the M2 tidal constituent, which has a period of approximately 12.4206 hours, one simply
takes a water level time series and breaks it up into consecutive pieces, each 12.4206 hours long.
One then superimposes these 12.4206-hour-long pieces (i.e., adds them up) and averages them. For
each of these pieces the M2 contribution will be in sync, that is, the maximum M2 contribution will
be at the same time within each piece of the chopped up data time series, but the other tidal
constituent frequencies will not stay in sync. For S2, which has a period of 12.0000 hours, the time
of the maximum S2 contribution to the tide (within each 12.4206-hour piece) will slowly shift earlier
as one moves from piece to piece. This is illustrated in Figure 3.1. The top plot in this figure shows
the contributions of M2 and S2 for the first 12.4206-hour piece. The M2 maximum and the S2
maximum start out in sync, but by the end of the second segment the S2 maximum is slightly to the
left of the M2 maximum. In the next plot below it, showing the next consecutive 12.4206-hour piece
from a couple of days later, the S2 maximum has shifted even more to the left. By the fourth plot
down (about 7.38 days after the first data point), the S2 maximum occurs at the same time as the M2
82
Figure 3.1. A vert simple demonstration of extracting tidal constituents from a water level data
time series (see text).
minimum. After enough pieces (about 14.76 days after the first data point), the S2 maximum will
come back to the same time (within the piece), and again be in sync with the M2 maximum.
When one adds up all the 12.4206-hour pieces of the data time series, the S2 portion of the tide
will cancel itself out. The same thing will happen with the other tidal constituents all that will be
left will be the M2 contribution. For the moment it has been assumed that one has a very long time
series so the contributions of all the other tidal constituents will cancel out. The result of averaging
all these M2-period-long pieces is an average M2 cycle, from which one can obtain the M2 amplitude
(half the M2 range, i.e., half the height difference between the M2 high water and the M2 low water).
One can also obtain the M2 phase lag, which is the time of the maximum M2 high water from the
beginning of the mean curve (which itself will have a time relationship to some other time reference
point, such as the time of the moons transit over a particular time meridian).
So far it has been assumed that one has a very long data times series, so that all the other
constituents would cancel themselves out over that long period of time. However, when one does
not have such a long data record, one may not want to use the entire length of the record when
extracting (in this case) the M2 signal. For example, in the above procedure, one needs a 14.76-day
long time series of data in order to cancel out the S2 tidal constituent. This 14.76-day length of
series is determined by the difference in the periods of M2 and S2. The difference between the M2
83
period (12.4206 hours) and the S2 period (12.0000 hours) is 0.4206 hours, and thus the maximum
S2 contribution shifts to the left in each successive piece of the time series by 0.4206 hours. It
therefore takes (12.0000/0.4206=) 28.53 M2 cycles (or 14.765 days) for the S2 maximum to sequence
through completely (as was seen in Figure 3.1). If one had a time series that was 20 days long, one
would still only want to use 14.76 days of that time series, because using the entire 20 day would
leave in some of the S2 contribution. One could, however, use 29.53 days, or any multiple of 14.76
days.
The closer in tidal period a tidal constituent is to M2, the longer the data times series must for
its effects to be completely canceled out. The period of N2 is 12.6583 hours. The difference
between the periods of M2 and N2 is 0.2377 hour. This is smaller than the difference between the
period of M2 and S2, so it will require a longer time series to cancel out the N2 contribution. In this
case it requires 27.560 days long (12.6583/0.2377= 53.25 M2 cycles), or some multiple of that. The
period of K1 (23.9345 hours) is not close to that of M2, and thus only 1.1 days of data (or some
multiple of that) are needed to cancel out the effect of K1. The period of K1 is closer to the period
of O1 (25.8193 hours) and so 13.66 days are needed to cancel out the O1 effect when calculating K1
and vice versa.
One can obtain the contribution of another tidal constituent by going through the same
procedure, but choosing a different time period for chopping the time series up into pieces. For S2,
one would chop the time series up into consecutive pieces, each with a length of exactly 12 .00 hours
long. For N2, one would chop the times series up into pieces, each with a length of 12.658 hours
long. And so on. When doing this procedure for K1 (23.094 hours), the two closest important tidal
constituents are O1 (25.819 hours) and P1 (24.066 hours). For O1 therefore, 13.66 days of data (or
some multiple) are required to cancel out the O1 contribution when trying to extract K1. For P1, since
it is much closer to K1, 182.6 days of data (or some multiple) are required.
3.3 The Synodic Period Length of Time Series Needed To Separate Two
Constituents
The time (typically) required to separate the effects of two nearby tidal constituents is called the
synodic period, and it is defined as the interval between two consecutive conjunctions of phase of
the two constituents (Schureman, 1940, p 51). This time period includes N oscillations of the lower
frequency constituent wave and N+1 oscillations of the higher frequency constituent wave. This
criterion for the minimum length of series required to resolve two constituents is also called the
Rayleigh criterion (Godin, 1972), because it was first proposed by Lord Rayleigh in optics, dealing
with the criterion for resolving two adjacent frequency components when light is shone on a
diffraction grating. Although the demonstration in Section 3.2 was carried out in the time domain,
the most straight forward way to calculate the synodic period , Tsyn , is simply to find the inverse of
the difference in the frequency between two tidal constituents. Thus, the difference in the M2 and
N2 frequencies is 0.0015 cycles per hour, so the synodic period for M2 and N2 is Tsyn = (1/.0015)cph =
661 hours = 27.560 days. Or likewise, in the classical notation of angular speeds, the synodic period
is 360o divided by the difference in the angular speeds of the two tidal constituents. Another way
of stating Rayleighs criterion is: to resolve two frequencies their difference must be greater than
the inverse of the length of the data time series analyzed, i.e.,
|2 ! 1| > T -1
84
Table 3.1. Schuremans Table 39 showing the synodic periods for a number of pairs
of semidiurnal and diurnal tidal constituents. (From Schureman, 1958.)
or, in other words, the length of the time series must be greater than (or, in reality, approximately
equal to) the synodic period, i.e., T > Tsyn .
Schuremans Table 39 (reproduced here as Table 3.1) shows the synodic periods for a number
of pairs of semidiurnal tidal constituents (bottom table) and for a number of pairs of diurnal tidal
constituents (top table). (There is no need to consider the synodic period between a semidiurnal
constituent and a diurnal constituent, since it will be on the order of a day.) For a particular
constituent, the most important synodic period will the longest synodic period with another tidal
constituent that is expected to have a larger amplitude than it. For example, in Table 3.1, one sees
that to separate P1 from J1, K1, M1, O1, and OO1 (the fifth row of the diurnal table) one needs data
times series of lengths 23.942, 182.621, 32.451. 14.765, and 12.710 days, respectively, and to
separate P1 from Q1, 2Q1, S1, and 1 (the sixth column of the diurnal table), one needs data times
series of lengths 9.614, 7.127, 365.243, and 10.085 days, respectively. But because K1 is the diurnal
constituent with the largest amplitude, it is the 182.632-day synodic period that is important. If one
has less than 183 days of data, according to the Rayleigh criterion (but see caveat below) one could
analyze either for K1 or for P1, but not both. One would naturally choose K1 because it will always
have a larger amplitude than P1. But the K1 harmonic constants (amplitude and phase lag) that come
out of such an analysis will include an error caused by the effect of the P1 energy that could not be
separated out. Table 3.2 shows the synodic period (fourth column in that Table) for each tidal
constituents with respect to a larger tidal constituent from which it must be separated.
85
In both Tables 3.1 and 3.2 one notices that once there is at least a month of data there are no
significant astronomical tidal constituents that adversely affect M2 or O1. However, for the rest of
the primary tidal constituents there are interfering tidal constituents with half-yearly and yearly
synodic periods. Thus, unless one has (approximately) a half year of data, K1 will be affected by
P1, and S2 will be affected by K2, and N2 will be affected by 2. Unless one has (approximately) a
year of data K1 will be affected by S1, and S2 will be affected by both R2 and T2.
If one was interested in separating only two tidal constituents the minimum length of the time
series would be straightforward, i.e., it must have a length equal to the synodic period for those two
constituents. But since one needs to correctly separate many pairs of constituents, all with different
synodic periods, the key question is what should the length of a time series be in order to separate
the most tidal constituents? For a particular data time series one could use a length that is the best
compromise for the most important tidal constituents. This involves not just the longest synodic
period found in Table 3.1 for key tidal constituents, but also multiples of other key synodic periods.
For example, in Table 3.2 one sees some natural groupings near 15 days, 29 days, a half year, and
a year. Schureman suggests that 369 days should be considered as a standard length for harmonic
analysis because it is supposed to conform very closely to multiples of the synodic periods of
practically all of the short-period constituents. In practice, however, and especially if one is using
a least squares harmonic analysis technique (see Section 3.4.5) 365 days of data should work fine
(Schureman used a Fourier series technique).
However, a number of authors have felt that the Rayleigh criterion is overly restrictive. Munk
and Hasselman (1964) suggested that the length of the time series determined by the Rayleigh
criterion could be shortened if there was not much noise in the data time series (either from
meteorological effects of from the instrument itself). In fact, they specifically suggested that the
required data series length can be reduced by the ratio of the signal-to-noise standard deviation, and
they referred to this as super resolution. Thus, Munk and Hasselman replaced the Rayleigh
criterion with another relation. They said that meaningful information can be gained about the
frequencies, 1 and 2, provided that
|2 ! 1| > T -1
(signal/noise level)
This seems to imply that for a signal-to-noise ratio of 1 that the classic Rayleigh criterion is a
good rule. It also seems to imply that if one has (e.g.) a water level record with a very strong tidal
signal then one may be able to extract more tidal constituents than would be expected for the length
of time series available. The extreme example would be a time series produced by a model forced
at the entrance by only tidal predictions, which does not contain any noise. In this case it may be
possible to use much shorter records than specified in Table 3.1 (see Foreman and Henry, 1989).
Similarly, if one has a weak tide signal, one may not be able obtain good harmonic constants even
if one has a record length that is longer than the synodic period. It also implies that for a data record
with some noise a large tidal constituent might be obtained very accurately, but a small constituents
down in the nontidal noise might be determined less accurately (or not at all). This would be a
separate effect from this small constituent being overwhelmed by meteorological fluctuations that
happened to fall at its particular frequency, yet it might not be distinguishable from that effect, since
one would expect the noise closest in frequency to the tidal line to have the most effect. Either way,
the only good way to check the quality of harmonic constants (if one has enough data available) is
86
to analyze several different data sets (with the same data length) and see how consistent the
amplitudes and epochs are from data set to data set.
However, there still seems to be a lot of room for interpretation. Pugh (1987) suggests that in
practice, the Rayleigh criterion is a good guide for tidal analyses of continental shelf data from
middle and high latitudes, but finer resolution is feasible in ideal conditions such as tropical oceanic
sites. There is also experience to show that the nature of the least squares technique used in most
modern harmonic analysis programs sometimes allows one to use time series lengths shorter than
that determined by the Rayleigh criterion. In some cases, for example, P1 has been reasonably
accurately determined from length of series a month or two shorter than the six months suggested
by the Rayleigh criterion, but this has not happened for all such cases, and it may possibly be
affected by the relative positions of the K1 and P1 maximums in the particular time series used. (See
also Godin, 1972, Foreman, 1977, and Jay and Flinchem, 1999).
A reasonable approach is to initially use the synodic period (Rayleigh criterion) as a guideline
when carrying out a harmonic analysis. Then, especially if ones time series is a little shorter than
one of the key lengths of series (see Table. 3.2), rerun the harmonic analysis with one or more
additional constituents included that one should be able to obtain with the next highest length of
series.
The procedure described in the simple demonstration in Section 3.2 is not a very efficient one,
especially when many tidal constituents have to be resolved from each other, and as will be seen
there are more sophisticated ways to accomplish this. Each of the tidal analysis methods described
in the following sections has a different technique for extracting the amplitude and phase lag (epoch)
of each tidal constituent. But the above demonstration provides the basic idea behind most tidal
analyses whose results are presented in the frequency domain, and the Rayleigh criterion (at least
as a useful guide) is true for all tidal analyses in the frequency domain the longer the data time
series the more tidal constituents that can be accurately resolved, that is, whose effects can be
separated from each other. Another way of saying this, the closer in frequency two important tidal
constituents are, the longer the time series that is needed to resolve them.
Although the most accurate tidal analysis methods will deal with the frequency domain (that is,
they will produce results for particular tidal frequencies), there are also tidal analysis methods that
deal only with the time domain. Such nonharmonic methods simply compare the data time series
at one location (station) with the data time series at another. This is usually done only for key
recognizable points in the time series, such as high waters and low waters on a water level curve,
or maximum floods, maximum ebbs, and slacks waters (or minimum flows) on a tidal current curve.
Thus, the time of high water at a subordinate station will be calculated as occurring a certain amount
of time after the corresponding high water at a reference station, and being a certain number of feet
higher or lower than the high water at the reference station. The results of such nonharmonic
methods are used for stations listed in Table 2 of the Tide and Tidal Current Tables. Such time
domain methods were historically the first kinds of tidal analysis used. The earliest methods used
the time of each of the moons transit overhead as the time reference points, a certain number of
hours after which high water would be calculated to occur. [A nonharmonic comparison method is
also the basis for how tidal datums are computed at a short-term station and referenced to a 19-year
period (see CO-OPS, 2003).]
In the following sections various methods will be described that are used for tidal analysis and
prediction. The most commonly used methods are based on the theory of harmonic analysis, not
much different from that developed by Thomson, Ferrel, and Darwin in the late 1800s, except that
the Fourier analysis solution technique (Schureman, 1958) has generally been replaced by a least
87
NL
Tidal Shallow- mechanism Angular Synodic Cartwright Amplitude
Harmonic Origin of water sym or Speed Freq. Period Period wrt Potential at
Constituent Constituent Cartwright No equivalent asym (o/hour) (cpd) (hours) (days) const. Coeff. Trenton
.
M2 lunar 2 0 0 0 0 0 KO2 asym 28.9841042 1.9323 12.4206 --- 0.90809 3.547
M4 shallow 4 0 0 0 0 0 M4 asym 57.9682084 3.8645 6.2103 0.5 M2 0.517
M6 shallow 6 0 0 0 0 0 M6 sym 86.9523127 5.7968 4.1402 0.5 M4 0.266
M8 shallow 8 0 0 0 0 0 M8 both 115.9364169 7.7291 3.1052 0.5 M6 0.120
K1 luni-solar 1 1 0 0 0 0 MO1 asym 15.0410686 1.0027 23.9345 1.1 M2 0.53011 0.349
S6 ** shallow 6 6 -6 0 0 0 S6 sym 90.0000000 6.0000 4.0000 4.9 M6 0.005
O1 lunar 1 -1 0 0 0 0 MK1 asym 13.9430356 0.9295 25.8193 13.7 K1 0.37694 0.288
OO1 lunar 1 3 0 0 0 0 16.1391017 1.0759 22.3061 13.7 K1 0.01624 0.030
2MK3 (MO3) shallow 3 -1 0 0 0 0 2MK3 (MO3) sym (asym) 42.9271398 2.8618 8.3863 13.7 MK3 0.120
MK3 (2MO3) shallow 3 1 0 0 0 0 MK3 (2MO3) asym (sym) 44.0251729 2.9350 8.1771 13.7 2MK3 0.116
2Q1 lunar 1 -3 0 2 0 0 12.8542862 0.8570 28.0062 13.8 O1 0.0955 0.028
S2 solar 2 2 -2 0 0 0 KP2 asym 30.0000000 2.0000 12.0000 14.8 M2 0.42248 0.461
2SM2 shallow 2 4 -4 0 0 0 2SM2 sym 31.0158958 2.0677 11.6070 14.8 S2 0.025
88
MS4 shallow 4 2 -2 0 0 0 MS4 asym 58.9841042 3.9323 6.1033 14.8 M4 0.148
S4 shallow 4 4 -4 0 0 0 S4 asym 60.0000000 4.0000 6.0000 14.8 MS4 0.005
M1 lunar 1 0 0 1 0 0 NO1 asym 14.4920521 0.9661 24.8412 27.3 K1 0.02964 0.027
M3 lunar 3 0 0 0 0 0 43.4761563 2.8984 8.2804 27.3 MK3 0.01188 0.034
N2 lunar 2 -1 0 1 0 0 28.4397295 1.8960 12.6583 27.6 M2 0.17386 0.553
Q1 lunar 1 -2 0 1 0 0 13.3986609 0.8932 26.8684 27.6 O1 0.07217 0.022

J1 lunar 1 2 0 -1 0 0 15.5854433 1.0390 23.0985 27.6 K1 0.02964 0.016
MN4 shallow 4 -1 0 1 0 0 MN4 asym 57.4238337 3.8283 6.2692 27.6 M4 0.176
L2 lunar 2 1 0 -1 0 0 2MN2 sym 29.5284789 1.9686 12.1916 31.8 S2 0.02567 0.409
2 lunar 2 -2 2 0 0 0 2MS2 sym 27.9682084 1.8645 12.8718 31.8 N2 0.02776 0.219
Mm lunar (*met) 0 1 0 -1 0 0 MN asym 0.5443747 0.0363 661.3092 31.8 MSF 0.08254 0.124
P1 lunar 1 1 -2 0 0 0 SK1 asym 14.9589314 0.9973 24.0659 182.6 K1 0.17543 0.110
K2 luni-solar 2 2 0 0 0 0 30.0821373 2.0055 11.9672 182.6 S2 0.11498 0.094
MSf lunar (*met) 0 2 -2 0 0 0 MS asym 1.0158958 0.0677 354.3671 182.6 Mf 0.01369 0.186
Mf lunar (*met) 0 2 0 0 0 0 KO asym 1.0980331 0.0732 327.8590 182.6 MSf 0.15647 0.132
2 lunar 2 -1 2 -1 0 0 28.5125831 1.9008 12.6260 205.9 N2 0.03302 0.210
2 lunar 2 1 -2 1 0 0 29.4556253 1.9637 12.2218 205.9 L2 0.00670 0.102
2N2 lunar 2 -2 0 2 0 0 2NM2 sym 27.8953548 1.8597 12.9054 205.9 2MS2 0.02301 0.042
2N2 lunar 2 -2 0 2 0 0 2NM2 sym 27.8953548 1.8597 12.9054 205.9 2MS2 0.02301 0.042
1 lunar 1 -2 2 -1 0 0 13.4715145 0.8981 26.7231 205.9 Q1 0.01371 0.013
Sa solar (*met) 0 0 1 0 0 0 0.0410686 0.0027 8765.8211 365.2 Ssa 0.01156 0.430
Ssa solar (*met) 0 0 2 0 0 0 0.0821373 0.0055 4382.9052 365.2 Sa 0.07281 0.169
S1 solar (*met) 1 1 -1 0 0 0 15.0000000 1.0000 24.0000 365.2 K1 0.00416 0.062
R2 solar 2 2 -1 0 0 -1 SK2 30.0410667 2.0027 11.9836 365.3 S2 0.00355 0.028
T2 solar 2 2 -3 0 0 1 29.9589333 1.9973 12.0164 365.3 S2 0.02476 0.056
Table 3.2. The 37 tidal harmonic constituents that typically had been included in a harmonic analysis at CO-OPS and its predecessor
organizations (unless it was an extreme shallow-water situation).
In the 2nd column (origin of constituent), the term shallow indicates that the constituent was produced by nonlinear mechanisms in shallow water. A
(*met) indicates that the calculated harmonic constants for this constituent usually include mostly quasi-periodic meteorological effects. (S6 is marked with a
**, because this is one constituent that was a poor choice to be included in the standard 37, since it is usually much smaller than two other sixth diurnal
constituents which could have been chosen instead, MS6 and MN6.)
The 4th column (shallow-water equivalent) gives the shallow-water constituent with the same frequency as the astronomical constituent listed in the 1st
column. The 5th column (NL mechanism, sym or asym) indicates whether this shallow-water constituent is produced by the symmetric nonlinear mechanism
(u*u*) or by the asymmetric nonlinear mechanisms (M(u)/Mx, uMu/Mx, and u*u*) as indicated in Table 2.4. Constituents listed as asymmetric can also be
produced by the lateral initial terms as discussed in Section 7.6.7.
89
The 3rd column (Cartwright number) gives the six digits that indicate which combination of astronomical frequencies (from Table 2.1) produces that
astronomical tidal constituent (examples are shown in Table 2.2). These six digits are the multiplying coefficients in front of the six frequencies L, and 1
through 5 from Table 2.1. The classic Doodson numbers are the same as the Cartwright numbers but with a 5 added to each digit (except the first), which
Doodson did to keep these digits from being negative.
The 6th column gives the angular speed of the constituent in degrees per hour. Dividing the angular speed by 360o (= one cycle) and multiplying by 24
gives the frequency of the constituent in cycles per day (cpd), shown in the 7th column. Dividing 360o by the angular speed gives the period of the constituent in
hours, shown in the 8th column.
The 9th column gives the synodic period for the constituent with respect to (wrt) the constituent which is closest in frequency among the constituents with a
larger amplitude (shown in the 10th column). Table 3.1 showed several synodic periods for each constituent, each paired with another constituent, but only the
longest synodic period for the larger constituents is shown in the 9th column.
The 11th column shows the relative coefficient of the strength of the tide potential (the tide producing force) for each astronomical constituent as determined
by Cartwright and Edden (1973), in order to give one a first guess at the expected relative sizes of the tidal constituents (at least the astronomical ones).
However, hydrodynamics will change these relative strengths, and the 12th column gives, as just one example, the constituent amplitudes for a water level station
at Trenton, NJ, on the Delaware River. In this latter case, one will notice that N2 is larger than S2, unlike in the tide producing force. Also, for that station one is
really calculating 2MN2 and not L2, and likewise 2MS2 not 2.
The tables in Appendix A include many additional tidal constituents.
squares solution technique (Harris, 1965; Foreman, 1977; Foreman and Henry, 1989; Dronkers,
1964). The daily predictions for reference stations in all nationally published Tide and Tidal Current
Tables are produced from the harmonic constants that come out of tidal harmonic analysis.
Nonharmonic analysis, which is simply one of several methods of comparing the tide or tidal current
at two locations, is used to calculate the time differences, height differences, velocity ratios, etc. for
the thousands of subordinate stations in these tables that are referred to the reference stations.
Several refinements to the harmonic methods were developed in recent decades, as well as the cross-
spectral approach of the response method developed by Munk and Cartwright (1966). All of these
methods will be described below. Whatever the method used, some principles will always apply.
As mentioned above, the analysis of longer data times series will lead to better predictions no matter
which type of analysis is used, because more tidal frequencies can be resolved with a longer time
series.
3.4 Harmonic Analysis
3.4.1 Introduction
Tidal harmonic analysis was first developed by William Thomson (later Lord Kelvin) in England
in 1867, but was developed independently in 1874 in the U.S. by William Ferrel. In England,
Thomsons work was modified and improved by George Darwin (1883) and Arthur Doodson (1921,
1928) and others, and in the U.S. Ferrels work was modified and improved by Rollin Harris (1897-
1907), Paul Schureman (1924) and others in the U.S. Coast and Geodetic Survey. Harmonic
analysis takes advantage of the fact that one knows a priori all the frequencies at which tidal energy
will be found in a data times series. As already mentioned, most of these frequencies are
astronomically caused but many are due to the nonlinear hydrodynamic effects of shallow water.
The demonstration in Section 3.1 provides the basic idea of how to find the amplitude and epoch
(phase lag) of each tidal constituents contribution to the tide or tidal current. But when one looks
more specifically at the exact mathematical technique for finding the amplitudes and epoch of tidal
harmonic constituents, there are differences in the various forms of harmonic analysis techniques.
The biggest difference (between older and more recent techniques) is that the Fourier analysis
solution technique (e.g., Schureman, 1958) has generally been replaced by a least-squares solution
technique (e.g., Harris et al, 1963; Foreman, 2004), the latter based on minimizing the squared
differences between tidal data and computed tidal predictions.
Before looking at the actual methods for extracting the tidal harmonic constants from a time
series of data, one must first look at the harmonic tidal prediction equation. This is the equation into
which the tidal harmonic constants (amplitudes and phase lags) are put in order to make a tidal
prediction.
3.4.2 The Harmonic Tide Prediction Equation
A tide prediction can be made by summing up the oscillating contributions of some number of
tidal constituents. The tide, being is a single scalar quantity, requires only a single tide prediction
equation. However, Section 3.4.3 will deal with prediction of the tidal current, a vector quantity,
which must be broken up into two orthogonal components, so that two prediction equations are
needed. (Another option for the tidal current case, however, is to treat the two components together
using complex algebra, in which case one can use one prediction equation instead of two.)
90
Figure 3.2. A graphical representation of the

amplitude and epoch (phase lag) of a single
tidal constituent and its time relationship to the
moons transit (over the tide station). (From
Schureman, 1958.)
The formula for the height of one tidal constituent, h1 , at time equal zero, can be written as
h1 = f H cos (Vo + u - )
which is represented graphically in Figure 3.2. The contribution of a tidal constituent for a specific
tidal frequency (each frequency being classically symbolized by the angular speed, a; see Section
2.1.1) is represented by a cosine curve with an amplitude and phase. The amplitude is often called
H, but as will be seen, that H is often multiplied by a node factor, f, so that the modulation effect of
the 18.6-year lunar nodal cycle can be included (see Sections 2.2.4 and 4.1.5). The phase is usually
made up of three parts, the first changing with time and determined by the frequency (angular
speed), the second a phase relationship for the idealized equilibrium tide based on astronomical
variations (discussed below), and the third, a phase lag relative to the equilibrium tide. This third
part, a phase lag for that particular constituent (relative to the moons transit over the tide station),
is usually called the epoch, .
It is the amplitude H and the epoch that are needed for each tidal constituent to make a
prediction (and it is H and for each constituent that are calculated from a data time series using
harmonic analysis). In Figure 3.2, T is a particular time of interest, and M is the time when the
constituent argument equals zero, so that the interval from M (the moons transit over the tide
station) to the following high water (HW) is the epoch . The time interval from the previous high
water to M is measured by the explement of ( i.e., 360o - ), which can be expressed as - . The
phase of the constituent argument at time T is reckoned from M and is equal to Vo + u. The phase
of the constituent at this time is reckoned from the previous high water and is equal to Vo + u - .
The epoch of the tidal constituent determines when the maximum effect of the tidal constituent
occurs, relative to the maximum effect of this same constituent in the theoretical (and idealized)
equilibrium tide (which, as was seen in Section 2.1.2, is the tide that would occur on an Earth
covered with water and without continents and with the tide responding instantly to the tide
producing forces). More specifically, the epoch is the angular retardation of the maximum of a
tidal constituent of the observed tide behind the corresponding maximum of the same constituent
of the theoretical equilibrium tide. Thus it may also be defined as the phase difference between a
tidal constituent and its equilibrium argument. The equilibrium argument is the theoretical phase
of a constituent of the equilibrium tide. It is usually represented by the expression (V + u), in which
V is a uniformly changing angular quantity involving multiples of the hour angle of the mean Sun,
the mean longitudes of the Moon and Sun, and the mean longitude of lunar or solar perigee. As was
91
mentioned in Section 2.2.4, u is a slowly changing angle depending upon the longitude of the
Moon's node and varies over the 18.6-year lunar nodal cycle. When pertaining to an initial instant
of time, such as the beginning of a series of observations, the equilibrium argument is expressed by
(Vo+ u).
To account for the effect of the 18.6-year lunar nodal cycle (also see Section 2.2.4) on the
amplitude of the tidal constituent, H is usually multiplied by a node factor, f. However, f is not used
(nor is u) if the so-called satellite tidal constituents are included in the prediction (see Section 4.1.5).
The height of the tide, h, at any time, t, is typically represented by a formula summing up the
contributions of the individual tidal constituents, such as the formula below from Schureman (1958):
h(t) = H o + 3 f i H i cos (a i t + {Vo + u} i - i ) (3.1)

i =1 to n
where
h(t) = height (above some reference datum) of the tide at any time t
n = the number tidal constituents being used to make the prediction
Ho = mean height of water level above the datum
Hi = amplitude of tidal constituent i
ai = angular speed of tidal constituent i (i.e., its frequency) [in
degrees/hour]
t = time, reckoned from some initial epoch (e.g., t=0 at the beginning
of the year) [in hours]
i = epoch (phase lag) of tidal constituent i [in degrees] relative to the
moons transit over the tide station
fi = node factor for tidal constituent i
{Vo+u} i = equilibrium argument for tidal constituent i at t=0. [in degrees]
(a i t + {Vo + u} i - i ) = the phase at any time t (sometimes called the argument) relative to
the moons transit over the tide station
Usually an entire predicted time series of tidal heights is produced using this equation, and then
some standardized technique is used to pick off the times and heights of the high waters and the
times and heights of low waters (see Section 3.9). Figure 3.3 shows four days of the predicted tide
at San Diego, CA (second to bottom plot), compared with the observed water level curve (bottom
plot), along with the plotted contributions of four semidiurnal constituents (M2, S2, N2, K2) and three
diurnal constituents (K1, O1, and P1).
In the U.S. the reference datum is typically the mean lower low water (MLLW) datum, which
is referred to as chart datum, because all the depths on U.S. nautical charts are depths below chart
datum. Tidal predictions are heights above chart datum, so the total water depth at any moment in
time at a specific location on a nautical chart is the tidal height added to the charted depth.
92
Figure 3.3. Example of a tide curve (second from the bottom) produced with the harmonic tide
prediction equation, along with curves for seven of the tidal constituents used to make the
prediction. The bottom curve is from water level observations for the same time period as the
tide predictions.
As mentioned in Section 2.1.1, the angular speed of a constituent is merely a traditionally used
form of frequency and is given in terms of degrees per solar hour, where 360o is one complete
constituent cycle. M2, for example, has a speed of 28.984104 o/hour. One can obtain the frequency
of a tidal constituent in cycles per hour by dividing its speed by 360o. Thus, dividing 28.984104
o
/hour by 360o gives the M2 frequency of 0.081 cycles per hour, or 1.932 cycles per day. One can
obtain the period of a tidal constituent by dividing 360o by its speed. Thus, diving 360o by
28.984104 o/hour gives the M2 period of 12.4206 hours.
The tidal harmonic constants that come out of a harmonic analysis are the n pairs of amplitudes
and epochs, H and . The tidal amplitude H is modified by the node factor, f , which allows one to
include the modulating effect of the 18.6-year lunar nodal cycle without having to use numerous
additional tidal harmonic constant (the satellite tidal constituents). Schureman (1958) provides in
his Table 14 the values of f for 37 tidal constituents for the years 1850 through 2000. Zetler (1982)
extended this table to the year 2025. In this table f is considered to be constant for each entire year
(using the midyear value of f ). However, if more accuracy is desired, one can use the original
93
astronomical formulas to calculate f for any time and any time series length. These formulas are
provided by Schureman in his Table 2. These node factor formulas have been built into some
harmonic analysis and tidal prediction programs, while other programs require the user to read in
the node factors as data obtained from Tables like those in Schureman.
The tidal epoch (phase lag) is referenced to the time when the moon passes directly over the
exact location of the tide station, the data from which the tidal harmonic constants were calculated.
However, using is very inconvenient, since time is always referenced to a local time meridian (or
sometimes to the Greenwich time meridian). Also, comparison of phase lags at different locations
is not possible without the phase lags having the same time reference system. When the epoch is
given in terms of the local time meridian, it is usually called (or g). When the epoch is referred
to the Greenwich time meridian (0o) is it is called G. The relationship among these three defined
epoch terms is described by the following formula:
( = g ) = + pL - aS
15 (3.2)
where
L = west longitude of the tide station (in degrees); it will be negative
if the station has east longitude
a = angular speed of the tidal constituent (in degrees/hour)
S = west longitude (in degrees) of the local time meridian
p = the species of the tidal constituent (i.e., = 0 for long-period, = 1 for
diurnal, = 2 for semidiurnal, = 4 for quarter-diurnal, etc.)
+ pL = G = the epoch relative to the Greenwich meridian
Time meridians around the world are every 15o of longitude, that is, one hour of time change for
every 15o of longitude. In the above formula, the 15 is actually 15 degrees/hour. Thus, for example,
the time meridian of 75oW, for the East Coast Standard Time in the U.S., is (75/15 =) 5 hours earlier
than at the Greenwich time meridian (0o). Typical tidal analysis and tidal prediction programs allow
the use of any of these three versions of the epoch, but one should know the relationship among the
three versions. They become especially relevant when considering the equilibrium argument,
{Vo+u}.
{Vo+u} is not only different for each tidal constituent, it is different for every longitude on Earth.
Schureman (1958) provides in his Table 15 the values of {Vo+u} for 37 tidal constituents at the
Greenwich meridian for the years 1850 through 2000. Zetler (1982) extended this table to the year
2025. As mentioned earlier, Vo is a uniformly varying part of the equilibrium argument, and is
referenced by Schureman to the beginning each year. u is a very slowing varying part of the
equilibrium argument due to changes in the moons node, and is (like the node factor, f ) referenced
by Schureman to the middle of the year and assumed to be constant throughout the year. {Vo+u}
must, therefore, be corrected for the longitude of the tide station being predicted and for the time
meridian used for time referencing. This is accomplished using a formula that looks similar to the
one above, namely, the equilibrium argument at a location that is Lo west of Greenwich, using the
local time meridian, which is So west of Greenwich, is
{Vo+u} = {Vo+u}Greenwich + aS
15 - pL
The tidal prediction equation can thus now be written:
94
h(t) = H o + 3 f i H i cos (a i t + [{Vo+u}Greenwich + aS

15 - pL] i - i )
i =1 to n
which simplifies to
h(t) = H o + 3 f i H i cos (a i t + {Vo+u} i Greenwich - i ) (3.3)

i =1 to n
so one can now use Schuremans Greenwich equilibrium arguments as well as epochs referenced
to the local time meridian (shown here as , but also called g).
However, as mentioned above, the value for the u portion of the equilibrium argument due to
changes in the moons node is assumed to be the midyear value and constant for the entire year (as
is f). If one would likely a little more accuracy one can use the actual (fairly messy) astronomical
formulas for V, u, and f provided by Schureman. These have been computerized and used in some
tidal analysis and tidal prediction programs.
3.4.3 The Harmonic Tidal Current Prediction Equations
The tidal current is a vector quantity usually given in terms of a speed and a direction. To
harmonically analyze current data one must first transform each speed and direction pair into two
orthogonal components, such as north and east components (see Figure 3.4). If si is the current
speed and i is the current direction clockwise from north ( i indicating a particular data point pair),
then the north and east components are simply
Ni = si cos i and Ei = si sin i
Any pair of orthogonal components can be used, and it is common to use major and minor
components, where the direction of the major axis is usually the direction of maximum flood, and
the direction of the minor axis is 90o clockwise from the major axis (see Figure 3.4), namely
Mji = si cos (i - major) and Mni = si sin (i - major)
Each current vector data time series will be turned into two current component data time series,
such as Ni and Ei (or Mj i and Mn i ), and thus for each tidal current constituent there will be two
harmonic constant pairs, namely, an amplitude and epoch for (e.g.) the north component, and an
amplitude and epoch for the east component (or likewise for the major and minor components).
And thus for tidal currents, the harmonic prediction equation (3.1) will be used twice, once for
each orthogonal component (with the h replaced by an N or E, or by a Mj or Mn). Two predicted
component data times series will be produced, and then will be combined to produce predicted speed
and direction time series using
si = ( Ni2 + Ei2 ) and i = tan-1 (Ei/Ni)

or
si = ( Mji2 + Mni2 ) and i = tan-1 (Mni/Mji) + major
Times and amplitudes of maximum floods and maximum ebbs can then be picked off the speed time
series, si (see Section 3.9.2). The minimums (or slacks, if the current speed goes down to zero and
95
Figure 3.4. A current vector with speed S and direction , broken up

into north and east components (N and E) and into major and minor
components (Mj and Mn).
the tidal current becomes truly reversing) are usually found by looking at the speed values when the
current flow is directly perpendicular to the major axis as flow (see Section 3.9.2).
The orthogonal pair of harmonic constants for each tidal current harmonic constituent can also
be combined (in a way analogous to combining the predicted component time series). This will
allow us to see how the speed and direction of flow due to a particular tidal constituent changes each
hour, and to watch it rotate around the compass over a tidal constituent cycle. The result will be an
elliptical representation of the flow due to that tidal constituent. The technique for producing these
harmonic constituent ellipses will be described in Section 5.2.
If the two orthogonal components chosen were major-minor components, and the major axis
was chosen to be the direction of maximum flood, then by looking at these harmonic constants one
can get some understanding of the harmonic makeup of the tidal current in the up-channel direction
(i.e., the primary flood and ebb flow), as well as the harmonic makeup of the cross-channel/cross-
bay component of the tidal current. [If one originally chose north-east components (as has often
been done by default) and these do not line up with the predominant major-minor axes of flow, then
interpretation of the results is limited until the harmonic constant pairs are transformed into a major-
minor system (see Section 5.2).] The tidal constituent ellipse produced from the pair of orthogonal
harmonic constants, provides more information, showing how the flow from that tidal current
constituent changes over the whole constituent cycle, which way it rotates, and the direction of its
major axis (which may be somewhat different than the mean flood direction and also may be
different than that for other constituents, even though one would think that the bathymetry would
affect all the constituents the same). One can also decompose each constituent ellipse into different
pairs of harmonic constants for two polarized counter-rotating circular motions, one for clockwise
motion and one for counter-clockwise motion. This is often done with spectra obtained from
currents, and are called rotary spectra (see Section 3.10.3).
One can also treat the two orthogonal components together using complex algebra (Pawlowicz,
et al, 2002). In this case there is one prediction equation instead of two, since the current vector is
96
represented by the complex number u + iv (where u and v are the orthogonal components of the
tidal current, but the complex algebra allows them to be treated together). This complex prediction
equation looks like
v(t) = v o + 3 a i ei i t + a i e i i t
i =1 to n
where v(t) is used to indicate velocity, but this equation could also be used for the scalar tidal
heights. a i and ai are complex conjugates of the amplitude if v(t) is a real series. For more
information see Pawlowicz, et al, (2002).
3.4.4 Fourier-based Harmonic Analysis
The tidal prediction equation itself appears fairly simple and it is, because various scientists
have already worked out the great mathematical complexity behind it, i.e., that dealing with the
derivation of the tidal frequencies from astronomical equations, and with the variation of the tide
producing forces with time, again based on astronomical equations. The present chapter has
unfortunately provided only a little insight into the complexity of this astronomically based work.
For a proper appreciation of it one should read Doodson (1921 and 1928) and Schureman (1958,
originally published in 1924), Godin (1972), and the more recently Cartwright and Tayler (1971),
and Cartwright and Edden (1973). With that work accomplished, one is left with simply having to
put some tidal harmonic constants into the tidal prediction equation, and then predicting the tide for
any time in the future or the past. These tidal harmonic constants have to be determined from water
level data (or current data) because their amplitudes and epochs are determined by the
hydrodynamics of the waterways from where the data were obtained, and not by the astronomy.
The more demanding part of the job of tidal analysis and prediction is the analysis part, that is,
the calculation of the most accurate tidal harmonic constants possible from whatever data time series
may be available. Tidal analysis, of course, also relies on these same astronomical equations,
looking for energy at those same astronomical frequencies, and using the equations that determine
the variation in tidal forcing with time. And again, that part has been worked out. But extracting
accurate tidal harmonic constants from a data time series is trickier than making tide predictions
using those harmonic constants.
There have been two primary techniques within the idea of harmonic analysis for extracting
harmonic constants from a data time series using a Fourier series technique or using a least
squares technique. In Fourier-based harmonic analysis, each tidal constituent is solved for
separately, and this type of analysis resembles to some extent our demonstration in Section 3.2. In
the least-squares-based harmonic analysis (which is described in detail in the next Section), all the
tidal constituents are solved for simultaneously, the approach being to minimize the squared
differences between measurements and computed tidal predictions. As will be seen, the least
squares method has many advantages and is the method generally used today. But the Fourier-based
method, which was the first method to be developed, is still worth looking at, at the very least as a
lead in to the least squares technique treated in the next section.
Although in CO-OPS a harmonic analysis program based on the least squares technique is the
primary method for extracting tidal harmonic constants from data, there is still in use a program
based on the Fourier analysis technique developed by Schuremen (1958) for time periods of 15 or
29 days. This latter program, written by Dennis and Long (documented in 1971) has probably
remained in use because it is easy to use. Its inference and elimination routines infer some
97
constituents that cannot be solved for with only 15 or 29 days of data and correct the solved-for
constituents for the adverse effects of the unsolved-for constituents (see Section 4.2.1).
Sir Isaac Newton, who in 1687 first explained that the tide was a consequence of the gravitation
attraction between the moon and the Earth, was certainly aware of the numerous astronomically
caused periodicities in the tide. However, it was the French scientist Pierre-Simon Marquis de
Laplace, who in 1775, following up on Newtons work, first suggested the idea of harmonic
analysis. He had derived the hydrodynamic equations of motion for tides on a rotating earth, and
solved them for the special case of an Earth entirely covered by one ocean. Using that solution he
became the first to separate the tide into three species semidiurnal tides, diurnal tides, and long-
period tides, and he went on to say that the tides could be represented by the sum of cosines
representing the many periodic forces involved. However, it was not until 1867 that Sir William
Thomson (later Lord Kelvin) actually developed and used a practical harmonic analysis method.
In the U.S. only seven years later, William Ferrel, a scientist in the U.S. Coast and Geodetic Survey,
unaware of Thompsons work, independently developed the harmonic analysis technique in 1874,
also basing his method on Laplaces work. In each country, those original harmonic analysis
methods were further modified and improved, in England by George Darwin (1883) and later Arthur
Doodson (1921), and in the U.S. by Rollin Harris (1897-1907) and Paul Schureman (1924) of the
U.S. Coast and Geodetic Survey.
Much of the work of these scientists was astronomical and oceanographic. To actually extract
the tidal harmonic constants from the tide data, they all used some form of Fourier series technique.
As an example the form of the Fourier-base technique in the U.S. Coast and Geodetic Survey
developed by Schureman (in 1924, but revised in 1940) and computerized by Dennis & Long (1971)
will be briefly looked at.
Traditionally, before a Fourier-series-based harmonic analysis is carried out, the tidal prediction
equation is changed to an equivalent form involving the sums of sines and cosines. Doing this, the
tidal prediction equation (3.1) becomes (for a prediction taking into consideration n tidal frequencies ai )
ai )
h(t) = H o + 3 c i cos a i t + 3 s i sin a i t (3.4)

i =1 to n i =1 to n
where
f i H i = (c i2 + s i2)
(3.5)
and (Vo + u) i = tan-1 s i
ci
Then ci and si an be found from the following two summations:
ci = 2 3 h cos a i k and si = 2 3 h k sin a i k (3.6)

N k =0 to N-1 k N k =0 to N-1
where time t is now represented by (k ), being the sampling interval of the data time series and
k being the number of the data point (from 0 to N1). Although these summations can be carried
out in seconds in the Dennis&Long program (see their subroutine FORAN), in the pre-computer era
they had to be done by hand. Years before Schuremans work, special paper stencils or keys(i.e.,
sheets with holes in it that were laid over sheets of tabulated hourly heights) were invented by
Leland P. Shidy in 1885, and Schureman incorporated them into his analysis scheme. As Schureman
98
put it, they resulted in a very great saving of labor, but it could still take a week or more to do a
harmonic analysis by hand compared with seconds on a computer.
As mentioned earlier, the length of the data time series, N, is chosen to be one of the key synodic
period lengths (although in the Dennis & Long program that can only be 15 or 29 days). The pairs
of ci and si would then be converted to pairs of amplitudes and epochs using formulas (3.6). The
epochs are then adjusted for proper time reference using the equilibrium argument {Vo+u}, which
includes the nodal effect of u. The amplitudes are likewise adjusted by dividing them by the node
factor f for the time period at the particular data time series that was analyzed. In the Dennis &
Long program (1971), the f and u are determined directly from the appropriate astronomical
equations (from Harris, 1897, and Schureman, 1941). The program then goes on to implement
Schuremans inference and elimination routines (see Section 4.2.1).
The next section will describe the numerical technique most commonly used today for extracting
tidal constituent amplitudes and epochs from a data time series the least squares technique.
3.4.5 Least-squares-based Harmonic Analysis
As mentioned above, most harmonic analysis programs today use the least squares technique
instead of the Fourier series technique described in the last section. Because of its computational
demands the least squares technique could not really be used until the computers became available
in the 1960s. The least squares technique simultaneously solves for all harmonic constituents (that
can be separated with the length of the available data time series). The harmonic constants
determined are those that minimize the sum of the square of the residuals, that is, minimize the sum
of the squared differences between the original data time series and the predicted time series (i.e.,
the fit to that series using the calculated harmonic constants). This method has several advantages
over the Fourier series method, including: (1) one is not restricted to continuous equally spaced data
with no gaps; in fact, the data can be completely random in time and can have large gaps; (2) there
is more flexibility on the length of the time series; one does not have to stick strictly to lengths that
match a particular synodic period, or a multiple of synodic periods; and (3) one can determine how
the variance between the data time series and the predicted time series from the calculated harmonic
constants is reduced by each additional constituent included in the analysis.
The least squares harmonic analysis used in CO-OPS is the program originally written by D. Lee
Harris in the 1960s. Most other harmonic analysis programs commonly used today also use some
type of least squares technique (e.g., Foreman, 2004a). First, the typical least squares technique will
be discussed, and then the Harris version will be briefly looked it. In that technique the harmonic
constants are derived using a multiple correlation screening process which can be terminated when
the regression equation contains a specified number of terms or when the next constituent will not
explain some preselected fraction of the variance.
To reiterate the problem, one has a time series of (in this case) water level data, h(t) with N
observations, from which one wants to extract the amplitudes (Hi) and phase lags ( i) of the
greatest possible number (n) of tidal harmonic constituents, whose frequencies are known. Fitting
n harmonic curves to the time series with N observations (where n << N) is an over-determined
problem, so there is no solution which could exactly match all the data points, and one looks instead
for an approximate solution that best matches the data points. To do this one uses some type of
optimization technique. Most common of these techniques is the least squares technique, in which
one will estimate the amplitudes and phase lags by minimizing the squared difference (i.e., the least
99
squares) between the original data time series and the predicted time series created using these
calculated amplitudes and phase lags.
Again one starts with the tide prediction equation (3.1) but here one has h(t) as the actual
observed data time series, and so an additional term is included to represent the residual time series,
hr(t):
h(t) = H o + 3 f i H i cos (a i t + {Vo + u} i - i ) + hr(t)
i =1 to n
with the definitions of the other parameters given in Section 3.4.2. The residual time series is equal
to the observed data series h(t) minus the predicted time series created using the n harmonic
constants (Hi and i). The residual series contains nontidal water level variations due to
meteorological effects (such as wind, atmospheric pressure, and river discharge) as well as tidal
oscillations caused by any harmonic constants not among the n resolvable constituents used to make
the prediction. The residual time series may or may not include Ho, but one typically demeans the
data time series before the harmonic analysis is carried out (in order to minimize roundoff errors),
in which case Ho is taken out of the problem. (This is especially important to do for water level data,
which may be referenced to a datum which can be quite a large value.)
And again (as was done in Section 3.4.4) the above water level equation can be transformed into
an equivalent form involving the sums of sines and cosines (but with the residual series added)
h(t) = H o + 3 c i cos a i t + 3 s i sin a i t + hr(t) (3.7)

i =1 to n i =1 to n
where
f i H i = (c i2 + s i2)
(3.8)
and (Vo + u) i = tan-1 s i
ci
Since one is dealing with digital data, with individual data points, it is clearer to put (3.7) into a
digital form. The equation for each individual data point, the kth data point, is
hk = H o + 3 c i cos a i t k + 3 s i sin a i t k + hr (3.9)

k
i =1 to n i =1 to n
where k goes from 1 to N. Our objective is to minimize the variance of the residual series, e2, i.e.,
minimize
e2 = 3 hr 2
= 3 { h k - 3 [ c i cos a i t k + s i sin a i t k] }2 (3.10)
k
k =1 to N k =1 to N i =1 to n
To find the minimum of e2 one takes partial derivatives of the residual equation (3.10) with respect
to the unknown coefficients ci and si, and set the results equal to zero. This produces 2n+1
simultaneous equations for n+1 tidal constituents. These equations are solved using matrix algebra
[solution of a (n+1) by (n+1) matrix equation], but the specific solution technique can vary, and
various ways of simplifying the matrix can be used to improve the ease and numerical efficiency of
the solution. For example, although most least-square-based harmonic analysis programs can handle
random data and data with large gaps, one usually has to turn on that option, since there are more
100
efficient matrix solutions that can be used when the data is continuous and equally spaced. (See
pages 18-23 of Foreman, 2004; as well as Emery and Thomson, 2001.)
As has been mentioned, with a Fourier-series-based harmonic analysis, one should stick fairly
closely to the guidance of synodic periods, but when using a least squares harmonic analysis, there
seems to be some leeway, and often less data (than based on synodic periods) will still do a good
job. Some oceanographers have managed to obtain good harmonic constants with almost half as
much data as would seem to be required by the Rayleigh criterion, and not necessarily with noiseless
data (see the end Section 3.3). But this has not always been the case, and there is some thinking that
it may depend on the relative phases of the two constituents trying to be resolved with too little data.
Now one particular type of least squares harmonic analysis will be briefly looked at, namely, one
using a numerical approach developed by D. Lee Harris (1965). A program using Harris method
is used in CO-OPS. Harris original Fortran code can be found in Zetler (1982). This method uses
a step-wise linear regression algorithm and a multiple-correlation screening process. In the first step
of the analysis, the tidal constituent which is most highly correlated with the observed tide is
selected and the program derives the harmonic constants for this constituent which provide the best
prediction (in a least squares sense, i.e., the residual is minimized). In the second step, the next
constituent that will make the greatest improvement to the first prediction is selected, and then the
program computes the harmonic constants for both these constituents which will provide the best
prediction (again in the least squares sense). The next step is the same, except the third most
important constituents is included, and so on. At each step the reduction in variance brought about
by the last constituent is calculated. This process continues until either all the user-specified
constituents have been considered, or until the reduction in variance brought about by the last tidal
constituent considered is below some user-specified value. In other words, if none of the remaining
tidal constituents make an appreciable difference, the calculation will end and these remaining
constituents will be left out.
[In the program implementing Harris method, subroutine SCREEN (see Zetler, 1987) solves
a large multiple regression equation, but treats the variables one at a time or in pairs, in order to
avoid stability problems. Two algebraically equivalent solution procedures are used, one based on
a correlation matrix and the other on a covariance or product matrix, because they have different
convergence properties, making each preferable for certain calculations. The matrix (of covariance
or of correlation) is inverted to solve for the amplitudes and phases of the harmonic constituents.
The contribution of the constituent with the highest correlation is then subtracted from the observed
data time series and the matrix is recalculated with the residual time series (in place of the observed
time series). In each iteration all unused variables are examined to determine which will make the
greatest contribution to the solution. The matrix is rearranged if necessary to facilitate the solution
for selected variables. A check for potential instability is made along with this procedure and if a
potential or real instability can be recognized with any predictor, that predictor is set aside for
consideration after all the apparently stable predictors have been considered. Calculations are ended
when none of the remaining predictors will make a significant contribution.]
A users guide for running the most recent version in CO-OPS of Harriss least square harmonic
analysis program (called LSQHA) can be found in Zervas (1999). It is presently set up to handle
175 tidal constituents. The program can handle water level or current data. In addition to the
harmonic constants (amplitudes and epochs) the program also outputs the variance accounted for
by each constituent, the cumulative variance, the observed mean, the observed variance, the
observed standard deviation, the residual variance, the residual standard deviation, and the
regression constant.
101
3.5 Other Frequency Domain Methods
3.5.1 The Response Method
Of all the alternatives to harmonic analysis for tidal analysis and prediction the response method
is the technique most often used, and the one often felt to be an improvement over the harmonic
technique, although such improvements are actually minor in terms of producing accurate
predictions. The response method is simpler in some ways (requiring fewer arbitrary constants), but
the results are usually not as easy to interpret as those from a harmonic analysis, whose amplitudes
and phase lags for each tidal frequency, whether astronomical or of nonlinear shallow-water origin,
are quite straight forward to interpret with regard to the effects of the hydrodynamics of the ocean
and connecting waterways. Expert use of the response technique requires some background in
spectral analysis and other areas of time series analysis. Here only a short summary of this
technique is provided, along with references to numerous papers.
The response method was first proposed by Walter Munk and David Cartwright in their well-
known paper Tidal Spectroscopy and Prediction (Munk and Cartwright, 1966). They note that
in the harmonic method the Kepler-Newtonian mechanics (representing the many aspects of the
Earth-moon orbit, the Earths orbit around the sun, and the Earths rotation and wobble) is only used
to identify the principal tidal frequencies (Table 2.1). In their response method, however, they
directly take into account the actual Kepler-Newtonian mechanics. From the basic astronomical
equations they generate a time series of the tidal potential itself (i.e., the tide-producing forces) for
the same time period as a water level data time series. They then treat the problem essentially like
a black box system, where the tide potential is the input to the system, the water level time series
is the output of the system, and the response of the system is determined by comparing the output
with the input in some manner. The response of the system once determined (for the particular
location of the water level station) can then be applied to a tide potential time series generated for
another time period to produce tide predictions for that time period.
The black box system essentially represents all the hydrodynamic effects of the ocean and any
connecting waterways on the tide (at that specific location), yet it is determined strictly statistically,
by cross correlating the time variation of the potential with the observed time variation in the water
level. This involves the use of a weighted sum of past and present values of the potential. If (t)
is the water level time series and V(t) is the astronomically generated tidal potential time series, then
a predicted water level time series (t) can be produced according to
(t) = 3 w(s) V(t-s) ,

s
where the weights w are determined so that the prediction error (t)(t) is a minimum in the least-
square sense. [This is referred to as taking the convolution of w and V, where V has been reversed
and shifted in time.] These weights represent the sea level response at the location of the water
level gauge to a unit impulse V(t)=(t), which is where Munk and Cartwright got the name response
method. The actual input potential V(t) was regarded as a sequence of such impulses. Munk and
Cartwright expanded V(t) in spherical harmonics, which converged rapidly so only a few terms were
needed. This input potential, in the form of time-variable spherical harmonics confined to the tidal
bands (0, 1, and 2 cycles per day), is cross-correlated with the output, and the response is expressed
in terms of frequency-dependent admittances (amplitudes and phase lags). (The admittance is the
cross spectrum of the input and output functions divided by the power spectrum of the input
102
function, the cross spectrum being a measure of the part of the output spectrum that is coherent with
the input spectrum. Each admittance has an amplitude and a phase.)
A key assumption of this method, called the credo of smoothness, is that these admittances
should vary smoothly over each narrow tidal band. This appears to work fine for deep water
stations, but in shallow-water areas, where the admittances are no longer as smooth, one has to
include some additional nonlinear input functions to handle the shallow-water tides (which is done
as a second step in the analysis).
One can also include other inputs, for example, a radiational input included by Munk and
Cartwright to represent some solar/meteorological effects (some of which show up in the Sa, Ssa,
and S1 harmonic constants in a harmonic analysis). (See also Zetler, 1971.)
The variability of the admittances for a tidal band varies with the number of complex weights
in the response analysis. The number of complex weights that can be resolved depends on the length
of the series being analyzed (Zetler and Munk, 1975), just as do the number of constituents that can
be resolved with the harmonic method. In fact, Zetler, et al (1969) provides a method for deriving
harmonic constants from response admittances. The response method has proved to be very useful
for tidal research, but only slightly improves tidal prediction accuracy (Zetler, et al, 1979; Amin,
1977, and Smith, et al, 1997). See also discussions in Pugh, 1987 and Godin, 1988.
From the standpoint of doing a self-prediction, the response method with a radiational input may
do a better job than a harmonic analysis because it does a better job at handling the low-frequency
(annual, semiannual, and other periods) variations due to nontidal/meteorological effects (wind,
pressure, temperature, river runoff). However, these low-frequency nontidal variations vary
considerably from year to year (as will be seen in Section 3.7.1), and so for predictions into the
future, when one does not know what the response to the radiational input will be (the chaotic
atmosphere being the key intermediary, since it produces the winds, atmospheric pressure, etc), one
would not expect as much prediction skill. Also with regards to prediction skill, there is always an
effect of the length of series (also affected by noise) no matter what the method of analysis (response
method or harmonic analysis). The response method has been used with short data time series by
using the water level time series (or predicted tide series) from a nearby longer reference station as
its input instead of the tidal potential. As was mentioned already (and will be discussed in more
detail in Section 4.2) there are also various ways to use a reference station with harmonic analysis.
3.5.2 The Harmonic Shallow-Water Corrections Method
In shallow water the nonlinear hydrodynamics transfers energy to many additional frequencies,
which must be included in a harmonic analysis if accurate predictions are to be made. For shallow
waterways with large tide ranges many dozens of additional harmonic constants are required, the
majority in higher frequency bands. Today with modern computing capabilities one can easily
handle a hundred or more tidal constituents, but in the pre-computer era this presented a serious
problem. Tide predicting machines could not handle that many tidal constituents and the effort to
determine the additional harmonic constants manually was excessive. This was a significant
problem in the United Kingdom, which has many bays and estuaries that are shallow and have large
tide ranges.
To deal with this problem Arthur Doodson developed a special technique, which came to be
called the harmonic shallow water corrections (HSWC) method (Doodson, 1957). This method was
intended to improve the predicted heights and times of high and low waters for locations in
103
waterways with strongly nonlinear tides. The HSWC method used a two-step process. First a tide
prediction was made using the best available harmonic constants, from which high and low waters
were calculated. Then in the second step, corrections to these high and low waters were made that
took into account the distortions of the tide caused by the shallow water. To make these corrections
Doodson actually took advantage of the aliasing caused by the energy at frequencies greater than
the Nyquist frequency (see Section 2.2.3). In this case, because (in the second step) his data points
were only high and low waters, he was dealing with a sampling interval of a lunar day. He was able
to derive formulas where the differences in the heights and times of the high and low waters
depended only on the folded frequencies obtained by this mean-lunar-day sampling. (See also
Godin, 1972, pp 154-61) Doodson also was able to do the initial harmonic prediction at a reference
station (e.g., at the entrance to a bay where the shallow-water effects were not strong) instead of at
the actual station where the strong shallow-water effects were, and then could correct those
predictions for the shallow-water effects at this second station. This made for a more efficient
process if there were many shallow-water stations to be predicted, since they could all use the same
reference station for the harmonically predicted first step of the process (especially since all
calculations were done by hand and all harmonic predictions were done on a tide predicting
machine).
In the modern computer age the HSWC is still a useful method for extreme situations where one
does not want to deal with hundreds of harmonic constituents. Amin (1977) did a comparison
between the HSWC method, the response method, and an extended harmonic method (i.e., one using
more than a 100 harmonic constituents) and found that all three were comparable to each other with
respect to the accuracy of the predicted high and low waters. However, he also points out that the
HSWC method does not work well in waterways with strong diurnal signals, i.e., where K1
approaches the magnitude of M2.
3.5.3 Species Concordance Method
Extreme shallow-water situations can be handled by the harmonic method of analysis and
prediction, but the number of additional shallow-water tidal constituents needed (to capture all of
the energy that the nonlinear hydrodynamic mechanisms have transferred to new frequencies) can
become quite large for a shallow waterway with a large tide range. For example, Zetler and
Cummings (1967) used 114 constituents for the tide at Anchorage, Alaska, as did Rossiter and
Lennon (1967) for the tide in the Thames estuary in the United Kingdom (although only 74
constituents of each 114 were the same for both groups).
In the Gironde estuary in France, George and Simon carried out spectral analyses of water level
data that showed tidal lines out to 38 cycles per day (cpd) at Bordeaux, about 100km (39 miles) up
the estuary. They were already using 100 constituents to predict the distorted tide at Bordeaux, but
they increased that to several hundred. With so many constituents needed it gets very messy trying
to determine the particular constituents that should be included in the analysis, since there can be
many different combinations of the astronomical frequencies that can produce the same high-
frequency compound tidal constituents. And it takes more than a year of data to separate many of
them. In 1978 the Bordeaux Harbor Authority asked the French Hydrographic Service to improve
the accuracy of the tide table predictions for the Gironde estuary including Bordeaux Harbor. The
result was the species concordance method, which was first used in 1980 to produce those improved
tables (George and Simon, 1984; Simon, 1991).
104
The species concordance method relies on the nonlinear relation between species of the tide at
the station to be analyzed (and predicted) and at a reference station where the tide is well known and
easily predicted (often at the mouth of an estuary). The efficiency of the method results from its
application to the predictable part of the water level variation, which is confined to narrow bands
of the spectrum. Each species may be analyzed individually. This method is similar to the response
method in that it involves a convolution (see Section 3.5.1), but without using the tidal force (or
potential) as the input as the latter method does.
Like Doodsons harmonic shallow water corrections method (Section 3.5.2) this is a two-step
method, and it also uses a reference station (e.g.) near the entrance of an estuary where the shallow-
water effects have not built up. The harmonic method is used to predict the relatively undistorted
tide at the mouth of the estuary, and a secondary technique is then used to predict the tide at the
stations up the estuary where the shallow-water effects are significant. The relationship between
the tide at the reference station and the tide at a station up the estuary is expressed in terms of so-
called reduced vectors, which are the complex amplitudes of the tide in particular species, and which
vary slowly. Reduced heights are computed from these using the Fast Fourier Transform, and then
real heights by quadratic interpolation. (See George and Simon, 1984; Simon, 1991).
The species concordance method has been used successfully not only to account for the many
higher frequency tidal species (e.g., up 36 cycles per day for Rouen on the upper Seine River), but
also to account for the effects of river discharge. Average seasonal variations in river discharge can
be input, but river discharges can vary considerably throughout the year, so in the species
concordance method the fresh water level may be explicitly introduced by the user so that events
like freshets can be handled.
3.5.4 Method for Superfine Resolution of Harmonic Constituents
Amin (1976, 1991) developed a refinement to the harmonic method, the method for superfine
resolution of tidal harmonic constituents, in which a corrective step is added to the harmonic
method. The tidal species are divided into subgroups of common origin, such as those due to
gravitational (astronomical) or advective or frictional forces. Then, in conjunction with functional
representation of the response of the sea and of nonlinear interaction coefficients within a subgroup,
harmonic analysis is used to improve the resolved tide and to compute estimates for some additional
terms (constituents). The interaction coefficients of the resolved nonlinear tides are computed using
the analytical solution of the hydrodynamic equations (see Chapter 7). Then the unresolved
nonlinear terms are estimated by interpolating these coefficients. Using this method, estimates of
nodal terms can be obtained from analysis of a few years of tide gauge records and the variance of
the principle tides can be reduced. See Amin (1976, 1991) for more details.
3.5.5 Continuous Wavelet Transform Method
As was shown in Section 2.3.3, there are times when in shallow water the nonlinear interaction
between the tide and a nontidal phenomenon (such as river flow or storm surge) can lead to a
temporary change in the tide range, as well as a distortion in the tide indicated by a change in the
shape of the tide curve and an increase in the overtide constituents (M4, M6, and M8). This, of
course, can also be seen in the tidal current, which in addition can also affected by changes in
density (due to changes in salinity or temperature) through the temporary generation of internal tide
waves or other baroclinic effects.
105
Any harmonic constants calculated from data obtained when nontidal effects are very significant
cannot be used at other times to predict the tides. They might only be usable for a self prediction
during the time period from which they were derived, and possibly for a future period when the same
nontidal effect was predicted to occur. For example, as was seen Section 2.3.3 a high fresh water
flow in the early spring in some rivers will lead to a dramatic shrinkage in the tide range and a
distortion of the tide curve (a higher M4/M2 ratio). The harmonic constants derived from (e.g.) a 15-
day harmonic analysis for a period covering this freshet, if used to make predictions for the rest of
the year would greatly underpredict the tide ranges. But even using these harmonic constants to try
to predict the water level during the freshet event may not work well either, since the event most
likely was much shorter than 15 days, the 15-day analysis came up with average harmonic constants
for the whole 15-days (hence spreading the two- or three-day effect over the entire 15 days). The
effect of the shorter period freshet was not captured, since one could not do a quality harmonic
analysis with only a few days of data.
For this type of situation, i.e., a relatively short nontidal event which is known to affect the tide
through some type of nonlinear interaction mechanism (usually due to shallow water, or for currents
perhaps a lateral inertial effect as seen in Section 2.36e) a special type of tidal analysis was
developed by David Jay called the continuous wavelet transform (CWT) method (Jay and Flinchem,
1999; Jay and Kukulka, 2003). This method is used for those special applications where one wants
to better understand the dynamics of a short nontidal event that significantly affects the tide or tidal
current, or one wants to be able to do a self prediction for that time period. Also, if one has an
independent way of accurately predicting the nontidal event, then the CWT method could possibly
be used to make a more accurate future prediction of the total effect of such an event, involving both
the tide and the nontidal influence. [This, of course, can also be done using a nonlinear numerical
hydrodynamic model, if one has such a calibrated and verified model for the particular waterway
of interest.]
Like the species concordance method, the CWT method does not use tidal constituents, but
instead uses just tidal species, usually the semidiurnal band, the diurnal band, and the higher
harmonic (overtide) bands. For short nontidal events, one cannot separate out the effects of (e.g.)
the various semidiurnal constituents, but the effect of not knowing N2 and S2 in addition to M2 is
much smaller than the nonlinear effect of the nontidal phenomenon on the tide. Thus, for this
special situation a species approach is fine.
3.5.6 Harmonic Constants From a Long Series of High and Low Waters
Before modern automated water level gauges were invented, water level measurements were
made using a tide staff attached to a pier or some other permanent structure. Because the
measurements had to be made by sight and recorded by hand, it could be quite labor intensive to
record hourly water level heights. At some locations therefore only the height and time of high
waters and low waters were recorded, but sometimes such data series lasted for years. Such
historical data time series may only have one or two data points a day, but they can still be used to
extract harmonic constants. Although, as will be seen elsewhere in this book, the high and low
waters (and especially their timing) are very sensitive to the effects of nonlinear shallow-water
dynamics as well as to the accuracy and/or bias of particular observers, useful harmonic constants
should still be obtainable, if the time series is long enough (see more below).
When such data are found today (through some data archeological effort) it can easily be
harmonically analyzed with a least squares technique written for handling random data. However,
106
in the pre-computer era when these high and low water data series were created, other methods had
to be devised, but these were still quite elaborate and labor intensive to carry out.
The first in the U.S. to derive harmonic constituents from a series of high and low waters was
probably William Ferrel (1874, pp 160-83), who applied his method to a 19-year data series from
Brest, France, that had originally been analyzed by Laplace. [One result of Ferrels work was his
calculation that the moons mass was 1/78.0 of the Earths mass (p.189), a value that is closer to
todays accepted value of 1/81.3 than the value that Laplace had calculated, 1/75. He went on to
obtain an even closer value, 1/81.7, from a 19-year data series from Boston Harbor. (see Cartwright,
1999, p 104).] In 1890 George Darwin published his method of harmonic analysis of tidal
observations, which he said was needed because Extensive use of the tide gauge has only been in
recent years, and by far the largest number of tidal records consist only of observations of high and
low water. (Darwin, 1890) He also states My objective has been to make the whole process a
purely mechanical one, and, although nothing can render the reduction of tidal observations a light
piece of work, I believe that it is here presented in a form which is nearly as short as possible.
However, 61 years later Arthur Doodson would say that due to the complexity of the method
designed by Sir George Darwin he would rule it out as a practical method (Doodson, 1951).
Doodson came up with a clever method that treated the high and low water times and heights as low-
frequency harmonic series, but it still took him 65 pages to explain and demonstrate his method.
Twenty years earlier Doodson published a method of analyzing and predicting tidal currents based
on analysis of only the slack waters (Doodson, 1928b), which also could be applied to the tide.
In 1924 Paul Schureman also referred to Darwins method as an elaborate mode of analysis
and provided his own simpler method of harmonically analyzing high and low waters (Schureman,
1924, rev. in 1940), which was partially based on a method developed by Rollin Harris at the turn
of the century. Only a few highlights of that method will be mentioned here (see pages 100-104 in
Schureman for more details) to give the reader an idea of the process, which involved the use of the
old Coast & Geodetic (C&GS) forms and stencils. The first step was simply to make the usual high
and low water reductions, including the computation of the lunitidal intervals. This involved
tabulating the times and heights of all the high and low waters, along with the times of the moons
transits, and the calculation of the interval between each transit and the following high or low water
(he used C&GS Form 138 for this). Then separate means were found of all the low water intervals
and of all the high water intervals. For semidiurnal tides the approximate amplitude and epoch for
M2 could be obtained directly from these results. The amplitude of M2 was found by multiplying
the mean tide range by 0.47 (or by the ratio of M2 amplitude to mean tide range at a nearby water
level station), and then corrected for the longitude of the moons node using factor F from
Schuremans Table 12.. The epoch of M2 was obtained from the high and low water lunitidal
intervals, HWI and LWI, using the formula
M2o = (HWI + LWI) x 28.944 + 90o
where HWI must be larger than LWI, and so 12.42 hours could be added to HWI if necessary. M4
is the main cause of the difference between the average duration of the rise and fall of the tide, and
this was used to calculate an approximate M4. Schureman ended up with the following formula for
the amplitude of M4

M4 = + cos a DR M2
cos a DR
107
where a = the angular speed (frequency) of M2 and DR = the average duration of the rise of the tide.
Schureman then found approximate values for S2, N2, K1, and O1 using the following procedure.
First he added the mean tide range value to each low water height, and copied the high and modified
low water heights onto the form usually used for hourly height (C&GS Form 362). Then he summed
for the desired constituents using the same stencils normally used for analyzing the regular harmonic
analysis of hourly heights. He then went through a second procedure to improve these results,
making use of a tide predicting machine. With this machine he made tide predictions with the
harmonic constants calculated so far (plus some smaller inferred constituents) and tabulated the
differences between the heights of the predicted high and low waters and the heights of the observed
high and low waters. These differences were then summed and analyzed in the same manner as the
original observed heights of high and low water. The result from this analysis of the residuals was
then combined with the previously calculated constants. For the combination of the previously
calculated harmonic constant with the amplitude and epoch corrections, Schureman used the
following two formulas:
A = [ (A cos + A cos )2 + (A sin + A sin )2 ]

and
= tan1 A sin + A sin
A cos + A cos
where A and are the final corrected constituent amplitude and epoch, A and are the original
uncorrected amplitude and epoch, and A and are the corrections.
Another specialized example of deriving harmonic constants from nonharmonic tidal quantities
was published in 1952 by Bernard Zetler, in this case a relatively simple procedure to infer M2 and
(K1+O1) amplitudes from the mean diurnal high water inequality (DHQ) and the mean diurnal low
water inequality (DLQ). DHQ is the difference between the mean higher high water (MHHW) and
the mean high water (MHW), and DLQ is the difference between the mean lower low water
(MLLW) and the mean low water (MLW).
The ratio of (K1+O1) to M2 is often used for the classification of the type of tide (see Section
2.2.5), but in this case there was a different motivation for this technique, which was developed
during World War II. About 250 secondary stations in the Philippines had to be assigned to several
new reference stations (which were required because of the great variability in the diurnal-to-
semidiurnal ratio throughout the islands). Because of this great variability, the values for M2 and
(K1+O1) were needed in order to assign a secondary station to the most appropriate reference station.
However, no water level data were available for these stations, only the DHQ and DLQ values in
the tide tables. And so Zetler developed the following method, publishing it six years after the war
ended.
Zetler ended up with two seemingly very simple formulas for the (K1+O1) amplitudes
if DHQ > DLQ, then (K1+O1) = DHQ/F1

and
if DHQ < DLQ, then (K1+O1) = DLQ/F1
where F1 is provided by Table 1 found in Zetler (1952) and ranged from 0.46 to 0.64, chosen
according to a DHQ/DLQ (or DLQ/DHQ) ratio ranging from 1.0 to 3.6. He also presented a
seemingly simple formula for the M2 amplitude from the mean range (Mn)
108
M2 = Mn/(2.19 + F2)
where F2 is provided by Table 2 in the same publication and ranged from 0.00 to 0.65. Of course,
the complexity came in developing the Tables for F1 and for F2, which is explained in Zetler (1952)
and makes use of Tables in S.P. 260 (C&GS, 1952).
As mentioned at the beginning of this section, with todays modern computer capabilities, the
use of the least squares harmonic analysis technique is the best way to derive harmonic constants
from a time series of high and low waters. Zetler, et al (1965) were one of the first to assess the
usefulness of this approach. They created one month predicted tide records with 100 or less random
data points selected over that period but restricted to 11 hours of daylight (which comes down to 3
or less data points per day). Although these random data points were not necessarily high or low
waters, they were equally as sparse. (They actually represented the time density of observations of
currents measured photogrammetrically from airplanes which flew only during the daytime.) They
were able to obtain accurate harmonic constants for a variety of cases. Most cases were noise free
(since the data points selected were actually tide predictions), but in four cases they added noise by
reducing the number of significant figures in the data points.
Foreman and Henry (2004) also carried out numerical tests, this time specifically analyzing one
month of observed high and low waters with a least squares technique. Good harmonic constants
were obtained for all the test cases, which included: adding gaps, adding timing errors, using
inference and not using inference, and using daytime only values. They also added an option to the
least squares error-minimization equation in which they included a zero derivative representing the
fact that they were looking only at high and low waters. Including this option gave better results,
but the results without this option were still quite good. Foreman and Henry also provide a program
and manual for carrying out the harmonic analyses of high and low waters. Because there are much
fewer data points when analyzing high and low waters (versus analyzing hourly data) their high-low
water analysis program uses a different least squares solution technique than the program they use
to analyze hourly data (Foreman, 2004a).
Theoretically, with such least squares methods, one may not be able to accurately calculate
overtides or other higher frequency compound tides, since the (approximate) sampling rate
determines the highest frequency that can be calculated and when using only high and low waters
(an approximate sampling interval of 6 hours) the upper limit is somewhere between the semidiurnal
and quarterdiurnal bands. However, was seen with Schuremans method, one should at least be able
to approximate the M4 constituent by looking at the duration of the rise of tide from low water to
high water.
3.6 Nonharmonic Comparison Methods
3.6.1 Overview and short historical background

In the past, tidal analysis and prediction manuals or books have usually paid most attention to
harmonic or other frequency-domain methods. That is understandable from the standpoint that such
methods provide the greatest prediction accuracy and also the greatest understanding of the tidal
dynamics. However, throughout history the methods that have been used most often for tidal
prediction have been nonharmonic methods. Such methods relate the tide, both its height and its
timing, to either the passage of the moon overhead (and the day of the lunar month) or to the tide
109
at another location (where it is presumably well known and accurately predicted, being based on
harmonic analysis). The earliest methods of tide prediction by ancient man correlated the times of
high water with the moons passage overhead, and also, of course, correlated the size of the tide
range with the phases of the moon. (The latter was probably what convinced ancient man that there
was some type of connection between the moon and the tide.)
The worlds oldest known tide table, produced by the Chinese (more than 200 years before the
English tide tables for London Bridge) to predict the occurrence of the tidal bore in the Qiantang
River near Hangzhou, was based on a nonharmonic analysis. The tide table was first carved in stone
on the Zhejiang Ting pavilion on the bank of the Qiantang at Yanguan, and later a printed version
was produced in 1056 A.D. The tide table gave the time for two occurrences of the bore for each
day (to the nearest 2 hours) for each lunar month in each of four seasons of the year. The two-hour
time intervals of each day were called the twelve earthly branches, which began at 11:00pm, each
branch being named for an animal rat, ox, tiger, hare, dragon, snake, horse, sheep, monkey, cock,
dog, and pig. The heights of the bore, from lowest to highest, were seven characters translated as
lowest, very low, low, fairly high, high, very high, and highest (see Zuosheng, et al., 1989) The
English tide table for London Bridge was also based on a nonharmonic method and simply related
the tide to the age of the moon (the number of days since the last new moon). As was mentioned
in Section 1.4, John Lubbock greatly perfected the nonharmonic method for tide tables he produced
in England in 1832. Eighty years earlier Daniel Bernoulli also produced very useful tide tables in
France using a nonharmonic method.
Those earlier nonharmonic methods compared the elevation and times of high waters (and
sometimes low waters) to various astronomical occurrences such as the moons transit, times of new
moon and full moon, times of apogee and perigee, etc. More recently, and for many decades now,
the main use for nonharmonic tidal analysis has been to compare one or more aspects of the tide (or
tidal current) at two tide (or tidal current) stations. A typical nonharmonic comparison analysis
might, for example, calculate the time difference between the occurrence of high water at two tide
stations, as well as the height difference, and similarly for low waters at the two stations. Although
this type of analysis is not very sophisticated, it has been used to analyze data at thousands of
locations, i.e. at the so-called subordinate stations (or secondary ports) in national tide and tidal
current tables. In the U.S. Tide Tables, daily predictions of high and low waters for particular
reference stations (called standard ports in other nations), produced using harmonic constants, are
found in Table 1. The nonharmonic comparison analysis is used to compare a subordinate station to
a reference station, and the resulting time differences and height differences are listed in Table 2.
The same is true for current data, but here there are four sets of differences in Table 2, for maximum
flood, maximum ebb, slack before flood, and slack before ebb. In some cases, where the current is
not strictly reversing the slacks are replaced with minimums. In some cases current speed
comparisons between the subordinate and reference station are handled with ratios instead of
differences, and in some rare cases both differences and ratios are used.
The use of nonharmonic comparison methods has been necessary for two reasons. First, there
is limited space in the annually published paper Tide Tables. If daily predictions (i.e., the times and
heights of high and low waters) were included for the thousands of tide stations in these tables, they
would have many thousand of pages (typically there have been four pages per harmonically
predicted reference station in Table 1 of the Tide Tables). However, the time and height differences
for a subordinate station (relating it to a reference station) takes only one line in Table 2 (with
typically 60 stations to a page). Thus, the only way to manageably include the thousands of stations
is to calculate time and height differences using a nonharmonic comparison method. Second, in the
110
past at least, many stations had too few days of data to carry out a reliable harmonic analysis. This
was especially true of stations with current data. Those data records were typically short because
current meters were much more difficult to deploy and maintain compared with land-based coastal
water level gauges.
An average time or height difference that comes out of a nonharmonic comparison analysis is
implicitly assumed to be constant and to apply at all times so that it can be used to adjust the
harmonically-predicted (e.g) high waters at the reference station in order to predict the high waters
at the subordinate station. However, it is important to remember (and it will be demonstrated in
Section 3.6.3) that such differences and ratios actually vary throughout the month as astronomical
forces change. The closer the reference and subordinate stations are in their tidal characteristics the
smaller this variation will be. But often these two stations are not as similar as one would like, and
this variation throughout the month can be the greatest source of error in producing a tide or tidal
current prediction for a subordinate station. This is especially true in waterways with a significant
diurnal inequality. Such errors are minimized by using a reference station that is close enough to
the subordinate station to have very similar tidal characteristics, but as mentioned above this has not
always been possible in the past. It can be especially difficult for current stations, since the
frequency characteristics of the current can change significantly over fairly small distances by such
things as channel bends and points of land through inertial effects (as seen in Section 2.3.6).
3.6.2 Types of Nonharmonic Analyses
There are three basic types of nonharmonic analysis: (1) the monthly mean analysis; (2) the
mean cycle analysis (or rotary analysis when analyzing current data); and (3) the cycle-by-cycle
analysis (or reversing reduction analysis when analyzing current data). The first two analyses
derive differences between two stations from average values calculated at each station over some
time period (a month for the first analysis). The third method, however, calculates differences
and/or ratios for every individual tidal cycle in a data series. The first and third analyses look at only
key points in the tidal cycle (high and waters or maximum floods and ebb and slacks), while the
second analysis produces an entire mean tide or mean tidal current curve.
The monthly mean analysis is typically done on water level data (U.S. C&GS, 1965). Times and
heights of high water and low water are tabulated at a station and mean values are calculated for one
or more months. These mean results are then directly compared with the mean results from a
reference station to give time and height differences (and/or height ratios). A month-by-month
comparison ignores the greatest tidal variations, those that take place within each month from cycle
to cycle. It also provides no information between the high and low waters, and thus cannot represent
distortions in the tide curve caused by nonlinear shallow-water effects.
The rotary reduction analysis has been used primarily with current data (U.S. C&GS, 1950) and
is so named because it was originally used with data taken from offshore current stations where the
current flow was rotary. At such stations the direction of flow rotated around the compass in one
tidal cycle and no distinct direction of maximum flood or maximum ebb could be determined. In
a rotary analysis the velocity time series is divided into tidal cycles, which are superimposed and
vectorially averaged to produce mean tidal current curves (speed and direction plots, and a polar plot
with the equivalent of a current rose only the tips of each vector are usually connected). The
division of the data series into tidal-cycle pieces is determined by the times of a single key point in
the tidal cycle (e.g. times of maximum flood) at the reference station. The rotary analysis is useful
for all current stations where one is interested in the values in between the times of maximum flood
111
or ebb or the times of slack (minimum). It is especially useful in shallow-water areas where the tidal
current curve can be distorted (away from a perfect cosine curve) due to overtides, and it can even
show double floods or ebbs. The rotary analysis is also very useful in producing tidal current charts
that depict the tidal flow in a waterway for each hour of the tidal cycle; the actual rotation of the
current at each location can then be depicted. The equivalent mean cycle analysis is not often done
with tide data, although a mean tide curve would also be useful in shallow-water areas where the
curve is significantly distorted.
In the cycle-by-cycle analysis (also called the tide-by-tide analysis or, for currents, the reversing
reduction) key points are picked off two time series of data or predictions (U.S. C&GS, 1950). For
tide data or tide predictions, the key points are usually high water and low water. For current data
the key points are usually slack (or minimum) before flood, maximum flood, slack (or minimum)
before ebb, and maximum ebb. The time differences between corresponding key points at the two
stations for each tidal cycle are tabulated and then an average value is calculated. Heights at
corresponding high or low waters, or speeds at corresponding maximum floods or ebbs, are similarly
compared; the latter generally is in the form of ratios, while the former has been typically been in
the form of height differences, height ratios, or a combination of both. These methods depend on
having an accurate and consistent method for selecting maximums and minimums and slacks from
water level and current data times series. This especially becomes an issue when shallow water has
distorted the tide or tidal current curve. As will be seen in the next section, the cycle-by-cycle
analysis is very useful because it allows us to see how the differences between the two stations vary
throughout the month, thus giving us some idea of the quality of a tide prediction made using these
differences in Table 2 of the Tide Tables or Tidal Current Tables.
3.6.3 Variations In Nonharmonically Determined Time and Height Differences Throughout

a Month
There is one key fact that must always be remembered when carrying out a nonharmonic
comparison analysis (and especially when putting the results in Table 2 of a Tide or Tidal Current
Table):
the time and height differences between corresponding high waters (or low waters) at
two tide stations vary from tidal cycle to tidal cycle throughout the month (and similarly
for tidal current stations).
The greater the difference between the tidal harmonic characteristics of the two stations, the larger
the variation in time and height difference (or current speed ratios) will be throughout the month.
This variation can be seen when using cycle-by-cycle analysis; it cannot be seen when using a
monthly mean analysis or a rotary reduction analysis.
If a cycle-by-cycle analysis is done on observed data, then the time and height differences can
include the effects of nontidal influences such as wind or river discharge, in addition to the effects
of changing astronomical conditions. However, if a cycle-by-cycle analysis is done to compare tide
predictions from two stations, then the variation in these differences are solely the effect of changing
astronomical conditions at the two stations (which do not have identical tidal characteristics). In this
latter case, the variation in differences over the month, around a mean difference, corresponds to the
error that would result when making a prediction at a subordinate station using the mean differences
112
that come out of a nonharmonic comparison analysis (i.e., the value that is put in Table 2 of the Tide
Table or Tidal Current Table). Thus, the cycle-by-cycle analysis can serve an important role in the
quality assurance analysis of subordinate stations. If there is too much variation in the time or height
differences over a month, then one needs to use a different reference station, one with more similar
tidal characteristics (usually geographically closer).
The average time difference between, for example, high waters at the two stations will
approximately equal the time difference between the two respective M2 high waters. But other
constituents, each with different phase (time) differences, will also affect high water, so that the time
difference between high water at the two stations will vary throughout the month, i.e. from springs
to neaps, and from apogee to perigee, and from maximum lunar declination to equatorial declination
to southern declination and back.
Generally, the greater the geographical distance between two stations the greater the difference
in harmonic makeup and the greater the variation in time differences. The key consideration is how
quickly the tidal characteristics change with distance in the vicinity of the two stations. When
producing a tide or tidal current table, the intention, of course, is to have enough reference stations
so that no subordinate station is forced to be associated with a reference station whose tidal
characteristics are dissimilar enough that significant variations in time differences and other
differences or ratios will occur. In mixed tidal regimes the reference stations need to be closer
because the diurnal-to-semidiurnal ratios change more quickly with distance than do the ratios
between two semidiurnal constituents. The phase differences between K1 and between O1 at two
stations, have a significant effect on the average (essentially M2) time differences.
This will be illustrated with examples from San Francisco Bay, where the tide and tidal current
regimes are mixed and there is a significant diurnal inequality, so that the diurnal-to-semidiurnal
rations change relatively quickly with geographic distance. Figure 3.5 shows the tidal harmonic
characteristics for two water level stations and two current stations in San Francisco Bay. The two
tide stations, at Golden Gate and Port Chicago, are only about 18 nautical miles apart. One-year
least-squares harmonic analyses were carried out for both stations for the year 1980. The
(K1+O1)/M2 ratio is 1.048 at Golden Gate and 0.900 at Port Chicago; the phase differences,
(M2o-K1o-O1o)/2, are 247o and 241o, respectively. One-year predictions were made at both these
stations, and the times and heights of high water and low water were determined for each set.
Cycle-by-cycle comparisons of the two stations were made. Figure 3.6a shows a plot of the
differences between predicted low water times at Port Chicago and predicted low water times at
Golden Gate, for the first 120 tidal cycles of 1980. The solid line connects the first of each pair of
cycles, and the dashed line connects the second of each pair.
The average time difference between low water at the two stations is 2.99 hours; namely, low
water at Port Chicago is 2.99 hours later than low water at Golden Gate, on the average. It can be
seen from the plot, however, that this average value is only good at certain times of the month. Over
the first 120 cycles the time difference can be as large as 3.43 hours and as small as 2.47 hours. The
maximum time difference occurs in the first of each pair of cycles, just after maximum southern
lunar declination (S).
If Golden Gate is used as a reference station and Port Chicago is used as a subordinate station,
then 2.99 would be the value that would appear in Table 2 of the Tide Table for Port Chicago. A
prediction of low water at the subordinate station Port Chicago, calculated by using this 2.99-hour
time difference to adjust the predicted low waters at the reference station (Golden Gate), would give
tide predictions for Port Chicago that would have an error ranging from -0.44 hour to +0.52 hour.
113
Figure 3.5. The harmonic characteristics for two water level stations and two current stations
in San Francisco Bay, where the tide regime is mixed. (From Parker, 1991b)
Such errors would be even larger at locations further up the estuary. Similar variations in time
differences are also seen for high waters.
Figure 3.6b shows a similar plot of the differences between the heights of predicted low waters
at Port Chicago and the heights of predicted low waters at Golden Gate. The average height
difference for low waters at the two stations is -0.41 foot ; namely, low water at Port Chicago will
be 0.41 foot (0.12 m) lower than low water at Golden Gate, on the average. However, it can be seen
from the plot, that the height difference can as great as as -1.10 feet (with Port Chicago much lower
than Golden Gate) or as great as +0.70 foot (with Port Chicago actually higher than Golden Gate).
The greater height differences occur just after maximum southern lunar declination (S) for the first
of each pair of cycles. Again, -0.41 is the value that would appear in the tide table for Port Chicago.
A prediction of low water at Port Chicago calculated by adjusting the predicted height at Golden
Gate would have an error ranging from -1.11 feet to+0.69 foot. These San Francisco Bay results
114
Figure 3.6a. Plot of the differences between

harmonically predicted low water times at Port
Chicago and at Golden Gate, for the first 120
tidal cycles of 1980. The solid line connects the
first of each pair of tidal cycles, and the dashed
line the second. N and S are the times of
maximum north and south lunar declination.
(From Parker, 1991b)
Figure 3.6b. Plot of the height differences between

the harmonically predicted low waters at Port
Chicago and at Golden Gate. (From Parker, 1991b.)
115
suggest that the use of a second reference station in the northern part of the estuary would improve
the tide predictions for stations in that area. (This in fact did happen and Port Chicago was added
as a reference station in the U.S. Tide Tables because of the results displayed in Figures 36a and
36b.) But in some areas a second reference station may not always be possible until a new long-term
station is installed.
Shallow-water-generated distortions in the tide or tidal current may also lead to variations in
time differences between two stations. An example is provided by the results of a reversing
reduction analysis of two current stations in Delaware River and Bay that show sudden large
maximum ebb time differences at certain times of the month, for the first of each pair of cycles (see
Figure 3.7). These sudden jumps in the plots are caused by the effect of higher-frequency
shallow-water constituents (the overtides) on the times of maximum at Philadelphia, where the
overtides are larger than at the Bay entrance. Although these constituents are not large enough to
greatly change the speed value of maximum ebb, they do distort the shape of the tidal current curve
enough to shift the times of maximum ebb later or earlier by a considerable amount. Examples of
the shifting of high and low waters in the tide were seen in Section 2.2.5. Figure 3.8 shows an
example of this with the tidal currents near Philadelphia (see also Figure 9.3).
When the overtides were removed from the Philadelphia prediction the time differences between
Philadelphia and the bay entrance became quite small (Figure 3.9). Here the method of selecting
maximum flood (see Section 3.9.2) makes a difference. In cases like this where the time of (in this
case) maximum flood is shifted by the shallow-water distortion of the tidal current (but the
maximum speed is not changed much), it would make more sense not to use the time of the actual
maximum flood value, but to use a time more in the center of the flattened out area, e.g. the center
Figure 3.7. Plot of the time differences between

harmonically predicted maximum ebbs at two station in
Delaware River and Bay. The solid line connects the
first of each pair of tidal cycles, and the dashed line the
second. (From Parker, 1991b.)
116
time of all values greater than 90 percent of the actual maximum. This would provide more
consistency from cycle to cycle in such flattened out cases, while still accurately picking maximums
for strongly distorted, steeper cases.
Tidal currents have an additional problem. It was noted at the end of Section 2.3.6 that the
lateral advective/inertial terms in the 2-dimensional momentum equation (see also Section 7.3.2)
shift energy into the next harmonic (especially off points and around channel bends), changing, for
example, the (K1+O1)/M2 ratio or the M4/M2 ratio in the tidal current (see Figure 2.35). The area
around the San Juan Islands (between British Columbia and the state of Washington), including the
Strait of Juan de Fuca, Haro Strait, and many other connecting waterways, provides an extreme
example. Here a combination of the large diurnal inequalities and the inertial effects due to the
complex geometry lead to tidal current characteristics that can change dramatically over a short
distance. While the (K1+O1)/M2 ratio for the tide changes gradually and smoothly in this area, the
(K1+O1)/M2 ratio for the tidal current varies considerably (Parker, 1977). The six reference stations
in the NOS Tidal Current Tables for this area are not really sufficient and there are times during the
month when some subordinate stations will become diurnal a day earlier or later than the reference
station, making it almost impossible to make a prediction using a nonharmonically-derived
correction. Even at times of the month when both reference station and subordinate station have not
become diurnal, there is large variation in time differences and speed ratios. One really needs
several values for each tidal parameter (e.g. one set for maximum southern declination, one set for
equatorial lunar declination, and one set for maximum northern lunar declination), and that may still
not work well enough.
Although the Tide and Tidal Current Tables must use nonharmonically-produced correction
factors for thousands of stations in order to avoid adding thousands of extra pages to the Tables, as
will be seen in Section 9.7, there is no reason to continue using this method in digital tidal prediction
products, which with todays computer power can be based totally on harmonic methods. Most
subordinate (i.e., Table 2) water level stations usually have at least 29 days of data, and thus
harmonic constants are available. However, most subordinate stations in Tidal Current Tables have
less than 15 days of data, and so new current data would have to be obtained. The improvement in
the tidal prediction accuracy would be significant, especially in waterways with a strong diurnal
signal. CO-OPS is, in fact, moving toward delivering harmonically based predictions for more
locations than just reference stations.
Chapter 6 will look at how to carry out the various kinds of nonharmonic tidal analysis of water
level and current data times series, the best methods for including the analysis results in Table 2 of
the Tide and Tidal Current Tables, and ways to minimize the errors that can easily result from
making predictions using the corrections factors put in Table 2.
117
Figure 3.8. Predicted tidal current at Philadelphia on the

Delaware River illustrating the dramatically varying time
interval between successive maximum flood and between
successive maximum ebbs due to shallow-water effects.
(From Parker, 1991b.)
Figure 3.9. Plot of the time differences between

harmonically predicted (but leaving out the overtides and
compound tides) maximum ebbs at two current stations in
Delaware River and Bay.
118
3.7 Analyses For Determining Sa and Ssa7
3.7.1 The Meteorological Origins of Sa and Ssa
Sa and Ssa are long-term tidal constituents directly generated by the nonuniform changes in the
Sun's declination and distance (the perihelion-aphelion effect). The solar annual constituent, Sa, has
a period of 365 days (a frequency of 0.0027 cpd) and the solar semiannual constituent, Ssa, has a
period of 183 days (a frequency of 0.0055 cpd). Sa and Ssa are very small compared to other tidal
constituents, and insignificant for most applications. However, if one includes Sa and Ssa in a year-
long harmonic analysis of water level data one may initially be surprised that their calculated values
come out much larger than would be expected from astronomical forcing. This is because the
energy at the one cycle per year (Sa) and one cycle per half year (Ssa) found by the analysis is
actually meteorological in origin, namely, caused by the seasonal changes in wind, temperature, and
atmospheric pressure that affect water level.
For example, summer heating of the upper layer of ocean causes the water to expand, raising the
coastal water level. This is usually referred to as a steric sea level change. The expansion in the
summer and contraction in the winter puts energy into the Sa constituent. Since this annual raising
and falling water level due to annual warming and cooling is not perfectly sinusoidal (that is, there
is some asymmetry in this variation), some of the energy will also show up in the next higher
harmonic, i.e., at two cycles per year, which is the Ssa constituent. For example, if most of the
heating and expansion took place in just a few summer months and the rest of the year there was
slow cooling and contraction, that asymmetric effect would put some of the energy into Ssa.
Sa and Ssa can be produced by the seasonal cycle of any meteorological effect on water level.
There are, for example, annual patterns of changing coastal winds, that during part of the year can
push the water toward the shore, raising the water level, and during other parts of the year push the
water away from the shore, lowering the water level (putting energy into Sa). And again whatever
asymmetry is involved will also put energy into Ssa. Similarly, seasonal changes in atmospheric
pressure will push the water down or lift it up (the inverted barometer effects), and changes in river
discharge will also raise the water level in a river (during spring freshets), the latter effect being very
asymmetric, since the largest water level increase usually occurs during only one month of the year.
Similar seasonal effects can also be seen in long-term current records, where other effects, such as
seasonal changes in salinity, may also come into play.
One can get an idea of how large and varied (from location to location) meteorologically caused
Sa and Ssa can be by looking at Figure 3.10. This figure shows the mean annual sea level cycle at
13 water stations along the East Coast of the U.S. and part of the Gulf Coast, as well as at the island
of Bermuda east of the U.S. East Coast. (Since the tide is being averaged out as well as other higher
frequency water level variations, this is referred to as sea level.) This mean seasonal sea level cycle
was produced by taking monthly averages of 37 years of water level data at each station (1948-1984;
which includes two 18.6-year periods). One can see a clear annual cycle at each station, but many
of the stations have strong asymmetries. The northern most station, at Halifax, Canada, has an
almost sinusoidal shape, but the stations get more and more asymmetric as one moves south. For
most of these stations the maximum sea level value occurs in September or October. The asymmetry
gets so strong at Charleston and Fernandina Beach that their annual cycle curves show double
maximums, the larger one peaking in October, but a second smaller one peaking at the beginning
of June. The range of the annual cycle for each station, which is given in the figure to the right
119
of each station name, is largest at Charleston and Fernandina Beach, and generally larger in the south
than in the north.
Bermuda, which is far to the east of the U.S. East Coast, has a sinusoidal annual sea level cycle,
which is due to steric effects, namely, due to a similar variation in annual water temperature. With
most of the coastal stations, the annual cycle of varying wind speed and direction is the dominant
cause of the annual water level cycle. Seasonal changes in the Gulf Stream transport may also have
an effect on the seasonal water level curves at stations along the southern part of the U.S. East Coast
(see Noble and Gelfenbaum, 1992), with maximum transport causing lower sea levels at the coast,
and vice versa. The largest Gulf Stream transport usually occurs in July-August, which matches a
minimum sea level at Charleston, Fernandina Beach, and Miami, while the lowest Gulf Stream
transport usually occurs in October which matches the maximum sea level at these three stations.
Extensions of the Gulf Stream, in the form of the Florida Current and the Loop Current (in the Gulf
of Mexico) might also account for the slight dip in sea level in July at other stations heading into the
Gulf of Mexico.
The variation in annual cycle and semiannual cycle seen in Figure 3.10 (from Parker, 1994) is
not an isolated example. Similar variation in these cycles is seen in the results of Tsimplis and
Woodworth (1994), who used data from 1043 water level stations around the world to determine
the amplitude and phases of the annual and semiannual cycles at each location.
Seeing these examples of numerous meteorologically caused seasonal changes in sea level, one
could rightly decide not to use any Sa or Ssa values that come out of a harmonic analysis when
making a tidal prediction. Some oceanographers have, however, proposed that to make the best
water level prediction possible, one could include these meteorologically caused Sa and Ssa
harmonic constants. That would make sense if the calculated Sa and Ssa values were expected to
be consistent from year to year. In fact, they can vary fairly dramatically from year to year, as can
be see in Figure 3.11, which shows the variation from year to year of the seasonal sea level range
from 1935 to 1992 at four stations along the U.S. Atlantic coast. In this figure one can see that (e.g.)
the three southern stations had large seasonal sea level ranges in 1958 and small ranges in 1980. At
Fernandina Beach the mean sea level range was 0.99 feet (from Figure 3.10), but its seasonal range
in 1958 was 1.92 ft and in 1980 its range was 0.54 ft. From this one can see that Sa and Ssa
(obtained from one-year harmonic analyses) would vary considerably from year to year.
3.7.2 Considerations In Whether To Use Sa and Ssa In a Prediction
From the results presented in the last section, one would still not know what value of Sa and Ssa
should be used for a future year. Choosing an Sa and Ssa for next year, the mean values obtained
from Figure 3.10 could be a poor choice, because (from the variation seen in Figure 3.11) they might
do a poor job of representing that year. One might try projecting into the future based on a figure
like Figure 3.11 (but more specifically for values of Sa and Saa for each year), but, from the sudden
changes one sees in Figure 3.11, that may not do a very good job. Including Sa and Ssa in a self
prediction (i.e., predicting the same year as the data used to produce the harmonic constants) can be
justified, if one is looking to make a water level prediction (i.e., the tide plus any other known
periodic water level variation) as close as possible to an actual water level record, and to reduce the
energy in the residual time series. But for a prediction in another year, this is not likely to be the
case, because of the significant variation from year to year in the meteorologically caused variations
at the annual and semi-annual frequencies that are not predictable.
120
Figure 3.10. The mean annual sea level cycle at fourteen NOS stations along the East Coast and
Gulf Coast of the U.S. for the period 1948-1984. The number to the right of each station name gives
the mean range of that stations seasonal cycle. All values are in feet; increments along the vertical
axis are 0.2 foot. (From Parker, 1994.)
121
Figure 3.11. The variation from year to year of the seasonal sea level range from
1935 to 1992 at four NOS stations along the U.S. East Coast. Increments along the
vertical axis are 0.2 foot. (From Parker, 1994.)
Instead of using the Sa and Ssa values that come out of a least squares harmonic analysis of a
years worth of data, sometimes a special analysis is carried out with many years of data, which
produces much smaller values of Sa and Ssa, more representative of an average annual and semi-
annual cycle. Although much smaller than the values from an individual year analysis, these Sa and
Ssa values still contain meteorological energy and still are greater than if they truly represented only
the astronomical tidal energy at those frequencies. They are usually too small to make much
difference in a water level prediction.
So whether the calculated values of Sa and Ssa are used or not, becomes almost a philosophical
question, based on ones application, the amount of data one has, and the dynamics at the particular
location from which the data came. Often tide predictions are subtracted from actual water level
data in order to look at only the nontidal signal (and perhaps correlate it with changes in the wind
and barometric pressure or steric changes). In such a case, one would not want to use the essentially
meteorological Sa and Ssa harmonic constants in the prediction. Even for producing the most
accurate water level predictions for other purposes, for many locations the use of the Sa and Ssa
may not change the prediction that much, and may not represent the annual cycle for a particular
year very well anyway.
3.7.3 Methods for Calculating Sa and Ssa
Today the simplest (and perhaps best way) to calculate Sa and Ssa is simply to include them in
a one-year least squares harmonic analysis, and to carry out as many one-year analyses as possible
122
with the available data. Then one can see how Sa and Ssa varies from year to year. One can also
do longer analyses (e.g., two-year or five year analyses, or even a 19-year analysis) to see how much
the amplitudes are reduced and how the phase lags are changed in the longer analysis. Greatly
reduced amplitudes in the longer analysis compared with the shorter analyses indicates that the
inclusion of Sa and Ssa in an accepted set of harmonic constants for a particular station will not be
very useful for a tide prediction.
There are also methods to extract Sa and Ssa from tabulations of monthly means of mean sea
level. This again can be done with a least squares harmonic analysis. In the pre-computer era, other
methods were devised. Schureman (1958, page 114) provides a method for determining Sa and Ssa
from monthly means of sea level using C&GS Form 194. This form was originally designed for a
regular harmonic analysis using hourly means. Such analyses were done by hand so the various
steps in the arithmetic on the form were designed with labor savings in mind (and so the method
behind may not be very clear), and it also involves the use of logarithms (unnecessary now in the
computer age).
If one decides that they want to include Sa and Ssa in a set of harmonic constants, it is probably
wisest to analyze a 19-year set of monthly means, or to include Sa and Ssa in a least squares
harmonic analysis of a 19-year water level time series.
3.8 Tidal Filtering and Detiding
3.8.1 Introduction
There are often applications where one wishes to remove the tidal energy from a data time series.
A good way to accomplish that is to do a harmonic analysis of the available observed data time
series and then make a tidal prediction for the same time period. Subtracting the predicted times
series from the observed series produces a residual time series, which should contain only (or
predominantly) nontidal energy. This is usually referred to as detiding. Here, of course, one has
the same limitations as discussed throughout this book, namely, the shorter the time series the fewer
harmonic constants that can be used to make the prediction. However, this limitation may not be
quite as serious a problem for this application, since here one is making a self-prediction (i.e.,
predicting for the same time period as the analyzed data series). Energy from the unsolved-for
constituents is included in the solved-for constituents, and although the harmonic constants that
come out of a short harmonic analysis may not do a good job of predicting for other time periods
(since the phase relationship between the unsolved-for constituents and the solved-for constituents
will be different and unknown), they are able to do a fairly good job predicting for the period of the
analysis. Still, if the data time series is very short, for example less than 15 days, this harmonic
method of producing a residual series may not do a good enough job of removing the tidal energy.
The alternative (and much more direct and economical) method for removing tidal energy is to
use filtering. Filtering in its simplest form is merely taking running averages (running means) as
one moves through a data time series. For example, suppose one wants to remove M2 energy from
a data time series that has a sampling rate of one hour. M2 (frequency = 1.93 cpd) has a period of
12.42 hours, so if one averages 13 hourly data points, most of the M2 energy should cancel out, since
the M2 contribution would have gone through an entire M2 cycle. Since, for this example, a 13-hour
running mean filter (a 13-hour box filter) is being used, and the M2 period is 12.42 hours, there will
actually be a small amount of M2 energy left, an important concern as will be seen. To filter out (a
lot of) the M2 energy from a time series, one can average the first 13 data points (call them data point
123
Figure 3.12a. Filtering a data time series. In this simple

example each group of 13 points is averaged to produce one
data point in the filtered series (see text).
1 through 13) and save the averaged result (it will have the same time as original data point
number 7) , then average data points 2 through 14 and save the result (at the time of original data
point 8), then average data points 3 through 15, etc, etc (see Figure 3.12a). When this is finished
there will result a new (filtered) time series with most of the M2 energy eliminated (see Figure 3.12b,
but this actually shows better results than a simple box filter for M2 could accomplish.). Because
of the filtering process, this new filtered time series will be 12 points shorter than the original time
series, beginning 6 hours later and ending 6 hours sooner. As a result of this process, one has also
taken out some of the S2 tidal energy, but not as much as the M2 energy, since the S2 period is 12.00
hours and thus averaging over 13 hours would leave more of the S2 energy. To eliminate the energy
at all the tidal frequencies one would have to go through this process many times, using running
means of different lengths. This would be very time consuming, and with each pass through with
another running mean our filtered time series would become shorter and shorter. One could instead
use a 25-hour running mean filter, which loses only 24 data points and filters out energy from both
semidiurnal and diurnal tidal constituents. However, this filter still leaves too much tidal energy for
most applications.
To avoid this problem, special filters have been designed to eliminate most tidal energy on one
pass through the time series. In this case the process is to take weighted running averages as one
moves through the time series. The way this is usually represented is the following:
n
y(t) = + 3 W i x(t-i)
where i = -n
y(t) = a single data point in the output filtered time series (at time t)
x(t-i) = a single data point in the input data time series; 2n+1 data points
of these input data points (centered at time t) are averaged to
produce a single output data point, y(t)
Wi = one of the 2n+1 weights that multiply times the 2n+1 input data
points
2n+1 = the length of the filter (i.e., from i= n to i=n ), i.e., the number of
weights in the filter, and the number of input data points averaged
to create each new filtered output data point
An odd number of weights are usually chosen, with values symmetric about t , so that the filtered
time series will occur at the same times as the original data series (minus the 2n data points lost, of
124
Figure 3.12b. A water level time series (solid line)

filtered to remove the tidal oscillations and leave a
predominantly nontidal time series (dotted line).
course). Each weight has a value less than 1, and the sum of the 2n+1 weights must equal 1, so that
no energy is artificially taken away or added. These two requirements are written:
n
W i = W-i and 3 Wi = 1 .
i = -n
For the simple running mean the Wis are all equal to 1/(2n+1). Thus in the simple M2 example
above, the Wis are all equal to 1/13 (= 0.077). For a 25-hour running mean the Wis are all equal
to 1/25 (= 0.04).
The next section will look at several filters specially designed to filter out most tidal energy.
Each filter will have different capabilities and different drawbacks.
When using tidal filters it is very important to remember that a tidal filter will also eliminate
energy at other nontidal frequencies. Such filters reduce nontidal energy not only at frequencies
near tidal frequencies, but also at frequencies not close to tidal frequencies. For example, although
tidal filters are generally low pass filters (i.e., they eliminate most high frequency energy and allow
low frequency energy to pass through), a typical tidal filter can still reduce some subtidal energy
by as much as 60%.
When deciding to use tidal filtering (instead of detiding the data series by creating a residual
series using tidal harmonic prediction) one must look at the intended application. If one wishes
primarily to eliminate higher frequency energy (including tidal energy) from a data record, then tidal
filtering is a reasonable choice. However, if one does not want to lose 60% of the subtidal energy
as a result of this filtering, then it would be wiser to detide the data series using tidal harmonic
analysis. This detided series will still have higher frequency nontidal energy, but that can then
eliminated with a shorter filter.
There are other possible situations where tidal filtering would be better to use than creating a
residual series using tidal harmonic prediction. As was seen in Section 2.3.3, in shallow waterways
nontidal phenomena such as river flow and low-frequency storm surge can affect the amplitudes and
phases of tidal harmonic constants obtained from a harmonic analysis. Thus, one should not create
a residual series using accepted harmonic constants (probably calculated from a one-year analysis),
because those constants will not predict the tide accurately for a time period with strong nontidal
influences (because the tide was changed by that strong nontidal influence). Because those
125
Figure 3.13. Water level data from Trenton, NJ, filtered with
a tidal filter, and compared with river discharge data also
filtered with the same tidal filter. (From Parker, 1984.)
harmonic constants are wrong for the period of the nontidal event there will still be tidal energy in
the residual time series, whereas tidal filtering will eliminate it. Even trying to calculate new
harmonic constants just for the time period of the (e.g.) strong river flow will not work well, because
such events are usually shorter than the length of series needed for a reasonable harmonic analysis
(although one could try the continuous wavelet transform method; see Section 3.5.5).
If one decides to use tidal filtering, and one is going to compare the filtered tidal series to some
other data time series, it is usually wise to also filter that other data time series with the same tidal
filter (even if one does not expect there to be tidal energy in that other data time series), so that the
nontidal energy in both times series will be treated the same way. This is because, as mentioned
above, the tidal filter will also remove some nontidal energy, and if one is looking to compare the
results of a filtered water level time series to (e.g.) a wind or pressure data series, they should also
be filtered in order to take out the same percentage of nontidal energy. Figures 3.13 and 3.14
provide two examples of this.
Figure 3.13 shows a tidally filtered water level record from Trenton, NJ, during a heavy river
runoff period on the Delaware River and a river discharge record for the same time period (also from
Trenton, but above the rapids) which was also tidally filtered even though there was no tidal signal
in the river discharge data. The purpose of this application was to look for similarities in the two
resulting curves that might indicate how much the river discharge affects the water level at Trenton.
Figure 3.14 illustrates an application of tidal filtering in Massachusetts Bay, where tidally
filtered water level records are compared with tidally filtered wind records. In this case, two water
levels records (Boston and Sandwich) at opposite ends of the Bay (see map in Figure 3.14) were
tidally filtered and subtracted from each other to look at the changing sea level difference between
the two ends of the Bay. The wind record for Boston (Logan Airport) was also tidally filtered (and
squared, since wind stress was being looked at). The component of this wind stress along the axis
of the Bay is plotted in the figure along with the sea level difference curve (between Boston and
126
Figure 3.14. Plots (right panel) of the difference between two tidally filtered water level records
at opposite ends of Massachusetts Bay (left panel) compared with plots of wind stress along the
axis of the Bay, which were similarly filtered. (From Parker and Pearce, 1975.)
Sandwich) to see how much the wind stress along the longitudinal axis of the Bay affected the
longitudinal tilt of the sea surface.
3.8.2 Specially Designed Tidal Filters
Numerous filters have been developed to eliminate energy at the tidal frequencies (see Groves,
1955, and Godin, 1972), each having specific weights (W is) and specific lengths (2n+1). Such
filters are usually designed to try to accomplish three objectives: (1) to eliminate the maximum
amount of tidal energy; (2) minimize the amount of nontidal energy eliminated; and (3) to have
minimum adverse effects on other frequencies through side-band effects. The box filter, which is
the running mean (see Figure 3.15), is especially bad in terms of side-band effects, that is, it adds
energy to higher harmonics of the frequency being filtered for. Other filters can have similar effects.
In the right-hand plot in Figure 3.15, which shows the effect of the running mean in the frequency
domain, one sees that it passes through all the energy at zero frequency and almost none at the
frequency being filtered for, but it also filters out some of the low-frequency energy. One also sees
the side bands, which are a result of the abruptness of the box filter (running mean), i.e., it has a
value of 1.0 and then suddenly a value of 0.0. Filters which have a gradual decrease from the center
(e.g., Gaussian and cosine filters) show a frequency response that gradually decreases from the
frequency, without side bands effect, but unfortunately filtering out more low-frequency energy.
Filters generally are classified as: low-pass filters, i.e., they allow energy to be passed through
below a designated frequency; high-pass filters, i.e. they allow energy to be passed through above
a designated frequency; band-pass filters, i.e., they allow energy to be passed through between two
127
designated frequencies; or band-stop filters (or band-rejection filters), i.e., they pass through the
energy at most frequencies except those between two designated frequencies. A high-pass filter is
the opposite of a low-pass filter, and a high-pass filtered time series can be obtained by subtracting
a low-pass filtered time series from the original data time series, and vice versa. Similarly, a band-
stop filter is the opposite of a band-pass filter, and one can get one filtered time series by subtracting
the other filtered time series from the original data time series. Since tidal filters are designed to
eliminate the energy in the tidal bands (diurnal, semidiurnal, and higher harmonic bands) they are
band-stop filters, but if most of the surrounding frequencies and higher frequencies are also reduced,
then they are essentially low-pass filters.
One of the earliest and most widely used tidal filter is the Doodson filter (Doodson and Warburg,
1941; they referred to it as multipliers of sea level). The Doodson filter has a length of 39 hours
(i.e., 2n+1 = 39), and the following 39 weights:
0 for i = 0, 5, 8, 10, 13, 15, 16, 18
9 2/30 for i = 1, 4, 9
W i = 1/30 for i = 2, 3, 6, 7, 11, 12, 14, 17, 19
According to Groves (1955), the Doodson filter eliminates 99.94 percent of tidal energy at the
semidiurnal frequencies , 99.79 percent of the tidal energy at the diurnal frequencies, and 99.38
percent of the tidal energy at the overtide frequencies. However, it also reduces nontidal energy at
subtidal frequencies, e.g. by 60% at 0.5 cpd and by smaller amounts below that frequency. It has
generally smaller side bands (than the 24-hour running mean) at the first few harmonics, but
surprisingly high sidebands at the 9th harmonic, and so for some cases additional high-frequency
smoothing may be recommended if the original data series has energy in this area of the frequency
domain. (The Doodson filter was the filter used to produce the plots in Figures 3.13 and 3.14.)
The Doodson filter was originally developed by combining other simple filters, each of which
discriminated against one or more tidal constituents. Groves (1955) used a similar method of
combining simple filters to produce many other tidal filters. There are numerous papers on the
subject of tidal filtering and many opinions on which filters work the best. Emery and Thomson
Figure 3.15. The simple box filter (running mean) and its effect is the frequency
domain.
128
(2001) have a good discussion on filtering and seem to like Butterworth filters. In the end it comes
down to the particular application and probably personal taste.
3.9 Calculating Consistently Defined Maxima and Minima
3.9.1 Consistently Defined High and Low Waters
The tidal prediction equation is generally used to predict a series of tidal heights from which the
times and heights of high and low waters are calculated for each tidal cycle. It is these high and low
waters that are listed in Table 1 of the U.S. Tide Tables, and presented in numerous media outlets
such as in newspapers and on TV news shows during the weather report. Similarly, high and low
waters are also calculated for each tidal cycle of the actual observed water level data time series that
was analyzed. A typical quality control test for the tide prediction is the comparison of the high and
low waters from the predicted tide series with the high and low waters from the actual water level
data time series.
It would seem to be a relatively simple operation to pick off the maximums and minimums from
a tide curve, and that is usually true if the curve is relatively smooth (with little or no noise) and
undistorted by shallow water effects. Actual water level data curves (and especially tidal current
data curves) can be noisy due to wave effects, which can make the determination of high or low
water (or maximum flood or ebb) a little less obvious. There will be no noise in a predicted tide
curve, but it can have distortions due to the nonlinear effects of shallow-water (see Section 2.3.2),
as can the actual data curve. For example, the high water portion of the tide curve might be severely
flattened so that it not obvious what the time of the actual high water should be.
There have been various methods of selecting high and low waters (and maximum flood and
ebbs and slack waters) to handle noise and distortions. Which ever method is used, the important
thing is consistency, namely, to use the same method for all the data and predictions. If one is
comparing the high and low waters at two stations (for a nonharmonic comparison analysis) or
comparing the predicted and actual high and low waters at the same station (for quality controlling
the prediction), is important that the same method of calculating these high and low waters is used
for both time series. As will be seen below, failure to do this can lead to some very misleading
results.
In most cases today the water level data time series will generally be fairly smooth because
modern acoustic water level gauges average out most of the higher frequency waves. With older
water level gauges, a float (connected by a wire to a recording device) moved up and down inside
a stilling well, which protected it from waves, damping out much of the wave effects on the water
level data. Instrument errors, if not removed with some type of quality control methodology can
also add noise and possibly make selection of maximums and minimums difficult. (As will be seen
in the next section, current data is much noisier than water level data, and it can often be more
difficult to pick off the maxima.)
In very shallow waterways the tide curve is distorted by the nonlinear mechanisms (see Sections
2.3.2 and 7.6.2). In such cases the tide curve will no longer be a simple sine curve. Sometimes one
part of the curve will be flattened while the other half of the cycle will have a sharper peak.
Examples were given in Figure 2.26. The extreme case is when one sees double high waters, that
is, when the water level rises to a maximum, starts to fall, but then rises again to a second maximum,
before finally falling all the way down to a low water. Or there can be double low waters, or instead
the curve might flatten out near mean tide level.
129
Prior to the use of electronic computers and automated data processing, high and low waters
were selected by hand from a paper tide roll or some other paper graph. This method continued for
many years even after computers were used for other functions. In order to make this manual
operation as consistent as possible from person to person, the following instructions were given in
the Manual of Tide Observations (C&GS, 1965, page 48):
In selecting the time of high or low water from the tide curve attention should be
given to the general trend of the curve rather than the individual peaks arising for various
causes. The aim should be to take the middle of a smooth arc covering an hour or more
during the high or low water period. ........ In determining the times of the high and
low waters to the nearest tenth of an hour it may be found convenient to construct a small
scale 1 inch long and divided into 10 equal parts for use between hour marks on the curve.
An experience tabulator, however, will usually be able to estimate the tenth accurately
without use of such a scale.
For smooth predicted curves high and low water selection was generally not a problem.
However, for noisy data or for portions of tides curves flattened by nonlinear shallow-water effects,
there might be a difference of opinion on where the actual maximum or actual minimum is located
on the curve. One of the benefits of an accepted computerized routine for selecting high and low
waters, besides the much greater operational efficiency, is the consistency all high and low waters
are selected using the same method. But it is still important to make sure that the same method is
used on both predictions and data. Even after computer routines began to be used, the results were
often checked manually, by looking at the selected high and low waters plotted on top of the tide
curve, and sometimes manual corrections were made.
One basic aspect of all methods for selecting high and low waters from a predicted or observed
time series is keeping tracking of each tidal cycle, since usually there is only one true high water and
one true low water per cycle. But there are two key exceptions that must be accounted for in any
maximum/minimum selection routine.
The first exception is when the nonlinear effects of shallow water distort the shape of a tide
curve and lead to double high waters or double low waters (see Section 2.3.2). The high water/low
water selection method must be able to recognize when the tide curve is rising and when it is falling
and thus be able to keep track of all maximums and minimums throughout each cycle. And, if there
is more than one maximum (or minimum) it must decide whether the second maximum is significant
enough to be considered as a double high water.
The second exception occurs where there is a strong mixed tide regime (see Section 2.1.1), so
that the tide actually is diurnal part of the month. The high water/low water selection method must
be able to handle the transition from the tide having two high waters (and two low waters) per day
to the tide having one high water (and one low water) per day.
A less extreme and more common shallow-water effect leads to a situation that can be very
misleading. Shallow water can, for example, greatly flatten the high water part of the tide curve.
The actual high water is usually not much higher than the rest of the high water part of the tide
curve, but the time of high water can be shifted to the right end or the left end of the high water part
of the tide curve. Often the maximum will be near one end of this plateau, rather than in the middle
(see Figure 3.8), and if this is listed as the high water it can be very misleading (e.g.) to a mariner
using a Tide or Tidal Current Table, since high waters are usually thought of as being in the middle
130
of the high water part of the tide curve. In such cases one would be better off selecting the midpoint
of the high water part of the curve as the time of high water.
Another bad consequence of this type of situation is that the time of high water will often shift
around in time from cycle to cycle, as the relative phases of the (especially overtide) constituents
change. As was seen in Sections 3.6.3 this shifting of the time of the high water can lead to
problems with nonharmonic analyses. It could also lead to great variability in an error analysis
comparing predicted and observed high and low waters.
Many computerized high and low water selection routines use some type of curve fit to select
the maximum or minimum. Often these take into consideration the amount of noise in the data. The
higher the noise level the more data points that are used in the curve fit. Some also look for the type
of nonlinear asymmetric situation described above, where high water may cover a long period as flat
plateau, in which case high water is selected as being midway between the slacks.
Standard computerized algorithms are used by NOS to tabulate and quality control high and low
waters from 6-minute interval water level data. Six-minute data, even with the mechanical and
numerical filtering used in water level measurement systems, can be noisy due to waves, so a
polynomial curve fit is used to determine the times and heights of the high and low waters. The
tabulation routines contain error diagnostics for flat tides, curve fit failures, etc, for the analyst to
review. The algorithms also use some cut-off criteria to eliminate unwanted computer selections
during periods of high frequency noise (e.g., not tabulating preliminary computer selections if they
are not greater than 0.03m apart in elevation or greater than 2.0 hours in duration.
3.9.2 Consistently Defined Maximum Floods and Ebbs and Slacks (or Minimums)
All that has been said in the last section about consistently selecting high and low waters also
applies to finding maximum floods and ebbs and slacks (or minimums) before floods and ebbs (and
one should read that section before reading the present section). However, maximums and
minimums in tidal current predictions and especially current data are usually more difficult to
calculate for a variety of reasons.
First, current data is usually noisier than water level data, especially for offshore current stations.
Of course, this will depend a great deal on the type of current measurement device used, as was
discussed in Section 2.3.6.
Second, tidal currents are much more susceptible to distortion by shallow water and the
distortion can be much more extreme than that seen in the tide. Flattened ebb cycles are very
common, with the choice of the time of maximum ebb not always easy to determine (using the
midpoint of the ebb phase being the safest choice). In Section 2.3.6 a case was shown where the
tidal currents were distorted by shallow-water so much that they are actually quarter-diurnal (with
four tidal current cycles per day) few standardized max/min selection programs could handle that
special case.
Third, tidal currents are most often rotary (see Section 2.3.6 and Section 3.4.3), that is, the
direction of flow of the current rotates around the compass over one tidal cycle. In the extreme case,
in the open ocean far off the coast, the tip of the current vector as it changes direction traces out a
circle, and there is no obvious time of maximum flood or maximum ebb. When such stations are
put in the Tidal Current Tables, no maximums are picked and instead hourly values for a mean tidal
current cycle are put in Table 5 (see Section 9.3.3). In most other cases, ellipses are traced out by
the tip of the current vector as it changes direction, and a clear maximum flood and maximum ebb
are observable. In narrow channels the ellipse is usually so narrow, that the tidal current is
131
essentially reversing, with a complete stoppage of flow (slack water) halfway between the maximum
flood and the maximum ebb. In this case the tidal current speed curve looks like a tide curve (with
the zero speed line serving the role of a datum).
Most rotary tidal currents fall between these two extremes (i.e., between circular rotary flow and
reversing flow), with their flow vectors tracing out ellipses of varying widths, with slacks replaced
by flow minimum, when for a short time the flow is perpendicular to the main flood-ebb axis of
flow. However, geography (i.e., bathymetry and shoreline shapes) can make these rotary flows
complicated, adding asymmetry in a variety of ways, so that the curves traced out by the changing
flow vectors do not look like ellipses at all, and often are strange shapes (see for example the fifth
plot in Figure 2.4). Selecting maximums may not always be as straightforward as one would hope.
For example, if one has current data (or harmonically produced vector tidal current predictions that
show the rotating current) at a location where three channels meet, one may see three maximum
flows, one each for when the tidal current is flowing into each channel. For such a case, it may be
difficult to decide whether one of those maximums is a maximum flood or a maximum ebb.
Typically the methods for selecting maximum floods and ebbs from a predicted tidal current time
series and from an observed current data time series are similar to those used for the tide. Again,
one aspect of all methods is the keeping track of each tidal cycle. All of the complications just
discussed (as well as the two exceptions discussed for tides) must be somehow accounted for in any
maximum selection routine (or in a few of the extreme cases, they must be discovered manually,
after looking at the maxima and minima that the computerized routine selected).
There are different methods for selecting slack waters (if the tidal current is reversing) or
minimums (if the tidal current is rotary). For selecting slack waters (one before flood and one before
ebb) in a predicted tidal current curve, it is simply a matter of finding the smallest positive (flood)
speed value and the smallest negative (ebb) speed value (right next to it) and interpolating between
them to find the time of the zero speed. Or in some cases one may actually have a predicted value
that is zero. For selecting slack waters in a current data curve, the noise may lead to several
calculated times of slack, in which case some type of averaging/smoothing scheme over some
number of data points can be used to come up with only one slack water. For selecting the times,
speeds and directions of the minimum before flood and the minimum before ebb, one usually
vectorially averages a group of points with the smallest speeds. One should obtain directions for
the minimums that are approximately perpendicular to the flood-ebb axis.
3.10 Spectral Analysis and EOF Analyses
Spectral analysis was mentioned in Section 2.2.3 as a way of determining (and easily
visualizing) the amount of energy present in the tidal frequency bands as well as in the nontidal
continuum between those bands. In general the results of spectral analysis are not very useful for
accurately predicting tides or tidal currents. The exception to this statement, as was seen with the
response method, is when a cross-spectral analysis is done (e.g.) between a water level time series
and a time series of the tide potential (or the time series of a reference tide station), in which case
there will result amplitudes and phase lags that can be used to predict the tide (relative to the input
tide potential or reference tide station). So more precisely, it should be said that the results of an
autospectrum are not very useful for accurately predicting tides and tidal currents.
Such results are, however, very useful for assessing which tidal constituent signals in a data time
series are large enough so that they should be included in a harmonic analysis. Spectral analysis,
for example, comes in very handy in determining how important the higher order bands (i.e., the
132
quarter-diurnal and higher bands) will be at a station from a shallow-water area. One can then make
some decisions as to which additional tidal constituents will be needed to be included in the
harmonic analysis in order to accurately capture most of the tidal energy and ultimately to make the
most accurate tidal prediction possible.
3.10.1 Spectral Analysis of Water Level Data Time Series
Only a very brief description of spectral analysis techniques themselves will be given here. For
a much more thorough treatment of the material presented below see Emery and Thomson, 2001,
as well as Press, et al, 1992 (or one of the latters many later editions for different computer
languages). Although in many ways very different than harmonic analysis, some of the insights
gained by understanding those differences can be useful for appreciating what the very act of
sampling a data time series and analyzing it can do to that datas inherent frequency make up. This
section will concentrate on analyzing water level data, a scalar data times series, but in Section
3.11.3 one will see how this is extended to vector time series, i.e. current data time series.
Technically the spectrum has usually been called a energy spectrum or power spectrum, and it
involves at some point squaring the data values in the time series or squaring the individual results
in the frequency domain. A theorem called Parsevals theorem says that the total energy will be the
same whether computed in the time domain or in the frequency domain. A spectrum from only one
time series is referred to as an autospectrum. When one has two time series for the same time period
and one wants to see the frequency make up of what might be characterized as shared power
between them, one calculates a cross-spectrum (see Section 3.10.4). Prior to carrying out a spectral
analysis one first removes (subtracts) the mean from the data time series, as well as its trend, because
leaving them in can have an adverse effect on the estimated low-frequency components of the
spectrum. Sometimes a low-frequency component itself will be removed if it can be reliably
determined.
The frequency resolution of spectral analysis, like all time series analyses (including harmonic
analysis) depends on the length of the time series the longer the time series analyzed the finer the
resolution in the frequency domain (see Sections 2.2.3 and 3.3). However, with spectral analysis
one typically gives up some of the maximum possible resolution in order to improve the reliability
of the analysis result. This is because the data being analyzed with the spectral analysis has many
nontidal random and aperiodic fluctuations (which are stochastic, and thus unpredictable, except in
a probabilistic sense), along with the periodic tidal oscillations (which are deterministic and
predictable). To make the raw spectrum of a stochastic process more reliable some type of
averaging is done, either in the time domain or in the frequency domain. In the time domain this is
done by chopping the time series into some number of equal length segments, doing spectral
analysis on each segment, and averaging the results. In the frequency domain the same thing is
accomplished by averaging several adjacent raw spectral estimates into wider frequency bands.
Both processes reduce the frequency resolution in the final (but more reliable) result. (One does not
usually choose to do this with harmonic analysis because of the periodic deterministic character of
the tides.)
The original type of spectral analysis was the autocorrelation method (Blackman and Tukey,
1958), where the power spectrum is calculated by taking the Fourier transform of the autocorrelation
function using some number of lag intervals. (As the name implies, the autocorrelation function is
a way of showing how well the time series correlates with itself shifted in time.) When computers
began to be used, a numerical technique called the Fast Fourier Transform (FFT) allowed the use
133
of a better method usually called the periodogram method or the direct method using the FFT
(Cooley and Tukey, 1965). Here the data itself is directly transformed into its Fourier components
with the FFT. Both of these methods are sometimes referred to being nonparametric methods
(Emery and Thomson, 2001) because such methods are not data specific like the so-called
parametric methods, which use some kind of predetermined statistical model to modify the time
series.
The first of the parametric methods was the so-called maximum entropy method (MEM) (Burg,
1972), where the name comes from the fact that a maximum entropy condition is used, that is, it is
required that the spectral estimate be the most random of any power spectrum which is consistent
with the data. Autoregressive methods (AR) are very similar to MEM. MEM/AR methods provide
finer frequency resolution (although they lose some accuracy), and they allow one to better deal with
short time series. These methods can therefore better find the location of peaks, but may not do as
good a job calculating an accurate spectral energy at those peaks. Other types of parametric spectral
analysis have also been developed such as the maximum likelihood method.
The mathematical details of these (and other types of) spectral analysis methods are left to the
references mentioned above. [Emery and Thomson (2001, pp 461-2) provide a step-by-step standard
spectral analysis approach.] What the reader of this manual mainly needs to know are (1) the
advantages and disadvantages of a spectral analysis method that he or she may want to use; and (2)
how to interpret and use the resulting spectra that come from that method.
As was mentioned in Section 2.2.3 the spectrum for a typical water level time series (such as that
shown in Figure 2.16) shows a continuous curve with spikes above it at the tidal frequencies. The
portion of the spectra plot between these tidal spikes is the nontidal continuum and usually is higher
at the lower frequencies. This is referred to as a red spectrum (in analogy to optics where red light
has lower frequencies and longer wavelengths than the rest of visible light). For a long enough time
series the frequency resolution can be made small enough to allow the energy in each tidal
constituent to show up in its own vertical spectral line (see Figure 2.17).
3.10.2 Tidal Cusps In Spectra
If one looks at a water level spectrum with even greater resolution (e.g., from a time series length
of a year or greater) one sees that between the tidal spectral lines the nontidal continuum curve rises
up to meet each tidal spectral line on both sides, forming a cusp-like shape (this being usually most
noticeable around semidiurnal tidal lines). Such tidal cusps were first described in a papers by
Munk, et al (1965) and Munk and Cartwright (1966). They felt that the simplest explanation for
these cusps were that they were presumably due to non-linear interaction between the lines with
the low-frequency continuum, namely that they represented the sidebands due to the modulation
of a carrier (the tides) by a low-frequency noise (the continuum at very low frequencies). This
seems to make sense, but they could not identify any dynamic mechanism that seemed to be able to
produce the cusps seen in the water spectra from the locations analyzed by these authors and others.
As Cartwright (1999) would write 33 years later, Their precise cause was hard to identify, but the
most promising suggestion was that they were due to the surface manifestation of internal tides.
Such surface manifestations of internal tides would slowly vary as the stratification of the ocean
varied (or even disappear if the water became well mixed). This was, in fact, demonstrated by Ray
and Mitchum (1997) at Hawaiian water level gauges.
Parker (1991) suggested that, at least for shallow water areas, the same nonlinear mechanisms
that lead to the interaction of the tide with storm surge, should produce tidal cusps in a water level
134
spectrum. If a low-frequency storm surge was periodic, the Fourier decomposition analysis of the
hydrodynamic equations of motion (see Section 7.6) that showed how the various nonlinear terms
could produce overtides and compound tide, could also be used to determine the effect of such a
periodic storm surge on the tide. If periodic, such subtidal storm surges would produce side bands
around each astronomical constituent spectral line via primarily the M(u)/Mx and u|u| terms (the
effect of u Mu/Mx would probably be negligible because of the small current speeds associated with
low-frequency surges). The low-frequency water level spectrum has the highest amplitudes nearer
the zero frequency and then slowly decreases with increasing frequency. A periodic storm surge
with a frequency close to zero would lead to side bands very close to the tidal spectral line, and these
would be larger than the side bands caused by a slightly higher-frequency periodic storm surge
(which would be further from zero frequency and thus have a smaller amplitude). A visualization
of this effect in the frequency domain does, in fact, correspond to the idealized picture in a figure
presented by Munk, et al (1965), and recreated in Figure 3.16. But storm surges are, of course,
transient and vary in frequency. The result is a smearing of each spectral line to create a cusp-like
shape in the nonlinear continuum around the tidal spectral line.
Whatever their cause the presence of cusps is one more source of noise that might adversely
affect the calculation of the smaller harmonic constants. (See discussion in Section 3.3.) And it also
implies that some energy is lost from the tidal lines, and that harmonic constant amplitudes would
be somewhat larger if there was no nontidal noise and thus no tidal cusps.
3.10.3 Spectral Analysis of Current Data Time Series Rotary Spectra
Since tidal currents are vectors and since the speed and direction time series are typically
transformed into two orthogonal component time series (such as major-minor or north-east
component series), there are several choices on what type of spectrum can one can produce from a
current data time series. For very reversing tidal currents, or even for tidal currents where the flow
in flood or ebb directions is much larger than the perpendicular flow near times of minimums (i.e.,
for narrow elliptical rotation), one may wish to simply do one spectral analysis, on either the speed
time series (with ebb flows given negative values) or on the major-component time series only. For
other cases, however, one will want to carry out two spectra analyses, one on the major-axis time
series and one on the minor-axis time series. In some cases the spectra of the minor-axis time series
will look different than the major-axis time series, e.g., sometimes having more energy in the
quarter-diurnal tidal band.
Another option is to carry out a what is usually referred to as a rotary spectral analysis, first
developed by Gonella (1972) and Mooers (1973). The energy in a current vector time series (i.e.,
speed and direction time series) does not have to be partitioned into two orthogonal components;
it can instead be decomposed into two polarized counter-rotating components, that separately
represent the clockwise energy and the counterclockwise energy in the time series. At each
frequency, the two counter-rotating circular components can be combined to produce an ellipse for
that frequency (the same one that would be produced by combining the two orthogonal results at that
frequency). An ellipse from a rotary spectrum at a particular frequency is analogous to the ellipse
from a harmonic analysis, created by combining the harmonic constants for two orthogonal
components (see Sections 3.4.3 and 5.2). Rotary spectra can be calculated from their Cartesian
counterparts (i.e., the spectra for the north-east or major-minor components). (See Emery and
Thomson, 2001.)
135
Figure 3.16. An idealized low-frequency nontidal continuum decreasing from its

highest value near zero frequency, but then rising up around a tidal spectral line in the
shape of a so-called tidal cusp (see text). (From Munk, et al, 1965.)
Rotary spectra are useful because there are situations where energy may be predominant in either
the clockwise or the counter-clockwise component. Inertial currents (which can be close to tidal
frequencies at particular latitudes) rotate clockwise in the northern hemisphere and counter-
clockwise in the southern hemisphere due to Coriolis effects. Sometimes one will see the dominance
of one type of rotation offshore, and the opposite rotation or perhaps rectilinear motion (i.e., equal
amounts of energy in both clockwise and counter-clockwise) close to shore. Each of these types of
spectra will provide not only information on how much energy is left at the various tidal frequencies,
but also might also provide insights (some more than others depending on the situation and location)
into possible hydrodynamic reasons why the energy is there.
3.10.4 Cross-spectral Analysis
The calculation of an autospectrum deals only with the frequency characteristics of a single data
time series. The calculation of a cross-spectrum deals with the frequency characteristics common
to two data time series from the same time period. At each frequency in the cross spectrum one will
see the highest values at frequencies where there is energy in both the time series. At each
frequency there will be an amplitude and a relative phase, showing the relationship between the two
time series at that frequency. The two times series are often the same parameter but at two different
locations, such as two water level records from two locations along a waterway (from which, for
example, the cross spectrum might tell us something about the response of the tide at the second
station relative to the tide at the first station). However, the two data time series can be for two
different parameters, such as the water level and wind, or current and water temperature, again to
learn something about the connection between the two parameters over the frequency domain. One
often sees higher values only over certain frequency bands, implying a possible causal connection
between those parameters for those frequencies.
We have already seen one example of the use of cross-spectral analysis with tides, i.e., the
response method (see Section 3.5.1). Here a cross-spectral analysis is done (e.g.) between a water
level time series and a time series of the tide potential (or the time series of a reference tide station),
136
in which case there will result amplitudes and phase lags that can then be used to predict the tide
(relative to the input tide potential or reference tide station).
One can also carry out a cross-spectral analysis of more than two data time series, and in fact,
this approach is recommended if one believes that more than one input (e.g., tidal forcing, wind,
changing water temperature) to a physical system is important in producing the output (e.g.,
changing water level). Such a multi-input cross spectral analysis takes into account the mutual
correlation among all the inputs, and calculates the relative contribution of each input to the output.
Wunsch (1972) used this method to look at the sea level at Bermuda in relation to the tides, weather,
and fluctuations in water temperature. Cartwright (1968) also used it to carry out a unified analysis
of tides and surges along the coasts of Britain. Similarly, Groves and Hannan (1968) used it to look
at the effects of weather on sea level.
One can also calculate rotary cross spectra for two vector times series (Mooers, 1973). Here
one is essentially trying to ascertain the similarity between the two vector time series in terms of
their circularly polarized rotary components. This includes the common energy in the two co-
rotating components as well as in the two counter-rotating components. One might, for example,
want to look at the rotary cross spectra in currents and wind, as part of an investigation of the effect
of a sea breeze-land breeze on the currents (OBrien and Pillsbury, 1974), and perhaps also see how
this might affect nearby tidal frequencies.
(See Emery and Thomson, 2001, and Godin, 1972, for a much more thorough discussion of
cross-spectra and rotary cross-spectra.)
3.10.5 Empirical Orthogonal Function (Principal Component) Analysis
All the discussions in this chapter have only dealt with analysis techniques in the time domain
(and thus also in the frequency domain). Variations in the space domain (and analogously in the
wave number domain) were only mentioned in terms of displaying the time series analysis results
on a graph or on a chart, where the results at different locations are plotted. For example, for tides
one produces cotidal and corange charts, or graphs the increase in various tidal constituents
amplitudes as one moves up a waterway. This serves an important visualization function, and it also
is necessary for comparison with the output of hydrodynamic models used to explain the physics
behind the variations in the relevant physical oceanographic parameters, including the tides and tidal
current.
However, there is a whole field of spatial analysis of time varying data fields, that is, when there
are literally hundreds or thousands of time series being produced by an instrument or a model at
hundreds or thousands of (usually evenly spaced) locations. Such techniques become very useful
when one wishes to analyze such time varying data fields, for example, time varying surface current
fields from an HF radar system, or time varying elevation fields from satellite altimetry (from which
fields of tidal variation can be extracted), or various time varying fields from a hydrodynamic
numerical model of a waterway. Here one wishes to understand not only how the frequency makeup
varies in space, but one also wants to extract some type of wave number information (in the spatial
domain the wave number is the inverse of the wavelength, just as in the time domain the frequency
is the inverse of the wave period).
The most common technique used in the analysis of time-varying data fields is empirical
orthogonal function (EOF) analysis, also called principal component analysis (PCA). EOF analysis
essentially allows one to characterize the spatial variation in the time-varying fields using orthogonal
functions or modes. It does this in a more economical manner than one might expect, because often
137
most of the variance of a spatially distributed time series is captured by the first few orthogonal
functions (in many cases the first orthogonal mode can account for more than half of the variation).
For an EOF analysis of sea level data over a region, for example, the first mode might represent the
portion of the energy for the water level going up and down together at all the stations, while the
second mode might represent a standing wave oscillation where the water level at some of the
stations moves up together while at other stations it moves down. The higher modes may have
oscillations with decreasing wave length (increasing wave number).
One then usually tries to somehow interpret or correlate the patterns seen in these modes with
respect to the spatial variation in particular hydrodynamics mechanisms. This may not always be
easy, since these modes are a statistical entity and should not be confused with a particular
oscillation that comes out of the hydrodynamic equations. In that respect EOF modes are not the
spatial analog of the harmonic constants produced by a harmonic analysis. Where the analogy does
hold is that the few modes that come out of an EOF analysis can be used to economically reproduce
the spatial variation over a waterway that, for example, may previously been represented by data
from dozens of water level stations (just like a few harmonic constants can represent most of the
temporal variation in a water level time series). As another example, using an EOF analysis of
vertical or cross-sectional current velocity profiles it might be possible to separate the current flow
into a few modes of variance, from which one might be able to ascertain the relative importance of
the baroclinic contribution (due to perhaps internal waves) compared with the basic barotropic
contribution (of the dominant tide wave, wind, or other effects).
For more information on the math behind this technique and its practical application see
Preisendorfer (1988) and Emery and Thomson (2001).
138
4. Harmonic Analysis of Water Level Data
4.1 Considerations In Carrying Out the Analysis
Using the astronomical and hydrodynamic considerations discussed in Chapter 2 and the least
squares harmonic analysis method described in Chapter 3, this chapter will look specifically at the
things to be considered when harmonically analyzing water level data. (Harmonically analyzing
current data will be treated in Chapter 5, and nonharmonically analyzing water level and current data
will be treated in Chapter 6.) At the end of this chapter, in summary, all the steps to be considered
in a harmonic analysis of water level data will be listed.
4.1.1 Which Tidal Constituents Should Be Solved For
4.1.1a Based on Length of the Data Time Series (the Synodic Period)
As was seen in Section 3.3, the most important consideration affecting the accuracy of calculated
harmonic constants (and skill of the tide predictions made with those harmonic constants) is the
length of the data time series that is to be analyzed. The key to an accurate tidal prediction is
determining the amplitudes and epochs for the most tidal constituents that can be calculated with
a given length data time series. The longer the time series, the closer in frequency two tidal
constituents can be and still both be solved for. Or stated another way, if the synodic period (i.e.,
the time required to separate the effects of two nearby constituents) is shorter than the data record
length, then both those constituents can be included in the harmonic analysis.
If a time series is not long enough to separate two nearby tidal constituents, then only one of
these two constituents can be solved for in the harmonic analysis. This will have two effects. First,
some tidal energy of the unsolved-for constituent will not be captured. Second, some of the energy
of the unsolved-for constituent will be wrongly incorporated in the harmonic amplitude and phase
lag of the constituent that is solved for. Since the two close constituents will go in and out of phase
over their synodic period (see Section 3.3), the astronomical conditions during the data record will
determine how much of an adverse effect this will have on the harmonic constants. As will be seen
in Section 4.2, for cases where the time series is too short, methods have been developed to infer the
unsolved-for tidal constituents, and also to adjust the solved-for constituents for the degrading
effects of the unsolved-for constituents.
Table 3.2 lists the 37 tidal constituents that have typically been solved for in a standard one-year
harmonic analysis at CO-OPS (and its predecessor organizations). The list starts with M2, generally
the largest of all the constituents (except for some locations with an extreme diurnal situation) and
then follows with other constituents listed in the order of their most important synodic period, that
is, the length of the series required to separate them from a nearby (usually) larger constituent. Once
139
one gets past K1 (and overtides) there are four natural groupings with respect to the synodic period:
half month, month, half year, and year.
This table provides guidance for which constituents to include in a harmonic analysis, but (as
was discussed in Sections 3.3 and 3.4.5) when using a least square harmonic analysis there may be
some leeway, especially when there is a very strong tidal signal. Thus, if one has less data than
Table 3.2 says is required to be able to include a particular constituent in our harmonic analysis (so
that it can be separated from another key constituent), one may still be able to analyze for both
constituents because of the least squares technique. The best procedure is probably do several
analyses, following the recommendations of Table 3.2 with the first analysis, and then adding a key
constituent with the second analysis, and if necessary adding additional constituents in additional
analyses. For each analysis redone with an added constituent, examine carefully the other tidal
constituents to see how much they have changed compared with the results of the previous analysis.
Large changes probably indicate that the program has been pushed too far (and in fact, the program
will crash if one really pushes it too far).
As but one example, suppose one has 5 months of water level data from a waterway with a
strong enough diurnal signal that one wants to be able to solve for P1 , because it is the third largest
diurnal tidal constituent. However, Table 3.2 says that to be able to include P1 in a harmonic
analysis one should have 182.6 days of data (the synodic period for resolving P1 from K1). The time
series length is a month short, but since a least squares harmonic analysis technique is being used,
and since one expects P1 to be a strong signal (well above the noise level of the water level record),
one can decide to try a 150-day analysis. This analysis will most likely turn out fine, but one can
check this (as just mentioned) by comparing the K1 value from this analysis with the K1 value from
an analysis without P1 included. If they are reasonably similar, then there was little adverse affect
of P1 on K1. Also, if one has analysis results from a nearby longer period station, one can see if the
P1/K1 ratio at that station is similar to the P1/K1 ratio at this newly analyzed shorter station (and
likewise that the difference in the K1 and P1 epochs are similar). If the results do not look
reasonable, the analysis results without solving for P1 will have to be used, and then the amplitude
and phase of P1 can be inferred using one of the methods talked about in Section 4.2.
As another example, suppose one has only 15 days of data. Then a harmonic analysis will
provide values for the major constituents M2, S2, K1, and O1, plus a few higher harmonics and a
couple of less important constituents. However, these calculated values will also include energy
from the constituents that could not be separated out in only 15 days. Most important, M2 will
include energy from N2 (which could have been resolved from M2 if there had been 29 days of data).
This N2 contribution could make the M2 value calculated from 15 days of data larger than it should
be, or smaller than it should be, depending on when the data were measured. Likewise (as in the
above example but more extreme here), since P1 cannot be separated from K1, K1 will include the
effects of P1. If one had more data and harmonically analyzed successive 15-day periods, one would
see the amplitude of K1 slowly vary over a 6-month period because of the influence of P1. One
would also see the amplitude of M2 vary from 15-day to 15-day period, but to clearly show the 29-
day modulation one would need to start each 15-day analysis only a few days after the previous one.
For a record as short as 15 days, use of a method to infer important missing tidal constituents such
as N2 and P1 is definitely required.
Table 3.2 is quite adequate for many situations, but for shallow-water areas with large tide
ranges, or if one wants to include every possible constituent that can be calculated using a data time
series that is longer than a year, then one needs to look at Table A.2 in Appendix A, which includes
140
149 tidal constituents (including 103 shallow-water constituents) and extends beyond a year for the
synodic periods of some very small constituents.
4.1.1b Compound Constituents Versus Astronomical Constituents (with the same frequency)
There is another consideration when deciding which constituents to include in a harmonic

analysis. In shallow-water areas, compound tides will be produced (by the nonlinear processes
discussed in Sections 2.3.2 and 7.6) and some of these compound tides will have frequencies that
are the same as particular astronomical constituents. Although they have the same frequencies, they
may have different node factors (and equilibrium arguments), so that if the wrong choice is made
between using the astronomical constituent versus using the shallow-water compound constituent,
then there will be errors in future predictions (but not in the self-prediction for the year of the
analyzed data).
A usually important example of this is that the shallow-water constituent 2MN2 has the same
frequency as L2 , but has a different node factor. There is no way to separate two such constituents
unless one has 18.6 years of data, in which case the satellite constituents corresponding to these two
constituents would handle the different long-term variation (that is represented by the different node
factors when not using satellite constituents). To decide which constituent to use, one can initially
use L2 but then look at the analysis results for locations throughout the waterway. If one sees that
L2 increases in amplitude (as one moves up a waterway) at a rate that is greater than the rate of
increase in M2 or S2 (for example see Figure 7.5) then L2 is probably really 2MN2. [L2 would
increase at a rate similar to the rate of increase of M2 , because both would be affected by how close
the waterway is to a resonance condition (see Sections 2.3.1 and 7.4.1), but 2MN2 would increase
at a faster rate because nonlinear processes would be extracting energy from M2 and N2 and putting
it into 2MN2 all the way up the waterway.] If the constituent is 2MN2 , the analysis does not have
to be redone, but the final amplitude and epoch should be adjusted for the correct node factor and
equilibrium argument.
In some cases, where the amplitude of the constituent does not grow at a rate greater than the
rate of increase of M2 , it may not be possible to tell how much of it is L2 and how much of it is
2MN2. (If one had an numerical hydrodynamic model, one could tell by running the model with and
without the appropriate nonlinear terms turned on, but this is much too much effort.) However, in
this case the amplitude of the constituent is probably small enough that simply sticking with L2 is
fine. If the water level station is on an oceanic island or along an ocean coast (especially one with
a deep and/or narrow continental shelf) one is very safe in using L2. For most other shallower
waterways, since L2 is usually small, it has been simplest just to use only 2MN2.
Another example is that the shallow-water compound constituent 2MS2 and 2 have the same
frequency. In shallow estuaries one sees 2MS2 increase at a rate faster than the rate of increase of
M2 or S2 (also see Figure 7.5). Here, however, it does not matter whether the analyzed constituent
is 2MS2 or 2 , because they have the same node factor.
For very shallow-water areas with large tide ranges one adds many additional shallow-water
compound constituents, and some of these may end up with the same frequency. In such cases one
has to look at the nonlinear generating mechanism and try to decide which of the candidate
constituents to use. One should also first determine whether their node factors will even be
different. If they are not, it doesnt matter what name one gives that constituent.
141
4.1.2 Keeping or Rejecting Small Constituents
Many of the constituents included in the harmonic analysis will come out with very small
amplitudes. If these constituents are so small that they fall within the nontidal continuum, they may
not be real (that is, one has not been able to calculate the actual tidal constituent, only the amplitude
and phase lag of some transient nontidal periodic signal). Sometimes there has been an arbitrary
lower limit set, below which the harmonic constants are rejected (see Section 4.3.5). However, there
can be cases where harmonic constants are real even if their amplitudes fall below that lower cutoff
level. The best way to determine this, if one has enough data, is to run more than one harmonic
analysis and to look at the phase lag (epoch) and see if it is consistent from analysis to analysis. If
it is, then the harmonic constants probably do represent the tidal constituent solved for (and not just
some transient periodic nontidal energy).
The reason it is important not to reject small tidal constituents unnecessarily, is that the lost tidal
energy can add up if several small constituents are rejected. Then the predicted high waters (on
average) will be too small (and low waters on average will be too high) compared with the high and
low waters in the data (also on average, so that meteorological effects are averaged out). Such a
situation has sometimes been solved by simply increasing all the constituent amplitudes by some
percentage (see Section 4.3.5) in order to make the mean predicted high and low waters better match
the mean high and low waters from the data. This so-called build-up factor adds energy in the
wrong places and does not really make the tidal predictions more accurate (only the average of all
the high waters and the average of all the low waters are more accurate).
4.1.3 Instrument Errors and Their Effects On Analysis Results
When data is referred to as being bad or unreal, one is essentially talking about instrument
error, i.e., values that do not represent the actual water level at the time of measurement because of
some error related to the measurement process itself. There will also be natural noise in the data
record due to various nontidal phenomena, but being natural such noise is considered real and not
errors (although one may wish to filter them out at some point, if one feels they might adversely
affect our tidal analysis). Techniques for evaluating the quality of a data time series and techniques
for editing out (or possibly repairing) data points that are determined to be bad or unreal can
vary significantly, and there is no need to discuss any particular method here. Commonly one tries
to find and then remove (or correct) data points that are obvious outliers, which is fairly simple
when there is a clear tidal signal in the data. One also tries to discover if possible other types of
variations in the data that may not be of natural origin (and thus must have come from the
instrument) which might affect the analysis results.
What is important is to acquire some feeling for the potential sources of errors and how different
types of errors might affect the results of a harmonic analysis. Errors in a water level data time
series may be electronic, mechanical, the effect of temperature changes on the instrument (as
opposed to its effect on the water level), etc, or have to do with movement of the water level gauge
itself.
The types of potential errors are different for each different type of water level gauge. However,
with an eye toward their effect on harmonic analysis results, these types of error can be classified
as either:
(Type 1) random errors [which will tend to average out during a harmonic analysis];
142
(Type 2) sudden shifts, when from some point in time forward, the entire time series of measured
heights are all larger or smaller by some amount than they were before that point in time
[which would incorrectly change mean sea level, but probably not affect the amplitudes
of harmonic constants, except perhaps the longer-period constituents (such as Sa and
Ssa) depending on the length of the erroneous offset];
(Type 3) slowly changing errors, where an error might slowly grow over months or longer [which
might change the harmonic constants];
(Type 4) slowly changing time measurement, such as when a clock may be running slow [which
would essentially spread out time and reduce the frequencies of the tidal constituents, so
that when a harmonic analysis looks for energy at the correct tidal frequencies the result
will be smaller amplitudes and changed phase lags]; or
(Type 5) periodic errors, which might, for example, be due to meteorological effects on the
instrument (as opposed to meteorological effects on the water level itself) [which might
change particular tidal constituents].
From 1807 until the mid 1850's, in the U.S. Coast Survey the simple tide staff was the primary
method used for measuring water level. A tide staff had graduated markings on it in feet and tenths
of a foot, and was fastened firmly to a pier or piling or some other structure that was (hopefully)
unlikely to move vertically. In remote areas, supporting structures had to be constructed so that a
stable tide staff could be read during the time period of a hydrographic survey. In many instances,
the tide staff was not read continuously, often being read only around the times of high water or low
water. In such cases, either the harmonic methods discussed in Section 3.5.6 or nonharmonic
methods (Chapter 6) had to be used for analysis and prediction purposes.
To make a water level observation an observer simply tried to determine where the water level
fell on the staff, and wrote down the time and height of the water level elevation. That was often
made difficult by the fact that the water level was usually oscillating up and down due to the action
of (higher frequency) wind waves. The Manual of Tide Observations (U.S. C&GS, 1963, page 33)
provided the following guidance for the tide observer using a tide staff:
In taking the tide staff reading, both the highest and lowest points reached by the waves are to
be recorded, the two readings being separated by a dash or in some other distinctive manner.
For this type of water level measurement, errors were mostly likely human error during the visual
process of estimating each water level elevation (and any errors were hopefully of Type 1) . If the
observers changed, there might also be a change in bias of how each person tended to make the
observation (which would be error Type 2) . Another error could result if the tide staff suddenly
shifted position on the pier (again error Type 2), but it also could also have slowly slipped lower
over a period of time (error Type 3). [The pier itself could also be shifting, which is a potential
problem for all types of water level gauges (see below).]
One type of tide staff occasionally used had a glass tube attached to its full length. This glass
tube was partially closed at the lower end (which was always in the water), the purpose being to
dampen the action of waves so that the water in the tube would not oscillate as much as the water
outside the tube. This was intended to make it easier for the observer to make an accurate visual
measurement of the water level. Such a tube might get clogged which could lead to errors of Type
2, or might be prone in some areas to slow biological growth, which might causes errors of Type 3.
143
It was also subject to random errors by the observer (error Type 1), as well as to a change in bias (on
the way a particular observer made a visual measurement) when observers changed (error Type 2).
From the mid 1850's until the last couple of decades, the most common method for water level
measurement was the use of a float inside a stilling well (sometimes also called a float well). The
float was at the end of a wire which was attached (usually by pulleys) to some type of
measuring/recording device. Before it became an automated system and measurements were still
made visually by an observer, the wire was a tape which was either graduated itself (in feet and
tenths of a foot) or had a pointer which would move up and down along a graduated scale. This pre-
automated version was also subject to random errors by the observer (error Type 1), as well as to
a change in observer bias when observers changed (error Type 2). The stilling well was intended
to dampen (or still) wave effects on the vertical movement of the float. It had a small opening at
the bottom, typically a 1 inch opening in a well of from 4 to 12 inches in diameter, but the opening
could be as small as 3/4 inch in areas exposed to heavy seas. The stilling well could become
clogged with floating debris, which could lead to errors of Type 2, or it could be slowly clogged by
biological growth, which might causes errors of Type 3.
When the float tide gauge system became an automated system (not requiring an observer to
make a visual reading), the wire moving up and down operated a worm screw on the gauge, which
in turn moved a pencil back and forth across a moving strip of paper. The paper moved forward at
a known uniform rate by a clock motor, the result being a continuous graph of the rise and fall of
the water level. There was still a human aspect to making the measurement (and a chance for human
error of Types 1 or 2), but it was now in how the time and height were taken off the graph of the tide
on the paper roll, which involved not only determining the times and heights of high and low waters,
but also the heights at hourly intervals. Tide observers were still important in the field, however,
since once a day they made independent measurements using a tide staff, unclogged the stilling well,
and made clock adjustments to keep the tide gauge clock on the correct time. In the mid 1960's, the
analog strip charts were replaced by analog-to-digital (ADR) punched paper tape. The punched
paper tape allowed the use of an automated data processing system. This reduced some of the
processing and tabulation errors, tide gauges still had clock drift, and tide staffs were still required
to be read by tide observers. Slowly degrading bearings in the ADR gauge system due to the salt-
water environment could also lead to slowly changing errors (Type 3).
The stilling well was also a potential source of another type of error when there were fast tidal
currents (Shih and Baer, 1991). Such currents could cause a pull-down effect as the water flowed
past the smaller opening (the orifice) at the bottom of the well. Such an effect would be periodic
(error Type 5) with the frequency of the primary tidal constituent, and thus could affect that tidal
constituent and other constituents in the same frequency band. However, the maximum effect would
be at both maximum flood current and at maximum ebb current, so the result might also effect the
overtide constituents. As a solution to this problem, in locations with strong tidal currents, large
parallel plates were mounted below the orifice to reduce the pull down effect.
At locations where a tide station could not be installed on a solid structure along the shore, a gas-
purging pressure tide gauge (or bubbler gauge) was used, being placed on the sea bottom (usually
2000 feet or less from the shore, the maximum length of tube between the bubbler and the data
recorder onshore). As explained in The Manual of Tide Observations, when gas is bubbled freely
into a liquid from the fixed end of a tube, the pressure in the entire length of the tube is
approximately equal to the pressure head of the liquid over the bubble orifice. Any change in the
hydrostatic pressure (such as caused by the rise and fall of the tide) is transmitted by gas pressure
through the tubing to the transducer bellows of the recorder where pressure variations are recorded
144
as changes in depth of water. The changes in the pressure pushing down on the bubbler gauge
(sitting on the sea bottom) can also be caused by changes in water density due to changes in salinity
or water temperature. Since in estuaries tidal currents can periodically change the salinity and water
temperature (e.g., perhaps being saltier during the flood current phase and fresher during the ebb
current phase), this is a potential source of error(Type 5) that can affect the tidal harmonic constants
calculated from the water level data obtained with this instrument. Seasonal changes in water
temperature and salinity can also have an effect (also a Type 5 error, but with a longer period and
so more likely affecting the constituents Sa and Ssa; see Section 3.7). For bubbler gauges deployed
for short time periods, this seasonal effect would not be obvious in the data. However, in the early
1970's, bubbler tide gauges were installed at most long-term stations to serve as an independent
backup system to the float and stilling well system. More recently, the density effect problem has
been reduced by using dual bubbler orifice systems with the pressure fed into two highly precise
Paroscientific pressure transducers. The pressure difference between the two bubbler orifices is used
to estimate water density, which is then used to correct the elevations calculated from the pressure
data.
When automated systems began to be used, which involved time keeping with clocks, there was
the added possible problem of clock errors, both sudden changes and very slow changes in time that
occur over days, months, or years. Slow changes in time can have a significant effects on harmonic
analysis results (error Type 4). A slow clock would lead to data samples being taken at time
intervals slightly longer than the intended sampling interval. If this was not recognized and
corrected (and the time was erroneously assigned to the data points as though the clock were
correct), then the tidal energy in this data time series (with erroneously assigned times) would appear
at slightly higher frequencies than the tidal frequencies. (For a fast clock, the opposite would
happen.) Analyzing such data for the frequencies where one expects to find tidal energy could then
lead to erroneous harmonic constants (depending on how slow or fast the clock was). A sudden shift
in time also decreases the amplitudes of the constituents and changes the phase lags (like Type 4).
With todays real-time systems, timing errors are no longer a problem, as timing is controlled by a
small antenna that receives precise timing from the GPS system.
The present water level measurement system used by CO-OPS is an acoustic one, where again
there is a well, although more open than a stilling well to reduce the aliasing effect of the stilling
well when there are waves (Shih and Baer, 1991). Water level elevation is now measured by the
time it takes a sound pulse to leave a transducer, travel to the surface of the water, reflect off it and
return, the sound pulse actually traveling through a 13 mm diameter PVC tube. The speed of sound
in the tube is determined using a second reflection from a discontinuity in the tube at a know
distance from the transducer (this section of the tube being referred to as the calibration tube). There
is a new potential source of error, caused by changes in air temperature, which changes the density
of the air, which in turn changes the speed of sound, but the use of the calibration tube is designed
to take care of that. However, care must be taken to assure that the temperature in the calibration
tube is the same as in the rest of the tube, and this is accomplished by various means (including
ventilation, keeping the calibration tube in the same environmental conditions as the rest of the tube,
and painting the protective well a light reflective color). But for water level stations where there is
a large tide range, the PVC tube is very long and so there is more of a chance for temperature
effects. To deal with this problem, there are also two thermistors, one in the calibration tube and
one in the rest of the tube. If the temperature at these two locations is different enough, they can be
used to make corrections to the speed of sound (and thus of the water level measurement). If
temperature errors do occur, since the air temperature changes over a daily cycle, such errors would
145
be Type 5, and there could be seasonal variations as well. (See Gill, et al, 1993, and Porter and
Shih, 1996.)
The switch from the float gauge to an acoustic gauge at a particular stations can also potentially
cause an error Type 2. However, to minimize such an effect (which would not affect harmonic
constants, but could affect sea level determinations, and in fact look like a small change in sea level),
both gauges were run simultaneously side by side for a year, in order to make the necessary
adjustments, and assure continuation of the datums.
A potential problem for all types of water level gauges is that the pier itself could shift vertically,
either suddenly (perhaps due to a storm, which is easy to recognize; error Type 2 ) or slowly over
the years (error Type 3). The land itself, of course, can also move vertically, both suddenly (due to
an earthquake) or very slowly (due to things such as glacial rebound, sediment compaction,
extraction of fluids from the ground nearby). Operational standard operating procedures require
annual leveling to check for pier and/or local vertical movement in bench marks on land. Special
levels are run after earthquakes or known hits by storms or ships to check for vertical movement.
To track slow regional vertical land movement, tide stations are now being tied into nearby
continuously operating GPS systems that track land movement. These errors when they occur are
usually very small and mainly are of concern in the determination of long-term sea level change and
the updating of tidal datums. However, one should be aware of the possibility of some of these
errors affecting the results of a harmonic analysis of the data.
There are a variety of other very recent techniques for measuring water level, which are not yet
used on a regular basis, such as GPS receivers on buoys, laser systems, and land-based radar
systems. These will also have some type of potential instrument-caused errors, but they are still new
enough so that such potential errors have not yet been fully assessed. For a recent discussion and
assessment of water level measurement techniques see the Intergovernmental Oceanographic
Commissionss Manual of Sea Level Measurement and Interpretation, volume IV (IOC, 2006).
4.1.4 Assessing the Potential Effects of Nontidal Influences
As was seen in Section 2.3.3, in shallow-water areas the tide can be modified by the nonlinear
interaction of nontidal phenomena such as river discharge and storm surge. It was shown, for
example, that strong river discharge during a spring runoff period can shrink the tide range and shift
energy into the first overtide, increasing the M4/M2 ratio and distorting the shape of the tide curve
(perhaps leading to a more rapid rise to high water and a slower drop to low water). Storm surge
with frequencies below the tidal frequency bands (so that the water level change due to the storm
surge takes place slowly over a couple of days) also affects the tide through shallow-water nonlinear
mechanisms. The tide range might, for example, increase when the nontidal water level is higher
due to the storm surge, and decrease when the nontidal water level is lower due to the storm surge.
Changing water density (due to changing salinity in estuaries, or to changing water temperature
offshore) can cause various baroclinic effects (including internal tide waves), which though they
primarily affect tidal currents, could indirectly affect the tide through effects on energy dissipation
(see Section 2.3.6). There can also be seasonal effects on water level, such as higher water levels
during the summer due to the thermal expansion of the warmer upper layer of the ocean (see Section
3.7.1), during which time the tide range (and most tidal constituents) might be higher.
When one is harmonically analyzing a water level data time series one must therefore be aware
of the potential effects of nontidal phenomena on the harmonic constants that come out of the
analysis. If one needs harmonic constants from which accurate predictions can be made for most
146
of the year, then one might consider leaving out sections of the time series that include strong
nontidal events and using the gap option of the least squares harmonic analysis program. If one has
a short time series of water level data that includes a major nontidal event, one will probably not be
able to use the harmonic constants that come out of the analysis to make tide predictions for
normal time periods, i.e., for time periods without those strong nontidal phenomena. In fact, one
can probably not even use those harmonic constants to make a tide prediction for the time period
that included the nontidal event, because the effect of the nontidal event (which probably only lasted
a couple of days) was spread out over the length of the analysis. So even the results of a 15-day
harmonic analysis will not be useful. In such cases, one can try using the continuous wavelet
transform method (see Section 3.5.5).
To discover those periods in an observed water level data time series when a nontidal
phenomenon may have a strong influence, one can first tidally filter (see Section 3.8) the water level
time series. Although the filtering process may reduce the amplitude of such an effect, one should
still see periods when the nontidal water level is raised and then lowered (or vice versa). These may
be sections of the time series that one will ultimately decide to leave out of the harmonic analysis,
but initially at least one should analyze the entire time series, produce a predicted time series for the
entire period of the observations, and subtract the predicted series from the observed series to
produce a residual series. Then one should carefully examine the residual series. At those same
periods in time where one saw changes in nontidal water level in the filtered time series, one is
likely now to see oscillations at a tidal period appear and then disappear. One might even see
transient tidal oscillation at time periods without an obvious change in nontidal water level in the
filtered time series (which would imply the effects from nontidal currents or changing water
density).
When these transient tidal oscillations occur, it will usually be due to the effect of the nonlinear
interaction between the tide and a transient nontidal phenomenon, such as a large river discharge
(see Figure 4.1) or storm surge. During these periods of interaction, the nontidal phenomenon has
temporarily changed the hydrodynamics of the situation and thus changed the amplitudes and/or
Figure 4.1. The residual water level at Trenton, NJ, on the Delaware River, during a period
of increasing river discharge. The residual series was produced by subtracting the predicted
tide series (based on harmonic constants from a one-year analysis) from the water level data
time series. In addition to raising the nontidal water level, the high river discharge shrinks
the tide range and distorts the tide curve, so that tidal energy temporarily shows up in the
residual time series (see text, and also Figures 2.29 and 2.31).
147
phase lags of the tidal constituents. Thus, by subtracting the predictions based on the average tidal
constituents (obtained from the analysis of the entire time series) one leaves some tidal energy for
that particular time period. The transient tidal oscillation seen at time periods without an obvious
change in nontidal water level (in the filtered time series) would be time periods when there are
nonlinear effects of nontidal currents or perhaps even the indirect effect of a change in water density
that allowed a baroclinic effect to occur (although this effect is much more important for affecting
tidal currents).
When one sees nontidal events in the filtered time series and/or transient tidal oscillations in the
residual time series, one should look at wind or river or other data records for the potential nontidal
influences that may have caused the temporarily different tidal constituents. If such incidents occur
frequently, one may wish to add a footnote in a Tide Table about when such incidents occur.
(However, in cases where it is large nontidal water level changes that are nonlinearly modifying the
tide, those nontidal changes may overshadow any changes seen in the tide.) If such incidents occur
frequently, then water level prediction with a real-time driven water level forecast numerical model
may be the best route to take, because it can include not only the nontidal phenomena but also the
effects of their nonlinear interaction with the tide.
If there is a strong seasonal effect on nontidal water levels and, for example, one sees larger tide
ranges on average in the summer, one might need additional tidal harmonic constants in order to
include the essentially semiannual modulation of the tide range. The actual higher nontidal water
level in the summer might also be included in the Sa constituent, even though it is not tidal. (See
Section 3.7.)
4.1.5 Use of Node Factors and Equilibrium Arguments (or of Satellite Constituents)
As was seen in Section 2.2.4 there are some very slow modulations of the tide-producing forces
with periods of many years. The two most important are the 18.6-year lunar nodal regression, i.e.,
the variation in the angle between the plane of the moons orbits and the plane of the Earths
equator, and the 8.85-year variation in the lunar distance from the Earth due to the rotation of the
longitude of the lunar perigee. To include these effects in a complete harmonic analysis one must
add dozens of additional satellite tidal constituents, at least one for each of the lunar tidal
constituents, with frequencies very very close to the frequencies of those usual tidal constituents
(e.g., a separation of .0022o/hour or .0015 cpd), and one must analyze 19 years of data. Classically
such a method was not used because of the enormous required work effort (which could not even
be attempted prior to electronic computers) and because of the lack of 19-year long time series for
many stations.
The alternative approach, as explained in Section 2.2.4 is the use the node factor, f, a factor that
multiplies the amplitude of each tidal constituent (directly representing the modulation of the tidal
forces), as well as a phase difference, u, included in the equilibrium argument, that also varies over
the 18.6 years. Each node factor fluctuates around 1.0, with a variation of 4% for M2, 11% for
K1, 18% for O1, etc. The solar constituents such as S2 or P1 do not have node factors.
Because of the slowness of the variation in f and u, they would not change much in value over
the length of a typical harmonic analysis or tidal prediction, which is usually for a year or less.
Thus, for convenience, it became regular practice in the U.S. C&GS to use only a single value of
f and a single value of u for the entire analysis or prediction, and to use values for the center of the
time duration of the analysis or prediction. In some countries the values of f and u used in prediction
programs are changed every month (e.g., in the UK and in Canada). Typically the f and u values
148
(for the center of a particular year) were obtained from tables (such as Table 14 and 15 in
Schureman, 1958, or Zetler, 1982), but the astronomical equations with which they are calculated
can be included in any harmonic analysis or tidal prediction program and often were.
However, as was seen in Section 2.3.4, these astronomically determined node factors are based
on equilibrium theory and thus ignore the modifying influences of hydrodynamic effects. Although
the frequency of the typical tidal harmonic constituents is determined by astronomy, their amplitudes
and phase lags are determined by the hydrodynamics of the ocean and waterways on the Earth. The
same is true of the 18.6 and 8.85-year variations, and one should expect some modification of f and
u due to hydrodynamics. Because of the very low frequencies involved, however, the modifications
may be more due to nonlinear interactions than to typical near-resonance-type effects (see Section
2.3.1).
Figure 2.32 shows the true variation in the amplitude of six harmonic constants over 19 years
as determined from uncorrected one-year harmonic analyses (and compared with the variation based
on equilibrium theory), and in only one case does the variation in the node factor calculated from
actual data closely match the astronomically determined node factor (for O1, see Section 2.3.4 for
the reason). In these plots, besides the 18.6-year variation one sees other variations due to other
effects, some of which may be due to small wind-, river-, or temperature-caused changes in the
hydrodynamic system for a particular year.
The use of satellite constituents might capture some of this hydrodynamically caused
modification of the astronomical node factor, but only those variations that are periodic. Instead of
using the astronomically determined node factors, one could instead use calculated node factors (like
those in Figure 2.32), which could be from a nearby reference station if there is not enough data at
the station being analyzed or predicted. Using node factors calculated from actual water level data
will work for self predictions and predictions into the past, but not for predictions into the future.
In such a case, one would try to extrapolate forward the periodic signals seen in the most recent 19-
year or 38-year (or longer) analysis. But for many applications the increase in the accuracy obtained
may not be worth that effort.
Zetler, Long, and Ku (1985) did a study on the accuracy of tide predictions using satellite
constituents versus node factors (as well as comparing the use of an annual f and u, a bimonthly f
and u, and a monthly f and u). They did this by first doing a 19-year harmonic analysis of water
level data from Seattle, Washington, using 55 tidal constituents of which 30 were satellite
constituents. Tide predictions were made with these 55 constituents (the use of f and u was not
needed), and these predictions were subtracted from the observed time series to produce a residual
series. Similar predictions were then made using the primary 25 tidal constituents (i.e., no satellite
constituents), but using f and u, and again producing residual series. This was done three times, the
f and u held constant for each year, the f and u held constant for two month intervals, and the f and
u held constant for one month intervals. A fifth prediction was also made using satellite constituents
that were inferred (using the ratio of the satellite constituent to the main constituent as found in the
tide-generating potential in Cartwright and Edden, 1973). The residual variances were best for the
predictions using the satellite constituents obtained from the 19-year analysis, but only by a few
percentage points. Among the predictions using f and u, the results using f and u held constant for
one month or two month intervals were just barely better than the results when f and u were held
constant for a year, as were the results gotten from using the inferred satellite constituents.
Another thorough analysis of 19 years (and 38 years) of water level data using satellite
constituents was carried out by Foreman and Neufeld (1991). They used over 500 constituents.
They found that predictions based on a 19-year harmonic analysis with satellite constituents were
149
only slightly better than those based on averages from 19 one-year analyses using Godins (1972)
satellite correction algorithm and satellite inference (based on tide potential theory relationships).
To give the reader some feel for a 19-year analysis done using satellite constituents, Table 4.1
is reproduced from Zetler, Long, and Ku (1985) showing the 25 main tidal constituents plus the 30
satellite constituents that were calculated from 19 years of Seattle water level data. In this table one
can see the main harmonic constants (amplitudes and epochs) obtained from standard 19-year
analyses with and without satellite constituents, along with the inferred satellite constituents. The
very slight difference in angular speed (i.e., frequency) between corresponding main and satellite
constituents is apparent in the third column of the table, and the cause of this difference, either the
lunar node variation or the variation in lunar distance, is indicated by the change in 4th and/or 5th
Cartwright number (in the second column of the table) (see the end of Section 2.2.3). Some main
constituents only have lunar nodal satellite constituents (e.g., M2, N2, K1, O1, etc.), one or two
satellite constituents per main constituent. M1 and L2 have both lunar node and lunar distance
satellites, in which case there are four satellite constituents per main constituent. Solar constituents
have no satellite constituents.
In spite of the possible error in using a node factor (mentioned above and in Section 2.3.4), the
tendency has been to stay with that approach (versus using satellite constituents) for several reasons:
(1) there are still many water level stations without 19 years of data, and to be consistent the same
method has been used for all stations;
(2) it is a familiar approach that is more intuitively understandable than using dozens of additional
small satellite constituents;
(3) the additional accuracy does not seem to warrant (in most cases) the extra work of analyzing for
dozens of additional constituents and needing nearly 19 years of data for the analysis;
(4) even when there is some hydrodynamic nonlinear interaction that affects the 18.6-year
modulation so that it may not look very similar to the astronomically determined variation, that
nonlinearly-determined variation is more understandable in terms of its affect on the
astronomical nodal variation than trying to understand the changes in the many satellite
constituents.
4.2 Methods for Analyzing Short Time Series
As was seen earlier in this book (Sections 3.3 and 4.1.1) the number of tidal constituents that one
can accurately determine using harmonic analysis depends on the length of the data time series. The
length of the time series determines the frequency resolution of the analysis, that is, how close two
constituents can be in frequency and still be separated by the harmonic analysis. If one has two tidal
constituents that are closer in frequency than can be separated by the analysis (i.e., the time series
is too short), then only one of those two constituents (the one expected to be larger) can be solved
for. The calculated value for this larger constituent will not be as accurate as one would like,
because its value will include an unknown contribution from the second constituent that could not
be solved for. This contribution from the nearby unsolved-for constituent will vary from month to
month throughout the year.
When this occurs there are methods to try to improve the analysis results both to obtain values
for constituents that could not be included in the original harmonic analysis (called inference) and
to improve the value of constituents that were in the analysis but were adversely affected by the
constituents that were left out of the analysis (called elimination). Several of these methods for
inferring constituents that could not be solved for with the available time series, and for eliminating
150
Table 4.1. 55 tidal constituents from a 19-year harmonic analysis of Seattle

water level data, including 30 satellite constituents (indicated by a number in
parentheses following the name); see text. (From Zetler, Long, and Ku, 1985.)
151
the adverse effects of these missing constituents from those constituents that were solved for, are
presented in the next sections.
4.2.1 Schuremans Inference and Elimination Technique
4.2.1a Inferring Tidal Constituents Not Solved For
To infer tidal constituents that could not be solved for because the data times series was too
short, and to eliminate the effects of these missing constituents on those constituents that were
solved for, Schureman (1958, pages 78-87) uses a technique that first makes the assumption that the
amplitudes of the constituents of a similar type at any place, although differing greatly from their
theoretical values, have a relation that, in general, agrees fairly closely with the relations of their
theoretical coefficients. This technique also assumes that the difference in the epochs or lags of
the constituents have a relation conforming, in general, with the relation of the differences in their
speeds [i.e., frequencies]. Schureman recognized that the hydrodynamics of a waterway will make
the harmonic constants at a tide station quite different than they would be if equilibrium theory was
true (i.e., if the Earth was totally covered by one ocean and the tide responded instantaneously to the
astronomical tide producing forces; see Section 2.1.2), but for the purpose of his technique he
assumed that the relationships between pairs of particular harmonic constants will still be similar
to relationships between them if equilibrium theory were true. In reality this is only approximately
true, as the hydrodynamics of a waterway can significantly affect the relationships between two tidal
constituents, especially in shallow waterways. But to a first approximation this is still useful, and
for many areas, especially deep-water open-ocean areas, it can work reasonably well. (As will be
seen in the next section, there are ways to further improve this, if one has a long-term station
nearby.)
Thus, based on his first assumption above, to infer the amplitude, H(B), of a tidal constituent B,
based on another tidal constituent A with amplitude H(A), Schureman uses the formula
mean coefficient of B
H(B) = H(A) .
mean coefficient of A
The mean coefficients in this formula are values calculated by Schureman from astronomical
considerations (pages 21-24), which are listed in his Table 2 (pages 164-7). The relative magnitudes
of tidal constituents based on equilibrium theory are proportional to the mean constituent
coefficients.
Based on the second assumption above, to infer the epoch (i.e., the phase lag), (C), of a tidal
constituent C, whose angular speed c is close to the angular speeds a and b of tidal constituents A
and B [whose epochs are (A) and (B)], Schureman uses the formula
c-a [(B) (A)] .
(C) = (A) + b-a
By substituting the mean coefficient values (from Schuremans Table 2) into the first formula, and
the angular speeds into the second formula, Schureman produced the special formulas shown in
Figure 4.2.
Schuremans inference technique is built into the CO-OPS computer program version of
Schuremans Fourier harmonic analysis method, the original version written by Dennis and Long,
152
Figure 4.2. Schuremans formulas for inferring tidal

constituents from constituents already calculated. (From
Schureman, 1958, page 79, or Dennis and Long, 1971, page 6.)
1971 (the FORTRAN code for the inference technique is shown on pages 18 and 19 of that
publication). C&GS Form 452, page 115 in Schureman, 1958, was used in the hand analysis.
Schureman did a test of the reliability of his inference technique using 60 tide stations
representing various types of tide in different parts of the world. (He does not mention the exact
locations of the stations, nor how many stations were on the ocean coast versus inside shallow bays.)
He compared actual calculated values of M1, P1, Q1, K2, L2, and 2 (presumably from a one-year
harmonic analysis) with the inferred values obtained by using the appropriate formulas from Figure
4.2. He characterized his reliability test results (shown on page 80 of Schureman, 1958) as fairly
good, but admits that the inferred constants, especially the epochs, cannot be depended upon for
a high degree of refinement. For P1 , the most important constituent inferred in a 29-day analysis,
he found an average difference of 0.03 feet between the inferred amplitude and the actual amplitude,
and an average difference of 8o between the inferred epoch and the actual epoch. The largest
difference between inferred and actual amplitudes and epochs for P1 was 0.27 feet and 49o (one
would guess from a station in a shallow bay perhaps, but Schureman does not say). The results for
the rest of the test constituents were worse. Without knowing the locations of the 60 stations tested
one cannot make total use of these results. Hydrodynamics will change the relationships among
tidal constituents, even those close in frequency (speed), the magnitude of that change depending
on the particular hydrodynamic situation. As a general rule one would expect the most change in
long shallow-water bays, estuaries, or rivers (where the tidal wavelength has been shortened by the
shallow water).
153
4.2.1b Eliminating the Effects of Missing Constituents on Constituents That Were Solved For
Schuremans technique for eliminating the effects of tidal constituents that could not be solved
for on those tidal constituents that were solved for is complex enough that the numerous equations
will not be repeated here (see pages 80-87 of Schureman, 1958, as well as his Tables 21-26). The
FORTRAN code for this technique as implemented by Dennis and Long (1971) can be found within
their program on their pages 17, 18, 22, and 28 (which has Schuremans Tables 21-26). [C&GS
Form 245 on page 117 in Schureman was used in the hand analysis.] The most important
constituents considered are the disturbing effects on K1 by P1 and the disturbing effects on S2 by K2
and T2 . This is corrected (as best it can be by this equilibrium/astronomical-based method) by
accounting for the phase displacement and amplitude modification of the major constituents K1 and
S2 by the minor constituents P1, K2, and T2 , for which the inferred values obtained by Schuremans
technique are used. Since the method must begin with inferred values, a series of successive
approximations is used ( each time using the newly eliminated values for the disturbing constituents
in Schuremans formulas). Thus any degree of refinement could theoretically be achieved, but the
first approximation has generally been used, realizing that there are other effects that would
probably limit the benefit of continuing.
4.2.2 Use of a Reference Station To Improve the Results of a Short Analysis
Schuremans inference and elimination technique (as least for 15-day analyses) is better than
nothing, but its results are not as accurate as one would like for many applications. Schuremans
method assumes that equilibrium (i.e., astronomical) relationships among the tidal constituents hold,
that is, that the relative strengths and timing of the these constituents is the same as in the tide
producing forces. Thus, the hydrodynamics of the oceans and connected bays and rivers is ignored.
One can improve on this by making use of the tidal harmonic constants from a nearby long-term
water level station, as long as the two stations are close enough to have similar tidal characteristics.
In most cases it is more likely that, for example, the N2 /M2 ratio at the station with the short time
series will be closer to the N2 /M2 ratio at a nearby reference tide station than to the N2 /M2 ratio in
the astronomical tide producing forces.
The use of a nearby reference station is very straight forward. One merely needs a water level
data time series at the reference station for the same time period as the station with the short time
series. (If the reference tide station was not in operation during that period, one can use a
harmonically predicted time series for that period.) Then one harmonically analyzes that short series
from the reference station in exactly the same way the short data time series at the subordinate
station was analyzed. The results from this short analysis at the reference station are then compared
with the equivalent results from the one-year harmonic analysis at the reference station. For
example, the one-year M2 amplitude is divided by the short-analysis M2 amplitude (both from the
reference station), and the resulting ratio is then used to correct the M2 amplitude from the short
analysis at the subordinate station (by multiplying it with the ratio). Likewise, the short-analysis M2
epoch (phase lag) is subtracted from the one-year M2 epoch (both from the reference station), and
the resulting difference is used to correct the epoch at the subordinate station (by adding the phase
difference to it).
Another way to look at this, is that the results of, for example, a 29-day analysis at the
subordinate station will be different than the results that could have been obtained from a one-year
analysis (if one had had the data). This difference is due to the particular astronomical conditions
154
during those particular 29 days. One expects to see a similar difference at a nearby reference
stations during those same 29 days, but for this reference stations one does have one-year harmonic
analysis results, so one knows what that difference is. Thus one can correct the short subordinate
station using this knowledge from the reference station.
This method was used by Parker (1977) to correct the 29-day harmonic analysis results for
numerous tides stations along the Strait of Juan de Fuca Strait of Georgia, in order to construct
corange and cophase (cotidal) charts for the most important tidal constituents. Before using this
method it was found that the harmonic constants did not change smoothly along this waterway, and
it was suspected that the variation in the harmonic analysis results was due to the short records being
analyzed at most of the tide stations. Such variation was obvious because the waterway was deep
and thus the tide regime changed slowly, with tide stations separated along the coast by one or two
degrees of phase (which translates into only 2 to 4 minutes difference in M2 high waters, and 4 to
8 minutes difference in K1 high waters). The errors in a 29-day analysis, even one corrected using
Schuremans method based on astronomical/equilibrium theory can easily be larger than one or two
degrees of phase. And similarly for the amplitudes.
For this waterway there were ten longer-period tide stations for which one-year least squares
harmonic analyses were able to be carried out, and 58 stations for which 29-day harmonic analyses
had to be used (using Schuremans Fourier method and his inference and elimination technique).
Table 4.2 shows the correction factors that were calculated for one particular 29-day period at the
ten longer-period stations for the six largest tidal constituents. One sees that the correction factors
are different at each station, because of the changing tide regime with location. The largest changes
(i.e., ratios most different from 1.000) are found in the diurnal constituents, since for this waterway
P1 (which was inferred in the 29-day Schureman harmonic analysis) is the fourth largest tidal
constituent and requires a half year of data to separate it from K1.
Each of the 58 tide stations that had been analyzed using Schuremans 29-day analysis was
corrected using a pair of correction factors (for amplitude and epoch) from one of the ten long-
period stations (listed in Table 4.2), usually the closest geographically to the 29-day station. The
method appeared to work quite well. Prior to correction, harmonic constants from nearby 29-day
stations that had been operated during different time periods often were quite different. However,
once corrected they were closer in value and seemed to fit in much better with the geographic
change in values moving up the waterway. Phase corrections as large as 10o were made, with the
results fitting in better with surrounding stations. Three of the resulting charts from this process are
shown in Figures 2.5, 2.6, and 2.9.
Another way in which to accomplish the same type of correction of the harmonic constants of
a short station (secondary station) based on the harmonic constants at a nearby long-term station
(reference station), and a method that would make the process a little more automated, would be to
incorporate it into Schuremans inference and elimination technique, but to use the relationships
among the tidal constituents based on the nearby reference station instead of based on
equilibrium/astronomical relationships. This is, in fact, done in the least squares harmonic analysis
program by Foreman (2004) based on the method of Godin (1972) (see Section 4.2.4), as well as in
a recent version of the harmonic analysis program based on Schureman.
4.2.3 Inference In the Frequency Domain
By plotting the amplitudes of tidal constituents (of the same species) in the frequency domain
(and similarly for the epochs of the tidal constituents), one will gain some insights that one might
155
Table 4.2. Correction factors for ten different reference stations in the Strait of Juan de Fuca
Strait of Georgia for the 29-day period beginning March 1, 1974, for use in correcting the
harmonic constants from other stations with only 29 days of data. (From Parker, 1977.)
initially hope would allow one to infer the amplitude and epoch of missing tidal constituents, i.e.,
those that could not be analyzed with the available length of data time series. In doing so, one must
remember (again) that the variations in tidal constants throughout a waterway are due to
hydrodynamics, and barring the use of a hydrodynamic model, one is only looking for statistical
ways to essentially guess what those missing constituents will look like.
One would expect that the closer that two tidal constituents are in frequency the less the
relationship between those tidal constituents will change as one moves to another geographic
location, but it must be remembered that this is only true of the original astronomical tidal
constituents. In shallow waterways the nonlinearly produced shallow-water constituents (see
Sections 2.3.2 and 7.6.2) will typically grow much faster (and change phase faster) than nearby
astronomical constituents, so one must separate astronomical constituents from shallow-water
constituents when developing any inference and elimination scheme. The amplification and phase
change of all the astronomical constituents in a species (e.g., the semidiurnal band) will depend on
the nearness of the waterway to resonance conditions based on the length, depth, and (sometimes)
width of the waterway (see Sections 2.3.1 and 7.4.1). The amplification and phase change of all the
shallow-water constituents in a species (e.g., the semidiurnal band) will depend on the energy
transfer into them from the relevant astronomical constituents via one or more nonlinear
mechanisms. (The astronomical constituents will be reduced by this energy transfer, but not as
dramatically as the shallow-water constituents will be increased; see, for example, Figure 7.5.)
So the first question is, can one look at the tidal constituents (of similar nature, either
astronomical or shallow-water) surrounding a missing constituent and be able to infer a reasonable
amplitude and phase lag for that missing constituent. The answer is no. Without a nearby reference
156
station to look at, one will have no idea what the relative sizes of those astronomical constituents
should be. For example, in many waterways S2 is larger than N2, but in many others (including the
East Coast of the U.S.) N2 is larger than S2. This is because even for constituents fairly close in
frequency the hydrodynamics of a large waterway (like an ocean) can cause them to be fairly
different in certain locations (when tide waves travel long distances, there is plenty of time for two
constituents close in frequency to become quite different in amplitude and phase). Likewise
shallow-water constituents may grow at roughly the same rate in an estuary, but their energy comes
from astronomical constituents with different sizes, so here again one will have no idea what their
relative sizes should be (without looking at a nearby reference station).
Godin (1972, in his Figure 2.16, pp 180-1) shows how the relationship between close pairs of
constituents varies at 21 Canadian water level stations (from all its coasts). As would be expected,
there is a lot of scatter in his diagrams. Some have less scatter than others (e.g., the diagram for
P1K1 and the one K2S2), but these amplitude ratios and phase lags differences still depend on the
hydrodynamics of the particular locations chosen for the analysis, and there is no reason to assume
particular values for them (except initially the values from equilibrium theory, for lack of anything
else) unless one has values from a nearby (reference) station.
Thus, to infer missing constituents one is stuck with needing a nearby reference station (with
presumably similar harmonic characteristics). Schuremans reference station was actually the tide
producing forces themselves via equilibrium theory.
4.2.4 Other methods
Inference and elimination/correction is included in the least squares harmonic analysis program
of Foreman (2004) based on the method of Godin (1972). In that program one inputs, for a nearby
reference station, the amplitude ratio of the constituent to be inferred to the constituent that is
included in the actual harmonic analysis (which is close in frequency), and likewise the differences
in phase lags for the two constituents. The program calculates the amplitude and phase lag for the
missing constituent, as well as the corrected amplitude and phase lag for the close constituent that
was solved for in the analysis (and which had been influenced by the energy from the missing
constituent).
The mathematics behind this inference and correction process is given on pages 27-28 of
Foreman (2004), but it is useful to summarize it here (although using different notation). It involves
two assumptions (one not explicitly stated). First, it assumes that all the energy from the unsolved-
for constituent (the one to be inferred) was incorporated in the amplitude and phase lag for the
nearby solved-for constituent that came out of the least squares harmonic analysis. Thus, the
contribution to the tidal elevation of the uncorrected solved-for constituent (h1o) is assumed equal
to the contributions of the two constituents after inference and correction have been carried out
(h1+h2), namely,
h1o = h1 + h2
or
H1o cos(a1t1o) = H1 cos(a1 t 1) + H2 cos(a 2 t 2) (4.1)
where a1 and a2 are the angular speeds (frequencies) of the two constituents (the first constituent
being the one solved for in the least squares harmonic analysis), H1o, H1 , and H2 are the amplitudes
of the uncorrected solved-for constituent, the corrected solved-for constituent, and the unsolved-for
constituent (the one being inferred), and likewise for the phase lags (epochs) 1o, 1 , and 2 . Here
157
the node factors and equilibrium arguments have been left out for simplicity. Now let r12 be the
ratio of the inferred constituent amplitude to the corrected solved-for constituent, i.e.
r12 = H2
H1
and let be the difference in the phase lags of these two constituents, i.e.
= 1 2
Then equation (4.1) becomes
H1o cos(a1t1o) = H1 cos(a1 t 1) + r12 H1 cos{(a1 t1)+(a 2a1)t + )}

= H1 cos(a1 t 1) + r12 H1 cos(a1 t1) cos[(a 2a1)t + ]
r12 H1 sin(a1 t1) sin[(a 2a1)t + ]
H1o cos(a1t1o) = H1 cos(a1 t 1) {1+ r12 cos[(a 2a1)t + ]}
H1 sin(a1 t 1) {r12 sin[(a 2a1)t + ]} (4.2)
Since the length of the time series was not long enough to separate these two constituents, one
knows that the Rayleigh criteria was not met (see Section 3.3) and so the difference in the
frequencies was not greater than the frequency resolution of the analysis (i.e., 1/T, where T is the
length of the series). However, a second assumption is now made, that the difference in frequencies
is much smaller than 1/T. This is done to allow one to approximate the values in the { } in equation
(4.2), with their average value over the length of the series. After some further substitutions and
manipulations (see Foreman, 2004, page 28), one ends up with the amplitude and phase lag for the
corrected solved-for constituent as
H1 = H1o and = o
+ tan1 (S/C)
1 1
(C2+S2) 2
and the amplitude and phase lag for the inferred constituent as
H2 = r12 H1 and 2 = 1
where S and C are the terms that include the average value over the length of the series of the cosine
and sine terms in equation (4.2), which are
S = r12 sin[(a 2 a 2)T] sin and C = 1 + r12 sin[(a 2 a 2)T] cos

(a 2 a 2)T (a 2 a 2)T
(The equations in Foreman, 2004 used frequency instead of angular speed. The angular speed is in
o
/hour, where as the frequency in Foreman is in cycles per hour. One can go from angular speed to
frequency by dividing by 360o. For example, the angular speed for M2 is 28.984104, and its
frequency is 0.081 cycles per hour, or 1.932 cycles per day.)
Other discussions of methods for inferring tidal constituents from short data time series can be
found in Zetler, et al (1965), Dronkers (1964), and Zetler, et al (1985) in the latter case inferring
satellite constituents based on the tide potential. The continuous wavelet transform method (see
158
Section 3.5.5) is also useful for analyzing short data time series, although it does not use tidal
constituents (Jay and Flinchem, 1999; Jay and Kukulka, 2003).
4.3 Assessing the Quality of the Predicted Tide Series
4.3.1 Introduction
The typical way to evaluate the quality of the final set of harmonic constants (amplitudes and
epochs) that result from a harmonic analysis process (including any inference and elimination and
node factor application done after the actual least squares analysis) is to produce a tide prediction
time series with that set of those harmonic constants and then evaluate the quality of those tide
predictions. The need for a set of harmonic constants has been specifically referred to, because no
matter how accurate the constants may be for individual tidal constituents, it is the entire set that is
important. It does little good to have very accurate constants for M2 and K1, but not have other
constituents, since an accurate tide prediction can only be made if the harmonic constants for all the
important constituents are available (so that as much as possible of the tidal energy has been
accounted for). (Of course, for particular hydrodynamic research involving perhaps the
hydrodynamics of only M2, such a limited analysis might be acceptable, but even here most
hydrodynamics studies should really include all the important constituents.)
In addition to making tide predictions, one also subtracts these predictions from a simultaneous
observed water level data time series to create a so-called residual time series. This observed water
level time series might be the data that were analyzed to produce the set of harmonic constants, or
it might be from another time period. If it is the data that were analyzed, then the prediction for that
time period is referred as a self-prediction. As will be discussed, there are some limitations on using
a self-prediction to assess the quality of the set of harmonic constants, versus using predictions for
other time periods, but often one has no other data other than that which were analyzed. The
residual series will show all the variations in the time domain that are left over once the tide
predictions have been removed.
Obviously there will be water level variations in the observed series caused by nontidal
phenomena, such as wind, atmospheric pressure, river discharge, and changing salinity and water
temperature (steric effects), and which will not be in the predicted tide time series. The residual
series will have predominantly those nontidal water level variations (unless one has poor harmonic
constants, in which a great deal of tidal variations will show up in the residual series). Because of
these nontidal variations (which of course can be very large when there is storm or a heavy river
flow event) the tide prediction can be very different than the actual water level data on some days
and be very close to the water level data on other days. That is beyond ones control (and if one
wishes to do a good job of predicting the actual water level, one needs to use real-time based
hydrodynamic forecast models). The most one can do is hope to produce tide predictions which on
average match the water level time series as well as possible. However, as will be seen, it will still
be valuable to investigate the tide predictions on individual days as well as on the whole. For
shallow-water areas there will be times when such nontidal influences will change the tide through
nonlinear interaction (although in cases such as with storm surges, these changes will be still be
smaller than the nontidal variations that caused them).
No matter what type of evaluation method one uses, the issue of self-prediction versus prediction
for time periods other than that used to calculate the harmonic constants will be important. If other
data are available besides that which was used for the analysis, it is important to use them in a
159
thorough quality assessment. This is because one can often produce a high quality self-prediction,
even if one has not managed to calculate the optimum set of harmonic constants, because these
constants have been best fit to produce the best possible prediction for the analyzed data. Energy
from missing constituents (that should have been included in the harmonic analysis) will often still
be included in the constituents that were solved for. Those same constituents, however, may not do
as good a job predicting for other time periods.
If one does not have any data other than the data that were analyzed, one could in some
situations face making a choice between: (1) using all the data for the analysis and getting the
maximum number of constituents possible (but having no independent data to check the predictions
against); or (2) analyzing some portion of the data and using the other portion to check the
predictions (but perhaps having to solve for a smaller number of harmonic constituents). Generally,
where possible, one usually will opt for using as much of the data as possible to obtain as many tidal
harmonic constituents possible. If, for example, one had close to six months of data, one would
certainly want to use all the data and be able to calculate P1. But sometimes the length of the
available data time series will fall between two of the preferred lengths (e.g., between 6 months and
a year), in which case one can choose to use less than all the data (in this case, 6 months) and use
the remainder for testing the results. However, even here, when one is using the least squares
technique, one might still be able to successfully solve for additional tidal constituents with synodic
periods of one year, in which case all the data will again be used.
Even before getting to the point of making predictions with the best set of harmonic constants
that came out of the analysis, there is an advantage to having more than one time period with data.
In this case, however, one wants to do additional analyses to see if the epochs of the smallest
constituents are consistent from analysis to analysis. This consistency would be a sign that these
constants really represent those tidal constituents and are not just the amplitudes and phase lags for
some transient periodic nontidal phenomena.
Before doing the analysis, one might want to test the harmonic analysis program, to see how well
it does on a predicted tide time series generated from known harmonic constants. One would like
to see if the harmonic analysis program can reproduce exactly the harmonic constants that were used
to make the predicted tide time series, and if not, where energy is lost (or possibly gained).
However, a predicted tide time series has no noise, and as was discussed in Section 3.3 with
reference to Munk and Hasselman (1964), a low tidal signal-to-noise ratio can adversely effect the
calculation of the smaller tidal constituents, especially ones close in frequency to other constituents.
So if the harmonic analysis program passed the test with a pure predicted time series, one might
repeat the tests with additional noise added each time.
There are a number of ways to evaluate the quality of the tide predictions produced with a set
of harmonic constants. Three approaches will be primarily looked at:
(1) directly compare the high and low waters of the predicted time series with the high and low
waters of a simultaneous water level data time series;
(2) carry out and examine a spectral analysis of the residual time series; and
(3) examine the residual time series itself for periods with transient tidal oscillations.
160
4.3.2 Comparison of High and Low Waters In the Predicted Time Series Versus In the
Observed Time Series
A frequently used method for assessing the quality of a tide prediction time series is to compare
the heights and times of corresponding high waters (and low waters) in the predicted series and the
observed water level series. Here it is important to use the same method for calculating high and
low waters for both time series (see Section 3.9.1). Usually one looks at the height differences and
time differences between each pair of corresponding high waters (and low waters), the mean
differences, the standard deviations, and other statistical parameters.
This method is typically used to check the quality of the daily predictions for reference stations
in Table 1 of a Tide Table. The water level stations from which the harmonic constants for the
predictions were extracted are typically permanent stations, so that data are available every year to
evaluate the predictions. Each year some one checks how close the predicted high waters and low
waters are to the observed high and low waters. If each year they are getting further apart, then one
assumes that something is changing the hydrodynamic situation in the waterway. (And
investigations of possible changes to the bathymetry and/or shoreline of the waterway are carried
out.) One might keep the same harmonic constants until the differences between the predicted tide
series and the observed water level series are larger than some predetermined criteria based on
typical applications that use the predicted tide (such as safe navigation of deep-draft ships), but often
new harmonic constants will be produced before that point is reached.
It is important to remember that the observed series high waters and low waters includes
nontidal contributions, so there will be differences between the predicted high and low waters and
the observed high and low waters due to these nontidal contributions. These should mostly average
out when one is calculating a mean difference over a long enough time period. However, when
averaging over only a year, there can still be asymmetries in seasonal effects on the nontidal water
level that will not average out and will show up in the mean time and height differences. This can
be seen in Table 8 from the Tide Tables for the East Coast of North and South America (shown in
Figure 8.4). [Table 8 shows some statistical figures on tide prediction accuracy at 32 reference
stations in the Tide Table. However, technically, since one is comparing predicted and observed
values, one is not really looking at the accuracy of the tide prediction, but how well the tide
prediction can represent the actual water level (on average over a year). However, the statistical
information in Table 8 is useful for mariners who use tide prediction as the best estimate of actual
high and low waters that they will experience with their ships. See Section 9.2.3.]
Thus, in Figure 8.4, one sees larger mean differences between predicted and observed high
waters at Philadelphia, which is on the upper Delaware River. This station is influenced by the
changes in water level caused by the river flow, which will produce much higher water levels during
the strong freshets in the spring. Since the mean differences in Table 8, being based on only one
year of data, may primarily represent the nontidal contributions, the astronomical tide itself could,
in fact, be totally captured by the harmonically produced tide predictions, and thus could really be
99.99% accurate. If the comparison between predicted and observed high waters for Table 8 had
been done over many years, the mean difference values would be smaller, and probably more
representative of any missing tidal energy in the tide prediction (and thus a truer indication of the
accuracy of the astronomical tide prediction). In Section 3.7.1 and specifically in Figure 3.11 one
saw how much the seasonal sea level range can vary from year to year. The year chosen for the
statistics in Table 8 could have been a typical year or an unusual year. However, Table 8 is simply
an example provided in the Tide Tables. More than one year of high and low waters are typically
161
used to assess the quality of a tide prediction (and the set harmonic constants on which that
prediction is based).
Standard deviations (also presented in Table 8 of the Tide Tables) will always be affected by the
nontidal water level variations, no matter how long the time series of differences that is analyzed.
The four stations from the northern Gulf of Mexico (Pensacola, Mobile, Grand Isle, and
Galveston) shown in Table 8 in Figure 9.4 all have large standard deviations for the time differences
between predicted and observed times of high water and low water. At these locations the tide range
is small and is often dominated by water level variations caused by wind and pressure effects
(i.e., storm surge), which are usually stronger during the fall months.
Even with these nontidally caused differences it is still useful to look at how the individual time
and height difference values change throughout each month and from month to month. For example,
one might see a periodicity in the height (or time) differences that points to a problem with a specific
harmonic constant. Other longer periodicities or patterns in the height (or time) differences might
be due to nontidal seasonal effects. For example, both predicted high and low waters might appear
consistently too low in the summer because thermal expansion of the upper water layer has raised
sea level and thus raised both the high waters and the low waters. Such an effect, though not tidal,
might be captured to some degree in the Sa constituent, although its variance from year to year can
be considerable, and is another the reason why one cannot really include harmonic constants that
represent even quasi-periodic meteorological effects.
4.3.3 Spectral Analysis of the Residual Time Series
A typical way to check the quality of the calculated harmonic constants (i.e., some type of error
estimation) is to make a tide prediction for the time period of the data analyzed, subtract the
predicted time series from the original observed times to produce a residual time series, and then
to see how much energy is left in the residual time series at the tidal frequencies using a spectral
analysis. If there is a significant amount of tidal energy still present (relative to some predetermined
criterion), then possibly not enough tidal constituents were included in the harmonic analysis (or
possibly constituents with small amplitudes were erroneously rejected). If the number of
constituents was limited by the length of the available data time series, then additional constituents
probably need to be inferred. However, it is also possible that the tidal energy in the residual series
is due to erroneously inferred tidal constituents. Perhaps they were based only on the equilibrium
theory relationships, and one needs to use the relationships at a nearby longer-period station.
Although a spectral analysis is quite straight forward, when assessing the spectrum of the
residual series one should always remember (as mentioned above) that a self prediction can often
look good, even when an important constituent has been left out of the analysis (perhaps due to a
short data time series) and the harmonic constants that come out of the analysis are not as accurate
as they could be. This is because the constituents solved for are best fit to the data. The energy of
constituents not included in the analysis will be incorporated into one or more of the constituents
that are solved for, so that a self prediction for the same time period can look reasonable. The key
question, however, is whether one can predict accurately for other time periods. When predicting
for time periods other than the one analyzed, the way that the energy from the unsolved-for
constituents was incorporated into the solved-for constituents will not work for those other time
periods, because there will different phase relationships between the unsolved-for constituents and
the solved-for constituents for those other time periods.
162
4.3.4 Examination of the Residual Time Series For Periods With Transient Tidal Oscillations
In addition to doing the spectral analysis, one should also examine the residual time series itself
and look for time periods where oscillations with a tidal period appear and then disappear (see
Section 4.1.4). When these occur, it will usually be due to the effect of the nonlinear interaction
between the tide and some transient nontidal phenomenon, such as a large river discharge or storm
surge (or perhaps even the indirect effect of a change in water density that allowed a baroclinic
effect to occur, although this effect is much more important for tidal currents). During these periods
of interaction, the nontidal phenomenon has temporarily changed the hydrodynamics of the situation
and thus temporarily changed the amplitude and/or phase lag of one or more tidal constituents, and
so by subtracting the predictions based on the average tidal constituents (from the analysis of the
entire time series) one leaves some tidal energy for that particular time period. These various events
with temporary tidal energy will not show up in the spectral analysis because they will have
different phase relationships and cancel each other out in the overall average values determined by
the analysis.
When such incidents occur, one should look at the nontidal influences (such as wind, river
discharge, etc.) to try to determine the cause of the temporarily different tidal constituents. If such
incidents occur frequently, one may wish to add a footnote in a Tide Table about when such
incidents occur. However, some nontidal effects itself (such a storm surge or a river freshet) will
usually be larger than its nonlinear influence on the tide, and so the changes to the tide may be of
less importance. (If such incidents occur frequently, then water level prediction with a real-time
driven water level forecast numerical model may be the best route to take, because it can include not
only the nontidal phenomena but also the effects of their nonlinear interaction with the tide. See
Chapter 8.)
4.3.5 The Use and Misuse of Build-up Factors
In some cases, when one compares the high and low waters from a predicted tide time series with
those from an observed water level series, one will find that the high waters from the observed series
are on average higher than those from the predicted series and that the low waters from the observed
series are on average lower than those from the predicted series. In order words, the predicted tide
range (on average) is smaller than the actual measured mean tide range.
When this happens, one solution in the past has been to increase the amplitude of all the
harmonic constants (from which the prediction was made) by a percentage that will allow the
predicted high and low waters to have the same heights (on average) as the high and low waters in
the observed series. This has usually been referred to as the build-up factor. It was sometimes
reasoned that the harmonic analysis program itself probably lost some energy, and the assumption
was made that this energy was probably lost similarly over all tidal bands, so that one could
compensate by increasing all the tidal constituent amplitudes by the same percentage.
However, the more likely cause of this missing tidal energy is that certain tidal constituents were
probably not included in the analysis, and if they had been included that they would have accounted
for the missing energy. The missing constituents may have been left out because the analyzed data
series was too short (and not enough of them were inferred). Or, the missing constituents may have
been smaller constituents from the analysis that were rejected because their amplitude fell below the
criteria of 0.03 foot (about a cm), which was standard procedure at the U.S. C&GS since the time
of Schureman [who apparently in an informal office memorandum, rejected all analyzed amplitudes
163
less than 0.03 foot (Zetler, et al., 1985)]. If numerous small constituents were rejected that were
in fact perfectly good, that would account for much or most of the missing energy that would later
be compensated for with the build-up factor. For stations with enough data, a better criteria for
rejection of small constituents is a lack of consistency in the phase lag from year to year. If there
is consistency, then the constituent results are probably really tidal and should be used. When using
a least squares harmonic analysis program that determines the best order to solve for each additional
constituent and that calculates the reduction in variance after each additional constituent, one should
be able to ignore a constituent that does not add to the cumulative reduction of variance. However,
the opposite may not be true. A small constituent that reduces the variance could be representing
transient periodic nontidal energy. Again, if possible, one needs constituent results from the analysis
of at least one other time period.
Trying to compensate for missing tidal constituents by increasing the amplitudes of the other
tidal constituents will in fact lead to errors. Although the mean of the predicted high waters will
now match the mean of the observed high waters, some of the predicted high waters will end up
being too large and some will end up being too small (and likewise for low waters), because the
phases relationships involving the missing tidal constituents are different than the phase
relationships of the included (and built-up) tidal constituents.
The solution for a reference station or other permanent water level station (since one has plenty
of data) is to carry out a harmonic analysis on a longer time series and to include more tidal
harmonic constants. For a water level station with a limited time series length, the solution is more
difficult, and for many applications the predictions made using built-upharmonic constants may
be okay. If more accuracy is wanted, one can try inferring additional tidal constituents (see Section
4.2), and then re-evaluating the predicted and observed high and low waters.
4.3.6 Looking For Changes in Bathymetry or Shoreline
Although more important for tidal current predictions, changes in the depth of a waterway can
affect the tide, especially if the waterway is small and the change in volume of water in the
waterway during a tidal cycle is a significant portion of the total volume of water in the waterway.
Such changes in depth over time at a location may be rapid due to a hurricane or extratropical storm,
relatively fast due to a dredging project in a navigation channel, or fairly slow due to year-to-year
erosion and movement of sediment in the waterway.
When a dredging project to deepen a navigation channel in a small waterway has been
completed, one will know from the project report (and any follow-up hydrographic surveys) how
much the depths have changed. Tide stations will have been installed during and after the project,
so one will also be able to determine whether there has been a significant change in tide range has
occurred. If significant changes have occurred, new harmonic constants should be calculated from
a new (preferably at least year-long) data record taken after the dredging is complete. Similarly,
when the shoreline is changed significantly by filling in water areas to create land, or to create
artificial island, that may also change the tide in a small waterway.
When a hurricane or large storm has hit a small waterway, one should harmonically analyze
post-storm water level data and compare it to analyses of pre-storm water level data, to look for
possible changes in the tide which might warrant calculating new harmonic constants and new tide
predictions. The effect of slower changes (due to year-to-year erosion and sediment transport in a
small waterway) might not be seen unless one compares water level records and harmonic constants
separated by several years. In all these cases, the smaller the waterway the more likely that changes
164
in depth will have an effect on the tide (whereas for tidal currents, even local changes in depth in
a large waterway will affect the tidal currents).
4.4. Producing a Reference Station for Table 1 of a Tide Table
Very accurate tidal harmonic constants are required for a reference station that appears in Table
1 of the Tide Tables, since accurate tide predictions using those harmonic constants will be made
for many years in the future, and many subordinate stations in Table 2 will be referenced to that
station. The water level data time series for such a station must be long (at least a year) in order to
be able to use as many tidal constituents as possible in the harmonic analysis. Since most tide
reference stations in Table 1 coincide with one of the permanently installed water level stations
operated by CO-OPS, there will usually be many decades of data are available. However, the most
recent should be used especially in small waterways in case there have been bathymetric changes
which may have affected the tide regime. A long data time series not only allows one to calculate
as many tidal harmonic constituents as are needed to produce an accurate prediction (with the limit
on tidal constituents determined mainly by whether the smaller tidal spectral lines are sufficiently
above the nontidal continuum), but it also allows one to accurately determine the effects of the 18.6-
year lunar nodal cycle. In fact, if one wanted to, one could abandon the use of node factors and
simply add dozens of additional satellite constituents when carrying out the harmonic analysis and
subsequent tide predictions. However, the tendency has been to stay with the node factor approach
for reasons given in Section 4.1.5.
Typically the standard 37 constituents (listed in Table 3.2) have been used in CO-OPS (and its
predecessor organizations) unless the station is located in a shallow waterway with a large tide
range, in which case additional shallow-water tidal constituents are used (see Tables A.2). For
example, 114 tidal constituents were used for the reference station at Anchorage, Alaska. The 77
additional constituents were shallow-water constituents that were needed because of the 30-ft tide
range and the very shallow-water conditions at the upper end of Cook Inlet (there is even a tidal bore
that occurs at Turnagain Arm near Anchorage). However, when creating a new reference station,
or updating an old reference station, one should not feel it necessary to stick with the standard 37
constituents (even for waterways that are not as extreme as upper Cook Inlet). One should use the
methods mentioned in Section 4.3 for assessing the quality of the predicted tide series to decide
whether more constituents are needed (and whether a longer time series needs to be analyzed). One
should use as many tidal constituents as can be obtained from the available time series (plus inferred
constituents, if necessary) in order to accurately capture as much of the tidal energy as possible, and
to avoid the temptation to use a build-up factor (see Section 4.3.5). Constituents that come out
of a harmonic analysis with amplitudes less than some standard cutoff criterion for acceptance
(e.g., 0.03 feet) should not be rejected if they contributed to the cumulative reduction in variance
(usually calculated after each step by the least squares program). One should do analyses on
different time periods to see if the amplitude and epoch (phase lag) for these small constituents are
similar from year to year. Such consistency is a better acceptance criterion than using an arbitrary
cutoff criterion, or even looking for a reduction in variance during the analysis.
165
4.5 Summary Overview: Steps In Harmonically Analyzing Water Level Data
In this section we will summarize the steps in harmonically analyzing a water level data time
series. Here we will assume we are using a least squares harmonic analysis program. In many cases
more than one harmonic analysis will need to be carried out on the data, with adjustments being
made after each analysis if needed (such as adding additional constituents after assessing how much
tidal energy is left in a residual series). Some steps may not be necessary for every analysis.
Sections where more information can be found are in brackets [ ] at the end of each step.
These steps, in bullet form are:
(1) when first using a harmonic analysis program it may be a good idea to test the program on
a predicted tide time series based on known harmonic constants, to see if the program can
reproduce exactly the harmonic constants that were used to make the predicted tide time
series, and if not, to try to determine where energy is lost (or possibly gained). [Section
4.3.1]
(2) determine how much accuracy one needs in the harmonic constants (and in predictions to
be made with those constants) for your particular application; that will affect how many of
the steps below one will choose to use;
(3) assess the data available at the water level station; if more than one time series is available,
use the longest most recent data; (especially for small shallow waterways) check the records
for dramatic changes in bathymetry (e.g., due to storms or to dredging); if such changes have
occurred, select data (if possible) from after those changes occurred. [Section 4.3.6]
(4) determine if there are gaps in the water level data time series; still use all the data but make
sure to select the option in the least squares program that handles gaps (or random time data)
since some least square programs distinguish these two cases from a continuous data record
(because the for the latter, a different numerical scheme is typically used to analyze the data
more quickly). [Section 3.4.5]
(5) determine which astronomical tidal constituents will be included in the program, initially
based on the length of the water level time series; use Table 3.2 or Tables A.2 or to make this
determination, selecting all the constituents with synodic periods less than or almost equal
to the length of the data time series; in subsequent runs of the program more constituents
may be added, especially if the length of the data time series is fairly close to the next
highest synodic period. [Section 4.1.1]
(6) determine whether additional shallow-water tidal constituents should be included in the
analysis; this will be obvious for a very shallow waterway with a large tide range, but one
may first have to use the standard 37 constituents and see how much energy is left in the
higher harmonic tidal bands of a spectrum; for an extreme case, the highest order overtide
that can be solved for will depend on the sampling rate of the water level time series, but
hourly data is sufficient for almost all locations. [Section 4.1.1]
(7) determine how much of the data time series to use in the harmonic analysis; in Step (5) the
longest synodic period within the length of the available data time series was selected, but
one may have more data than that (but less than the next highest synodic period); run the first
analysis using a segment of the time series whose length is the selected synodic period; if
there is enough extra data, it will be useful to run one analysis from the beginning of the time
series, and one that includes the end of the time series (and look for consistency in the
harmonic constants for each, especially in the smallest constituents); if the length of the
available data series is more than half way to the next synodic period, the follow-up
166
analyses can use all the available data and include additional harmonic constituents that
theoretically require the next highest synodic period; if one has many years of data, one will
initially do several one-year analyses, and then decide based on how much tidal energy is
left in the residual series (and whether the smaller constituents are consistent from year to
year), whether to analyze a two-year or longer time series. [Section 4.1.1]
(8) decide whether to use a particular shallow-water compound tidal constituent or the
astronomical constituent with the same frequency (e.g., 2MN2 or L2); for deep-water stations
the astronomical constituent is fine, but for shallow-water stations (especially inside bays
and rivers) one will probably need to use the shallow-water constituent; if one is unsure, one
will have to look at the harmonic analysis results of several stations in the bay and see if the
rate of growth moving up the bay is greater than the rate of growth of an astronomical
constituent in the same tidal band (e.g., see if 2MN2 grows at a faster rate than M2); for
constituent pairs whose node factors are the same (e.g., 2MS2 and 2) it does not matter
which is selected; for pairs with different node factors, if one initially chose the astronomical
constituent, but later decides it should have been the shallow-water constituent, one does not
have to re-analyze, for one can merely correct for the different node factor and equilibrium
argument. [Section 4.1.1]
(9) check the data file for notes about instrument errors, datum shifts, or other potential
problems with the data; if such problems are noted, keep in mind the possible effects of the
five types of errors on the harmonic constants; [Section 4.1.3]
(10) tidally filter the water level time series to look for strong nontidal events (like spring river
runoff, or storm surge) that may nonlinearly affect the tidal analysis results; if one needs
harmonic constants from which predictions can be made for most of the year, one should
consider leaving out strong nontidal event periods and using the gap option of the least
squares program; if one has a short time series of water level data that includes a major
nontidal event, one will probably not be able to use the harmonic constants that come out of
the analysis for tide prediction; [Section 4.1.4]
(11) carry out the least squares harmonic analysis; use all the constituents that contributed to the
cumulative reduction in variance, even if their amplitudes are very small (e.g., lower than
some standard lower cutoff criterion such as 0.03 feet); if additional data are available, run
additional harmonic analyses to see if these smaller constituents are consistent from year to
year; inconsistent epochs are grounds for rejecting a constituent; [Section 4.3.5]
(12) with the harmonic constants from the first harmonic analysis, make a tide prediction for the
same period as the data, and subtract this predicted time series from the original observed
time series to produce a residual time series; if one has data for another time period besides
the one analyzed, make tide predictions and produce a residual time series for that other time
period; this other residual time series will provide more meaningful results in steps (13) and
(14) because it does not represent a self-prediction. [Section 4.3.1]
(13) carry out a spectral analysis on the residual time series to see if there is any remaining tidal
energy still in the record; also carry out a spectral analysis on the original observed time
series for comparison purposes. [Section 4.3.3]
(14) examine a plot of the residual time series looking for places where a transient tidal signal
may appear; compare to the results of Step (10); investigate the wind, river, or other
potential nontidal influences for those time periods; one may have to consider leaving strong
nontidal event time periods out of the analysis and rerunning the least squares harmonic
analysis using the gap option; if one has a short time series of water level data that includes
167
a major nontidal event, one will probably not be able to use the harmonic constants that
come out of the analysis for tide prediction. [Section 4.3.4]
(15) if there still is tidal energy in the residual time series try to determine what additional tidal
constituents might be added to the next harmonic analysis; most easy to spot will be energy
in higher species bands due to shallow-water effects; decide which additional shallow-water
constituents are most likely to be important [Table 3.2 or Table A.2];
for energy left in the semidiurnal or diurnal bands there may be compound tidal constituents
(due to shallow-water effects) that should be added to the next analysis; [Table 3.2 or Table
A.2]
(16) obtain values for additional constituents; if one does not have enough data to move to the
next synodic period, one can then either:
(a) use an inference (and elimination) technique to calculate the amplitudes and epochs for
the missing constituents, using a nearby reference station if possible; [Section 4.2]
(b) try rerunning the harmonic analysis with these constituents included using the full length
of the available time series, to see if the least squares technique can successfully
determine them;
if one originally carried out a one-year analysis, and has more available data, consider
running a longer analysis so as to include additional tidal constituents.
(17) when one has decided that they have the optimum set of harmonic constants for a particular
water level station, run a final prediction for a year of data (or for the full length of data, if
less than a year) and calculate all the high and low waters; compare the predicted high and
low waters with the high and low waters calculated from the data time series, determining
all the usual statistics; (if possible, use the same max/min routine for both predicted and data
series); if the mean high waters from the data are higher than the mean predicted high waters
(and likewise if the mean low waters from the data are lower than the predicted low waters),
consider going back to Step (15);
(18) for producing predictions for reference stations in the Tide Tables, use a build-up factor on
the final harmonic constants only as a last resort, and only if the average predicted high and
low waters differ so significantly from the observed mean high and low waters that they
would make a noticeable difference to the mariner. [Section 4.3.5]
168
5 Harmonic Analysis of Current Data
5.1 Special Problems With Current Data
The special aspects of tidal currents were described and explained in Section 2.3.6 (that section
should be read before reading this section). Those special aspects are now looked at again with an
eye toward the difficulties they might present when harmonically analyzing current data. Many of
these problems have led some oceanographers to be very pessimistic about how well one can
represent the tidal current using harmonic analysis or other techniques. Here we believe that most
of these problems can be overcome if treated carefully, assuming we have good quality data of
sufficient length. In some cases, as will be discussed below and in Chapter 8, the use of numerical
hydrodynamic models can do a better job than harmonic prediction methods (although, if not
properly used, e.g., if too coarse a resolution is used, models will provide less accurate results).
As mentioned in Sections 2.3.6 and 3.4.3, the primary difference between a time series of current
data and a time series of water level data is that water level is a simple one-dimensional scalar
quantity and the current is a two-dimensional vector quantity (that also varies along a third
dimension, i.e. depth). One can directly analyze the time series of water level heights, but to analyze
the time series of current speeds and directions one must first convert each speed and direction pair
into two orthogonal components, such as north and east components. Often major and minor
components are chosen, the major axis usually being the direction of maximum flood with the minor
axis 90o clockwise from the major axis. Thus, to harmonically analyze current data one must
harmonically analyze two data times series instead of one. This in itself is not a problem; it simply
adds little extra work, and it requires some additional knowledge with regards to interpreting the
analysis results. The pairs of harmonic constants that come out of a harmonic analysis of current
data can be combined into tidal current constituent ellipses (which are derived and explained in
Section 5.2), which show how the flow from each tidal constituent rotates around the compass
during a constituent cycle.
With regard to actual problems with current data, they will be discussed in roughly the same
order as they were presented in Section 2.3.6. This discussion will include the possible effect of
these problems on the accuracy of the analysis and prediction of the tidal current, and how to avoid
or mitigate such problems.
Current data time series have typically been much shorter than water level times series because
of the difficulty and expense of maintaining current meters in the open water, as compared with
maintaining a water level gauge on a pier or other land-based platform. Current meters also have
tended to have a worse fouling problem due to biological growth or debris clogging the sensor
(especially when propellers or rotors were used). However, with newer technology this is much less
of a problem today, and even if the time series are short, there are ways to get useful information out
of the data. So this is now not usually a critical problem.
169
Current data have also tended to be noisier (in some cases much noisier) than water level data,
and that noise could adversely affect the harmonic analysis results especially in locations where the
tidal signal is not that strong. The degree of this problem has depended on the type of current meter
used, and how much averaging was done to acquire each measurement. Mechanical current meters
that used an S-shaped savonius rotor to measure the speed and a separate (generally quite small)
vane to measure the direction usually had the worst noise problems, especially when hung from a
surface buoy when there were significant surface waves. There was less of an instrument-caused
noise problem when using a propeller and a larger vane, and still less when using two propellers at
right angles (to measure two orthogonal components directly, instead of speed and direction). There
was even less of a noise problem with electromagnetic in situ sensors, which also measured two
orthogonal components. Likewise, acoustic doppler and high frequency remote sensors have much
less noise problems, since there is vectorial averaging over time intervals and over volumes of water.
Thus, instrument-caused noise is less of a problem today. There is still natural noise due to
meteorological effects, and this is generally more significant for currents than for water levels, but
most effects of this noise are eliminated by tidal analysis because of its emphasis on energy at
particular frequencies and the inherent averaging. Those effects that can be significant will be talked
more about below.
More serious is the spatial variation in tidal currents, which is much more rapid than for the tide.
Tidal currents can vary rapidly both vertically and horizontally. Tidal currents vary with depth due
to the effects of (especially) bottom friction, surface friction (from the wind), and baroclinic
(density) effects, the latter often due to internal tide waves. The most dramatic changes in tidal
currents in the horizontal direction are typically in going from deep channels (with faster currents)
to shallow water next to the channels (with slower currents). (See examples in Section 2.3.6).
In some cases this spatial variation in the tidal current can be so rapid that the particular
placement of a current sensor can make a significant difference in the analysis results. If there is
an unintended change in the position of the current sensor during the period of measurement (e.g.,
it was moved by a storm or by a strong current) and one analyzes the entire record, then one is really
combining the results for essentially two different locations. In such cases, one should break the
current time series up into two pieces and analyze them separately. If the current sensor slowly
changes position through the measurement period (due to strong currents), that is a serious problem,
and all one can do is analyze different portions of the record and try to assess whether there is
enough similarity (or an obvious trend) so that the results are usable. Another serious problem is
when a current sensor is hung from a surface buoy with a long anchor line, and the buoy and sensor
swings around to different locations during flood and during ebb. This can be especially bad if the
buoy was anchored near the edge of a navigation channel and the buoy and sensor move into shallow
water during either the flood or ebb phase.
The solution to many of these problems is proper planning and good field work. One should try
to install a current station where one expects less horizontal variation in the tidal current field, such
as in the center of a channel, (rather than at the edge) or in the middle of a straight channel (rather than
in a channel bend). Although information is still certainly needed at the edge of a channel and in a
channel bend, it may be better to use a high-resolution hydrodynamic numerical model to produce
those tidal current predictions, and to use a towed ADCP and/or HF radar to obtain observations of
such horizontal variation. Most important, one needs to make sure that the current station is not
moved during the observation period. If such movement does occur, perhaps due to a storm or due
to being dragged by strong currents, then one may need to break up the current time series into one
or more pieces to be analyzed separately.
170
5. Harmonic Analysis of Current Data
Dealing with the change in tidal currents with depth requires a different approach. The effects
on tidal currents of internal tide waves or other baroclinic effects are not easy to handle because such
effects are generally intermittent. In areas where there is believed to be some stratification, or even
a lateral variation in salinity or water temperature, one should take vertical CTD (salinity and
temperature) profiles during the current measurement period to be able to assess the changing
density fields. In rivers and estuaries one can choose not to harmonically analyze current
measurements made during heavy river runoff periods. Offshore, internal tide waves move along
density fields due to vertical temperature gradients (of which the thermocline is the most dramatic),
but these are more likely to happen during warmer months and when the upper layer is warmer and
there is little wind to mix the water column. Data obtained during periods with less baroclinic
effects will provide reasonable tidal current predictions for most of the year. For dealing with those
periods with baroclinic effects one might consider trying the continuous wavelet transform analysis
method (Section 3.5.5), but any results can probably only be included in a Tidal Current Table as
a footnote for the particular current station.
Even without baroclinic/density effects, the tidal current speed and times of maximum flow will
change with depth due to bottom friction effects, as well as effects near the surface due to wind.
And so, one would hope that an in situ current sensor could be kept at the same depth throughout
the measurement period (or similarly, the bins of an ADCP should be kept consistent). However,
this may be more difficult than one would think, if there is a reasonable tide range. In shallow water
there can be a difference in the currents measured by a bottom-mounted sensor compared with the
currents measured by a sensor attached to a surface float, because the rise and fall of the tide changes
the distance from the surface-float-hung sensor to the bottom. The bottom-mounted sensor will be
measuring currents at different depths relative to the surface, whereas the sensor hanging from a
surface float will be measuring currents at different distances relative to the bottom. Throughout
a tidal cycle an HF radar system always measures currents at the surface. In most cases, however,
one would expect the differences due to such depth changes to be relatively unimportant for most
applications. However, with a bottom-mounted ADCP, one can see how the entire velocity profile
changes over a tidal cycle and thus get some idea of the depth dependencies.
Tidal currents are generally much more susceptible to the distortion by nonlinear shallow-water
effects. They are also affected by nonlinear lateral inertial effects (see Section 2.3.6e) that do not
affect the tide at all. For example, Figure 2.35 provided an example of dramatically changing
(K1+O1)/M2 and M4/M2 ratios in the tidal currents across a waterway, but there was no such variation
in the tide. This is not really a problem, as long as one is aware of this, and makes sure to look for
the results of this distortion when analyzing current data. Thus one would expect to see stronger
overtides (i.e., higher M4/M2, M6/M2, M8/M2 ratios, etc) and stronger compound tides. As already
mentioned, a particular compound constituent can have the same frequency as an astronomical
constituent but be much larger (e.g, 2MN2 is often larger than L2). These two constituents may have
different node factors and equilibrium arguments, and so give different predictions for years other
than the one from which the analysis came. (2MN2 and L2 have different node factors, while 2MS2
and 2 have the same node factor.) There may also be those very rare occurrences such as seen in
Ramshorn Creek (see Figure 2.36) where the shallow-water effect is so strong (in this case enhanced
by the superposition of two tide waves crossing) that the tidal current becomes a true quarter-diurnal
tidal current. Without recognizing the hydrodynamic cause of this (admittedly rare) occurrence, and
upon seeing the tidal current change directions four times a day instead of two, one might think that
there was a clock problem with the current sensor.
171
There is also the special situation which occurs when a current sensor is placed at a channel
bend, or any other location where the direction of maximum flood is not exactly 180o from the
direction of maximum ebb. When current data from a sensor in such a location are harmonically
analyzed, one will see an apparent mean current and artificially higher M4 values that are not caused
by the shallow water (nor even by an inertial effect if the channel is narrow), but are simply due to
the geometry of the situation. Figure 5.1 shows such a situation. When analyzing the current data
from this location, one would probably select the major axis along neither the flood direction nor
the ebb direction, but at some compromise position (in this case in the north direction, with the
minor axis pointing eastward). One can see that as the major-axis component goes through one
complete cycle, that the minor-axis component will actually go through two complete cycles (the
minor axis component will always flow toward the east, but will go through two oscillations for the
one oscillation of the major axis). The harmonic analysis would show an M4 and other apparent
overtides. The analysis would also calculate a mean current toward the east, which, of course,
cannot be interpreted as a real mean current upon which the tidal current was superimposed (as one
could interpret a tidal current superimposed on a river current) Again, this should not present a
problem, since predictions based on the orthogonal pairs of harmonic constants should reproduce
the tidal current accurately, but one must understand what is behind it.
Figure 5.1. A current station deployed at a bend in the waterway,

so that the flood direction and the ebb direction are not 180o apart.
Because of this geometry a harmonic analysis will show a mean
current toward the east and a significant value for M4.
5.2 Tidal Current Harmonic Constituent Ellipses
One way to recombine the results of analyzing the two orthogonal component time series is the
use of harmonic constituent ellipses. The result of the harmonic analysis of a time series of current
data is two sets of harmonic constants, one for each of the two orthogonal axes that were chosen
when the original speed and direction data time series were turned into two component time series.
If the two orthogonal components chosen were major-minor components, and the major axis was
chosen to be the direction of maximum flood, then by looking at these harmonic constants one can
get some understanding of the harmonic makeup of the tidal current in the up-channel-down-channel
direction (i.e., the primary flood and ebb flow), as well as the harmonic makeup in the cross-channel
direction. If the two orthogonal components chosen were north-east components (as has often been
172
done by default) and these do not line up with the predominant major-minor axes of flow, then
interpretation of the harmonic constants is limited until they are recombined.
To see how the speed and direction of flow due to a particular tidal constituent changes each
hour, and to watch it rotate around the compass over a tidal constituent cycle, one needs to combine
the two component harmonic constants for that constituent. The result will be an elliptical
representation of the flow due to that tidal constituent. The width of the ellipse will vary
considerably with location, from very narrow in a narrow waterway to very wide (and thus circular)
in the open ocean, with every possible situation in between.
To derive the formula for a tidal constituent ellipse from a pair of harmonic constants for two
orthogonal components, the derivation of Doodson and Warburg (1941, pp 180-1) is followed fairly
closely, but instead of restricting it to north and east components, the formula is derived for any pair
of orthogonal components, since often harmonic analyses are done using major-minor components.
However, to keep it as close as possible to Doodson and Warburg, the formula for north-east
components is derived first and then modified later for any orthogonal pair. This looks like a messy
derivation but it simply involves the use of angle-sum and angle-difference formulas from a standard
trigonometry text. The notation of Doodson and Warburg is used as much as possible, but it has
been modified slightly for clarity, and here directions are clockwise from north (instead of
counterclockwise from east) so that they match up with the standard compass directions.
Figure 3.4 shows a current vector, W(t), and its north and east components, WN(t) and WE(t),
which are functions of time (that figure also shows a pair of major-minor components, which will
correspond to the directions of the axes of the ellipse) . The time variation in the north and east
components is expressed in the normal fashion, similar to the harmonic prediction equation (3.1) in
Section 3.4.2 (but with the h replaced by an N or E, and without the node factor and equilibrium
argument, and using Doodson and Warburgs notation for frequency, n), as
N(t)= WN cos(nt - N) and E(t) = WE cos(nt - E) .

= WN cos nt cos N + WN sin nt sin N = WE cos nt cos E + WE sin nt sin E
= A2 cos nt + B2 sin nt = A1 cos nt + B1 sin nt
where the second line of each formula has been expanded out from the first (with an angle-difference
formula), and the third line has been simplified by using constants A1, A2, B1, and B2 for the known
quantities made up of the component harmonic constants and defined by:
A1 = WE cos E ; B1 = WE sin E ; A2 = WN cos N ; B2 = WN sin N
The square of the current speed at any moment in time can then be given by
W2 = (A1 cos nt + B1 sin nt)2 + (A2 cos nt + B2 sin nt)2

= C cos2nt + D sin2nt + 2E sin nt cos nt
where
C = A12 + A22 D = B12 + B22 E = A1B1 + A2B2 .
Letting F = (C-D) and G = (C+D), which equivalently is C=F+G and D=GF, then (using
double angle formulas) the equation for W2 can be written as
W2 = F cos 2nt + E sin 2nt + G .
173
This can be further simplified to
W2 = G + H cos2(nt - a)
where
F
H = cos2a E
= sin2a and H2 = E2 + F2 and a = tan-1 E
F .
and a is the phase of the maximum constituent current. This occurs when nt = a or when nt=a+180o,
in which case
Wmajor = (G + H)
This happens twice in a constituent cycle, when the constituent current vector is pointing toward
either end of the major axis of the constituent ellipse. The minimum constituent current occurs when
nt = a +90o or a90o, in which case
Wminor = (G H)
This also happens twice in a constituent cycle, when the constituent current vector is pointing toward
either end of the minor axis of the constituent ellipse.
This can be described with a drawn constituent ellipse, with the major axis of length twice Wmajor
and the minor axis length of twice Wminor . The major axis direction (clockwise from north) is
major = tan-1 ( A2 sin a + B2 cos a

A sin a + B cos a ) .
1 1
Likewise the direction of the minor axis (clockwise from north) is
minor = tan-1 ( A2 sin a + B2 cos a

A sin a + B cos a ) .
1 1
If minor = major +90 , then the rotation of the ellipse is clockwise,

minor = major 90 , then the rotation of the ellipse is counterclockwise.
The phase lag for Wmajor is

major = tan-1 E
F .
The phase lag for Wminor is
minor = major + 90 , when the rotation is clockwise
and minor = major 90 , when the rotation is counterclockwise.
The eccentricity of each ellipse is a measure of how narrow or wide the ellipse is (see Figure 5.2).
It is defined as
2
e = (W major W2minor) .
Wmajor
Eccentricity will go from 0.0 for a circle (where the major and minor axes are equal, and the current
is pure rotary) to 1.0 for a very narrow ellipse (essentially a reversing current)
174
Figure 5.2. A fairly narrow tidal constituent ellipse (i.e., with

an eccentricity approaching 1.0). See text for explanation of
parameters.
These results based on north-east components, called N(t) and E(t), will now be generalized for
any pair of orthogonal components, such as (user-selected) major-minor components, called MJ(t)
and MN(t). From Figure 5.3, which shows the geometry and trigonometry for converting between
north-east components and major-minor components, the north component in terms of the major and
minor components (where is the major axis direction clockwise from north) is:
N(t) = MJ(t) cos MN(t) sin

= WMJ cos(ntMJ) cos WMN cos(ntMN) sin
= WMJ cosnt cosMJ cos + WMJ sinnt sinMJ cos
WMN cosnt cosMN sin WMN sinnt sinMN sin
= (WMJ cosMJ cos WMN cosMN sin) cosnt
+ (WMJ sinMJ cos WMN sinMN sin) sinnt
= A2 cos nt + B2 sin nt
where
A2 = WMJ cosMJ cos WMN cosMN sin
B2 = WMJ sinMJ cos WMN sinMN sin
And likewise the east component in terms of the major and minor components is:
E(t) = MJ(t) sin + MN(t) cos

= WMJ cos(ntMJ) sin + WMN cos(ntMN) cos
= WMJ cosnt cosMJ sin + WMJ sinnt sinMJ sin
+ WMN cosnt cosMN cos + WMN sinnt sinMN cos
= (WMJ cosMJ sin +WMN cosMN cos) cosnt
+ (WMJ sinMJ sin +WMN sinMN cos) sinnt
= A1 cos nt + B1 sin nt
175
where
A1 = WMJ cosMJ sin + WMN cosMN cos
B1 = WMJ sinMJ sin + WMN sinMN cos.
The same derivation as previously used applies, but now these new values for A1, B1, A2, B2 are
used.
Sometimes it is convenient to be able to convert harmonic constants that were calculated using
north-east components directly into harmonic constants for major-minor components, which is a
form in which they can be more easily interpreted even without looking at ellipses. There might be
cases where the tidal signal in the cross-bay direction (the minor axis) is small enough that one
might be worried about extracting accurate harmonic constants for this component. Since there
usually is a very strong tidal signal in the flood-ebb direction (the major axis), it might be
advantageous to purposely run the analysis with north-east components in order to have a strong
tidal signal in both components. Then once the harmonic constants for the north-east components
are calculated, they are then converted to major-minor components.
From Figure 5.3 one can start with the formulas
MJ(t) = N(t) cos + E(t) sin and MN(t) = E(t) cos N(t) sin ,
which will be manipulated (in a manner similar to above) to obtain the formulas for the
epoch and amplitude for the major axis component in terms of the epochs and amplitudes of the
north and east components, and similarly for the minor axis component. Only a couple of steps are
Figure 5.3. The geometry and trigonometry for converting between north-east
components and major-minor components.
176
given in finding the major axis epoch and amplitude.

MJ(t) = N(t) cos + E(t) sin
WMJ cos(ntMJ) = WN cos(ntN) cos + WE cos(ntE) sin
WMJ (cosnt cosMJ + sinnt sinMJ) = WN (cosnt cosN + sinnt sinN) cos
+ WE (cosnt cosE + sinnt sinE) sin
WMJ cosMJ cosnt + WMJ sinMJ sinnt = (WN cosN cos + WE cosE sin) cosnt
+ (WN sinN cos + WE sinE sin) sinnt
then
WMJ cosMJ = WN cosN cos + WE cosE sin
WMJ sinMJ = WN sinN cos + WE sinE sin
and dividing the first equation by the second to get tan MJ on the left side, the major axis epoch
then is
MJ = tan1 WE sinE sin + WN sinN cos
WE cosE sin + WN cosN cos
Then the major axis amplitude is
WMJ = WE cos E sin + WN cosN cos

cosMJ
using the value just obtained for MJ.
Similarly for the minor axis component one obtains
MN = tan1 WE sinE cos WN sinN sin

WE cosE cos WN cosN sin
and
WMN = WE cos E cos WN cosN sin .
cosMN
Figure 5.4 shows an example of the constituent ellipses for M2, S2, N2, K1, and O1 for a current
station in the Strait of Georgia. One will notice that the eccentricities (and widths) of the constituent
ellipses are different. This may be the result of the effect of the nonlinear lateral inertial terms (see
Sections 2.3.6 and 7.4.2) being different for the different frequencies. Even more dramatic
differences in ellipse width are seen in Figure 7.3 (which shows additional M2 ellipses at the eastern
end of the Strait of Juan de Fuca, Haro and Rosario Straits, and the southern end of the Strait of
Georgia). The differences in ellipse widths from location to location are also due to basic continuity
effects; for example, the ellipse in a narrow entrance will be faster and narrower than the ellipse in
a more open water area.
One will also notice that the major axis directions for these five ellipses are not exactly the same.
This is at first somewhat surprising since one would expect that the bathymetry and shorelines would
steer all components of the tidal current in the same direction. One might begin looking for some
frequency-dependent tidal hydrodynamic effect that could cause the ellipse directions to vary, but
it is more likely to be a product of the harmonic analysis itself, where for the smaller constituents
the smaller of the two orthogonal components may be more affected by noise, resulting in a value
that might shift the direction of its ellipse.
177
Figure 5.4. Tidal constituent ellipses for M2, S2, N2, K1, and O1, for current station 70 (at
15 feet below the surface) in the Strait of Georgia, based on a 29-day harmonic analysis of
a data time series beginning February 4, 1973. (From Parker, 1977.)
178
As was mentioned in Section 3.4.3, one can also treat the two orthogonal components together
using complex algebra (Pawlowicz, et al, 2002). In this case there is one prediction equation instead
of two, since the current vector is represented by the complex number u + iv (where u and v are
the orthogonal components of the tidal current, but the complex algebra allows them to be treated
together). The complex prediction equation (for N tidal constituents) looks like
x(t) = x o + 3 a k ei k t + a k e i k t
k =1 to N
where here x(t) has been used to indicate velocity, and k to indicate the frequency of each
constituent, ak and ak to represent the complex amplitudes (ak and ak are complex conjugates if the
time series is real). The length of the semimajor axis of the ellipse is *ak*+ *ak* and the length of
the semiminor axis is *ak* *ak*. (See Pawlowicz, et al, 2002, for more details.)
5.3 Considerations In Carrying Out the Analysis
5.3.1 Which Tidal Current Constituents Should Be Solved For
Almost all the considerations in harmonically analyzing water level data discussed in Section
4.1.1 apply to harmonically analyzing current data. Thus, one should read Section 4.1.1 before
proceeding, and should use either Table 3.2 or Table A.2 to determine which tidal constituents to
solve for based on the length of the available data time series. Since current data time series are
generally shorter than water level time series (there being typically only a few current stations with
a year of data or longer), the decisions one will usually be making with regard to current data are
whether one can include key constituents like N2 or P1 in the analysis, and if not, what will be the
best method to infer them.
Since in shallow-water areas, current data are more likely to have larger compound tides and
overtides than water level data, extra shallow-water tidal constituents may be needed (some of which
can be solved for with the lengths of data time series available, and some of which cannot).
However, current data are often noisier than water level data, so these additional tidal constituents
will have to be clearly above the nontidal (noise) continuum to be accurately calculated. If one has
enough data available to do more than one analysis (perhaps the current station was occupied at two
different times, but both records were short) then consistency between the two calculations of the
amplitude and epoch for a small constituent will be important in deciding whether the result is
accurate. As mentioned before, some of the compound shallow-water constituents will have the
same frequency as an astronomically caused tidal constituents. For example, in a shallow waterway
there is a very good chance that 2MN2 in the tidal current will be a good deal larger than L2, and thus
2MN2 should be used in the harmonic analysis rather than L2.
5.3.2 Selection of Orthogonal Axes
One can choose any pair of orthogonal axes for the analysis, to convert the speed and direction
time series into two component time series. Often north-east components have been chosen, really
for no particular reason. More often major-minor components are chosen, with the direction of the
major axis being the mean direction of maximum flood (which often coincides or is very close to
the direction of maximum M2 flood). Accurate tidal current predictions can be made using any
choice of orthogonal components system (north-east works as well as major-minor), but for
179
interpreting the results one usually feels more comfortable with major-minor. And there may be
times when one may only wish to use the major axis component, because either the tidal current is
a reversing current, or it is rotary but there is so little energy in the minor component that one may
decide to leave out for convenience.
The harmonic analysis itself calculates the mean direction of maximum flood, and the directions
of all the tidal constituent flood directions. It also produces ellipses for all the tidal current
constituents that came out of the harmonic analysis (by combining the harmonic constants for the
two components). Unless one knows ahead of time what the mean direction of maximum flood is,
one may end up running the harmonic analysis a second time, using the mean maximum flood
direction that came out of the first harmonic analysis as the major axis. However, one does not
really need to run the harmonic analysis a second time, if one includes some trigonometric formulas
at the end of the harmonic analysis program, which will automatically convert all the harmonic
constant pairs (for the orthogonal components used in the analysis) into equivalent pairs for a major-
minor orthogonal system using the calculated mean maximum flood direction as the major axis.
Formulas for conversion from north-east components to major-minor components were given at the
end of Section 5.2.
Sometimes one does wish to include a relatively small minor component when making the
prediction, but one may be worried about the quality of the harmonic constants extracted by the
harmonic analysis from the minor component time series because the signal-to-noise can be quite
low (whereas in the major axis there is very high signal-to-noise). In such cases one could select
orthogonal axes so that the strong tidal signal is evenly split between the two components, and then
use formulas like those at the end of Section 5.2 to convert them into the preferred major-minor
components.
5.3.3 Types of Instrument Errors Possible and Their Effects On Analysis Results
As was mentioned in Section 4.1.3, when data points said to be bad or unreal, one is
essentially talking about instrument error, i.e., values that do not represent the actual current speed
and direction at a particular location at the time of measurement because of some error related to the
measurement process itself. In Section 5.1 some types of current sensors were mentioned along with
the problems they could present (and this will be discussed more below), which included adding
instrument noise to the data record. There is also natural noise in a current data record due to
various nontidal phenomena (especially wind waves), but being natural they are real and are not
considered errors (but one may have to filter out that noise prior to our analysis).
There are many methods for evaluating the quality of a data time series and for editing out data
points that are determined to be bad or unreal, and we will not discuss any particular methods
here. Usually one tries to find and then remove (or correct) data points that are obvious outliers,
which is fairly simple when there is a clear tidal signal in the data, but can be more difficult if the
current data is noisy. One also tries to discover other types of variations in the data that may not be
of natural origin, and thus must have come from the instrument, which might affect the analysis
results.
It is important to acquire some feeling for the potential sources of errors and how different
types of errors might affect the results of a harmonic analysis. The errors in a current data time
series may be due to electronic, mechanical, or many other causes. It may be helpful to first read
Section 4.1.3 for the general discussion on types of errors in water level gauges and the possible
effects they can have on the results of a water level harmonic analysis. The five error types
180
discussed in that section are repeated here. The various types of current measurement devices will
be looked at, along with the types of errors they can produce. With an eye toward their effect on
analysis results, these five types of error are:
(Type 1) random errors [which will tend to average out during a harmonic analysis];
(Type 2) sudden shifts, when from some point in time forward, the entire time series of measured
currents are all larger or smaller (or in a different direction) by some constant amount
than they were before that point in time, which would change the amplitudes of harmonic
constants as well as the measured mean flow (this is a different effect than a sudden shift
in tide gauges);
(Type 3) slowly changing errors, where an error might slowly grow over months or longer [which
would change the harmonic constants];
(Type 4) slowly changing time measurement, such as when a clock may be running slow [which
would essentially spread out time and reduce the frequencies of the tidal constituents, so
that when a harmonic analysis looks for energy at the correct tidal frequencies the result
will be smaller amplitudes and changed phase lags]; or
(Type 5) periodic errors, which might, for example, be due to meteorological effects on the
instrument (as opposed to meteorological effects on the water level itself) [which might
change particular tidal constituents].
First, it should be mentioned that many of the methods below for measuring currents involved
making those measurements from an anchored boat, or from an anchored buoy, which would swing
around when the tidal current changed from flood to ebb or from ebb to flood. Since, as was seen
in Section 2.3.6, current speeds and directions can vary significantly over geographic distance (due
to changing bathymetry, among other reasons), the measurements made during the flood phase were
likely made at a location some distance (depending on the anchor line length) from the location of
the measurements during the ebb phase. Thus error Type 5 was quite prevalent. In some cases two
or three anchor lines were used to try to minimize the movement of the boat or buoy.
The earliest method for measuring currents used by the Coast and Geodetic Survey on a regular
basis was the current pole and log line. (See Manual of Current Observations, U.S. C&GS, 1950,
for more details on the current pole and some other early instruments mentioned below.) This was
a truly manual method to measure currents, which involved putting a 15-foot-long current pole
(weighted at one end to keep it vertical, with the top about a foot above the waters surface) in the
water behind an anchored boat with an attached graduated log line that was pulled out as the current
moved the pole away from the boat. Using a stop watch to measure how much of the line was pulled
out over a specific interval of time, the speed of the current was calculated. A compass or some
other device such as a sextant or pelorus was used to measure the direction that the pole moved.
[This method of measuring currents is said to go back to the 16th century, when the log line was
knotted at regular intervals so that the sailor could easily tell how much line was let out in a set
amount of time by counting how many knots passed through his hands, thus leading to the term
knots as a unit of speed (which came to mean one nautical mile per hour). On the early Coast
Survey log lines, the unit graduations were marked by pieces of cotton string in which knots were
tied, the number of knots in any string indicating the speed of the current expressed in knots.]
This method using a current pole and log line did not really measure the current at one specific
location in space, since the current pole moved some distance in space during the measurement
period. It was really a measurement of the average speed over the distance that the pole traveled,
over a depth comparable to the length of the pole, over the time period of the measurement. Since
the current speed and direction can vary significantly over space (both with depth and with
181
geographic distance), the current pole often moved through regions with different current speeds.
Using the current pole it was hard enough to accumulate a current data time series long enough for
a harmonic analysis, but it was even more difficult to have each data point represent the current for
different times over the exact same location. An error analysis of the results of a harmonic analysis
of current data measured with a current pole was almost not worth the effort (and all the types of
error were possible, most of it human error in a situation that had a low chance of repeatability).
The next improvement in current measurement was to hang over the side of an anchored boat
an instrument that could measure the current speed at one location, lowering it to a specific depth.
The first of these used regularly by the Coast and Geodetic Survey was the Price Current Meter,
which had a wheel made of conical shaped cups angled horizontally around the wheel in such a way
that the water flow would make the wheel spin. The current speed was calculated by counting the
number of rotations of the wheel that occurred during a specific time period. Originally the counting
of rotations was done manually by an observer on the deck of the ship with headphones connected
electronically to the instrument. A later version was connected to an automatic recording device.
Direction was not measured, but was estimated by the observer on the boat. Another current meter
used by C&GS that was lowered over the side of a boat and could measure direction as well as
speed at one location was the Ekman current meter. The Ekman current meter had a screw propeller
to measure speed, and a compass-ball receptacle to measure direction, by catching small bronze balls
in one or more of 36 ten-degree receptacles around a compass. Although a big improvement over
the current pole, this was still a manually intensive way to measure currents, with many chances for
human error or human inconsistency.
The next advance in current measurement at C&GS was the Roberts Radio Current Meter, which
measured speed and direction, was fully automated, and could be hung from an unmanned buoy.
A rotating impeller measured the current speed. A large vane pointed the current meter into the
direction of flow, with a magnetic compass device that determined what that direction was relative
to magnetic north. The current meter was connected electronically to a radio transmitter on the buoy
which sent the data to a nearby receiving station where time was assigned to the data with a
chronometer and where a chronograph tape was produced with the times and the data. That tape had
to be converted manually into a tabulation of pairs of current speeds and directions.
With later more modern versions of current meters (such as the Geodyne photographic current
meter, the TICUS current meter, the Plessy Current meter, and the Aandera Current meter), current
meter systems became more automated, with various data storage devices (e.g., photographic film
and magnetic tape) to save the data internally for a month or longer (depending on the sampling
rate). In some cases the method of measurement could cause serious errors in the data, especially
when there was significant wave action during the deployment. The Geodyne and TICUS current
meters measured current speed with a small S-shaped savonious rotor (which could feel water flow
from all directions and so could easily be turned by wave action) and current direction with a small
vane (which was easily flipped around, again by wave action). This had a great effect on the
harmonic make up of the measured tidal currents, and could often totally mask slack waters in a tidal
current. Vector averaging current meters overcame this problem, and deploying these in situ current
meters from a submerged tautline buoy helped keep the current meters out of the surface wave zone.
Current measurement based on remote sensing is now the preferred method, with the two most used
systems being the acoustic Doppler profiling current meter (for measuring currents along a depth
profile at a specific location, or a horizontal profile at one depth from a pier) and the high frequency
radar system (for measuring surface currents over a large area).
182
The following discussion mentions how the five types of errors might be caused by the various
current measurement systems. Random errors (Type 1) can be caused by all current sensors. Such
errors were probably more frequent and/or larger for the early manual current measurement
techniques (such as the current pole and some of the earliest current meters). As long as those errors
are random, they tend to be averaged out in a harmonic analysis, but if the human errors become in
any way consistent, they could add some type of bias to the measurements, which then would affect
the harmonic constants from the analysis.
Sudden shifts or changes in the data (Type 2) were probably most prevalent with mechanical
current meters due to the rotor or propeller being clogged by debris in the water. If the clogging
totally stopped the rotor or propeller from turning, the current speeds would suddenly go to zero.
This was actually preferable to a partial clogging that caused slower speeds to be registered, because
the zero-speed data would be recognized and not used, but the slower speeds (especially if it still
showed a tidal cycle) might not be noticed, and the data would be used. In the latter case, the
amplitudes of all the harmonic constants would be reduced. Such (relatively) sudden changes are
still possible with newer instruments, for example, if a great deal of sand suddenly was moved on
top of a bottom-mounted acoustic doppler current profiler (ADCP). However, for bottom-mounted
ADCPs and for moored in situ current sensors, the most common cause of a sudden change in the
data is when the instrument is suddenly moved by strong currents during a storm (or perhaps
inadvertently by a fishing boat), since the current can change dramatically over horizontal distances.
Slowly changing errors (Type 3) for all current measurement systems (except land-based remote
systems) are usually caused by biological fouling, namely, the slow growth of an organism on the
sensor in a way that slowly changes the current measurements, usually slowly reducing the measured
speeds. For the early current meters fouling would slowly inhibit rotors and propellers or even the
movement of the smaller vanes. Such changes would affect all the calculated harmonic constants,
mainly reducing their amplitudes. If the change was slow enough (and did not last long enough to
bring the current meter to a halt), it might go completely unnoticed, and if noticed might even be
assumed to be due to the changing hydrodynamics rather than fouling. Luckily, fouling is less of
a problem with modern non-mechanical current sensors.
Slowly changing time measurement (Type 4) was a prevalent problem with the early internal
recording current meters. In fact, all such current meters were assumed to have clock drift, and
methods were developed to determine how much the clock drifted during a current meter
deployment. The usual way was to carefully write down the exact time the current meter was turned
on and the exact time the current meter went below the water surface, and likewise the times when
the current meter first left the water and when it was turned off. These points in time could be
spotted in the current data record (recognizing the water entry and exit was helped considerably if
the current meter also had a temperature sensor), and then one could determine how many data
points should have been recorded versus how many actually were recorded. One had to assume that
the clock drift was constant throughout the data record, but one could then adjust the sampling
interval accordingly. Of course, there was always the possibility that individual data points were
for some reason not measured or not recorded when they should have been (due to electrical
problems). To recognize those occurrences, the instrument makers began adding mechanisms to
either mark the hours (since sampling was usually several data points per hour) or to record the time
itself for each data point. All this was necessary because effects on the automated time keeping
could adversely affect the harmonic constants from an analysis. A slow clock would lead to data
samples being taken at time intervals slightly longer than the intended sampling interval. If this was
not recognized and corrected (and the time was erroneously assigned to the data points as though
183
the clock were correct), then the tidal energy in this data time series (with erroneously assigned
times) would appear at slightly higher frequencies than the tidal frequencies. (For a fast clock, the
opposite would happen.) Analyzing such data for the frequencies where one expects to find tidal
energy could then lead to erroneous harmonic constants (depending on how slow or fast the clock
was).
There are a number of ways that periodic errors (Type 5) can be caused in various current
measurement systems. This includes the previously mentioned current sensors hung from an
anchored surface buoy (or boat) that are swung around by the change in current direction from flood
to ebb and back to flood. Since current speeds can change dramatically over horizontal distance,
especially when the depths are different, being moved horizontally by the currents will put the
current sensor in different current regimes. This can be most dramatic when the sensor is in a deep
channel (with faster currents) during one phase of the tide, but is over the shallows (with slower
current) during the other phase of the tide. This is a periodic effect that will have the most effect
on the harmonic constants of the dominant constituent (usually M2) and other constituents in the
semidiurnal band. The inherent asymmetry of this effect could increase the amplitude of M4. Since
the current regimes over the various depths (that the current meter will pass over as the buoy swings
around) may not change smoothly, there are likely to be other effects on the higher harmonic
constituents as well. Such changes would be difficult to determine in the data. To avoid this
problem methods were often used to try to limit the movement of anchored buoys, such as by using
two or three anchor lines.
Another periodic effect influencing sensors hung from both buoys and bottom-mounted tautline
moorings, is the effect of the drag on the mooring line (or the line from the surface buoy) due to the
currents. The drag of the currents on a tautline mooring will pull the current sensor deeper as the
current speeds increase, and thus the measured current speeds will actually be for different depths,
which likely will be different than if the sensor had not been pulled deeper, since current speeds
often change with depth (due to frictional effects, as well as possibly baroclinic effects; see Section
2.3.6, as well as Figure 7.27). For current sensors hung from surface buoys, the drag on the line due
to the currents will pull them shallower. In either case, the effect is periodic, the current sensor
moving deeper (for the tautline mooring) or shallower (for the surface buoy) as one approaches
maximum flood and again as one approaches maximum ebb. This will again have the greatest effect
on the harmonic constants of the dominant constituent (usually M2) and other constituents in the
semidiurnal band. The inherent symmetry of this effect could also affect M6. Other higher order
constituents might also be affected. Such effects are difficult to determine in a current record, so
methods were tried to limit the drag on the line by currents (such as using Kevlar lines to reduce
drag).
5.3.4 Assessing the Potential Effects of Nontidal Influences
As was seen in Section 2.3.3, in shallow-water areas both the tide and tidal current can be
modified by the nonlinear interaction of nontidal phenomena such as river discharge and storm
surge. See Section 4.1.4 for a discussion of the specific effects on the tide; similar effects can occur
on the tidal current. In addition, however, as seen in Section 2.3.6 there are effects on the tidal
current that one will not see in the tide. There is also periodic nontidal energy in the currents that
can occur at or near tidal frequencies, which can affect the results of harmonic analysis. Changing
water density due to changing salinity in estuaries or to changing water temperature offshore can
allow internal tide waves and other baroclinic effects, which primarily affect tidal currents.
184
When one is harmonically analyzing a current data time series one must therefore be aware of
the potential effects of nontidal phenomena on the harmonic constants that come out of the analysis.
If one needs harmonic constants from which accurate predictions can be made for most of the year,
then one might consider leaving out sections of the time series that include strong nontidal events
and using the gap option of the least squares harmonic analysis program. If one has a short time
series of current data that includes a major nontidal event, one will probably not be able to use the
harmonic constants that come out of the analysis for tide predictions for normal time periods, i.e.,
for time periods without those strong nontidal phenomena. In fact, one can probably not even use
those harmonic constants for the time period that included the nontidal event, because the effect of
the nontidal event (which probably only lasted a couple of days) was spread out over the length of
the analysis. So even the results of a 15-day harmonic analysis will not be useful. In such cases,
one can try using the continuous wavelet transform method (see Section 3.5.5).
To discover those periods in an observed current data time series when a nontidal phenomenon
may have a strong influence, one can first tidally filter (see Section 3.9) the current data time series.
Although the filtering process may reduce the amplitude of such an effect, one should still see
periods when the nontidal currents are increased and then decreased. These may be sections of the
time series that one will ultimately decide to leave out of the harmonic analysis, but initially one
should analyze the entire time series, produce a predicted time series for the entire period of the
observations, and then subtract the predicted series from the observed series to produce a residual
series. When examining the data, predicted, and residual time series, one will probably have to look
at plots of orthogonal components, as well as plots of speed and direction. At those same time
periods where one saw changes in nontidal currents in the filtered time series, one is likely to see
tidal oscillations appear and then disappear. One might see transient tidal oscillations at time
periods without an obvious change in nontidal currents in the filtered time series (which would
imply the effects of either changing water density or of changing water level, the latter easily
checked in a water level time series from a nearby location).
When these transient tidal current oscillations occur, it will often be due to the effect of the
nonlinear interaction between the tide or tidal current with a transient nontidal phenomenon, such
as a large river discharge or storm surge. During these periods of interaction, the nontidal
phenomenon has temporarily changed the hydrodynamics of the situation and thus changed the
amplitude and/or phase lag of one or more tidal current constituents. Thus, by subtracting the
predictions based on the average tidal current constituents (obtained from the analysis of the entire
time series) one leaves some tidal energy for that particular time period. The transient tidal current
oscillation seen at time periods without an obvious change in nontidal currents (in the filtered time
series) may be due to the nonlinear effects of a change in water density. Also low-frequency
changes in nontidal water level are not accompanied by large changes in nontidal currents, but can
have significant effects on both the tide and tidal current. (See Section 4.1.4 and 4.3.4.)
When one sees nontidal events in the filtered time series and/or transient tidal oscillations in the
residual times series (or in the residual time series from a nearby water level station), one should
look at wind, river, salinity or water temperature data records for the potential nontidal influences
that may have caused the temporarily different tidal current constituents. If such incidents occur
fairly frequently, one may wish to add a footnote in the Tidal Current Table about when such
incidents occur. For such incidents current prediction with a real-time driven forecast numerical
circulation model may be the best route to take, because it can include not only the nontidal
phenomena but also the effects of their nonlinear interaction with the tide and tidal current.
185
There can also be nontidal phenomena with (temporary) periodic behavior at frequencies near
tidal frequencies. A land breeze-sea breeze (see Section 2.3.6) changes its wind speed and direction
over a 24-hour period, and this can produce a wind driven current with the same periodicity. In a
tidal current harmonic analysis, this energy will show up in the S1 constituent, but because the
changes are asymmetric there can also be additional energy put into S2 and maybe even S4 . To
notice this effect on S1, one would need a year-long time series to separate this meteorological S1
from K1 . Since in many locations the land breeze-sea breeze usually only occurs in the summer, the
S1 value would be very different in the summer than in the winter (and one might see a noticeable
change in S2 as well). One should remember this effect when one is analyzing current data obtained
during the summer, because it may have a noticeable effect on the K1 value that comes out of a
harmonic analysis.
For current measurements made at latitudes near 30oN or 30oS, one may have to also consider
the possibility of inertial currents (see Section 2.3.6) affecting the measurement of diurnal tidal
currents. Inertial currents are nontidal but still periodic. At latitude 30oN inertial currents will have
their energy at exactly the K1 tidal frequency, and at 27.6oN the inertial energy will be at exactly the
frequency of O1. Although generally transient and usually lasting only a few days, inertial currents
can last long enough to affect a tidal current analysis. Inertial currents with frequencies in the
semidiurnal band will occur near 75oN or 75oS.
There can also be a strong seasonal effect on nontidal currents. For example, stratification
during the summer allows internal tide waves to occur which can greatly affect the tidal currents,
but such internal waves will not occur during the winter months because of the mixing of the upper
water column due to the higher winds from more frequent storms. Also, the sea-breeze-land-breeze
effects just mentioned occur most often in the summer, and so one might see some effect on the Sa
and Ssa constituents.
5.3.5 Whether to Include the Calculated Mean Current In a Tidal Current Prediction
There is one other aspect of nontidal effects in a current series that one must deal with, but from
a different perspective. The harmonic analysis will also calculate a mean current for the current time
series that was analyzed. That value is the mean current for only that particular time period. That
particular mean current could have been caused by one or more different possible phenomena, for
example, by the winds or river flow. It could be the gravitational current resulting from the salinity
gradient going up an estuary. Or it could be a tidally induced residual current produced by the
nonlinear lateral inertial terms in the momentum equation, usually due to a bend in the waterway
or a point of land sticking out into the waterway, which creates residual eddies due to a sheltering
effect (see Sections 2.3.6 and 7.6.7).
All of these mean currents, with the exception of the tidally induced mean current, are transient
and the exact value that comes out of the harmonic analysis is probably only good for the particular
time period of the analyzed data. Because of this, in most cases the mean current that comes out of
the harmonic analysis should not be included in tidal current predictions for future time periods, or
for use in Tidal Current Tables. There are two exceptions.
The first exception is if one can determine in some way that the mean current is likely to be
permanent because it was tidally induced. A look at the geography around the location of the
current station will give one the first clue. If, for example, the current station is in the center of a
long narrow waterway, it is highly unlikely that the lateral inertial mechanism could have come into
play. On the other hand, one might have a situation like in Figure 2.35, in which case it would make
186
sense to include the mean current in all tidal current predictions. [Also, if one used orthogonal
components to analyze the current data at a location as depicted in Figure 5.1 (with flood and ebb
directions that are not 180o apart), one must include the apparent mean current, for reasons explained
in Section 5.1. (However, for that type of location, an alternative approach is to treat the reversing
current as a scalar quantity, with positive flood speeds and negative ebb speeds, so that only one set
of harmonic constants is required, in which case there would be no apparent mean current.)
The second exception is not as strong a case. If one has done a harmonic analysis on a long
current data series, and one has reason to believe that they have captured a mean current that exists
most of the time (e.g., a typical river flow), then one may decide to include it in the tidal current
prediction. Strictly speaking, it is not part of the tidal current, but in some cases it might be included
in a prediction for a Tidal Current Table if one is trying to predict as much of the current as possible
(for the mariner). However, if it is included, there will be many time periods when the nontidal
current will be faster or slower than that mean current.
Actually, there is a third exception, but for a very different circumstance. When one is
evaluating the quality of the harmonic constants calculated by comparing the predicted tidal current
series with the original observed current data series, one must include the calculated mean current
in the prediction (or one must subtract the mean current from the observed series).
5.3.6 Use of Node Factors and Equilibrium Arguments (or of Satellite Constituents)
In Section 2.2.4 it was seen that there are some very slow modulations of the tide-producing
forces with periods of many years, the two most important being the 18.6-year variation in the angle
between the plane of the moons orbit and the plane of the Earths equator, and the 8.85-year
variation in the lunar distance from the Earth due to the rotation of the longitude of the lunar perigee.
To include these effects in a complete harmonic analysis one can either add dozens of additional
satellite tidal constituents, or use the concept of a node factor which represents the modulation of
the tidal constituents over these 18.6- and 8.85-year cycles. Consideration of these two approaches
for the harmonic analysis of water level data was discussed in Section 4.1.5. That discussion also
applies to the harmonic analysis of tidal current data, and so the reader should read that section.
5.4 Methods for Analyzing Short Time Series
In Section 4.2 various methods for analyzing a short time series of water level data were
discussed. That discussion applies equally well to currents and in fact is probably more relevant for
currents, since typically current data time series have been much shorter than water level times
series. As was seen in Section 3.3 the number of tidal constituents that one can accurately determine
using harmonic analysis depends on the length of the data time series. One can look at either Table
3.2 or Table A.2 to determine which constituents can be included in a harmonic analysis based on
the length of the time series. But there is some leeway in following the guidance in those tables,
both due to our use of the least squares harmonic analysis technique, and for high tidal signal-to-low
nontidal noise situations, and one should read Section 3.3. When the length of the time series limits
the number of tidal constituents one can solve for, there are methods to try to improve the analysis
results both to obtain values for constituents that could not be included in the original harmonic
analysis (called inference) and to improve the value of constituents that were in the analysis but were
adversely affected by the constituents that were left out of the analysis (call elimination). Several
of these methods for inferring constituents that could not be solved for with the available time series,
187
and for eliminating the adverse effects of these missing constituents from those constituents that
were solved for, were presented in Sections 4.2.1 through Sections 4.2.5. These methods were
applied to water level data, but they can be applied equally well to current data, and one should read
those sections.
5.5 Assessing the Quality of the Predicted Tidal Current Series
5.5.1 Introduction
To assess the accuracy of harmonic constants calculated using a harmonic analysis, one usually
compares a predicted tidal current time series (made using those harmonic constants) with the
observed current data time series from which the harmonic constants were extracted. Even though
there will be other nontidal signals in the observed current data time series, there are ways to
evaluate the quality of the predicted tidal current time series and thus of the accuracy of the
harmonic constants. In Section 4.3 several methods were looked at for assessing the accuracy of
the harmonic constants from water level data using the predicted tide time series. That section
should be read before reading the sections below.
The same approaches can be used for tidal currents, although there may be some differences due
to the vector nature of the tidal current and the need to harmonically analyze two orthogonal
component series. The other difference relates to the mean current. The mean current can be large,
but since it is usually nontidal and thus unpredictable, one should usually not include it in a tidal
current prediction for a Tidal Current Table. (As was discussed in Section 5.3.5, the mean current
could be permanent if it is tidally induced by the nonlinear inertial mechanism described in Section
2.3.6.) However, when one does a quality assessment of the harmonic constants by comparing the
predicted series to the observed series, the mean current must be included in the predicted series (or
subtracted from the observed series).
Also, one usually had a lot less current data than one has water level data, which limits the kind
of quality assessment that can be done. Sometimes one will not have a very long current time series,
perhaps a month-long record, but one may have two such one-month records. It is important to
analyze both records, for comparison, especially of the smallest constituents ones that might be
rejected for being very small unless they show up in both analyses with approximately the same
phase lags, which would indicate that they do represent the tidal constituent (and not just some
transient periodic nontidal energy).
For tidal currents (as for the tides) the three main approaches for assessing the accuracy of the
calculated harmonic using the predicted time series are:
(1) directly compare the speeds, directions, and times of maximum floods, maximum ebbs, and
slack waters (or minimum flows) in the predicted series versus in an observed time series
(2) carry out and examine a spectral analysis of the residual time series; and
(3) examine the residual time series itself for periods with transient tidal current oscillations.
5.5.2 Comparison of Maximum Floods and Ebbs and Slacks In the Predicted Time Series
Versus In the Observed Time Series
A frequently used method for assessing the quality of a tidal current prediction time series is to
compare the speeds, directions, and times of corresponding maximum floods, maximum ebbs, and
slack waters (or minimums) in the predicted tidal current series and the observed current series.
188
Here it is important to use the same method for calculating these key points on a tidal current curve
for both time series (see Section 3.9.2). Usually one looks at the time differences for each pair of
corresponding maximum floods, maximum ebbs, slack waters (or minimums) and the speed ratios
for the maximums. If the tidal current predictions were done using only major axis harmonic
constituents, there will be no variation in the predicted directions. If the tidal current predictions
were done using both major and minor constituents (or north and east constituents) then directions
will be predicted, but they usually do not vary enough to enter into a quality assessment.
As with the tide, this method is typically used to check the quality of the daily predictions for
reference stations in Table 1 of a Tidal Current Table. However, unlike the tide, there may not be
any current data beyond what was analyzed to calculate the harmonic constants, even for a reference
station.
As already mentioned, the observed current series includes nontidal effects on the maximum
floods and ebb and the slack waters, so there will be differences between the predicted tidal current
series and the observed current series due to those nontidal effects. These should average out when
one is calculating mean ratios and difference for these parameters. Even with these nontidally
caused differences it is useful to look at how the individual time difference and current speed ratio
values change throughout each month and from month to month. For example, one might see a
periodicity in the current speed ratios or time differences that points to a problem with a specific
harmonic constant. Other longer periodicities or patterns in the current speed ratios or time
differences might be due to nontidal seasonal effects.
5.5.3 Spectral Analysis of the Residual Time Series
As was seen in Section 4.3.3 one common way to check the quality of the calculated harmonic
constants (and to provide some type of error estimation) is to make a tidal current prediction for the
time period of the data analyzed, subtract the predicted time series from the original observed times
to produce a residual time series, and then to see how much energy is left in the residual time series
at the tidal frequencies using a spectral analysis. The details and comments presented in Section
4.3.3 for tides also apply to tidal currents, and that section should be read.
For tidal currents, since they are vectors and since the speed and direction time series are
typically transformed into two orthogonal component time series (such as major-minor or north-east
component series), there are several choices on which residual series to look at and what type of
spectrum to produce. For reversing tidal currents, or even tidal currents where the flood-ebb axis
is much longer than the perpendicular flow axis near minimums (i.e., for narrow elliptical rotation)
one may wish to look at the speed residual time series (with ebb flows given negative values) or the
major-component residual time series only. But for other cases one will look at both the major and
minor residual time series. One can also look at the residual rotary spectra (see Section 3.10.3), i.e.,
the clockwise component residual time series and the counterclockwise residual time series. Each
of these types of residual spectra will provide not only information about how much energy is left
at the various tidal frequencies, but might also provide insights (some more than others depending
on the situation and location) into possible hydrodynamic reasons why the energy is there.
189
5.5.4 Examination of the Residual Time Series For Periods With Transient Tidal Current
Oscillations
In addition to doing the spectral analysis, one should also examine the residual current time
series itself and look for time periods where oscillations at a tidal period appear and then disappear.
(See Sections 4.1.4, 4.3.4, and 5.3.4.) When these occur, it will usually be due to the effect of the
nonlinear interaction between the tide and some transient nontidal phenomenon, such as a large river
discharge or storm surge, or to a change in water density that allowed a baroclinic effect to occur
such as an internal tide. This latter effect can be important for tidal currents. During these periods
of interaction, the nontidal phenomenon has temporarily changed the hydrodynamics of the situation
and thus temporarily changed the amplitude and/or phase lag of one or more tidal constituents, and
so by subtracting the predictions based on the average tidal constituents (from the analysis of the
entire time series) one leaves some tidal energy for that particular time period. These various events
with temporary tidal energy will not show up in the spectral analysis because they will have
different phase relationships and cancel each other out in the overall average values determined by
analysis.
When such incidents occur, one should look at the nontidal influences (such as changes in wind,
river discharge, water density, etc.) to try to determine the cause of the temporarily different tidal
constituents. If such incidents occur frequently, one may wish to add a footnote in the Tidal Current
Table describing when such incidents occur. However, in some cases the nontidal effect (e.g. river
flow) that caused the change in tidal current harmonic constants will be larger than its nonlinear
influence on the tidal current, and so such changes to the tidal current may be less important. (If such
incidents occur frequently, then tidal current prediction with a real-time driven forecast numerical
circulation model may be the best route to take, because it can include not only the nontidal
phenomena but also the effects of their nonlinear interaction with the tide and tidal current. See
Chapter 8.)
5.5.5 Looking For Changes in Bathymetry or Shoreline
Changes in the depth of a waterway can significantly affect the tidal current. It was seen in
Section 2.3.6 how sensitive the currents are to depths, because of friction and other effects.
Changes in depth over time at a location may be rapid due to a hurricane or extra-tropical storm,
relatively fast due to a dredging project in a navigation channel, or fairly slow due to year-to-year
erosion and movement of sediment in the waterway. This effect is most important in shallow
waterways where the same amount of depth change will represent a larger percentage of the total
depth. Since sediment is almost always on the move, bathymetry is always changing. For very deep
bays and estuaries the depth changes will not have much effect on the tidal currents, but for shallow
bays and estuaries the same depth changes can significantly affect the tidal currents, so much so that
the predictions in the Tidal Current Tables for that bay may no longer be accurate enough, since they
are based on harmonic constants calculated from data measured before the depth changed.
When a dredging project to deepen a navigation channel in a small waterway has been
completed, it is very likely that the tidal currents will be different (probably faster) in the channel
than they were before the dredging project. It would therefore make sense to try to deploy a current
sensor in that channel to see how much of a change has occurred. If the change is determined to be
significant, then a new current survey is required, and the Tidal Current Tables should be footnoted
with a warning until new predictions can be included. Similarly, when the shoreline is changed
190
significantly by filling in water areas to create land, or to create an artificial island, there also will
likely be a change in the tidal currents. Storms and longer-term erosional effects due to the currents
themselves (that move sediment around, creating some shoals and eroding others) can significantly
change the tidal current regime in a shallow bay or estuary. For some locations, such as the shallow
salt-marsh estuaries of South Carolina and Georgia, the changes can be very difficult (and
expensive) to try to keep up to date in the Tidal Current Tables.
5.6 Producing a Reference Station for Table 1 of a Tidal Current Table
Very accurate tidal current harmonic constants are required for a reference station that appears
in Table 1 of the Tidal Current Tables, since accurate tidal current predictions using these harmonic
constants will be made for many years in the future, and many subordinate stations in Table 2 will
be referenced to that station. The current data time series for such a station should be as long as
possible in order to be able to use as many tidal constituents as possible in the harmonic analysis.
Current data time series have typically been much shorter than water level time series, and many of
the reference stations in Table 1 of the Tidal Current Tables are less than (some much less than) a
year.
Selecting the location for a tidal current reference station is often trickier than for a tide station.
One should try to select, if possible, a dynamically simple location, such as in the center of the
entrance to the waterway, or in the straight section of a channel, and away from channel bends,
junctures of channels, and points sticking out into the waterway. It is preferable, if possible, that
this location should also be important for navigational purposes, as many mariners will rely mainly
on the reference station and may not want to go through the adjustment process for predicting a
subordinate station using Table 2 time differences and speed ratios .
Typically the standard 37 constituents (listed in Table 3.2) have been used in CO-OPS (and its
predecessor organizations). However, since nonlinear shallow-water effects are usually larger in
tidal currents than in tides, and since nonlinear lateral inertial effects also move energy into the
higher harmonics, there is probably room for adding more shallow-water constituents into the
harmonic analysis of current data (see Table A.2). One should use the methods mentioned in
Section 5.5 for assessing the quality of the predicted tidal current series produced using 37
constituents to decide whether more constituents are needed (and whether a longer time series needs
to be acquired from a new deployment).
If the location for the tidal current reference station happens to have a fairly rotary current, that
is, its overall current ellipse is wide enough that the minimum flows (perpendicular to the flood-ebb
axis) are important enough to be included in the daily predictions, then one must use a set of pairs
of harmonic constants from a major-minor harmonic analysis (or a north-east harmonic analysis).
If there are analysis results from several current sensors at that location, each for a different depth,
one needs to assess each set of harmonic constants to decide which one is best to serve as the
reference station. If the current data came from an ADCP then there will be many bin depths from
which to choose. The best bin depth will probably be the one with the largest current speeds but far
enough below the water surface so that the current data will not be significantly affected by surface
wind waves. The other bin depths can be included in Table 2 and treated like subordinate stations.
191
5.7 Summary Overview: Steps In Harmonically Analyzing Current Data
In this section the steps to take in harmonically analyzing a current data time series will be
summarized. Here it is assumed that one is using a least squares harmonic analysis program. In
many cases, if the data are available, more than one harmonic analysis will need to be carried out
on the data, with adjustments being made after each analysis if needed (such as adding additional
constituents after assessing how much tidal energy is left in a residual series). Some steps may not
be necessary for every analysis. Sections where more information can be found are in brackets [ ]
at the end of each step.
These steps, in bullet form are:
(1) when first using a harmonic analysis program it may be a good idea to test it on a predicted
tidal current time series based on known harmonic constants, to see if the program can
reproduce exactly the harmonic constants that were used to make the predicted tidal current
time series, and if not, try to determine where energy is lost (or possibly gained); this might
have to be done on tide data in order to have enough data to do an adequate test; [Section
4.3.1]
(2) determine how much accuracy one needs in the harmonic constants (and in predictions to
be made with those constants) for your particular application; that will affect how many of
the steps below one will choose to use;
(3) assess the data available at the current station; if more than one time series is available, use
the longest most recent data; (especially for small shallow waterways) check for indications
of dramatic changes in bathymetry (e.g., due to storms or to dredging); if such changes have
occurred, select data (if possible) after those changes occurred. [Section 5.5.5]
(4) determine if there are gaps in the current data time series; still use all the data but make sure
to select the option in the least squares program that handles gaps (or random time data)
since some least squares programs distinguish these two cases from a continuous data record,
because the latter can be analyzed more quickly. [Section 3.4.5]
(5) determine which astronomical tidal constituents will be included in the program, initially
based on the length of the current data time series; use Table 3.2 or Table A.2 to make this
determination, selecting all the constituents with synodic periods less than or almost equal
to the length of the data time series; in subsequent runs of the program more constituents
may be added, especially if the length of the data time series is fairly close to the next
highest synodic period. [Section 5.3.1]
(6) determine whether an orthogonal component harmonic analysis should be done (major-
minor or a north-east analysis), or whether the tidal current is reversing enough to treat with
only one (major) component;
(a) if the waterway is very narrow and the tidal current is essentially reversing, so that a
single component analysis is to be done, make sure that the flood and ebb directions are 180o
apart; if they are not (because there is a bend in the narrow waterway) one should use only
the current speed time series, making all ebb speeds negative;
(b) if the waterway is wide enough for there to be some rotation in the current directions,
then an orthogonal component harmonic analysis will be used; one may initially run a
major-minor analysis or a north-east analysis, but let the analysis determine the actual mean
flood direction (probably the M2 max flood direction); the analysis can then either be rerun
with that new major axis direction, or the constituent can be converted to the new major axis
direction; if the signal-to-noise ratio looks small in the minor component, decide whether to
192
choose orthogonal components that put equal amounts of the tidal signal into each
component; [Sections 5.2 and 5.3.2]
(7) determine whether additional shallow-water tidal current constituents should be included
in the analysis; this will be obvious for a very shallow waterway with strong tidal currents,
but one should first use the standard 37 constituents and see how much energy is left in the
higher harmonic tidal bands of a spectrum; for an extremely nonlinear case, the highest order
overtide that can be solved for will depend on the sampling rate of the current data time
series, but the sampling rate of most current sensors (several sample per hour) is more than
sufficient. [Section 5.3.1]
(8) determine how much of the data time series to use in the harmonic analysis; in Step (5) the
longest synodic period within the length of the available data time series was selected, but
one may have more data than that (but less than the next highest synodic period); run the first
analysis using a segment of the time series whose length is the selected synodic period; if
there is enough extra data, it will be useful to run one analysis from the beginning of the time
series and a second analysis that includes the end of the time series (and look for consistency
in the harmonic constants for each, especially in the smallest constituents); if the length of
the available data series is more than half way to the next synodic period, the follow-up
analysis can use all the available data and include additional harmonic constituents that
theoretically require the next highest synodic period; if one has a second (also short) data
time series, also harmonically analyze that time series and look for consistency in the two
sets of constituents; [Section 5.3.1]
(9) decide whether to use a particular shallow-water compound tidal current constituent or the
astronomical constituent with the same frequency (e.g., 2MN2 or L2); for deep-water stations
the astronomical constituent is fine, but for shallow-water stations (especially inside bays
and rivers) one will probably need to use the shallow-water constituent; if one is unsure, one
will have to look at the harmonic analysis results of several current stations in the bay and
see if the rate of growth moving up the bay is greater than the rate of growth of an
astronomical constituent in the same tidal band (e.g., see if 2MN2 grows at a faster rate than
M2); for constituent pairs whose node factors are the same (e.g., 2MS2 and 2) it does not
matter which is selected; for pairs with different node factors, if one initially chose the
astronomical constituent, but later decides it should have been the shallow-water constituent,
one does not have to re-analyze, for one can merely correct for the different node factor and
equilibrium argument. [Section 5.3.1]
(10) check the data file for notes about instrument errors, movement of the current sensor, or
other potential problems with the data; if such problems are noted, keep in mind the possible
effects of the five types of errors on the harmonic constants; [Section 5.3.3]
(11) carry out the least squares harmonic analysis; if an orthogonal component analysis has been
run, plot the current constituent ellipses; decide whether to rerun the analysis with a new
major axis direction;
(12) tidally filter the current data time series to look for strong nontidal events (like spring river
runoff, strong wind currents, or internal tides) that may nonlinearly affect the tidal analysis
results; one should probably also look at a tidally filtered water level data series from a
nearby station (for storm surges, which do not usually have strong currents associated with
them, but which can significantly affect both the tide and the tidal current); if one needs
harmonic constants from which predictions can be made for most of the year, one might
consider leaving out the nontidal event periods and rerunning the least squares harmonic
193
analysis using the gap option; if one has a short time series of current data that includes a
major nontidal event, one will probably not be able to use the harmonic constants that come
out of the analysis for tidal current prediction. [Section 5.3.4]
(13) with the harmonic constants from the first harmonic analysis, make a tidal current prediction
for the same period as the data, and subtract this predicted time series from the original
observed time series to produce a residual time series; if one has more data than were used
to do the harmonic analysis, then also produce predicted and residual time series for that
second time period (this other residual time series will provide more meaningful results in
steps (15) and (16) than the residual series from the self prediction); [Section 5.5.3]
(14) carry out a spectral analysis on the residual time series to see if there is any remaining tidal
energy still in the record; also carry out a spectral analysis on the original observed time
series for comparison purposes. [Section 5.5.3]
(15) examine a plot of the residual time series looking for places where a transient tidal signal
appears; compare to the results of Step (12); investigate the wind, river, water density, or
other potential nontidal influences for those time periods; one may have to consider leaving
time periods with strong nontidal events out of the analysis and rerunning the least squares
harmonic analysis using the gap option; if one has a short time series of current data that
includes a major nontidal event, one will probably not be able to use the harmonic constants
that come out of the analysis for tide prediction. [Section 5.5.4]
(16) if there still is tidal energy in the residual time series try to determine what additional tidal
constituents might be added to the harmonic analysis; most easy to spot will be energy in
higher species bands due to shallow-water effects; decide which additional shallow-water
constituents are most likely to be important; [Table 3.2 or Table A.2]
for energy left in the semidiurnal or diurnal bands, one should determine which are the next
semidiurnal or diurnal constituents that were left out of the analysis because of insufficient
length of the time series; there may also be compound tidal constituents (due to shallow-
water effects) that should be added to the next analysis; [Table 3.2 or Table A.2]
(17) obtain values for additional constituents; if one does not have enough data to move to the
next synodic period, one can then either:
(a) use an inference (and elimination) technique to calculate the amplitudes and epochs for
the missing constituents (and to correct the amplitudes and epochs of the solved-for
constituents; [Section 4.2]
(b) try rerunning the harmonic analysis with these constituent included using the full length
of the available time series, to see if the least squares technique can successfully
determine them; and
(18) when it is decided that one has the optimum set of harmonic constants for a particular current
station, run a final prediction for the full length of data and calculate all the maximum
floods, maximum ebbs, and slack waters (or minimums); compare these predicted maximum
floods, maximum ebbs, and slack waters (or minimums) with the equivalent values
calculated from the data time series, determining all the usual statistics; (if possible, use the
same max/min routine for both predicted and data series); if the mean maximum floods (and
the mean maximum ebbs) from the data are larger than the mean predicted maximum floods
(and mean predicted maximum ebbs) consider going back to Step (17).
194
6. Nonharmonic Analysis of Water Level
and Current Data
Nonharmonic comparison analysis methods for analyzing the tide and tidal current were
introduced and briefly described in Section 3.6. This type of analysis is not very sophisticated,
simply comparing one or more aspects of the tide (or tidal current) at two stations. But it has been
used to analyze data at thousands of stations which have been put into the Tide or Tidal Current
Tables. A nonharmonic comparison analysis is used to compare a subordinate station to a reference
station, with the resulting time differences and height differences (for tides) or current speed ratios
(for tidal currents) listed in Table 2 of the Tide or Tidal Current Tables. In Section 3.6.3 a key fact
was demonstrated (using several detailed examples) about the time and height differences and speed
ratios that come out of a nonharmonic comparison analysis:
the time and height differences between corresponding high waters (or low waters)
at two tide stations vary from tidal cycle to tidal cycle throughout the month (and
similarly for tidal current stations).
If the tidal characteristics at the two stations being compared are very similar, then this variation
throughout the month will not be significant (which is the desired result). Often, however, this is
not the case. The tidal characteristics of the two stations may not be similar enough because either
they are not close enough geographically or because the waterway has a mixed tide with a
significant diurnal inequality, which causes the tidal characteristics to change significantly over even
fairly short geographic distances. If this is the case then these differences can vary significantly
from the mean value when, for example, the moon is at maximum declination.
This fact is important because the main use for the mean time and height differences that come
out of a comparison tidal analysis is to produce correction factors for a subordinate station in Table 2
of the Tide Tables. These correction factors are used to produce tide predictions at the subordinate
station by applying them to the (harmonically produced) tide predictions at a reference station. A
single mean value of the time or height difference is usually used to make these subordinate station
predictions, so errors in the prediction will occur at certain times of the month (such as maximum
lunar declination, as mentioned above).
This chapter will present more details about using the various types of nonharmonic comparison
tidal analysis, including a discussion of the consequences of the variation in nonharmonically
determined time and height differences throughout the month. The most important consequence will
be errors in making tide predictions using the correction factors found in Table 2 of the Tide and
Tidal Current Tables, and ways will be looked at to minimize such errors.
195
6.1 The Need For Nonharmonic Tidal Analyses
As was mentioned above and in Section 3.6, nonharmonic comparison analysis does not provide
as accurate a representation of the tide or tidal current as does harmonic analysis, nor are the
predictions made using the time and height differences (or tidal current speed ratios) that come out
of a nonharmonic comparison analysis as accurate as predictions made using tidal harmonic
constants. Tide predictions made with such time and height differences are made by applying these
difference to the harmonically produced tide predictions at a reference station, and the fact that these
differences are not accurate for all times of the month degrades the tide prediction to some degree.
In spite of its lesser accuracy, nonharmonic analysis is still required in order to put stations into
Table 2 of the Tide and Tidal Current Tables. Table 2 contain thousands of subordinate stations
which could not feasiblely be included as daily predictions in Table 1, even if one wanted to. At
four pages per station in Table 1 (versus one line in Table 2), it would take thousands of pages to
include these stations in Table 1. For example, in the 2007 edition of Tide Tables, East Coast of
North and South America, produced by CO-OPS for NOAA, Table 1 consists of daily predictions
for 76 harmonically analyzed reference stations, while Table 2 consists of time differences and
height differences and/or ratios for 2589 subordinate stations.
Nonharmonic comparison analysis is also required for stations with data times series too short
to be accurately analyzed harmonically (usually less than 15 days). Short time series have been
especially common for stations with observed current data, which have been much more difficult
and expensive to deploy and maintain compared with land-based water level stations. For short
stations there is no choice but to analyze such stations nonharmonically and then put them into Table 2.
Tide and Tidal Current tables are still the primary means for providing tidal predictions to users
in the marine community, and often hard copy versions are carried by the mariner and recreational
boater. However, even if there is an eventual transition to fully digital means of providing tidal
predictions, with all predictions produced harmonically (see Section 9.7), there are thousand of
stations already in Table 2 of the Tide and Tidal Current Tables that are based on short data times
series, and these stations would need to be re-occupied (for at least 15 days, and preferably 30) in
order to include them in a harmonically-based digital prediction product. Table 2 is therefore likely
to remain a part of all Tide and Tidal Current Tables in the near future, and so nonharmonic
comparison analyses will continue to be used.
6.2 Hydrodynamic Considerations
A key fact first mentioned in Section 3.6.3 is repeated here again, which should always be
remembered when carrying out a nonharmonic comparison analysis (especially when the results are
going to be put in Table 2 of the Tide or Tidal Current Tables):
the time and height differences between corresponding high waters (or low waters)
at two tide stations vary from tidal cycle to tidal cycle throughout the month (and
similarly for tidal current stations).
This variation in time and height differences is a consequence of hydrodynamic effects on the times
and heights of high waters and low waters.
Two stations in different parts of a waterway will have a different harmonic makeup because,
as was seen in Section 2.3.1, various hydrodynamic effects cause the amplitudes and phase lags of
196
6. Nonharmonic Analysis of Water Level and Current Data
the harmonic constituents to change with geographic distance, along both the length and width of
a waterway, and these changes are frequency dependent. Thus, as one moves along the waterway
the ratio of two harmonic constants changes. The larger the difference between the frequencies of
the two constituents the faster their ratio changes with distance. The ratio of a diurnal constituent
to a semidiurnal constituent, for example, K1/M2, changes much faster than, for example, S2/M2
changes. As was seen in Section 3.6.3, in a waterway with a mixed tide and a significant diurnal
inequality, two stations do not have to be separated geographically by many miles to lead to large
variations throughout the month in the differences between the times and heights of high waters at
the two stations. It was also seen in Section 3.6.3 that the distortion of the tide curve or tidal current
curves due to nonlinear shallow-water effects can also lead to variations in the time and height
differences and speed ratios.
In essence, by using a constant time or height difference or a constant speed ratio throughout the
month one is assuming that only M2 is important, when, in fact, other tidal harmonic constituents are
also important. A constant time difference between times of high water at the two locations would
also be possible if the two stations in a bay had exactly the same relative harmonic makeup, that is,
if the amplitude ratios of and the phase differences between corresponding harmonic constituents
were exactly the same. But the hydrodynamics of the bay makes this impossible. Numerous
hydrodynamic effects have a frequency dependence that causes them to affect each harmonic tidal
constituent differently (see Sections 2.3 and Chapter 7). This includes the effects of partial and full
reflections with which the bay affects the amplitude and phase lag of each tidal constituent, as well
as frictional effects, continuity effects, and inertial effects, all of which are frequency dependent.
As mentioned above, the average time difference between high waters at the two stations equals
approximately the time difference between the two respective M2 high waters, but other tidal
constituents, each with their own phase lag (time) difference, also affect the time and height of high
water (and likewise the maximum floods and ebbs in a tidal current). The time difference between
high water at the two stations therefore varies throughout the month, varying from springs to neaps
or from maximum lunar declination to equatorial declination. Generally, the greater the
geographical distance between two stations the greater the change in the tide regime and the greater
the difference in harmonic makeup, and thus the greater the variation in time differences. The
fundamental consideration is how quickly the tidal characteristics change with distance in the
vicinity of the two stations being compared.
Four different types of nonharmonic comparison analyses will now be looked at in some detail,
the first two for tides and the second two for tidal currents.
6.3 Monthly Mean Tide Analysis
6.3.1 Details of the Analysis
The monthly mean tidal analysis is described in the Manual of Tide Observations (U.S. C&GS,
1965, pages 48, 51-57); see also CO-OPS, 2003. The relevant section in that manual is called High
and Low Water Tabulations because the bulk of the work was in tabulating the times and heights
of high water and low water at a water level station, usually a month at a time. Prior to computer-
assisted data processing and analysis, such tabulations were put on a special form, C&GS Form
2211, a copy of which is shown in Figure 6.1 (taken from C&GS, 1965, page 55). One can see in
that figure that Form 2211 provided two lines for each calendar day, the morning high water and low
water times and heights entered on the first line, and the afternoon values on the second line. A
197
Figure 6.1. A copy of Form 2211, on which tabulations were done in the pre-computer era
when carrying out a monthly mean analysis.
198
dashed line indicated when there was no high water (or low water) during that 12-hour period, which
happens every 14 or 15 days since the time of high water (or low water) gets progressively later
every day (usually by roughly 50 minutes). (The dashed line distinguished this situation from
missing data, which was simply left blank.)
Also tabulated on this form were the times of Moons Transit over the Greenwich time meridian
(these transits also being available in the American Ephemeris and Nautical Almanac). Lower
transits (i.e., the moon passing over that part of the meridian lying below the polar axis) were
enclosed in parentheses to distinguish them from upper transits (i.e., the moon passing over that part
of the meridian lying above the polar axis). The lunitidal interval for each high water (or a low
water) was calculated by subtracting the latest moons transit just prior to the time of high water (or
low water). The lunitidal intervals calculated using lower transits were indicated by parentheses.
Thus, there were seven values listed for each half calendar day, in order: moons transit, time of
high water, time of low water, lunitidal interval for high water, lunitidal interval for low water,
height of high water, and height of low water. At the bottom of the right half of this form the
lunitidal intervals were summed, as were the heights of high water and low water; then means were
calculated for each. Since the moons transits were given in Greenwich time but (for convenience)
the times of high and low waters were given in Standard Time (for the local time meridian) the
means had to be corrected to become Greenwich lunitidal intervals, by adding one hour for every
15o difference between the local time meridian and Greenwich (if the local time meridian is west of
Greenwich; if it is east of Greenwich then subtraction is done, and sometimes one had to subtract
12.42 hours if the result is 12.42 hours or greater).
The local interval is relative to the moon passing over the longitude of the tide station location
(rather than passing over Greenwich). To calculate this one had to know the lunar interval, i.e., the
time it would take the moon to pass from the Greenwich meridian to the station location. This
astronomically-determined value was obtained from a table (Table 7 in U.S. C&GS, 1965, pages 53)
but it essentially meant subtracting approximately 0.069 hour for every degree of longitude change
(when the station was west of Greenwich), plus subtracting approximately 0.00115 hour for every
additional minute of longitude difference.
Form 2211 was also used to calculate the mean high water (MHW) and the mean low water
(MLW) for the month of the tabulations. The mean tide range (Mn) was also calculated by
subtracting MLW from MHW, and the mean tide level (MTL) was calculated by taking the average
of MHW and MLW. For tide stations with a significant diurnal inequality, there were places on the
form to sum all the higher high waters and all the lower low waters, so that the mean higher high
water (MHHW, but called HHW on Form 2211) and mean lower low water (MLLW, but called
LLW on Form 2211) could be calculated. Then the diurnal high water inequality (DHQ) was
calculated by subtracting the mean of all the high waters from the mean of all the higher high waters,
and likewise for the diurnal low water inequality (DLQ).
The values of Mn, DHQ, and DLQ were the mean range, diurnal high water inequality, and
diurnal low water inequality for that particular month of water level data. These values were then
corrected for the 18.6-year variation in the procession of the lunar node (see Sections 2.2.4 and
3.4.2), which was accomplished using other tables in U.S. C&GS, 1965 (Tables 8 and 9, pages 56-7),
that were based on tables originally found in Harris Manual of Tides (1897-1907). However, these
tables also included the effects of the relation of the diurnal tide to the semidiurnal tide at the station,
which was represented by the formula 2(DHQ + DLQ)/Mn, which should be approximately
equivalent to (K1+O1)/M2.
199
Figure 6.2. A copy of Form 657, which was used in the pre-computer era for calculating
and recording comparison of monthly means.
In order to find time differences and height differences between a subordinate station (to be put
in Table 2 of the Tide Tables) and a reference station, another form, C&GS Form 657, was used.
Form 657 is shown in Figure 6.2 (taken from Tidal Datum Planes by Marmer, 1951, page 139).
This form simply provided three columns for each of seven parameters, the first two columns being
values for the subordinate station and for the reference stations (which is referred to as the standard
station on the form) and the third column being for the difference or ratio of the two values. The
seven parameters for which this was done were: mean tide level (MTL), mean sea level (MSL), high
water interval (HWI), low water interval (LWI), mean range (Mn), diurnal high water inequality
(DHQ), and diurnal low water inequality (DLQ). For the first two parameters the third column is
the height difference in feet, for the next two parameters the third column is the time difference in
hours, and for the last three parameters the third column is the height ratio. There is one line for
each set of monthly means being compared.
200
Figure 6.3. A copy of Form 248, which was used in the pre-computer era to carry out
a comparison of simultaneous observations (a tide-by-tide analysis).
201
Essentially the same procedures are now done routinely using computers, as part of the standard
water level data processing and analysis in CO-OPS.
6.3.2 The Limitations of the Monthly Mean Tide Analysis
The monthly mean tide analysis is routinely carried out for all tide stations for which there is a
month of data or more. Most tide stations are occupied for at least a year, and a couple of hundred
stations are permanently installed as part of the National Water Level Observation Network
(NWLON). Thus many monthly tabulations and monthly means have accumulated at these stations.
It is part of standard data processing for water level data.
The major drawback with this method for putting tide stations in Table 2 is that only the monthly
means are used to calculate the time and height differences between the subordinate station being
put in Table 2 and the reference station. The differences between the two stations are not looked
at cycle by cycle, and so there is no way of knowing how much the differences actually vary
throughout the month due to changing lunar declination or the change from neaps to springs. One
can see variations from month to month, but those variations are much smaller than those that occur
within a month.
If the reference station is nearby and has tidal harmonic constant ratios and phase lag differences
that are similar to those at the subordinate station, the results from the comparison of the monthly
means will be of good quality, but one can not tell how good that is unless one also carries out a
tide-by-tide analysis (see next Section).
6.4 Tide-by-Tide Analysis
A tide-by-tide analysis is the cycle-by-cycle analysis applied to water level data (see CO-OPS,
2003). Classically in the Coast and Geodetic Survey it was usually referred to as a comparison of
simultaneous observations (Manual of Tide Observations, C&GS, 1965, page 58), although one
could also compare observations from one station to the tide predictions at a nearby reference
station, for the same time period. In the pre-computer era a special form was used, C&GS Form
248, which was titled: TIDES: Comparison of Simultaneous Observations. This analysis tended
to be used only for short series of observations, the monthly mean analysis being typically used
when there was a longer data time series. According to the above manual, for series extended over
longer periods, Form 657 for the comparison of monthly means will be found more convenient.
However, as will be seen in the next section, the tide-by-tide analysis has advantages over the
monthly mean analysis, and in that pre-computer era it was probably its labor intensiveness that led
to it being used only for short time series.
To begin a brief description is given about how the analysis was carried out using Form 248,
since that procedure was essentially transferred to computer operation. Form 248 is shown in Figure
6.3. In the form there is one line for each half calendar day, and twelve values on each line. The
first four columns are the times of high water and low water for the two stations being compared.
The next two columns are the time differences between the high waters and between the low waters
at the two stations. The last six columns are the equivalent for the heights of the high and low
waters and the height differences.
202
The sums and means of all the time differences for high water and for low water are calculated
and written at the bottom of the sixth and seventh columns. The sums and means for the heights and
the height differences for high water and for low water are done separately for: higher high waters,
lower high waters, higher low waters, and lower low waters. Further calculations in the spaces
provided below the tabulation lead to calculation of mean high water height difference, the mean
range ratio, the mean tide level ratio, the DHQ ratio and the DLQ ratio, and the high water interval
and low water interval.
The tide-by-tide analysis has been computerized (see CO-COPS, 2003) and a short analysis
using that program is shown in Figure 6.4 The analysis is straight forward, the only somewhat
tricky parts being: keeping the respective cycles in sync at the two stations; allowing for cases
where the tide goes diurnal; and dealing with distorted tide curves in very shallow water. It is also
important to use of the same method for selecting the high waters and low waters, if possible.
6.4.2 The Benefits and Limitation of the Tide-by-Tide Analysis
As was seen in Section 3.6.3, the difference between the height of a high water at a subordinate
station and the height of a high water at a reference station will vary throughout the month. This
variation can be significant if the tidal characteristics of the two stations are not as similar as one
would like it to be. Because of the hydrodynamics of the waterway, the further apart two stations
are geographically the greater the difference in their tidal characteristics. But such distances need
not be very great if the waterway has mixed tides with a strong diurnal signal. It is this mean
difference in high water heights between the two stations that ends up in Table 2 of a Tide Table,
Figure 6.4. A short tide-by-tide analysis produced with a recent computer program. (From
CO-OPS, 2003)
203
and which will be used to produce tide predictions at the subordinate station by modifying the tide
predictions at the reference station. Since only the mean value of the difference is used, there will
be errors at certain times of the month corresponding to those times when the height difference was
different than the mean value (usually near maximum lunar declination in mixed tide areas, or during
spring tides when there is a significant difference between springs and neaps, that is, when the effect
of S2 is large). This variation in high water height difference (as well as low water height
differences, and the time difference for high water and lower) from cycle to cycle cannot be seen
in a monthly mean analysis, since only monthly averages are produced and compared. However,
the tide-by-tide analysis, as its name would imply, does show the cycle-by-cycle variation.
Stations in waterways with a significant diurnal inequality are usually the ones most poorly
represented in Table 2 of the Tide Tables, especially if there are only one or two reference stations
for that waterway. An example of this was given in Section 3.6.3 for the comparison of two tide
stations in San Francisco Bay, where the tides are mixed and there is a strong diurnal signal (see
Figure 3.5). If a tide-by-tide analysis is done on actual observed data, then the differences can result
from the effects of nontidal influences such as wind, atmospheric pressure, or river discharge, as
well as from changing astronomical conditions. Typically the differences due to nontidal influences
will not be cyclical, whereas the cyclical nature of the changing astronomical conditions will stand
out. However, to make very clear the effect of the changing astronomical conditions on the
differences between two tide stations that do not have similar enough harmonic characteristics, the
tide-by-tide analysis can be done on predictions for the two stations. Then the differences are solely
the effect of changing astronomical conditions on the tide. This was done in Section 3.6.3.
Figure 3.6a shows a plot of the results of a tide-by-tide analysis, which exhibited significant
variation in the time difference throughout the month between low waters at the two tide stations.
Figure 3.6b showed a significant variation throughout the month in the height difference between
low waters at the two tide stations. In each plot one sees the maximum differences a short while
after maximum lunar declination. If the mean time difference and the mean height difference were
put into Table 2 for the subordinate station, and then used to make predictions of low water at the
subordinate station by modifying the times and heights of low waters at the reference station, then
these maximum differences translate into prediction errors ranging from -0.44 hour to + 0.52 hours
and from -1.11 feet to +0.69 foot. A new reference station for the subordinate stations in this
waterway would thus be called for, and in fact, Port Chicago was added to the U.S. Tide Tables as
a reference station because of the results displayed in Figure 3.6.
This analysis is probably the most important of all types of nonharmonic comparison analyses
because it shows how the various differences and ratios vary from tidal cycle to tidal cycle. Thus,
this technique can be used for quality assurance analyses of subordinate stations in Table 2.
However, the tide-by-tide analysis has one limitation. It deals only with high and low waters
and does not provide any information about tidal heights in between the high and low waters. For
data from deep-water stations this is not a problem, since the tide curve will look like a cosine curve.
However, for shallow-water stations the tide curve is distorted away from a pure cosine curve by
the nonlinear hydrodynamic effects. It can have a variety of shapes, often with more rapid rises to
high water and slow falls to low water. In extreme cases there can be double high waters or double
low waters. The only type of nonharmonic comparison analysis that could be provide this
information would be a mean tidal cycle analysis (analogous to the rotary reduction analysis,
Section 6.6). Such an analysis would chop up the water level times series into tidal cycles (based
on, for example, the times of high water at a reference station), add them together, and average them.
The resulting mean tide curve could be used for interpolating between the predicted high and low
204
waters, instead of using a cosine curve, which would provide erroneous values for a shallow-water
station.
6.5 Reversing Reduction Analysis for Tidal Currents
The reversing reduction analysis is the cycle-by-cycle analysis applied to current data. Its name
comes from the fact that this type of analysis was intended to be used on reversing tidal currents,
that is, tidal currents that are bidirectional. In this case the tidal current flows upstream or into a
bay (the flood current), reaches a maximum flood speed, then slows up until it reaches zero speed
(slack water), then flows in the opposite direction going downstream or out of a bay (ebb current),
reaches a maximum ebb speed, then slows up, again reaches another slack water, and reverses flow
to the flood direction again. The tidal current speed curve for a reversing tidal current looks like a
tide curve, with the zero speed (slack water) line serving the role of a datum, and with flood currents
above that zero line (and considered positive) and ebb currents below the line (and considered
negative).
However, only in the narrowest waterways are the currents truly reversing. In most cases the
tidal current is rotary (see Section 6.6), but since the shape traced out by the end of the rotating
current vector over a tidal cycle is usually elliptical (except in the open ocean where it is circular),
with a clear maximum flood and maximum ebb, the current can still be analyzed as though it was
reversing. (In such cases a tidal current speed curve as described above, with flood currents above
the zero line and ebb currents below the zero line, will not smoothly pass through this slack-water
line, because for such cases there is no real slack water, only minimum currents flowing
perpendicular to the flood-ebb axis. Instead two plots are typically produced, one for the absolute
current speed, and one for the current direction.) A rotary reduction analysis would be more
appropriate, but before the computer era the reversing reduction was used instead because it was
much less labor intensive to do by hand. Once computers came on the scene the opposite became
true and rotary reduction analyses were predominantly done because that analysis was much easier
to program for the computer. Each type of analysis has its benefits and drawbacks, and both really
should be done when analyzing a current station.
In a reversing reduction analysis key points are picked off from two time series of current data
or tidal current predictions, specifically: maximum flood, maximum ebb, slack water (or minimum
current) before flood, and slack water (or minimum current) before ebb. Time differences and speed
ratios between corresponding key points (such as maximum flood) at the two current stations are
calculated for each tidal cycle. Means for these time differences and speed ratios for each key point
are calculated for the entire data record. These means should be close to those that would come out
of a rotary reduction analysis, but the real benefit of the reversing reduction analysis is that is also
allows one to see how the time differences and speed ratios vary throughout the month.
In the pre-computer era the reversing reduction analysis used was different in a couple of
important details. The manual undertaking of this analysis used C&GS Form 451, shown in Figure
6.5 (taken from U.S. C&GS, 1950, page 54). This form had six columns, for the tabulation of the
205
times of slack before flood, strength of flood, slack before ebb, and strength of ebb (where strength
of means the same as maximum), and for the speed and direction of the strength of flood and
strength of ebb (on the form the word velocity was used for speed). On this form the reference
station used was a tide station and so the times of high water and low water for each cycle were also
tabulated in two of the columns. However, as was seen in Section 3.6, referencing a current station
to a tide station can cause problems, because generally their tidal harmonic makeup will be different,
but back when this form was used there were not many long-term current reference stations. Form
451 was modified in the 1960s to restrict the reference station to being a current station. Thus the
two columns for the high and low waters were replaced with six columns for the reference current
station, for the same six parameters as in the six columns for the subordinate current station. Both
versions of Form 451 showed only time differences for each cycle; speed ratios were not shown for
each cycle, and the mean maximum flood speed ratio and mean maximum ebb speed ratio between
the two stations were calculated from the calculated mean flood speed and calculated mean ebb
Figure 6.5. An early version of Form 451, which was used in the pre-
computer era to carry out a reversing reduction analysis of current data (from
Manual of Current Observations, 1950).
206
speed for each station. Thus, there was no information about how the flood speed ratio or ebb speed
ratio changed from tidal cycle to tidal cycle.
When the author of this book computerized the reversing reduction in the 1980s, he added this
capability, thus allowing the reversing reduction to be able to serve a quality control function, by
determining how much the time differences and speed ratios varied throughout the month, and thus
how large errors would be in predictions made using the mean time differences and mean speed
ratios that came from the analysis.
The basic reversing reduction analysis is very straightforward, but its accuracy and usefulness
depends on having an accurate and consistent method for selecting maximum floods and ebbs, and
minimums before flood and ebb, from a current data times series. A combination of sometimes
noisy data and sometimes significant distortion by shallow-water effects made this analysis more
difficult to program than the rotary reduction analysis (which did not need to pick maximum and
minimums until at the end when the data had already been vectorially averaged into one mean tidal
current curve).
The computerized version of the reversing reduction analysis was used to produce the plots
shown in Figures 3.7 and 3.9 showing the differences in maximum ebb times for tidal currents in
the Delaware River near Philadelphia relative to the reference tidal current station at the Delaware
Bay Entrance. (The tides option in this program was also used to produced the plots in Figures 3.6a
and 3.6b showing the difference in low water times and heights for tides at Port Chicago in San
Francisco Bay relative to the reference tide station at Golden Gate Bridge.)
Figure 6.6 shows the tabular output from the computerized reversing reduction analysis, in this
case for a current station at Kent Island in Chesapeake Bay relative to the reference tidal current
station at the Chesapeake Bay Entrance. Only one page of tabular results is shown, with the first
few tidal current cycles and the last few, plus the final mean results. One can see that the time of
slack before ebb at Kent Island was on average 8.63 hours later than the time of slack before flood
at the Chesapeake Bay Entrance, but that it could be as much as an hour later than that or an hour
earlier (see also the plots in Figure 6.7). Thus, if +8.63 hours was used in Table 2 as a correction
factor with which to produce tidal current predictions at Kent Island based on the (harmonically
produced) tidal current predictions at the Chesapeake Bay Entrance, there would be certain times
of the month where the Kent Island prediction could be off by up to an hour, either earlier or later.
This is because Kent Island is too far from the Chesapeake Bay Entrance for the two stations to have
similar enough tidal harmonic makeup. (And, in fact, in the Tide Tables, Kent Island is referenced
to the much closer reference station at Baltimore Harbor Approaches.)
Another example of the results of a reversing reduction, this time for two current stations in San
Francisco Bay (Carquinez Strait and Alcatraz-North Point) are given in Figures 6.8 and 6.9. The
locations of these two stations are shown in Figure 3.5, along with their diurnal-to-semidiurnal
ratios. The (K1+O1)/M2 ratio at Carquinez Strait is 0.594, while it is 0.471 at Alcatraz-North Point,
and the phase lag differences at the two stations differ by only 7o, yet one sees significant variation
in the time differences and speed ratios between the two stations. In Figure 6.8 the difference in
maximum ebb times at the two stations varies throughout the month, averaging 3.38 hours, but
varying from a half hour smaller than that to a half hour greater. In Figure 6.9 the ratio of maximum
ebb speeds at the two stations varies through the month, averaging 1.36, but varying from 0.77 to
1.86.
207
Figure 6.6. Results of a computerized reversing reduction analysis

for Kent Island, MD, relative to a reference station at Chesapeake
Bay Entrance. Only 4 days from the 365-day analysis are shown,
along with the results.
Figure 6.7. Plotted time differences between slack before

flood at Kent Island, MD, and a reference station at
Chesapeake Bay Entrance.
208
Figure 6.8. Plotted results from a reversing reduction analysis

showing the time differences between maximum ebbs at Carquinez
Strait and Alcartraz-North Point, both in San Francisco Bay.
Along the horizontal axis, N=maximum northern lunar declination,
S=maximum southern lunar declination, and E=equatorial lunar
declination. (From Parker, 1991b.)
Figure 6.9. Plotted results from a reversing reduction analysis

showing the ratio of maximum ebb speeds at Carquinez Strait
and Alcatraz-North Point in San Francisco Bay. See Figure 6.8
for the meaning of N, S, and E. (From Parker, 1991b.)
209
6.5.2 The Advantages and Limitations of the Reversing Reduction Analysis
As mentioned earlier, the main advantage of using the reversing reduction analysis is that one
can see cycle by cycle how the time differences and speed ratios vary throughout the month. This
knowledge allows one to quality control the resulting mean values that come out of the analysis and
that will be put into Table 2 of the Tidal Current Tables. Since these time differences and speed
ratios will be used to make predictions for a subordinate station in Table 2 (by applying them to the
daily predictions at the nearest reference station in Table 1), the reversing reduction analysis
provides a quality analysis of those predictions. (The other nonharmonic comparison analysis used
on tidal currents, the rotary reduction analysis, does not provide this capability.)
As discussed before, if the variation in the time difference and speed ratios (which equate to
the errors one would see in the predictions using the mean time differences and mean speed ratios),
is larger than acceptable (for a particular application, such a maritime safety) then one may need to
establish a new (harmonically produced) reference station for this (and other nearby ) subordinate
current stations. However, if that is not possible, there is another possibility, using the results of the
reversing reduction analysis. One can try to use these results to calculate additional types of mean
time difference and mean speed ratio, for example, for times of maximum lunar declination. Such
values might be added to the Endnotes for Table 2. If there are many subordinate in need of those
additional time differences and speed ratios, one could even change the format of the Table 2 page
(for those stations) to more efficiently handle it.
The reversing reduction analysis (like the tide-by-tide analysis) does have a limitation, which
can be significant for shallow water areas. It deals only with the maximum floods and ebbs and the
slacks, and cannot provide information on the currents in-between those points. Except for narrow
waterways, the tidal current is rotary, with the tip of its current vector rotating around the compass
in one tidal cycle and tracing out an ellipse. None of this information is captured in a reversing
reduction. For offshore areas, this tidal current ellipse can even be circular, with no clear maximum
flood direction and no clear maximum ebb direction. (Even in the pre-computer era, such ocean
stations had to be analyzed using the rotary reduction analysis, done by hand.) Also, the tidal
current in shallow waterways is distorted by the nonlinear mechanisms discussed in Sections 2.3.2
and 2.3.6. The tidal current speed curve in such areas does not look at all like a pure cosine curve.
It can have short flood durations but with high speeds, and long ebb durations with lower speeds.
It can have double maximum floods or double maximum ebbs. None of this information is shown
by a reversing reduction analysis. Such information is, however, provided by the rotary reduction
analysis described in the next section.
6.6 Rotary Reduction Analysis for Tidal Currents
6.6.1 The Benefits of Using the Rotary Analysis
The rotary reduction analysis (described briefly in Section 3.6.2) is a form of nonharmonic
analysis that was originally used only for offshore current stations where the current was rotary, that
is, where the direction of the current flow rotated completely around the compass over one tidal
cycle. This can be represented by a current rose, but more often the tips of the moving current
vectors are connected, and usually trace out an ellipse. For areas away from shore the ellipse is
often almost circular, making it difficult to decide what to call the flood direction and what to call
the ebb direction (which is the reason the rotary reduction analysis was used instead of the reversing
210
reduction analysis). Whether circular or elliptical, there are no slack waters. Current always flows
in some direction. In locations where there is an ellipse, and for which there is an obvious flood and
ebb direction, about half way between maximum flood and maximum ebb the current flows
perpendicular to the flood-ebb direction and reaches a minimum speed. The narrower the ellipse
the smaller the minimum current speed. (See Figure 2.4.)
In reality, most tidal currents are rotary, with only tidal currents in very narrow channels having
true slack waters and being true reversing currents. The rotation of the Earth is the main cause of
the rotation offshore, but geography and bathymetry also play a role especially within bays and
estuaries.
Before computers, it took much more manual effort to carry out a rotary analysis than a reversing
reduction, so rotary analyses were usually carried out only for offshore current stations where there
were no obvious flood and ebb directions. Table 5 in each Tidal Current Table (see Section 9.3.3)
is a special table for those offshore stations with rotary currents, where the speed and direction of
current flow is provided for an entire mean tidal cycle in hourly increments (the time increment can
be smaller if needed). Prior to 1972 all current stations put into Table 2 were analyzed using the
reversing reduction.
The author of this book first computerized the rotary analysis in 1972. From that point on most
current stations that were put into Table 2 were analyzed using the rotary analysis. There were
several reasons for this change over.
First, although the rotary analysis took more time to carry out manually, ironically it was much
easier to computerize than the reversing analysis (which would not be computerized for another
decade, also by the author of this manual). The time savings in using a computerized rotary analysis
instead of a manual reversing analysis were enormous.
Second, as mentioned above, all tidal currents are really rotary and the rotary analysis gives a
better description of the current, showing the exact speed and direction of flow for every hour of the
tidal cycle. The reversing analysis only gives the times and speeds of maximum flood and maximum
ebb and the times of the (supposed) slack waters. It gives no information about what is going on in
between those times. In the past if such information was wanted it was assumed that a cosine curve
could be fitted to those four known points on the tidal current curve which, in fact, was almost
always an erroneous assumption and in shallow waterways dramatically so. After analyzing many
current data records with the rotary analysis it became apparent that not only did the current rotate,
but often the tips of the current vectors did not always trace out simple ellipses. As a result of the
geography of the waterway and especially of the bathymetry, the shapes of the path traced out by
the current vector tips could be quite varied, especially at locations where there were channel bends
or where several channels met at the locations of the current station. Figure 2.4 shows some
examples, including a station where a figure eight was traced out.
Third, for shallow-water areas the tidal current (even more than the tide) is distorted by nonlinear
effects. The rotary analysis clearly shows this distortion because the result is an entire mean tidal
current curve (not just maximum floods and ebbs and slacks as gotten from a reversing reduction).
When the current speeds are plotted (like a tide curve) they rarely look like a smooth cosine curve,
but are often flattened out in one phase (flood or ebb) and show a sharper peak in the other phase,
or other times may stretch out the times of minimum flow significantly. In some cases there may
be double floods or double ebbs. These are the same effects represented by the overtide tidal
constituents (e.g., M4, M6, M8, and others) in a harmonic analysis. As was seen in Section 2.3.6,
superposition of two tide waves can greatly enhance these overtide effects, as can the inertial effects
seen at channel bends and in the sheltered areas near points.
211
Fourth, the rotary analysis is vastly superior for producing Tidal Current Charts. In the past,
using a reversing reduction, the 12 charts representing the tidal current flow for each hour increment
of a mean tidal cycle were determined by fitting a cosine curve to the maximum flood, slack before
ebb, maximum ebb, and slack before flood for each current station. Thus, even in the widest part
of a waterway, the tidal current was erroneously shown as being reversing, with the direction of flow
represented as being exactly the same throughout the flood phase and then in the opposite direction
throughout the ebb phase, and with speeds changing like a cosine curve from slack to maximum to
slack. Only after the change over to using the rotary analysis did Tidal Current Charts show rotation
of flow, the distortion due to shallow-water effects, and the effects of the bathymetry and geography.
6.6.2 The Limitations of the Rotary Analysis
The rotary analysis calculates the mean tidal current (speeds and directions at a specific time
interval) for a mean tidal cycle, the average done for the length of the current data time series
available. The main limitation of a rotary analysis is that one cannot see the variation from cycle
to cycle throughout the month (and year). That variation is caused by whatever differences there
are between the current station being analyzed and the reference station used to break up the current
data times series into tidal cycles. If the reference has a very similar tidal harmonic makeup to that
of the station being analyzed, this will not be a problem. Such cycle-to-cycle variations can,
however, be seen with the reversing reduction analysis, and both types of current analysis should
be carried out on every current station.
In a rotary reduction analysis the current velocity time series is divided into tidal cycles, which
are superimposed and vectorially averaged to produce a mean tidal current. To do the vectorial
average all speeds and direction pairs are first converted to a major-minor component pair, such as
north component and east component, or any other orthogonal pair. The resulting mean tidal current
is plotted as both speed and direction curves (and can also be plotted as a polar plot with the
equivalent of a current rose but with the tips of each vector are usually connected) covering one
(mean) tidal cycle..
The division of the data time series into tidal-cycle pieces is usually determined by the times of
a single key point in the tidal current cycle at the reference station, for example, the times of all the
predicted maximum floods at the reference station that fall within the time period of the data (or the
times of the maximum ebbs or of the slacks). However, using two key points, such as using the
times of both maximum flood and maximum ebb, is also an option to be considered (which will be
discussed in the next section). Each tidal current cycle of data is divided up into increments, which
are typically half-hour intervals centered on (e.g.) the time of maximum flood at the reference
station, designated here as MF, and each half hour after that, i.e., MF+ hour, MF+1, MF+1 ,
MF+2, ....., MF+12. (The last interval may not be exactly a half hour, and this may vary from cycle
to cycle.) Every data point in the time series will fall into one of these 25 time intervals. After all
the data points have been assigned to a half-hour interval, an average current speed and direction
is obtained for each of these intervals. The calculation is done vectorially, so first average major
and minor components are obtained and then the averages are converted back into speeds and
directions.
212
At this point a vectorial mean current for the entire mean tidal cycle is obtained by averaging
the 25 half-hour increment values. This mean current, being only for the that particular period of
time, requires careful interpretation (discussed in the next section) as to whether it will be
incorporated into the final values that are turned into time and height difference for inclusion in
Table 2 of the Tidal Current Tables. Next this mean current is subtracted from each of the 25
individual current values covering the tidal cycle, to produce a tidal-only current, which again is
usually plotted.
The next step is to compensate for the fact that the data times series does not usually cover a time
period long enough for all the tidal frequencies to be equally represented. Thus, some type of
correction must be made to the analysis results to make them more closely represent the values that
would be obtained from a full-year analysis (or actually from a full 19-year analysis, that also
eliminates the varying affects of the lunar node). In this step the analysis takes all the speeds of
(e.g.) the reference stations maximum floods during the time period of the data from the subordinate
station and averages them. Then the true 19-year mean maximum flood speed value for the
reference station is divided by this reference station average for the time period of the data to
produce a correction ratio. This ratio is then used to multiply all 25 of the tidal-only speed results,
to give a corrected tidal-only result. Then the mean current is vectorially added back in to give a
final total current. This total current is usually plotted, both on speed and direction plots, and on
a polar plot, the latter being the equivalent of a current rose but with the tips of each vector
connected, so one can easily see how the flow rotates. This polar plot makes the most sense for
interpretation purposes if it is superimposed on a chartlet showing the current station location and
the bathymetry.
At this point the analysis picks off the values from the mean tidal current curves for maximum
flood, maximum ebb, minimum before flood, and minimum before ebb, providing a time, speed
and direction for each. In the case of a double flood or ebb, it will also pick off the time, speed, and
direction for that also.
In NOS the method presently used for the maximums is as follows. For most cases, the
maximum flood and ebb are each determined using a parabolic fit to the 3 highest current speed
values. If 3 or more speed values are within 5 cm/sec (0.1 knot) of the highest value, then the curve
around the maximum value is fairly flat, and in such cases the time of maximum flood (or ebb) is
chosen to be the midpoint of this flat interval. The direction of the maximum flood (or ebb) is the
average of the directions for the speed values used in the previous calculation. The method presently
used for determining the minimums before flood and ebb begins by finding the five current vectors
which vectorially are closest to zero. The minimum current vector is the shortest vector
perpendicular to one of the four lines connecting the tips of these five current vectors. One should
always examine the calculated maximums and minimums and compare them with the linear and
polar plots.
The final step is to determine the Greenwich Intervals for the maximum flood, maximum ebb,
minimum before flood, and minimum before ebb for the analyzed stations. Those values are known
for the reference station and are input into the rotary analysis program. The results of the analysis
give the times of these four points of the tidal current with respect to one of these reference times
at the reference station, so the Greenwich Interval for that point in the tidal current at the reference
station is simply added to the four times from the rotary analysis to produce the Greenwich Intervals
at the subordinate station. If the analysis results are to be used in Table 2 of Tidal Current Tables,
then the time differences to go into Table 2 are obtained by subtracting the Greenwich Intervals of
the reference station from the Greenwich Intervals of the subordinate station.
213
Figure 6.10a. Speed and direction plots of the

total current (tidal current plus mean current)
resulting from a rotary reduction analysis of
current data from Boston Harbor (Station 6 at 10
feet below the surface, May 28 June 1, 1971).
Times are relative to the slack before flood (S) at
the reference station (Deer Island Light).
Figure 6.10b. Polar plot showing the rotation of

the total current (tidal current plus mean flow)
resulting from the same rotary reduction analysis
as in Figure 6.10a. Each hourly current vector
would be an arrow from the center of the plot to
a dot labeled with a time relative to S. Note that
the tidal current rotates clockwise from S+11.5 to
S+4.5, but counterclockwise for the rest of the
tidal cycle.
214
The plotted results of a typical rotary analysis are shown in Figures 6.10a and 6.10b. Figure
6.10a shows the speed and direction plots for total current (i.e., tidal current plus the mean nontidal
flow for the period of the data). In this example the (harmonically) predicted times of slack before
flood (S) at the reference station were used to break up the current data time series into tidal cycles.
Thus the half hour intervals of the analysis go from S+0.0, S+0.5, S+1.0, ......., to S+12.0 hours. For
each of these 25 half hour intervals a speed and direction are plotted. Below the plots are shown the
time (relative to S), speed and direction of the maximum flood and maximum ebb, as well as the
maximum observed current data point for that period of the data, and the mean nontidal current, also
for the period of the data. In the particular plots shown in Figure 6.10a one sees that the ebb phase
lasts much longer than the flood phase, which is due to the mean nontidal current.
Figure 6.10b shows these same results on a polar plot. Here one can get a better idea of the
direction in which the current flows, and how that direction changes over a tidal cycle. This
particular example shows that the current does not necessarily rotate as a simple ellipse. Here the
tip of the current vector actually traces out a figure eight, which is due to the geography (shoreline,
channels, and bathymetry) of the location of the current station. The mean nontidal current has
shifted the center of the figure away from zero.
6.6.4 Special Considerations and Choices To Be Made For the Rotary Analysis
6.6.4a Editing Out Data During the Analysis
Current data can be quite noisy, but the averaging process of the analysis can take care of most
real and reasonable noise. Extreme outliers that do not look real will usually be taken care of in the
quality control part of the data processing procedure prior to being analyzed with the rotary method.
However, within the rotary analysis itself there is the option to edit out data determined to be bad
by some criteria. One of two methods is typically used.
The first method edits out speeds and directions separately. (This method is justified for some
current meters that measure speed and direction separately.) Speeds are thrown out where the speed
is outside some range (typically two standard deviations from the calculated average speed for the
interval). Directions are thrown out in a three-step process, first all directions more than 90o from
the average direction for the interval, then after recalculating the average direction, the process is
repeated for 60o and 45o. If more than 80% of the directions in an interval are thrown out, that
interval is flagged (for narrow current ellipses this often happens near times of slack water since the
direction of flow changes rapidly around that time).
The second editing method removes velocity outliers vectorially in a two-step process. All
current velocity values that are beyond a certain distance vectorially from the average current vector
for an interval are removed. That distance value is initially the absolute value of the average current
speed, and then after recalculating the average current vector the distance is half the absolute value
of the average current speed. This method tends to be used for low signal-to-noise currents. Near
times of slack water it generally throws out less than 80% of the values.
6.6.4b Selecting the Start and Stop Times For the Analysis
It is important to use a whole number of tidal cycles in the analysis, especially for short analyzes,
where the fragment of a tidal cycle could bias the results. One should determine the largest number
of whole tidal cycles that can fit within the time period of the data, and analyze that many cycles,
215
selecting what appears to be the best quality data (leaving data off at the end or at the beginning, or
some off from both the beginning and end.)
6.6.4c Selecting which aspect of the tidal current (maximum flood, maximum ebb, or
slack/minimum) to use to divide up the data record into tidal cycles
The most likely value to be adversely affected by an erroneous rotary analysis (for example,
when one to uses a reference station that is not similar enough in tidal characteristics) are the speeds
of the maximum floods and ebbs. The time shifts in the data for each superimposed tidal cycle do
not significantly hurt the calculation of the minimums, or even the times of the maximums, but they
can make the maximums smaller than they should be. The true average maximum flood or ebb will
come out of the analysis only if all the individual maximums line up in time perfectly with each
other. (This is a problem that does not occur in the reversing reduction, since the actual maximum
flood and ebb values are chosen to be averaged.) Thus, it is usually preferable to use reference times
(to break up the data time series into tidal cycles) that will put all the maximum floods in the same
half-hour interval. This often means that the maximum flood times and/or the maximum ebb times
are best to use as the reference point for the analysis.
If one chooses to use the times of maximum flood and there is some variation in the length of
the tidal cycle, then while the maximum floods (or each tidal cycle) are likely to line up with each
other fairly well, the maximum ebbs will be less likely to line up with each other, and the result will
be a smaller mean maximum ebb speed than should be. Thus, it may actually be better to use both
the maximum flood (MF) times and the maximum ebb (ME) times, and have two reference times
per tidal cycle, to try to minimize the time shifting around the two maximums. In this case one
might have for example, MF-3, MF-2, MF-2, MF-1, MF-1, MF-, MF, MF+, MF+1, MF+1,
MF+2, MF+2, and MF+3, and likewise for maximum ebb. If there ends up being 26 intervals,
then there will likely be some overlap between MF+3 and ME-3, which will have to be worked out.
6.6.4d Selecting the Size of the Time Intervals Used in the Analysis.
The most typical selection of time interval for a rotary analysis is to use 25 half-hour intervals
to cover the tidal cycle. The 25th interval will usually vary in length, often being less than a half-
hour. If two reference points are used per tidal cycle (such as maximum flood and maximum ebb)
then the intervals near the change over from flood to ebb may vary in size. The length of the time
intervals, however, does not have to be a half hour and can be varied as desired, depending on the
sampling interval of the current data time series. However, it should be remembered that the wider
the interval the more the average maximum flood and ebb will be reduced by the analysis (especially
if the rise to the maximum is steep) because slightly smaller values near each actual maximum in
the data will be included in the interval average.
It is very difficult to accurately represent a mixed tidal current with the rotary reduction analysis,
because with a mixed tidal current the length of the duration of the flood phase and the duration of
the ebb phase will vary from tidal cycle to tidal cycle, sometimes significantly. To try to make this
method work for the diurnal situation, one must not use a constant time interval, because for this
situation this will cause some half-hour segments to be in the flood phase during part of the data and
in the ebb phases during other parts of the data. As an alternative scheme, one can instead divide
the duration of flood phase into the same number of interval (and likewise the duration of the ebb
216
phase), no matter what the length in time of the flood (or ebb). Then, each interval will vary in
width, but will always represent the same portion of the flood (or ebb) phase. To break the data up
into flood phases and ebb phases, one can use the slack (minimum) times at the reference station.
However, if the subordinate and reference stations are too far apart, using the reference station slack
times may not break up the subordinate station appropriately. In this case, one might try to break
up the subordinate station time series using its own slack(minimum) times. Whether this will help
better represent the varying diurnal inequality of the mixed tidal current will have to be looked at
closely. It cannot work when a mixed tidal current goes diurnal for part of the month.
6.6.4e Interpreting the mean nontidal current from the analysis
The rotary analysis provides a mean nontidal current value for the time period of the particular
data time series analyzed. The mean current value for another data time series from that location
might be quite different. It depends on what caused the mean flow. In a river or estuary the mean
flow could be due to river flow. River flow varies considerably over the year, from high values
during spring runoff to low values during the summer. The mean flow might also be due to the
wind, in which case, it will vary significantly with time.
There are, however, situations where the geography of the waterway (for example a point jutting
out into the waterway, or a bending channel) causes tidally-induced eddies to occur (sometimes
only during one phase of the current). This results from a nonlinear lateral inertial effect (see
Section 2.3.6). When averaged over a tidal cycle, the result is a mean current. Such mean currents
are fairly permanent, and being tidally induced should be included in the final total current of the
rotary analysis. An example of such tidally induced mean currents is shown in Figure 2.35.
6.7 Selecting a Reference Station
It is important to select a reference station with similar tidal characteristics. While for tide
stations it is usually enough to select a geographically close station, this is not always the case for
tidal currents. The frequency characteristics of tidal currents can change dramatically over a short
horizontal distance, as a result of bathymetry and the shape of the shoreline. Channel bends and
points jutting out into channels can (via nonlinear inertial mechanism, see Section 2.3.6) transfer
diurnal energy into the semidiurnal band and transfer semidiurnal energy into the quarter diurnal
band, e.g., in the latter case, from M2 to M4 (see Figure 2.35), which will make the time differences
and speed ratios between two stations vary considerably throughout the month.
One should also not use a tide station as a reference station for a tidal current analysis without
careful consideration. In many areas the type of tidal current can be quite different than the type of
tide. For example, in the eastern end of the Strait of Juan de Fuca, where it meets Haro Strait the
tide is mixed mainly diurnal and the tidal currents are mixed mainly semidiurnal.
When the tidal characteristics are different, one will see changing differences between, for
example, successive times of maximum flood, which will make the superimposed tidal cycles in a
rotary reduction analysis have a variety of lengths. In that case, the mean maximum flood may still
come out of the analysis with reasonable accuracy, but the mean maximum ebb velocity will
probably come out smaller than it should (because all the individual maxi ebbs were shifted in time
relative to each other). As was discussed, the use of a reference station with tidal characteristic not
similar to those at the subordinate station will show up clearly in a reversing reduction analysis, but
it will not be apparent in a rotary reduction analysis.
217
To try to select the best reference station, carry out the following steps: (1) identify the nearest
reference stations that might be used; if analyzing water level data, choose a reference tide station;
if analyzing current data, choose a reference tidal current station; dont refer a current station to a
tide reference station, or vice versa; (2) compare the harmonic constants at the station to be
analyzed (if available) with the harmonic constants at the candidate reference stations; (3) select
the reference station with the most similar harmonic constant relationships, that is, with ratios of
(K1+O1)/M2, M4/M2, S2/M2, etc. that are most similar, as well as phase lag (epoch) differences,
e.g.,(M2o-K1o-O1o), that are most similar. If there is a reasonably strong diurnal signal, then the
(K1+O1)/M2 ratio should be given precedence, since that ratio will change more quickly with
geographic distance between the two stations. However, M4/M2 ratios for tidal currents can vary
significantly from location to location, and this overtide effect can shift the time of maximum flood
and/or maximum ebb significantly; and (4) if no harmonic constants are available for the station to
be analyzed, initially choose the reference station that is closest geographically (however, if one is
analyzing a current station, the closest reference station geographically may not be the one with the
most similar harmonic constants; if possible, do not choose a reference station that is in a different
channel, or around a bend or on the other side of a point of land).
6.8 Putting a Subordinate Station In a Tide or Tidal Current Table and On a

Tidal Current Chart
Today the two primary uses for a nonharmonic comparison analysis are for the inclusion of a
subordinate station in Table 2 of a Tide or Tidal Current Table (or sometimes in Table 5 of a Tidal
Current Table) and for the construction of Tidal Current Charts. Based on what has been learned
in this chapter, the use of nonharmonic analyses for these purposes is briefly summarized.
Putting a subordinate station in Table 2 of a Tide Table has been fairly routine for decades. For
many water level stations there are many years of data, numerous harmonic analysis results, and
numerous results of monthly mean analyses. Only the limited pages allowable in a Tide Table
prevents most tide stations from becoming (harmonically predicted) reference stations in Table 1.
The standard procedure for putting the mean time and height differences of one of these stations into
Table 2, is merely to use the differences between corresponding monthly mean analyses from the
subordinate station and from the reference station. However, for short tide stations, the tide-by-tide
analysis (usually called Comparison of Simultaneous Observations) has been used to find the
mean time and height differences. This last analysis has typically also shown the time differences
for each tidal cycle (although until recently the height differences for each cycle were not shown)
and so one could see how much they varied from cycle to cycle.
As quality control, to make sure that the subordinate tide station is referred to a reference tide
station with similar enough tidal characteristics, and to calculate the expected error in predictions
made with the mean time and height differences, one should run a tide-by-tide analysis on the
subordinate station. One should make sure to tabulate all the height differences (and not just the
time differences) for each tidal cycle.
Putting a subordinate station in Table 2 of a Tidal Current Table has in the past been done by
either using the reversing reduction analysis or the rotary reduction analysis (see previous sections
for the circumstances under which each was used). In reality both methods should be used, because
each has a unique benefit that one should take advantage of. The rotary analysis will provide entire
mean tidal current curves (a speed curve, a direction curve, and a polar plot) for a mean tidal cycle.
Thus, one will not only have mean values for maximum flood, maximum ebb, slack (minimum)
218
before flood, and slack (minimum) before ebb, but one will able to see the complete rotation of the
tidal current over the whole tidal cycle, plus any nonlinear distortions due to shallow-water or lateral
inertial effects. It will also provide the time differences and speed ratios that will go into Table 2.
For current stations off the coast, where the current ellipse widens out to a circle, and thus where
there are no clear flood or ebb directions, one can put all the hourly (or half-hourly) results from the
rotary reduction analysis into Table 5 of the Tidal Current Tables.
What the rotary reduction analysis does not tell us is how much variation there will be in the
time differences and speed ratios throughout the month (and how similar the tidal characteristics of
the tidal current are at the two stations, subordinate and reference). This information must come
out of the reversing reduction analysis, which is used for quality control purposes. This analysis
tells us what kind of errors one can expect in the tidal current predictions made with the above mean
values put into Table 2. This is especially important for currents, for as was seen, there are many
nonlinear hydrodynamic mechanisms (but especially the lateral inertial effect) that can cause a great
deal of variability in the harmonic makeup of currents stations, even ones that are geographically
close to each other (see Section 2.3.6).
If the variation in the time differences and speed ratios (which equate to the errors one would
see in the predictions using the mean time differences and mean speed ratios) is larger than
acceptable (for a particular application, such a maritime safety), then one may need to establish a
new harmonically produced reference station for this (and other nearby) subordinate current stations.
If that is not possible, one might try to use the results of the reversing reduction analysis to calculate
additional types of mean time difference and mean speed ratios, for example, for times of maximum
lunar declination. Such values might be added to the Endnotes for Table 2. If there are many
subordinate stations in need of those additional time differences and speed ratios, one could even
change the format of the Table 2 page (for those stations) to more efficiently handle it, for example,
adding columns to handle the extra time differences and speed ratios for the maximum declination
situation.
It should be added that sometimes the mean values for maximum floods and ebbs and slacks
(minimums) that come out of the two types of analysis (rotary and reversing) are not exactly the
same. Most likely the mean maximum flood and ebb speeds obtained form the rotary reduction
analysis will be a little smaller than those obtained from the reversing reduction. That is because
the maximum values in the rotary reduction are obtained by averaging over half-hour (or worse,
hour) intervals, so some speeds lower than the actual maximum get included in the calculation. The
difference between the rotary value and the reversing value is usually not great, but when one has
both, one should probably use the values from the reversing reduction.
For constructing a set of Tidal Current Charts for a waterway, the rotary reduction analysis is
the best nonharmonic comparison method to use (such charts can also be produced with numerical
hydrodynamic models, or with tidal current predictions from harmonic constants). Only the rotary
reduction analysis can give the current speed and direction for each hour (or half-hour) of the tidal
cycle. This is discussed in more detail in Section 9.4.
6.9 Minimizing the Errors Due to the Variations Throughout the Month Of
Nonharmonically Determined Time and Height Differences
When producing a Tide or Tidal Current Table the intention is to have enough reference stations
so that no subordinate station is forced to be associated with a reference station whose tidal
characteristics are dissimilar enough that significant variations in time differences and other
219
differences or ratios will occur. However, this may not always be possible. In mixed tidal regimes
the reference stations need to be closer because the diurnal-to-semidiurnal ratios change more
quickly with distance than do the ratios between two semidiurnal constituents. The phase
differences between the K1's and O1's, respectively, at two stations, have a significant effect on the
average (essentially M2) time differences. In addition, shallow-water generated distortions in the
tide or tidal current (the effect of M4 in particular) may shift the times of (e.g.) high water or
maximum flood around during the month, also leading to variations in time differences between two
stations.
Tidal currents have an additional problem. The lateral advective (inertial) terms in the
conservation of momentum equation (see Sections 2.3.6 and 7.3.2), which generate tidally induced
residual currents off points and around channel bends, also shift energy into the next harmonic,
changing the K1/M2 ratio or the M2/M4 ratio in the tidal current. The area around the San Juan
Islands, including the Strait of Juan de Fuca, Haro Strait, and many other connecting waterways,
provides an example (see Figure 2.35). A combination of the large diurnal inequalities and the
inertial effects due to the complex geometry lead to tidal current characteristics that can change
dramatically over a short distance. While the K1/M2 ratio for the tide changes gradually and
smoothly in this area, the K1/M2 ratio for the tidal current varies considerably (see Parker, 1977).
Even though there are six reference stations in the NOS Tidal Current Tables for the Strait of
Juan de Fuca Haro Strait San Juan Islands area, there are times during the month when the tidal
current at some subordinate stations will become diurnal a day earlier or later than at the reference
station, making it almost impossible to make a prediction using a Table 2 correction. Even at times
of the month when both reference station and subordinate station have not become diurnal, there is
significant variation in time differences and speed ratios. For such situations one really needs
several sets of time differences and speed ratios for each tidal parameter in Table 2 for example, one
set for maximum southern declination, one set for equatorial lunar declination, and one set for
maximum northern lunar declination.
There may also be subordinate stations in the Tide or Tidal Current Tables which are associated
with an inappropriate reference station merely because their relationship was not examined closely
enough. It is not enough that two stations are both semidiurnal or that their harmonic constituents
look similar. It is also not enough that two stations are geographically close, especially in the case
of current stations, where two stations right around a channel bend or on opposite sides of a point,
can have different tidal characteristics.
At each subordinate station variation in time differences and height differences throughout the
month should be calculated and compared against an accepted standard.
6.10 Summary Overview:: Steps In Analyzing Water Level or Current Data With
a Nonharmonic Comparison Method
In this section the steps to take in nonharmonically analyzing a water level or current data times
series will be summarized. Sections where more information can be found are in [ ] at the end of
each step. These steps, in bullet form are:
220
(1) identify the nearest reference stations that might be used; if analyzing water level data,
choose a reference tide station; if analyzing current data, choose a reference tidal current
station; dont refer a current station to a tide reference station, or vice versa. [Section 6.7]
(2) compare the harmonic constants at the station to be analyzed (if available) with the harmonic
constants at the candidate reference stations; [Section 6.7]
(3) select the reference station with the most similar harmonic constant relationships, that is,
with ratios of (K1+O1)/M2, M4/M2, S2/M2, etc. that are most similar, as well as phase lag
(epoch) differences, e.g.,(M2o-K1o-O1o), that are most similar. If there is a reasonably
strong diurnal signal, then similarity in the (K1+O1)/M2 ratio should be given precedence,
since that ratio will change more quickly with geographic distance between the two stations.
However, M4/M2 ratios for tidal currents can vary significantly from location to location, and
this overtide effect can shift the time of maximum flood and/or maximum ebb significantly.
(4) if no harmonic constants are available for the station to be analyzed, initially choose the
reference station that is closest geographically; (however, if one is analyzing a current
station, the closest reference station geographically may not be the one with the most similar
harmonic constants; if possible, do not choose a reference station that is in a different
channel, or around a bend or on the other side of a point of land) [Section 6.7]
(5a) for water level data, if one intends to use a mean monthly analysis, one should still also run
at least a month-long tide-by-tide analysis (comparison of simultaneous observations) in
order to assess the size of the variation throughout the month in the time differences between
high waters (and also between low waters) at the two stations, as well as the height
differences; [Section 6.3 and 6.4]
(5b) for current data, if one intends to use a rotary reduction analysis, one should still also run
a reversing reduction analysis in order to assess the size of the variation throughout the
month in the time differences between maximum floods (and also between maximum ebbs,
and between slacks) at the two stations, as well as the speed ratios. [Section 6.5]
(6) if the variation in these differences throughout the month, as determined by a tide-by-tide
analysis of water level data or a reversing reduction analysis of current data, is large
(according to a criterion based on the intended users of the resulting tide or tidal current
predictions) then there are three options with regard to including the subordinate station(s)
in the Tide or Tidal Current Tables: [Section 6.8 and 6.9]
(a) use another reference station that might have a more similar harmonic makeup;
(b) establish a new reference station from a suitable nearby long-term station; or
(c) use the tide-by-tide analysis (for tides) or the reversing reduction analysis (for currents)
to calculate mean time and height differences or speed ratios for special astronomical
conditions (such as maximum lunar declination), and put these in the Endnotes section for
Table 2; or if there are many stations with this problem, reformat a Table 2 page with extra
columns to handle these additional values;
(7a) to get a full picture of the tidal current, one should use both a reversing reduction analysis
and a rotary reduction analysis on the current data series; the rotary analysis will capture the
rotation in the tidal current flow, as well as any distortions in the shape of the tidal current
curve due to shallow-water effects or lateral inertial effects, and the reversing analysis will
capture the variation in time differences and speed ratios throughout the month; both are
important; Table 2 Endnotes may have to be used to indicate such occurrences as double
max floods or ebbs, longer than normal periods of slack water (or minimum flow), flattened
maximums, and/or variations in the time differences and speed ratios over a month etc.; if
221
the current ellipse is very wide, the station might have to be put in Table 5; if the current
data is being analyzed for a Tidal Current Chart, one must always use a rotary reduction
analysis, since values are required for each hour of the tidal current cycle.
(7b) to get the full picture of the tide, one should usually carry out (at least initially) a month-long
tide-by-tide analysis (along with the numerous mean monthly analyses that will routinely be
carried out) to show the variation in time differences and speed ratios throughout the month;
if the water level data is in shallow water, one should also carry out (at least initially) a one-
month mean tide curve analysis (analogous to the rotary analysis in currents) from which
information about the average distortion of the tide can be ascertained (including possible
double high or low waters); the results of these analyses can be included in a Table 2
Endnote for the station. [Section 6.9]
222
7 Interpretation of Tidal Analysis Results Based On
Hydrodynamics
The accurate prediction of the tide or tidal current, though a major objective of tidal analysis,
is not the only objective. Another important goal is to obtain a better understanding of the
hydrodynamics of a waterway. This may in turn allow one to improve the analysis or better
implement a numerical hydrodynamic model, which then ultimately could produce better
predictions. Although tidal analysis results are used to understand the tidal dynamics of a particular
waterway, the opposite can also be true a basic understanding of tidal hydrodynamics in a variety
of waterways can help one better interpret and quality control the tidal analysis results from a newly
surveyed waterway. For example, the calculated harmonic constants should change spatially over
this waterway in a way that makes sense with respect to hydrodynamics.
In this chapter, therefore, the tidal hydrodynamics of a waterway will again be looked at, but this
time in more detail. There may be some redundancy with Section 2.3 and with some parts of other
sections in this book, but those sections were meant to provide just enough hydrodynamic
background to allow one to make the best choices when statistically analyzing the tides and tidal
currents. Now the hydrodynamics will be looked at more thoroughly (and mathematically) in order
to help in the interpretation of the results of the statistical analysis.
In this chapter some simple analytical models will be used to illustrate typical variations in tide
and tidal current characteristics for a variety of conditions and situations. Analytical models, though
representing simple special cases in basin shape and depths (which is the reason the equations of
motion can be solved analytically rather than numerically) have the advantage of being able to
illustrate many cases simply by changing parameters in one or two formulas. To accomplish the
same thing with a numerical hydrodynamic model, one must rerun the model many times. However,
use of a numerical model is unavoidable when dealing with nonlinear effects (such as in shallow
waterways) or dealing with complicated geometries, neither of which can be solved analytically.
7.1 Products for Showing Tidal Analysis Results
Prior to looking at what analytical models can tell one about the tidal hydrodynamics of a
waterway, a few typical ways of graphically displaying tide and tidal current analysis results will
be looked at.
7.1.1 Cotidal, Corange, and Coamplitude Charts
A corange tidal chart is nothing more than a group of contour lines representing different tide
ranges throughout a waterway. Each corange line is a line of constant tide range. These corange
lines can be for the total tide, or they can be for single tidal constituent, such as M2, in which case
223
sometimes the tidal constituent amplitudes (half the constituent ranges) are contoured instead of the
ranges. An example of a corange chart for the M2 tidal constituent was shown in Figure 2.5 for the
Strait of Juan de Fuca - Straight of Georgia based on the harmonic analysis of 95 tide stations along
that waterway (Parker, 1977). Another example of a corange chart, for Chesapeake Bay, was shown
in Figure 2.24.
A cotidal chart is nothing more than a group of contour lines representing different hours (or
phase lags) of high water throughout a waterway. Each cotidal line is a line of constant phase lag
or of constant high water time, relative to the same time meridian (usually the local time meridian,
but sometimes the Greenwich time meridian). These cotidal lines can be for the total tide, or they can
be for single tidal constituent. An example of a cotidal chart for the M2 tide was shown in Figure 2.6.
One more example of cotidal and coamplitude lines is given in Figure 7.1 for Delaware River
and Bay, in this case for M2 again. In this figure the coamplitude (half the tide range) and cophase
lines are put on the same chart, with the actual values at each tide station (the black dots) also given.
Because the upper part of the waterway is so narrow, there are contour lines only in the lower bay.
The phase lags relative to the local time meridian can be converted to hours after moon transit over
that time meridian by dividing by 360o and multiplying by 12.42 hours. In this figure one sees that
it takes 193o or 6.7 hours for high water to move from the bay entrance to the head of tide near
Trenton, NJ. (These M2 results will be referred to in later sections of this chapter.)
When one has harmonic analysis results from a sufficient number of water level stations along
a waterway and one wishes to produce cotidal and corange (or coamplitude) charts for the total tide
and/or for various tidal constituents, one begins by putting the ranges and high water times, or the
constituent amplitude and epoch (i.e., phase lags), on the chart at the exact locations of the water
level stations. Then one must decide how best to draw the contours to represent the entire tide
regime based on these results at these particular locations. The more water level stations one has,
of course, the easier this job is. But even when there are many stations, those stations will be mainly
along both shorelines, as well as on whatever islands may be available, and so the middle of the
waterway is usually not well covered. For a narrow waterway where Coriolis does not have much
of an effect, drawing (the short) contour lines will be straight forward, and one is basically left with
interpolating between neighboring stations to determine where to put the lines. However, for wider
bays one must consider the fact that the contour lines will likely cross the waterway at an angle due
to effect of Coriolis (see Sections 2.3.1 and 7.5). With enough stations (and especially with some
stations in the middle of the waterway, e.g., on islands or structures) one may still be able to obtain
the correct angles. If there are not enough tide gauges on both sides and/or the waterway is long
enough to have a quasinode (a location of minimum tide range where the tide range is larger both
up the waterway and down the waterway from that location), then one may need the help of a
hydrodynamic tide model to help decide how to angle the lines.
A second problem occurs in deeper waterways, where the ranges and time of high water (or the
constituent amplitudes and epochs) may change too slowly over geographic distance. For this
situation the construction of the chart may put demand a level of accuracy from the harmonic
analysis results that may not be attainable if one has to rely on many stations that are less than six
months long. In other words, the results from these short tide stations will (because of their
shortness) have various errors (unless all the stations happened to be installed over the same time
period, which is usually unlikely) that will make it difficult to determine accurate contour lines.
In Section 4.2.2 a reference station approach was used to improve the results from short harmonic
analyses, so that coamplitude and cophase charts could be produced. This method was, in fact, used
to construct the charts in Figures 2.5 and 2.6 (and other charts found in Parker, 1977).
224
7. Interpretation of Tidal Analysis Results Based On Hydrodynamics
The creation of cotidal and corange charts is a form of quality control of the harmonic constants
for the stations in a waterway, because these constants must make sense with respect to each other
over space. If the harmonic constants from a particular station do not seem to make sense relative
Figure 7.1. Coamplitude and cophase chart for the M2 tide constituent for the Delaware River and
Bay, showing the geographic variation in M2 amplitude and epoch (phase lag). Coamplitude lines
are dashed; cophase lines are solid. (From Parker, 1984.)
225
to the harmonic constants at surrounding stations, there either may be a problem with the data from
that station, or the results are actually good and one must figure out the hydrodynamic cause of the
observed variation over space relative to the other stations. These results will help verify
hydrodynamic models (analytical or numerical), and likewise the models will then help us
understand the tidal analysis results.
Later in this chapter it will be seen how a number of simple analytical models, as well as more
complex numerical models, explain how the tide and tidal current are likely to vary over space and
time for a variety of different conditions, such as different basin depths, widths, length, and shapes.
Corange and cotidal charts also have other practical uses. Some form of them (whether hardcopy
or in some electronic form) are normally used for carrying out hydrographic surveys, where the
depth soundings made during the survey must be adjusted so that they are referenced to chart datum,
which in the United States is mean lower low water (MLLW). This will be discussed more in
Section 8.7.2, because tide models now play an important role for this application.
7.1.2 Cospeed and Cophase Charts
Cospeed and cophase charts for tidal currents in a waterway are much more difficult to produce,
and often end up not looking like their tide equivalents. This is because so many factors can affect
the speed amplitude and phase lags of the tidal current constituents that there is often so much
spatial variability among the results from the the current stations that it is impossible to produce
meaningful contour lines. This is especially true for waterways with bends, sudden-width changes,
and variable bathymetry, since the tidal current speeds and phase lags are very sensitive to such
changes. Current sensors are often at different depths, and current speeds and lags are very depth
dependent. Tidal current speeds will be much faster in a narrow entrance (because of continuity)
compared with the speeds outside the entrance (see for example, the very long tidal current ellipses
in Figure 7.3 in the next section).
Current station data time series are also frequently short, on the order of only 15 to 30 days, and
correction using a reference station is often difficult or impossible, because there are few long-
enough current stations, and those that are long enough may not have similar enough tidal
characteristics. Thus, one will often see just the values at the specific current stations put on the
charts, with no contour lines. For example, Figure 7.2 shows the cophase chart for the M2 tidal
current in the Strait of Juan de Fuca Strait of Georgia. Only in the western end of the Strait of Juan
de Fuca were cophase line attempted. In the rest of the waterway the phase lags (epochs) from the
harmonic analyses were so variable in space that no such lines could be attempted. This variation
is primarily due to the effect of the lateral inertial nonlinear terms described in Sections 2.3.6 and
5.1.
7.1.3 Tidal Current Ellipses
Tidal current ellipses were described in Section 5.2 with a sample result for several tidal current
constituents from a single current station showed in Figure 5.4. Placing many such ellipses on a
chart of the waterway at the locations of the current meters can provide insights into the dynamics
of the tidal currents in that waterway. Such an example is provided in Figure 7.3, which shows
numerous M2 ellipses (from Parker, 1977) for the eastern end of the Strait of Juan de Fuca, Haro and
226
Figure 7.2. Chart with epochs (phase lags) for maximum flood for the M2 tidal currents in the
Strait of Juan de Fuca Strait of Georgia, given in degrees of phase lag relative to the local time
meridian (120oW).
Rosario Straits, and the southern end of the Strait of Georgia. All the ellipses are drawn to the same
scale, so longer ellipses indicate higher M2 maximum flood and ebb speeds. For each ellipse,
connecting the center dot to the dot for a particular hour indicates the speed and direction of the tidal
constituent for that hour of the constituent cycle. Zero hour indicates the lunar transit over the local
time meridian (in this case 120oW). For ellipses that are wide enough for the rotation of the tidal
current to be clear, one can see whether that rotation is clockwise or counterclockwise by looking
at the hour numbers. The hour number at the end of the major axis of the ellipse that comes after
the zero hour is the hour of maximum flood current (relative to 120oW).
Most the current data analyzed to produce this chart were measured at 70 feet below MLLW
(because the current meters were mounted from bottom-mounted tautline buoys). Looking at the
many ellipses shown in this figure, one can see a great deal of variability in their length, width, time
of maximum flood, and direction of rotation. There are very long narrow ellipses in narrow and
channels and in small entrances. The M2 flood-ebb orientation tends to follow the geography or the
bathymetry. Wider ellipses tend to be in wider parts of the waterway, but not always. The wide
227
Figure 7.3. Numerous M2 tidal current constituent ellipses for the eastern end of the Strait of
Juan de Fuca, the southern end of the Strait of Georgia, and around and between the San Juan
Islands. See key on chart, and the text. For the flow for a particular hour of the M2 cycle,
draw a vector from the center of the ellipse to the dot with the number of the hour of the cycle.
Note the different directions of rotation of the flow and the different degrees of narrowness of
the ellipses.
228
ellipse just south of Victoria, BC is probably affected by the lateral inertial effects of the channel
bending north into Haro Strait (see Section 2.3.6 and 5.1). The timing of each M2 maximum flood
can vary quite a bit from station to station, due to differing water depths, as well as due to the lateral
inertial effects of channel bends, points sticking out into the waterway, and other geographic effects.
The majority of the ellipses show a clockwise rotation, but there are also some that are
counterclockwise (lateral inertial effects again probably playing a role in this, but bottom friction
may also play a role). Some of this variation may be the result of the short data time series (from
different time periods) harmonically analyzed to produced the ellipses (most of the current stations
analyzed only had 15 to 29 days of data). The possible causes of the variability in tidal currents will
be described in the following sections of this chapter (see also Section 2.3.6).
7.1.4 Various Types of Graphs
For wider waterways, corange and cotidal charts created from harmonic analysis results may be
most useful for visualization of the tide regime over the waterway and for comparison to numerical
models results, but for a narrower waterway a simple plot of (e.g.) the tide ranges along the length
of the waterway may be just as useful. For example, Figure 7.4 shows the variation in tide
amplitude (half the tide range) for the five largest semidiurnal constituents in Delaware River and
Bay M2, N2, S2, 2MN2, and 2MS2. The M2 values are the same as those on the coamplitude chart
in Figure 7.1 (and the values for the other four constituents can be found on corange charts in Parker,
1984). Clearly M2 is the dominant constituent, being much larger than the other four semidiurnal
constituents. One also sees that in the lower bay (which is wider) the tidal amplitude is higher on
the right side of the waterway (looking up the estuary) due to the Coriolis effect.
More useful is Figure 7.5, which shows the amplification of these same five constituents (i.e.,
for each constituents each amplitude value is divided by the amplitude value at the entrance). Here
M2 has an amplification of 1.7 at Trenton, but N2 and S2 have smaller amplifications (approximately
1.3 and 1.4, respectively) and 2MN2, and 2MS2 have much greater amplifications (5 and 6,
respectively). Other plots from Parker (1984) show K1 with even less amplification than N2 and S2,
and M4 and M6 with even larger amplifications than 2MN2, and 2MS2. The reasons for these
differences in amplification will be explained in Section 7.6.2. The shallow-water constituents
2MN2, and 2MS2 have the same frequencies as the astronomical constituents L2 and 2, and the
original harmonic analyses may have had L2 and 2 listed as the constituents being solved for, but
these amplification plots would have clearly indicated that 2MN2 and 2MS2 were the actual
constituents being calculated by the analysis.
There are, of course, many other graphs of various statistical analysis outputs that can be useful,
including frequency domain plots of the harmonic constants, various spectra, EOF analysis results,
and graphs from various nonharmonic analyses (such as the rotary and reversing reductions).
7.2 Spatial Variation in Tidal and Tidal Current Parameters In a Waterway Due
to Hydrodynamic Processes
Throughout physical oceanography (and geophysics in general) data analysis and dynamical
modeling go hand in hand and complement each other. A model is a mathematical representation
of the particular laws of physics that are believed to govern the specific dynamics with which we
are concerned. The data (and the data analysis results) are used in that model to see how well that
229
3.5 M2
Right side
3.0 of Bay
2.5
Narrow River
Section
2.0 Left side

of Bay
1.5
1.0
N2
0.5 S2
RIGHT SIDE
LEFT SID
E
RIGHT
SIDE
LEFT SIDE
2MN2
2MS2 LEFT SIDE
RIGHT SIDE
LEFT SID
E
0.0 FIELDSBORO BILLINGSPORT DEEPWATER FORTESCUE LEWES

CAPE
TRENTON PHILADELPHIA MARCUS ARTIFICIAL ABIDWELL
MAY
HOOK ISLAND CREEK
FERRY
TERMINAL
Figure 7.4. Variation in the amplitude (half range) of the five largest
semidiurnal constituents in the Delaware River and Bay, from the ocean
entrance on the right to the head of tide on the left. (From Parker, 1984.)
model can describe that particular dynamic phenomenon and how well it can make predictions.
Once a model has been validated using the data, it can in turn be used to better understand the data.
It can provide values at locations where there are no data. And it can provide better insights with
regard to how the data should vary over space and time. Models can provide synoptic pictures of
tide regimes that can help us understand our data analysis results. It can, in fact, provide some
quality assurance with regard to the results of our harmonic or nonharmonic tidal analyses, for
example, giving one an idea whether harmonic constants at a particular station look reasonable
relative to harmonic constants found at other nearby stations. Hydrodynamics can help one answer
questions like, how do we expect the M2 amplitude or epoch to change along a particular waterway,
or how do we expect the M4/M2 ratio or the K1/O1 ratio to change with distance?
Hydrodynamic models can provide insights about the how the tide regime varies in different
types of waterways. In this chapter many of these models will have been greatly simplified by using
an idealized geometry or by ignoring a particular aspect of the physics (such as the nonlinear terms),
but the features they demonstrate usually are a good first approximation and for some (no too
shallow) waterways, they can even fairly accurately represent the tide. The advantage of these
230
Figure 7.5. The amplification of the five largest semidiurnal constituents in the Delaware
River and Bay. M2, N2, and S2 are astronomical constituents, while 2MN2 and 2MS2 are
compound constituents generated by nonlinear mechanisms in the shallow water (see text).
(From Parker, 1991a.)
simple models is that they can be solved analytically, with the resulting solution being one or more
formulas, which can then be used to directly provide information about the tide or tidal current over
a whole waterway for a whole host of situations. Incorporating the more complete physics requires
fully nonlinear numerical solutions, often with high-resolution grids. The analytical models allow
us to easily see how the tide or tidal current changes for different water depths, frictional dissipation,
etc. by simply changing a few parameters in the formulas. The numerical models, while much more
accurate, require rerunning the model many times to be able to summarize the different cases.
One way to interpret the results of a tidal harmonic analysis is to see whether those results make
sense within a synoptic description of the tidal hydrodynamics of the waterway from which the
analyzed data came. Even if one is analyzing data from a particular waterway for the first time, and
expects to describe the tidal hydrodynamic system in that waterway for the first time (assuming one
has enough stations to do so), there will be certain expectations (based on the physics of the
waterway) of how the various tidal harmonic constituents should vary in horizontal (geographic)
space, and in vertical space (i.e., with depth) when looking at tidal currents. However, the use of
such information, primarily from tidal hydrodynamic models, can only be carried so far from a
quality control point of view, since the data analysis results are usually meant to validate a model,
rather than the other way around. Models can be of all varieties and levels of resolution and
accuracy, and thus may not provide the detail or accuracy to finely quality control the analysis
results, especially if one is interested in tidal current analysis results (knowing how quickly such
results can change with distance; see Section 2.3.6). But as an approximate indicator of quality
231
control, this tidal hydrodynamic understanding from models can point out large errors in analysis
results. Simply from the standpoint of better understanding what a waterway can do to the tide wave
that enters it from the ocean, it is important to understand what such models can tell us. The
following sections of this chapter are a more detailed extension of what was described in Section
2.3, but with reference to the equations of motion (on which the various models are based, to
different degrees of approximation), which will be discussed next.
7.3 The Equations of Motion As the Basis for Tidal Hydrodynamic Models
Tidal models are mathematical representations of the physics of the waterway that affect the tide
and tidal current (which are two manifestations of the same phenomenon, one vertical one
horizontal). They make use of conservation equations that are usually referred to as the equations
of motion. These equations are derived based on the assumptions that mass (or volume) and
momentum must be conserved, as well as energy in some cases, with additional equations of state
if one includes salinity and water temperature effects. The various tidal models that will be looked
at in this chapter will each involve some kind of simplification of these equations of motion (such
as ignoring certain nonlinear terms), and their analytical solution will usually be made possible by
using some kind of idealized geometry for the waterway (e.g., a rectangular basin). Numerical
solutions, however, will also be used in Section 7.6 when looking at nonlinear effects.
Derivations of the basic original equations of motion (before being simplified for particular
cases) will not be given here, but some key information will be provided about where the equations
came from, as well as references where one can go to see the derivations. These are partial
differential equations. If the reader lacks experience with advanced calculus it will hopefully not
be too much of a liability because the physical meaning of these equations will be described. The
final outcome, at least for the analytical solutions, will be fairly straightforward formulas.
7.3.1 The Equation of Continuity
The conservation of mass is described by the following equation:
M M(u) M(v) M(w) . (7.1)

+ + + = 0
Mt Mx My Mz
where t is time, x, y, and z are the three orthogonal dimensions in space, is the water density (mass
per unit volume; a scalar quantity) and u, v, and w are the three orthogonal components of the
current speed (a vector quantity). [For those not familiar with partial differential equations, one can
simply think of the M/Mt as a change in density over a small time interval. Similarly M(u)/Mx can
be thought of as a change in mass over a small distance in the x direction.]
The visualization typically used in the derivation of this equation is to picture a cubic box (as
a control volume) with water flowing through it, such as in Figure 7.6. The second term in equation
(7.1) is the difference between the mass flowing into the box (in the x-direction) through one face
of the box and the mass flowing out of the box through the opposite face (and likewise for the y and
z directions). The first term in equation (7.1), the change in density with time, will go up if more
mass enters the cube than leaves the cube, and vice versa.
Here, for simplicity, one assumes that the density remains constant everywhere and for all time.
232
Figure 7.6. A cubic control volume used in deriving

the conservation of mass equation. (From Daily and
Harleman, 1966.).
Then equation (7.1) reduces to the basic continuity equation (which now really represents a
conservation of volume):
Mu Mv Mw
+ + = 0 . (7.2)
Mx My Mz
(There is actually a less drastic assumption that will also lead to equation (7.2), i.e., the assumption
that water is incompressible. This is approximately true for water, but is not true for air and so
cannot be used when the fluid dynamic equations are applied to the atmosphere.)
7.3.2 The Momentum Equations
The conversation of momentum equation comes from Newtons second law, F = ma, i.e., force
(F) equals mass (m) times acceleration (a). Momentum is mass times velocity, so ma is actually
the rate of change of momentum. To obtain a momentum equation useful in a tidal model,
Newtons second law is put into a different form, where the rate of change of momentum is balanced
with the various forces that may be acting on the water (such as gravity and friction).
To visualize the derivation of this equation one can again use a cube as a control volume, and
look at how various changing forces on a cube (similar to the one in Figure 7.6) will affect the rate
of change of momentum of the water moving through the cube. There are pressure forces pressing
in on the cube due to the affect of gravity (i.e., the pressure due to the weight of the water above it)
as well as shear stress forces rubbing on the faces of the cube (due to such things as the affect of
bottom friction, or the wind stress from the water surface, both transmitted through the water, from
water particle to water particle). One usually deals again with density () instead of mass, and one
makes another simplifying assumption in order to more easily handle in the equations. The so-
called Boussinesq Approximation says that one can ignore small changes in density except as it
affects the weight/pressure of the water. This lets one divide all the terms in the momentum
equation by , so that the acceleration is being balanced by changing forces divided by mass.
At this point, in order to jump quickly to a momentum equation usable for our purposes (and
skipping many steps, which can be found in the references), use will be made of several geometric
assumptions dealing with the length scales of the waterway and the length scale of the tide wave in
233
the waterway. If one does a scaling analysis of the equations of motion, and some of these length
scales are disparate from each other, then one will see that certain terms will be much smaller than
others and so can be left out of the equations, greatly simplifying them, and making them easier to
solve.
For example, since the tide wave is a very long wave, one can make the assumption that its wave
length () is much greater than the water depth (D). When this is the case (>>D) the vertical
acceleration (of moving water particles) will be negligibly small and the third of the three
momentum equations will be greatly simplified (see below). As a result of this assumption, the
change in pressure in the vertical direction will be due solely to the effect of gravity on the water
(i.e., the deeper one goes the greater the pressure will be due to the weight of the water above). This
is usually referred as either the long-wave approximation or the shallow-water approximation or the
hydrostatic approximation. Other length-scale based simplifications will be seen in later sections.
Another length-scale assumption doesnt eliminate any of the terms, but does have another
benefit. One assumes that the length of our waterway is much smaller than the radius of the Earth,
and this allows the use of Cartesian coordinates instead of (the much more cumbersome) spherical
coordinates.
And finally (to skip one more step in the derivation of the equations below) a shift is made from
a fixed reference frame (relative to a distant star in outer space) to a reference frame that rotates with
the rotating Earth. The result of this transformation of the equations to a rotating reference frame
adds terms to the equations that represent an apparent force, i.e., the Coriolis force. The Coriolis
force appears to push moving objects to the right in the Northern Hemisphere and to the left in the
Southern Hemisphere (relative to a fixed reference frame, however, those objects would still move
straight and the Earth would turn under it; see Parker, 1998). The slower the motion of the object
and the longer distance it goes, the more it pushed to the right (or left). The Coriolis force increases
from zero at the equator to a maximum near the north or south pole. However, for simplicity, it is
assumed that the waterway is small enough that one can ignore the variation in the Coriolis force
with latitude. This is the so-called f-plane approximation (where f = 2sin and is the latitude
and 2 is the planet vorticity of the Earth, i.e., its rotation), and f is assumed to be constant. (For
hydrodynamic models covering a large area of the ocean, one instead makes the so-called beta-plane
approximation, which keeps the latitudinal variation of the Coriolis force, but says that over short
distances one can assume it to be linear.)
Thus, the momentum equations become (in three dimensions):
Mu + u Mu + v Mu + w Mu f v = 1 Mp (frictional forces)x
Mt Mx My Mz Mx
Mv + u Mv + v Mv + w Mv + f u = 1 Mp (frictional forces)y (7.3)

Mt Mx Mz My
0 = 1 Mp g
Mz
where p = pressure and g = the acceleration due to gravity, and the x and y components of the
frictional forces have been left vague for the moment.
So far in these equations, while the currents have been clearly indicated (u, v, w), there has yet
been no mention of the elevation of the water level surface. This will come next, with the help of
the third equation (which is in a greatly simplified form since vertical acceleration was neglected
234
as a result of the long-wave approximation). This third equation is called the hydrostatic equation,
and if one integrates it over depth (i.e., with respect to z) and uses the boundary condition that the
pressure at the water surface is equal to the atmospheric pressure, pa one obtains
p = g ( z) + pa (7.4)
where is the water level elevation. From this equation one can see that when z = , namely,
when one is at the water surface, the pressure equals atmospheric pressure, but as one goes deeper
the pressure increases linearly (in this equation, z=0 at mean sea level and z is positive above mean
sea level and negative below it). By neglecting the horizontal variation in pa and differentiating
equation (7.4) with respect to x and y, the pressure terms in equation (7.3) can be replaced with
terms that have the water level elevation, , namely :
1 Mp = g M and 1 Mp = g M .
Mx Mx My My
The frictional forces are usually treated by handling the turbulence (that friction causes) with
the so-called Reynolds equations. The typical approach is to break up each velocity component into
a so-called a mean portion and a rapidly fluctuating portion (the latter due to the turbulence caused
by frictional effects, such as the current flowing along a rough sea bottom). The so-called mean
portion is a mean only in the sense of short-term averaging and is really the current that is measured
and modeled. One ends up (again skipping the steps in the derivation) with Reynolds stress terms,
often indicated as x and y. At this point in our discussion the first two momentum equations now
are:
Mu + u Mu + v Mu + w Mw f v = g M + 1 Mx
Mt Mx My Mz Mx Mz
(7.5)
Mv + u Mv + v Mv + w Mv + f u = g M + 1 My
Mt Mx My Mz My Mz
There has been much research on the best ways to represent the energy loss due to the friction
terms on the extreme right ends of these two equations (which again must be skipped over here).
The type of approach usually depends on whether one stays with the full three-dimensional
equations (in which case some type of a so-called turbulence closure scheme would be used) or go
down to two dimensions (e.g., by integrating the equations over the depth) or go down to one
dimension (by integrating over the depth and the width). As will be seen below, the integration over
depth (with boundary conditions at the bottom and water surface) ends up giving us some type of
parameter for the bottom friction and another for the surface friction (which would be due to the
wind blowing on the surface). The most frequently used representation for bottom friction is a
quadratic law such as b= cf u*u* , where cf is a friction coefficient that is determined empirically.
(The absolute sign * * is necessary to keep the frictional stress in the same direction as the current
flow, but still with a quadratic character, i.e., still like u2.)
So at this point one has a continuity equation (7.2) and two momentum equations (7.5), the later
having the water level elevation () as a variable (as a result of using the hydrostatic equation),
along with the currents, whose three components are u, v, w. (The water level elevation will also
235
come into the continuity equation once it is integrated over depth.) In each component equation of
(7.5) there is a balance between the change in current speed with time and space (i.e., the
acceleration ) and three forces: (1) the Coriolis force (actually an apparent force due to being in a
rotating reference frame); (2) the pressure forces of the weight of the water (now represented by
elevation times the acceleration due to gravity, because of the hydrostatic equation); and (3) the
frictional forces that dissipate energy. The spatial change in the current speed (terms two, three, and
four), called the inertial terms, are nonlinear since the velocity values multiply each other. These
nonlinear terms will have to be neglected in order to find simple analytical solutions, but in Section
7.6 numerical solutions will be used to deal with them because of their importance in producing
overtides, compound tides, and nonlinear interactions with nontidal phenomena like storm surge and
river flow. The representations for the friction terms will also end up being nonlinear, but in the
analytical models friction will be linearized, not neglected, since it is too important. Nonlinear terms
will still come into play in the continuity equation when it is integrated over depth. Scaling
arguments will be used in order to determine when such simplifications are allowable. For
example, one will be able to neglect the inertial terms when the depth of the waterway is much
greater than the amplitude of the tide wave and the waterways is straight.
In the above discussion many mathematical steps have been skipped over to get from the most
general and most comprehensive equations of motion down to much simpler equations. Many
assumptions have been made (and there is more simplification still to come). For the derivation of
these equations, and a much more thorough discussion of them, there are many texts in fluid
dynamics, geophysical fluid dynamics, and physical oceanography to which one can go. Only a few
are mentioned here, and even in these there are differences in their mathematical complexity. On
the less complex side is the physical oceanography text by Knauss (1978). The more engineering
oriented fluid dynamics text by Daily and Harleman (1966) provides a clear derivation of the
equations of motion, as do the fluid mechanics texts by Yih (1969) and Batchelor (1967). A more
comprehensive (and more complex) discussion can be found in the geophysical fluid dynamics text
by Pedlosky (1987). There are, of course, many others from which to choose.
7.3.3 Simplifying the Equations of Motion For Idealized Cases
In order to obtain some useful easy-to-use analytical tide models one must further simplify the
continuity equation (7.2) and the momentum equations (7.5), again by using scaling arguments that
take advantage of disparate length scales.
First, the very simplest case will be looked at, that is, the case of a long narrow straight
waterway, where one expects there to be no appreciable cross-channel motion. Here another
length-scale assumption is made, namely, that the length of the waterway (L) is much greater than
the width (B), i.e., L >> B, which is called the narrow basin approximation. Adding this assumption
to the previous assumptions that have been made about length scales, they can be summarized by
the following inequalities:
tidal length of width of depth of

> >> >>
wavelength waterway waterway waterway
236
As a result of the narrow basin assumption the second (v) momentum equation in (7.5) reduces to:
f u = g M
My
For this case, due to Coriolis, the water level elevation on one side of the waterway will be higher
than the elevation on the other side of the waterway, and this will depend on the current speed u.
The length scale associated with the effect of Coriolis is the Rossby radius of deformation. The
constant-density version, usually called RO, is defined as

RO = (gD) = c = L
f f f
where c is the propagation speed of the crest (high water) of the tide wave, which for frictionless
situations is (gD) , and is the tidal frequency. In order to be able to neglect the effect of Coriolis,
the width of the waterway must be narrow enough such that B/Ro << 1 , or equivalently B/L <<
/f . This assumption will be made now (but in Section 7.5 the case where the waterway is wide
enough for Coriolis to be important will be treated). [However, one still needs to retain the
nonlinear inertial terms, in which case the condition for being able to neglect Coriolis, but keep the
nonlinear terms is B/Ro << N/D or equivalently B/L << (/f)(N/D), where N is the scale of the
amplitude of the tide wave. For more information on a complete scaling analysis, see Parker
(1984).]
To make the continuity equation and the remaining momentum equation truly one-dimensional,
one must integrate over the cross-sectional area of the waterway. The cross-sectional area (A) is
assumed to be rectangular with A=b(h+), where the width b and the depth h are constants, and
is the tidal amplitude, which must be still included since the nonlinear terms have not been neglected
(i.e., one has not yet said that N/D << 1).
For the continuity equation, many steps will be again skipped over (which make use of
Leibnitzs Rule several times; see Parker, 1984), and again the surface boundary condition is used
to bring in the water level elevation, . Finally, one ends up with
M + 1 1 (b (h+) u) = 0 (7.6)
Mt b Mx
Similarly integrating the first momentum equation in (7.5) over the cross-sectional area one ends
up with
Mu + u Mu = g M 1 c u*u*
f (7.7)
Mt Mx Mx h+
where the cf u*u* is the classic representation of the bottom friction, and surface friction has been
neglected by assuming there will be no wind.
Equations (7.6) and (7.7) will be looked at more closely in Section 7.6, where the effect of the
various nonlinear terms will be looked at, including the generation of overtides and compound tides,
and the nonlinear interaction of the tide with nontidal phenomena such as storm surge and river flow.
But for now one assumes that the depth of the waterway is much greater than the amplitude of
the tide (N<<D), so that all the nonlinear terms can be neglected except friction. There are two
nonlinear aspects to friction term in equation (7.7), the (h+) in the denominator, in which the will
be neglected, and the quadratic part, i.e., the u*u* (which is like u2 except the use of the absolute
value allows the energy loss to be in the direction of flow). It is important to keep some term for
237
the energy dissipation, so the quadratic friction is linearized and the energy loss is made proportional
to simply u (with N as a frictional coefficient, which will be determined empirically). The linearized
continuity and momentum equations, for which one will be able to find analytical solutions (for
simple geometries) then become
M + 1 1 (bhu) = 0
Mt b Mx
(7.8)
Mu = g M N u
Mt Mx
7.4 Tidal Variations Along a Very Narrow Waterway
7.4.1 Tidal Variations In a Narrow Rectangular Waterway
In this section the simplest possible analytical tidal model is looked at, namely, a model
describing the tide and tidal current in a long very narrow shallow waterway, and assuming that the
tidal amplitude is much smaller than the depth. In terms of length scales one is assuming that
tidal length of width of depth of tidal

> >> >> >>
wavelength waterway waterway waterway amplitude
With the assumption of a small tidal amplitude relative to the water depth one can then neglect all
the nonlinear terms except friction. Friction is linearized, that is, it is assumed that energy loss is
directly proportional to the current speed (rather than to u*u*). One also assumes that the waterway
is rectangular with a constant width b and depth h, in which case equations (7.8), become:
M = h Mu
Mt Mx
(7.8a)
Mu = g M N u
Mt Mx
where N is some type of a frictional coefficient, which will be determined empirically.

A simple solution of these two equations (see Redfield, 1950; Ippen and Harleman, 1966; Parker,
1984) is an exponentially damped progressive wave (see Figure 2.21) whose water level elevation
is described by the formula
i = ao cos (t kx) ex (7.9)
where is the frequency of the progressive wave, k is the wave number (the inverse of the
wavelength , i.e., k=1/), is a frictional damping coefficient, and ao is the tidal amplitude at the
x=0 (x=0 is at the closed end of the waterway). The frictional term ex wears down the wave as
it moves up the waterway. If one labels the frictionless values for wave number, wavelength, and
238
wave speed (also called celerity) as ko , o , and co , one effect of friction can be seen from the
formula
ko = = c = 1
k o co [ 1 + (/ko)2 ]
From this formula once sees that friction shortens the tidal wavelength and slows up the tide wave.
As friction increases (i.e., as increases), the tidal wavelength decreases, as does the propagation
speed of the tide wave. Since friction increases as the water depth decreases, one will expect to see
shorter tidal wavelengths in shallower waterways. ( is related to N coefficient originally used in
the momentum equation (7.8) by the formula N = 2kco/ko)
For this same exponentially damped progressive wave, the current velocity can be obtained by
putting (7.9) in the continuity equation, the result being the formula
ui = a o 1
2 2
cos (t kx + ) ex (7.10)
h ( +k )
where = tan1 (/k) is the phase difference between the time of high water and the time of
maximum flood current. For no friction (=0) high water and maximum flood occur at the same
time. As friction increases (as the depth decreases) the time difference between high water and
maximum flood current increases. However, this only applies for a damped progressive wave,
which primarily exist in essentially open-ended waterways like rivers where there is no reflection
of the tide wave and the progressive wave moves up the waterway until friction wears it down to
nothing.
To look at the case where the waterway has a closed end (see Figure 7.7 for schematic and
coordinate system), one simply adds a reflected damped progressive wave to the incident
progressive damped wave, with the boundary condition that the incident wave must equal the
reflected wave at the closed end, where the reflection takes place. Thus,
= i + r = ao cos (t kx) ex + ao cos (t + kx) e x (7.11)
Figure 7.7. Schematic of waterway with a Cartesian

coordinate system.
239
where i = r at x=0. To find the time and height of high water, one takes the first derivative of
(7.11) and sets it equal to zero, solves for the time of high water, and puts it back into (7.11).
M = a sin (t kx) ex a sin (t + kx) e x = 0

o o
Mt
The ao drops out, and expanding the sine and cosine terms and rearranging, one has
0 = sint coskx (e x + ex) + cost sinkx (e x ex) .
Dividing through by cost coskx one has
sint sinkx (e x ex)

=
cost coskx (e x + ex)
which is
tan t = tankx tanh x
where tanhx is the hyperbolic tangent. The time of high water is given by the formula:
tH = tan1 (-tankx tanhx) (7.12)
where tH is the time of high water, and tH is the phase of high water (both are relative to the time
or phase of high water at the reflection point at the closed end of the waterway). Putting (7.12) into
(7.11) one obtains the following formula for the high water level elevation
H = 2ao [cost coskx coshx sint sinkx sinhx ] (7.13)
where cosh and sinh are hyperbolic sine and cosine, and 2ao is the high water elevation at the closed
end . (7.12) and (7.13) are the equations for a frictionally damped standing wave (see Figure 2.22).
tH is plotted versus kx in Figure 7.8 for various values of the damping coefficient . In this
figure, kx represents distance from the closed end (x is distance in portions of a tidal wavelength).
In Figure 7.8, kx equals 0o at the left end of the horizontal axis (representing the closed of the
waterway) and becomes more negative as one moves to the right toward the mouth of the waterway
(which can be placed anywhere along the horizontal axis). From this figure one can see that for the
case of very little friction (e.g., =0.5), such as in a deeper waterway, one has almost a pure standing
wave, where high water occurs at approximately the same time everywhere on one side of kx=90o,
and occurs a half cycle (180o) later on the other side of kx=90o. Here kx=90o is one quarter of a tidal
wavelength from the closed end. When it is high water to the left of kx=90o it is low water to the
right of it (and vice versa); see Figure 2.22. For much higher frictional damping (e.g., = 8.0, such
as in a very shallow waterway) one has almost a pure damped progressive wave, where the time of
high water get progressively later as one moves up the waterway. (For this last case, it looks
basically like the incident damped progressive wave, because the reflected wave is worn down so
much that it has no effect except very close to the point of reflection.) For both these cases, and all
the cases in between, the phase of high water is always the same at kx= 90o (but this location
changes with changing because the wavelength changes with , and thus k changes with ).
240
Figure 7.8. The phase (time) of high water, tH, versus

distance from the closed end (extreme left) of the
waterway in terms of wavenumber phase, kx, for
different values of the frictional damping coefficient .
See text. x is distance in portions of a tidal
wavelength. kx= 90o is one quarter of a tidal
wavelength from the closed end. (Very similar to
Figure 1 in Redfield, 1950.)
One can use some trigonometric angle relationships to simplify (7.13) and obtain the formula
for the amplitude of the tide (i.e., half the tide range) at high water (H) relative to the high water
amplitude at the closed end [o = H(0) = 2ao], i.e.,
H (7.14)
= [ (cos 2kx + cosh 2x) ]
o
a result obtained by Fjeldstad (1929), Redfield (1950), and Ippen and Harleman (1966). This
amplitude ratio (which is the same as the ratio of the tide ranges) is plotted in Figure 7.9 for various
values of the damping coefficient , again plotted relative to the closed end of the waterway at the
left end of the figure. Again, for very little frictional damping (i.e., =0.5, such as in a deeper
waterway) one sees an almost standing wave with an almost zero tide range at kx = 90o, that
location being a quasinode (or a node, for the frictionless situation). For a (frictionless) standing
wave and frictional cases where does not get too large, the largest tide range occurs at the closed
end of the waterway. For very large (e.g., =8.0, such as for a very shallow waterway) one sees
that the smallest tide ranges are at the closed end of the waterway and one again has almost a pure
damped progressive wave that decreases in tide range as it moves up the waterway. (For this last
case, it again looks basically like the incident damped progressive wave, because the reflected wave
is worn down so much that it has no effect except very close to the point of reflection.)
241
Figure 7.9. The amplitude ratio (=tide range ratio)

plotted versus distance from the closed end
(extreme left) of the waterway in terms of
wavenumber phase, kx, for several values of the
frictional coefficient, . (See text and Figure 7.8.)
(Very similar to Figure 2 in Redfield, 1950.)
From Figure 7.9 one can also see how the length of waterway affects the amplification of the
tide. If one has a waterway whose length is, for example, one quarter of a tidal wavelength long,
then in Figure 7.9 one would mark the entrance at kx = -90o on the horizontal axis. One would then
pick the curve for a particular frictional situation (lower for deeper waterways; higher for
shallower). The amplification for the tide would be the closed end value (1.0) divided by the value
at kx = -90o for that curve. Thus one sees very large amplification for small . For =0 (no
friction) there is infinite amplification at kx = -90o, which is called resonance. That never happens
in the real world (because of friction) but very large amplifications do occur in bays with lengths
that are near a quarter tidal wavelength (e.g., the Bay of Fundy). For waterways that are not as long,
the amplification is not as great (for the small case) .
Figure 7.9 is a convenient way to put many cases on the same plot, because every case can be
plotted relative to the closed end, but such things as amplification may be easier to understand if
values are plotted relative to the entrance. Figure 7.10 shows such a set of plots, for one frictional
damping value (=1.5, typical for many bays) and for six different waterway lengths (here the
entrance to the waterway is now on the left). This gives a better idea of how the length of the
waterway can affect the amplification of the tide. The greatest amplification occurs at a length of
kx=90o (1/4 of a tidal wavelength), as one would expect from the discussion in the last paragraph.
Smaller amplifications result when the length of the waterway is 60o (1/6 of a tidal wavelength) or
30o (1/12 of a tidal wavelength). For a length of 180o (1/2 a tidal wavelength) there is no
amplification, the tide range actually decreasing halfway up the waterway, then increasing again but
not getting back to the value at the entrance. For a length of 120o (1/3 a tidal wavelength) the tide
range at first decreases but then increases and eventually is larger than the value at the entrance.
242
Figure 7.10. The tide range ratio (relative to the

value at the entrance to the waterway on the left)
for one frictional damping coefficient (=1.5) for
six different length waterways. A 90o waterway
length is of a tidal wavelength. A 60o length is a
of a tidal wavelength, etc. (Very similar to Figure
15 in Redfield, 1950.)
Figure 7.11. The phase (time) difference between

high water and maximum flood versus distance form
the closed end of the waterway for different values of
the frictional coefficient, . (Very similar to Figure 5
in Redfield, 1950.)
243
To look at the tidal currents in our waterway with a closed end, one simply adds the formulas
for the tidal current of the incident and reflected damped progressive waves, i.e.,
u = ui + u r = ao 2 1 2 [cos (t kx + ) ex + cos (t + kx + ) e x] (7.15)

h ( +k )
(where again = tan1 (/k) is the phase difference between high water and maximum flood in the
incident damped progress wave). The phase of maximum flood is obtained by setting the derivative
of this equation equal to zero, ending up with
tM = tan1 (-tankx cothx) . (7.16)
Figure 7.11 shows plots of the difference between times of high water and maximum flood
(tH tM). Between the closed end and 1/4 of a tidal wavelength downstream (i.e., kx= 90o, the
location of the quasinode) the maximum flood current occurs before high water, but further
downstream from kx= 90o it occurs after high water. For the almost standing wave case (=0.5)
maximum flood occurs approximately 1/4 of tidal period before high water (-90o) when the tide is
at mean tide level (which means that slack water occurs at high water and at low water). Beyond
this location it occurs 1/4 of tidal period after high water (+90o) when the tide again is at mean tide
level heading for low water). For a high frictional case (=8.0, almost a pure damped progressive
wave) the times of maximum flood and high water get closer and closer together, especially as the
waterway becomes longer. Here again, it looks basically like the incident damped progressive wave,
because the reflected wave is worn down so much that it has no effect except very close to the point
Figure 7.12. Maximum flood speed versus distance

from the closed end of the waterway for various
values of the frictional damping coefficient, . (See
text.)
244
To obtain the current speed value for the maximum flood, (7.16) is substituted into a simplified
form of (7.15), again for various values of , from which is obtained
2ao 1
uM = 2 2 [ (cosh 2x cos 2kx) ] (7.17)
h ( +k )
The maximum flood speed is plotted in Figure 7.12 for various frictional damping values, . (In this
figure, the speed value is divided by 2ao/h.) For all values of the current speed equals zero at the
closed end (x=0), as it must. For small to moderate frictional damping, the highest values of
maximum flood speed occur at 1/4 of a tidal wavelength downstream (i.e., kx= 90o), the location
where the tide range value is the smallest. Again for the highest frictional values (e.g., =8.0, an
almost a pure damped progressive wave) the maximum flood speeds decrease all the way up the
waterway (as do the tide range values).
One can also use these simple analytical formulas to look at the relationship between the
semidiurnal and diurnal tide waves with changing frictional dissipation and different length
waterways. For example, the K1/M2 tide range ratio, relative to the K1/M2 tide range ratio at the
closed end of the waterway, can be expressed as
K1 ]
M2 = [ (cos 2kx/1.927 + cosh 2x/1.927) (7.18)
[ (cos 2kx + cosh 2x) ]
Figure 7.13. The K1/M2 ratio (relative to the K1/M2

ratio at the closed end of the waterway) plotted
versus distance from the closed end, for several
values of the frictional damping coefficient, . kx=
90o is of a tidal wavelength from the closed
end. (Very similar to Figure 16 in Redfield, 1950.)
245
This is plotted in Figure 7.13, where the horizontal axis is in terms of the M2 phase number change
(kx). This figure must be viewed remembering that the 1.0 value at the closed end can be misleading
if one forgets that all these plotted ratios are relative to the K1/M2 ratio at the closed end. That being
said, one can see from this figure that for low values, the K1/M2 ratio tends to be largest near 1/4
of an M2 tidal wavelength from the closed end of the waterway (i.e., kx= 90o). This is the location
of the M2 quasinode, but the K1 quasinode is near kx= 180o; the M2 tide wave has reached its
minimum but the K1 tide wave has not.
The wavelengths of the M2 and K1 constituent waves depend on the depth of the waterway. Thus
the changing tide regime moving along a waterway depends on: (1) the depth of the waterway; (2)
the length of the waterway; and (3) the K1/M2 ratio at the entrance to the waterway. Figure 7.14
shows six K1/M2 ratio curves plotted relative to the K1/M2 ratio at the entrance, for one frictional
damping value (=1.5, typical for many bays) and six different waterway lengths (in the figure the
entrance to the waterway is on the left). In this figure one can see that if the length of the waterway
is less than 80o of phase number length, then the K1/M2 ratio will decrease from the entrance up the
waterway. For longer waterways one sees that the K1/M2 ratio increases as one moves up the
waterways from the entrance. (This figure must be viewed remembering that all plotted ratios are
relative to the K1/M2 at the entrance, and the K1/M2 value shown at the entrance is 1.0. Whatever
the actual K1/M2 ratio is at the entrance should therefore multiply the values on this figure. If the
tide is, for example, very semidiurnal at the entrance (e.g., the entrance value of K1/M2 is 0.25), then
even after the amplification shown on the plot the tide might still be semidiurnal.)
[A caution here on the interpretation of model results when dealing with more than one tidal
frequency. Although here we have acted as though there are two separate long waves, one for M2
and one for K1 , the tide-producing forces actually act on the water in a waterway to produce only
one long wave, but that one wave changes shape with time because it has many frequency
components. Breaking that one wave into waves representing its frequency components is a useful
Figure 7.14. K1/M2 tide range ratio (relative

to the ratio at the entrance, on the left) versus
distance from the entrance, for six different
length waterways, for a value of =1.5 for the
frictional damping. (See text.)
246
way to represent it statistically (as was seen in using harmonic analysis and prediction) and it also
seems to work fine when using linear hydrodynamic models, where those component waves do not
interact and can simply be added to recreate that one long wave that changes shape with time.
However, when the nonlinear terms have been included, there will be energy transfer to new
frequencies (and to some already existing frequencies), so that the waves representing those
components will be changing amplitude and phase, not only because of the effects of the basin (just
looked at) but because of the energy they will be receiving from (or giving up to) other component
waves through the nonlinear interactions. Numerical models, on the other hand, act in the time
domain, time stepping forward with all hydrodynamic effects acting at the same time on the water.
This is much closer to reality, but then one usually still ends up looking at the results in the
frequency domain to better understand what has happened.]
Although the relationships obtained in this section were for very narrow situations, one also
finds that they match the values found at the center of wider waterways where the Coriolis effect has
been included. The case including Coriolis effects is treated in Section 7.5.1, but first the effect of
changing widths on the tide will be investigated.
7.4.2 Tidal Variations In a Narrowing Waterway
If there is a change in width as one moves up a waterway, that will also affect the tide and tidal
current. Generally if the waterway becomes narrower the tide range increases. If the decrease in
width can be approximately described with an exponential function (as some natural waterways can,
especially rivers) then one can find a simple analytical solution to the equations of motion that will
provide insights into how changing width can affect the tide regime in such a waterway.
Starting with the version of the continuity and momentum equations in equation (7.8)
M + 1 1 (bhu) = 0
Mt b Mx
(7.8)
Mu = g M N u
Mt Mx
Neglecting nonlinear terms, linearizing friction, assuming constant depth h and an exponential width
decrease described by b=boe x/R , where R is the length scale associated with the decreasing width
(see Figure 7.7), one obtains
M + h 1 Mu = 0
Mt 2R Mx
(7.19)
Mu = g M N u
Mt Mx
Solutions to these two equations will be of the form (see Parker, 1984)
(x) % ex/2R eix ex

2 2 2 2
where = k [ (1+ N 4 ) + ] and = k [ (1+ N ) 4
] and
k k
247
where k = [1 1 2 ] is the wave number for the frictionless case [assuming > 1/(2R), in
(2R)
which case the frictionless solution is oscillatory; otherwise the frictionless solution is monotonic].
If > 1/(2R) , then is the wave number and is the damping factor. As the rate of width-decrease
increases (i.e., as R decreases) the wave number decreases and the wavelength increases as does the
wave propagation speed. Thus, a decreasing width has the opposite effect of increasing friction
(since the latter decreases the wavelength and the wave propagation speed). Thus, as a tide wave
propagates up a shallow narrowing river, friction will be wearing it down and the decreasing width
will be building it up.
From the solution of equations (7.19) one obtains two messy looking formulas, but at least an
analytical solution is possible, and using them one can see a limitless number of cases by simply
changing parameters in these formulas. The formula for the phase of high water (relative to the
phase of high water at the closed end of the waterway) is:
x x
tH = tan1 ( ex sec2 e x sec1 tan(x+) ) tan1 ( sec2 sec1 tan ) (7.20)
e sec2 + e sec1 sec2 + sec1
and the formula for the high water elevation relative to the high water elevation at the closed end
of the waterway (which equals the tide range relative to the tide range at the closed end) is:
(x) = 1 2 ex/2R [ e2x sec22 + e 2x sec21 + 2 sec1 sec2 cos2(x + )] (7.21)

2 2
(0) 2 ( + )
where (to make it a little less messy) the following parameters were defined
= (1 2) , 1 = tan1 ( 1/(2R) + ) , and 2 = tan1 ( 1/(2R) + ).

The width change in the upper 70 percent of the Delaware estuary can be described fairly well
with an exponential formula using a value of R=0.136. Using that value, formula (7.20) is plotted
in Figure 7.15a for eight different values of the frictional coefficient N. The asterisks (*) on the plot
represent the center-waterway M2 high water phase lag (relative to the value at the closed end) for
nine locations along the estuary, from actual harmonically analyzed water level data. They fall
nicely along the N=8 curve, showing that this simple analytical model can represent the M2 phase
change reasonably well. (For a constant-width model, there would have been no N curves along
which these data values would have fallen.)
Similarly, in Figure 7.15b formula (7.21) is plotted for eight different values of the frictional
coefficient N again for a width decrease represented by R=0.136. The asterisks (*) on the plot this
time represent the center-waterway M2 tidal amplitudes (relative to the value at the closed end) for
the nine locations along the estuary. Again they fall nicely along the N=8 curve (and again for a
constant-width model there would have been no N curves along which these data values would have
fallen.). The same proved to be true for the K1 amplitudes and phases, only they fell nicely along
the N=10 curve (see Parker, 1984). (As will be seen in Section 7.6, the larger value of N for K1 is
due to the nonlinear effect of the larger M2 constituent increasing the frictional momentum loss from
the smaller K1 constituent.)
248
Figure 7.15a. The phase (time) of M2 high water (relative to

the value at the closed end, on the left) for eight different values
of the frictional coefficient N in an bay with an exponential
width decrease (R=0.136) matching Delaware River. The *s are
M2 phase lags from harmonically analyzed water level data in
the Delaware River (see text). (From Parker, 1984.)
Figure 15b. The M2 tide range (relative to the value at the closed
end of the waterway, on the left) versus distance from the closed
end (in portions of a tidal wavelength) for eight different values of
the frictional coefficient N in a bay with an exponential width
decrease (R=0.136) matching that in the Delaware River. The *s
indicate the M2 tide range (relative to the value at the closed end)
from harmonically analyzed water level data from stations along
the Delaware River. (From Parker, 1984.)
249
Figure 7.16. Tide range (relative to the range at the closed end, on the
left) for five different exponential width decreases, for the frictionless
case (N=0). R=0.136 represents the width decrease of the Delaware
River. Dashed line are for monotonic solutions. Solid-dotted curves
are for oscillatory solutions, which represent standing waves (since this
is a frictionless case). (See text.)
Figure 7.16 shows how different rates of width-decreases (i.e., different values of R ) can affect
the M2 tidal amplitude along the waterway. To isolate the width effect one looks at the case for no
friction (N=0). The frictionless case is interesting in that it actually has two types of solutions, one
oscillatory [if >1/(2R)] and one monotonic [if <1/(2R)]. For the M2 tidal frequency the solution
goes monotonic for R<0.08 (represented by dashed lines in the figure). Each oscillatory solution is
represented by a solid line up to a node (zero tide range) and then by a dotted line (where the phase
is now a half cycle out of phase with the solid curve portion). For R=10 the width is essentially
constant, and the node occurs at 1/4 of a constant-width frictionless M2 tidal wavelength from the
closed end. As the rate of width-decrease increases (i.e., R decreases) one sees that the wavelength
increases and the node moves further away from the closed end. (This again is the opposite effect
of friction, which shortens the tide wave length.) By the time R decreases to 0.08, the wavelength
has become infinite and the water level oscillation is in phase everywhere in the waterway (the
monotonic solution).
For the R=0.136 value in the Delaware estuary, the frequency at which the frictionless solution
goes monotonic (the so-called cutoff frequency) happens to fall between the frequency of the K1 and
O1, with O1 falling below it. In Section 2.3.4 we saw that in the Delaware estuary the calculated
lunar nodal factor curve for O1 very closely matched the astronomically determined curve, while the
K1 somewhat matched, and the curves for the semidiurnal and higher order constituents did not
match at all. It was mentioned that the exponential width decrease probably inhibited the tidal
hydrodynamics for the diurnal constituents and the nonlinear interactions that would have modified
the astronomically-based lunar nodal factor curves.
In this section an exponential width change has been used to illustrate the effects of width
changes on the tide, but also see Prandle and Rahman (1980) and Prandle (1991) for the effects on
the tide when the width varies according to a power law. The classic paper on the effects of width
250
(and depth) changes is by Green (1837) in which he showed that the wave amplitude is inversely
proportional to the square root of the width of the basin (and inversely proportional to the fourth root
of the depth of the basin), but that relationship ignores friction and assumes slowly changing width
and/or depth. Jay (1991) looked at cases with more rapid changes in width and/or depth.
For very sudden changes in width and/or depth, the tide wave is partially reflected and partially
transmitted. A sudden reduction in width (looking up a bay) leads to a partial reflection-transmission
that amplifies the tide wave, leading to larger tide ranges at the head of the bay. A sudden widening
of the bay (again, looking up the bay) has the opposite effect, reducing the tide range at the head of
the bay. One can analytically model the tide for this type of situation by using transmission and
reflection wave coefficients similar to those used by Eagleson and Dean (1966, pp 71-7).
Some large tide ranges are actually due to several successive sudden reductions in width moving
up a waterway with partial reflection-transmissions at each location, which amplify the tide wave.
This contributes to the 38.4 mean tide range (43.5 mean spring range; with ranges sometimes
reaching 50 feet) at the head of Minas Basin at the northeastern end of the Bay of Fundy. There is
an amplification from the partial reflection-transmission at the transition from the continental shelf
to the Gulf of Maine-Bay of Fundy, and then again at the transition from the Bay of Fundy into
Minas Basin, before being further amplified by the total reflection at the head of Minas Basin. In
each case the length of that portion of the waterway is conducive to amplification (as explained in
Section 7.4.1), each portion of the waterway falling somewhere among the first three curves of
Figure 7.10, with the Bay of Fundy portion covering almost 90o of wave number phase (kx), namely,
almost of an M2 tidal wavelength. One can, in fact, see these three regions in Figure 9 from
Redfield (1950) (reproduced along with Redfields Figure 10 in Figure 7.17 below), which shows
the ratio of the tide range for stations along the Bay of Fundy, Gulf of Maine, and part of the
continental shelf (relative to the tide range at the head of the Bay of Fundy) versus the time (phase)
of high water at those stations (relative to the time of high water at the head of the Bay). Stations
16 through 21 are along the Atlantic coast of Nova Scotia and represent some of the amplification
on the continental shelf. Stations 24 and 23 are in Minas Basin and 22 is at the end of Cumberland
Basin, a little to the west of Minas Basin. The stations along the northwest coast of the Bay of
Fundy are shown lying approximately along a curve representing a frictional coefficient of =1.0,
while those along the southeast coast lie somewhere between =1.5 and =2.0. However, this
supposed difference in frictional dissipation is not real, being a result of Redfield not including the
Coriolis force in his analytical model but still using coastal tide stations, when center-bay values
would have been more appropriate. He also did not include the slow decrease in width going up the
Bay of Fundy, which if included would have probably resulted in a higher value.
7.5 Tidal Variations In Wider Waterways the Effect of Coriolis
If the waterway is not narrow enough that so that the following equalities are true, B/Ro<<1 ,
or equivalently B/L<</f (see Section 7.3.3), where Ro is the constant-density version of the Rossby
radius of deformation, i.e.

RO = (gD) = c = L
f f f
251
Figure 7.17. (Top panel) The tide range for stations along the Bay of Fundy,
Gulf of Maine, and part of the Atlantic Coast along Nova Scotia (relative to the
tide range at the head of the Minas Basin at the northeastern end of the Bay of
Fundy) plotted versus the time (phase) of high water at those same stations
(relative to high water time at the head of Minas Basin). See text. (Bottom panel)
Map showing water level station locations. (From Redfield, 1950, with some
labels added.)
252
then one must include the effect of the Coriolis force (f = 2 sin; see Section 7.3.2). The equations
of motions for this case (neglecting the nonlinear terms and linearizing the friction term) are:
M = h Mu
Mt Mx
(7.22)
Mu = g M N u
Mt Mx
f u = g M
My
which is the same as (7.8a) except for the cross-bay momentum equation (the 3rd equation above).
The first solution for this case was probably Fjeldstad (1929); the solution for the frictionless case
was given by Taylor (1920) (see also Hendershott and Speranza, 1971). The solution is still an
exponentially damped progressive wave but now there is a term representing the cross-bay variation
in water level due to the Coriolis force. The water level elevation for a damped progressive wave
under the influence of Coriolis is described by the formula
i = ao cos (t kx y) e x e k y (7.23)
where is the frequency of the progressive wave, k is the wave number (the inverse of the
wavelength , i.e., k=1/), is a frictional damping coefficient, co is the frictionless wave speed
(gD), ao is the tidal amplitude at the x=0, y=0 (located at the center of the closed end of the
waterway), and is defined by
= f = f k2 2 .
co2 (k2+2) k2 + 2
The wave described by formula (7.23) is called a frictionally damped Kelvin wave (the frictionless
case was first looked at by Lord Kelvin). From the last formula it can be seen that as the frictional
damping () increases the Coriolis effect on the water level elevation decreases, so even if a
waterway is wide the effect of Coriolis may be small if the waterway is very shallow. As in Section
7.4.1, to find the elevation in a closed waterway, one adds incident and reflected damped progressive
Kelvin waves, i.e.,
= i + r = ao cos (t kx y) e x e k y + ao cos (t + kx + y) e x e+ k y
and simplifies to obtain
= 2ao [cost cos(kx+ y) cosh(x + y) sinnt sin(kx+ y) sinh(x+k y)] (7.24)
By setting the derivative of this equation to zero one obtains the phase of high water
tH = tan1 [-tan(kx + y) tanh(x + k y)] (7.25)
which when substituted into (7.24) gives the high water elevation (relative to the high water
elevation at the center of the closed end) as
253
H (7.26)
= [ {cos 2(kx + y) + cosh 2(x + k y)} ]
o
This amplitude ratio is the same as the tide range relative to the tide range at the center of the closed
end. For this case, however, the reflection isnt perfect. i does not equal r at the closed end except
at y=0, nor is the tidal current equal to zero at the closed end (again) except at y=0. To make this
reflection of Kelvin waves perfect, one must add additional waves [of the type known as Poincar
waves, as Taylor (1920) first did with the frictionless case]. However, the effect of these added
waves is small for a narrow waterway and also disappears as one moves away from the closed end.
The only difference between equation (7.26) above and equation (7.14) in Section 7.4.1 for the
non Coriolis case, is that kx is replaced by (kx + y) and x is replaced by (x + k y). The same
is true for the equation for the tidal current. One will also notice that at y=0, i.e., down the center
of the waterway, these equations reduce to those found in Section 7.4.1. Thus, those equations in
Section 7.4.1 can still be used to represent the tide and tidal current down the center of a waterway,
even if that waterway is wide enough for Coriolis to have an effect.
Figure 7.18 shows the cotidal and corange lines in the closed waterway for different values of
the frictional damping coefficient . In each case the width is 1/10th of a frictionless tidal
wavelength and the length is a frictionless tidal wavelength. For the frictionless case (=0, i.e.,
a very deep waterway) there is a node, i.e., a point of zero tide range, 1/4 of a tidal wavelength from
the closed end, in the center of the waterway. As friction increases ( increases) this node moves
to the right (when looking down the waterway), until for the =1.0 case it moves onto land and
becomes a virtual node (for this particular width).
The cotidal and corange lines in many waterways approximately match those seen in one of the
examples in Figure 7.18. In Figures 2.5 and 2.6, even though the Strait of Juan de Fuca -Strait of
Georgia waterway bends around, one can see the similarity in the M2 cotidal and corange lines with
those for the case =1.0 in Figure 7.18, including a virtual node at the southeastern tip of Vancouver
Island. Likewise, in Figure 2.24 one can see a virtual node on the western shore of Chesapeake Bay
near South Point. In corange charts for the open ocean (which are so deep that =0 is not a bad
approximation) one does see nodes, which are usually called amphidromic points (see Section 7.8).
[Tahiti happens to be located near an M2 amphidromic point in the Pacific Ocean, allowing S2 to
be of comparable size, and therefore leading to the very unusual situation of high water there
occurring at approximately the same time every day (see Marmer, 1926).]
In a wide waterway, there are other mechanisms besides Coriolis that can affect the tide or tidal
current, including the effects of changing bathymetry and the nonlinear effects of the lateral inertial
terms (see Section 7.6.7 ).
7.6 Nonlinear Shallow-water Effects on Tides
7.6.1 The Nonlinear Terms In the Equations of Motion
Returning now to the nonlinear one-dimensional continuity and momentum equations from
Section 7.3.3, equations (7.6) and (7.7) are
M + 1 M (b (h+) u) = 0 (7.27)
Mt b Mx
254
Figure 7.18. Cotidal and corange lines in a semi-enclosed

rectangular waterway (the closed end on the right) caused by the
Coriolis force, for four different values of the frictional damping
coefficient (including the frictionless case (top panel).
and
f (7.28)
Mt Mx Mx h+
where is the tidal elevation, u is the cross-sectionally averaged tidal current velocity, b is the width
of the basin, h is the width-averaged depth of the basin, and cf is the friction coefficient (which is
usually determined empirically for a particular waterway, but is often on the order of 0.0025).
There are three nonlinear terms in these two equations:
(1) u Mu , the inertial (advective) term in the momentum equation;
Mx
(2) M(bu), in the continuity equation (where one lets b=constant, for simplicity); and
Mx
(3) 1 c u*u*, the friction term in the momentum equation
f
h+
255
The first two terms were the terms considered in the classical works on shallow-water tides. The
third term, the friction term, actually has two nonlinear aspects, the quadratic part u*u* and the effect
of in the denominator. The effect in the denominator of the friction term is the effect of
elevation on the frictional momentum loss per unit volume of fluid. The energy dissipation due to
bottom friction has less of an effect if spread over a deeper depth than over a shallower depth. When
the tidal elevation is comparable in size to the water depth there is good deal less energy loss per
unit volume of fluid under the crest (high water) than under the trough (low water) of the tide wave.
These two frictional effects can be separated into two separate terms. First (3) is put into the form
1 1 cf u*u* = 1 (1+/h)1 cf u*u*

h (1+/h) h
and then a binomial expansion of (1+/h)1 is carried out , namely,
(1+/h)1 = 1 /h + (/h)2 - (/h)3 + ......
Then, assuming <<h and neglecting terms with (/h)2 and higher order, the friction term can be
written as
1 c u*u* 1 2 c u*u*
h f h f
The u*u* is the elevation effect just discussed, and it will be seen below that its effects are similar
to the effects of terms (1) and (2) above, all three having asymmetric effects, while the effect of
u*u* is a symmetric effect. These nonlinear terms transfer energy from one tidal frequency to
another, and also cause the interaction between the tide and nontidal phenomena such as storm surge
and river flow.
Because of these nonlinear terms the equations of motion, (7.27) and (7.28), cannot be solved
analytically, and so there cannot be any formulas comparable to those in Sections 7.4 for telling how
(for example) high waters or maximum floods will be affected by different conditions. These
equations can be solved using numerical methods, but in order to obtain some idea of the effect of
these terms one must run a numerical model many times a very time consuming process.
However, there is one alternative method, which will provide at least an approximate idea of the
effects of each of these nonlinear terms, and that is to carry out a Fourier decomposition of equations
(7.27) and (7.28).
7.6.2 The Generation of Compound Tides and Overtides
A Fourier decomposition of the equations of motion will first be used to determine which
nonlinear mechanisms generate which compound tides and overtides. First the total tidal current
is represented as the sum of two astronomical tidal constituents, i.e.:
u(x,t) = u1(x) cos(t 1(x)) + u1'(x) cos(t 1'(x))
where , are the frequencies of the two tidal current constituents, and u1, u1', and 1, 1' are the
tidal current constituent amplitudes and phase lags. To simplify this for use in later calculations let
r = , i.e., the difference in frequency between the two constituents. Then rewrite u(x,t) as
256
u = u1 cos(n r t 1) + u1' cos([n+] r t 1') (7.29)
where = + 1 and n is an integer. Also let n = (n r t 1), again to make later calculations a little
simpler to keep track of. Then rewrite u(x,t) as
u = u1 cos n + u1' cos([n+] [1' n+ 1]) (7.30)

n
Similarly for the total tidal elevation:
(x,t) = 1(x) cos(t 1(x)) + 1'(x) cos(t 1'(x))
again using r = and n = (n r t 1) and = +1 ,one has:
= 1 cos (n [11]) + 1' cos([n+] [1' n+ 1]) (7.31)

n
One also makes the assumption that u1' << u1 and 1' << 1 , which is true for semidiurnal tide
regime, since in such an area M2 is much larger than the other tidal constituents. So for
convenience the larger constituent will be called M2.
The inertial term then becomes:
2 n+ 2
u Mu = 1 M(u ) = 1 M ( [u1 cos n + u1' cos([n+] [1' n 1])] )
Mx 2 Mx 2 Mx
n+
= 1 M ( u1 + u1u1' cos ( [1' n 1]) + u1 cos 2n
2 2
2 Mx (7.32)
+ u1u1' cos([2n+] [1' n+ 1'])
n
The first term inside the ( ) on the right-hand side of (7.32), involving u12, is a mean flow generated
by the inertial terms effect on M2 , namely, the so-called tidally induced residual current. The third
term is the second harmonic of M2, namely, the overtide M4. (The higher amplification of M4
mentioned in Section 7.2 results from this generation of M4 along the length of the waterway.) The
second and fourth terms are new compound tides generated by the interaction between M2 and the
second constituent. If this second constituent is N2 then the second term is a low-frequency shallow-
water tidal constituent, MN, which happens to have the same frequency as the astronomical
constituent Mn. The fourth term is a higher-frequency shallow-water tidal constituent, MN4. (The
inertial term can also produce M6, but this results from an interaction between M2 and M4, and thus
is a smaller second-order effect. As will be seen below, M6 is primarily generated by the quadratic
friction term from M2 alone.)
The nonlinear term from the continuity equation becomes:
M (u) = M ( { cos(n[ ]) + ' cos([n+] [ ' + n+ ] ) }

Mx Mx
1 1 1 1 1 n 1
{ [u1 cos n + u1' cos([n+] [1' n+

n 1]) } )
= 1 M ( 1 u1 cos [11] + 1 u1' cos(+[11'+1])
2 Mx
+ 1'u1 cos( [1' n+
n 1]) + 1u1 cos(2n [11])
257
+ 1u1' cos([2n+] [11' 2n+

n 1])
+ 1'u1 cos([2n+] [1'1' n+
n 1]) ) (7.33)
The same new tidal constituents are generated as by the inertial term, but the phase lag
relationships are different, and the amplitudes are also different. In situations where one expects
slow currents this nonlinear continuity term will be larger than the inertial term. One will notice that
this term also produces a shift in mean sea level (the first term to the right of the second equal sign).
Thus, if the tide range of a waterway changes for some reason, then one will also see a change in
the mean sea level of the waterway (especially near the closed end) due to this nonlinear effect.
The two above derivations were quite straight forward, but for the two friction terms, u*u* and
u*u*, a complete Fourier decomposition of each term must be done. u*u* can be represented as
a Fourier series, i.e.,
4
u*u* = a o + 3 [am cos m + bm sin m ] (7.34)
2 n
To find am one multiplies equation (7.34) by cos m and integrates over all possible frequencies, and
to find bm one does the same but instead multiplies the equation by sin m, i.e.:
2
am = 1 I (u1 cos(n r t 1) + u1' cos([n+] r t 1'))
o
* (u1 cos(n r t 1) + u1' cos([n+] r t 1'))* cos m d
2
bm = 1 I (u1 cos(n r t 1) + u1' cos([n+] r t 1'))
o
*(u1 cos(n r t 1) + u1' cos([n+] r t 1')) * sin m d ,
The tedious steps in carrying out this Fourier expansion can be found in Parker, 1984 (see also
Dronkers, 1964; LeProvost, 1976; and Fang, 1987). Here only the resulting values for am and bm will
be given. First one notes that ao = 0, so no mean flow is generated by the effect of u*u* for this case.
For the larger tidal constituent (M2), with frequency n, one obtains
an = 8 u12 , which represents the first-order momentum loss due to friction. One also gets
3
a3n = 8 u12 , with frequency 3n, which represents the generation of a third harmonic of M2,
15
namely, the second overtide, M6. (The higher amplification of M6 mentioned in Section 7.2 results
from its nonlinear generation along the length of the waterway.)
If the second smaller astronomical constituent has a frequency of (n-1) then one has
a(n-1) = 8 u u ' cos [ ' n1 '] and b(n-1) = 8 u u ' sin [ ' n1 ']
3 1 1 1
n 1 3 1 1 1
n 1
which represents its first-order momentum loss from the smaller tidal constituent due to friction, but
this loss is greater because of the presence of the larger tidal constituent, i.e., the u1u1' is larger than
(u1')2. (In Figure 7.5 this is why the amplifications of S2 and N2 are smaller than amplification of
M2.)
258
The interaction of the two astronomical tidal constituents also produces a new compound tidal
constituent, with frequency (n+1), whose formulas for a(n+1) and b(n+1) look the same as those just
shown for a(n 1) and b(n 1) except with the n1 replaced with n+1. One can see that if the smaller
astronomical tidal constituent is semidiurnal then the compound tidal constituent will also be
semidiurnal. If the smaller astronomical constituent is N2 then the generated compound tidal
constituent will be 2MN2 . If the smaller astronomical constituent is S2 then the generated compount
constituent will be 2SM2. (2MN2 and 2MS2 have the same frequencies as the two astronomical
constituents L2 and 2. The large amplifications of 2MN2 and 2MS2 seen in Figure 7.5 are because
they are nonlinearly generated along the length of the waterway. Such amplifications are much
larger than would be the amplifications of L2 and 2, which would be comparable to the
amplifications of N2 and S2 seen in that figure.) There are also high frequency compound tidal
constituents generated that are close in frequency to M6 (for example, 2MN6 and 2NM6).
For the u*u* friction term one must also do a complete Fourier decomposition, where u*u*
is represented as a Fourier series, i.e.,
4
u*u* = co + 3 [cm cos m + dm sin m ] (7.35)
2 n
and the coefficients cm and dm are found in a similar manner (see Parker, 1984) with even more terms
involved. Again only the results are summarized. This time there is a mean effect, since for co we
get a nonzero value of
co = 4 u 2 cos( )
2 3 1 1 1 1
where (11) is the phase difference between high water and maximum flood. One also obtains
values for c2n and d2n , which also generate the second harmonic, M4 , but with a different amplitude
and phase lag contribution (than those from the inertial or nonlinear continuity terms), i.e.,
c2n = 8 u 2 cos( ) and d2n = 8 u 2 sin( )

15 1 1 1 1
15 1 1 1 1
In addition the same compound tidal constituents (both low frequency and higher frequency) are
generated as by the inertial term and the nonlinear continuity term, but again with different
amplitude and phase lag contributions. (For c1 and d1 , and the other compound coefficients, see
Parker, 1984.)
Table 2.4 (presented in Chapter 2) summarizes the nonlinear interactions between M2 and four
other astronomical tidal constituents (N2, S2, K1, O1), as well as with itself, and shows what
compound and overtide tidal constituents are generated by these interactions, and which nonlinear
terms can be responsible. There is a distinct difference between the symmetric nonlinear effect of
quadratic friction, u*u*, and the asymmetric nonlinear effect of the nonlinear continuity term,
M(u)/Mx, the inertial term uMu/Mx, and the elevation effect on friction, u*u*. The semidiurnal
compound tides (such as 2MN2 and 2MS2) are only generated by u*u*. The asymmetric nonlinear
terms generate low-frequency compound tides, such as MN and MS, which have the same
frequencies as the astronomical constituents Mm and MSf (and which are often dominated anyway
by meteorological noise). They also generate overtides (e.g., M4, S4, etc.) and quarter-diurnal
compound tides (e.g., MN4 and MS4). Terdiurnal constituents can be generated by all the nonlinear
terms (i.e., by both symmetric and asymmetric mechanisms), through the interaction of M2 with
either K1 or O1. For example, MK3 is generated by one or more asymmetric nonlinear mechanisms,
259
while 2MO3 is generated by the symmetric quadratic friction mechanism; both have the same
frequency (but different node factors). In some research papers one will see either MK3 or 2MO3
listed, when in fact, both are really present; and likewise for MO3 and 2MK3.
The symmetric quadratic frictional interaction between M2 with either K1 or O1 also produces
fifth and seventh diurnal constituents. There are also many more compound tidal constituents (not
shown in the Table) that can be generated through the nonlinear interaction of two or more tidal
constituents. Only in waterways with very shallow depths and a large tide range will these other
constituents be important enough to use in a tide prediction, but there are cases where over a
hundred tidal constituents (more than half of them shallow-water constituents) are used. For
example, Zetler and Cummings (1967) used 114 constituents for the tide at Anchorage, Alaska, as
did Rossiter and Lennon (1967) for the tide in the Thames estuary in the United Kingdom. However,
only 74 constituents of each 114 were the same for both groups. Thus each group, included 40
different constituents in their analysis that the other group did not use, which points to the confusion
that can occur in deciding which shallow-water constituents are most important to include.
Understanding the nonlinear mechanisms that generated them, and which one are first, second, or
third order helps in that decision making.
M8 is often larger than many of the shallow-water constituents shown in Table 2.4. This table
only goes up to seventh diurnal tidal constituents, because it includes only constituents generated
by a first-order interaction between two astronomical constituents. M8 is generated by the
interaction of M6 with M2 via the asymmetric nonlinear terms and is thus a second-order interaction.
It often reaches a significant size only because M2 is usually so much larger than the other tidal
constituents.
As mentioned earlier, M6 can also be generated by the interaction of M4 with M2 via the
asymmetric nonlinear terms, but this second-order contribution is significantly smaller than the
effect of quadratic friction directly on M2. Although the quadratic friction mechanism was first
pointed out by Proudman in 1923, for many decades M6 was treated as though it was produced by
the classic shallow-water nonlinear terms through the interaction of M2 and M4. (One can see this
in Schuremans (1958) choice of a node factor for M6, which he erroneously specifies as being the
cube of the node factor for M2 instead of the square of the node factor).
Table 3.2 lists some of the shallow-water constituents typically included in a harmonic analysis.
It also shows the shallow-water compound constituent equivalents for many of the astronomical
constituents. Tables A.1 and A.2 include many more shallow-water constituents. In all three of
these tables the asymmetric or symmetric nonlinear mechanisms are also indicated for each shallow-
water constituent.
The same nonlinear mechanisms that generate overtide and compound tide constituents also
cause the interaction between the tide and various nontidal phenomena such as river flow and storm
surge. The next two sections will look at such interactions and the roles played by the specific
nonlinear terms.
7.6.3 The Nonlinear Interaction of the Tide With River Flow
In Table 2.4 there a fourth way listed to generate the asymmetric nonlinear effects produced by
the other three nonlinear mechanisms (the continuity term M(u)/Mx, the inertial term uMu/Mx, and the
elevation effect on friction u*u*). It turns out that quadratic friction u*u* can produce asymmetric
effects if there is a mean flow present, such as river flow. This can be shown by doing a Fourier
decomposition similar to the one done in the previous section, but including a mean flow, i.e., let
260
u(x,t) = uo(x) + u1(x) cos(t 1(x)) and again for u*u* one has
4
u*u* = a o (x) + 3 [an(x) cos n 1 + bn(x) sin n 1]].............. (7.36)
2 n
To find an one multiplies equation (7.36) by cos n1 and integrates over all possible frequencies, and
to find bn one does the same but instead multiplies the equation by sin n1, i.e.:
2
an = 1 I (uo + u1 cos1) *(uo + u1 cos1 )* cos n1 d1
o
2
bn = 1 I (uo + u1 cos1) *(uo + u1 cos1 * sin n1 d1
o
The steps in the complete Fourier decomposition can be found in Parker (1984) (and see also
Dronkers, 1964, and Fang, 1987). Here only some of the results are presented.
Since a mean current flow has been included one expects to get a value for ao, but it also includes
an effect from the tidal current, namely,
ao = (8/) u1uo when uo<<u1 or

ao = 2 (uo2 + u12) when uo>u1
namely, in the second case, when the river flow is fast enough so that the direction of flow is always
downstream (although current speed oscillates, becoming faster and then slower due to the effect
of the tidal current).
For the main tidal constituent one obtains the same result as in Section 7.6.2 when uo<<u1, i.e.,
a1 = (8/3)u12 (a result first gotten by Proudman, 1923). But when uo ~ u1 we obtain
2 2
a1 = 4 u12 [ uo sin1 uo + 1 ( uo 2 + 2) (1 uo2 ) ]
u1 u1 3 u1 u1
(a result obtained by Bowden, 1953) and one sees that the river flow will increase the frictional
momentum loss from the main tidal constituent, thus reducing the tide range. This is a little more
obvious when uo=u1 , for which one gets a1 = 2uou1.
The result for a2 shows that second harmonics (e.g., M4) can be generated by u*u* when there
is a mean flow present. For the case where uo<<u1 one obtains a2 = (8/3) uou1. For the case when
uo ~ u1 one obtains 2 2
a2 = 1 u12 [sin1 uo + 1 uo (5 2 uo 2) (1 uo2 ) ]
u1 3 u1 u1 u1
When uo=u1 this reduces to a2 = u12, and maximum generation of M4.

As was seen in the last section, u*u* directly generates third harmonics such as M6. From the
result here for a3 one sees that adding a mean flow reduces this third harmonic generation. For
uo<<u1 one again gets a3 =(8/15) u12 (a result also obtained by Proudman, 1923), but for uo ~ u1
one obtains 2 4 2
a3 = 8 u12 [(1 2 uo2 + uo 4 ) (1 uo2 ) ]
15 u1 u1 u1
and once the flow reaches uo = u1 , then a3=0 and all M6 generation by u*u* disappears (because
with the river speed now the same as the tidal current amplitude there is no more flood and ebb and
261
u*u* has become u2). [However, one also sees an increase in M6 generation by u*u* with a
small increase in river flow; see Parker, 1984.]
Figure 7.19 graphically illustrates the results for a1, a2, and a2, resulting from the interaction
between the tide and river flow as a result of the quadratic friction term u*u*. It shows how these
three coefficients change as uo/u1 increases from zero to 2 (that is, from no river flow to river flow
with a current speed twice that of the tidal current). The friction coefficient a1 increases with
increasing uo/u1 indicating increased momentum loss from the main tidal constituent due to the
increased river flow. River flow decreases the tide range through this frictional effect. Coefficient
a2 represents momentum input into the second harmonic (e.g., M4), which increases as the river flow
increases (but only up to the point where uo = u1). Coefficient a3 represents momentum input into the
third harmonic (e.g., M6), which decreases as the river flow increases, reaching zero when uo = u1.
7.6.4 The Nonlinear Interaction of the Tide With Low-Frequency Storm Surge
The same nonlinear mechanisms that generate compound and overtide constituents and that
cause the interaction between the tide and river flow, also cause the interaction between the tide and
low-frequency (i.e., subtidal) storm surge. Of the four nonlinear terms that have been looked at (the
nonlinear continuity term M(u)/Mx, the inertial term uMu/Mx, the quadratic friction term u*u*, and the
elevation effect on friction u*u* ), uMu/Mx and u*u* do not usually play a major role here because
a low-frequency storm surge changes the water level elevation slowly and the accompanying
Figure 7.19. The effect of mean river flow

(uo) on the first three tidal harmonics from a
Fourier decomposition of the quadratic
friction term, u*u*. u1 is the M2 tidal current
amplitude. a1 and a2 represent momentum
input to the M4 and M6. a1 represent frictional
momentum loss from M2. (See text.)
262
currents are therefore usually very small. Thus, at least for subtidal storm surge, nonlinear
continuity term M(u)/Mx and the elevation effect on friction u*u* are the two nonlinear terms one
will need to consider. (For storm surge with frequencies higher than the tidal frequency uMu/Mx and
u*u* are much more important. For a discussion of the interaction of the tide with storm surge that
have frequencies higher than the tidal frequency see Prandle and Wolf (1978). This interaction
depends more on the quadratic friction mechanism because the higher frequencies of the surge mean
stronger currents.)
At the crest of a storm surge the total depth will greater than if the storm surge were not there.
This will increase the tidal wavelength and may also slightly affect the proximity of the
hydrodynamic system to a (damped) resonant condition. The opposite occurs at the trough of the
storm surge. The larger depth under the crest of the storm surge may also lead to slower tidal
current velocities and therefore less (first-order) frictional attenuation, and likewise more frictional
attenuation under the trough of the storm surge. The tidal amplitude is thus larger at the surge crest
than it would be if no surge were present, and smaller at the surge trough than it would be if the
surge were not present. These are the effects of the nonlinear continuity term, M(u)/Mx , with
first-order friction playing a role. The u|u| term also causes the same effects. The greater depth
at the crest of the storm surge leads to less frictional momentum loss per unit volume of fluid and
thus less attenuation of the wave and larger tidal amplitudes. At the trough of the surge the smaller
depth leads to greater frictional momentum loss per unit volume of fluid, and thus greater
attenuation and smaller tidal amplitudes. An example of such effects by a low-frequency subtidal
storm surge was seen in Figure 2.31, in which one can see smaller tide ranges along the entire
Delaware River and Bay coinciding with the trough of a subtidal storm surge, and larger tide ranges
coinciding with the crest of the storm surge.
If low-frequency (subtidal) storm surge were truly periodic (they are really transient oscillatory
phenomena caused by the wind, and to some degree by atmospheric pressure), one could use a
Fourier decomposition similar to that in Section 7.6.2 to show that, in water level spectra from
coastal areas and bays, such periodic subtidal storm surges would produce side bands around each
astronomical constituent spectral line via these nonlinear mechanisms. The lower the frequency of
such a "periodic" storm surge the closer the side bands would be to each spectral tidal line.
However, since storm surges are indeed transient with their energy spread over many frequencies,
the result is a smearing of each spectral line, called a "tidal cusp" by Munk, et al. (1965) (the name
coming from the shape of the spectrum as it rises up to each tidal spectral line on both sides. (See
Section 3.10.2 for more discussion of tidal cusps.)
7.6.5 A Physical Explanation of the Nonlinear Mechanisms
The effect of the nonlinear terms in the equations of motion on a tidal constituent is to cause a
modulation or distortion of that constituent, which can be represented by the combination of that
constituent with one or more new constituents. The Fourier decomposition in the previous sections
reveals what these new tidal constituents are, but does not provide physical insight into the cause
of such modulations or distortions. Here some physical insight will be provided, which will be seen
to involve predominantly wave propagation velocity of the tide wave and frictional attenuation.
The generation of second harmonics (e.g. M4) in shallow water by the so called "shallow-water"
terms (the inertial term in the momentum equation and the nonlinear term in the continuity equation)
has been studied at least since Airy (1845). The wave propagation velocity for long waves without
friction [co= (gh)1/2 ] is approximately constant over a tidal cycle only if the tidal amplitude, , is
263
much smaller than the depth, h, i.e., if /h <<1. For a progressive long wave in shallow water,
where /h is not negligibly small and where will therefore significantly affect the total depth, the
wave crest will travel faster than the trough. Thus the faster moving crest (high water) will move
closer to the trough ahead of it, which at the same time is falling more behind the next crest. Thus,
high water arrives sooner than it would in deeper water, and one sees a faster rise to high water and
a slower fall to low water. This can be seen in the water level curves in Figure 2.25 as one moves
further up the Gironde River (in France). The resulting wave profile will be distorted from a perfect
sinusoid. Subtracting the original sinusoid from this distorted profile, one finds energy in the second
harmonic. This is the asymmetric effect of the nonlinear continuity term, M(u)/Mx.
The tidal current amplitude is approximately u=(/h)co, ignoring friction. When /h is not very
small, u cannot be neglected with respect to co, so the wave propagation velocity at the crest is
actually co+u and at the trough is co-u. The result again is a similarly distorted wave profile. This
is the effect of the inertial term in the momentum equation, u Mu/Mx. The importance of the inertial
term in distorting the tidal elevation is reduced by friction, by a width-decrease, and by reflection
(making the tide wave closer to a standing wave). As will be seen in Section 7.6.7, however, the
lateral inertial terms in the two-dimensional momentum equation have can have a very important
effect on the tidal current.
The first-order effects of friction are to decrease the wave propagation velocity and to attenuate
the wave amplitude. This linear effect will not distort the wave profile; both high and low waters
will be delayed and decreased in amplitude. However, the two nonlinear aspects of the frictional
momentum loss, u|u| and u|u|, will lead to distortion. The effect of u|u| is that the frictional loss
of momentum per unit volume of fluid is smaller for greater depths and greater for smaller depths,
so the crest will travel faster than the trough. This asymmetric effect can thus generate M4 and other
even harmonics just like the classic shallow-water terms, and in fact is usually more important than
the inertial term.
Second harmonics (first overtides) are generated by the above three mechanisms because the
maximum wave propagation velocity and minimum attenuation occur at the crest, and the opposite
occurs at the trough. The tide is distorted in an asymmetric manner (see Figure 7.20).
Quadratic friction, u|u|, on the other hand, has a symmetric effect, causing maximum attenuation
(and minimum wave propagation velocity) at both maximum flood and maximum ebb (which
coincides approximately with the crest and trough if the wave is progressive), and minimum
attenuation (and maximum wave propagation velocity) at slack waters. The result of this symmetric
effect is a third harmonic (e.g. M6) and other smaller odd harmonics (see Figure 7.21).
When two astronomical tidal constituents are present quadratic friction modulates their
combined effects, producing new compound tidal constituents. For example, when M2 and N2 are
in phase, so that their respective high waters occur at the same time, the total depth under the crest
of the combined wave will be greater than when they are out of phase 13.8 days later. Similarly, the
total depth under the trough when M2 and N2 are in phase will be smaller than when they are out of
phase. Thus, M4 generation due to greater depth at the crest than at the trough will be modulated
by the 27.6-day variation of the combined M2+N2 effects, leading to a 27.6-day modulation of M4
and a new constituent, NM4.
264
Figure 7.20. Diagram illustrating the asymmetric effects of

the nonlinear continuity term, the inertial(advective) term,
and the term representing the elevation effect on frictional
momentum loss per unit volume of fluid. (See text.)
Maximum flood and ebb currents will be greatest when M2 and N2 tidal currents are in phase.
Because of the quadratic nature of frictional momentum loss, the loss when M2 and N2 are in phase
will be greater than the sum of the individual losses if M2 and N2 existed independently. The
combined wave will travel slower and be damped more. The opposite will be true 13.8 days later
when M2 and N2 are out of phase and the combined maximum currents are less. The increased
frictional momentum loss at in-phase times, will be greater than the decreased loss at out-of-phase
times, and each constituent will lose more momentum than if it existed alone. Since M2 is larger
than N2, M2 will reduce N2 by a higher percentage than N2 will reduce M2 (see Figure 7.5).
There will also be a 27.6 day modulation of the frictional momentum loss from M2 and N2 , and
thus a 27.6-day modulation of their frictional damping and wave propagation speed. The result is
represented by two new compound tidal constituents, 2MN2 and 2NM2 (which have the same
Figure 7.21. Diagram illustrating the symmetric nonlinear

effect of quadratic friction.
265
frequencies as the astronomical constituents, L2 and 2N2). Third harmonics (e.g. M6) will also be
modulated to produce new sixth diurnals (such as 2MN6 and 4MN6).
In all of the above examples a progressive wave was used for simplicity. For tidal waves in real
estuaries, which, due to first-order "linear" frictional effects are always somewhere between a
damped progressive wave and a damped standing wave, the effects are similar but the phase
relationships and relative amplitudes will vary.
The effects on the tide of a mean flow (e.g. river flow) can also be explained in terms of changes
in wave propagation velocity and frictional attenuation. Quadratic friction generally has a greater
effect than the other nonlinear terms, especially at high river current speeds. A mean river flow
makes ebb current speeds larger and flood current speeds smaller. Due to the quadratic nature of
the frictional momentum loss, the increased loss during the ebb phase is greater than the decreased
loss during the flood phase. The result is a greater energy loss than if the mean flow were not
present, and thus the river flow dampens the tidal wave reducing the tide range.
Quadratic friction effects, being greater during ebb than during flood and thus being asymmetric,
also lead to the generation of M4. Low waters are delayed and high waters come earlier. Frictional
generation of M6 will continue as long as the river current speed is less than the tidal current speed
and there are still slack waters (see upper half of Figure 7.22). When the river flow is greater than
the tidal current the flow becomes unidirectional (there are no more slack waters) and the minimum
attenuation then occurs at what would have been maximum flood (if the river flow hadnt
overwhelmed the tidal current); see lower half of Figure 7.22. Figure 7.19 shows the effect of a
Figure 7.22. Diagram illustrating the nonlinear

interaction of the tide with river discharge via
quadratic friction, for two cases: (1) the river
flow is less than the maximum tidal current; and
(2) the river flow is greater than the maximum
tidal current. (See text.)
266
mean flow on the first three harmonics from a Fourier decomposition of u|u|. The increase in the
friction coefficient a1 with increasing uo/u1 increases momentum loss from the main tidal constituent.
Coefficients a2 and a3 represent momentum input into the second harmonic (e.g., M4) and third
harmonic (e.g., M6), the former increasing with increasing uo/u1 and the latter decreasing.
The quadratic frictional effect described above, may be slightly modified by the other nonlinear
terms, but these smaller effects will probably depend on the specific situation and may be competing
with each other. For example, when increased river discharge raises the mean sea level due to
frictional momentum loss from the mean flow, the resulting increase in water depth will (via the
nonlinear continuity term) increases the wave propagation speed and the tide wavelength to some
degree. Yet, increased (linear) friction tends to decrease the wave propagation speed and
wavelength. A changing wavelength due to one or both of these effects can move the location of
a quasinode (i.e. the point of minimum tide range) and actually increasing the tide range at that
location. This appears to happen in Figure 7.24 (in Section 7.6.6c) south of Philadelphia.
The effect on the tide of storm surge can also be explained in terms of changes in wave
propagation velocity and frictional damping. The relative importance of the four nonlinear terms
depends on the frequency of the surge. The lower the frequency of the surge the smaller the current
velocities accompanying the water level changes, and the less important are u Mu/Mx and u|u|. A
low-frequency storm surge affects the tide primarily through the slowly changing depth on which
the tide propagates. At the crest of the storm surge the total depth will greater and the tidal wave
propagation velocity will be greater. This will increase the tidal wavelength and may slightly affect
the proximity of the hydrodynamic system to a (damped) resonant condition. The opposite occurs
at the trough of the storm surge. The larger depth under the crest of the storm surge also leads to
slower tidal current velocities and therefore less (first-order) frictional attenuation, and likewise
more frictional attenuation under the trough of the storm surge. The tidal amplitude is thus larger
at the surge crest than it would be if no surge were present. Likewise, the tidal amplitude at the
surge trough is smaller than it would be if the surge were not present. These are the effects of the
nonlinear continuity term, M(u)/Mx , with first-order friction playing a key role. The u|u| term also
causes the same effects. The greater depth at the crest of the storm surge leads to less frictional
momentum loss per unit volume of fluid and thus less attenuation of the wave and larger tidal
amplitudes. At the trough of the surge the smaller depth leads to greater frictional momentum loss
per unit volume of fluid, and thus greater attenuation and smaller tidal amplitudes.
7.6.6 A Real-World Example of Nonlinear Shallow-water Effects Using a Numerical Model
The nonlinear effects demonstrated by the Fourier analysis and explained above can be
demonstrated using a numerical hydrodynamic model. That model must be calibrated and verified
with harmonic analysis results obtained from actual data. The model must be run long enough to
produce a times series (at its hundreds of grid points) that can be harmonically analyzed. Since for
simplicity we have been working with a nonlinear one-dimensional model, we will use here a
numerical hydrodynamic model for a long narrow shallow estuary, one that approximately meets
the criteria used in the above sections for simplifying (at least geometrically) the equations of
motion. The Delaware River and Bay estuary is a convenient choice; not only is it long and narrow
and shallow, but it also has an exponential width decrease as one moves up the estuary. Using a
numerical model for this waterway, one can turn the nonlinear terms on and off for different runs
of the model, and not only confirm the effects of each nonlinear term, but also estimate the relative
importance of several nonlinear mechanisms that might have the same effect.
267
7.6.6a The Numerical Model
There are, of course, a whole host of hydrodynamic numerical models, using a wide variety of
numerical schemes, and with also a wide variety of assumptions made about the equations of
motion. Here we use one of the simplest models that still includes the nonlinear terms that we have
been discussing. The model used in this section is a finite-difference numerical model of the
nonlinear one-dimensional equations of motion (7.6) and (7.7) applied to the Delaware River and
Bay, which matches the geometric assumptions of these equations fairly well. For those familiar
with basic numerical modeling a few of the details of this simple model will be described (no
attempt is made to define the terms mentioned, see modeling texts such as Haidvogel and Beckman,
1999, and Kantha and Clayson, 2000); see also Section 8.2).
This numerical model uses a staggered grid spatial scheme (elevation is determined at odd grid
points and velocity is determined at even grid points). Centered second-order spatial differences are
used for all spatial derivatives. The explicit time integration scheme uses centered second-order
time differences (the "leapfrog" scheme), except for the friction term, for which a backward
first-order time difference scheme is used. The friction term (in this very shallow waterway) has
such a stabilizing effect that the model will reach the correct solution from a motionless flat initial
condition in just a couple of tidal cycles; the larger the friction term the faster the correct solution
is reached. Because the friction term has such a strong damping effect, nonlinear instabilities, if they
occurred, never cause the model to blow up, but instead show themselves as very short wavelength,
high-frequency spurious oscillations. Such oscillations result only from the nonlinear terms
involving spatial gradients, i.e. the inertial term and the nonlinear continuity term, leaking in from
the boundaries because of the uncentered second-order spatial differences required there. Once
these two terms are turned off (only) at the boundaries, all spurious oscillations cease (see Parker,
1984). Other details about this simple numerical model are give in Section 8.2.
The model was calibrated for the Delaware River and Bay using the M2 harmonic constants at
15 locations along the estuary. The model was forced at the ocean entrance with the appropriate
M2 tide at Lewes, Delaware and Cape May, New Jersey. The boundary at the upriver end of the
model was a closed end with respect to the tide, which could not get past the rapids north of Trenton
(see Figure 7.1), but still allowed river discharge to enter the model. The model was run for 8 days
and the last 4 days were harmonically analyzed for M2 and overtides.
Although M2 greatly dominates all other constituents, the cumulative effect of the other
constituents is to reduce M2 via nonlinear frictional interaction. If those constituents are left out of
a calibration run (as was done above), then their effect on M2 will be compensated for by a larger
friction coefficient, when one tries to match the model to the harmonic analysis results from the data.
Thus, calibrating the model only with M2 will lead to an empirically determined friction coefficient
that is too large, and when one then adds all the other tidal constituents into the model, there will
be too much frictional dissipation. The ideal calibration would have the model forced by all
important constituents (for every run), but then each run would have to generate a month of
calculated tidal heights if one used, M2, S2, N2, K1, and O1, or a half year of heights if one used these
five plus P1, in order to separate all the constituents with harmonic analysis. [ Actually one might
be able to analyze a shorter time series because there will be no noise, and a least squares harmonic
analysis would be used; see Section 3.3.] Because of this additional (erroneous) frictional effect,
the model (calibrated only with M2) will also generate M6 amplitudes that are too large, as well as
compound tides (such as 2MN6 and 2MS6) that are too large (although the shape of the model-
produced constituent plots along the estuary will look similar in shape to the plots from the harmonic
268
analysis of data. Figure 7.23 shows plots of M2, M4/M2, and M6/M2 produced by the one-
dimensional nonlinear model of the Delaware waterway described above (solid lines) compared with
harmonic analysis results (the * and + symbols) from Delaware water level data. One can see that
the model-predicted M6/M2 graph is larger than the M6/M2 values from the harmonic analyses of the
data (although both curves have very similar shapes). [A better way to calibrate a hydrodynamic
numerical tide model is to start with an M2, but then add the additional important constituents and
re-calibrate (even though one will have to generate and harmonically analyze longer model-
produced time series); see Section 8.6.]
7.6.6b Overtides Generated By the Numerical Model Compared With Analysis Results
Repeated model runs with different nonlinear terms turned on and off, confirmed that quadratic
friction, u|u|, was the dominant M6 generating term. When the rest of the nonlinear terms were also
turned on, producing M4 , the third-order interaction between M2 and M4 (also via these terms)
produced an M6 out of phase with the frictionally generated M6 , such that the total model-produced
M6 decreased a little.
The M4 values produced by the model were close to the actual Delaware data analysis results
(see Figure 7.23 and also Parker, 1984). Running the model with different nonlinear terms turned
on, showed that the nonlinear continuity term, M(u)/Mx, accounted for 73% of the M4. The term
representing the elevation effect on frictional momentum loss (per unit volume of fluid), u*u*,
accounted for 20% of the M4. The inertial term, uMu/Mx, accounted for only 7%.
These same three terms also caused a small shift in mean sea level, although in this case the
inertial term had a very small effect. The largest change in mean sea level occurred near
Figure 7.23. Plots of tidal M2 amplitude and M4/M2 and M6/M2 amplitude ratios
produced by a calibrated one-dimensional nonlinear numerical model of the
Delaware River and Bay (solid lines) compared with harmonic analysis results for
stations along the waterway (*s and +s). (See text) (From Parker, 1984.)
269
Philadelphia, an increase of approximately 0.16 ft (5cm) (i.e., 7.7% of the M2 amplitude at the
entrance to the bay). The term representing the elevation effect on frictional momentum loss was
most important, accounting for 69% of the change in mean sea level; the nonlinear continuity term
accounted for the rest. The term representing the elevation effect on frictional momentum loss had
its greatest effect at Trenton (6% of the M2 amplitude), while the nonlinear continuity term had its
greatest effect near the Chesapeake and Delaware Canal (3.4% of the M2 amplitude).
The model, forced with M2 and N2 , was run for 28 days and the resulting data were harmonically
analyzed. The results confirmed that u|u| generated the compound tide 2MN2. The results also
confirmed that u|u| reduced the amplification of N2 due to the presence of M2.
7.6.6c Numerical Model Results Illustrating Nonlinear Tidal Interaction With River Flow
The numerical model was also run with seven different (constant) river discharges (with only
M2 forcing at the entrance). Figure 7.24 shows the M2 amplitude curve for the runs with u|u| as the
only nonlinear term turned on; all values are plotted relative to the entrance amplitude value. The
river effect is localized in the narrower upper river north of Philadelphia, where the river current
speeds are greater than in the wider portions of the river and bay (due to continuity). In the upper
river one sees a significant decrease in the M2 amplitude. [Downstream of Philadelphia the M2
amplitude actually increases slightly for higher river flows. This is the location of a quasinode under
no flow conditions, but for higher flows there is a shortening of the M2 wavelength and a marked
reduction in the amplitude of the reflected wave by the time it reaches this location, which leads to
a small increase in amplitude there.] Model runs with all nonlinear terms turned on, gave only
slightly different M2 amplitudes (approximately 1-2% larger) than the u|u|-only run.
Figure 7.24. M2 amplitudes produced by a one-dimensional nonlinear

numerical model along the Delaware River and Bay for seven different
amounts of river discharge, with quadratic friction as the only nonlinear term
turned on in the model. U= 1 corresponds to a discharge of approximately
10,000 ft3/sec (283m3/sec). All values are relative to the M2 at the entrance.
270
Increased river discharge delays M2 high and low waters. For example, near Trenton the
u|u|-only run with discharge equivalent to approximately 50,000 ft3/sec (1416 m3/sec) produced a
17.2o later phase lag. With all the other nonlinear terms turned on this lag decreased to 9.4o because
the increased depth (due to the increased river flow) increased the wave propagation speed of the
M2 wave (via the nonlinear continuity term). Figure 7.25 shows the M4/M2 amplitude ratio resulting
from the river flow effect (only the u*u* terms was turned on; all other nonlinear terms were turned
off). As river discharge increased so did the M4/M2 ratio. Eventually, however, at the highest
discharges, the M4/M2 ratio begins to decrease, because the greater (first-order) frictional damping
of M4 than M2 finally overcomes the M4 generating effect, which has leveled off. The increased
M4/M2 ratio was also reflected in an increased delay in the time of low water relative to the high
water.
Model runs with only u|u| also showed the expected decrease in M6/M2 amplitude ratio with
increased river flow (see Figure 7.26.). The mean sea level increased with increased river flow as
expected (see Parker, 1984).
7.6.6d Numerical Model Results Illustrating Nonlinear Tidal Interaction With Storm Surge
Storm surges are transient and approximately periodic for short time periods. However, to
simulate the effects of a low-frequency storm surge the model was forced with an M2 tide plus a
low-frequency oscillation with the same amplitude but with 1/10th the frequency (roughly a 5-day
period of oscillation). Runs were made with tide and surge separately and combined, with each
nonlinear term turned on individually and then with all the nonlinear terms turn on together. The
Figure 7.25. Numerical model-produced M4/M2 tide range ratios along

the Delaware River and Bay for seven different river discharge amounts,
with quadratic friction as the only nonlinear term turned on in the model.
(From Parker, 1984.)
271
Figure 7.26. Numerical-model produced M6/M2 tide range

ratios along the Delaware River and Bay for seven different
river discharge amounts, with quadratic friction as the only
nonlinear term turned on in the model. (See text.) (From
Parker, 1984.)
results are summarized in Table 7.1. With all nonlinear terms turned on there was a reduction in tide
range at the trough of the low-frequency storm surge, which increased as one moved up the estuary
(the results in Table 7.1 are near Trenton). There was also an increase in tide range at the crest of
the surge. The model with quadratic friction as the only nonlinear term showed none of these
effects, nor did it with the inertial term added to quadratic friction. The decrease in the tide range
under the trough of the storm surge was due solely to the effects of the nonlinear continuity term and
the term representing the elevation effect on frictional momentum less, with both terms being of
Table 7.1. Effect of a low-frequency storm surge on the tide near the
head of tide in the Delaware River, using a nonlinear numerical model.
The amplitude of the surge equaled the tide amplitude at the entrance to
Delaware Bay, but had a frequency of 1/10th of the M2 tide frequency.
(See text.) (From Parker, 1991a.)
272
approximately equal importance. The same two terms were also responsible for the increase in tide
range under the crest of the storm surge.
The model was also run with a down-estuary wind blowing over only the lower bay for one tide
cycle and then turned off. With quadratic friction as the only nonlinear term turned on, there was
simply a shift down in water level that propagated up the estuary, but with no effect on the tidal
range. Turning on the inertial term did not change this. But turning on the nonlinear continuity term
and/or the term representing the elevation effect on frictional momentum loss produced a tidel range
reduction (in addition to the shift down in water level) that became more significant as one moved
up the estuary. The nonlinear continuity term had the greater effect of the two terms, but both were
significant.
7.6.7 The Effect of the Nonlinear Lateral Inertial Terms
In the one-dimensional approach in the previous sections the third and fourth terms in each of
the momentum equations in (7.3) were neglected; the (7.3) equations are repeated below.
Mu + u Mu + v Mu + w Mw f v = 1 Mp (frictional forces)x
Mt Mx My Mz Mx
(7.3)
Mv + u Mv + v Mv + w Mv + f u = 1 Mp (frictional forces)y
Mt Mx My Mz My
Here one will still neglect the fourth term in each, still assuming that the vertical acceleration is
small. Now, however, one needs to consider the effects of the lateral inertial terms, which as was
seen in Section 2.3.6e can have a very important effect on tidal currents when there are irregular
geometries such as bends in the waterway or points of land sticking out into the waterway. A good
example was shown in Figure 2.36, which showed the change in the (K1+O1)/M2 ratio, the M4/M2
ratio, and the mean (residual) current across the entrance to Haro Strait due to the effects of the
lateral inertial terms.
As with the nonlinear shallow-water terms, there is no simple analytical solution that allows one
to see the effects of the nonlinear lateral inertial terms (and one can best see these effects with a
high-resolution two- or three-dimensional numerical hydrodynamic model). However, as was done
in Section 7.6.2, one do a Fourier decomposition of these terms to get an understanding of what they
will do to tidal harmonic constituents. For example, for the third term in the second momentum
equation in (7.3) one can do the following Fourier decomposition. As usual the tidal current will
be broken into two orthogonal components, u and v, such as in the major axis and minor axis
directions. Representing the total cross-waterway component of the tidal current, v, as the sum of
two astronomical tidal constituents, one has:
v(x,y,t) = v1(x,y) cos(t 1(x,y)) + v1'(x,y) cos(t 1'(x,y))
where , are the frequencies of the two tidal current constituents, and v1, v1', and 1, 1' are the
tidal current constituent amplitudes and phase lags for the cross-waterway component. To simplify
this for use in later calculations let r = , i.e., the difference in frequency between the two
constituents. Then rewrite v(x,y,t) as
273
v = v1 cos(n r t 1) + uv1' cos([n+] r t 1')
where = + 1 and n is an integer. Also let n = (n r t 1), again to make later calculations a little
simpler to keep track of. Then rewrite v(x,y,t) as
v = v1 cos n + v1' cos([n+] [1' n+ 1])

n
One also makes the assumption that v1' << v1 , which is true for semidiurnal tide regime, since in
such an area M2 is much larger than the other tidal constituents (it may be less true in some areas
when the second tidal constituent is K1, but here we are only trying to demonstrate effects in the
frequency domain and we dont necessary expect very accurate answers). So for convenience we
will call the larger constituents M2.
This lateral inertial term then becomes:
2 n+ 2
v Mv = 1 M(v ) = 1 M ( [v1 cos n + v1' cos([n+] [1' n 1])] )
My 2 My 2 Mx
2 n+ 2
= 1 M ( v1 + v1v1' cos ( [1' n 1]) + u1 cos 2n
2 My
+ v1v1' cos([2n+] [1' n+ 1'])
n
The first term inside the ( ) on the right-hand side of the second = sign is a mean flow generated by
the inertial terms effect on M2. This helps generate the tidally induced residual currents seen in
Figure 2.3.6 (along with the other lateral inertial terms). This is also called tidal rectification. The
third term is the second harmonic of M2, namely, the overtide M4. This term helps produce the
increased M4/M2 ratio seen on the west side of Haro Strait in Figure 2.3.6. It also helps produce the
decreased (K1+O1)/M2 ratio in that figure [but to show that with this Fourier decomposition, one
would have to let v1 represent K1 and let v1' represent O1 ; the result would produce the compound
tidal current KO2, which has the same frequency as M2]. The second and fourth terms are new
compound tidal currents generated by the interaction between M2 and the second constituent. If this
second constituent is N2 then the second term is a low-frequency shallow-water tidal constituent,
MN, which happens to have the same frequency as the astronomical constituent Mn. The fourth
term is a higher-frequency shallow-water tidal constituent, MN4. The nonlinear lateral inertial terms
have the same asymmetric effect as the asymmetric nonlinear mechanisms (M(u)/Mx, uMu/Mx, and
u*u*) discussed in Section 7.6.2 and summarized in Table 2.4.
7.7 Variation of Tidal Currents With Depth
Tidal currents vary with depth, due to frictional momentum loss at the sea bottom. The current
speed increases from zero right at the bottom to a maximum often near (but not always at) the water
surface. The shape of the current profile can vary, depending on the roughness of the bottom and
other factors. There is usually some distance from the bottom over which there is a fairly rapid
increase in current speed (which is usually called the bottom boundary layer), and then a more
gradual increase upward after that. The turbulence caused by the rough bottom makes the thickness
of the bottom boundary layer larger than it would be with slow flow over a smooth bottom (in which
case there is no turbulence and one has what is called a laminar boundary layer), but this is found
only in a laboratory, and not in the ocean or bays, where the flow is always turbulent. In Figure 2.33
274
we saw the vertical variation in tidal current speed and direction in a cross section in Chesapeake
Bay, which showed that at particular times the tidal current can be flowing in opposite directions
near the bottom versus near the water surface. Bottom friction causes the current to slow up and
change direction sooner near the bottom, and then this change slowly works its way up the water
column.
There is also vertical variation in the current due to density effects when there are vertical
gradients in salinity and/or water temperature. This latter baroclinic effect includes the effects of
(usually intermittent) internal tide waves. When there is sufficient stratification an internal tide
wave will propagate along an interface between an upper fresher layer and a lower saltier layer (the
interface does not actually have to be sharp, but can be blurred by mixing). This interface moves
up and down (in much the same way as the interface between air and water moves up and down at
the waters surface), at the same time causing an oscillating current back and forth horizontally.
These oscillations have a tidal period and so the baroclinic tidal currents are superimposed on the
barotropic tidal currents (i.e., those associated with the regular tide, that we have been looking at
in this book). The tidal currents will therefore vary most around the interface. [A horizontal density
(salinity) gradient up an estuary can also generate a so-called estuarine or gravitation circulation,
where there is outflow near the water surface and inflow near the bottom, this nontidal effect being
superimposed on the tidal current.]
To accurately account for these effects on the tidal currents one must go back to the full three
dimensional equations of motion [i.e., equations (7.1) and (7.3)]. To include the baroclinic effects
one can no longer keep density constant, and one must add a conservation of salt equation and/or
conservation of heat equation (equations of state). Needless to say, this makes solutions more
complex, and obviously solvable only with numerical methods. The frictional momentum loss (and
its effect on the currents along the water column) must be represented with some kind of a
turbulence closure scheme (e.g., Mellor and Yamada, 1982). This is beyond the scope of this book.
For a treatment of possible shapes of the vertical current profile see Dronkers (1964, pp145-152),
but these are generally simpler than one will usually see in a real waterway.
In a real waterway there are a variety of hydrodynamic effects that can alter the vertical current
profile, such as bathymetry, which will have more of an effect on deeper currents than on currents
closed to the surface. Because of this (and the changing stratification) there can be great variety in
measured current profiles. This can be seen in Figure 7.27, which shows current profiles measured
with bottom-mounted upward-looking acoustic Doppler current profilers (ADCPs) at eight different
locations in Chesapeake Bay, all taken at the same time (but the bay is so large, each of these
locations is at a different stage of the tide).
As was seen in Sections 5.2 and 7.1.3, tidal constituent ellipses can vary significantly, and that
includes with depth. Frequently one will see tidal current ellipses become narrower with depth and
then change from clockwise rotation to counterclockwise rotation (or vice versa). For example in
Figure 7.28 one sees the M2 ellipse at the upper two depths (near the surface) rotating clockwise but
then the ellipse at a depth much closer to the bottom rotates counterclockwise. At the same station
the N2 ellipse starts out with a fairly wide ellipse at the depth closest to the surface, then becomes
vary narrow a little deeper, both these having clockwise rotation, and then near the bottom getting
even narrower, but rotating counterclockwise. The K1 at the same station (not shown in the figure)
had fairly narrow ellipses at all three depths, but it started out with counterclockwise rotation at the
top two depths and changed to clockwise rotation at the bottom depth.
When the wind blows on the water surface it produces a wind-driven current that decreases with
depth as the frictional stress is transmitted from layer to layer. This, of course, is a nontidal current
275
Figure 7.27. Current profiles from bottom-mounted upward-looking ADCPs at eight locations in
Chesapeake Bay, all measured on October 26, 2006 at noon. The solid line is the along-channel
component, and the dashed line is the cross-channel component. Negative speeds are in the ebb
direction.
but as was seen, there can be interactions with the tidal current, and this could possibly affect the
tidal current, also giving it a depth dependency, in addition to the depth dependency it has from
bottom friction and from density effects. In waterways wide enough for Coriolis to have an effect,
one sees that not only does the wind-driven current decrease with depth, but it slowly changes
direction, each lower layer of water moving slower and further to the right of the wind (in the
Northern Hemisphere), describing the classic Ekman spiral. A similar effect can happen with the
bottom frictional effect on tidal currents, and this may be responsible for the change in rotation of
tidal current ellipses with depth (but there may also be topographic effects involved also).
For more discussion of and approaches to handling the vertical structure of tidal currents see also
Prandle (1982).
276
Figure 7.28. M2 and N2 constituent ellipse

for three different depths for a current station
in the Strait of Georgia. (Reworked from
Parker, 1977.)
7.8 Tidal Dynamics of the Oceans of the World
Even though the oceans are where the tide is first generated, they are the most difficult
waterways to model hydrodynamically. The tide in other bodies of water (the bays and estuaries
connected to the oceans), being too small for the tide-producing forces to generate a tide, result from
the tidal forcing at their entrances by the ocean tide (modified by the continental shelves). This
makes the hydrodynamic modeling of a bay or estuary much easier, for the tidal forcing at the
entrance can be done with accurate harmonic constants obtained from water levels stations (which
often are on both sides of the entrance). (Even when one wishes to move the open boundary
conditions offshore from the entrance there are usually ways to figure out what those offshore
boundary conditions should be. See Section 8.5.) No such convenient open boundary conditions
exist when modeling the tide in the Atlantic Ocean or in the Pacific Ocean. In fact, some modelers
have felt that it is simpler to model all the ocean basins of the world as one, which avoids the
problem of open boundaries that are not well known. For even if one had good harmonics constant
for some line of latitude or longitude that one would like to make an open boundary, technically the
ocean is not being forced at that boundary. The tide in a particular ocean is a combination of the
tide created directly by the tide-producing forces in that ocean, and the tides that propagate into that
ocean from other ocean basins (where they were generated). This is another reason why modeling
all the oceans together as one large connected ocean makes sense (Kuo, 1991).
Another problem has been the data against which to compare ones ocean model. For decades,
oceanographers have been trying to construct accurate cotidal and corange charts for the oceans,
against which models can be compared. However, water level gauges along the coastlines of the
world have data that have been affected by the continental shelves, and so in order to use harmonic
constants from these stations to develop a model, that model must have the continental shelves
277
represented in the model. That increases the resolution requirements of the model. Water level
gauges on islands provide the best data to harmonically analyze, usually not having large enough
shallow water areas around them to dramatically change the ocean tide signal. Only in the last
couple of decades has satellite altimetry provided a way to accurately measure the tide in the middle
of the ocean. For although there is a very large sampling period for the data from satellite altimeters,
this does not keep one from harmonically analyzing the data (one does, however, have to worry
about periodicities in the satellite orbits that may interfere with particular tidal constituents); see
Cartwright, 1991; Egbert, et al, 1994.
The first attempts to understand the tides in the open ocean were based on whatever cotidal and
corange charts could be produced from the available water level data (with all its coastal limitations,
as mentioned above). (See Chapter 9 in Cartwright, 1999.) In that pre-computer era, analytical
solutions were attempted by adding together combinations of Kelvin waves, Poincare waves, and
other specialized solutions of some version of the long wave equation on a rotating Earth (which
often obtained the name of the first person to solve them). It was not until the advent of high-
resolution global tide models and satellite altimetry (to provide global tidal data) that
oceanographers were able to gain some real understanding of the tidal dynamics of the oceans, and
were able to produce reasonably accurate cotidal and corange charts for the oceans (see Figure 7.29
for one example). For more information on the modeling of oceans see Doodson, 1958; Cartwright,
1977; Schwiderski, 1978,1980,1991; Platzman, 1991; Le Provost and Vincent, 1991; Kuo, 1991;
Andersen, et al, 1995; and Shum, et al, 1997.
The continental shelves are sometimes modeled separately from the rest of the much deeper
ocean, the open ocean providing the boundary conditions along the continental slope. A continental
shelf, especially if it is it is reasonably shallow, can be a significant co-oscillating tidal system in
itself. Looking at the tidal system over a continental shelf in the simplest possible way, a long tide
wave from the ocean will undergo a partial reflection-transmission at the depth discontinuity along
the continental slope, increasing the tide range over the shelf. Another full reflection of the tide
wave at the coast will further increase the tide range, that increase depending on the width of the
shelf compared with the tidal wavelength over the shelf (that wavelength being dependent on the
water depth over the shelf). A shallower continental shelf has a shorter tidal wavelength and its
width is thus more likely to be closer to of a tidal wavelength, leading to a larger tide range. The
largest tide range along an open coast is along the coast of Argentina because of the very wide
continental shelf, reaching to the Falkland Islands. The above discussion, of course, is an over
simplification. Tide waves can propagate along the continental shelf as Kelvin waves or other types
of long waves, with the tidal frequencys relationship to the Coriolis parameter (which depends on
latitude) having an important effect. Strong tidal amplification can occur across wide continental
shelves in mid and low latitudes. For more information on tides over the continental shelf see the
review on the subject by Clarke (1991).
278
Figure 7.29. An example of an ocean M2 amplitude chart, this one for the Atlantic Ocean
produced by Schwiderksi (1979). Amplitudes are in cm. Circles with a dot in the center denote
amphidronic points (i.e., points where the amplitude is zero).
279
280
8. The Use of Numerical Hydrodynamic Models For
Predicting Tides and Tidal Currents
8.1 Introduction
The alternative to making a tide or tidal current prediction at a particular location using a
statistical technique (such as harmonic analysis) is to make a tide or tidal prediction using a
numerical hydrodynamic model of the waterway in which that particular location is found. As has
been seen, the harmonic analysis technique relies solely on the knowledge of the frequencies where
the astronomically and shallow-water produced tidal energy is found. The numerical hydrodynamic
model, on the other hand, makes the prediction using the physics of the water movement as
described by a discretized version of the equations of motion (such as those seen in Section 7.3).
Such a model still relies on astronomical forcing, but for a bay or river this forcing is needed only
at the entrance (usually in the form of harmonically produced tide predictions) and the model then
uses physics to determine how the tide wave propagates up the waterway and what the tide and tidal
current will be at the location(s) of interest. [For a global ocean tide model, instead of tide
predictions at an entrance, the model uses the actual tide producing forces throughout the ocean(s),
with the forced equations of motion determining how the tide wave propagates throughout the ocean
and up the bays and rivers.]
As one steps forward in time when making a harmonically-based tide prediction, one is, for each
time step, merely adding up the contributions of the cosine values for each tidal frequency for which
one has harmonic constants. This is done independently of the previous time step. However, as one
steps forward in time when making a prediction using a numerical hydrodynamic model, the
calculation at each time step makes use of values at one or more previous time steps both at the
location (grid point) where the prediction is being made and at one or more nearby locations (grid
points). The formulas for calculating the tidal height or tidal current at a particular time step (using
the values at previous time steps) come from the rearranged discretized equations of motion. The
next section will show a simple example so that those not familiar with numerical modeling theory
can gain some insight into how this type of prediction process works.
8.2 The Workings of a Very Simple Numerical Hydrodynamic Model
It is beyond the scope of this book to go into great detail about how numerical hydrodynamic
models are created and how they work. There are thousands of papers on the subject, and a few
good textbooks such as Haidvogel and Beckman, 1999, and Kantha and Clayson, 2000. Here a little
background is merely provided for those without any experience with numerical models.
Numerical hydrodynamic models are computer programs that calculate numerical solutions of
the equations of motion [such as equations (7.1) and (7.3)] for a specific waterway, whose shoreline
and bathymetry determine the solid boundaries of the model, and whose forcing at its entrances
(such as the tide at the ocean entrance and discharge from rivers that flow into the waterway)
281
provides the momentum and energy to drive the motion inside the waterway. The model may also
include additional equations for salinity and temperature/heat, if one wants to include baroclinic
(density) effects on the tide and tidal current.
Chapter 7 provided examples of a few analytical hydrodynamic models, that is, very simple
mathematical equations that can be solved using algebra and trigonometry to produce formulas with
which one can calculate the tide or tidal current at any point in time at any place within the idealized
geometry of the waterway. However, such analytical models do not accurately represent any real-
world waterways. In order to solve the equations of motion one had to leave out all but the slightest
variation in the geography of the waterway, and one had to leave out all the nonlinear terms in the
equations. When those nonlinear terms and the complex geometry are left in, the only alternative
is to solve the equations numerically.
When one uses the complete three-dimensional equations of motion (including the extra
equations to include density effects) to describe the tidal hydrodynamic over the full three-
dimensional geometry of a waterway, the resulting numerical model is large and complex, as one
might expect. However, to get across the basic idea of how a numerical solution works, a very
simple geometry will be used, along with a very simple numerical solution technique, the finite-
difference technique. (There are a whole host of numerical modeling techniques, most more
sophisticated than the finite-difference technique, but for our purposes this simple technique gets
across the basic idea of numerical models.)
This basic finite-difference numerical modeling technique will be applied to a simple one-
dimension nonlinear set of equations of motion, equations (7.27) and (7.28), which is repeated here
M + 1 M (b (h+) u) = 0 (7.27)
Mt b Mx
f (7.28)
Mt Mx Mx h+
where is the tidal elevation, u is the cross-sectionally averaged tidal current velocity, b is the width
of the basin, h is the width-averaged depth of the basin, and c f is the friction coefficient (which is
usually determined empirically for a particular waterway, but is often on the order of 0.0025).
Let x be the distance between points on a grid representing the waterway (which for a one-
dimensional model is nothing more than a single string of boxes running up the waterway). Let t
be the time interval between time steps, as the predictions are created by time stepping through the
solution. Each time step makes use of the values calculated in the previous time step. The partial
derivatives in equations (7.27) and (7.28) are replaced with small differences in speed (u) and
elevation () divided by small differences in space (x) and time (t), creating finite-difference
analogs of these equations. For example, M/Mt becomes /t, but for numerical reasons involving
stability (discussed in the above mentioned numerical modeling texts) it is better to use what is
called a centered difference, i.e. [u(x,t+t) u(x,tt)] and divide it by 2t. [In other words, one
takes the u value at time interval (t+t) and subtracts from it the u value at time interval (tt), and
then divides by 2t.] Likewise, uMu/Mx, first becomes M(u2)/Mx, this spatial derivative being
replaced by (u2)/x, but a centered spatial difference is used, i.e., [u2(x+x,t) u2(xx,t)] and
it is divided it by 2x. [In other words, one takes the u2 value at the grid point after x, i.e., (x+x),
and subtracts from it the u2 value at the grid point before x, i.e., (xx), and then divides by 2x.]
282
8. The Use of Numerical Hydrodynamic Models For Predicting Tides and Tidal Currents
The finite-difference analogs of these one-dimensional nonlinear equations of motion then become
(x,t+t) (x,tt) + 1 1 { b(x+x) [h(x+x) + (x+x,t)] u(x+x,t) (8.1)

2t b(x) 2x
b(xx) [h(xx) + (xx,t)] u(xx,t) } = 0
u(x,t+t) u(x,tt) + 1 u2(x+x,t) u2(xx,t) = g (x+x,t) (xx,t) (8.2)

2t 2 2x 2x
1 c f u(x,tt) *u(x,tt)*
h(x)+(xx,t)
This particular model often uses a so-called staggered spatial scheme, where the elevation,
(x,t+t), is determined at odd grid points using equation (8.1) and the velocity, u(x,t+t) , is
determined at even grid points using equation (8.2). One essentially calculates a value for (x,t+t)
from the rest of the terms in equation (8.1), which all involve values at two previous time steps, t
and (tt) [at three grid points, xx, x, and x+x]. Likewise, one calculates a value at u(x,t+t)
from the rest of the terms in equation (8.2), which also involves values at t and tt [again at three
grid points, xx, x, and x+x]. And one continues to step forward in time, producing as many
predicted elevations and velocities as one needs.
The model is forced at the entrance with predicted tide elevations (from accurate harmonic
constants) at grid point 1, and with river flow (or a no flow condition) at the highest numbered even
grid point in the model. Centered second-order spatial differences are used for all spatial
derivatives. The inertial term, which is the second term in equation (8.2), requires velocities at odd
grid points, which are obtained by averaging the velocities at each surrounding pair of even grid
points. The nonlinear continuity term in equation (8.2) requires elevation values at even grid points,
which are obtained by averaging the elevations at each pair of surrounding odd grid points. [At the
two open boundaries one is forced to use uncentered second-order spatial differences, which can
lead to instabilities, so these two nonlinear terms are turned off at these two points.]
The explicit time integration scheme, using the rearranged equations (8.1) and (8.2), uses
centered second-order time difference (the so-called leapfrog scheme), except for the friction term
(for which such a scheme would cause the numerical solution to be unstable), so for this term a
backward, first-order time difference scheme is used. In fact, the friction term has a very stabilizing
effect, which allows this model to reach a correct solution after (time stepping forward) from a
completely flat initial condition (a zero startup) in just a couple of tidal cycles. The larger the
friction term (i.e., the shallower the waterway being modeled) the faster the correct solution is
reached. There is a commonly used stability criterion (the Courant-Friedrichs-Lewy criterion) that
specifies how small the time step should be for a particular grid spacing, so that the model will stay
stable. The higher the resolution of a model (i.e., the smaller x is) the smaller the time step t must
be and the longer it will take the model to run.
There are two types of two-dimensional models. The most typical solves the depth-integrated
two-dimensional equations of motion [of which equations (7.22) are a very simplified form]. For
a narrow waterway, a model might instead use the width-integrated two-dimensional equations of
motion, in order to better look at how current speed changes with depth. With a two-dimensional
depth-integrated model (and again using a basic finite-difference technique), one can have hundreds
283
or many thousands of square grid cells filling up the entire the entire waterway (see Figure 8.1,
which shows only a part of a grid). Using square grid cells does not make it easy to accurately
match the complicated shoreline at the models boundaries. One approach is to increase the
resolution, i.e., have smaller cells, for specific parts of a waterway (a so-called high-resolution
nested grid), for example, needing more cells in a narrow channel where it is important to predict
accurately the tidal current. But there are other forms of finite-difference models where the grid
cells do not have to be squares, and can be shapes that help the model better match the shorelines,
for example, using a boundary fitted coordinate system or an unstructured coordinate system. With
models that use a finite-element solution technique the grid cells are (usually) triangles, which makes
it very easy to match the shorelines, and very easy to change the resolutions of a particular part of
the model.
For three-dimensional models there can be many layers of grid cells, and there are many
approaches with regard to the vertical spacing of the grid points (see Mellor, et al, 2002). The
frequently used sigma coordinate grid allows higher resolution near the bottom (to better resolve
the bathymetry and the bottom boundary layer) and near the surface (to better handle wind effects),
but can cause problems when there are large scale changes in depth, and so other systems for vertical
spacing have also been developed. For models covering a waterway with high horizontal resolution
and many layers, there can be hundreds of thousands of grid cells, but with todays computer power,
that is usually not a serious problem.
Today there many types of numerical hydrodynamic models that are used to model the tide, or
to model water level and currents with the tide being a predominant driver of this motion (along with
changing wind, atmospheric pressure, river flow, and water density). These models use a whole
host of numerical solution techniques, grid coordinate systems, and/or specialized forms of the
equations of motion (only a few of which include: Blumberg and Mellor, 1987; Le Provost and
Vincent, 1991; Lynch and Werner, 1991; Cheng, et al, 1991; Hess, 1989; ). Some of these models
have been standardized, with the same basic model being used by many oceanographers. Yet even
with this increased standardization, the approaches to calibrating and verifying a model can still
vary.
8.3 Advantages of Using a Numerical Hydrodynamic Model
When making tide or tidal current predictions using harmonic constants extracted from water
level data or current data, there can be a number of limitations or problems. Now we will look at
how we can overcome those limitations by making tidal predictions using numerical hydrodynamic
computer models. [In the following discussion, the term model is used to mean a well calibrated
and accurately verified three-dimensional high-resolution numerical hydrodynamic model, that
accurately captures the relevant physics for the type of application for which the model is being
used. Sections 8.5 and 8.6 will present a discussion about how to assure that a model meets these
criteria.]
The first limitation using the results of a harmonic (or nonharmonic) tidal analysis is that one
can make tide or tidal current predictions only at the locations for which one has data. Often one
wants predictions at locations where one doesnt have data. With tide predictions that is usually not
too serious a problem, because generally the tide regime changes slowly enough that either a nearby
station can be used, or one can do reasonably well linearly interpolating between two nearby tide
stations. However, this is not the case for tidal currents, which (as has been discussed many times
284
Figure 8.1. (Upper panel) A fine grid section of the NOS New York New Jersey
Operational Forecast Model. (Lower panel) The predicted currents in the same section of
the model illustrating the complex horizontal structure of the tidal currents (including
current shears and eddies). (From Wei, 2003.)
in this book) can vary dramatically with distance, both horizontally and vertically. The lower panel
in Figure 8.1 shows a section of the Port of New York and New Jersey where one can see a complex
structure in the tidal currents, including current shears and eddies. This figure was produced by a
numerical model of the port (a portion of the model grid is shown in the upper panel); see Wei,
2003. To try describe this complex a current pattern based on harmonically analyzing data, one
would need hundreds of bottom-mounted ADCPs, or many towed ADCPs (e.g., see Figure 2.35),
or still better, a high-resolution microwave radar system (which, however, provides only surface
currents), and even then, to harmonically analyze all these data would be extremely labor intensive
and inefficient compared to using a numerical hydrodynamic model.
Numerical hydrodynamic models can provide tide and tidal current predictions at many
thousands of locations. The finer the resolution of the model grid the more places one can provide
285
predictions. Normally one will not want predictions at all the possible locations but it makes it
easier to select them for specific locations of interest.
Models are especially important for tidal current predictions. It has been seen that tidal currents
near the sea bottom can be very different in speed and direction and timing than tidal currents near
the water surface. They can also be quite different in a channel and in the shallows just a few feet
away. It is very difficult to obtain enough data to adequately describe all the spatial variability of
the tidal currents. With a bottom-mounted upward-looking ADCP one can get only a single vertical
velocity profile over time. With a side-mounted ADCP one can get only a single horizontal velocity
profile at over time (or possibly sweep it around to get coverage of one water depth for one small
area). With HF radar current measurement systems one can get spatial variation in the surface layer
over a larger area, but not with enough resolution for a small bay or harbor unless one uses very high
frequencies like microwaves. Only a validated high-resolution numerical model can really provide
accurate tidal current predictions over an entire waterway.
Because of their great spatially variability, if one wants tidal current predictions for a particular
location in the waterway, one will want to look at predictions at all the grid points around that
location, and in all the (depth) layers of the model at that location, to see how much the current
speed and direction changes from grid cell to grid cell (see Figure 8.1 again for an example of
horizontal variation) and at various depth intervals. One can then decide how best to present the
information to a user, like a pilot or ships captain. One can graphically present synoptic views of
nowcast and forecast model outputs (and even animate them) and/or select time series from key grid
points to harmonically analyze for inclusion in the Tidal Current Tables (instead of basing it on a
current sensor record that perhaps was not installed at a good location to best represent the tidal
currents for that area).
The second limitation using the results of harmonic (or nonharmonic) tidal analysis, is the
problem with making accurate tidal current predictions in waterways like that shown in Figures 7.2,
7.3, and 2.34 that was discussed in Section 3.6.3 (see also Section 9.2.6). Because of the complex
geography affecting the mixed tidal currents in this waterway, there are such rapid changes in the
harmonic make up of the tidal currents over short distances, that the six reference stations for that
area now in the Tidal Current Tables are not nearly enough. The (K1+O1)/M2 ratios and the M4/M2
ratios change from location to location, and permanent tidally induced mean currents are also
generated (that should be included in the tidal current predictions). The only way to reasonably
handle such a complex situation is with a high-resolution numerical hydrodynamic model. The
predictions from the model may have to be provided only electronically, since convenient graphical
representations in hardcopy form of all the possible dynamic situations remains difficult. A
numerical model was developed in Canada to make predictions for the above complicated dynamic
area (Crean, et al, 1988), but when they tried to turn the results in to a Tidal Current Chart it took
more than a hundred charts to fairly accurately represent the results in a way that a mariner could
(try to use to) make a prediction (Huggett, 1998). See also Section 9.4.3b.
With regard to our third limitation, using the results of harmonic (or nonharmonic) analysis it
has been seen in Sections 2.3.3, 4.1.4, 5.3.4, 7.6.3, and 7.6.4 that in shallow waterways the tide
nonlinearly interacts with nontidal phenomena such as river flow and storm surges. These nontidal
phenomena can modify the harmonic constants during strong events, so that they cannot be used to
make tide or tidal current predictions for other time periods (and they are generally not even usable
for the time period of the nontidal event, because such events are too short for a reliable harmonic
analysis to be done). However, a numerical hydrodynamic model can easily handle the nonlinear
interactions between the tide and nontidal phenomena. It can model the tide and the various nontidal
286
phenomena at the same time (including their interactions), and predict the total water level and the
total currents. If the model includes the appropriate equations for salinity and water temperature,
it can also include all the transient baroclinic effects on the tide and tidal current (such as internal
tide waves). It can even include the gravitational (estuarine) circulation due to salinity gradients up
the waterway, another nontidal effect.
As mentioned in the first paragraph, when we say model, we are talking about validated three-
dimensional numerical hydrodynamic models with good resolution. With todays computer power,
there is no reason to settle for less. Sometimes more important than having access to predictions at
many locations is the fact that high-resolution three-dimensional numerical models usually do a
much better job of representing the physics of the waterway, than do one- or two-dimensional
models, especially when it comes to the tidal currents, and so they provide more accurate
predictions.
8.4 Disadvantages of Using a Numerical Hydrodynamic Model
With all the advantages discussed in the last section, one must remember that a lot of work and
skill goes into developing a numerical hydrodynamic model that can provide accurate tide and tidal
current predictions when forced only with the tide, or accurate water level and current predictions
when forced at the entrance with actual water levels, plus inputs of wind, atmospheric pressure, and
river discharge.
There are many ways that an inadequate model can be developed. The harmonic constants for
the entrance may not be accurate enough, because there was not a long enough data time series to
harmonically analyze. There may not be enough data at enough locations in the waterway to
adequately calibrate it and verify it. It may not have high enough spatial resolution to do a good job
on the tidal currents. Sometimes the available bathymetry is too old and the depths have changed
(which can be a serious problem for shallow waterways). Or the calibration and verification
procedure itself may not have been done in a way that resulted in a model that could reproduce all
the important tidal frequencies accurately. When models are not producing as accurate results as
desired, sometimes certain coefficients are played with and sometimes bathymetry is changed, and
wrong choices can be made. One may make changes that allow the model to reproduce a particular
data set, but those changes might not work well for a different set of data (covering a different
dynamic situation).
One error that was often made in the past occurred when depths were taken off nautical charts
for assigning depths to all the grid cells of the model. Sometimes the modeler forgot that the depths
on a nautical chart are referenced to mean lower low water (MLLW) and not to mean sea level
(MSL), which the model is usually referenced to. For example, a six-foot depth is six feet below
MLLW; it would be larger if it were referenced to MSL. Such an error essentially reduces the
depths over the entire model by half the tide range, making the waterway shallower by that amount.
This might not matter much if the waterway is very deep, but for a shallow waterway such an error
can make a significant difference (and when the model did not produce the desired results, the
modeler then often played with various coefficients to make the model match the data).
Of course, to correct these depths from MLLW to MSL, one must know the tide range at every
point in the waterway, which in fact, is what is to be calculated by the numerical model and so is not
known (at least accurately) when the model is first set up. One can initially do this depth correction
with some approximate spatial description of the changing tide range, and then re-correct the depths
287
with the new spatial description of the tide range that comes out of the first model run, and then
rerun the model (perhaps re-correcting one more time, if necessary).
Of course, the charted depths themselves were originally calculated by adjusting the depth
soundings for the tide at the time they were measured, and only recently have models been used in
that correction process (see Hess et al, 1999). So there is some unknown error built into those
charted depths (see Section 8.7.2). The most promising approach for accurately measuring
bathymetry with respect to a known datum is to make the depth measurements with respect to a
known ellipsoid datum using GPS, and then to transform them to a tidal datum (e.g. MLLW) using
a tool like VDatum (see Parker, 2002, and Parker, et al, 2003). A tool like VDatum requires some
type of a numerical tide model to produce the horizontal variation in the tidal datums, either by
dynamically interpolating the harmonic constants from all the available water level stations (Hess
et al, 1999) or by a more typical calculation based on the tidal forcing at the entrance (but with data
assimilation to make sure that the model matches the tide exactly at the available water level
stations).
A three-dimensional numerical hydrodynamic model that is well calibrated and verified using
a good deal of high quality water level and current data, can usually do a better overall job of tide
prediction and especially tidal current prediction. However, at a particular station where there are
many years of water level data that can be analyzed, a prediction based on a quality set of harmonic
constants obtained from those data is usually more accurate than a model-produced tide prediction.
It takes a lot of effort to develop a well calibrated and verified numerical hydrodynamic model.
It takes much less effort to harmonically analyze a water level time series. So if one has a long
enough time series of quality water level data, and one needs tide predictions at only a few locations,
the harmonic analysis approach is probably best. However, If one needs tidal current predictions,
the choice is not as clear. In some cases, one may have a long data time series from a current sensor
that was deployed in a good location (i.e., in an area with little horizontal variation in the tidal
current) and covers enough depths. The harmonic constants from this current station will be very
useful and will provide tidal current predictions that are as good or even better than those produced
by a numerical hydrodynamic model. But for many other situations, one has to believe that the
model (again if well calibrated and verified) will do the better job.
8.5 Proper Tidal Forcing of a Numerical Hydrodynamic Model
After the above discussions of the advantages and disadvantages of using a numerical
hydrodynamic model to produce tide and tidal current predictions, one must remember that such
models still must be forced at their open boundaries to the ocean with very accurate tide predictions.
And even if the model is eventually going to be forced by actual real-time water level data at the
ocean open boundary (e.g., as part of a real-time data-based nowcast-forecast system to provide
water levels and currents), it is important to first show that the model can produce accurate tide and
tidal current predictions throughout the waterway when forced at the entrance with accurate tide
predictions.
To provide proper tidal forcing at the ocean entrance(s) to the waterway, one should have at least
one water level station with at least one years worth of data for each ocean entrance. If the entrance
is reasonably wide, it is important to have at least two water level stations, one on each side of each
entrance. For each of these stations one should follow essentially the same procedures for producing
a reference station in a Tide Table (see Section 4.4).
288
The water level data time series for such a station must be at least a year long in order to be able
to use as many tidal constituents as possible in the harmonic analysis. The most recent data should
be used (especially in small waterways) in case there have been bathymetric changes which may
have affected the tide regime. Typically the standard 37 tidal harmonic constituents listed in Table
3.2 have been used in CO-OPS (and its predecessor organizations) unless the station is located in
a shallow waterway with a large tide range, in which case additional tidal constituents are used (see
Table A.2). At Anchorage, Alaska, 114 tidal constituents were used to produce accurate tide
predictions, the 77 additional shallow-water constituents being needed because of the 30-ft tide
range and the very shallow-water conditions at the upper end of Cook Inlet. (The tidal bore that
occurs at Turnagain Arm near Anchorage will be a challenge for most numerical hydrodynamic
models.) However, at entrances to shallow waterways, the extra constituents are usually not needed,
because the shallow-water overtides and compound tides are usually not large at the entrance (it
requires some distance along the shallow waterway for them to grow large).
As with analyzing data for a reference station in the Tide Tables, one must try to include as many
tidal constituents as possible. This may include a number of very small constituents, that are
sometimes rejected because of their small size (0.03 feet being a typical arbitrarily chosen cutoff),
the reasoning being that they are unreliable because they are small enough to be down in the
nontidal noise continuum. However, to avoid throwing out possibly real tidal constituents, one
should do a number of one-year analyses on different time periods to see if the amplitude and epoch
(phase lag) for these constituents are similar from year to year. Such consistency is a better
acceptance criterion than an arbitrary lower cutoff value
One might be tempted to put the open boundary condition, with the tidal forcing, further up the
waterway (perhaps) to make the model smaller, but this is never a good idea. The ideal place for
an open boundary condition is at a location that will not be affected by the dynamics of the
waterway itself. One is, in fact, assuming this to be the case when one forces a waterway at the
entrance; nothing is allowed to change the tide predictions (or water level values) used at the
entrance. Only for the ideal case of a waterway attached to a semi-infinite ocean (i.e., with infinite
width and depth) is this absolutely true. For waterways with deep entrances directly to the ocean,
it may be very approximately true. But for a waterway whose entrance is shallow in places and
connects to a shallow continental shelf, there are likely to be some local hydrodynamic effects, and
putting the forced boundary there might not produce as good results as one would like.
For this reason, some modelers move the open boundary offshore of the entrance onto the
continental shelf. That usually solves the problem of local hydrodynamic effects at the open
boundary, but it presents another problem, which can be much worse. Offshore there are unlikely
to be any long water level data time series for the points along this new open boundary on the
continental shelf, from which accurate harmonic constants can be derived for tide predictions. (It
is even less likely that real-time water levels will be available there to drive a nowcast-forecast
model.)
The accuracy of the predictions inside the waterway from the numerical hydrodynamic model
is critically dependent on the forcing at the entrance, and by going offshore one adds a great deal
of uncertainty to that forcing. There are several possible approaches to developing a good offshore
tidal forcing boundary condition for the numerical hydrodynamic model of a bay or estuary.
First, one can actually put several water level measuring devices out on the continental shelf at
the locations appropriate for the extended model grid. Bottom pressure sensors have been used for
this purpose, but they have had a variety of problems (including water density effects, and the lack
of a datum) , which can take some effort to overcome (Filloux, 1991). (Recently additional bottom
289
pressure sensors have been deployed as part of the DART system to detect tsunamis. The water
level data from these bottom pressure sensors have been harmonically analyzed, so that tide
predictions can be subtracted from real-time water level data and the real-time residual data will
better show tsunami signals.) A more promising technique for measuring offshore water level data
is the use of GPS receivers on buoys. Such systems make water level measurements relative to a
known ellipsoid datum, and they are not affected by the changing density of the seawater as pressure
sensors are.
Second, one can try to use satellite altimetry data. Even though the sampling interval is very
large (on the order of a day or more at cross-over points), one can still extract tidal constituents from
these data using a least squares harmonic analysis technique or a modified response method (see
Cartwright, 1991; Egbert, et al, 1994; Woodworth and Thomas, 1990; and Cherniawsky et al, 2001).
Results have typically been better in the deep ocean compared with over the continental shelves and
near the coasts, because of the shorter tidal wavelengths over the shelf, which require better spatial
resolution in the data, and there are also less altimetric data over the shelves (see LeProvost, 2001).
Third, one can try to use the output from a shelf model. An often-used approach is to use such
a model with altimetry data analyzed for the tides. It is also likely be forced with an ocean or global
tide model (e.g., Schwiderski, 1980 and 1991; Bennett, 1992). Another approach is to start with a
such a shelf or coastal model and then use some type of data assimilation technique (using the tides
predicted from accurate harmonic constants at stations along the coast) to improve the model (see
Zhang, Wei and Parker, 2003; Chen and Mellor, 1999; Myers and Baptista, 2001.)
For many waterway entrances (to a shelf area) none of the offshore modeling approaches may
turn out to be as good as simply using tide predictions right at the entrance based on harmonic
constants from long data time series (and perhaps not using the tidal current predictions in the
immediate vicinity of the entrance. However, the shelf modeling approach is likely a better choice
if one is predicting total water levels, including wind and pressure driven changes over the shelf.
Wind and pressure data fields and forecast wind and pressure fields are used in the model, and then
can be adjusted using a data assimilation technique to better match the nontidal coastal water levels
(see Zhang, Parker, and Wei, 2002).
However it is accomplished, once the best possible tidal forcing is provided at the ocean open
boundary to the model of the bay or estuary, then one must go through a careful process to calibrate
and validate the model, which will be discussed in the next section.
8.6 Tidal Validation a Numerical Hydrodynamic Model
With most numerical hydrodynamic models there are particular constants/coefficients that can
be adjusted to try to make the predictions of the model better match the available data. (For fully
nonlinear three-dimensional models, this theoretically should not be necessary, but that may still
depend on the type of turbulence closure scheme used.) If the predictions of the model do not match
the available data, there must be a reason, and the first thing often tried is to play with whatever
coefficients are in the model. (This is assuming that one has done the best possible job of matching
the shoreline and bathymetry of the waterway, using adequate resolution.) The second thing often
tried is to adjust the depths in particular grid cells (which might be wrong if the bathymetry used is
old). One may also decide to increase the resolution of the model, especially if the shorelines of the
waterway are complex, but this takes a lot of effort and so may not be done until other approaches
are tried. Adjusting coefficients and depths sometimes seems more like art than science, because
it is sometimes not obvious which way to change the coefficients or the depths (since the
290
hydrodynamics of a shallow waterway with a complex geometry is usually complicated). For deep-
water waterways there is usually much less of a problem.
The shallow-water areas will always be the biggest problem in calibrating and verifying a
numerical model. The worst situation is when there are many small little bays, inlets, and streams
connected to the main waterway, most of which do not have water level data, or even worse, perhaps
not even accurate bathymetry. Even when the rest of the model is producing accurate tides (and
even good tidal currents) one cannot know how well the model is doing in these shallow bays, inlets,
and streams where there is no corroborating data. One cannot assume the model does well there
because it does well elsewhere, because these shallow-water areas are usually at the edge of the
model (as opposed to being areas between stations with data). Also, if the model is not doing as well
in main areas of the waterway, one cannot be sure that the cause may not be these shallow areas
(which can often absorb energy and store water in significant amounts).
Except for tides, all the energy in a water level record or current record are stochastic, i.e.,
random or quasi-periodic (e.g., periodic for a short while, but then gone). The tides are deterministic
and truly periodic (because of the astronomical forcing behind them), and one can figure out from
harmonically analyzed data records exactly what the response of a particular waterway is to the
periodic tidal forcing at its entrance. There are a number of different tidal frequencies one can look
at with respect to how well the model is doing, combined with a knowledge of the frequency effects
of each term in the equations of motion on which the model is based (from Fourier decompositions
such as done in Section 7.6). Thus, in carrying out the detective work necessary to make a model
match the available data, the periodic tide provides many useful hints. One can learn a lot from
looking at cotidal and corange maps of the waterway for all the important tidal constituents.
Corange ratio maps of M4/M2 for the tide, for example, will give one an idea of where nonlinear
shallow-water effects are important.
It is difficult to calibrate and verify a hydrodynamic model for stochastic nontidal effects,
because there are too many different situations to cover (with a whole host of wind, pressure,
density, and river effects). One ends up dealing in probabilities. But luckily one has the tides, with
known energy at known frequencies, with which to test a model. If a model can handle the tides and
tidal currents, it will most likely be able to handle wind effects, river discharges, and storm surge
(as long as one has good boundary conditions for the those effects and accurate nontidal forcing
data).
In calibrating and verifying a numerical hydrodynamic model for tides and tidal currents, one
should first get the tides right (because they are affected primarily by the entire waterway), and then
deal with the tidal currents (which are also greatly influenced by local conditions). The local
conditions may be adjustable to make the tidal currents at a particular location better, but that should
not be done until one knows that the overall hydrodynamic system is being represented well by the
model.
In calibrating the model to do a good job with the tides, modelers often started by first running
the model with only M2, typically largest tidal constituent. This was done to save analysis time.
Some percentage of the hundreds of predicted time series output at hundreds of grid points must be
harmonically analyzed to see if the predicted M2 amplitudes and epochs (phase lags) match those
from the harmonically analyzed water level data at the various stations around the waterway. By
predicting with only M2 one only needs to analyze only a couple of days of predicted time series.
If one adds the next four important tidal constituents (S2, N2, K1, and O1), one needs to predict and
analyze month-long time series. Including the 37 most important constituents requires year-long
times series for analysis in order to resolve some of the smaller constituents. This was usually done
291
in stages, first using M2 only, but then adding the other constituents. However, with todays
computer power, forcing with all available tidal constituents and analyzing longer predicted times
series is not a problem.
It is important to remember that one should never calibrate a numerical model only using M2
(and then not going on to more refined calibrations by adding the other constituents). As has been
seen in Sections 2.3.2 and 7.6, there are important nonlinear interactions between tidal constituents
(even in areas as deep as the continental shelf) which cause additional energy loss from each
constituent. In other words, because of the presence of M2, the constituents N2 and S2, will be
smaller than they would be if M2 was not present. This is because of the (nonlinear) quadratic aspect
of frictional energy loss. The M2 effect on N2 or S2 is much greater than the effect of N2 or S2 on M2
(since M2 is so much larger than N2 and S2), which explains the smaller amplifications of N2 and S2
in Figure 7.5 compared with the amplification of M2. However, the cumulative effects of all the
constituents on M2 is significant. Thus, if one calibrates a model only using M2, then the additional
energy loss due to the nonlinear effect on M2 of the other tidal constituents will not be present in the
model (but it will be in the data), and so to make the model match the data, it will be calibrated to
frictionally reduce M2 more than it should be. Then, when the other constituents are included in the
model, too much energy will be lost. [For the very simple case of the one-dimensional nonlinear
model used in Section 7.6.6 one can see the result of this in Figure 7.24, where the friction
coefficient was made too large (trying to compensate for the added energy loss due to all the other
constituents), and as a result the model over predicts M6 (which is generated by the same quadratic
frictional effect that reduces these other constituents).
The relative size of certain constituent ratios, such as M4/M2 or (K1+O1)/M2, in the data versus
in the predictions from the model, should also provide clues as to where the model might possibly
be adjusted. There are many aspects of the tidal hydrodynamics of a waterway with unique
frequency dependencies, that once learned can also help with the detective work.
Section 8.4 described the consequence of using the depths from a nautical chart (which are
referenced to the MLLW datum) without adjusting them to a MSL datum, and therefore essentially
changing all the depths in the model by half the tide range, i.e., making the waterway shallower by
that amount. Making the entire waterway shallower shortens the tidal wavelength and slows the
wave propagation velocity. Depending on the length of the waterway (relative to 1/4 of a tidal
wavelength), this shortening of the tidal wavelength could move the situation closer to resonance
or farther from resonance. Learning whether this increases the tide range or decreases is a useful
piece of information, and one might in fact artificially decide to decrease the overall average depth
of the waterway for one model run, and then artificially increase it for another, just to see how the
tide ranges (as well as the amplitudes and epochs of the different tidal constituents) are affected.
This is just one way to do a sensitivity analysis on the model, in order to have some idea of what
kinds of adjustments might be made to fix a model that is not performing well enough. Closing off
small inlets or adjoining bays for one model run is another type of artificial adjustment whose effects
can be enlightening. Such small inlets and adjoining bays can soak up energy, and may have a
cumulative effect similar to adjusting the effect of bottom friction over the entire waterway.
As mentioned above, making adjustments to improve tidal current predictions is usually much
more of a local accommodation. The speed and direction and timing of the tidal current at a specific
location (and at a specific depth below the water surface) is greatly affected by the water depth at
that location, the distance from that point in the water column to the sea bottom, the orientation of
the bathymetry at that exact location, and the nearby water depths. As has been discussed the tidal
currents are very sensitive to frictional, continuity, and inertial/advective effects. And examples
292
have been give illustrating how complex the structure of the current field can be. For example, the
currents in a channel can be much faster than in the shallows right next to it, and yet varying
considerably from top to bottom in the channel, sometimes flowing in opposite directions due to
frictional effects. It is even more complicated at channel bends or near points of land, due to inertial
effects that create eddies (resulting in zero frequency "residual currents" and shifts of tidal energy
into higher harmonics).
One should not attempt to make local adjustments that might improve tidal currents in a specific
area until after all necessary adjustments have been made to get the tide right, because this also
affects the tidal current. For the local effects, grid resolution will be a critical factor in how
accurately the tidal currents can be predicted, and one common fix for poor tidal current
predictions is to increase the spatial resolution in the areas where the predicted tidal currents are not
matching the data well. This can become almost more art than science, since the situations can be
so varied that it is difficult to develop too many rules of thumb for making adjustments to improve
predicted tidal currents.
As has been already mentioned, trying validate the numerical model for its prediction of currents
is difficult because of the lack of suitable current data fields that can show the current flow structure
with suitable resolution, and over long enough time periods. There are limitations to the information
one can obtain even from ADCPs (bottom-mounted , side-mounted, or towed) or from HF radar
systems. Although limited to the surface layer, and often without high enough resolution for some
waterways, an HF Radar system is the closest thing one has to being able to see if a three-
dimensional numerical hydrodynamic model is outputting current fields that are close to the real-
world (or at least the real world as measured by an HF radar). Figure 8.2 shows one typical
example of a comparison between an HF radar synoptic view of a surface current field at a specific
day and time and the output of a three-dimensional finite-element hydrodynamic model of
Chesapeake Bay for the same day and time (Gross, 2002). There are similarities in the two
circulation fields, and there are many differences. In the model output one sees the signature of the
navigation channels in the current field, which does not appear in the output from the HF radar. This
may be a resolution issue, the radar not having a high enough resolution to resolve the current over
the navigational channel.
When trying to validate a model with data, especially data from a relatively new remote
measurement technique, there will always be some uncertainty. In the example in Figure 8.2, is the
model right or is the radar right? In this case, one believes the model because one knows about the
navigation channels and the fact that currents are fast in and over those deeper depths, but one may
not always be able to so easily make a judgement.
8.7 The Operational Use of Numerical Hydrodynamic Models
There are a number of applications where it is usually better to obtain tide and tidal current
predictions (or water level and current predictions that include the tide and tidal current) from a
numerical hydrodynamic model. A few of these will be briefly discussed below.
8.7.1 Creation of Tidal Current Charts, Forecast Atlases and Digital Tidal Prediction
Products
A numerical hydrodynamic tide model can be used to produce Tidal Current Charts, Tidal
Circulation and Water Level Forecast Atlases, and Digital Tidal Prediction Products (see Sections
293
Figure 8.2. Comparison between a surface current field measured by an HF radar system (left
panel) in lower Chesapeake Bay and the output of a three-dimensional finite element
hydrodynamic model for the same time period. (from Gross, 2002)
8.3, 8.4 and 8.6). Such a model (Patchen, 1986) was used to produce the Delaware River and Bay
Tidal Circulation and Water Level Forecast Atlas (NOAA, 1987); see Figures 9.10 and 9.11, and
Parker, 1988.
With the hundreds of locations for which a model can predict tidal currents, one can produce a
Tidal Current Chart with many more current vectors covering more of the waterway, and covering
multiple depths. The large number of vectors allows one to choose the current vectors that are most
representative of the circulation at each location. When producing a Forecast Atlas, the hundreds
of locations for which the model can predict tidal heights allows one to produce very detailed height
contours for the entire waterway. The numerical hydrodynamic tide model can also produce such
charts for any astronomical situation. Thus for waterways with mixed tides or mixed tidal currents
(the most difficult situation to accurately represent in Tidal Current Charts or in Tide or Tidal
Current Tables), one can create different charts for (e.g.) equatorial lunar declination and for
maximum lunar declination, or (e.g.) for spring, mean, and neap situations.
As will be discussed in Section 9.7, a digital tidal prediction product can provide more accurate
tide and tidal current predictions than a Tide Table or a Tidal Current Table or a Tidal Current Chart.
This is because predictions based on harmonic constants can be used for every station, and thus the
problems with Table 2 nonharmonic-based predictions discussed in Sections 3.6.3 and 6.8 (and to
be discussed further in Sections 9.2.5, 9.3.6, and 9.4.2) can be avoided. The one problem, however,
is obtaining harmonic constants for all the necessary locations, especially for the tidal currents.
Because of the greater spatial variability in currents, current data are required for more locations
than are needed for water level data, but it is more difficult and expensive to acquire current data.
And so here again, a numerical hydrodynamic model is the best way to obtain the required time
294
series for harmonic analysis. A model also gives one the flexibility to provide tide or tidal current
predictions at locations where one does not even have data.
8.7.2 Tide-Based Corrections In Hydrographic Surveys
When hydrographic (bathymetric) surveys are carried out and millions of depth soundings are
acquired using a multibeam sonar system, each of those soundings is a measure of the total water
depth at the exact time the measurement was made. That total water depth measurement includes
the water level elevation (of which a major portion is usually the tide) at that exact time. At another
time, during a different phase of the tide, a total water depth measurement at that same location will
be different. The difference between the two water depth measurements at the same location is due
to the different water level heights at the two different times (and again, most of which is usually
due to the different tidal heights). To be able to provide depths on charts that are good for any time,
all depth sounding must be corrected for the water level elevation at the time the depth sounding
measurement was made. Both the eventual charted depth and the water level must be referenced to
a vertical datum. In the U.S., that vertical datum (the chart datum) is the mean lower low water
(MLLW).
The key question then is: what is the water level elevation (relative to MLLW) at the exact
location where the depth sounding measurement is being made? If one subtracts the water level
elevation referenced to MLLW, from the total depth, then one will have a water depth referenced
to MLLW, i.e., the charted depth. (See Parker and Huff, 1998.) There are, however, no water level
measuring devices at all the locations where soundings are being made. For the earliest
hydrographic surveys a single water level station might be installed in the area of the survey, and
then it was assumed that the water level at that station applied to the entire area being surveyed.
Since the tide regime changes with horizontal distance (and the shallower it is, the more quickly the
tide regime changes), most hydrographic surveys have used more than one water level gauge, and
a method of interpolating between the gauges was used to provide water level heights for the entire
survey area.
This was called tidal zoning, because the variation in the water level over the survey area was
assumed to be the same as the variation in the tide (there being no equally reliable way to consider
wind or other effects on the water level; and the tide was usually the dominant part of the water level
change). But there could still be problems. Zoning techniques often involved representing the
spatial variation of the water level in terms of connected polygon zones based on the variation in
tide range and times of high and low water. The drawbacks of such methods included: (1) treating
the variation of the water level as though it were exclusively tidal (when the wind-induced
component could vary spatially and temporally in a much different way than the tide); (2) the
sudden jumps in value when crossing the boundary between two polygon zones; and (3) the
constant time differences assumed between the tide at two geographic points over all astronomical
tidal situations over a month (the same problem looked at for Table 2 stations in the Tide Tables).
The use of numerical tide models can improve this situation, using two different techniques
(Parker and Huff, 1998). In the first technique, a model is used to provide the spatial variation of
all major tide constituents over the survey area, which can then be used to accurately predict the tide
at any point in space at any time. To these tide predictions one can then add the nontidal part of the
water level signal (due to the wind, atmospheric pressure, density changes, and/or river flow), as
interpolated between the nontidal elevations determined at the installed water level gauges (by
subtracting the tide predictions at those gauges from the water levels). Such continuous water level
295
zoning can also be provided using less computer intensive (but still numerical) interpolation
techniques, such as a Laplacian technique (Hess, et al, 1999).
The second technique is an even more attractive approach for accurately measuring depths with
respect to a common (chart) datum. This technique uses differential GPS on a moving ship, and
techniques for rapid ambiguity resolution, often referred to as on-the-fly (OTF) GPS (Parker and
Huff, 1998). Since the transducer of a multibeam sonar system is at a known position below a GPS
receiver on the ship, then the depth measurements are known relative to the an ellipsoid datum. If
the spatial variation of the MLLW tidal datum with respect to the ellipsoid has been determined in
advance, then the measured depths can be directly referenced to the chart datum as they are
measured (and no corrections based on water level zoning are necessary). The spatial variation in
the chart datum (with respect to the ellipsoid) can be determined with a numerical hydrodynamic
model (Schmalz, 1996) or with a numerical interpolation technique such as the Laplacian technique
mentioned above (Hess et al, 1999). ( Using OTF-GPS also eliminates the need to correct the depth
soundings for effects on the survey vessels dynamic draft.)
A depth sounding measurement made relative to an ellipsoid datum can be transformed to a
value relative to MLLW using NOSs Vertical Datum Transformation Tool (VDatum), in which is
included the spatial variation of all the various tidal, ellipsoid, and orthometric datums for the
waterway being surveyed (see Parker et al, 2003, and Parker, 2002). The geographic distribution
of the tidal datums in VDatum is calculated using a numerical tide model.
VDatum can also be used to calculate MHW shoreline from a digital elevation model (DEM)
by transforming the DEM from the ellipsoid datum (used when the Lidar was taken from a GPS-
positioned aircraft) to the MHW tidal datum, the MHW shoreline being the zero elevations points
in the MHW reference frame (see Parker, 2003). This is a more accurate and efficient way to
measure shoreline than traditional photogrammetric methods.
8.7.3 Real-time Based Nowcast/Forecast Model Prediction Systems
Although the tide and tidal current are usually the largest cause of changes in water level and
currents, the nontidal contributions (due to wind, atmospheric pressure, water density changes, and
river flow) can still be very important, depending on the needs of the mariner. (Of course, there are
times, such as during a storm or a river flow event, when the nontidal part can dominate the tidal
part.) For example, the pilot bringing into port a deep-draft commercial ship wants to know exactly
the underkeel clearance that his ship will have, so that it does not run aground, and to know this he
must know the total water level, not just the tide prediction. Or, a HAZMAT team wants to know
which direction an oil spill will move, and to predict this they need (in addition to good wind
information) accurate information on the currents, not just a tidal current prediction.
In respond to such needs, systems such as CO-OPS the Physical Oceanographic Real-Time
System (PORTS) were developed to provide real-time water level and current information to the
maritime community (Appell, et al, 1994; Parker, 1995; Parker, 1986). Such systems could provide
real-time data and information, but they originally could not provide short-term forecasts of water
level and currents, which were important for short-term on-the-water decisions (e.g., the ships pilot
and the HAZMAT team benefit greatly from short-term forecasts of water level and currents). More
recently PORTS was expanded to include a real-time based nowcast/forecast model prediction
system (Parker, 1998a; Parker, 2002; Aikman et al, 1996), which consists of an accurately calibrated
and validated numerical hydrodynamic model, as well as a surrounding system for managing the
real-time input of data and output of predictions, including graphics, quality control assessments,
296
etc. For nowcasts, such models are driven by real-time water levels as well as real-time surface
wind fields (Patchen, 1986, describes one of the earliest experimental real-time models). For
forecasts, sometimes the tidal and nontidal component are provided separately, with the forecast
nontidal water level signal provided by a coastal nowcast/forecast model (Aikman, et al, 1996;
Zhang, Parker, and Wei, 2002), which is driven by the predictions from a numerical weather forecast
model. For such coastal models other real-time data are often necessary, including satellite sea
surface temperature and salinity and water temperature profiles from ships (Szabados, et al, 1987).
Nowcast/forecast models are first calibrated and validated for accurately predicting the tide and
tidal current. If the real-time instrumentation fails one can still provide tide and tidal current
predictions over the entire waterway. Over the last decade a number of accurately calibrated and
validated hydrodynamic models for operation in the real-time based nowcast/forecast mode have
been developed and validated (see: Bosley and Hess, 1997; Wei, 2003; Schmalz, 1996; Kelley, et
al, 20002; Gross, et al, 2000; Patchen and Blaha, 2002). Such real-time based nowcast/forecast
model systems either support or become a component of other forecast models, such as storm
surge models, oil spill trajectory models, and water quality models.
297
298
9. Products For Presenting Tide and Tidal Current
Predictions
9.1 Introduction
Since the first time that mankind began to regularly predict the tide or the tidal current (or the
arrival of a tidal bore in a few rivers) there have been products devised to provide these predictions
to the general public in some type of convenient form. Most maritime nations have been producing
such products for at least a century, and in some cases for many centuries. Because of the nature
of the tide and tidal current these products have been very similar from nation to nation, and they
generally fall into one of only few categories. This chapter will briefly look at some of these tide
and tidal current prediction products. The products of the Center for Operational Oceanographic
Products and Services (CO-OPS) in the National Ocean Service (NOS) in NOAA are used as typical
examples, but they are very similar to the analogous products from other nations and from some
private companies. The advantages and disadvantages of each type of product, discussed in this
chapter, generally apply no matter which country produces that product. There has been, of course,
an occasional special variation of a particular product designed to overcome one of the typical
shortcomings of that type of product, which will be mentioned where relevant. This chapter only
very briefly touches on some of the history of such products. For a much more thorough and
fascinating coverage of the history of tide and tidal current products one should read Cartwright,
1999, which also includes many historical references. Sections 9.2 through 9.6 cover what might
be termed the classic hardcopy/printed tidal prediction products (even if delivered on the Internet),
while Section 9.7 discusses the improvements possible in tidal predictions delivered digitally.
9.2 Tide Tables
The Tide Table is the oldest method of providing tide predictions to mariners, and even today
it is still the primary way in which most users acquire tide predictions, whether obtained as hard
copy books, on CDs, or on the World Wide Web of the Internet. Prior to being published in their
own volumes, tide predictions were provided in other publications. The first published tide
predictions in the U.S. appeared in 1830 as part of The American Almanac and Repository of Useful
Knowledge for the Year 1830, published by Gray and Bowen in Boston. Tide notes on nautical
charts began in 1844. Tables for obtaining tide predictions appeared in 1853 in the Appendix to the
Annual Report of the Coast & Geodetic Survey. In 1864 a thousand copies of tide tables for that
year were provided during the Civil War to the Union naval forces. In all these early tables the tide
predictions were produced using a nonharmonic interval method
In 1867 the U.S. Tide Tables were published in their own separate volumes. There were two
volumes with daily predictions for 19 reference stations and time and height differences for 124
subordinate stations. Those first tide tables provided only high water predictions (since most large
vessels usually left a port on the high tide). Low water predictions were included beginning in 1887.
299
Two years earlier, in 1885, the U.S. Tide Tables began to be produced harmonically using the first
U.S. tide predicting machine, Ferrels maxima and minima tide predictor. In 1912 the Harris-
Fischer tide predicting machine was introduced and used to produce the Tide Tables. That machine
could predict actual hourly tidal heights (and not just the high and low waters).
By the 1920s the Tide Tables looked fairly similar in format to those used today. But even
before Ferrel developed his harmonic analysis and prediction method in the late 1800s (later
improved by Harris and then Schuremen), Tide Tables tended to be made up of two type of tables.
The first, referred to as Table 1, contains daily tide predictions for the most important ports and
harbors. These locations are referred to as reference stations in the U.S. (and as primary ports in
the U.K. and some other countries]). The second type of table, referred to as Table 2, contains time
and height differences for thousands of other locations, which are referred to as subordinate stations
in the U.S. (and as secondary ports in the U.K. and some other countries). The predictions at a
subordinate station are produced by applying the time and height differences to the daily predictions
at the reference station with the most similar tidal characteristics (usually the closest reference
station geographically). The daily predictions in Table 1 are produced with the harmonic tide
prediction equation (see Section 3.4.2 and Chapter 4), using the constituents calculated from a
harmonic analysis of at least a years worth of water level data. The time and height differences in
Table 2 are produced using one of several nonharmonic methods (see Section 3.6 and Chapter 6),
which compare the high and low waters of a water level data time series at the subordinate station
to the predicted high and low waters for the same time period at the station.
At the time of the writing of this book there are four volumes of Tide Tables produced by CO-
OPS that cover the entire globe. In the 2007 Tide Tables, East Coast of North and South America
there are 76 reference stations and 2589 subordinate stations. The 2007 Tide Tables, West Coast of
North and South America have 60 reference stations and 1318 subordinate stations. In the 2007 Tide
Tables, Central and Western Pacific Ocean and Indian Ocean there are 95 reference stations and
1998 subordinate stations. The 2007 Tide Tables, Europe and West Coast of Africa have 38
reference stations and 982 subordinate stations. The predictions for locations in other countries
included in these volumes are generally received through bilateral exchange with each nation.
Table 1 and Table 2, as well as other tables that have been added to the Tide Tables, will be
described in the following sections.
9.2.1 Table 1 Daily Predictions for References Stations
Figure 9.1 shows one page of the four pages of a typical Table 1 reference station, in this case
showing the daily predictions of high and low waters for Boston, covering January, February, and
March 2006. On each day of each month the predicted times and heights of high waters and low
waters are provided in chronological order, high water alternating with low water. The times of high
and low waters are given in hours and minutes (relative to the local time meridian, in this case 75o
west). The heights of high and low waters are given in both feet and centimeters (relative to
MLLW). To get a height at some point between a high water and a low water the user of the Tide
Table is referred to Table 3 (see Section 9.2.4), which provides a cosine-based interpolation scheme.
This scheme works reasonably well for deep-water tide stations, but less well for shallow-water
stations where the tide curve is distorted away from a cosine shape by nonlinear hydrodynamic
mechanisms (see Sections 2.3.2 and 7.6.2).
300
9. Products For Presenting Tide and Tidal Current Predictions
Figure 9.1. A page from the Table 1 daily predictions for Boston, MA, from the Tide Tables,
East Coast of North and South America.
301
The phases of the moon are also indicated on the days that they occur with symbols for new
moon, first quarter, moon, and third quarter. The rest of the important astronomical data for the
entire year is provided on the inside of the back cover of the Tide Table, including: moon in apogee,
moon in perigee, moon farthest north of Equator, moon on equator, moon farthest south of equator,
March equinox, June solstice, September equinox, and December solstice. (See Section 2.2.1 for
definitions.) Boston is one of 76 reference stations in the Tide Tables, East Coast of North and
South America. These reference stations are listed in the front of the Tide Tables, along with the
year(s) of the data from which the harmonic constants were calculated that were used to make the
daily predictions. For example, the Boston daily predictions in the 2006 Tide Tables (Figure 9.1)
were produced using harmonic constants from the harmonic analysis of five years of water level
data, from 1994 through 1998.
9.2.2 Table 2 Time and Height Differences for Subordinate Stations
Figure 9.2 shows one typical page of the 42 pages of Table 2 in the Tide Tables, East Coast of
North and South America. This page provides tidal information for 59 subordinate stations, one
station per line, 43 stations of which are referred to the Boston reference station (which also has a
line in the table) and 16 stations which are referred to the Portland reference station (listed on the
previous page). For each station the information includes: its latitude and longitude, the time and
height differences for the high waters and the low waters, the mean and spring ranges, and the mean
tide level. All height differences, ranges, and levels are given in feet. Sometimes height ratios are
provided instead of height differences (in rare cases both differences and ratios are used). The tide
at the 60 stations shown on the page in this figure are predominantly semidiurnal, so the spring range
is given. At stations where there is a significant diurnal inequality, the diurnal range is given instead
of the spring range.
The subordinate stations are split up into groups of stations referred to the same reference
station, which is listed above each group. There are two groups on this page, one group referred to
Portland and one to Boston. To make a prediction on a particular day at a subordinate station, the
time and height differences (or ratios) are applied to the daily prediction for that day at the reference
station listed above that group of subordinate stations. For stations with special situations that
cannot be totally handled by this format, there are Endnotes with additional information. In some
special cases an entire page will be devoted to the so-called peculiarities in the behavior of the tide
at a station, an example of which is shown in Figure 9.3. In this case the nonlinear effect of shallow
water has distorted and flattened out the low water half of the tide cycle, but that distortion changes
with different astronomical conditions, so the user is warned not to use Table 3 to interpolate
between low and high waters. Such peculiarities are actually fairly common, although few are
highlighted in the Tide Tables as in Figure 9.3. They are even more common in tidal currents.
The time and height differences in Table 2 were produced using one of several nonharmonic
comparison methods, which compare the high and low waters of a water level data time series at the
subordinate station to the predicted high and low waters for the same time period at the most
appropriate reference station (see Sections 6.3 and 6.4).
302
Figure 9.2. A page from Table 2 of the Tide Tables, East Coast of North and South America
showing 60 subordinate tide stations.
303
Figure. 9.3. A caution in the Tide Tables about variations in low

water times due to shallow-water effects.
9.2.3 Table 8 Tide Prediction Accuracy
Table 8 provides some statistical figures on the tide prediction accuracyof the reference
stations in a Tide Table. Figure 9.4 shows Table 8 from the Tide Tables, East Coast of North and
South America, which includes 32 stations. The statistical values provided for the times and heights
of the high and low waters at each station include the:
(1) 90% distribution level, which means that 90% of the absolute values of the difference are less
than or equal to the value shown in the column;
(2) standard deviation of differences, which is the square root of the average of all the squares of
the differences; and
(3) average difference, which is the average of the all the differences (with their signs taken into
consideration).
These values are determined by comparing predicted and observed high waters (and low waters) for
one year of data (the specific year is listed in the third column of the table), the observed values
being subtracted from the predicted values.
Technically, since one is comparing predicted and observed values, one is not really looking
at the accuracy of the astronomical tide prediction, but at how well this tide prediction can represent
304
Figure 9.4. An example of a Table 8 showing statistical figures related to the tide prediction
accuracy of the reference stations from the 2006 Tide Tables, East Coast of North and South
America. The values in this table, however, are actually greatly influenced by nontidal changes
in water level (see text).
the actual water level (on average over a year). The water level observations include nontidal
influences, such as the effects of winds and atmospheric pressure and river flow on water level. For
the average difference values, most of these nontidal contributions are averaged out over a year, but
there can still be asymmetries in seasonal effects on the nontidal water level that will not average
out and will show up in the time and height differences. Thus, for example, in Figure 9.4 one sees
larger average differences for high and low waters at Philadelphia, which is on the upper Delaware
River and thus is influenced by river flow. During high runoff periods the river flow raises the mean
sea level, reduces the tide range, and distorts the tide curve asymmetrically, all of which affect the
average differences in Table 8 (see Sections 2.3.3 and 7.6.6) .
Being based on only one year of data, the average differences in Table 8 may primarily
represent the nontidal contributions, and, in fact, the astronomical tide itself could be totally
captured by the harmonically produced tide predictions, and thus could really be 99.99% accurate.
(One can say, at least, that the average difference due to errors in predicting the astronomical tide
will be no worse than the value in Table 8.) If the comparison between predicted and observed high
waters is done over many years the mean difference values may be smaller and closer to representing
any missing tidal energy in the tide prediction (and thus a truer indication of the accuracy of the
astronomical tide prediction). But seasonal effects might still have an influence. Section 3.7.1 and
305
specifically Figure 3.11 showed how much the seasonal sea level range can vary from year to year.
It is possible that the year chosen for the statistics in Table 8 could have been a typical year or it
could have been an unusual year.
The standard deviations listed in Table 8, however, will include nontidal effects no matter how
many years of data are used, because the positive and negative differences are squared and do not
cancel each other out. One sees larger standard deviations for high water differences at Philadelphia
(as expected based on the discussion in the last paragraph). The four stations shown in Table 8 from
the northern Gulf of Mexico (Pensacola, Mobile, Grand Isle, and Galveston) all have even larger
standard deviations for the time differences between predicted and observed times of high water and
low water (and there is an explanatory note for this Table 8 mentioning the influence of weather
conditions). At these locations the tide range is small and is often dominated by water level
variations caused by wind and pressure effects (even a front passing over the waterway can
significantly affect the water level). Likewise the 90% distribution level values also reflect the
nontidal influences, no matter how long the length of the observed and predicted time series
analyzed, because absolute values of the differences are used (and thus the nontidal variations do
not average out).
Even though the statistical values in Table 8 are not strictly speaking a measure of how well the
astronomical tide has been predicted, the statistical information in Table 8 has been useful for
mariners who use tide prediction as the best estimate of actual high and low waters that they will
experience with their ships. Even then, however, there will always be come confusion because the
average differences shown in Table 8 (being averaged over a year) are less than the differences
between the predicted tide and the observed water level seen on any given day (and much less than
the differences seen when there is a large storm surge or strong river flow).
9.2.4 Other Tables and Information in the Tide Tables
There are other small tables in the Tide Tables that provide additional information such as how
to determine: the height of the tide between a high water and a low water, the local mean time of
sunrise and sunset, the reduction of local mean time to standard time, the conversion from feet to
centimeters, and the lowest and highest astronomical tide and other tidal datums.
Table 3 provides a method for getting a tidal height at some point between a high water and a
low water using a cosine-based interpolation scheme. This scheme works reasonably well for deep-
water tide stations, but less well for shallow-water stations where the tide curve is distorted away
from a cosine shape. A graphic method is also provided for the same purpose, also assuming cosine
interpolation.
Table 4 is nine pages of tables that give the local mean time of the rising sun and the setting of
the suns upper limb (i.e., the suns upper outer edge) for every fifth day of the year for every 5
degrees of latitude.
Table 5 allows a value from Table 4 to be converted to standard time.
Table 6 gives the time of the rising and setting of the moons upper limb for every day of the
year, but for only 8 locations (Boston, New York, Washington, D.C., Charleston, Savanah,
Galveston, and the Panama Canal) in the Tide Tables, East Coast of North and South America.
Table 7 provides a way to convert feet to centimeters.
Table 9 shows five tidal datum values for all the reference stations, including: the lowest
astronomical tide (LAT); mean lower low water (MLLW); mean low water (MWL); mean high
water (MHW); mean higher high water (MHHW); and the highest astronomical tide (HAT). LAT
306
and HAT are the highest and lowest predicted tide values at a given location over a 19-year period
(i.e., over the 18.6-year lunar nodal cycle). LAT and HAT are calculated by generating tide
predictions for the most recent National Tidal Datum Epoch (presently 1983-2001) using the latest
set of tidal harmonic constants.
9.2.5 Limitations of Tide Tables
Tide Tables are hardcopyproducts that have been printed for more than a century, and only
in recent years have been extended to electronic media such as CDs and the World Wide Web on
the Internet. As printed documents there were certain limitations put on what could be included in
a tide table, because of page limitations and because of the lack of ease with which certain
information could be conveyed to a user. With electronic media those limitations go away, but the
traditional format for tide tables continues to survive, if only because of its familiarity, but also
because extensive new data may be required to produce digital tide prediction products based on the
harmonic analysis of all stations (see Section 9.7). [One alternative to this is the use of numerical
hydrodynamic models forced with accurate tide predictions at the entrance based on long-term tide
stations (see Chapter 8). However, that requires more data than for just the entrance. The extensive
validation of the model, that is required to make sure it produces accurate predictions everywhere
in the modeled waterway, requires a reasonable amount of data for those validation sites.]
The first limitation of a tide table is that (harmonically produced) daily predictions can only be
provided for a relatively few stations out of the thousands in the table. This is a page limitation,
since the daily predictions for a reference station take four pages in Table 1, while the time and
height differences for a subordinate station take only one line in Table 2, with roughly 60 stations
per page. Most stations must therefore be included as subordinate stations in Table 2. As was seen
in Section 3.6.3, in the introduction to Chapter 6, and in Section 6.9, serious prediction errors can
occur if the harmonic makeup of the subordinate stations in Table2 is not similar enough to the
harmonic makeup of the reference station in Table 1. In this situation the mean time and height
differences in Table 2 (that come out of a nonharmonic comparison analysis) are really good only
for certain times of the month. This problem is most likely to occur in a waterway with a mixed tide
and a strong diurnal signal, because the tidal characteristics will change relatively quickly with
geographic distance, and the subordinate station will often be too far from the reference station. In
this case, tide predictions using Table 2 time and height differences will be least accurate around the
times of maximum lunar declination, and most accurate near times of equatorial declination (see
Sections 3.6.3 and 6.5.1for examples). But this problem also arises when shallow water distorts the
tide curve, shifting around the times of high water or low water so that the time differences between
the subordinate station and the reference stations can vary considerably (see Figures 3.7, 3.8, and
3.9 for one example). If this problem occurs, one solution is to add an additional reference station
to the Tide Table, one that is closer to the subordinate stations and has more similar tidal
characteristics. (But sometimes more additional reference stations may be needed than one would
like, considering the four additional pages required for each new reference station).
The second limitation of a tide table is that only high and low waters predictions are given. For
predictions of tidal heights between a low water and a high water the user is sent to Table 3 and
forced to use a cosine-based interpolation scheme. However, the tide curve only looks like a cosine
curve for deep-water stations (for example, stations on oceanic islands and continental coasts), and
such an interpolation scheme may not work well for stations inside bays and estuaries and rivers
where nonlinear shallow-water effects distort the tide curve away from a pure cosine. Figure 9.3
307
shows one example of this, but there are others, and in a Tide Table it can only be treated with an
Endnote or a diagram such as in Figure 9.3.
Both these problems can be solved in electronic/digital tide predictions products where for every
the station an entire tide curve is predicted harmonically (at any time interval desired). This
eliminates the need for time and height differences, and it eliminate the need for interpolation
between high and low waters.
9.2.6 The Construction of Tide Tables
Here a short overview is given on the constructions of tide tables (that repeats some of what was
said in Sections 4.4 and 6.8.). There are two primary activities: (1) producing accurate daily
predictions (of high and low waters) for references stations in Table 1 using harmonic analysis and
prediction, and (2) calculating time and height differences for subordinate stations in Table 2 using
nonharmonic analysis. Predictions at a subordinate station are made by applying its time and height
differences to the daily predictions at the reference station that was used in the nonharmonic
analysis.
To produce the daily predictions for a reference station for Table 1 one must have very accurate
tidal harmonic constants, since such predictions will be made for many years in future, and many
subordinate stations in Table 2 will be referenced to that station. Thus, the water level data time
series for a reference station must be long (at least a year) in order to be able to use as many tidal
constituents as possible in the least squares harmonic analysis. If many years of water level data are
available for the station, the most recent will be the best (since bathymetry and shorelines can
change in waterways), but rather than have to revise harmonic constants every year, it makes more
sense to do quality analyses on the predictions (versus observations) each year, to see how well the
present set of harmonic constants is doing. (When one revises the harmonic constants at a reference
station one must then carry out new nonharmonic comparison analyses for all the subordinate
stations referred to that reference station.)
Typically the standard 37 constituents (listed in Table 3.2) are used, but for a station far inside
a very shallow waterway with large tide range, one may need to use additional shallow-water tidal
constituents (the 114 constituents needed for Anchorage, Alaska being an example). However,
when creating a new reference station, or updating an old reference station, one should not feel it
necessary to stick with the standard 37 constituents even for stations with situations less extreme
than Anchorage. One should use the methods mentioned in Section 4.3 for assessing the quality of
the predicted tide series to decide whether more constituents are needed, in which case a time series
longer than a year can be analyzed, if required. By calculating as many tidal constituents as can be
obtained from the available time series, one will probably eliminate the need to use a build-up factor
(see Section 4.3.5). But to assure the use of the maximum number of tidal constituents possible, one
must look carefully at the smallest constituents that come out of the harmonic analysis. In the past
such small constituents have often been rejected because they fell below an (arbitrarily selected)
standard cutoff criterion for acceptance (e.g., 0.03 feet). However, one should run several one-year
harmonic analyses and look at the variation in the amplitude and epoch (phase lag) for these
constituents from year to year. If there is not much variation (especially in the epochs), then such
consistency indicates that these constituents though small are real (and not simply a representation
of quasiperiodic nontidal noise at that frequency). They should be kept and used in the tide
predictions. The inclusion of several small constituents that fall below a standard rejection criterion
(e.g., 0.03 foot historically in NOS) but appear to be consistent from year to year and thus are real,
308
may allow one to resist the temptation to use a build-up factor. If one wishes to go still further in
improving tide predictions, one can also assess the true variation in constituent amplitudes over a
19-year period versus the astronomically determined variation represented in the node factor, f ,
obtained from Table 15 in Schureman (1958). (See Section 2.3.4) One might even decide to use
dozens of additional satellite constituents in the harmonic analysis and prediction, and do away with
the use of node factors (but this requires a 19-year harmonic analysis, and usually does not improve
the predictions enough to be worth the effort).
To put a subordinate station in Table 2 of a Tide Table one must use some type of nonharmonic
analysis method. For many decades the standard procedure for putting the mean time and height
differences into Table 2 has been merely to use the differences between corresponding monthly
mean analyses from the subordinate station and from the reference station. Only for a water level
station with a short data time series was the tide-by-tide analysis (usually called Comparison of
Simultaneous Observations) used. This tide-by-tide analysis also shows the time and height
difference for each tidal cycle, and so one can see how much these differences vary from cycle to
cycle throughout the month. This makes the tide-by-tide analysis valuable as a quality analysis tool.
When one finds significant periodic variations in the time and/or height differences from cycle to
cycle this indicates that the selected reference station does not have similar enough tidal
characteristic to the subordinate station. These periodic variations equate to errors in tide
predictions made with these mean time and height differences (in Table 2) when applied to the daily
tide predictions at that reference station (in Table 1). In such cases, selection of a better reference
station is recommended, which may mean creating a new reference station in that Tide Table to
handle that subordinate station and probably other subordinate stations around it.
9.3 Tidal Current Tables
Tidal Current Tables are not as old as the Tide Tables, but they still have a long history. In the
U.S. the first published tidal current predictions were in 1890 for New York Harbor and vicinity.
The Tidal Current Tables were first published separately from the Tide Tables in 1923, as two
volumes, one for the Atlantic coast of the U.S. and one for the Pacific coast, as they are today. In
the 2007 Tidal Current Tables, Atlantic Coast of North America there are 31 reference stations and
1787 subordinate stations. The 2007 Tidal Current Tables, Pacific Coast of North American and
Asia have 32 reference stations and 1126 subordinate stations. Both include some foreign stations
(18 in the Atlantic tables, and 162 in the Pacific tables).
As with the Tide Tables there are two primary kinds of tables within the Tidal Current Tables.
Table 1 contains daily tidal current predictions for a restricted number of reference stations, which
includes the times and speeds of maximum floods and ebbs and the times of slack water. Table 2
contains time differences and current speed ratios for thousands of subordinate stations. The
predictions at a subordinate station are produced by applying these time and speed ratios to the daily
tidal current predictions at the reference station with the most similar tidal characteristics (usually
the closest reference station geographically). The daily predictions in Table 1 are produced with the
harmonic tidal current prediction equations (see Section 3.4.3 and Chapter 5), using the constituents
calculated from a harmonic analysis of (preferably but not always) a years worth of current data.
The time differences and speed ratios in Table 2 are produced using one of two nonharmonic
methods (the rotary reduction and the reversing reduction), which compare the maximum floods and
ebbs and slack waters (or minimum currents) in a current data time series at the subordinate station
309
with those predicted for the same time period at the most appropriate reference station (see Sections
3.6, 6.5, and 6.6).
In addition to Tables 1 and 2, there is another table (Table 5) to handle rotary tidal currents that
are so circular that one cannot decide on a flood or ebb direction. Such current stations usually are
in open water off the coast. Stations close to shore and in bays and estuaries are also rotary, but their
shape is usually elliptical so that maximum flood and maximum ebb directions are clear.
The following sections will look at Table 1, Table 2, and Table 5, as well as other tables that
have been added to the Tidal Current Tables for other purposes.
9.3.1 Table 1 Daily Predictions for References Stations
Figure 9.5 shows one page of the four pages of a typical Table 1 tidal current reference station,
in this case showing the daily predictions of slack waters and of maximum floods and ebbs for
Chesapeake Bay Entrance, covering January, February, and March 2006. On each day of each
month the predicted times of slacks (first column) and maximums (second column) are provided in
chronological order, maximum floods waters (F) alternating with maximum ebbs (E). The third
column contains the maximum speeds (in knots). The times are given in hours and minutes relative
to the local time meridian (in this case 75o west). To get a current speed at some point between a
slack and a maximum flood or ebb, the user is referred to Table 3 (see Section 9.3.4), which provides
a cosine-based interpolation scheme. This scheme may work reasonably well for a current station
in a deep narrow channel, but often much less well for other stations, since the tidal current curve
is often fairly rotary and is often distorted away from a simple cosine shape by nonlinear shallow-
water effects and lateral inertial effects.
The phases of the moon are also indicated in Table 1 on the days that they occur with symbols
for new moon, first quarter, full moon, and third quarter. The rest of the important astronomical data
for the entire year is provided on the inside of the back cover of the Tidal Current Table, including:
moon in apogee, moon in perigee, moon farthest north of Equator, moon on equator, moon farthest
south of equator, March equinox, June solstice, September equinox, and December solstice (see
Section 2.2.1 for definitions). Chesapeake Bay Entrance is one of 31 reference stations in the Tidal
Current Tables, Atlantic Coast of North America. These reference stations are listed in the front of
the Tide Tables with the length of the data time series from which the harmonic constants were
calculated that were used to make the daily predictions. Chesapeake Bay Entrance was based on
a least squares harmonic analysis using 330 days of current data from 1982. Because of the shortage
of long-term current stations, many other tidal current reference station daily predictions are based
on harmonic constants from shorter time series.
9.3.2 Table 2 Differences and Ratios for Subordinate Stations
Figure 9.6 shows one typical page of the 47 pages of Table 2 in the Tidal Current Tables,
Atlantic Coast of North America. This page provides tidal current information for 30 subordinate
stations that are referenced to Chesapeake Bay Entrance, one station per line. (The page also
includes 13 stations that are referred to the reference station at Delaware Bay Entrance, including
two offshore current stations that were rotary enough that the reader is referred to Table 5.) For each
station the information includes: the depth of the current station, its latitude and longitude, the time
differences for minimum (slack) before flood, maximum flood, minimum (slack) before ebb, and
maximum ebb, the speed ratios for maximum flood and maximum ebb, and the average speeds and
310
Figure 9.5. One page from Table 1 predictions for a tidal current reference station at Chesapeake
Bay Entrance from the Tidal Current Tables, Atlantic Coast of North America.
311
Figure 9.6. A page from Table 2 from the Tidal Current Tables, Atlantic Coast of North America
showing subordinate tidal current stations.
directions for the minimum before flood, maximum flood, minimum before ebb, and maximum ebb.
All speeds are given in knots (1 knot = 51.4 cm/sec). The subordinate stations are split up into
groups of stations referred to the same reference station, which is listed above each group (two
groups are shown on this page). To make a prediction on a particular day at a subordinate station,
the time differences and speed ratios are applied to the daily predictions for that day at the reference
station listed above that group of subordinate stations. For stations with special situations that
cannot be totally handled by this format, there are Endnotes with additional information, and for
stations that are rotary enough there is no clear flood or ebb direction the reader is referred to Table 5.
The time differences and speed ratios in Table 2 were produced using the rotary reduction
analysis and/or the reversing reduction analysis, which are two nonharmonic methods that compare
the slacks/minimums and maximum floods and ebbs in a current data time series at the subordinate
station with the predicted slacks/minimums and maximum floods and ebbs for the same time period
at the most appropriate reference station (see Sections 6.5 and 6.6). The column headings say
Minimum before Flood instead of Slack before Flood and Minimum before Ebb instead of
Slack before Ebb because the more recent current data were analyzed with a computerized rotary
reduction analysis. Most current stations are rotary to some degree, although the ones listed in
Figure 9.6 trace out fairly narrow ellipses so that the minimum current flow perpendicular to the
main flood-ebb axis is only on the order of a tenth of a knot or two. The dashes in those columns
are for older stations that were analyzed using a manual reversing reduction analysis.
312
9.3.3 Table 5 Rotary Tidal Currents
Figure 9.7 shows a typical page of Table 5 in the Tidal Current Tables, Atlantic Coast of North
America, which shows tidal current stations that are very rotary, i.e., the tips of the rotating vector
trace out a very wide ellipse or even a circle. For such current stations there are either no clear flood
and ebb directions, or if such directions can be determined, the minimum flows (perpendicular to
the flood-ebb axis) are not much smaller than the maximum flows. Since most current stations are
rotary to some degree, almost all the stations could be put into Table 5 if one wanted (and that would
serve a useful purpose for seeing how each station rotates, and also how it is distorted by shallow-
water effects). However, usually only offshore stations have been put in Table 5, as well as some
inshore stations with very wide ellipses. For these stations Table 5 gives a current speed and
direction for each hour of a mean tidal current cycle. These hours are relative to a particular time
at a particular reference station. For example, the hourly time increments in Figure 9.7 for Frying
Pan Shoals are after the time of maximum flood at the reference station, Charleston Harbor. (That
reference station had been used in a rotary reduction analysis of the current data at Frying Pan
Shoals, to come up with the twelve speed and direction pairs found in this table for this station.) The
direction of each hourly value is given in degrees true clockwise from north.
9.3.4 Other Tables and Information
There are other small tables in the Tidal Current Tables that provide additional information such
as: a table for determining the speed of current at any time (between maximum and slacks); a table
Figure 9.7. Rotary tidal currents presented in Table 5 of the Tidal Current Tables, Atlantic Coast
of North America.
313
for determining the durations of slack; and descriptive information about coastal tidal currents and
wind-driven currents and how to combine them.
Table 3 provides a method for calculating the tidal current speed at some point between a slack
water and a maximum flow based on cosine interpolation of the speeds of a reversing tidal current.
There are actually two tables, one for most of the stations listed in Table 2, and the second for a few
particular stations. This cosine-based interpolation scheme may not work very well for many current
stations, because in shallow water the tidal current speed curve is distorted away from a cosine shape
and the current direction also rotates around the compass, which is not considered in Table 3.
Table 4 gives information about the duration of slack water at any station in Table 2. Although
slack water is defined as one point along a tidal current speed curve where the speed is exactly
zero, there will be some period of time before and after this zero-speed point that the speeds are so
low that it will appear to be slack water to the mariner. Table 4 helps calculate how long a period
of time that the currents at a particular station will stay below some current speed value (the choices
being, 0.1, 0.2, 0.3, 0.4 or 0.5 knot), based on the maximum current at that particular station. As
with Table 3, there are two tables for this, one for most of the stations listed in Table 2, and the
second for a few particular stations. Table 4 is also based on cosine interpolation of the speeds of
a reversing tidal current and therefore will not necessarily give accurate results for stations with
rotary currents or shallow-water distortions.
9.3.5 Current Diagrams
The Tidal Current Tables, Atlantic Coast of North America also include five examples of another
type of presentation for tidal currents, the current diagram. A current diagram is a graphic that
shows the velocities of the flood and ebb currents (and the time of slacks and maximum floods and
ebbs) over a considerable stretch of the channel of a tidal waterway. At definite intervals along
the channel the current velocities are shown with reference to the times of slack waters (when the
current changes direction from flood to ebb or vice versa) at a particular reference current station.
An example, for Chesapeake Bay, from the entrance (Cape Henry) to Baltimore, a distance of about
150 nautical miles, is shown in Figure 9.8. Along the top and bottom horizontal axes one sees
hourly time intervals referred to slack flood begins (i.e., slack before flood, SBF) or to slack ebb
begins (i.e., slack before ebb, SBE) at the reference tidal current station, in this case Chesapeake Bay
Entrance (CBE). The first six intervals are 3 hours before SBF at CBE, 2 hours before SBF, 1 hour
before SBF, SBF, 1 hour after SBF, and 2 hours after SBF. The next six intervals are 2 hours before
SBE at CBE, 1 hour before SBE, SBE, 1 hour after SBE, 2 hours after SBE, and 3 hours after SBE.
And then the first set of intervals repeats. This added set of time intervals is needed in the diagram,
because the distance from the entrance to Baltimore is so long that it takes more than a tidal cycle
for a particular slack water to progress up the Bay from the entrance to Baltimore. The same tidal
cycle starts progressively later going up the Bay. The changing current speed for each station is
shown on a horizontal line to the right of the station. Flood and ebb phases are shown with
appropriate shading and labeling. The speed lines to the right of the diagram were for the captain
of a ship moving at a particular speed (from 5 to 15 knots, which correspond to the slanted speed
lines, one set for the ship sailing northward and one set for southward) to use to figure out (with
parallel rulers) what the current speeds will be along the ships route.
314
Figure 9.8. A tidal current diagram for Chesapeake Bay included in the Tidal Current Tables,
Atlantic Coast of North America.
315
9.3.6 Limitations of Tidal Current Tables
Tidal Current Tables, like Tide Tables (see Section 9.2.5), are hardcopyproducts that have
been printed for many decades, and only in recent years have been extended to electronic media such
as CDs and the World Wide Web on the Internet. As printed documents these tables have had
certain limitations on what could be included in them, because of page limitations and also because
of the lack of ease with which certain information could be conveyed to a user. With electronic
media those limitations go away, but the traditional format for tidal current tables continues to
survive, both because of its familiarity but also because of the extensive new data that will be
required to produce digital tidal current prediction products based on the harmonic analysis of all
stations (see Section 9.7).
One alternative to this, mentioned in Section 9.2.5, is the use of numerical hydrodynamic models
forced with accurate tide predictions at the entrance based on long-term tide stations. Such a method
for predicting tidal currents has many advantages (see Section 8.3) since they provide tidal current
predictions at hundreds or even thousands of locations, and they can show how the tidal currents
vary spatially (often in a dramatic fashion) over short distances. For example, they can show the
very different speed and directions near the bottom compared with near the mid and surface levels,
as well as the dramatic differences between tidal currents in a channel and in the nearby shallows.
However, a model requires new data, and for more than just the entrance, because of the extensive
validation of the model that is required to make sure it produces accurate predictions everywhere
in the modeled waterway. Thus, a reasonable amount of new data is necessary for numerous
validation sites. This is especially true for current data since the currents can change so dramatically
from place to place, which puts a great pressure on suitable validation.
The same two primary limitations apply to a tidal current table as to a tide table, only they are
usually more serious for a tidal current table (See Section 9.2.5.). The first limitation is that in a
tidal current table (harmonically produced) daily predictions can only be provided for a relatively
few of the thousands of tide stations in the table. This is a page limitation, since the daily
predictions of a reference takes four pages in Table 1, while the time differences and speed ratios
for a subordinate station takes only one line in Table 2, with roughly 50 stations per page. Most
stations must therefore be included as subordinate stations in Table 2. As was seen in Section 3.6.3,
in the introduction to Chapter 6, and in Section 6.9, serious prediction errors can occur if the
harmonic makeup of the subordinate stations is not similar enough to the harmonic makeup of the
reference station. In this situation the mean time differences and speed ratios in Table 2 (that come
out of a nonharmonic comparison analysis) are really good only for certain times of the month.
This can be a very serious problem for tidal currents especially in a waterway with a mixed tidal
current and a strong diurnal signal, because the tidal characteristics will change relatively quickly
with geographic distance, and the subordinate station will often be too far from the reference station.
With tidal currents this can happen much more frequently than with the tide, because for currents
in channel bends and around points sticking out into a waterway, nonlinear lateral inertial effects
transfer energy from diurnal frequencies to semidiurnal frequencies (see Section 2.3.6e and Figure
2.36). In this case tidal current predictions made using Table 2 time differences and speed ratios will
be least accurate around the times of maximum lunar declination, and most accurate near times of
equatorial declination. If this occurs, one solution is to add an additional reference station to the
Tidal Current Table, one that is closer to the subordinate stations and has more similar tidal
characteristics. But for an area like the Strait of Juan de Fuca, the San Juan Islands, and the Strait
of Georgia (see Figures 2.36, 7.2 and 7.3) there is so much variation in diurnal-to-semidiurnal ratios
316
in the area, that dozens of new tidal current reference stations would have to be created to handle
the situation.
The second limitation of a tidal current table is that only predictions for slacks/minimums and
maximum floods and ebbs are given. For predictions of the tidal current speed and direction
between the slacks/minimums and the maximum floods or maximum ebbs the user is sent to Table
3 and forced to use a cosine-based interpolation scheme (only a very few stations are included in
Table 5, for very rotary tidal currents). However, the tidal current curve looks like a cosine for only
a few special cases such as a reversing current in a deep narrow channel or waterway. And even if
Table 3 worked fine for speeds, it does not give the direction of flow, which for most current stations
is different for each hour. Most current stations are rotary to some degree and are distorted by
nonlinear shallow water and lateral inertial effects, for which a cosine curve will not work at all.
For example, of the tidal current examples shown in Figure 2.4 a cosine curve interpolations scheme
might only work for the left-hand bottom example (and then only if the speed curve is not too
distorted by shallow water, which is difficult to tell from that plot).
Both these problems can be solved using electronic/digital tide predictions products where all
the stations are predicted harmonically, which eliminates the need for time differences and speed
ratios, and which allows prediction of an entire tidal current curve for every station (showing all
rotations of direction as well as the distortions by shallow water and lateral inertial effects) so that
no interpolation is needed.
9.3.7 The Construction of Tidal Current Tables
Here a short overview on the constructions of Tidal Current Tables is presented that repeats
some of what was said in Sections 5.6 and Sections 6.8. There are two primary activities: (1)
producing accurate daily predictions (of slacks/minimums and maximum floods and ebbs) for
references stations in Table 1 using harmonic analysis and prediction; and (2) calculating time
differences and speed ratios for subordinate stations in Table 2 using nonharmonic comparison
analysis. Predictions at a subordinate station are made by applying its time differences and speed
ratios to the daily predictions at the reference station that was used in the nonharmonic analysis.
To produce reference stations for Table 1 of the Tidal Current Tables, one should use as long
a current data time series as possible, preferably at a dynamically simple location (such as in the
center of the entrance to the waterway or in the straight section of a channel; away from channel
bends, junctures of channel, points sticking out into the waterway, etc.). It is preferable, if possible,
that this location should be important for navigational purposes, as many mariners will rely mainly
on the reference station and may not go through the adjustment process using Table 2 time
differences and speed ratios. One should carry out a harmonic analysis using as many tidal
constituents as the length of the time series will allow (see Section 5.3.1), perhaps being able to push
this limit in some situations if one needs to include an important tidal constituent (see Section 3.3).
Because of the importance of this reference station, and because predictions for this station will be
produced for many years, one should do a careful quality analysis (see Section 5.5).
Using this reference station one should then analyze the numerous subordinate current stations
nonharmonically. One should use both the rotary reduction analysis and the reversing reduction
analysis on every subordinate station, since they both provide different but necessary information.
Since the tidal current at most locations (other than very narrow channels) is rotary to some degree
(the tip of the current vector approximately tracing out ellipses of various widths), a rotary reduction
must be run to show this rotation in direction over a tidal current cycle. The rotary reduction
317
analysis also shows how the current speeds are distorted by shallow-water effects (so that the speed
curve is often far from looking like a pure cosine curve). The reversing reduction must also be run,
primarily for quality assurance purposes, in order to see how much the time differences and speed
ratios between the subordinate station and the reference station for each tidal cycle change
throughout the month. For example, if there is too much periodic variation from a tidal cycle near
equatorial lunar declination to a tidal cycle near maximum lunar declination, then the tidal
characteristics of the reference station are not similar enough to those at the subordinate station. In
this case a different reference station should be used if possible, or even a new reference station
developed. If neither of these is an option, then one should determine (from the reversing reduction
analysis) how much the time differences and speed ratios will vary (and under what astronomical
conditions) and put that information in an Endnote for Table 2. If there are many stations with this
problem, one can even modify the format of one page of Table 2 for those stations, and add columns
for differences and ratios that apply to, for example, maximum lunar declination.
9.4 Tidal Current Charts
9.4.1 Description of Tidal Current Charts
Typically a set of Tidal Current Charts consists of a chart of the waterway repeated twelve times
(sometimes thirteen), one for each hour of the tidal cycle, each chart having numerous current
vectors showing the speed and direction of the tidal current flow for that hour of the tidal cycle at
various locations in the waterway. (In other countries similar charts can be called Tidal Stream
Atlases or Tide Stream Charts.) These charts allow one to produce tidal current predictions at
numerous locations in a waterway (using a reference station from the Tidal Current Tables), but one
of the main advantages of these charts is their visualization of the tidal current flow over the entire
waterway at hourly intervals throughout a tidal cycle. Figure 9.9 shows an example of (a portion
of) one tidal current chart from the 12 that make up the Upper Chesapeake Bay Tidal Current Charts
(2nd Edition, 1973). Some of the more recent Tidal Current Charts show current vectors for more
than one depth.
The first time that tidal currents (tidal streams) were presented on a chart may have been in 1702
when Edmond Halley (of comet fame) published a chart based on five months of observations that
he made in the English Channel the year before using the Admiralty ship Paramour. What he
actually recorded for many locations in the Channel (and put on the chart) were the times when his
anchored ship swung around when the tidal current changed from flood direction to ebb direction
(the end of the stream that sets to the eastward), as well as the direction of this flood current
(predominantly parallel to the coasts, and thus eastward); these would be essentially slack times if
the currents were not rotary. At numerous locations on his tidal stream chart, the hours for these
turning times were given for the first day of the new moon or full moon (and they were generally
one or two hours earlier than high water times, which were measured on the coasts). It would not
be for another 130 years before others would also produce some type of tidal stream chart. (See
Proudman, 1942)
In the U.S. in the Coast Survey, tidal current charts first appeared in a series of Special
Publications (S.P.) in the 1920s and 1930s, as part of some very thorough and detailed studies of the
tides and tidal currents in particular bays and harbors (which usually also included water temperature
and density data, as well). For example, fairly crude looking tidal current charts appeared in Tide
and Currents in San Francisco Bay by Disney and Overshiner (1925, S.P.115) in their Figures 19
318
9. Products for Presenting Tide and Tidal Current Predictions
through 30. Three years later more precise looking tidal current charts appeared in an S.P. by Paul
Schureman entitled Tide and Currents in Boston Harbor (Schureman, 1928, S.P. 142). In later S.P.s
the tidal current charts became larger and more detailed and required fold-out pages, such as in Tide
and Currents in Long Island and Block Island Sounds by LeLacheur and Sammons (1932, S.P. 174).
In more recent decades Tidal Current Charts were published as separate documents, one set each for
various bays and harbors around the United States.
Each individual tidal current chart (of the 12 or 13 provided for a particular waterway) is
referenced to one or two key points in the tidal current cycle [e.g. slack before flood (SBF) and/or
slack before ebb (SBE), or maximum flood (MF), etc.] or sometimes in the tide cycle [e.g. high
water (HW) and/or low water(LW)]. For some tidal current charts, each chart was for one or more
hours after or before the key point used. For example, the Upper Chesapeake Bay Tidal Current
Charts had 12 charts depicting the currents at: two hours before maximum flood at Baltimore
Harbor Approach (F-2), F-1, F, F+1, F+2, F+3, two hours before maximum ebb (E-2), E-1, E, E+1,
E+2, and E+3. The Narragansett Bay Tidal Current Charts (1st edition, 1971) had 13 charts
depicting the currents at: the time of high water (HW) at Newport, one hour after high water
(HW+1), HW+2, ......, HW+12. Several tidal current charts used slack, flood begins (SFB) and
slack, ebb begins (SEB), but instead of each chart being one or more hours before or after SFB or
SEB, the flood phase (i.e., from the time of SFB to the time of SEB) was divided into six equal
periods, as was the ebb phase. The Boston Harbor Tidal Current Charts had 13 charts depicting the
currents at: SFB, SFB+1, SFB+2, SFB+3, SFB+4, SFB+5, SEB, SEB+1, SEB+2, SEB+3, SEB+4,
SEB+5, and SEB+6, but here the 1, 2, 3, etc, refer to equal intervals after SFB or SEB, not hours.
Making tidal current predictions using a Tidal Current Chart still requires use of a tidal current
reference station (or sometimes a tide reference station) for the waterway depicted on the chart.
Although the tidal current chart applies to any year, each year a new set of daily predictions for the
reference station is needed. Such reference station daily predictions, of course, appear in Table 1
of the annual Tidal Current Table (or Tide Table), but were sometimes provided separately.
To make a prediction of the tidal current at a particular time on a particular day, the user would
go to the daily predictions at the reference station, from which the user first would determine which
of the 12 or 13 tidal current charts is needed to make the prediction. For Chesapeake Bay, which
had hourly charts before and after maximum flood at Baltimore Harbor Approach, one would look
at the reference station predictions for the day of interest and see how many hours before or after
the nearest maximum flood or maximum ebb the time of interest fell and then select the
corresponding chart.
The tidal speeds depicted on the charts are usually for mean springs (i.e., near times of full moon
or new moon). Since the tidal current varies throughout the month (being faster near times of
springs, perigee, and maximum lunar declination, and slower near times of neaps, apogee, and
equatorial declination) a method was used to adjust the spring values shown on the charts for other
astronomical conditions. A factor for correcting the speeds on the selected chart was chosen from
a table on the inside front cover of the Tidal Current Charts, where one looked up the predicted
speed of the nearest maximum flood or maximum ebb at the reference station and then selected the
corresponding correction factor (from the column to the right) and applied it to the speeds on the
selected chart.
319
Figure 9.9. A portion of one of the 12 hourly tidal current charts for the Upper Chesapeake Bay.
320
For the tidal current charts that used equal intervals for the ebb and flood phases (rather than
simply used hours after or before) the procedure was similar to that above, with one difference. To
select the correct chart to use one had to first figure whether the time of interest fell in the flood
phase or the ebb phase. Using the two slack water times that straddled the time of interest to
determine the duration of the flood or ebb phase, and then dividing the duration by six to determine
the start and stop times for each of the six charts for that phase, one selected the chart that the time
of interest fell in. To make this simpler, for some Tidal Current Charts a special set of 12 Tidal
Current Diagrams was published annually, one diagram for each month of the year. On each
diagram the days of the month ran down the left side, and the hours of the day ran across the top
side. By lining up the day of interest and time of interest one could directly obtain the correct tidal
current chart to use, those being indicated by contour lines generally running in parallel from upper
left to low right. (These diagrams should not be confused with the diagrams in Section 9.3.5 or
Section 9.6.)
9.4.2 The Advantages and Limitations of Tidal Current Charts
Since Tidal Current Charts show the tidal current for each hour of a mean tidal current cycle it
has the advantage of being able to accurately show the changing current direction (as well as the
changing current speed) as the current rotates around compass over one tidal cycle. For currents in
narrow channels the ellipses traced out by the tips of the hourly current vectors will be narrow (and
so almost a reversing currents) but for wider waterways and for areas where various channels meet,
the tips of the hourly current vectors can trace out wide ellipses as well as some other shapes (even
figure eights), which can be correctly depicted on the Tidal Current Charts. Also, tidal current
curves are frequently distorted due to shallow-water and lateral inertial effects (the M4/M2 ratio is
often larger in the tidal current than in the tide), and the changing speeds (which will look quite
different than a cosine curve) can also be represented accurately on Tidal Current Charts. Thus,
Tidal Current Charts should be able to provide better tidal current predictions than a Tidal Current
Table, since with the latter, cosine-base interpolations are used to provide the tidal current flows
between the slacks and maximum floods or ebbs that are provided in the Tables. (Unfortunately,
because the reversing reduction analysis was most frequently used during the pre-computer era,
because it took much less manual effort, there appear to have been a few examples of Tidal Current
Charts where the currents, even in wide waterways or areas with complicated geometry, are depicted
as strictly reversing and where cosine interpolation was probably used to produce many charts
representing the times between slacks and maximum floods and ebbs. A rotary reduction analysis
should always be used for analyzing all current stations put on a Tidal Current Chart.)
Tidal Current Charts also have some limitations. For example, they cannot easily be made to
represent mixed tidal currents, that is, tidal currents with a significant diurnal inequality. For such
situations not only does this inequality vary throughout the month (from a minimum near times of
equatorial declination to a maximum near times of maximum lunar declination), but the length of
the flood phase and the length of the ebb phase will also vary throughout the month (often being
quite different for the morning tide versus the afternoon tide). One can try to use the above
mentioned equal-interval method, with each flood or ebb phase divided up into six equal intervals
(instead of simply using hours before or after the slacks), so that all the shorter phases could still be
averaged with the longer phases. However, even if that method produced consistent reasonable
values, it could not be used when the tidal current goes diurnal during part of the month (as can
happen in mixed tidal currents that have a strong diurnal signal).
321
The distortions to the mean tidal current curve caused by nonlinear shallow-water effects or
lateral inertial effects are represented on a tidal current chart, but not easily visualized because the
actual mean tidal current curves are not provided (but probably should be added at the back of the
charts). These distortions, however, also vary throughout the month, and this variation would be
very difficult to represent with some kind of an correction technique.
9.4.3 The Construction of the Tidal Current Charts
9.4.3a Using the Rotary Reduction Analysis the Traditional Method
To construct a set of tidal current charts for a particular waterway one must analyze all available
current data of accepted quality from that waterway. Unless the waterway is very deep the current
data should be fairly recent. Currents are greatly affected by changes in bathymetry, changes in
shoreline or other geographic features, changes in average river discharge (perhaps due to a dam),
etc. Thus, current data from a time period before major dredging or shoaling occurred should not
be used.
One will use the rotary reduction to analyze these current data (reading Section 6.6 may be
necessary to fully appreciate some comments below). One must also have a long enough current
station time series to be able to harmonically analyze it and to produce a reliable reference tidal
current station, predictions from which will be used in the rotary analysis. Often that has already
been done for the Tidal Current Tables, in which case predictions from that reference station will
be used in the analysis. One should also do a few reversing reduction analyses for quality control
purposes, to make sure that the tidal characteristics of each subordinate station are similar enough
to those of the reference stations (i.e., that the time differences and speed ratios do not vary too much
throughout the month). If for that particular waterway a second reference station is required for
some of the subordinate stations, then essentially two sets of tidal current charts will have to be
produced, one set referred to each reference station (but they could perhaps be put on opposite facing
pages in the finished product).
One must next decide which (one or two) key points on a tidal current curve one would like to
use in the rotary analysis (the rotary analysis can handle both one or two key points). As was seen
in Section 9.4.1, in the past different Tidal Current Charts used different key points on a tidal current
curve, for example, the times of maximum flood, the times of maximum floods and ebbs, the slacks
before flood and slacks before ebb, and even the high waters and/or low waters at a nearby tide
gauge (which probably should only be used if there are absolutely no usable tidal current reference
stations; see Section 6.9). The rotary reduction analysis has a tendency to reduce the values of
maximum flood and maximum ebb if the individual maximums (from each tidal cycle) shift around
in time. To minimize this effect it might be better to choose maximum flood (MF) and maximum
ebb (ME) times at the reference station, as the way of breaking the subordinate time series into tidal-
cycle-sections of the data time series that will be superimposed and averaged (in this case, the flood
phase section around maximum flood and the ebb phase section around maximum ebb).
When using the rotary reduction analysis one usually divides these sections of data into half-hour
intervals, but one will probably decide to use the analysis results for only the hour intervals when
creating the actual tidal current charts. These charts might, for example, be for: MF-3, MF-2, MF-1,
MF, MF+1, MF+2, MF+3, ME-3, ME-2, ME-1, ME, ME+1, ME+2, and ME+3, where perhaps only
one of MF+3 or ME-3 might be used (or similarly only one of ME+3 or MF-3). The rotary analysis
puts out mean results, and if one wants to have spring results depicted on the Tidal Current Charts,
322
one will have to adjust the analysis results accordingly (perhaps using the spring-to-mean
relationship at the reference station), or one could selectively run the rotary analysis only for periods
of time surrounding full moon and new moon.
9.4.3b Using Numerical Hydrodynamic Models
High-resolution numerical hydrodynamic models now can do a good job at representing tidal
currents, as long as the resolution is high enough for the particular geographic and bathymetric
situation in the waterways where the currents are being represented. Such models, after being fully
calibrated and verified, can be forced with tide predictions at the waterways entrance in order to
carry out a tidal-only simulation. This simulation should be for at least a month, but could be run
for an entire year, from which specific time periods to be analyzed could be extracted (e.g., perhaps
all the time periods bracketing full and new moon). These simulations could then be analyzed using
a rotary reduction in the same way as described in the previous section. One would have hundreds
(even thousands) of locations to choose from, and those that seem most representative in depicting
the tidal currents throughout the waterway would be chosen. However, a much less labor intensive
(and probably more accurate) approach is to instead select a tidal-cycle-length time period within
the year that is shown to be closest to what would be (e.g.) a mean spring situation, and then simply
directly use the model output for that time period to create the Tidal Current Charts.
A calibrated and verified numerical hydrodynamic model was used by a predecessor
organization of CO-OPS in the National Ocean Service in 1987 to produce the Delaware River and
Bay Tidal Circulation and Water Level Forecast Atlas, which is described in Section 9.5.
To take this modeling approach a step further, one can use it to try to solve one of the primary
limitations of the traditionally produced Tidal Current Charts, i.e., dealing with a mixed tidal current
situation with a strong diurnal signal. One can select a tidal-cycle-length time period that represents
a mean equatorial declination situation and another that represents a mean maximum declination
situation (or one might need to have two maximum declination situations, one for northern
declination and one for souther declination, if they have different effects). Something like this was,
in fact, attempted for the very dynamically complex area of the Strait of Juan de Fuca-Strait of
Georgia, which had a strong diurnal signal and much variance of diurnal-to-semidiurnal ratios and
M4/M2 ratios due to lateral inertial effects (Crean, et al, 1988; Huggett, 1988). But, because of that
complexity, more than a hundred individual tidal current charts had to be produced to try to cover
all possible dynamic situations.
9.5 Tidal Circulation and Water Level Forecast Atlases
The Tidal Circulation and Water Level Forecast Atlas was a new type of product for providing
tide and tidal current predictions first published by the National Ocean Service in 1987 for the
Delaware River and Bay (NOAA, 1987). Its basic design was similar to that of the Tidal Current
Charts, but in addition to providing a tidal current chart for each hour of the tidal cycle, it also
provided a tidal height chart for each hour of the cycle. (See Parker, 1988) (see Figures 9.10 and
9.11)
323
Figure 9.10. One of the hourly tidal height charts from the Delaware River and Bay
Tidal Circulation and Water Level Atlas.
324
Figure 9.11. One of the hourly tidal circulation charts from the
Delaware River and Bay Tidal Circulation and Water Level Atlas.
325
The same methods that allowed fairly easy prediction of the tidal current could now be used to
predict tidal heights, with the same added benefit, namely, that one could more easily visualize the
change in tidal heights over the entire waterway. This was done in three ways, using: (1) tidal height
contours; (2) tide curves for the center of the entire length of the waterway, for mean high water, for
mean tide level, and for the actual water level heights for that hour of the tidal cycle; and (3)
symbols for a few key locations along the waterway that by means of a histogram showed the height
for that hour of the tidal cycle relative to mean high water, mean tide level, and chart datum (i.e.,
mean lower low water).
A calibrated and verified numerical hydrodynamic model was used to produce the atlas (Patchen,
1986). With the hundreds of locations for which the model could predict tidal currents, one could
produce tidal circulation charts with many more current vectors covering more of the waterway.
With the hundreds of locations for which the model could predict tidal heights, one could produce
detailed height contours for the entire waterway.
In addition to the tidal height charts and tidal circulation charts, the atlas also provided corange
and cotidal charts for the tide, and coamplitude and cophase charts for the tidal currents, as well as
a chart showing the bathymetry in the bay. The atlas also provided five years of daily predictions
for the reference station, in this case Breakwater Harbor, Delaware, so that predictions could be
made without having a separate Tide Table or Tidal Current Table.
9.6 The Tide and Light Diagram a bit of history
A special type of product for providing tide predictions was devised in the Coast and Geodetic
Survey during World War II. Tide predictions had always been important for navigating large ships
out of shallow harbors at times of high water, but with the outbreak of World War II, tide predictions
became extremely important for a truly life-or-death purpose amphibious landings. Capturing
enemy-held islands in the Pacific often meant knowing whether there was enough water covering
treacherous coral reefs so that U.S. Marines could safely cross them and reach the beaches. In
Europe the Allies often wanted to land near high tide to reduce the length of beach their soldiers
would have to run across under fire (but actually a little before high tide so that a landing craft
would have time to leave and not be left high and dry when the tide receded).
But often the stages of tide were not the only consideration. The day and time for landing would
be selected based on the combination of a certain stage of the tide occurring at a particular time of
the day (e.g., just after sunrise) after a night with a full moon or a new moon, or some other
combination. The Coast and Geodetic Survey produced Tide and Light Diagrams for most islands
in the Pacific and for North Africa (Zetler, 1991; Hicks, 1967).
An example of a Tide and Light Diagram (from Zetler, 1991) is shown in Figure 9.12 (this one
was for October 1945 and so was never used). It depicted graphically for one month the daily times
of high and low water, moonrise, moonset, sunrise, sunset, and various degrees of twilight. These
top-secret diagrams, which were primarily conceived by Walter Zerbe, Chief of the Predictions
Section at that time, and worked on also by Bernard Zetler, Chief of the Tides Section. The Tide
and Light Diagram became the principal product produced by these Sections during the war, which
grew to 25 people and eventually came under the Joint Army Navy Intelligence Service. Almost
one thousand Tide and Light Diagrams were produced during the war for the Pacific and for North
Africa. For security reasons many Diagrams were produced that the U.S. had no intention of using,
while others were requested on very short notice.
326
Figure 9.12. A Tide and Light Diagram produced during World War II to show the
times of high and low water, along with the times of moonrise, moonset, sunrise,
sunset, and various degrees of twilight. (From Zetler, 1991.)
327
Figure 9.13a. A Tidal Illumination Diagram for LeHarve, France, for July
1944, produced from tide predictions calculated by the British for Allied
forces.
328
In the upper portion of the Tide and Light Diagram (see Figure 9.12) there is a low-resolution
tide curve for the particular month of the diagram, primarily to show the general tidal characteristics,
such as days when there were spring tides or neap tides, or when the tide might become more diurnal
or more semidiurnal. The majority of the diagram is devoted to a two-dimensional graphic on which
exact daily times are provided for high water, low water, moonrise, moonset, sunrise, sunset, and
various degrees of twilight (both morning and evening). Various degrees of darkness, which
depended on the phase of the moon, were outlined by dotted overlays. Using all this information
found on one sheet, military planners could select the best time during the month when both the tide
and light conditions were optimum for a landing operation.
Similar diagrams, called Tidal Illumination Diagrams, were produced by the British for the
Allies to use in Europe, including for the beaches of Normandy for the D-Day landings. Figure
9.13a shows a Tidal Illumination Diagram for LeHarve, France, a little east of Normandy, for July
1944. The explanation of the diagram, found on the back, is shown in Figure 9.13b. The actual tide
predictions used in these diagrams were produced by Arthur Doodson at the Liverpool Observatory
and Tidal Institute and provided to the Commander William Ian Farquharson, the Superintendent
of Tides at the Hydrographic Department of the Admiralty in Bath.
Figure 9.13b. The Explanation of Diagram found on the

back of the Tidal Illumination Diagram in Figure 9.13a.
329
9.7 Digital Tidal Prediction Products
In Sections 9.2.5 and 9.3.6 we looked at the two primary limitations of the standard
hardcopy/paper tidal prediction products, the Tide Tables and Tidal Current Tables.
The first limitation is that nonharmonic comparison methods have to be used for thousands of
subordinate stations in Table 2, because it would take far too many pages to include them as
harmonically predicted reference stations. (A reference station takes four pages in Table 1,while
a subordinate station takes one line of 60 lines on one page in Table 2.) As was seen in Sections
3.6.3 and 6.9, the use of nonharmonic comparison analyses can cause prediction errors if the
reference station does not have tidal characteristics similar enough to the secondary station (which
is most likely to happen in waterways with mixed tides or tidal currents and a strong diurnal signal).
These errors oscillate throughout the month, usually being worse near times of maximum lunar
declination. If this occurs, one solution is to add an additional reference station to the Tide or Tidal
Current Table, one that is closer to the subordinate stations and has more similar tidal characteristics.
But for some areas with complex hydrodynamics there is so much variation in diurnal-to-
semidiurnal ratios in the area, that dozens of new tidal current reference stations would have to be
created to handle the situation.
The second limitation is that only high and low waters predictions are presented in Tide Tables
and only predictions of slacks and maximum floods and ebbs are presented in the Tidal Current
Tables. For predictions at other times (between these key points) the user is forced to use a cosine-
based interpolation scheme. Such an interpolation scheme only works well for deep-water stations.
It does not work well for stations in bays and estuaries and rivers where nonlinear shallow-water
effects distorts the tide or tidal current curve away from a pure cosine shape. It is especially bad for
tidal currents since most are rotary to some degree and the Tidal Current Tables do not show this
rotation in Table 1 or Table 2.
Both these problems can be solved by producing digital tidal predictions products where all the
stations are predicted harmonically, which eliminates the need for Table 2-like time and height
differences and speed ratios. It also allows prediction of the entire tide or tidal current curve for
every station, showing all rotations of direction for tidal currents, as well as the distortions in the tide
and tidal current by nonlinear shallow water and lateral inertial effects (so that no cosine-based
interpolation is needed). Such digital products, whether provided to the user on CDs for use on
their home computer or provided via a special Website on the World Wide Web, provide more
accurate harmonically-based predictions for all locations and times, using a variety of graphical
outputs. If one still wants hardcopy, one can print out a detailed tide or tidal current curve for the
particular place and time of interest, or even special tidal current charts for a specific day.
Predictions from such harmonically-based products are much easier to use than Tide or Tidal
Current tables or even Tidal Current Charts; they would require no calculations or corrections by
the mariner. Such products can also provide additional useful features such as providing predicted
values directly on charts on the screen, as animated results (such as current vectors changing over
time), and even Lagrangian drift calculations (in areas with enough density of current stations).
To produce a digital tidal prediction product, one needs harmonic constants for all water level
and current stations. For most tide stations in the Tide Tables this is not be a problem, because most
usually have long enough time series for adequate harmonic analyses (and most already have
calculated harmonic constants). However, most current stations in the Tidal Current Tables have
short data time series, so producing a digital tidal prediction product can mean the expense of
installing new current sensors for periods of at least a month (although for many of these stations,
330
one might be able to infer some of the harmonic constants). One alternative to this is to produce
long time series of predicted tidal currents using an accurately calibrated and verified high-
resolution three-dimensional numerical hydrodynamic model forced with accurate tide predictions
at the entrance (see Chapter 8). Then selected current time series for all locations of interest (to be
included in the digital tidal prediction product) could be harmonically analyzed.
CO-OPS is, in fact, moving towards electronically delivering harmonically based tidal
predictions for many more locations than just the standard reference stations in the Tide and Tidal
Current Tables. Other nations and private companies are also developing new ways to conveniently
present digital tide and tidal current predictions.
331
332
Appendix
Table A.1. 149 tidal constituents in order of increasing frequency.
Table A.2. 149 tidal constituents in order of increasing synodic period.
See explanation in the detailed key following the second Table.
333
Table A.1 Tidal Constituents In Order of Increasing Frequency
Tidal Origin Cartwright Shallow- Nonlinear wrt

Harmonic of (Doodson) water mechanism, Angular Speed Freq. Period Synodic Const. Cartwright
Constituent Constituent Number equivalent sym or asym (o/hour) (cpd) (hours) Period (name) Potential
Sa solar (*met) 0 0 1 0 0 0 0.0410686 0.0027 8765.8211 365.2 Ssa 0.01156
Ssa solar (*met) 0 0 2 0 0 0 0.0821373 0.0055 4382.9052 365.2 Sa 0.07281
Msm 0 1 -2 1 0 0 0.4715521 0.0314 763.4364 206.0 Mn 0.01579
Mm lunar (*met) 0 1 0 -1 0 0 MN asym 0.5443747 0.0363 661.3092 31.8 MSF 0.08254
MSf lunar 0 2 -2 0 0 0 MS asym 1.0158958 0.0677 354.3671 182.6 Mf 0.01369
Mf lunar 0 2 0 0 0 0 KO asym 1.0980331 0.0732 327.8590 182.6 MSf 0.15647
1 1 -4 2 1 0 0 12.3827652 0.8255 29.0727 1.3 Mf 0.00278
2Q1 lunar 1 -3 0 2 0 0 12.8542862 0.8570 28.0062 13.8 O1 0.00955
1 lunar 1 -3 2 0 0 0 12.9271398 0.8618 27.8484 14.8 Q1 0.01152
Q1 lunar 1 -2 0 1 0 0 NK1 asym 13.3986609 0.8932 26.8684 27.6 O1 0.07217
334
1 lunar 1 -2 2 -1 0 0 13.4715145 0.8981 26.7231 205.9 Q1 0.01371
O1 lunar 1 -1 0 0 0 0 MK1 asym 13.9430356 0.9295 25.8193 13.7 K1 0.37694
1 lunar 1 -1 2 0 0 0 MP1 asym 14.0251729 0.9350 25.6681 23.9 Q1 -0.00493
1 1 0 -2 1 0 0 14.4145548 0.9610 24.9748 193.6 M1 -0.00278
M1 luni-solar 1 0 0 1 0 0 NO1 asym 14.4920521 0.9661 24.8412 27.3 K1 0.02964
1 lunar 1 0 2 -1 0 0 14.5695476 0.9713 24.7091 193.6 M1 -0.00567
1 solar 1 1 -3 0 0 1 TK1 asym 14.9178647 0.9945 24.1321 365.3 P1 0.01023
P1 lunar 1 1 -2 0 0 0 SK1 asym 14.9589314 0.9973 24.0659 182.6 K1 0.17543
S1 solar (*met) 1 1 -1 0 0 0 15.0000000 1.0000 24.0000 365.2 K1 0.00416
K1 luni-solar 1 1 0 0 0 0 MO1 asym 15.0410686 1.0027 23.9345 1.1 M2 0.53011
1 solar 1 1 1 0 0 -1 RP1 asym 15.0821353 1.0055 23.8693 365.2 1 -0.00422
1 solar 1 1 2 0 0 0 2KP1 sym 15.1232058 1.0082 23.8045 182.6 K1 -0.00756
1 lunar 1 2 -2 1 0 0 15.5125897 1.0342 23.2070 205.9 J1 -0.00567
J1 lunar 1 2 0 -1 0 0 MQ1 asym 15.5854433 1.0390 23.0985 27.6 K1 -0.02964
2PO1 shallow water 2PO1 sym 15.9748272 1.0650 22.5355 38.5 J1
SO1 lunar -1 3 -2 0 0 0 SO1 asym 16.0569644 1.0705 22.4202 182.6 2PO1 -0.00492
OO1 lunar 1 3 0 0 0 0 2KO1 sym 16.1391017 1.0759 22.3061 13.7 K1 -0.01624
1 1 4 0 -1 0 0 16.6834764 1.1122 21.5782 27.6 OO1 -0.00078
2MNS2 shallow water 2MNS2 26.4079379 1.7605 13.6323 31.8 2NS2
2NS2 shallow water 2NS2 sym 26.8794590 1.7920 13.3931 14.8 2N2
3M2S2 shallow water 3M2S2 26.9523126 1.7968 13.3569 205.9 2NS2
2NK2S2 shallow water 2NK2S2 26.9615963 1.7974 13.3523 1615.7 3M2S2
OQ2 shallow water OQ2 asym 27.3416965 1.8228 13.1667 182.6 2
2 2 -3 2 1 0 0 MNS2 27.4238337 1.8283 13.1273 31.8 2N2 0.00671
MNK2S2 shallow water MNK2S2 27.5059710 1.8337 13.0881 182.6 2
2MS2K2 shallow water 2MS2K2 27.8039338 1.8536 12.9478 182.6 O2
O2 shallow water O2 asym 27.8860712 1.8591 12.9097 1615.8 2N2
2N2 lunar 2 -2 0 2 0 0 2NM2 sym 27.8953548 1.8597 12.9054 205.9 2 0.02301
2 lunar 2 -2 2 0 0 0 2MS2 sym 27.9682084 1.8645 12.8718 31.8 N2 0.02776
N2 lunar 2 -1 0 1 0 0 28.4397295 1.8960 12.6583 27.6 M2 0.17386
2 lunar 2 -1 2 -1 0 0 28.5125831 1.9008 12.6260 205.9 N2 0.03302
MKL2S2 shallow water MKL2S2 28.5947204 1.9063 12.5897 182.6 2
2KN2S2 shallow water 2KN2S2 28.6040041 1.9069 12.5857 164.1 2
OP2 shallow water OP2 asym 28.9019669 1.9268 12.4559 38.5 2
335
2 2 0 -2 2 0 0 28.9112508 1.9274 12.4519 205.9 M2 -0.00273
2 2 0 -1 0 0 1 SO2 asym 28.9430356 1.9295 12.4382 365.2 M2 -0.00313
M2 lunar 2 0 0 0 0 0 KO2 asym 28.9841042 1.9323 12.4206 ---- *** 0.90809
MKS2 shallow water MKS2 29.0662415 1.9377 12.3855 182.6 M2
2 lunar 2 0 2 0 0 0 M2(KS)2 29.1483788 1.9432 12.3506 91.3 M2 0.00104
2SN(MK)2 shallow water 2SN(MK)2 29.3734880 1.9582 12.2559 182.6 2
2 lunar 2 1 -2 1 0 0 29.4556253 1.9637 12.2218 205.9 L2 -0.00670
L2 lunar 2 1 0 -1 0 0 2MN2 sym 29.5284789 1.9686 12.1916 31.8 S2 -0.02567
2SK2 shallow water 2SK2 sym 29.9178627 1.9945 12.0329 38.5 L2
T2 solar 2 2 -3 0 0 1 29.9589333 1.9973 12.0164 365.3 S2 0.02476
S2 solar 2 2 -2 0 0 0 KP2 asym 30.0000000 2.0000 12.0000 14.8 M2 0.42248
R2 solar 2 2 -1 0 0 -1 SK2 asym 30.0410667 2.0027 11.9836 365.3 S2 -0.00355
K2 luni-solar 2 2 0 0 0 0 30.0821373 2.0055 11.9672 182.6 S2 0.11498
MSN2 shallow water MSN2 30.5443747 2.0363 11.7861 32.5 K2
2 2 3 -2 1 0 0 30.5536572 2.0369 11.7826 205.9 2 0.00123
2 lunar 2 3 0 -1 0 0 KJ2 asym 30.6265119 2.0418 11.7545 27.6 K2 0.00643
2KM(SN)2 shallow water 2KM(SN)2 30.7086493 2.0472 11.7231 48.8 2SM2
2SM2 shallow water 2SM2 sym 31.0158958 2.0677 11.6070 14.8 S2
SKM2 shallow water SKM2 31.0980331 2.0732 11.5763 182.6 2SM2
2NP3 shallow water 2NP3 sym 41.9205276 2.7947 8.5877 32.5 NO3
NO3 shallow water NO3 sym 42.3827651 2.8255 8.4940 27.6 2MK3
2MK3 (MO3) shallow water 2MK3 (MO3) sym (asym) 42.9271398 2.8618 8.3863 13.7 MK3
M3 lunar 3 0 0 0 0 0 43.4761563 2.8984 8.2804 27.3 2MK3 0.01188
NK3 shallow water NK3 sym 43.4807981 2.8987 8.2795 3231.5 M3
SO3 shallow water SO3 sym 43.9430356 2.9295 8.1924 182.6 MK3
MK3 (2MO3) shallow water MK3 (2MO3) asym (sym) 44.0251729 2.9350 8.1771 13.7 2MK3
SP3 shallow water SP3 sym 44.9589314 2.9973 8.0073 16.1 MK3
SK3 shallow water SK3 sym 45.0410686 3.0027 7.9927 182.6 SP3
NO4 shallow water NO4 56.3258007 3.7551 6.3914 27.6 MO4
MO4/3MK4 shallow water MO4/3MK4 56.8701754 3.7913 6.3302 27.1 MN4
N4 shallow water N4 asym 56.8794590 3.7920 6.3292 205.9 3MS4
3MS4 shallow water 3MS4 56.9523127 3.7968 6.3211 31.8 MN4
336
MN4 shallow water MN4 asym 57.4238337 3.8283 6.2692 27.6 M4
M2 shallow water 2MLS4 57.4966873 3.8331 6.2612 205.9 MN4
MNKS4 shallow water MNKS4 57.5059710 3.8337 6.2602 1615.7 M2
SO4 shallow water SO4 57.8860712 3.8591 6.2191 182.6 M4
M4 shallow water M4 asym 57.9682084 3.8645 6.2103 0.5 M2
SN4 shallow water SN4 asym 58.4397295 3.8960 6.1602 31.8 M4
3MN4 shallow water 3MN4 58.5125831 3.9008 6.1525 205.9 SN4
KN4 shallow water KN4 58.5218668 3.9015 6.1515 32.5 MS4
MS4 shallow water MS4 asym 58.9841042 3.9323 6.1033 14.8 M4
MK4 shallow water MK4 59.0662415 3.9377 6.0949 182.6 MS4
SL4 shallow water SL4 asym 59.5284789 3.9686 6.0475 32.5 MK4
S4 shallow water S4 asym 60.0000000 4.0000 6.0000 14.8 MS4
MNO5 shallow water MNO5 71.3668693 4.7578 5.0444 27.6 2MO5
2MO5/3MO5 shallow water 2MO5/3MO5 sym 71.9112440 4.7941 5.0062 13.7 2MK5
3MP5 shallow water 3MP5 sym 71.9933813 4.7996 5.0005 182.6 2MO5
MNK5 shallow water MNK5 72.4649023 4.8310 4.9679 27.6 2MK5
2MP5 shallow water 2MP5 sym 72.9271398 4.8618 4.9364 182.6 2MK5
2MK5/3MO5 shallow water 2MK5/3MO5 sym 73.0092770 4.8673 4.9309 1.1 M6
MSK5 shallow water MSK5 74.0251728 4.9350 4.8632 14.8 2MK5

3KM5 shallow water 3KM5 74.1073100 4.9405 4.8578 182.6 MSK5
N6 shallow water N6 sym 85.3191885 5.6879 4.2194 205.9 3MNS6
3MNS6 shallow water 3MNS6 85.3920421 5.6928 4.2158 1615.7 3NKS6
3NKS6 shallow water 3NKS6 85.4013258 5.6934 4.2154 32.5 2NM6
2NM6 shallow water 2NM6 sym 85.8635632 5.7242 4.1927 205.9 4MS6
4MS6 shallow water 4MS6 sym 85.9364168 5.7291 4.1891 14.8 M6
2NMKS6 shallow water 2NMKS6 85.9457005 5.7297 4.1887 1615.7 4MS6
2MN6 shallow water 2MN6 sym 86.4079380 5.7605 4.1663 27.6 M6
2M6 shallow water 2M6 sym 86.4807916 5.7654 4.1628 205.9 2MN6
2MNKS6 shallow water 2MNKS6 86.4900752 5.7660 4.1623 32.5 M6
M6 shallow water M6 sym 86.9523127 5.7968 4.1402 0.5 M4
MSN6 shallow water MSN6 87.4238337 5.8283 4.1179 205.9 4MN6
4MN6 shallow water 4MN6 87.4966873 5.8331 4.1144 27.6 M6
MKN6 shallow water MKN6 87.5059710 5.8337 4.1140 1615.7 4MN6
337
2MK6 shallow water 2MK6 88.0503457 5.8700 4.0886 182.6 2MS6
3MSN6 shallow water 3MSN6 88.5125831 5.9008 4.0672 32.5 2MK6
NSK6 shallow water NSK6 88.5218668 5.9015 4.0668 32.5 2SM6
2SM6 shallow water 2SM6 sym 88.9841042 5.9323 4.0457 7.4 M6
MSK6 shallow water MSK6 89.0662415 5.9377 4.0419 7.1 M6
S6 shallow water S6 sym 90.0000000 6.0000 4.0000 4.9 M6
4MO7/3MK7 shallow water 4MO7/3MK7 sym 101.9933600 6.7996 3.5296 13.7 4MK7
4MJ7 shallow water 4MJ7 sym 100.3509735 6.6901 3.5874 9.1 4MO7
2MNK7 shallow water 2MNK7 100.9046318 6.7270 3.5677 27.1 4MJ7
4M7 shallow water 4M7 sym 101.9112440 6.7941 3.5325 14.9 2MNK7
4M7 shallow water 4M7 sym 103.0092771 6.8673 3.4948 13.7 4MTAU7
2(MN)8 shallow water 2(MN)8 114.8476674 7.6565 3.1346 27.6 3MN8
3MNKS8 shallow water 3MNKS8 115.4741794 7.6983 3.1176 182.6 3MN8
M8 shallow water M8 both 115.9364169 7.7291 3.1052 0.5 M6
2MSN8 shallow water 2MSN8 116.4079380 7.7605 3.0926 182.6 2MNK8
2MNK8 shallow water 2MNK8 116.4900752 7.7660 3.0904 32.5 3MS8
3MS8 shallow water 3MS8 116.9523127 7.7968 3.0782 14.8 M8
MSNK8 shallow water MSNK8 117.5059710 7.8337 3.0637 31.8 3MK8
2(MS)8 shallow water 2(MS)8 117.9682084 7.8645 3.0517 14.8 3MS8
2MSK8 shallow water 2MSK8 118.0503457 7.8700 3.0495 182.6 2(MS)8
2M2NK9 shallow water 2M2NK9 129.8887360 8.6592 2.7716 13.8 4MK9
3MNK9 shallow water 3MNK9 130.4331108 8.6955 2.7600 27.6 4MK9
4MK9 shallow water 4MK9 130.9774855 8.7318 2.7486 1.0 M8
3MSK9 shallow water 3MSK9 131.9933813 8.7996 2.7274 14.8 4MK9
M10 shallow water M10 144.9205211 9.6614 2.4841 0.5 M8
3MNS10 shallow water 3MNS10 145.3920422 9.6928 2.4761 27.6 4MS10
338
2MNSK10 shallow water 2MNSK10 146.4900752 9.7660 2.4575 27.1 4MS10
2MNSK1
3M2S10 shallow water 3M2S10 146.9523127 9.7968 2.4498 32.5 0
4MSK11 shallow water 4MSK11 160.9774855 10.7318 2.2363 0.9 M10

3MNKS12 shallow water 3MNKS12 175.4741794 11.6983 2.0516 32.5 4M2S12
4M2S12 shallow water 4M2S12 175.9364169 11.7291 2.0462 14.8 5MS12
Table A.2 Tidal Constituents In Order of Increasing Synodic Period
Tidal Origin Cartwright Shallow- Nonlinear wrt

Harmonic of (Doodson) water mechanism, Angular Speed Freq. Period Synodic Const. Cartwright
Constituent Constituent Number equivalent sym or asym (o/hour) (cpd) (hours) Period (name) Potential
M2 lunar 2 0 0 0 0 0 KO2 asym 28.9841042 1.9323 12.4206 *** *** 0.90809
M4 shallow water M4 asym 57.9682084 3.8645 6.2103 0.5 M2
M6 shallow water M6 sym 86.9523127 5.7968 4.1402 0.5 M4
M8 shallow water M8 both 115.9364169 7.7291 3.1052 0.5 M6
M10 shallow water M10 144.9205211 9.6614 2.4841 0.5 M8
4MSK11 shallow water 4MSK11 160.9774855 10.7318 2.2363 0.9 M10
K1 luni-solar 1 1 0 0 0 0 MO1 asym 15.0410686 1.0027 23.9345 1.1 M2 0.53011
4MK9 shallow water 4MK9 130.9774855 8.7318 2.7486 1.0 M8
339
1 1 -4 2 1 0 0 12.3827652 0.8255 29.0727 1.3 Mf 0.00278
S6 shallow water S6 sym 90.0000000 6.0000 4.0000 4.9 M6
MSK6 shallow water MSK6 89.0662415 5.9377 4.0419 7.1 M6
2SM6 shallow water 2SM6 sym 88.9841042 5.9323 4.0457 7.4 M6
4MJ7 shallow water 4MJ7 sym 100.3509735 6.6901 3.5874 9.1 4MO7
O1 lunar 1 -1 0 0 0 0 MK1 asym 13.9430356 0.9295 25.8193 13.7 K1 0.37694
OO1 lunar 1 3 0 0 0 0 2KO1 sym 16.1391017 1.0759 22.3061 13.7 K1 -0.01624
2Q1 lunar 1 -3 0 2 0 0 12.8542862 0.8570 28.0062 13.8 O1 0.00955
2MK3 (MO3) shallow water 2MK3 (MO3) sym (asym) 42.9271398 2.8618 8.3863 13.7 MK3
MK3 (2MO3) shallow water MK3 (2MO3) asym (sym) 44.0251729 2.9350 8.1771 13.7 2MK3
2MO5/3MO5 shallow water 2MO5/3MO5 sym 71.9112440 4.7941 5.0062 13.7 2MK5
4MO7/3MK7 shallow water 4MO7/3MK7 sym 101.9933600 6.7996 3.5296 13.7 4MK7
4M7 shallow water 4M7 sym 103.0092771 6.8673 3.4948 13.7 4MTAU7
2M2NK9 shallow water 2M2NK9 129.8887360 8.6592 2.7716 13.8 4MK9
S2 solar 2 2 -2 0 0 0 KP2 asym 30.0000000 2.0000 12.0000 14.8 M2 0.42248
2SM2 shallow water 2SM2 sym 31.0158958 2.0677 11.6070 14.8 S2
3MSK9 shallow water 3MSK9 131.9933813 8.7996 2.7274 14.8 4MK9
2NS2 shallow water 2NS2 sym 26.8794590 1.7920 13.3931 14.8 2N2
S4 shallow water S4 asym 60.0000000 4.0000 6.0000 14.8 MS4
MSK5 shallow water MSK5 74.0251728 4.9350 4.8632 14.8 2MK5
MS4 shallow water MS4 asym 58.9841042 3.9323 6.1033 14.8 M4
1 lunar 1 -3 2 0 0 0 12.9271398 0.8618 27.8484 14.8 Q1 0.01152
2(MS)8 shallow water 2(MS)8 117.9682084 7.8645 3.0517 14.8 3MS8
4M7 shallow water 4M7 sym 101.9112440 6.7941 3.5325 14.9 2MNK7
4M2S12 shallow water 4M2S12 175.9364169 11.7291 2.0462 14.8 5MS12
SP3 shallow water SP3 sym 44.9589314 2.9973 8.0073 16.1 MK3
1 lunar 1 -1 2 0 0 0 MP1 asym 14.0251729 0.9350 25.6681 23.9 Q1 -0.00493
M1 luni-solar 1 0 0 1 0 0 NO1 asym 14.4920521 0.9661 24.8412 27.3 K1 0.02964
lunar 3 0 0 0 0 0 43.4761563 2.8984 8.2804 27.3 0.01188
340
M3 2MK3
2MNSK10 shallow water 2MNSK10 146.4900752 9.7660 2.4575 27.1 4MS10
2MNK7 shallow water 2MNK7 100.9046318 6.7270 3.5677 27.1 4MJ7
MO4/3MK4 shallow water MO4/3MK4 56.8701754 3.7913 6.3302 27.1 MN4
N2 lunar 2 -1 0 1 0 0 28.4397295 1.8960 12.6583 27.6 M2 0.17386
J1 lunar 1 2 0 -1 0 0 MQ1 asym 15.5854433 1.0390 23.0985 27.6 K1 -0.02964
Q1 lunar 1 -2 0 1 0 0 NK1 asym 13.3986609 0.8932 26.8684 27.6 O1 0.07217
MNK5 shallow water MNK5 72.4649023 4.8310 4.9679 27.6 2MK5
MNO5 shallow water MNO5 71.3668693 4.7578 5.0444 27.6 2MO5
2MN6 shallow water 2MN6 sym 86.4079380 5.7605 4.1663 27.6 M6
MN4 shallow water MN4 asym 57.4238337 3.8283 6.2692 27.6 M4
NO4 shallow water NO4 56.3258007 3.7551 6.3914 27.6 MO4
2(MN)8 shallow water 2(MN)8 114.8476674 7.6565 3.1346 27.6 3MN8
2 lunar 2 3 0 -1 0 0 KJ2 asym 30.6265119 2.0418 11.7545 27.6 K2 0.00643
1 1 4 0 -1 0 0 16.6834764 1.1122 21.5782 27.6 OO1 -0.00078
3MNK9 shallow water 3MNK9 130.4331108 8.6955 2.7600 27.6 4MK9
NO3 shallow water NO3 sym 42.3827651 2.8255 8.4940 27.6 2MK3
L2 lunar 2 1 0 -1 0 0 2MN2 sym 29.5284789 1.9686 12.1916 31.8 S2 -0.02567
2 lunar 2 -2 2 0 0 0 2MS2 sym 27.9682084 1.8645 12.8718 31.8 N2 0.02776
Mm lunar (*met) 0 1 0 -1 0 0 MN asym 0.5443747 0.0363 661.3092 31.8 MSF 0.08254
2MNS2 shallow water 2MNS2 26.4079379 1.7605 13.6323 31.8 2NS2
3MS4 shallow water 3MS4 56.9523127 3.7968 6.3211 31.8 MN4
MSNK8 shallow water MSNK8 117.5059710 7.8337 3.0637 31.8 3MK8
2 2 -3 2 1 0 0 MNS2 27.4238337 1.8283 13.1273 31.8 2N2 0.00671
SN4 shallow water SN4 asym 58.4397295 3.8960 6.1602 31.8 M4
MSN2 shallow water MSN2 30.5443747 2.0363 11.7861 32.5 K2
3MNKS12 shallow water 3MNKS12 175.4741794 11.6983 2.0516 32.5 4M2S12
2MNKS6 shallow water 2MNKS6 86.4900752 5.7660 4.1623 32.5 M6
3M2S10 shallow water 3M2S10 146.9523127 9.7968 2.4498 32.5 2MNSK10
341
3NKS6 shallow water 3NKS6 85.4013258 5.6934 4.2154 32.5 2NM6
3MSN6 shallow water 3MSN6 88.5125831 5.9008 4.0672 32.5 2MK6
SL4 shallow water SL4 asym 59.5284789 3.9686 6.0475 32.5 MK4
2NP3 shallow water 2NP3 sym 41.9205276 2.7947 8.5877 32.5 NO3
NSK6 shallow water NSK6 88.5218668 5.9015 4.0668 32.5 2SM6
2MNK8 shallow water 2MNK8 116.4900752 7.7660 3.0904 32.5 3MS8
KN4 shallow water KN4 58.5218668 3.9015 6.1515 32.5 MS4
OP2 shallow water OP2 asym 28.9019669 1.9268 12.4559 38.5 2
2SK2 shallow water 2SK2 sym 29.9178627 1.9945 12.0329 38.5 L2
2PO1 shallow water 2PO1 sym 15.9748272 1.0650 22.5355 38.5 J1
2KM(SN)2 shallow water 2KM(SN)2 30.7086493 2.0472 11.7231 48.8 2SM2
2 lunar 2 0 2 0 0 0 M2(KS)2 29.1483788 1.9432 12.3506 91.3 M2 0.00104
2KN2S2 shallow water 2KN2S2 28.6040041 1.9069 12.5857 164.1 2
P1 lunar 1 1 -2 0 0 0 SK1 asym 14.9589314 0.9973 24.0659 182.6 K1 0.17543
K2 luni-solar 2 2 0 0 0 0 30.0821373 2.0055 11.9672 182.6 S2 0.11498
Mf lunar 0 2 0 0 0 0 KO asym 1.0980331 0.0732 327.8590 182.6 MSf 0.15647
3MP5 shallow water 3MP5 sym 71.9933813 4.7996 5.0005 182.6 2MO5
MKS2 shallow water MKS2 29.0662415 1.9377 12.3855 182.6 M2
1 solar 1 1 2 0 0 0 2KP1 sym 15.1232058 1.0082 23.8045 182.6 K1 -0.00756
SO1 lunar -1 3 -2 0 0 0 SO1 asym 16.0569644 1.0705 22.4202 182.6 2PO1 -0.00492
3KM5 shallow water 3KM5 74.1073100 4.9405 4.8578 182.6 MSK5
3MNKS8 shallow water 3MNKS8 115.4741794 7.6983 3.1176 182.6 3MN8
2SN(MK)2 shallow water 2SN(MK)2 29.3734880 1.9582 12.2559 182.6 2
SO4 shallow water SO4 57.8860712 3.8591 6.2191 182.6 M4
MKL2S2 shallow water MKL2S2 28.5947204 1.9063 12.5897 182.6 2
SK3 shallow water SK3 sym 45.0410686 3.0027 7.9927 182.6 SP3
MK4 shallow water MK4 59.0662415 3.9377 6.0949 182.6 MS4
OQ2 shallow water OQ2 asym 27.3416965 1.8228 13.1667 182.6 2
342
SO3 shallow water SO3 sym 43.9430356 2.9295 8.1924 182.6 MK3
SKM2 shallow water SKM2 31.0980331 2.0732 11.5763 182.6 2SM2
2MSK8 shallow water 2MSK8 118.0503457 7.8700 3.0495 182.6 2(MS)8
2MSN8 shallow water 2MSN8 116.4079380 7.7605 3.0926 182.6 2MNK8
2MS2K2 shallow water 2MS2K2 27.8039338 1.8536 12.9478 182.6 O2
MNK2S2 shallow water MNK2S2 27.5059710 1.8337 13.0881 182.6 2
1 1 0 -2 1 0 0 14.4145548 0.9610 24.9748 193.6 M1 -0.00278
1 lunar 1 0 2 -1 0 0 14.5695476 0.9713 24.7091 193.6 M1 -0.00567
2N2 lunar 2 -2 0 2 0 0 2NM2 sym 27.8953548 1.8597 12.9054 205.9 2 0.02301
1 lunar 1 -2 2 -1 0 0 13.4715145 0.8981 26.7231 205.9 Q1 0.01371
2 lunar 2 -1 2 -1 0 0 28.5125831 1.9008 12.6260 205.9 N2 0.03302
2 lunar 2 1 -2 1 0 0 29.4556253 1.9637 12.2218 205.9 L2 -0.00670
MSN6 shallow water MSN6 87.4238337 5.8283 4.1179 205.9 4MN6
3M2S2 shallow water 3M2S2 26.9523126 1.7968 13.3569 205.9 2NS2
2M6 shallow water 2M6 sym 86.4807916 5.7654 4.1628 205.9 2MN6
1 lunar 1 2 -2 1 0 0 15.5125897 1.0342 23.2070 205.9 J1 -0.00567
N6 shallow water N6 sym 85.3191885 5.6879 4.2194 205.9 3MNS6
2 2 0 -2 2 0 0 28.9112508 1.9274 12.4519 205.9 M2 -0.00273
3MN4 shallow water 3MN4 58.5125831 3.9008 6.1525 205.9 SN4
M2 shallow water 2MLS4 57.4966873 3.8331 6.2612 205.9 MN4
N4 shallow water N4 asym 56.8794590 3.7920 6.3292 205.9 3MS4
2 2 3 -2 1 0 0 30.5536572 2.0369 11.7826 205.9 2 0.00123
2NM6 shallow water 2NM6 sym 85.8635632 5.7242 4.1927 205.9 4MS6
Msm 0 1 -2 1 0 0 0.4715521 0.0314 763.4364 206.0 Mn 0.01579
Sa solar (*met) 0 0 1 0 0 0 0.0410686 0.0027 8765.8211 365.2 Ssa 0.01156
Ssa solar (*met) 0 0 2 0 0 0 0.0821373 0.0055 4382.9052 365.2 Sa 0.07281
R2 solar 2 2 -1 0 0 -1 SK2 asym 30.0410667 2.0027 11.9836 365.3 S2 -0.00355
T2 solar 2 2 -3 0 0 1 29.9589333 1.9973 12.0164 365.3 S2 0.02476
S1 solar (*met) 1 1 -1 0 0 0 15.0000000 1.0000 24.0000 365.2 K1 0.00416
2 2 0 -1 0 0 1 SO2 asym 28.9430356 1.9295 12.4382 365.2 M2 -0.00313
1 solar 1 1 1 0 0 -1 RP1 asym 15.0821353 1.0055 23.8693 365.2 1 -0.00422
343
1 solar 1 1 -3 0 0 1 TK1 asym 14.9178647 0.9945 24.1321 365.3 P1 0.01023
MNKS4 shallow water MNKS4 57.5059710 3.8337 6.2602 1615.7 M2
3MNS6 shallow water 3MNS6 85.3920421 5.6928 4.2158 1615.7 3NKS6
2NK2S2 shallow water 2NK2S2 26.9615963 1.7974 13.3523 1615.7 3M2S2
2NMKS6 shallow water 2NMKS6 85.9457005 5.7297 4.1887 1615.7 4MS6
MKN6 shallow water MKN6 87.5059710 5.8337 4.1140 1615.7 4MN6
O2 shallow water O2 asym 27.8860712 1.8591 12.9097 1615.8 2N2
NK3 shallow water NK3 sym 43.4807981 2.8987 8.2795 3231.5 M3
Tables A.1 and A.2. 149 tidal harmonic constituents listed first in order of increasing frequency (Table A.1) and then in order of
increasing synodic period (Table A.2).
103 of these constituents are shallow-water constituents up through the 12th diurnals. There are a few cases where the astronomical
constituent is listed, but the shallow-water constituent at the same frequency can often be larger (e.g., 2MN2 is larger than L2 in shallow
bays and estuaries). No satellite constituents (representing the 18.6-year nodal cycle) are included in these tables (see table 4.1, and
Zetler et al, 1985) .
In the 2nd column (origin of constituent), the term shallow water indicates that the constituent was produced by nonlinear
mechanisms in shallow water. A (*met) indicates that the calculated harmonic constants for this constituent usually include mostly
quasi-periodic meteorological effect energy. The 4th column (shallow-water equivalent) gives the shallow-water constituent with the
same frequency as the astronomical constituent listed in the 1st column (or if the constituent is only a shallow-water constituent, it simply
repeats the name). The 5th column (NL mechanism, sym or asym) indicates whether the shallow-water constituent is produced by the
symmetric nonlinear mechanism (u*u*) or by the asymmetric nonlinear mechanisms (M(u)/Mx, uMu/Mx, and u*u*) as indicated in Table
2.4 and discussed in Section 7.6. Constituents listed as asymmetric can also be produced by the lateral initial terms as discussed in Section
7.6.7. (A sym or asym assessment is not given for higher order interactions, especially involving more than two constituents.)
The 3rd column (Cartwright number) gives the six digits that indicate which combination of astronomical frequencies (from
Table 2.1) produces that astronomical tidal constituent (examples are shown in Table 2.2). The classic Doodson numbers are the same
344
as the Cartwright numbers but with a 5 added to each digit (except the first), which Doodson did to keep these digits from being negative.
These six digits are the multiplying coefficients in front of the six frequencies L, and 1 through 5 from Table 2.1 (the classic Doodson
labeling for these six astronomical frequencies was , S, h p, N, p1) . The angular speeds (frequencies) for the six astronomical motions
listed in Table 3 2.1 are (in o/hour): L=14.49205236, 1=0.54901656, 2= 0.0410688, 3 =0.00011408, 4=0.0045422 ,
5=0.00220644 (Table 2.1 gives the equivalent astronomical periods). As an example, the Cartwright number for N2 is 2 -1 0 1 0 0; to
obtain the angular speed for N2 one multiplies L by 2, subtracts 1 from it, and then adds 3. Cartwright numbers are not given for
shallow-water constituents (even though a combination of Cartwright numbers could be constructed that would produce the correct
frequency of that constituent) because shallow-water constituents are not produced directly by a combination of those six astronomical
frequencies; they are produced hydrodynamically by energy transferred from other constituents through the nonlinear mechanisms
discussed in Sections 2.3.2 and 7.6.2.
The 6th column gives the angular speed of the constituent in degrees per hour. Dividing the angular speed by 360o (= one cycle)
and multiplying by 24 (hours) gives the frequency of the constituent in cycles per day (cpd) shown in the 7th column. Dividing 360o by
the angular speed gives the period of the constituent in hours shown in the 8th column.
The 9th column gives the synodic period for the constituent with respect to (wrt) the constituent which is closest in frequency
among the constituents with a larger amplitude (shown in the 10th column). Table 3.1 shows several synodic periods for each constituent,
each paired with another constituent, but among the possibilities only the longest synodic period for a larger constituent is shown in the
9th column. The synodic period given in the 9th column depends on which larger nearby constituent (listed in the 10th column) it is decided
is most important; for the shallow-water constituents it may not always be clear which nearby constituent should be in the 10th column
(unless one has a very high resolution spectra to look at).
The 11th column shows a coefficient of the relative strength of the tide potential (the tide producing force) for each astronomical
constituent as determined by Cartwright and Edden (1973), in order to give one a first guess at the expected relative sizes of the tidal
constituents (at least the astronomical ones). However, hydrodynamics can change these relative strengths. (In Table 3.2, the 12th column
gives, as just one example, the constituent amplitudes for a water level station at Trenton, NJ, on the Delaware River. In this case, N2
is larger than S2, unlike with the tide producing forces. Also, for that station one is really calculating 2MN2 and not L2, and likewise 2MS2
not 2.) This Cartwright potential is not given for the shallow-water constituents, since they are not produced directly from the
astronomical tide producing forces.
The relative size of a shallow-water term will be different for every hydrodynamic situation, but as an approximate estimate can
be initially based on: (1) the sizes of the astronomical constituents from which it was produced; (2) the particular nonlinear mechanism
that produced it; and (3) and the order of the nonlinear interaction (i.e., was it a first-order interaction or second-order, or even a third-
order). Generally one can guess that a shallow-water constituent will be larger (compared with other shallow-water constituents) if its
name includes fewer letters, and one or more of those letters represents one of the larger astronomical constituents. The examples of
compound tides and overtides shown in Table 2.4 were all first-order terms involving only two constituents at a time (one of them M2),
and thus are likely to be the most important of the shallow-water constituents. Compound tides and overtides similarly generated with
345
K1 or O1 are also likely to be important.
In the past ,shallow-water frequencies were found by simply adding and/or subtracting the frequencies for two, three, and even
four astronomical frequencies. A large number of combinations were possible, and the choices did not always represent a particular
nonlinear generating mechanism. To choose which shallow-water constituents should be included in Tables A.1 and A.2, first various
references were looked at including Godin (1988), Foreman (2004), Zetler et al (1967), and Rossiter and Lennon (1967), and others (most
references agreed on most of the astronomical frequencies, but there was not agreement on which shallow-water constituents to use);
then the nonlinear generating mechanisms were considered (such as in Table 2.4). However, Tables A.1 and A.2 should still not be
considered definitive, and a more precise and thorough Fourier decomposition (similar to that in Section 7.6) needs to be done. However,
even with that done, the relative sizes of the shallow-water constituents, and thus the order in which they should be included in a harmonic
analysis, will vary depending on the sizes of the astronomical constituents in the particular waterway from where the data came.
346
References
Aikman, F. III, G.L. Mellor, T. Ezer, D. Sheinin, P. Chen, L. Breaker, K. Bosley, and D.B. Rao,
1996. Towards an operational nowcast/forecast system for the U.S. East Coast. In:
Modern Approaches to Data Assimilation in Ocean Modeling, P. Malanotte-Rizzoli (ed.),
Elsevier Oceanography Series, 61, pp. 347-376.
Andersen, O.B., P.L. Woodworth, and R.A. Flather, 1995. Intercomparison of recent ocean tide
models. Journal of Geophysical Research, 100(C12):25261-25282.
Amin, M., 1976. The fine resolution of tidal harmonics. Geophysical J. Royal Astronomical
Soc., 44:293-310.
Amin, M., 1977. Some recent investigations into the harmonic shallow water corrections
method of tidal predictions. International Hydrographic Review, 56(1): 87-108.
Amin, M., 1991. Superfine resolution of tidal harmonic constants. In: Tidal Hydrodynamics,
(ed.) B.B. Parker, John Wiley & Sons, New York, pp 771-724.
Appell, G.F., T.N. Mero, and T.D. Bethem, and G.W. French, 1994. The development of a
real-time port information system, IEEE Journal of Oceanic Engineering, 19(2):149-
157.
Batchelor, G.K., 1977. An Introduction to Fluid Mechanics, Cambridge University Press,
Cambridge, UK., 615 pages.
Bennett, A.F., 1992. Inverse Methods in Physical Oceanography. Cambridge University Press,
Cambridge, UK..
Blackman, R.B. and J.W. Tukey, 1959. The Measurement of Power Spectra: from the point of
view of communications engineering, Dover Publications, New York.
Blumberg, A.F. and G.L. Mellor, 1987. A Description of a Three-Dimensional Coastal Ocean
Circulation Model. In: Three-Dimensional Coastal Ocean Models, (Ed.) N. Heaps, pp 1-
16, American Geophys. Union.
Bosley, K T and K.W. Hess, 1997. Development of an Experimental Nowcast/Forecast System
for Chesapeake Bay Water Levels. Estuarine and Coastal Modeling, pp. 413-426.
Bowden, K.F., 1953. Note on wind drift in a channel in the presence of tidal currents, Proc.
Royal Soc. London, A219:425-446.
Burg, J.P., 1972. The relationship between maximum entropy spectra and maximum likelihood
spectra. Geophysics, 37:375-6.
Cartwright, D.E., 1968. A unified analysis of tides and surges round north and east Britain.
Philosophical Transactions Royal Society London, A263:1-55.
Cartwright, D.E., 1977. Oceanic tides. Reports On Progress In Physics, 40:665-708.
Cartwright, D.E., 1991. Detection of tides from artificial satellites (review). In: Tidal
Hydrodynamics, (ed) B.B. Parker, John Wiley & Sons, New York, pp 547-567.
347
Cartwright, D.E., 1999. Tides A Scientific History, Cambridge University Press, Cambridge,
UK, 292 pages.
Cartwright, D.E. and M. Amin, 1986. The variances of tidal harmonics. Deutsche Hyrogr. Zeit.,
6:235-253.
Cartwright, D.E. and R.J. Tayler, 1971. New computations of the tide-generating potential.
Geophys. J. Royal Astronomical Society, 23:45-74.
Cartwright, D.E. and Anne C. Edden, 1973. Corrected tables of tidal harmonics. Geophys. J.
Royal Astronomical Society, 33:253-264.
Chen, P, and G. Mellor, 1999. Determination of tidal boundary forcing using tide station data.
In: Coastal Ocean Prediction, Christopher N.K. Mooers (Ed.), Coastal and Estuarine
Studies, 56, AGU Publications, Washington, D.C., pp 329-351..
Cheney, R., L. Miller, R. Agreen, N. Doyle, and J. Lillibridge, 1994. TOPEX/POSEIDON: the
2-cm solution. Journal of Geophysical Research, 99(C12):24555-24563.
Cheng, Ralph T., Jon R. Burau, and Jeff W. Gartner, 1991. Interfacing data analysis and
numerical modeling for tidal hydrodynamic phenomena. In: Tidal Hydrodynamics, (ed.)
B.B. Parker, John Wiley & Sons, New York, pp 201-220.
Cherniawsky, J.Y., M.G.G. Foreman, W.R.Crawford, and R.F. Henry, 2001. Ocean tides from
TOPEX/POSEIDON sea level data. Journal of Atmospheric and Oceanic Technology,
18(4):649-664.
Clarke, A.J., 1991. The dynamics of barotropic tides over the continental shelf and slope
(Review). In: Tidal Hydrodynamics, (ed.) B.B. Parker, John Wiley & Sons, New York,
pp 79-108.
Cooley, J.W. and J.W. Tukey, 1965. An algorithm for machine calculation of complex Fourier
series. Mathematical Computations, 19:297-301.
CO-OPS, 2000. Tides and Currents Glossary. Center for Operational Oceanographic Products
and Services, National Ocean Service, NOAA, U.S. Dept. of Commerce.
CO-OPS, 2003. Computational Techniques For Tidal Datums Handbook, NOAA Special
Publication NOS CO-OPS2, Center for Operational Oceanographic Products and
Services, National Ocean Service, NOAA, U.S. Dept. of Commerce, 90 pages.
Crean, P. B., T. S. Murty, and J. A. Stronach, 1988: Mathematical modeling of tides and
estuarine circulation: The coastal seas of southern British Columbia and Washington
State. Lecture Notes on Coastal and Estuarine Studies, M. Bowman et al.,Eds., Springer-
Verlag, 471 pp.
Daily, James W. and Donald R.F. Harleman, 1966. Fluid Dynamics. Addison-Wesley
Publishing Company, Reading, MA, 454 pages.
Darwin, George H., 1883. The harmonic analysis of tidal observations. British Assoc. Report for
1883, pp 49-118.
Darwin, G. H., 1883. On the harmonic analysis of tidal observations of high and low water.
Proc. Royal Society, 48:278-340.
Dean, R.G., 1966. Tides and Harmonic Analysis. In: Estuary and Coastline Hydrodynamics,
(Ed.) Arthur T. Ippen, McGraw-Hill Book Company, New York, pp. 197-230..
348
References
Defant, Albert, 1961. Physical Oceanography, Vol.II, Pergamon Press, The Macmillan Co,
New York, pp 254-516.
Dennis, R.E. and E.E Long, 1971. A Users Guide to a Computer Program for Harmonic
Analysis of Data at Tidal Frequencies, NOAA Technical Reort NOS, National Ocean
Survey, NOAA, Dept. of Commerce, 31 pages.
Disney, L.P. and W.H. Overshiner, 1925. Tides and Currents in San Francisco Bay. Special
Pub. 115, U.S. Coast and Geodetic Survey, Dept. of Commerce, Washington, DC, 125 pages.
Dietrich, G., 1967. General Oceanography, Wiley, New York, 588 pages.
Doodson, A.T., 1921. The harmonic development of the tide-generating potential. Proc. Royal
Society A, 100:305-329.
Doodson, A.T., 1928a. The analysis of tidal observations. Philosophical Transactions of the
Royal Society of London, 227:223-279.
Doodson, A.T., 1928b. The analysis and prediction of tidal currents from observations of time of
slack water. Proceedings of the Royal Society of London. Series A, 121(787):72-88.
Doodson, A.T., 1951. The analysis of high and low waters. International Hydrographic Review,
28(1):13-77.
Doodson, A.T., 1957. The analysis and prediction of tides in shallow water. International
Hydrographic Review, 34:85-111.
Doodson, A.T. and H.D.Warburg, 1941. Admiralty Manual of Tides. Hydrographic Department,
Admiralty, London, 270 pages.
Dronkers, J.J., 1964. Tidal Computations In Rivers and Coastal Waters. North-Holland
Publishing Co., Amsterdam, 518 pages.
Eagleson, P.S. and R.G. Dean, 1966. Small amplitude wave theory. In: Estuaurine and
Coastline Hydrodynamics, (Ed) Arthur T. Ippen, McGraw-Hill Book Company, New
York, pp 1-92.
Egbert, G.D., A.F. Bennett, and M.G.G. Foreman, 1994. TOPX/POSEIDON tides estimated
using a global inverse model, Journal of Geophysical Research, 99(C12):24821-52.
Emery, William J. and Richard E. Thomson, 2001. Data Analysis Methods in Physical
Oceanography. Elsevier, New York, 638 pages.
Fang, G., 1987. Nonlinear effect of tidal friction. Acta Oceanologica Sinica, vol 6, suppl.1, pp
105-122.
Fjeldstad, J.E., 1929. Contributions to the dynamics of free progressive tidal waves. In: The
Norwegian North Polar Expedition with the Maud, 1919-1925, Scientific Results, vol. IV,
no. 3, Geofysisk Institute, Bergen.
Ferrel, William E., 1874. Tidal Researches. United States Coast Survey Report, 1874.
Appendix, Washington, DC., 270 pages.
Filloux, J.H., D.S. Luther,and A.D.Chave,1991. Update on seafloor pressure and electric field
observations from the north-central and noreastern Pacific: Tides, Infratidal fluctuations,
and barotropic flow. In: Tidal Hydrodynamics, (ed.) B.B. Parker, John Wiley & Sons,
New York, pp 617-640.
Fisher, Carl, 1986. Tidal Circulation In Chesapeake Bay. Ph.D. Dissertation, Old Dominion
University, 255 pages.
349
Foreman, M.G.G., 2004a. Manual for Tidal Heights Analysis and Prediction. Pacific Marine
Science Report 77-10, Inst. of Ocean Sciences, Patricia Bay, Victoria, B.C., 101 pages.
(Online revised version of 1977 original)
Foreman, M.G.G., 2004b. Manual for Tidal Currents Analysis and Prediction. Pacific Marine
Science Report 78-6, Inst. of Ocean Sciences, Patricia Bay, Victoria, B.C., 70 pages.
(Online revised version of 1978 original)
Foreman, M.G.G. and R.F. Henry, 1989. The harmonic analysis of tidal model time series.
Advances In Water Resources, 12(3):109-120.
Foreman, M.G.G. and R.F. Henry, 2004. Tidal Analysis Based On High and Low Water
Observations. Pacific Marine Science Report 79-15, Inst. of Ocean Sciences, Patricia
Bay, Victoria, B.C., 36 pages. (Online revised version of 1979 original)
Foreman, M.G.G., and E.T. Neufeld, 1991. Harmonic analysis of long time series. International
Hydrographic Review, 68(1):85-108.
Forrester. Warren D., 1983. Canadian Tide Manual, Dept. of Fisheries and Oceans, Ottawa, CA,
137 pages.
George, K.J. and B. Simon, 1984. The species concordance method of tide prediction in
estuaries. International Hydrographic Review, 65(1):121-145.
Gill, S.K., T.N. Mero, and B.B. Parker, 1992. NOAA operational experience with acoustic sea
level measurement. In: Sea Level Measurement and Quality Control. Joint IAPSO/IOC
Workshop, October 12-13, 1992, Intergovernmental Commission, Paris, France.
Gill, S.K. and J.R. Schultz (eds), 2001. Tidal Datums and Their Applications. NOAA Special
Publication NOS CO-OPS1, Center for Operational Oceanographic Products and
Services, National Ocean Service, NOAA, U.S. Dept. of Commerce, 111 pages.
Glenn, S.M., T.D. Dickey, B. Parker, and W. Boicourt, 2000. Long-term real-time coastal ocean
observation networks, Oceanography, 13(1):24-34.
Glenn, S.M., W. Boicourt, B. Parker, and T.D. Dickey, 2000. Operational observation networks
for ports, a large estuary and an open shelf, Oceanography, 13(1):12-23.
Godin, G., 1972. The Analysis of Tides. University of Toronto Press, Toronto, Canada, 264
pages.
Godin, G., 1986. The use of nodal corrections in the calculation of harmonic constants.
International Hydrographic Review, 63(2):143-162.
Godin, Gabriel, 1988. Tides. CICESE, Ensenda, Mexico, 290 pages.
Godin, G., 1991. The analysis of tides and currents. In: Tidal Hydrodynamics, (ed.) B.B.
Parker, John Wiley & Sons, New York, pp 675-709.
Gonella, J., 1972. A rotary-component method for analysing meteorological and oceanographic
vector time series, Deep Sea Research., 19:833 846.
Green, G., 1837. On the motion of waves in a variable canal of small depth and width.
Transactions, Cambridge Philosophical Society, 6:457-462.
Gross, T.F., K.T. Bosley, K.W. Hess, 2000. The Chesapeake Bay Operational Forecast System
(CBOFS): Technical Documentation, National Ocean Service, NOAA, U.S. Dept. of
Commerce.
Gross, T.F., 2002. (Personal communication)
350
References
Groves, G., 1955. Numerical filters for discrimination against tidal periodicities. Trans.
American Geophysical Union, 36(6):1073-84.
Groves, G.W. and E.J. Hannan, 1968. Time series regression of sea level on weather, Reviews of
Geophysics, 6:129-174.
Haidvogel, D. B., and A. Beckman, 1999. Numerical Ocean Circulation Modeling, Imperial
College Press, River Edge, N. J.,318 pp.,
Harris, D.L., 1962. The equivalence between certain statistical prediction methods and linearizd
dynamical methods, Monthly Weather Review, vol. 90, No. 8, Aug. 1962.,
Harris, D.L., N.A. Pore, and R. Cummings, 1963. The Application of High Speed Computers to
Practical Tidal Problems. Abstracts of Papers, vol 6, IAPO, XIII General Assembly,
IUGG, Berkeley, VI-16.
Harris, Rollin A., 1887-1907. Manual of Tides. Parts 1, 2, 3, 4, and 5, originally Appendices to
the U.S. Coast and Geodetic Annual Reports.
Hendershott, M.C., Speranza, A., 1971. Co-oscillating tides in long narrow bays; the Taylor
problem revisited. Deep Sea Research, 8:959-980.
Hess KW. 1989. MECCA Program Documentation. NOAA Technical Report NESDIS 46,
NOAA, U.S. Department of Commerce, 156 pp.
Hess, K., R. Schmalz, C. Zervas, W. Collier, 1999. Tidal Constituent And Residual
Interpolation (TCARI): A New Method for the Tidal Correction of Bathymetric Data,
NOAA Technical Report NOS CS 4, National Oceanic and Atmospheric Administration.
Hess, Kurt, 2003. Water level simulation in bays by spatial interpolation of tidal constituents,
residual water levels, and datums. Continental Shelf Research, 23:395-414.
Hicks, Steacy D. 1967. The prediction centenary of the United States Coast and Geodetic
Survey, International Hydrographic Review, 44:123-131.
Hicks, Steacy D. 2006. Understanding Tides, Center for Operational Oceanographic Products
and Services, National Ocean Service, NOAA, U.S. Dept. of Commerce, 66 pages.
Huggett, W.S., 1988. Juan de Fuca Strait and Strait of Georgia Current Atlas. The
Hydrographical Journal, 47:5-16.
IOC, 2006. Manual On Sea Level Measurement and Interpretation, Vol. IV, an update to 2006,
Intergovernmental Oceanographic Commission, JCOMM Technical Report No. 31,
WMO/TD No. 1339, UNESCO, 88 pages.
Ippen, A.T. and D.R.F Harleman, 1966. Tidal dynamics in estuaries. In: Estuary and Coastline
Hydrodynamics, (Ed.) Arthur T. Ippen, McGraw-Hill Book Company, New York, pp.
493-545.
Jay, David A., 1991. Greens law revistied: tidal long wave propagation in channels with strong
topography. Journal of Geophysical Research, 96:20,585-20,598.
Jay, David A. and Edward P. Flinchem, 1999. A comparison of methods of analysis of tidal
records containing mult-scale non-tidal background energy. Continental Shelf Research,
19:1695-1731.
Jay, David A. and Tobias Kukulka, 2003. Revising the paradigm of tidal analysis the uses of
non-stationary data. Ocean Dynamics, 53:110-125.
351
Kantha, Lakshmi H. and Carol Anne Clayson, 2000. Numerical Models of Oceans and Oceanic
Processes. International Geophysics Series, Vol 66, Academic Press, 940 pages.
Kelley, John G.W., D.W.Behringer, J.H. Thiebaux, and B. Balasubramaniyan 2002.
Assimilation of SST data into a real-time coastal ocean forecast system for the U.S. East
Coast. Weather and Forecasting, 17(4):670-690.
Knauss, John A., 1978. Introduction To Physical Oceanography. Prentice-Hall, Englewood
Cliffs, NJ, 338 pages.
Kuo, John T., 1991. Synoptic predictions of tide and currents everywhere in the world. In: Tidal
Hydrodynamics, (ed.) B.B. Parker, John Wiley & Sons, New York, pp 61-76.
Laplace, P.S., 1775. Mem. Acad. R. Sci., 88:75-182, 89:177-267.
LeLacheur, E.A. and J.C. Sammons, 1932. Tide and Currents in Long Island and Block Island
Sounds, Special Publications 174, U.S. Coast and Geodetic Survey, Dept. of Commerce,
Washington, DC, 187 pages.
Le Provost, Christian, 1976. Theoretical analysis of the structure of the tidal wave spectrum in
shallow water areas, Mmoires Socit Royale des Sciences de Lige, 6e srie, tome C, pp
97-111.
Le Provost, Christian, 1992. On the use of sea-level gauge data for satellite altimetry validation,
a review. Oceanologica Acta, 15:431-440.
LeProvost, Christian, 2001. Tide over ridges, shelves and near coasts. In: Report of the High-
Resolution Ocean Topography Science Working Group Meeting, (Ed) D.B. Chelton,
convened at U. of Maryland, College Park, MD.
Le Provost, Christian and Patrick Vincent, 1991. Finite-element method for modeling ocean
tides. In: Tidal Hydrodynamics, (ed.) B.B. Parker, John Wiley & Sons, New York, pp
41-60.
Loder, J.W. and C. Garrett, 1978. The 18.6-year cycle of sea surface temperature in shallow seas
due to variations in tidal mixing. Journal of Geophysical Research, 83(C4):1967-1970.
Lynch, Daniel R. and Francisco E. Werner, 1991. Three-dimensional velocities from a finite-
element model of English Channel/Southern Bight tides. In: Tidal Hydrodynamics, (ed.)
B.B. Parker, John Wiley & Sons, New York, pp 183-200.
Marmer, Harry A., 1926. The Tide. D. Appleton and Company, New York, 282 pages.
Marmer, Harry A., 1951. Tidal Datum Planes, U.S. Coast and Geodetic Survey, Dept. of
Commerce, Washington, DC, 142 pages.
Mellor, G.L., and T. Yamada, 1982. Development of a turbulence closure model for geophysical
fluid problems, Review Geophys. Space Physics, 20:851-875.
Mellor, George, Sirpa Hakkinen, Tal Ezer, and Richard Patchen, 2002. A generalization of a
sigma coordinate ocean model and an intercomparison of model vertical grids. In: Ocean
Forecasting: Conceptual Basics and Applications, N. Pinardi and J. Woods (Eds.),
Springer, Berlin, pp 55-72.
Myers, E.P. and A.M. Baptista, 2001. Inversion for tides in the Eastern North Pacific Ocean.
Advances in Water Resources, 24(5):505-519.
352
References
Mooers, C.N.K., 1973. A technique for the cross spectrum analysis of pairs of complex-valued
time series, with emphasis on properties of polarized components and rotational
invariants. Deep-Sea Research, 20:1129-1141.
Munk, W.H. , B. Zetler, and G.W.Groves, 1965. Tidal cusps. Geophysical Journal, 10:211-219.
Munk, W.H. and D.E. Cartwright, (1966). Tidal spectroscopy and prediction. Phil. Trans. R.
Soc. A, 259:533-581.
Munk, Walter and Klaus Hasselmann, 1964. Super-resolution of tides. Studies on
Oceanography, University of Tokyo Press, Tokyo, Japan, pp 339-344.
National Ocean Service, 1987. Delaware River and Bay Tidal Circulation and Water Level
Forecast Atlas. National Ocean Service, NOAA, U.S. Dept. of Commerce, 49 pages.
Noble, M.A. and G.R. Gelfenbaum, 1992. Seasonal fluctuations in sea level on the South
Carolina Shelf and their relationship to the Gulf Stream. Journal of Geophysical
Research, 97(C6):9521-9529.
OBrien, J.J. and R.D. Pillsbury, 1974. Rotary wind spectra in a sea breeze regime. Jour. Of
Applied Meteorology, 13(7):820-825.
Parker, Bruce B. 1977. Tidal Hydrodynamics In the Strait of Juan de Fuca-Strait of Georgia,
NOAA Technical Report, NOS 69, 56 pages.
Parker, Bruce B., 1984. Frictional Effect on the Tidal Dynamics of a Shallow Estuary. Ph.D.
Dissertation, The Johns Hopkins University, Baltimore, Maryland, 292 pages.
Parker, Bruce B., 1986. Applications of real-time oceanographic data systems, The Marine
Technology Society Journal, 20(3):18-27.
Parker, Bruce B., 1988. An atlas for forecasting circulation and water levels. Sea Technology,
29(8): 21-25.
Parker, Bruce B., 1991a. The relative importance of the various nonlinear mechanisms in a wide
rage of tidal interactions. In: Tidal Hydrodynamics, pages 237-268, John Wiley & Sons,
New York..
Parker, Bruce B., 1991b. On the use of nonharmonic tidal analyses and their effect on the
quality of subordinate station predictions. In: Tidal Hydrodynamics, pages 771-788, John
Wiley & Sons, New York.
Parker, Bruce B., 1994. Sea level variability along the East Coast of the United States. In:
Studies In Eastern Energy and the Environment, AAPG Eastern Section Special Volume,
(Eds.) A.P. Schultz and E.K. Rader, Virginia Div. of Mineral Resources Publication 132,
pp 66-72.
Parker, Bruce B., 1995. PORTS makes navigation safer. Proceedings of the Marine Safety
Council, 52(5):30-2, Sept-Oct, 1995.
Parker, Bruce B., 1996. Monitoring and modeling of coastal waters in support of environmental
preservation. Journal of Marine Science and Technology, 1(2):75-84.
Parker, Bruce B., 1997. Coastal Currents, Mariners Weather Log, 40(3):24-28.
Parker, Bruce B., 1998a. Development of Model-Based Regional Nowcasting/ Forecasting
Systems. In: Spaulding, M.L. and A.F. Blumberg (Eds.), Estuarine and Coastal
Modeling, American Society of Civil Engineers, New York, refereed proceedings of the
5th International Conference, Alexandria, VA, pp 355-373.
353
Parker, Bruce B., 1998b. The Coriolis Effect: Motion On a Rotating Planet, Mariners Weather
Log, 42(2):17-33.
Parker, Bruce B., 1998c. Why Are the Tides So Predictable? Mariners Weather Log, 42(3):21-
27.
Parker, Bruce B., 1999. Tides In Shallow Water Mariners Weather Log, 43(3):16-23.
Parker, Bruce, B., 2002. The integration of bathymetry, topography, and shoreline and the
vertical datum transformations behind it. International Hydrographic Review, vol.3,
no.3(New Series), pp 35-47, Nov 2002.
Parker, Bruce B., 2002. Integrated real-time/forecast information for safe, efficient, and
environmentally-friendly ports, Singapore Maritime and Port Journal, pp 1-10.
Parker, Bruce B., 2003. The difficulties in measuring a consistently defined shoreline -- the
problem of vertical referencing. Journal of Coastal Research, Special Issue 38:44-56.
Parker, Bruce B., 2004. Tides. In: Encyclopedia of Coastal Science, (ed.) M. Schwartz, Kluwer
Academic Publishers, Encyclopedia of Earth Sciences Series, pp 987-996..
Parker, Bruce B. and Bryan R. Pearce. 1975. The Response of Massachusetts Bay To Wind
Stress, Massachusetts Institute of Technology Technical Report No. 198, Dept. of Civil
Engineering, 109 pages.
Parker, Bruce B. and Lloyd C. Huff, 1998. Modern under-keel clearance management.
International Hydrographic Review, LXXV(2):143-165.
Parker, Bruce B., Alan Davies, and Jiuxing Xing, 1999. Tidal Height and Current Prediction.
Chapter 12 in: Coastal Ocean Prediction, Christopher N.K. Mooers (Ed.), Coastal and
Estuarine Studies, 56, pages 277-327, AGU Publications, Washington, D.C..
Parker, B., K. Hess, D. Milbert, and S. Gill, 2003. A National Vertical Datum Transformation
Tool. Sea Technology, 44(9):1-15.
Patchen, R. C., 1986. A real-time operational model/measurement system for Delaware River
and Bay. In Applications of Real-Time Oceanographic Circulation Modeling, (ed) Bruce
B. Parker, pp. 47-58.
Patchen, Richard C., Elmo E. Long, and Bruce B. Parker. 1976. Analysis of Current Meter
Observations In New York Bight Apex, August 1973-June 1974, NOAA Technical Report
ERL 368-MESA 5, 24 pages.
Patchen, R. and J. Blaha, 2002. Implementation of an infrastructure to support operation and
evaluation of Gulf of Mexico models. In: Oceans02 MTS/IEEE, vo. 2, pp 803-809.
Pawlowicz, Rich, Bob Beardsley, and Steven Lenz, 2002. Classical tidal harmonic analysis
including error estimates in MATLAB using T_TIDE, Computers & Geosciences,
28:929-937.
Pedlosky, Joseph, 1987. Geophysical Fluid Dynamics. Spring-Verlag, New York, 624 pages.
Platzman, George, 1971. Ocean tides and related waves. Lectures in Applied Math, vol. 14, pp
239-291.
Platzman, George, 1991. Tidal evidence for ocean normal modes. In: Tidal Hydrodynamics,
354
References
Porter, David L. and H.H. Shih, 1996. Investigations of temperature effects on NOAAs Next
Generation Water Level Measurement system. Jour. of Atmospheric and Oceanic
Technology, 13:714-725.
Prandle, David, 1982. The vertical structure of tidal currents and other oscillatory flows.
Geophysical and Astrophysical Fluid Dynamics, 22:29-49.
Prandle, David, 1991. Tides in estuaries and embayments (Review). In: Tidal Hydrodynamics,
Prandle, D. and J. Wolf, 1978. The interaction of surge and tide in the North Sea and River
Thames. Geophysical Journal of the Royal Astronomical Society, 55:203-216.
Prandle, D. and M. Rahman, 1980. Tidal response in estuaries. Journal of Physical
Oceanography, 10(10):1552-73.
Press, W.H., B.P. Flannery, S.A. Teukolsky, and W.T. Vetterling, 1992. Numerical Recipes:
The Art of Scientific Computing. (2nd. Edition) Cambridge University Press, 992 pages.
Preisendorfer, R.W., 1988. Principal Component Analysis in Meteorology and Oceanography,
Developments in Atmospheric Sciences, 17, Elsevier, Amsterdam.
Proudman, J., 1923. Tides: Appendix. Reports on the State of Science, etc., Section A -
Liverpool, pp 1-6, British Association.
Proudman, J., 1942. Halleys tidal chart. The Geophysical Journal, 100(4):174-6.
Pugh, D.T., 1987. Tides, Surges, and Mean Sea Level. John Wiley & Sons, New York, 472
pages.
Pugh, D.T., 2004. Changing Sea Levels. Effects of Tides, Weather, and Climate, Cambridge
University Press, Cambridge, UK, 265 pages.
Ray, Richard D. and Gary T. Mitchum, 1997. Surface manifestation of internal tides in the deep
ocean: observations from altimetry and island gauges. Progress In Oceanography,
40:135-162.
Redfield, A., 1950. The analysis of tidal phenomena in narrow embayments, Papers In Physical
Oceanography and Meteorology, 11(4), Massachusetts Inst. of Tech. and Woods Hole.
Rossiter, J.R. and G.W. Lennon, 1967. An intensive analysis of shallow water tides.
Geophysical J. Royal astronomical Soc., 16275-293.
Schureman, P., 1958. Manual of Harmonic Analysis and Prediction of Tides. Special
Publication 98, Coast and Geodetic Survey, U.S. Dept. of Commerce, Washington, D.C.,
317 pages.(Revised 1940 Edition with corrections; first published in 1924)
Schwiderskii, E.W., 1978. Global Ocean Tides, Part I: A Detailed Hydrodynamic Interpolation
Model, NSWC/DX TR-3866, U.S. Naval Surface Warfare Center.
Schwiderskii, E.W., 1979. Global Ocean Tides, Part II: The Semidiurnal Principle Lunar Tide
(M2), Atlast of Tidal Charts and Maps, NSWC TR-79-414, U.S. Naval Surface Warfare
Center.
Schwiderski, E.W., 1980. On charting global ocean tides, Rev. of Geophysics and Space
Physics, 18(1):243-268.
Schwiderski, Ernst W., 1991. High-precision modeling of mean sea level, ocean tides, and
dynamic ocean variations with GEOSAT altimeter signals. In: Tidal Hydrodynamics,
(ed.) B.B. Parker, John Wiley & Sons, New York, pages 593-616.
355
Schmalz, Richard, 1996. DGPS-Supported Hydrosurvey, Water Level Measurement, and

Modeling of Galveston Bay: Development and Application of the Numerical Circulation
Model. NOAA Technical Report NOS OES 012, Silver Spring, Maryland, 136 pages.
Schmalz, Richard, 2000. Three-dimensional Hydrodynamic Model Development For a
Galveston Bay Nowcast/Forecast System, NOAA Technical Report NOS CS9, Silver
Spring, MD, 167 pages.
Shih, H.H. and L. Baer, 1991. Some errors in tide measurement caused by dynamic
environment. In: Tidal Hydrodynamics, (ed.) B.B. Parker, John Wiley & Sons, New
York, pp 641-671.
Shum, C.K., P.L. Woodworth, O.B. Anersen, G.D. Egbert, O. Francis, C. King, S.M. Klosko, C.
LeProvost, X. Li, J-M Molines, M.E. Parke, R.D. Ray, M.G. Schlax, D. Stammer, C.C.
Tierney, P. Vincent, and C.I. Wunsch, 1997. Accuracy assessment of recent ocean tide
models. Journal of Geophysical Research, 102(C11)::25173-25194.
Simon, B., 1991. The species concordance method of tide prediction. In: Tidal Hydrodynamics,
(ed.) B.B. Parker, John Wiley & Sons, pp 725-735.
Smith, A.J., B.A.C. Ambrosius, K.F. Wakker, P.L. Woodworth, and J.M. Vassie, 1997.
Comparison between the harmonic and response methods of tidal analysis using
TOPEX/POSEIDON altimetry. Journal of Geodesy, 71:695-703.
Speer, Paul, David G. Aubrey, and Carl T. Friedrichs, 1991. Nonlinear hydrodynamics of
shallow tidal inlet/bay systems. In: Tidal Hydrodynamics, (ed.) B.B. Parker, John Wiley
& Sons, New York, pp 321-340.
Szabados, Michael, Chuck Roman, and Bobby Taylor, 1987. Transmission of real time
oceanographic and meteorological data from ships. Oceans, vol. 19, pp 863-869.
Taylor, G. I., 1920. Tidal oscillations in gulfs and rectangular basins. Proceedings London
Mathematical Society, 20:148181.
Thomson, Sir William (Lord Kelvin), 1869. Report of the Committee for the purpose of
promoting the extension, improvement and harmonic analysis of Tidal Observations. In:
Report of the 38th Meeting of the BAAS, 1868, pp 489-510, John Murray, London.
Thomson, Sir William (Lord Kelvin), 1881. The tide gauge, tidal harmonic analyzer and tide
predicter. Proceedings Institute of Civil Engineering, London, 65:4-74.
Tsimplis, M.N. and P.L. Woodworth, 1994. The global distribution of the seasonal sea level
cycle calculation from coeastal tide gauge data. Journal of Geophysical Research,
99(C8):16031-16039.
U.S. Coast and Geodetic Survey, 1950. Manual of Current Observations, Special Publication
No. 215, Washington, D.C., 87 pages.
U.S. Coast and Geodetic Survey,1952. Manual of Harmonic Constant Reductions, Special
Publication 260, U.S. , U.S. Dept. of Commerce, 74 pages.
U.S. Coast and Geodetic Survey, 1965. Manual of Tide Observations, Publication 30-1,
Washington, D.C., 72 pages.
Wei, Eugene, 2003. The new Port of New York and New Jersey Operational Forecast System
(NYOFS), Bulletin of the American Meteorological Society, Sept 2003, pp 1004-6.
356
References
Woodworth, P.L. and J.P. Thomas, 1990. Determination of the major semidiurnal tides of the
northwester European continental shelf from Geosat altimetry. Journal of Geophysical
Research, 95(C3):3061-3068.
Wunsch, C., 1972. Bermuda sea level in relation to tides, weather, and baroclinic fluctuations.
Reviews in Geophysics, Space Physics, 10:1-49.
Yih, Chia-Shun, 1969. Fluid Mechanics. McGraw-Hill Book Company, New York, 622 pages.
Zervas, Chris, 1999. Tidal Current Analysis Procedures and Associated Computer Programs.
NOAA Technical Report NOS CO-OPS 0021. National Ocean Service, NOAA, U.S.
Dept. of Commerce, 101 pages.
Zetler, Bernard D, 1953. A method of inferring the amplitudes of the tidal constituents M2 and
(K1+O1), International Hydrographic Review, vol. 30.
Zetler, Bernard D, 1959. Tidal characteristics from harmonic constants. J. of the Hydraulics
Div, Proc American Soc. Civil Engineers, vol 85, HY12, pp 77-87.
Zetler, Bernard D, 1982. Computer Applications to Tides in the National Ocean Survey.
National Ocean Survey, NOAA, Dept. of Commerce, 85 pages.
Zetler, Bernard.D, 1987. The evolution of modern tide analysis and prediction some personal
memories. International Hydrographic Review, 64(1):124-139.
Zetler, Bernard.D, 1991. Military Tide Predictions in World War II.. In: Tidal Hydrodynamics,
Zetler, B.D., M.D. Schuldt, R.W. Whipple, and S.D. Hicks, 1965. Harmonic analysis of tides
from data randomly spaced in time. Journal of Geophysical Research, 70(12):2805-2811.
Zetler, Bernard D. and Robert E. Cummings, 1967. A harmonic method for prediction shallow-
water tides. Journal. of Marine Research, 25(1):103-114.
Zetler, B.D., D.E. Cartwright, and W.H. Munk, 1969. Tidal constants derived from response
admittances. Symp. on Earth Tides, Strasbourg, 175-178.
Zetler, B.D., D.E. Cartwright, and S. Berkman, 1979. Some comparisons of response and
harmonic tide predictions. International Hydrographic Review, 56(2):105-115.
Zetler, B.D. and W.H. Munk, 1975. The optimum wiggliness of tidal admittances. J. of Marine
Research, 33(Suppl.):1-13.
Zetler, B.D., E.E. Long, and L.F. Ku, 1985. Tide predictions using satellite constituents.
International Hydrographic Review, 62(2):135-142.
Zhang, Aijun, Bruce B. Parker, and Eugene Wei, 2002. Assimilation of water level data into a
coastal hydrodynamic model by an adjoint optimal technique. Continental Shelf
Research, 22:1909-1934.
Zhang, Aijun, Eugene Wei, and Bruce B. Parker, 2003. Optimal estimation of tidal open
boundary conditions using predicted tides and adjoint data assimilation technique.
Continental Shelf Research, 23:1055-1070.
Zhang, Aijun, K.W. Hess, E.Wei, and E.Myers, 2006. Implementation of Model Skill
Assessment Software for Water Level and Current In Tidal Regions, NOAA Technical
Report NOS CS 24., 59 pages.
Zuosheng, Yang, K.O. Emery, and Xui Yui, 1989. Historical development and use of thousand-
year-old tide-prediction tables. Limnology and Oceanography, 34(5):953-957.
357
358
List of Figures
Figure 1.1. Harris tide predicting machine 4

Figure 1.2. Gear and pulley system of an early analog tide predicting machine. 9
Figure 2.1. Simple tide curve with definitions. 16

Figure 2.2. Four tide curves showing how the tide range (high water minus low water)
varies throughout the month. 17
Figure 2.3. Simple tidal current curve with definitions for a reversing tidal current. 18
Figure 2.4. Examples of three rotary tidal currents and one reversing tidal current. 18
Figure 2.5. Corange chart showing the geographic variation of the tidal range of the
largest tidal constituent, M2 , for the Strait of Juan de Fuca-Strait of Georgia, 20
Figure 2.6. Cotidal chart showing the geographic variation in the time of high water for
the largest tidal constituent, M2 , for the Strait of Juan de Fuca-Strait of
Georgia. 21
Figure 2.7. The relationship between tidal datums and U.S. marine boundaries. 23
Figure 2.8. The Earth-Moon system (viewed from very high above the North Pole)
revolving around a common point (inside the Earth) with two tidal bulges. 25
Figure 2.9. The tide generating forces on the Earth resulting from the difference between
gravitational attraction and centrifugal force. 25
Figure 2.10. A very long tide wave of relatively small amplitude produced in the ocean by
the tide-generating forces caused by the moon and the sun. 26
Figure 2.11. Asymmetric tidal bulges when the moon is at maximum declination north or
south of the equator, resulting in a diurnal component in the tidal forces 28
Figure 2.12. The orbits of the moon around the Earth and of the Earth around the sun, as
viewed from a distant star. 30
Figure 2.13. The moons orbit moon and the suns orbit (the ecliptic) with respect to the
Earths equatorial plane. 30
Figure 2.14. The combined effect of the moon and the sun at Full Moon and New Moon
(causing larger tidal ranges, spring tides), and at First Quarter and Last
Quarter (causing smaller tidal ranges, neap tides). 33
Figure 2.15. A demonstration of aliasing. 36
Figure 2.16. Water level spectrum from near Anchorage, Alaska, at the end of a shallow
waterway with a large tide range. 36
Figure 2.17. Tidal Spectra, with diurnal species and semidiurnal species further split into
tidal spectral lines. 38
359
Figure 2.18. A 100-year record of monthly and annual mean tide ranges at Seattle, WA,
illustrating the 18.6-year lunar nodal cycle. 42
Figure 2.19. Tide curves for particular combinations of (K1+O1)/M2 amplitude ratios and
phase differences defined as (M2o-K1o-O1o). 44
Figure 2.20. Chart illustrating the geographic variation in the (K1+O1)/(M2+S2) tidal
constituent amplitude ratio, as well as the type of tide classification for each
location, in the Strait of Juan de Fuca - Strait of Georgia. 45
Figure 2.21. The tide propagating up a river as a progressive wave. 50
Figure 2.22. The tide in a bay as a standing wave 51
Figure 2.23. The effect of Coriolis force on M2 tidal range (corange lines) and the times
of high water (cotidal lines) for an idealized rectangular bay. 53
Figure 2.24. M2 coamplitude chart for Chesapeake Bay. 54
Figure 2.25. Typical tide curves for an area with no shallow-water effects (top panel) and
for three areas with different degrees of distortion caused by the shallow water. 56
Figure 2.26. Water level curves from the Gironde River in France showing the effects of
nonlinear shallow-water distortion. 57
Figure 2.27 Tide curves for particular combinations of M4/M2 amplitude ratios and phase
differences between M4 and M2 (defined as 2M2o-M4o). 60
Figure 2.28. Tide curves for particular combinations of M6/M2 amplitude ratios and phase
differences between M6 and M2 (defined as 3M2o-M6o). 61
Figure 2.29. Water level data from the tide gauge at Trenton, NJ, during a high river
discharge period. 63
Figure 2.30. Comparison of 15-day harmonic analyses of water level data from Delaware
River and Bay during a high runoff period and during a low runoff period,
both periods with similar astronomical conditions. 64
Figure 2.31. The tidal portion of a water level record from Trenton, NJ, on the Delaware
River, during a high river discharge period, plotted with the tide predictions
for the same period as if there were no river discharge. 65
Figure 2.32. The tidal portion of water level recorded plotted with tidal predictions for
three locations along the Delaware River and Bay and a station on the
Atlantic Coast showing the effect on the tide of subtidal storm surge. 66
Figure 2.33. Calculated node factors for six tidal constituents from 19 one-year
(uncorrected) least squares harmonic analyses of Philadelphia water level
data plotted with node factors determined from astronomical/equilibrium-tide
considerations. 68
Figure 2.34. Current flow through a cross section near the Chesapeake Bay Bridge. 71
Figure 2.35. The spatial variation in tidal currents near the surface to the west of Bergen
Point in the Port of the New York and New Jersey, acquired from a towed
ADCP. 72
Figure 2.36. Variation in (K1+O2)/M2 and M4/M2 tidal current amplitude ratios, and
tidally-induced residual current, across the entrance to Haro Strait, due to the
effect of the lateral inertial terms in the equations of motion 74
Figure 2.37. Speed and direction plots of the unusual occurrence of dominant
quarter-diurnal tidal currents at an NOS current station in Ramshorn
Creek, SC. 75
360
List of Figures
Figure 3.1. A very simple demonstration of extracting tidal constituents from a water
level time series. 83
Figure 3.2. A graphical representation of the amplitude and epoch of a single tidal
constituent and its time relationship to moons transit. 91
Figure 3.3. Example of a tide curve produced with the harmonic tide prediction equation,
along with seven of the individual tidal constituent curves. 93
Figure 3.4. A current vector with speed and direction, broken up into north and east
components and into major and minor components. 96
Figure 3.5. The harmonic characteristics for two water level stations and two current
stations in San Francisco Bay, where the tide regime is mixed. 114
Figure 3.6a. Plot of the differences between predicted low water times at Port Chicago
and predicted low water times at Golden Gate, for the first 120 tidal cycles
of 1980. 115
Figure 3.6b. Plot of the differences between the heights of predicted low waters at Port
Chicago and the heights of predicted low waters at Golden Gate. 115
Figure 3.7. Plot of time differences between harmonically predicted maximum ebbs at
two current stations in Delaware River and Bay. 116
Figure 3.8. Predicted tidal current at Philadelphia on the Delaware River illustrating the
dramatically varying time interval between successive maximum floods and
between successive maximum ebbs due to shallow-water effects. 118
Figure 3.9. Plot of time differences between harmonically predicted (but leaving out
the overtide and compound tidal constituents) maximum ebbs at two
current stations in Delaware River and Bay. 118
Figure 3.10. The mean annual sea level cycle at 14 NOS stations along the East Coast
and Gulf Coast of the U.S. for the period 1948-1984. 121
Figure 3.11. The variation from year to year of the seasonal sea level range from 1935
to 1992 at four NOS stations along the U.S. East Coast. 122
Figure 3.12a. Filtering a data time series. In this simple example each group of 13 data
points is averaged to produce one data point in the filtered data series.
. 124
Figure 3.12b. A water level time series filtered to remove the tidal oscillations and leave
a predominantly nontidal time series. 125
Figure 3.13. Water level data from Trenton, NJ, filtered with a tidal filter compared with
river discharge data at Trenton filtered with the same filter. 126
Figure 3.14. The difference between two tidally filtered water level records at opposite
ends of Massachusetts Bay compared with the wind stress component along
the axis of the Bay, which was similarly filtered. 127
Figure 3.15. The simple box filter (running mean) and its effect in the frequency domain. 128
Figure 3.16. An idealized low-frequency nontidal continuum decreasing from its highest
value near zero freqency, but then rising up around a tidal spectral line in
the shape of a so-called tidal cusp. 136
Figure 4.1. The residual water level at Trenton, NJ, on the Delaware River, showing
the temporary appearance of tidal energy in the residual during a period of
increasing river discharge. 147
361
Figure 4.2. Schuremans formulas for inferring tidal constituents from constituents
already calculated. 153
Figure 5.1. A current station deployed at a bend in the waterway, so that the flood
direction and the ebb direction are not 180o apart; because of this geometry
a harmonic analysis will show a mean current toward the east and a
significant second harmonic, M4. 172
Figure 5.2. A fairly narrow tidal constituent ellipse (i.e., eccentricity approaching 1.0). 175
Figure 5.3. The geometry and trigonometry for converting between north-east
components and major-minor components. 176
Figure 5.4. Tidal current constituent ellipses for M2, S2, N2, K1, and O1, for a current
station in the Strait of Georgia, based on a 29-day harmonic analysis of a
data time series. 178
Figure 6.1. A copy of Form 2211, on which tabulations were done in the pre-computer
era when carrying out a monthly mean analysis. 198
Figure 6.2. Form 657, which was used in the pre-computer era for calculating and
recording comparison of monthly means. 200
Figure 6.3. Form 248, which was used in the pre-computer era to carry out a
comparison of simultaneous observations (a tide-by-tide analysis). 201
Figure 6.4. A short tide-by-tide analysis produced with a recent computer program. 203
Figure 6.5. An early version of Form 451, which was used in the pre-computer era to
carry out a reversing reduction analysis of current data. 206
Figure 6.6. Results of a computerized reversing reduction analysis for Kent Island,
MD, relative to a reference station at Chesapeake Bay Entrance. 208
Figure 6.7. Plotted time differences between slack before flood at Kent Island, MD,
and a reference station at Chesapeake Bay Entrance. 208
Figure 6.8. Plotted results from a reversing reduction analysis showing the time
differences between maximum ebbs at Carquinez Strait and Alcatraz-North
Point, both in San Francisco Bay. 209
Figure 6.9. Plotted results of a reversing reduction showing the ratio of maximum ebb
speeds at Carquinez Strait and Alcatraz-North Point, both in San Francisco
Bay. 209
Figure 6.10a. Speed and direction plots of the total current (tidal current plus mean
nontidal current) resulting from a rotary reduction analysis of current data
from a station in Boston Harbor. 214
Figure 6.10b. Polar plot showing the rotation of the total current (tidal current plus mean
nontidal current) resulting from a rotary reduction analysis of current data
from a station in Boston Harbor. 214
Figure 7.1. Coamplitude and cophase chart for the M2 tide constituent for the Delaware
River and Bay waterway, showing the geographic variation in M2 amplitude
and epoch. 225
362
List of Figures
Figure 7.2. Chart with epochs (phase lags) for maximum flood for the M2 tidal currents
in the Strait of Juan de Fuca Strait of Georgia waterway, given in degrees
relative to the local time meridian (120oW). 227
Figure 7.3. Numerous M2 ellipses for the waterways around the San Juan Islands. 228
Figure 7.4. Variation in the amplitude (half-tide range) of the five largest semidiurnal
constituents in the Delaware River and Bay. 230
Figure 7.5. The amplification of the five largest semidiurnal constituents in Delaware
River and Bay, namely, M2, N2, S2, 2MN2, and 2MS2. 231
Figure 7.6. A cubic control volume used in deriving the conservation of mass equation. 233
Figure 7.7. Schematic of waterway with Cartesian coordinate system. 239
Figure 7.8. The phase (time) of high water for various values of the friction coefficient
plotted along the length of the waterway from the closed end versus the
change in wave number phase 241
Figure 7.9. The tide range ratio for various values of the friction coefficient plotted
along the length of the waterway from the closed end versus the change in
wave number phase. 242
Figure 7.10. Plots of the tide range ratio (relative to the value at the entrance) for six
different length waterways (and the same frictional damping coefficient). 243
Figure 7.11. Plots of the phase (time) difference between high water and maximum
flood for various frictional damping coefficients. 243
Figure 7.12. Maximum flood speed for various frictional damping values plotted
relative to the closed end of the waterway. 244
Figure 7.13. Plots of the K1/M2 tide range ratio (relative to the ratio at the closed end)
for various values of the frictional damping coefficient. 245
Figure 7.14. K1/M2 ratios for six different waterways lengths, for frictional damping
coefficient =1.5 (typical for many waterways), plotted relative to the
K1/M2 value at the entrance. 246
Figure 7.15a. The phase (time) of M2 high water (relative the value at the closed end) for
eight different values of the frictional coefficient N in an estuary with an
exponential width decrease matching that in the Delaware River and Bay. 249
Figure 7.15b. The M2 tide range (relative the value at the closed end) for eight different
values of the frictional coefficient N in an estuary with an exponential
width decrease matching that in the Delaware River and Bay. 249
Figure 7.16. Tide range (relative to that at the closed end) for five different values of
exponential width decrease, for the frictionless case (N=0). 250
Figure 7.17. Tide range for stations in the Bay of Fundy plotted verus the time (phase)
of high water, and a map of the locations of the tide stations. 252
Figure 7.18. Cotidal and corange lines in a semi-enclosed waterway for four different
values of the frictional damping coefficient (including the frictionless case). 255
Figure 7.19. The effect of a mean river flow on the first three tidal harmonics, from a
Fourier decomposition of the quadratic friction term, u*u*. 262
Figure 7.20. Diagram illustrating the asymmetric effects of the nonlinear continuity term,
the inertial term, and the term representing the elevation effect on frictional
momentum loss per unit volume of fluid. 265
Figure 7.21. Diagram illustrating the symmetric nonlinear effects of quadratic friction. 265
363
Figure 7.22. Diagrams illustrating the nonlinear interaction of the tide with river flow
via the quadratric friction term, for two cases: the river flow is less than the
maximum tidal current, and the river flow is greater than the maximum
tidal current. 266
Figure 7.23. Plots of the tidal amplitudes M2, M4/M2, and M6/M2 produced by a one
dimensional nonlinear model of Delaware River and Bay compared with
harmonic analysis results from Delaware River and Bay water level data. 269
Figure 7.24. M2 amplitudes produced by a nonlinear numerical model along the
Delaware River and Bay for seven amounts of river discharge, with only
the quadratic friction term turned on. 270
Figure 7.25. Numerical model-produced M4/M2 amplitude ratios along the Delaware
River and Bay for seven different river discharge amounts, with quadratic
friction as the only nonlinear term turned on in the model. 271
Figure 7.26. Numerical model-produced M6/M2 amplitude ratios along the Delaware
River and Bay for seven different river discharge amounts, with quadratic
friction as the only nonlinear term turned on in the model. 272
Figure 7.27. Current profiles from bottom-mounted upward-looking ADCPs at eight
locations in Chesapeake Bay, all measured on October 26, 2006 at noon. 276
Figure 7.28. M2 and N2 constituent ellipses for three different depths for a current station
in the Strait of Georgia. 277
Figure 7.29. An example of an ocean M2 co-amplitude chart for the Atlantic Ocean. 279
Figure 8.1. A fine grid section of the NOS New York -New Jersey Operational Forecast
Model System, and the predicted currents in the same section of the model,
illustrating the complex horizontal structure of the tidal currents including
current shear and eddies. 285
Figure 8.2. Comparison between a surface current field measured by an HF radar
system in lower Chesapeake Bay and the output of a three-dimensional
finite element hydrodynamic model for the same time period. 294
Figure 9.1. A Table 1 page of daily tide predictions for Boston, MA. 301
Figure 9.2. A Table 2 page showing subordinate tide stations. 303
Figure 9.3. A caution in the Tide Tables about the variation in low water time due to
shallow-water effects. 304
Figure 9.4. An example of Table 8 from the 2006 Tide Tables for the East Coast of
North and South America, showing statistical figures related to the tide
prediction accuracy of the reference stations. 305
Figure 9.5. Table 1 daily predictions for a the tidal current reference station at
Chesapeake Bay Entrance from the Tidal Current Tables for the Atlantic
Coast of North America. 311
Figure 9.6. A Table 2 page of subordinate tidal current stations from the Tidal Current
Tables for the Atlantic Coast of North American. 312
Figure 9.7. Rotary tidal currents presented in Table 5 of the Tidal Current Tables for
the Atlantic Coast of North America. 313
364
List of Figures
Figure 9.8. A tidal current diagram for Chesapeake Bay included in the Tidal Current
Tables for the Atlantic Coast of North America. 315
Figure 9.9. A portion of one of the 12 hourly tidal current charts for the Upper
Chesapeake Bay. 320
Figure 9.10. One of the hourly tidal height charts from the Delaware River and Bay
Tidal Circulation and Water Level Atlas. 324
Figure 9.11. One of the hourly tidal circulations charts from the Delaware River and
Bay Tidal Circulation and Water Level Atlas. 325
Figure 9.12. A tide and light diagram produced during World War II to show the times
of high and low tides, plus the times of moonrise, moonset, sunrise, sunset,
and various degrees of twilight for a month. 327
Figure 9.13a. A Tidal Illumination Diagram for LeHarve, France, for July 1944, produced
from tide predictions calculated by the British for the Allied forces. 328
Figure 9.13b. The Explanation of Diagram found on the back of the Tidal Illumination
Diagram in Figure 9.13a. 329
365
List of Tables
Table 2.1. Fundamental astronomical periods of the motions of the earth, moon, and sun. 29
Table 2.2. Tidal constituents and their origin from key astronomical frequencies. 40
Table 2.3. Table showing the effect of the phase difference P = (M2o-K1o-O1o) on the
sequence of tide and the acceleration or retardation of the high or low water. 46
Table 2.4. Compound and overtide constituents generated by four nonlinear terms in the
one-dimensional equations of motion. 59
Table 3.1. Schuremans Table 39 showing the synodic periods for a number of pairs of
semidiurnal tidal constituents and diurnal tidal constituents. 85
Table 3.2. The 37 tidal harmonic constituents that typically had been included in a
harmonic analysis at CO-OPS and its predecessor organizations (unless it
was an extreme shallow-water situation). 88-9
Table 4.1. 55 tidal constituents from a 19-year harmonic analysis of Seattle water level
data, including 30 satellite constituents. 151
Table 4.2. Corrections factors for ten different reference stations in the Strait of Juan
de Fuca-Strait of Georgia for the 29-day period beginning March 1, 1974,
for use in correcting the harmonic constants from other stations with only 29
days of data. 156
Table 7.1. A low-frequency storm effect on tide amplitude near the head of tide in the
Delaware River produced with a nonlinear numerical model. 272
Table A.1. 149 tidal harmonic constituents listed in order of increasing frequency. 334-8
Table A.2. 149 tidal harmonic constituents listed in order of increasing synodic period. 339-5
366
Acknowledgments
Mike Szabados provided the support for this project, and totally believed in the need for this
book; without his support this book would not have been written. Brenda Via managed the
publication and printing process for the book and also improved many of the graphics. Steve Gill
coordinated the internal review process for the book and provided many valuable comments himself.
Chris Zervas, Pat Burke, and Chris Paternostro of CO-OPS provided significant comments and useful
suggestions. Others inside CO-OPS who reviewed the book were Kate Bosley, Karen Earwaker,
Kathleen Fisher, and Laura Rear. Members of the Coast Survey Development Laboratory who also
provided useful comments were Kurt Hess, John Kelley, Richard Patchen, Dick Schmalz, Eugene Wei,
and Aijun Zhang. Outside of NOAA, very beneficial comments and suggestions were provided by four
true experts in the field, Mike Foreman (Institute of Ocean Sciences, Canada), Philip Woodworth
(Proudman Oceanographic Laboratory, UK), David Jay (Portland State University), and Alan Blumberg
(Stevens Institute of Technology). Over the years, working in tides in the Coast and Geodetic Survey
(C&GS) and its later organizational forms in the National Ocean Service in NOAA, the author has had
the privilege of working with many dedicated tidalists, who have contributed much to tidal science and
operations, many of whose names one will see in the References section of this book. In addition to
those mentioned above, this included early in his career, scientists who were carrying on the
advancement of tidal science within the C&GS such as Bernard Zetler, Robert Cummings, D. Lee
Harris, Elmo Long, Robert Dennis, and Steacy Hicks.
About the Author

Dr. Bruce Parker spent most of his career in NOAA and much of his time working on tide related
problems as a specialty even while tackling jobs with a much broader scope. Positions he held at
NOAA included: Chief Scientist of the National Ocean Service; Director of the Coast Survey
Development Laboratory; Director of the World Data Center for Oceanography; Principal
Investigator for the NOAA Global Sea Level Program; and head of the U.S. national tides and
currents program (in a earlier organizational form before it became CO-OPS). Among his awards
are the NOAA Bronze Medal, the Department of Commerce Silver and Gold Medals, and the
Commodore Cooper Medal from the International Hydrographic Organization. Dr. Parker is
presently a Visiting Professor at the Center for Maritime Systems at the Stevens Institute of
Technology. Dr. Parker has written many papers on tidal subjects, some of which are included in
the References section of this book, as well as many tidal analysis programs, some still being used
in some form in CO-OPS. He also had the privilege of organizing the program for the International
Conference On Tidal Hydrodynamics in 1988 and editing the book that resulted. Dr. Parker
received his Ph.D. in physical oceanography from The Johns Hopkins University, and prior to that
an M.S. in physical oceanography from the Massachusetts Institute of Technology, and a B.S./B.A.
in biology and physics from Brown University.
367
368
Index
A C
accelerated tide 46 Cartwright tide potential of 39, 89, 149,
age 333-45
the tide (age of phase inequality) 47 Cartwright number 39, 88-89, 150, 333-45
of diurnal inequality 47 (see also Doodson number)
of parallax inequality 47 centrifugal force 24-26
of the moon 110 compound tide 1, 11, 31, 37, 49, 58-60,
aliasing 35-36, 103 104, 108, 141, 167, 171, 179,
amphidromic point 254, 279 193-94, 231, 236-67, 256-62,
(see node, point of zero range) 264-65, 268, 270, 274, 289, 303
analysis generation of 58-60, 256-62, 285, 289
(see tidal analysis, types of) with same frequency as an astronomical
angular speed 15, 39-40, 47, 84, 89, 91-94, constituent 141, 333
152, 157, 333-45 conservation of mass, volume, and
(see also tidal frequencies) momentum
apogee, lunar 30, 32-35, 38, 49, 58, 113, (see equations of motion)
302 consistency
apogean tides 32-34 in calculated amplitudes and epoch 87,
apogean neap tides 34 142, 160, 164-66, 179, 193, 289
aphelion (solar apogee) 31, 37, 39, 119 in calculating maxima 129-32
astronomical frequencies 11, 29, 35- 40, in calculated Sa and Ssa from year to
91, 108, 333-45 year 120
astronomical periods (inverse of constituent ellipse
astronomical frequencies) 29-30 (see ellipse)
constituents (see tidal constituents)
B continental shelf, tides on 278
baroclinic effects 71, 105, 171, 185, 275 continuous wavelet transform 65, 105-6
benchmark 22 126, 147, 158, 171, 185
bore, tidal (see tidal bore) continuum (nontidal) 35, 37, 82, 132,
bottom boundary layer 274, 284 134-36, 142, 165, 289
Boussinesq Approximation 233 (see also spectra)
build-up factor 142, 163-65, 308 corange (or coamplitude) chart 20, 54,
137, 224-25, 223-26, 255, 279
369
Coriolis force 27, 53, 234 solar declination 29-31, 34

effects in wider waterways 27, 53, definitions 15-24, 42-48
251-55 deterministic (data) 81, 133
effect on inertial currents 78, 136 detiding 7, 123-29
effect on land breeze-sea breeze 78, differences (time, height, speed)
136 from nonharmonic analyses 111-12,
f-plane approximation 234 195-222
Ekman spiral 276 in Table 2 112, 195, 218-19, 302-3,
Rossby radius of deformation 237 308-9, 310, 312, 317-18
cospeed chart 226 variation throughout month 112-19,
cotidal (or co-phase) chart 20-21, 52-53, 195-97, 219-220
223-26 digital tidal prediction products 330-31
current data, analysis of diurnal inequality 19, 34,
(see harmonic analysis) diurnal tide 17, 35, 43-46
(see nonharmonic comparison Doodson tidal potential 39
analysis) Doodson number 30, 89, 333-45
(see tidal analysis, types of) (see also Cartwright number)
current diagrams, 314-15
current meters E
errors from 180-84 ebb current 16, 50
types of 180-84 eccentricity 174
current pole and log line 181 ecliptic 29-31, 34,
cusp (see tidal cusp) obliquity of 31
cycle-by-cycle analysis 111, 112-19, 202-4 (see also orbits)
(see tide-by-tide analysis) elimination 154
(see reversing reduction analysis) (see inference and elimination)
Ekman spiral 276
D (see Coriolis force)
datum (vertical) ellipse
chart 22, 92, 295 rotary tidal current 18
ellipsoid (3-D) 22-24, (see tidal current constituent ellipse)
marine boundaries based on 23, empirical orthogonal function analysis
orthometric 22-24, 137-38
tidal 6, 22-24, (also see separate entry epoch 8, 28, 31, 39, 43, 47, 91-95, 97, 107-8
under tidal datum) (see harmonic constants)
vertical datum transformation tool Epoch, National Tidal Datum 22-23, 46
(VDatum) 24, equilibrium argument 39, 41-42, 69, 91-95,
day 99, 141, 148-49, 152, 158
lunar 15, 29 equilibrium theory 26, 28, 42, 67-68,
sidereal 29 91-92, 152, 154-55, 162
solar 15, 29 equations of motion 232-38
declination continuity equation 232-33
lunar declination finite-difference analog 281-84
equatorial 28-31, 34, momentum equations 233-35
northern or southern 28-31, 34 nonlinear terms in 254-56
(see also tropic tide) physical explanation of 263-67
370
Index
simplifying for idealized cases 223-28 G

equatorial declination 28, 34, geoid 23
(see lunar declination) gravitation (estuarine) circulation 275
equinox
autumnal 29, 34, H
vernal 29, 34, harmonic analysis 8, 39, 42, 91-101,
errors, instrument complex algebra form 97, 179
in current data time series data 180-84 constituents solved for 139-41
in water level data time series data (see also synodic period)
142-46 Fourier-based method 97-99
establishment of the port 47 historical development 8, 10
harmonic tide prediction equation 90-95
F harmonic tidal current prediction
fair tide 7, equation 95-97
filtering 123 inference and elimination 152-59
band pass 128 least-squares method 99-101
band-stop 128 of a long series of high and low waters
high pass 128 106-9
low pass 125-28 of current data 169-94
weights 124, 128 constituent ellipses from harmonic
(see tidal filtering) constants 96, 132, 169, 172-80,
finite-difference technique 283-84 226-29, 275-79
(see numerical hydrodynamic models) mean current from 186
finite-element technique 284 orthogonal axes 179-80
(see numerical hydrodynamic models) summary overview 192-94
flood current 16, 50 special problems with 169-72
folding (see aliasing) of water level data 139-68
folding frequency (see Nyquist frequency) summary overview 166-68
Fourier decompostion of the equations of reference station, use of 154-55
motion 256-62, 273-74 Sa and Ssa, analysis for determining
Fourier harmonic analysis 97-99 122-23
frequency satellite constituents, using 150-51
astronomical 29, 39 short time series, analyzing 150-59, 187
Nyquist 35 small constituents, keeping or rejecting
tidal (see tidal frequency) 142
friction (bottom) 52, 57 harmonic constants 1, 8, 42-48, 90
energy loss due to 57 harmonic constituents (see tidal
damping of tide wave 52, 238 constituents)
generation of compound tides and harmonic-shallow-water corrections method
overtides 59-60, 256-60 103
linearized 238 high and low waters 15
nonlinear effects, types of 255-60 calculating (consistently defined)
frictionally damped progressive wave 52, 129-31
238 diurnal inequality 19
frictionally damped standing wave 52, 240
371
harmonic analysis of a long time series interval, lunitidal (see lunitidal interval)
of 106-9
types of 17-18, 46-47 J
hydraulic tidal currents 76
hydrodynamic models 223-79, 281-97 K
analytical hydrodynamic models 223, Kelvin wave 253-55, 278
238-54
long narrow rectangular waterway L
238-47 lateral inertial effects (on tidal current)
exponentially narrowing waterway 73-75, 273-74
247-50 long-period tides 40-41
numerical hydrodynamic models lunar fortnightly constituent (Mf )
267-73, 281-84 37-38, 40
use in tidal prediction 281-97 lunar monthly constituent (Mm )
advantages 284-87 37-38, 40
disadvantages 287-88 lunar synodic fortnightly constituent
operational uses of 293-97 (MSf) 37-38
sensitivity analysis 292 (see Sa and Ssa)
tidal forcing, proper 288-90 long wave approximation 234
validation of 290-93 (see hydrostatic approximation)
real-time based nowcast/forecast model (see equations of motion)
prediction systems 296-97 low waters (see high and low waters)
wider waterways (Coriolis effects) lunar declination (see declination)
251-55 lunar nodal regression 31, 40,
(see equations of motion) ascending lunar node 31, 40
(see tidal prediction, numerical model) (see node factor)
hydrodynamics 10-11, 15, 48-79, 223-79 lunitidal interval 47, 205,
importance of 10-11 (see also cotidal chart)
hydrographic surveys 295-96
tidal modeling for 295-96 M
tide-based corrections in 295-96 marine boundaries 22-23
VDatum used in 296 maximum floods and ebbs 16
hydrostatic approximation 234 calculating consistently defined 131-32
hydrostatic equation 235 mean current
from a harmonic analysis 172, 186-88
I from a nonharmonic analysis 212-14
inference and elimination 150, 152-58, tidally induced 69, 73-74
187 (see also tidally induced residual
inertial currents 78-79, 186, current)
inertial effects mean cycle analysis 111
(see lateral inertial effects) (see rotary reduction analysis)
(see shallow-water tides) mean sea level 15, 19
inertial terms in equations of motion 236 tidally induced change in 69
internal tides 135, 146, 171, 185, 275 mixed tide 18, 43-46, 135,
interpretation of tidal analysis results models (see hydrodynamic models)
223-79
372
Index
modulation of tidal force 32-34, 37 nonlinear interaction of tides with nontidal

(see also splitting of tidal species) phenomena 63-67
momentum 233 with river flow 63-66, 260-62
(see equations of motion) with storm surge 66-67, 262-63
month nonlinear mechanisms 55, 59, 263-67,
synodic 30 333-45
tropical 29 (see nonlinear tides)
monthly mean analysis 111, 197-202 nonlinear tides
moon asymmetric effect 55-62, 69, 73-75,
(see lunar) 88-89, 259-60, 264-65, 274, 333-46
(see orbits) lateral inertial effects (see lateral
inertial effects on tides)
N physical explanation of nonlinear
narrow basin approximation 236 mechanisms 263-67
(see equations of motion) shallow-water (see shallow-water
narrowing waterway, effect on tide 247-51 tides)
natural period of oscillation 27 symmetric effect (of quadratic friction)
nodal regression (see lunar nodal 56-62, 259-60
regression) zero-frequency effects 70
neap tide 16, 32-33 (see compound tides and overtides)
node (point of zero tide range) 50-53, 55, nontidal phenomena (river flow and storm
77-78, 241, 250, 254, 267 surge)
amphidromic point 268 nonlinear interaction with tide and tidal
quasinode 52-53, 77-78, 224, 241, 244, current 63-67, 146, 260-63, 266-67,
246, 254, 257, 259, 267, 270 270-73
virtual 52, 254 nontidal water level change
node factor (lunar) 37, 39-42, 60-61, 67-69, causes of 19
91-94, 99, 141, 148-50, 152-59, 165, numerical hydrodynamic models (see
167, 171-73, 187, 193 hydrodynamic models)
hydrodynamic effects on 67-69 Nyquist frequency 35
nonharmonic comparison tidal analysis 8,
87, 109-19, 195-222 O
hydrodynamic considerations 196-97 orbits
need for 196 Earth orbit around sun (ecliptic) 27-34
problems with 112-18, 219-20 Moon orbit around Earth 27-34
variation of values throughout a month overtide 1, 11, 36, 55-61, 78, 116, 171-72,
112-18, 219-20 179, 256-60, 251-52, 264, 269, 274
minimizing errors due to 219-20 generation of 256-60
reference station, selecting a 217-18
subordinate station, putting in a Tide or P
Tidal Current Table 218-19 parallax 32
types of 111-12, 197-217 perigee, lunar 30-35, 37, 40-41, 47, 58
monthly mean analysis 197-201 perigean spring tides 34
tide-by-tide analysis 202-5 (see also apogee, lunar)
reversing reduction analysis 205-9 perihelion (solar perigee) 31, 37-38, 119
rotary reduction analysis 210-17
373
period 15 Rayleigh criterion 36, 84-87, 101, 158

natural period of oscillation 27 (see synodic period)
phase age 47 reduction factor 42
phase lag (see epoch) (see node factor)
Poincar waves 254, 278 reference station 2, 87, 90, 103-4, 108,
predicted series 110-11, 113, 116-17, 149, 154-57, 161,
assessing quality of 159-63, 188-90, 164, 189, 191, 195-96, 204
304-6 producing a reference station
changes in bathymetry and shoreline, for Table 1 of a Tide Table 165, 308
looking for 164, 190 for Table 1 of a Tidal Current Table
comparison of high and low waters 191, 317
(predicted versus data) 161-62, selecting for a nonharmonic comparison
304-6 analysis 217-18
comparison of maximum floods and use of in improving a short harmonic
ebbs (predicted versus data) analysis 154-56
188-89 reflected tide wave 51
residual series, spectral analysis (see standing wave)
spectral analysis of 162, resonance 27, 50-51
189 residual current
transient tidal oscillations, (see tidally induced residual current)
looking for 169-70, 198-99 residual time series 100-1, 121, 123,
build-up factors, use and misuse of 125-26, 147-49, 159-68, 185-86, 188-90,
163-64 192, 194
reference station, for 165, 191 spectral analysis of 161, 191,
prediction (see tidal prediction) transient tidal oscillations, looking for
principal component analysis 137-38 162, 192
products, tidal prediction (see tidal response method 5, 9, 90, 102-3, 104, 105,
prediction products) 133, 137, 290
progressive tide wave admittance 102-3
frictionally damped 49, 51, 238-41, credo of smoothness 103
244-45, 253, 264, 266 radiational input 103
frictionless 49-52, 250 resonance 27, 49, 50, 55, 67, 72, 242
retarded tide 50
Q Reynolds equations 235
quality assessment (see equations of motion)
of harmonic analysis of current data reversing reduction analysis 111-12, 116,
190-94 205-10, 212, 216, 219, 221
of harmonic analysis of water level data reversing tidal current
158-63 defined 16, 18,
of nonharmonic analysis 219-20 river flow (river discharge)
of tidal current predictions 190-94 effect on tide (see nonlinear interaction
of tide predictions 158-63 of tide with nontidal phenomena)
quasinode (see node, point of zero range) effect on nontidal water level 267
Rossby radius of deformation 237, 251
R rotary reduction analysis 111-12, 204-5,
range (see tide range) 207, 210-19, 221
374
Index
rotary tidal current spectra 36-37

defined 17, 18, 214, (see tidal spectra)
spectral analysis 132-37
S autocorrelation method 133
Sa and Ssa 38, 88, 103, 119-23, 143, 145, autospectrum 133
148, 162 cross-spectrum 133
considerations in whether to use 120-22 cross-spectral analysis 135-37
direct method using the FFT 134
meteorological origins of 119-23 maximum entropy method 134
methods for calculating 122-23 maximum likelihood method 134
satellite constituents 37, 39, 41, 92-93, 141, of water level data 133-34
148-50, 151, 158, 165 of current data 135-36
(see node factor) Parsevals theorem 133
satellites 39 rotary spectral analysis 135-36
sea level 15, 19, 22 (see also tidal cusps)
(also see mean sea level) spectroscopy 37
sea level change splitting of tidal species 37
absolute 22 (see also modulation of tidal force)
relative 22 spring tide 16, 32-34, 46-48,
steric sea level change 119-20, 122, 159 standard port (see reference station)
tidally induced 69 standing tide wave
secondary port (see subordinate station) frictionally damped 50, 51, 57, 240-41,
semidiurnal tide 18-19, 43-46, 244, 264, 266
sequence of tide 45-46 frictionless 50, 250
shallow-water approximation 234 stand of tide 61
(see hydrostatic approximation) stationary data 81
(see equations of motion) stochastic (data) 81, 133
shallow-water constituents 11, 38, 39, storm surge 6, 19, 66-67, 262-63
57-78, 60-2, 75, 141, 156-67, 165, 166, storm tide 6,
179, 191, 193-94, 229, 257, 260, 274 subordinate station 2, 87, 90,110-11, 113,
(see compound tides and overtides) 117,154-55, 165, 191, 195-96, 200,
(see also nonlinear tides) 202-4, 206, 210, 213, 216-221, 299-300,
simultaneous observations, comparison of 216-21, 299-300, 302-3, 307-310, 312,
(see tide-by-tide analysis) 316-18, 322, 330
shoreline putting
legal U.S. shoreline 22-23 in Table 2 of a Tide Table
mean high water shoreline defined 22 218-19, 302, 308
slack water 16, 48-50, 57, 76, 107, 112, in Table 2 of a Tidal Current Table
129, 132, 188, 194, 205, 208, 211, 218-89, 310, 317
214-15, 244, 264, 266 in Table 5 of a Tidal Current Table
solar declination (see solstice) 218-89, 313, 317
solstice on a Tidal Current Chart 218-89, 322
summer 29, 34, sun
winter 29, 34, (see solar)
(see declination) (see orbits)
species concordance method 104 superfine resolution method 105
375
synodic period 84-89, 99, 101, 139-41, flood current 16

160, 166-68, 192-94, 333-45 hydraulic tidal currents 76
(see also Rayleigh criterion) limitations on predictions 78-81
locations with high speeds 20-21
T mean current
tidal analysis, types of 81-138 tidally induced 69, 73-74, 195
simple demonstration of extracting tidal whether to include 195-96
constituents from a data time series nonlinear lateral inertial effects on
83-85 73-75, 273-74
(see continuous wavelet transform nontidal influences, potential effects of
analysis) 184-86
(see harmonic analysis, Fourier) orthogonal axes, selection of 179-80
(see harmonic analysis, least squares) reversing 16, 18, 205
(see harmonic-shallow-water rotary 18, 210
corrections method) shallow-water effects on 75-76
(see nonharmonic analysis) slack water 16, 48-50, 57, 76, 107, 112,
(see response method) 129, 132, 188, 194, 205, 208, 211,
(see species concordance method) 214-15, 244, 264, 266
(see spectral analysis) special aspects of 69-80
(see superfine-resolution method) special problems when analyzing
(see tidal filtering) 169-72
tidal bore 7, 55, 57, 299 types of 19
tidal circulation and water level forecast variation over depth 71-72, 274-77
atlases 323-26 variation over horizontal distance 72-
tidal bulges (in equilibrium theory) 24-25 75
tidal constituents tidal current charts 318-23
amplification of 201, 243 advantages and limitations of 321-22
constituents solved for 139-41, 166, construction of 322-23
179, 192 tidal current tables 309-18
(see synodic period) construction of 317-18
origin of 31-39, 88-89, 333-45 current diagrams 314-15
naming convention 39 limitations of 316-17
calculation of from data (see harmonic Table 1 (harmonically produced
analysis) reference stations) 310
shallow-water (see shallow-water Table 2 (nonharmonically produced
constituents) subordinate stations) 310-12
spectral lines for 38 Table 5 (for rotary tidal currents) 313
table of 37 constituents 88-89 tidal cusp 65, 134-36, 263
table of 149 constituents 334-45 tidal datum 7, 22
tidal current types of 22-24,
constituent ellipses 96, 136, 169, lowest astronomical tide 23
172-79, 180, 191, 193 mean high water 22-23
defined 15 mean lower low water 22-23
ebb current 16, 50 relationship to marine boundaries 22-23
errors, types of instrument errors relationship to shoreline 22-23
180-84 (see also datums (vertical))
376
Index
tidal filtering 123-29 tidal zoning 295

Doodson filter 128 continuous water level zoning 295
tidal frequencies 15, 94 (see hydrographic surveys)
derivation of 35-40 tidally induced mean sea level change 69,
(see also angular speed) 258
tidal harmonic constituent (see tide tidally induced residual current 69, 73-74,
constituent) 257, 273-74
tidal illumination diagram, 328-29 tide
tidal period 15-16 astronomical versus hydrodynamic
tidal prediction aspects 15, 19-20
complex tidal prediction equation astronomical considerations 27-48
97, 179 hydrodynamic considerations 48-80
harmonic tide prediction equation defined 15
90-95 generation of 24-27
harmonic tidal current prediction sequence of 46
equation 95-96 type of 17-19, 43
numerical model, using a 281-97 versus water level 6
advantages 284-87 tide and light diagram 326-27
disadvantages 287-88 tide-by-tide analysis 112, 115, 202-5
sensitivity analysis 292 tide gauge (tide gage)
tidal forcing, proper 288-90 (see water level gauge)
validation of 290-93 tide-generating potential 39, 88-89, 149,
operational uses of 293-97 333-45
uses of 6-7 tide predicting machine 3-4, 8-9
(see predicted series, assessing quality tide-producing forces 25, 39
of) tide range 15, 48
tidal prediction products 299-331 hydrodynamics effects on 49-45,
(see tide tables) 241-43, 248-50, 251-54
(see tidal current tables) large ranges, locations with 21, 52
(see tidal current charts) types of 48
(see tidal circulation and water level (see also corange chart)
forecast atlases) tide staff (see water level gauge)
(see tide and light diagram) tide tables 299-309
(see digital tidal prediction products) construction of 308-9
tidal range (see tide range) limitations of 307-8
tidal rectification (see tidally induced Table 1 (harmonically produced
residual current) reference stations) 300-2
tidal spectra 35-39, 82 Table 2 (nonharmonically produced
tidal spectral lines 38, 41, 82, 136 subordinate stations) 302-4
tidal stream (see tidal current) tropic tide 43
tidal stream atlas (see tidal current chart) turbulence closure scheme 235, 275, 290
tidal wavelength 50, 55, 239-46, 249-51, (see equations of motion)
254, 263, 267, 278 type of tide 17-19, 43-45
tidal wave number 238, 248, 251, 253 (see also semidiurnal, mixed, and
diurnal)
377
U
uses of tidal analysis and prediction 6-7
V
vertical datum
(see datum, vertical)
VDatum (vertical datum transformation
tool) 24, 296
virtual node (see node, point of zero
range)
vulgar establishment 47
W
wavelength
(see tidal wavelength)
water level
nontidal effects on 19, 267
versus tide 6
water level data, analysis of
(see harmonic analysis)
(see nonharmonic comparison
analysis)
(see tidal analysis, types of)
water level gauge 22, 143-46
errors from 143-46
types of 143-46
wave number (see tidal wave number)
Y
year
tropical 31
Z
zero-frequency effects from nonlinear tidal
interaction 69, 73-74, 257-78, 273-74
378

Tidal Analysis and Predictions PDF

Uploaded by

Copyright:

Available Formats

Tidal Analysis and Predictions PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Tidal Analysis and Predictions PDF

Uploaded by

Copyright:

Available Formats

What are some of the main tidal concepts discussed in the document?

What are some of the main tidal concepts discussed in the document?

How are tidal frequencies derived from basic astronomical frequencies?

How are tidal frequencies derived from basic astronomical frequencies?

U.S.

Tidal Analysis and Prediction

Tidal Analysis and Prediction

Bruce B. Parker, Ph.D.

Silver Spring, Maryland

Library of Congress Control Number: 2007925298

U.S. DEPARTMENT OF COMMERCE

National Oceanic and Atmospheric Administration

National Ocean Service

Center for Operational Oceanographic Products and Services

Mention of a commercial company or product does not constitute an endorsement

Request for copies should be addressed to

2. Theory Behind Tidal Analysis and Prediction .15

3. Methods of Tidal Analysis and Prediction .81

4. Harmonic Analysis of Water Level Data .139

4.2.1 Schuremans Inference and Elimination Technique 152

5. Harmonic Analysis of Current Data .169

6. Nonharmonic Analysis of Water Level and Current Data .195

6.4 Tide-by-Tide Analysis 202

7. Interpretation of Tidal Analysis Results Based On

7.7. Variation of Tidal Currents With Depth 274

8. The Use of Numerical Hydrodynamic Models For

9. Products for Presenting Tide and Tidal Current Predictions .299

List of Figures .359

1.1 Purpose of This Book

Figure 1.1. The Harris-Fischer tide predicting machine.

1.3 The Uses for Tidal Analysis and Prediction

1.4 Brief Historical Background

Figure 1.2. Gear and pulley system of an early analog tide

1.5 The Importance of Hydrodynamics

1.6 Organization of This Book

2.1 Short Introductory Overview the Big Picture

2.1.1 Some Definitions

Figure 2.1. A simple tide curve with some definitions of

2.1.2 The Generation of Tides

2.2 Astronomical Considerations

Figure 2.11. When the moon is at maximum declination north or south

2.2.2 The Origins of the Largest Tidal Harmonic Constituents

2.2.3. Derivation of Tidal Frequencies from the Basic Astronomical Frequencies

Figure 2.15. A demonstration of aliasing. The higher-frequency sine wave

Figure 2.16. A water level spectrum from near Anchorage,

2.2.4. The Origin of Node Factors and Equilibrium Arguments

2.2.5 Definitions Revisited, With Relationship to Harmonic Constants

Figure 2.19. Tide curves for particular combinations of (K1+O1)/M2 amplitude

Age of the tide (in hours) = 0.984 (S2o - M2o)

Age of parallax inequality (in hours) = 1.837 (M2o - N2o) .

Age of diurnal inequality (in hours) = 0.911 (K1o - O1o)

Mean range (long period average) = 2.2 M2

2.3 Hydrodynamic Considerations

Figure 2.21. An idealized frictionless tide wave propagating up a river as a progressive

2.3.2 Nonlinear Effects of Shallow Water Overtides and Compound Tides

Figure 2.25. Typical tide curves (over 1 tidal

M2 interacting Tidal Constituents Generated With Frequency:

N2 MN (Mm) 2MN2 (L2) MN4 2MN6 4MN6

S2 MS (MSf) 2MS2 (2) MS4 2MS6 4MS6