International Association

of Geodesy Symposia
Klaus-Peter Schwarz, Series Editor

Springer-Verlag Berlin Heidelberg GmbH

 International Association
of Geodesy Symposia
Klaus-Peter Schwarz, Series Editor

Symposium 101: Global and Regional Geodynamics

Symposium 102: Global Positioning System: An Overview
Symposium 103: Gravity, Gradiometry, and Gravimetry
Symposium 104: Sea Surface Topography and the Geoid
Symposium 105: Earth Rotation and Coordinate Reference Frames
Symposium 106: Determination of the Geoid: Present and Future
Symposium 107: Kinematic Systems in Geodesy, Surveying,
and Remote Sensing
Symposium 108: Applications of Geodesy to Engineering
Symposium 109: Permanent Satellite Tracking Networks for
Geodesy and Geodynamics
Symposium 110: From Mars to Greenland: Charting Gravity with
Space and Airborne Instruments
Symposium 111: Recent Geodetic and Gravimetric Research in Latin America
Symposium 112: Geodesy and Physics of the Earth: Geodetic Contributions
to Geodynamics
Symposium 113: Gravity and Geoid
Symposium 114: Geodetic Theory Today
Symposium 115: GPS Trends in Precise Terrestrial, Airborne,
and Spaceborne Applications
Symposium 116: Global Gravity Field and Its Temporal Variations
Symposium 117: Gravity, Geoid and Marine Geodesy
Symposium 118: Advances in Positioning and Reference Frames
Advances in Positioning
and Reference Frames

lAG Scientific Assembly

Rio de Janeiro, Brazil, September 3-9, 1997

Convened and Edited by

Fritz K. Brunner

Springer
Volume Editor Series Editor
Professor Dr. Fritz K. Brunner Professor Dr. Klaus-Peter Schwarz
Engineering Surveying and Metrology University of Calgary
Technical University of Graz Department of Geomatics Engineering
Steyrergasse 30 2500 University Drive N. W.
A-8010 Graz Calgary, Alberta T2N 1N4
Austria Canada

ISBN 978-3-642-08425-6 ISBN 978-3-662-03714-0 (eBook)

DOI 10.1007/978-3-662-03714-0

Library of Congress Cataloging-in-Publication Data

Advances in positioning and reference frames : international symposium no. 118 : Rio de ]aneiro, Brazii,
September 3-9, 1997 1 convened and edited by Fritz K. Brunner.
p. cm. -- (International Association of Geodesy symposia : Symposium no. 118)
Includes bibliographical references.

1. Global Positioning System--Congresses. 2. Geodesy--Congresses.

I. Brunner, F. K. (Fritz K.), 1943-
. II. Series: International Association of Geodesy symposia:
symposium 118
G109.5.A39 1998 623.89 '3--dc21 98-25812 CIP
This work is subject to copyright. All rights are reserved, whether the whole or part of the material is concerned,
specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on
microfilms or in other way, and storage in data banks. Duplication of this publication or parts thereof is permitted only
under the provisions of the German Copyright Law of September 9, 1965, in its current version, and permission for use
must always be obtained from Springer-Verlag. Violations are liable for prosecution act under German Copyright Law.
© Springer-Verlag Berlin Heidelberg 1998
Originally published by Springer-Verlag Berlin Heidelberg New York in 1998.
Softcover re print of the hardcover 1st edition 1998
The use of general descriptive names, registered names, trademarks, etc. in this publication does not imply, even in
the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and
therefore free for general use.
Typesetting: Camera ready by editor/author
Cover layout: design & production GmbH, Heidelberg
SPIN: 10680111 32/3020 - 5 4 3 2 1 o - Printed of acid-free paper
 FOREWORD

Klaus-Peter Schwarz, lAG President

The University of Calgary

The Scientific Assembly of the International Association of Geodesy (lAG) was held
from September 3 to 9, 1997 in Rio de Janeiro, Brazil, in conjunction with the 18th
Brazilean Congress of Carthography. This was the first time that one ofthe major lAG
meetings took place in Brazil. It provided an opportunity to showcase the progress of
geodetic work in South America through campaigns such as SIRGAS. It also provided an
opportunity for a large group of international experts to present the state of the art in ge-
odesy and geodynamis and to interact with their hosts on possibilities of future coopera-
tion. For the lAG, it continued a trend, started in Beijing four years ago, to hold major
geodetic meetings outside of Europe and North America. The International Geoid School
which was held in Sao Paulo following this meeting showed another facet of this grow-
ing internationalization ofIAG activities and services.

The scientific program of the meeting consisted of three symposia and two special ses-
sions, namely
Symposium 1: Advances in Positioning and Reference Frames
Symposium 2: Gravity and Geoid
Symposium 3: Geodynamics
Special Session 1: lAG Services
Special Session 2: Geodesy in Antartica.
Papers presented at the first symposium are published in volume 1 of these proceedings,
while papers of symposia 2 and 3 as well as special session 2 are contained in volume 2.
Papers presented at special session 1 will be published separately. More details on the
individual volumes are given in the prefaces written by the convenors.

The meeting was jointly organized by the lAG and the Brazilean Society of Cartography,
Geodesy, Photogrammetry and Remote Sensing. The two local organizing committees
worked closely together to economize on some of the organizational aspects and to guar-
antee a smooth running of two conferences in the same place. Thanks go to our Brazilean
colleagues for their hard work and their warm hospitality. On the geodetic side, special
thanks go to Professor D. Blitzkov, the national representative of the lAG who chaired
the lAG Local Organizing Committee (LOC), and the dedicated group of individuals
working working with him. Similarly, Professor W. Torge, past president of the lAG,
who provided the liason between the LOC and the lAG Executive deserves a special vote
of thanks. Finally, the symposium and session convenors who not only organized the
scientific program but also took care of organizing the review rocess and editing these
proceedings, are thanked for their outstanding efforts.
 PREFACE

Over the past few years we have experienced tremendous progress in GPS positioning.
Therefore, it seemed appropriate to organise a special symposium for the review of current
developments in precise GPS positioning with special emphasis on the work of the related
Special Study Groups oflAG.

"Advances in Positioning and Reference Frames" was one of the symposia held during the
Scientific Assembly of lAG in Rio de Janeiro in September 1997. This symposium com-
prised six sessions: Maintenance and Densification of Reference Frames, GPS Reference
Networks, the SIRGAS Project, Current Developments in Precise GPS Positioning, GPS
Kinematic Applications, and a very successful Poster Session. During the symposium, 25
invited papers were presented and 70 posters were displayed.

This volume contains the reviewed contributions to the symposium. Its contents reflect the
exciting and steadily growing developments of fundamental GPS work as well as novel
applications of static and kinematic GPS surveying techniques. The maintenance and the
densification of reference frames are treated for the purpose of establishing global and
regional GPS networks. The scientific achievements of the South American Geocentric
Reference System project (SIRGAS) are discussed. Congratulations to all SIRGAS con-
tributors for their outstanding achievements! Several contributions review the state of the
art of GPS analysis techniques, ambiguity resolution methods, as well as GPS antenna and
site problems. New applications of kinematic GPS positioning and the quality control is-
sues of real-time GPS positioning are presented.

The individual sessions of the symposium were organised by the conveners Yehuda Bock,
Claude Boucher, Luiz Fortes and Chris Rizos, who also carried the main burden of re-
viewing the manuscripts. Ms. S. Schmuck assisted in the organisation of the papers. I
would like to thank everybody who helped with the preparation of the symposium and the
proceedings, especially the authors for their excellent contributions.

May 1998 Fritz K. Brunner

 CONTENTS

Session 1: Maintenance and Densification of Reference Frames

Convener: Claude Boucher

WGS 84 - Past, Present and Future 1

JA. Slater and S. Malys (invited)

ITRF Densification and Continuous Realization by the IGS 8

G. Blewitt, C. Boucher, P.B.H Davies, MB. Heflin, TA. Herring and J Kouba
(invited)

The Use of the EUREF Permanent GPS Network for the Maintenance of the 18
European Terrestrial Reference Frame
C. Bruyninx, J Do usa, W Ehrnsperger, N Fachbach, J Johansson, F. Vespe,
L. Ferraro, A. Nardi, M Figurski, T Springer, G. Weber and A. Wiget (invited)

The European Vertical GPS Reference Network Campaign 1997 - 27

Concept and Status
J Ihde, J Adam, W Gurtner, B.G. Harsson, W SchlUter and G. Woeppelmann
(invited)

From UELN-95 to EVS 2000 - European Activities for a Continental Vertical 35

Datum
W Augath, F. Brouwer, H Lang, J van Mierlo and M Sacher

Global GPS Networks and the Determination of Earth Rotation Parameters 43

V Ashkenazi, T Moore, L.J Symons, W Chen and MH 0. Aquino

Realization of the ITRF-94 in Thailand and Malaysia by Use ofa Combined 49

Network for Geodynamics and National Survey
M Becker, E. Reinhart, H Seeger, S. Mingsamon, C. Boonphakdee and S. Abu
The ITRF96 Realization of the International Terrestrial Reference System (Abstract) 57
C. Boucher, Z. Altamini and P. Sillard

Modernizing the Reference Framework for Canada's Maritime Provinces 59

R.J. Gaudet

Permanent GPS Tracking Network including the Mediterranean Area 66

G. Weber, M Becker, P. Franke, W SchlUter and H Seeger

Session 2: GPS Reference Networks

Convener: Fritz K. Brunner

The Brazilian Network for Continuous Monitoring of GPS (RBMC): Operation 73

and Products
L.P.s. Fortes, R. T Luz, K.D.Pereira, S.MA. Costa and D. Blitzkow (invited)

Towards an Austrian GPS Reference Network 79

R. Weber (invited)

Real-Time Differential GPS Error Modelling in Regional Reference Station 86

Networks
L. Wanninger (invited)

Processing Strategies for Regional GPS Networks 93

M Rothacher, TA. Springer, S. Schaer and G. Beutler (invited)

A Continental Wide Area Differential GPS Strategy 101

MH 0. Aquino, V Ashkenazi and T Moore

Possibility of a Dynamic Cadastre for a Dynamic Nation 107

G. Blick and D. Grant

Preliminary Stability Test for the Regional GPS Tracking Stations in Taiwan 114
C. C. Chang and R. G. Chang

The Use of Heuristics in the Design ofGPS Networks 120

P. Dare and H Saleh

Transformations of GPS Coordinates and Heights in Trinidad 125

K. Edwards, J. Barbaste and J. Saleh

Vectors Connecting the Geodetic Points at Metsahovi and Sjokulla 131

J. Jokela

x
Use of the Finnish Pennanent GPS Network (FinnNet) in Regional GPS Campaigns 137
H Koivu/a, M Ollikainen and M Poutanen

The New Swiss National Height System LHN95 143

U Marti and D. Schneider

The Non-Fiducial Approach Applied to GPS Networks 149

JF.G. Monico and E.S da Fonseca Junior

The Statefix West Australian GPS Network 155

MP. Stewart, H Houghton and X Ding

POLREF-96 - The New Geodetic Reference Frame for Poland 161

JB. Zielinski, A. Lyszkowicz, L. Jaworski, A. Swiatek, R. Zdunek and S. Gelo

Session 3: SIRGAS Project

Convener: Luiz Paulo S. Fortes

An Overview of the SIRGAS Project (Abstract) 167

L.P.S Fortes (invited)

The Definition and Realization of the Reference System in the SIRGAS Project 168
M Hoyer, S Arciniegas, K Pereira, H Fagarc!, R. Maturana, R. Torchetti,
H Drewes, M Kumar and G. Seeber (invited)

Time Evolution of the SIRGAS Reference Frame 174

H Drewes (invited)

Results of the SIRGAS 95 GPS Network Processing at DGFIII 180

K Kaniuth, H Drewes, K Stuber, H Tremel and J Moirano (invited)

The Integration of Brazilian Geodetic Network into SIRGAS 187

SMA. Costa, KD. Pereira and D. Beattie

Adjustment of the New Venezuelan National GPS Network within the SIRGAS 193
Reference Frame
H Drewes, H Tremel and IN Hernandez

Realization of a Geocentric Reference System in Argentina in Connection 199

with SIRGAS
J Moirano, C. Brunini, H Drewes and K Kaniuth

XI
Some Considerations Related to the New Realization ofSAD-69 in Brazil 205
L. C. Oliveira, J.F G. Monico, M C. Santos and D. Blitzkow

The IGS Regional Network Associate Analysis Center for South America at DGFIII 211
W Seemuller and H Drewes

The Redefinition of the Geodetic Reference System of Uruguay into SIRGAS Frame 217
WH Subiza, R.R. Perez, F Barbato and S.MA. Costa

Session 4: Current Developments in Precise GPS Positioning

Convener: Yehuda Bock

GPS Ambiguity Resolution for Navigation, Rapid Static Surveying, and Regional 223
Networks
P.J. de Jonge (invited)

GPS Antenna and Site Effects 229

J.M Johansson (invited)

Precise GPS Positioning Improvements by Reducing Antenna and Site Dependent 237
Effects
G. Seeber, F Menge, C. Valksen, G. Wubbena and M Schmitz (invited)

IGS Orbit, Clock and EOP Combined Products: An Update 245

J. Kouba and Y Mireault (invited)

The Use ofGPS for Monitoring of the Ionospheric Disturbances 252

L. W Baran and II Shagimuratov

An Integrated GPS Monitoring System for Site Investigation of Nuclear Waste 259
Disposal
R. Chen and J. Kakkuri

Determination of Stochastic Models of GPS Baselines in GPS Network Adjustment 265

X Ding, M Stewart and J. Chao

The Assisi Landslide GPS Network 271

D. Dominici, F. Radicioni, S. SeW and A. Stoppini

Use of a Regional Ionospheric Model in GPS Geodetic Applications 277

M Gende, C. Brunini and A. Kleusberg

XII
Instantaneous Ambiguity Resolution for Medium-Range GPS Kinematic 283
Positioning Using Multiple Reference Stations
S. Han and C. Rizos

Exploiting the SIRGAS Colocations for Determining Elevation Dependent Phase 289
Center Variations of Geodetic GPS Antennas
K. Kaniuth

Real-Time Failure Detection and Repair in Ionospheric Delay Estimation Using 295
GPS by Robust and Conventional Kalman Filter State Estimates
L. -s. Lin, C. Rizos and S. Mertikas

GPS Levelling Results from Two Test Areas in Finland 301

M Ollikainen

Quality Control Algorithms for Permanent GPS Receiver Applications 307

C. Rizos, L. -s. Lin, S. Han, L. Troyer and S. Mertikas

Fast Ambiguity Resolution in Network Mode 313

P.J G. Teunissen, P.J de Jonge, D. Odijk and C. C.JM Tiberius

Session 5: GPS Kinematic Applications

Convener: Chris Rizos

High Precision GPS Kinematic Positioning: Progress and Outlook 319

C. Rizos and S. Han (invited)

Kinematic Positioning Using Adaptive Filters and Multiple DGPS Receiver 325
Configurations
A.M Bruton (invited)

Estimating the Residual Tropospheric Delay for Airborne Differential GPS 331
Positioning
JP. Collins and R.B. Langley (invited)

High-Accuracy Airborne Integrated Mapping System 337

D.A. Grejner-Brzezinska (invited)

Airborne GPS Performance during a Photograrnmetric Project 343

P. Baldi, M Marsella and L. Vittuari

Signal Distortion in High Precision GPS Surveys (Abstract) 349

F.K. Brunner and H Hartinger

XIII
Testing GPS Approaches for Civil Aviation in the Swiss Alps 351
A. Geiger, M Scaramuzza, M Co card, H -G. Kahle, H Lang, R. Aebersold,
B. Meier and A. Dose

Static and Kinematic Positioning with GPS for the Construction and Maintenance 357
of High Speed Railway Lines
H Kahmen

Kinematic GPS for Ice Sheet Surveys in Greenland 363

K. Keller, R. Forsberg and c.s. Nielsen

A Reweighted Filtering Algorithm and its Application to Open Pit Deformation 369
Monitoring
M Jia, M Tsakiri and M Stewart

Permanent Automatic GPS Deformation Monitoring Systems: A Review of System 375

Architecture and Data Processing Strategies
C. Roberts and C. Rizos

Modified GPS-OTF Algorithms for Bridge Monitoring: Application 381

to the Pierre-Laporte Suspension Brigde in Quebec City
R. Santerre and L. Lamoureux

The GPS Component of the Project for Digital Mapping of the KARST Aquifer 387
System Near Curitiba, Brazil (Abstract)
M C. Santos, A.J.B. Vieira and H Firkowski

Kinematic GPS Positioning with Adaptive Kalman Filtering Techniques 389

J. Wang, M Stewart and M Tsakiri

XIV
 WGS 84 - PAST, PRESENT AND FUTURE
James A. Slater and Stephen Malys
National Imagery and Mapping Agency
Bethesda, Maryland 20816-5003, U.S.A.

ABSTRACT

The World Geodetic System (WGS) was conceived as a practical geodetic reference
system that maintains consistency with the best scientific terrestrial reference system at the
time but also retains some stability. It has evolved from its beginnings as WGS 60 to its
present manifestation as WGS 84. A significantly improved Earth Gravitational Model
and global geoid were released in 1996, and the realization of the WGS 84 reference frame
is consistent with the International Terrestrial Reference Frame (ITRF) 1994 at the 5-cm
level. Evaluations of WGS 84 relative to ITRF94 have been made by comparing
International GPS Service (IGS) GPS orbits with the National Imagery and Mapping
Agency's precise orbits and by computing WGS 84 positions for a globally distributed set
of IGS stations with known ITRF94 coordinates. The results indicate that there is no
practical difference between ITRF94 and WGS 84 for mapping, charting, navigation and
many survey applications.

HISTORICAL PERSPECTIVE AND MOTIVATION

WGS 72 and Before

Almost 40 years have passed since the U. S. Department of Defense (DoD) began
developing a World Geodetic System (WGS) in the late 1950s. The impetus for this was
the need to inter-relate maps, charts and geodetic information from different local datums,
and the development of satellites and intercontinental ballistic missiles. WGS 60 was the
first system introduced by the DoD and only used satellite data to derive the ellipsoid
flattening. Large amounts of Doppler and optical satellite tracking data and advances in
computer technology prompted development of WGS 66 and then WGS 72. The WGS
72 ellipsoid and associated parameters were defined in a manner consistent with the
approach used by the International Union of Geodesy and Geophysics (IUGG) in its
establishment of Geodetic Reference System 1967 (GRS 67). An equipotential ellipsoid
was adopted along with four defining parameters -- semi-major axis, Earth's gravitational
constant and angular velocity, and the second degree zonal harmonic coefficient of the
geopotential. Datum transformations were derived to convert local coordinates to WGS
72 for stations where Doppler satellite receivers were colocated over local geodetic
control points. A geoid and geopotential model were also derived as a part ofWGS 72.
 Development of World Geodetic System 1984 (1984-1993)

The World Geodetic System 1984 (WGS 84) is a Conventional Terrestrial Reference
System which includes in its definition a geocentric coordinate system, a reference
ellipsoid, a consistent set of fundamental constants, and an Earth Gravitational Model and
associated global geoid. [DMA, 1991] During the last few years, several refinements have
been made to the system, but the name "WGS 84" has been retained. As has always been
the case, enhancements to the system have become necessary due to demands from users
for higher accuracy and precision, new and higher quality sources of data, and improved
data processing techniques. An additional factor has been the development and impact of
the Global Positioning System (GPS) .

. Inadequacies in WGS 72, better computational techniques, improved satellite data

(Doppler, satellite laser ranging, satellite radar altimetry), new surface gravity data, and
concurrent efforts to upgrade other geodetic reference systems in Europe, Australia and
North America all contributed to the development of WGS 84. The initial realization of
the WGS 84 reference frame was based on a world-wide set of coordinates derived from
Transit satellite Doppler data in the NSWC 9Z-2 reference frame. The origin and scale of
this frame were adjusted to align it with the Bureau International de I'Heure (BIH)
Terrestrial Reference System of 1984. Absolute positions computed in the new reference
frame had an accuracy (I-sigma) of ±1 m in each horizontal component and ±2 m in
height.

In order to maintain consistency with the recommendations of the IUGG, Geodetic

Reference System 1980 (GRS 80) was adopted for use in WGS 84. [Moritz, 1980] A
geocentric equipotential ellipsoid of revolution was chosen as the reference ellipsoid and
four fundamental constants were defined:

Semi-major axis of ellipsoid of revolution (a) = 6378137 m

Earth's gravitational constant including atmosphere (GM) = 3986005 x 108 m3s-2
Mean angular velocity of the Earth (co) = 7292115 x 10- 11 rad s-1
Normalized second degree zonal harmonic coefficient of the gravitational potential (C 20)
= -484.166685 x 10-6

The value for C 20 was computed from the GRS 80 value for geopotential coefficient J2
(108263 x 10-8 ) using the relationship C 20 =-Ji..J5 and truncating the result to 8 significant
figures. The permanent tidal deformation was excluded from C20 for compatibility with
GRS 80. Geometrical constants derived from C20 and J2 have minute differences;
therefore, the ellipsoid used with WGS 84 is referred to as the WGS 84 ellipsoid rather
than GRS 80. For all practical purposes, the GRS 80 and WGS 84 ellipsoids are identical.
The difference in the semi-minor axes is O. 1 mm and the difference in the flattening is less
than 1 x 10- 10 .

A new Earth Gravitational Model was developed using the latest satellite Doppler, GPS,
laser ranging and radar altimetry data along with surface gravity data. The model was a
spherical harmonic expansion of the gravitational potential to degree and order 180. A
global geoid was derived from this with a mean accuracy of 2-6 m.

WGS 84 was designed to be a practical standard global, geocentric horizontal datum for
mapping, charting, geodesy and navigation. Local datums could be mathematically related
to WGS 84 and to each other through datum transformations derived from colocated
positions. As of 1994, transformation parameters were available for over 115 local and
regional datums. In addition, WGS 84 point positions were accurate enough to provide

2
geodetic control for maps and charts up to 1:5,000 scale which satisfied most
requirements.

ENHANCEMENTS (1994-1996)

Reference Frame and Fundamental Constants

As of June 1987, the Defense Mapping Agency (DMA) had established five permanent
GPS monitor stations in Bahrain, England, Australia, Ecuador and Argentina. These were
geographically located to complement the U.S. Air Force's GPS monitor stations in
Diego Garcia, the United States (Colorado and Hawaii), Ascension Island, and Kwajalein.
WGS 84 coordinates for these stations were computed from Transit satellite Doppler
observations. Since the precise and broadcast orbits for GPS were defined in this Transit-
based realization of the reference frame, geodetic positions derived from them were also in
this frame.
In 1992, the International Earth Rotation Service (IERS) Standards recommended a value
of 3986004.418 x 108 m3s-2 for GM. [McCarthy, 1992] Comparisons using this value and
the original WGS 84 value indicated that the WGS 84 GM value was introducing a 1.3 m
radial error in DMA's and the Air Force's GPS orbits. Based on this information, DMA
and the Air Force adopted the new IERS GM value for high accuracy orbit determination
work. Users whose work was not affected by this small change in GM continued to use
the original GM value.
A number of studies between 1988 and 1994, showed evidence of a systematic ellipsoid
height bias between Doppler-derived coordinates and GPS-derived coordinates for the
same site. In order to eliminate the height bias and at the same time obtain a self-
consistent GPS-realization of the WGS 84 reference frame, DMA with the help of the
Naval Surface Warfare Center generated a revised set of station coordinates based solely
on GPS observations. [Swift et al., 1994] Data from 24 IGS stations and the 10 DoD
stations were used to estimate the DoD station coordinates, by constraining eight of the
IGS stations to their ITRF91 coordinates during the processing. The Nuvel NNR-l plate
motion model was used to propagate station coordinates to a common epoch. The
resulting WGS 84 station coordinates were aligned with the ITRF at about the 10-cm
level. Thus, through the use of GPS and a refined GM value, both the precision and
accuracy of the WGS 84 reference frame were improved. The new realization of the
reference frame was designated WGS 84(G730), the "G" denoting GPS-derived and "730"
denoting the GPS week number (2 January 1994) when the new coordinates were placed
into operational use at DMA. The GPS Master Control Station implemented these
changes in August 1994. [MalYs and Slater, 1994]
DMA added two monitor stations to its tracking network in 1995 -- Beijing. and
Washington, D.C. During the same period, the tracking station in Australia was moved
from Smithfield to Salisbury and the GPS antenna was repositioned at the station in
England. In order to maintain a self-consistent set of network coordinates and retain the
accuracy achieved in 1994, all the DoD WGS 84 station coordinates were recomputed in
1996 using the same approach that had been employed in 1994. This time GPS data from
18 IGS stations were used, with 11 of them constrained to their ITRF94 coordinates
during the data processing. [Cunningham and Curtis, 1996] Excluding the stations that
were physically moved, the coordinate adjustments for all the other stations were 11 cm or
less in each coordinate axis. This latest realization of the reference frame is designated as
WGS 84(G873) corresponding to GPS week number 873 (29 September 1996) when the
coordinates were implemented in the National Imagery and Mapping Agency's (NIMA,

3
new name for DMA) operations. These changes were implemented by the Air Force on
29 January 1997. The accuracy of the coordinates is estimated to be better than 5 cm (1-
sigma) in each component, reflecting an extremely good alignment with ITRF94.

New Earth Gravitational Model EGM96 and Geoid

An improved Earth Gravitational Model (EGM96) and an associated global geoid were
developed through the combined efforts of the National Aeronautics and Space
Administration's Goddard Space Flight Center (NASNGSFC), NIMA and Ohio State
University. [Lemoine et al., 1996] Large amounts of new surface gravity data were
incorporated into the model including many areas of the world not represented in the
earlier model. Many new sources of satellite data were also used including GEOSAT,
TOPEXIPOSEIDON, Lageos, Lageos-2, Ajisai, Starlette, Stella, GPSMET, and GEOS-l.
EGM96 is a spherical harmonic expansion of the gravitational potential, complete to
degree and order 360. The WGS 84 EGM96 geoid was generated from the spherical
harmonic model and WGS 84 parameters and is available as a 15' x 15' grid of geoid
heights. The new geoid has an average uncertainty of±0.5-1.0 m worldwide compared to
±2-6 m for the original WGS 84 geoid. A comparison of 1,038,240 15' x 15' gridded
geoid heights from the EGM96 geoid and the original WGS 84 geoid shows a mean
difference of 0.6 m (0- = 1.6 m) and minimum and maximum differences of -13.0 m and
16.8 m, respectively.

Revised Fundamental Parameters

EGM96 has a new estimated dynamic value for the second degree zonal harmonic
coefficient. This is inconsistent with the C20 value, and thus the flattening, of the original
definition of the WGS 84 ellipsoid. Since this inconsistency has no impact on practical
applications of WGS 84, a decision was made early in the gravity modeling project to
retain the original WGS 84 ellipsoid semi-major axis and flattening values. Furthermore, a
derived geometric value for the second degree zonal coefficient is used to define the
expression for "normal" gravity for the WGS 84 ellipsoid needed for creating a global
geoid. Based on this, the four fundamental parameters for WGS 84 are now defined as
follows:
Semi-major axis of ellipsoid of revolution (a) = 6378137.0 m
Earth's gravitational constant including atmosphere (GM) = 3986004.418 x 108 m3s-2
Mean angular velocity of the Earth (0) = 7292115.0 x 10- 11 rad s-l
Reciprocal flattening (lit) = 298.257223563

COMPARISONS WITH ITRF94

Although the realization of the WGS 84 reference frame is coincident with ITRF94 at the
5-cm level, NIMA's precise WGS 84 GPS orbits and absolute WGS 84 geodetic positions
computed from a single GPS receiver occupation of a site will exhibit somewhat more
variability in practice. This can be attributed to day-to-day variability in the amount and
quality of data received from each of the 12 permanent DoD tracking stations, and to
processing and modeling differences introduced by different positioning software and data
collection protocols.

Evaluation of GPS Orbits

IGS precise orbits (ITRF94) are routinely compared with NIMA's precise WGS 84 orbits
for all GPS satellites. For each day, a 7-parameter similarity transformation is computed

4
between the two sets of orbits. The mean parameter values for the aggregate constellation
are recorded daily. In order to specifically compare WGS 84(G873) and ITRF94, these
daily means have been averaged for the first 165 days of 1997 and are shown along with
their standard deviations in table 1. Mean translations (ilX,il Y,LlZ) are less than or equal
to 0.2 cm in each component and mean rotations (RX,RY,RZ) are between 0 and 0.6 mas
in a component.

Table 1. WGS 84(G873) to ITRF94 Similarity Transformation Based on

NIMA and IGS Orbits for 165 Days in 1997

ilX ilY LlZ RX RY RZ SCALE

(cm) (cm) (cm) (mas) (mas) (mas) (ppb)

Mean 0.1 -0.2 0.1 0.0 0.4 0.6 -0.5

StdDev 2.9 2.3 l.4 0.3 0.2 0.4 0.2

Evaluation of Geodetic Positions

NIMA normally computes absolute (single receiver) point positions using software and
procedures that are different from those used to realize the WGS 84 reference frame. In
order to evaluate WGS 84 positions relative to ITRF94, GPS data from a globally
distributed set of 17 IGS stations were used to compute WGS 84 positions using NIMA's
normal production software ("GASP") and the Jet Propulsion Laboratory's GIPSY-
OASIS II software for comparison. ITRF94 coordinates and velocities were obtained
from the IGS for all of the stations. NIMA's WGS 84 precise orbits and satellite clock
corrections were held fixed during the computations and Selective Availability corruptions
were removed from the data. The station velocities were used to propagate the ITRF94
coordinates from 1993.0 to 1997.0.
Seven non-consecutive days of data between 2 December 1996 and 9 January 1997 were
obtained for each site. For the GASP solutions, each 24-hour data set was broken into
three 8-hour segments which were processed separately and then averaged to produce one
daily solution. For the GIPSY solutions, each 24-hour data set was processed as a single
session. Some difficulties were encountered by both programs in processing some of the
data sets, resulting in the elimination of some of them in the results presented here. Table
2 shows the deviations of the GASP and GIPSY WGS 84 solutions from the ITRF94
coordinates in latitude, longitude and ellipsoid height for each site. On a global basis the
average difference between WGS 84 and ITRF94 coordinates over all sites aggregated
together is very small. In general the most disparity is in the height component, with
differences up to 72 cm. The GIPSY solutions are more precise than the GASP solutions
but still show relatively large unexplained differences with ITRF94 at some sites.

FUTURE EVOLUTION

It should be apparent that a consistent effort has been made over the years to maintain
WGS 84's compatibility with the best geodetic reference systems of the time. Advances in
technology and methodology, and demand for increased accuracy and precision will
continue to drive the development. There is always a need to maintain the reference
system at a higher level of fidelity than the applications it supports. Yet, WGS 84 is
intended to be a practical system as well, and as such, the number and frequency of
changes made to the system must be kept to a minimum. Every change can affect a vast

5
Table 2. WGS 84 Positions Compared to ITRF94 Positions for Selected IGS Sites
(WGS 84 - ITRF94)

Mean GASP Solutions GIPSY Solutions

No. of M.at M.ong LlliI.Ht. No. of M.at M.ong LlliI.H1.
Site Daily Solns. (cm) (cm) (cm) Daily Solns. (cm) (cm) (cm)

Arequipa 5 8.3 11.5 33.3 5 3.4 4.9 30.4

Fairbanks 7 -2.5 -24.5 2.8 7 -5.6 -21.1 -5.7
Fortaleza 7 -4.9 1.9 45.0 7 -1.7 2.4 53.4
Hartebeesthoek 7 -6.2 -6.1 -0.5 6 -2.9 -3.2 -15.2
Irkutsk 7 -9.8 20.7 13.2 7 -1.9 9.4 -4.1
Kerguelen 7 1.6 6.7 10.2 7 3.3 5.1 16.2
Kokee Park 7 -1.8 -30.1 10.9 6 -5.5 -25.3 20.2
Maspalomas 7 3.7 4.6 -6.3 6 -1.6 5.8 45.9
F1. Davis 7 -4.5 -14.9 -71.8 6 -8.5 -24.7 -31.0
Ny Alesund 6 -1.9 -2.6 -14.9 5 -0.8 8.8 5.6
Santiago 7 14.5 -36.3 27.1 6 3.9 1.6 -12.8
S1. John's 7 -4.6 12.3 -10.2 6 5.4 8.0 -24.6
Taipei 7 -12.3 44.0 29.9 6 -5.5 3.1 3.6
Tidbinbilla 7 4.5 32.8 -20.3 6 7.9 9.6 -1.1
Tsukuba 7 -8.3 18.7 52.2 7 -5.7 3.4 9.4
Wettzell 3 -4.0 -12.8 7.4 6 -4.5 -0.8 2.8
Yellowknife 7 -0.9 -25.4 18.4 7 5.9 -18.7 -6.5

Mean (all sites) -1.7 0.0 7.4 -0.8 -1.8 5.1

Standard Deviation 6.7 22.4 28.8 4.9 12.3 22.8

number of users. At this time, refinements to the reference frame are at the decimeter
level and really do not have an impact on most mapping, navigation and routine surveying.

In the future, NIMA will add stations to its GPS tracking network, which at some point
will result in new realizations of the WGS 84 reference frame. As more surface gravity
and satellite data become available, it will become possible to refine EGM96, explore
higher degree and order gravitational models, and improve the global geoid. Tectonic
plate motion is now being accounted for in WGS 84 geodetic positioning through the use
of station velocities or a plate motion model. NIMA will continue to develop and revise
datum transformations because, despite the availability of GPS and many national efforts
to convert to geocentric datums, much of the geospatial information in the world is still
referenced to local datums. It is also an opportune time to consider the practical
implementation of a global vertical reference system since GPS provides highly accurate
ellipsoid height measurements and geoid undulations are now available globally at meter-
level accuracies. In summary, WGS 84 will continue to be maintained as a practical
geodetic reference system that provides the accuracy, precision and consistency required
by the user community.

ACKNOWLEDGMENTS
The authors wish to thank: the NIMA Geodesy and Geophysics Department staff
responsible for the daily GPS orbit processing and comparisons with the IGS orbits. We

6
especially want to thank Mr. Robert Ramsey, Mr. Todd Sherman and Mr. Terry Timblin
for carrying out all of the point positioning data processing.

REFERENCES
Cunningham, 1. and Curtis, v.L., WGS 84 Coordinate Validation and Improvement for
the NIMA and Air Force GPS Tracking Stations, NSWCDD/TR-961201, Nov. 1996,
Dahlgren, Virginia.
Defense Mapping Agency, Technical Report, Department of Defense World Geodetic
System 1984, DMA TR 8350.2, Second Ed., 1 Sept. 1991.
Lemoine, F.G., Smith, D.E., Smith, R, Kunz, L., Pavlis, E.C., Pavlis, N.K., Klosko, S.M.,
Chinn, D.S., Torrence, M.ll., Williamson, RG., Cox, C.M., Rachlin, K.E.,Wang, YM.,
Kenyon, S.c., Salman, R, Trimmer, R, Rapp, RH., and Nerem, RS., The
Development of the NASAIGSFC and NIMA Joint Geopotential Model, Proc. of the
International Symposium of Gravity, Geoid, and Marine Gravity 1996 (GraGeoMar96),
Tokyo, Japan, 30 Sep - 5 Oct 1996 (in preparation).
Malys, S. and Slater, J., Maintenance and Enhancement of the World Geodetic System
1984, Proc. ION GPS-94, Salt Lake City, Utah, Sept. 1994, pp.17-24.
McCarthy, D.D., Ed., IERS Standards (1992), IERS Technical Note 13, July 1992,
Observatoire de Paris, France.
Moritz, H., "Geodetic Reference System 1980", Bulletin Geodesique, Vol. 54, No.3,
Paris, France, 1980.
Swift, E.R, Gouldman, M.W., Merrigan, M.1. and Curtis, V.L., GPS Orbit Estimation
and Station Coordinate Improvement Using a 1992 IGS Campaign Data Set,
NSWSCDD/TR-94/267, Oct. 1994, Dahlgren, Viriginia.

7
 ITRF DENSIFICATION AND CONTINUOUS
REALIZATION BY THE IGS

G. Blewittt, C. Boucher, P. B. H. Davies, M. B. Heflin, T. A. Herring and J. Kouba

tDepartment of Geomatics, University of Newcastle upon Tyne

Newcastle upon Tyne, NE1 7RU, United Kingdom

Abstract

The tools and infrastructure are now in place to realize a global, terrestrial,
kinematic reference frame with few millimeter precision, which is both
spatially dense, and can be reliably updated on a frequent basis (e.g., monthly).
A procedure is outlined by which IERS can continuously realize the ITRF
using methodology developed by the International GPS Service (IGS).
However, spatial densification not only requires such a processing scheme, but
also requires the recruitment of interested groups to participate in this venture.

INTRODUCTION

It is appropriate that Geodesy, a discipline over 2000 years old, is approaching a

watershed as we move into the new millenium. Around 200 B.C., Eratosthenes used the
difference in direction of the sun's rays between Syene and Alexandria to infer the radius of
the Earth, producing an answer only 2% in error [Torge, 1980]. (Careful as he was, this
surely required a degree of luck). Within the last two decades, the development of space
geodesy has allowed us to measure the 12700 km distance across the Earth between any
two points with an accuracy of 1 cm, which is 0.0000001%, or one part per billion. This
requires no luck, in the sense that the results are repeatable at this level; the various
techniques (SLR, VLBI, GPS, and DORIS) also agree at this level.
The real breakthrough in this last decade of the millenium, is that positioning with I-cm
accuracy can now be routinely achieved on a daily basis, by anyone armed with a single
dual-frequency GPS receiver, appropriate software, and products from the International
GPS Service for Geodynamics (IGS). With this revolution in the making, it is right for us
to step back and reconsider our, perhaps, outdated notions of reference frames. The idea
 of control networks has evolved with space geodesy, but the basic idea has until recently
remained as Bomford [1980] describes triangulation control networks: "To constitute the
main framework on which less precise observations may be based... "
Indeed, this describes the "fiducial concept", the idea prevalent in the 1980's that precise
VLBI or SLR measurements could provide the framework within which densification could
be achieved using GPS. Although GPS has since become a truely global technique, similar
ideas are also reflected in the IGS program to densify the IERS Terrestrial Reference
Frame (ITRF). This program is again based on the principles of heirarchical networks,
starting with the IGS Global Network, which is then densified by regional networks to
produce the IGS Polyhedron (100-200 stations) [Blewitt et al., 1993a, 1995].
This IGS Polyhedron can then act as "active" control points, relative to which users can
position themselves. However, it is a misconception, perhaps too commonly held, to think
that the reasons for the Polyhedron are analogous to the triangulation network heirarchy
described by Bomford. GPS relative positioning is now essentially independent of distance,
using the right tools and techniques, so why do we need the Polyhedron?
The real issue is one of reference frame, and the fact that all stations are moving at a
level which is very significant compared to the positioning precision. We need the
Polyhedron as an active reference frame, so that we can better interpret what our positional
coordinates actually mean. As a concrete example, major earthquakes can displace these
so-called "stable reference points" on a regional scale [Blewitt et al., 1993b, Bock et al.,
1993]. In this example, we must be careful to define the time evolution of the frame such
that points well outside the area of co-seismic deformation do not appear to be displaced
due to the earthquake. This sounds so obvious, yet it is no longer obvious what to do
when unpredictable motions are not strictly regional phenomena, but extend over all scales.
This fact has been realized ever since the network design stage ofIGS:
"We do not enjoy the benefit of potential fields or elastic waves to define a
physical averaging of pointwise properties of the Earth, so we must rely on a
hierarchy of networks, with sparser regional networks serving as fidUcial sites
in the survey of denser local networks" /Minster et aI., p.24, 1993J
We can therefore view the Polyhedron as providing a primary kinematic description of
the Earth's shape as it changes. Although the "reference system" can account for
predictable changes as prescribed by conventional models, only the "reference frame" can
account for true motions. We should therefore emphasize the kinematic aspects of
reference frames if we are to keep pace with these technological advances. How do
develop and realize this is the purpose of this paper.

IGS DENSIFICATION PROGRAM

After several years of planning [Mueller and Beutler, 1992], the International GPS
Service for Geodynamics (IGS) was officially established in 1993 by the International
Association of Geodesy. Ever since an initial pilot phase beginning June 1992, the IGS has
been coordinating the operations and analysis a global network of GPS stations. The IGS
officially commenced operations in January 1994, by which time approximately 40 to 50
IGS stations had become operational.

9
 The expanding global network of high precision GPS receivers (Figure 1) was seen to
present an opportunity to produce a reference frame which is (i) dense, (ii) of a reasonably
homegeneous quality, (iii) of few-millimeter accuracy on a global scale, (iv) readily
accessible to GPS users, and (v) ideal for monitoring variations in the Earth's shape, and for
providing kinematic boundary conditions for regional and local geodetic studies [Blewitt et
al. 1993a, 1995]. The challenge was to be able to analyze cohesively the data from an ever
increasing number of receivers, such that near-optimal solutions could be produced.
Although ideally all data should be analyzed simultaneously to produce a single solution, in
practise this is computationally prohibitive
This led to the "distributed processing approach," which, at the algorithm level, partitions
the problem into manageable segments [Figure 1], and, at the organizational level,
delegates responsibility to analysis centers who would naturally have an interest in the
quality of the solutions. Another characteristic of this approach is a level of redundancy,
such that a meaningful quality assessment can be made by other, independent groups.
Distributed processing was developed as a method which could be carried out as a natural
extension to the existing operations of the IGS.

\.-------------..
IGS Orbit Analysis III® l1li

IGS Global Analysis (AC)

/' IGS Regional Analysis (RNAAC) User Analysis

•
more
• later

•
\v. .-------.-.-').----------/ ,
/' "
~.
\
Global Stations:
At least 3 estimates
.-. ~1C\:::3.;r;:.-.....--::::::£7'fNt.~ IITRF I
of each
\ pOI')edrOn
'-, );;'~~tations
"'~NAAC: Polyhedron Assembly ..____',,/

------_r... . ._------
Figure 1: Schematic explanation of the distributed processing approach. Our proposal is
for science groups to operate as RNAACs. The GNAACs would then take care of
reference frame consistency, and input into ITRF.

10
 Following a planning workshop at JPL in December 1994 [IGS, 1995], a pilot program
was initiated in September 1995 to test these ideas. Global Network Associate Analysis
Centers (GNAACs) were set up at Newcastle University, MIT, and JPL. A format was
developed for the exchange of coordinate solutions, covariance matrices, and site
information (SINEX format) [SINEX Working Group, 1996]. Initially these GNAAC's
combined solutions for global network station coordinates provided every week by the
seven Analysis Centers, producing a single unified SINEX file. Approximately one year
later, Regional Network Analysis Centers (RNAACs) began submitting regional GPS
solutions, computed using weekly published IGS orbit solutions. These regional solutions
were then assimilated into the unified global solution by the GNAACs, what is known as
the "IGS polyhedron solution."
Although currently undergoing final review, the pilot program has been viewed broadly as
a success, demonstrating few-millimeter repeatability in weekly solutions for geocentric
coordinates of not only the global stations, but also the regional stations. However the
actual process of densification (new GPS stations) is still less than adequate in many parts
of the globe. For example, tide-gauge benchmark monitoring could help. Additional GPS
stations installed at island tide-gauge sites will undoubtedly be greatly welcomed by IGS,
especially as oceanic regions of the globe are systematically undersampled (which is the
primary reason for the lack of stations in the ocean-rich southern hemisphere).
Furthermore, the IGS Densification Program provides a natural way for science groups to
participate in IGS. We expand on this point in a later section.

IGS Global Network Associate Analysis Centers: Methods and Results

Blewitt et al. [1995] discuss the following components of the GNAAC actiVItIes
(previously called "Type Two Analysis" during the planning stages): (i) detection of inter-
agency information discrepancies (e.g. in antenna heights); (ii) monitoring of solution
consistencies (inter-agency, and with respect to ITRF); (iii) weekly publication of a
combined global solution; (iv) weekly publication of an IGS polyhedron solution (global
plus regional networks); (v) periodic publication of kinematic solutions (e.g., station height
velocity, plate tectonic Euler vectors, etc.), with submission to the International Earth
Rotation Service (IERS) with the goal of improving the ITRF.
Global Network Associate Analysis Centers (GNAACs), at Newcastle University (NCL),
MIT, and JPL have so far been operating during the pilot phase of the IGS Densification
Program. All three GNAACs have in principal been using similar (but not identical)
approches and that results are quite comparable. For the purpose of this paper, methods
and results from NCL are briefly highlighted
Now two years since the inception of the IGS Densification Pilot Program, the NCL
GNAAC is continuously achieving all stated objectives [Davies and Blewitt, 1996, 1997].
Taking the most recent submission at the time of writing, coordinate solutions for 132
stations are presented, of which approximately 50% are global stations (defined as being
analyzed by at least 3 Analysis Centers), and 50% are regional. A total of 54 regional
station solutions derive from 3 RNAACs which cover South America, Europe, and Japan.

11
 We have developed combination procedures [Davies and Blewitt, 1996, 1997] which aim
to (1) minimize bias from datum assumptions, (2) minimise bias from unrealistic covariance
matrices; (3) utilize the inherent redundancy of overlapping networks to remove outliers
objectively. The first is achieved by applying a loosening transformation to each input
covariance matrix [Blewitt, 1997], which can be interpreted as the inverse of reference
frame projection [Blewitt, 1992]. The second is achieved by variance component
estimation [Rao and Kleffe, 1988; Sahin et al. 1992]. The third is achieved by applying
reliability analysis theory [Kosters and Kok 1989].
Our weekly, long-term repeatability in station height has a best case value of 3 mm,
median of 7 nun, and worst case of 19 mm. This is to be compared with the best Analysis
Center solutions (best case 4 nun, median 9 mm, worst case >30 mm). We conclude that
GNAAC analysis not only provides a consistent unique solution, but also a more reliable
solution (in the statistical sense of the word). The IGS Densification Program
methodology should not be viewed as compromising solution quality, but rather as a
preferred alternative to unilateral analysis.
In summary, GNAAC results show that the combined solutions produce coordinate time
series with a smaller variance than any individual AC network solution, even more so for
stations with the worst levels of precision. This indicates that reliable outlier detection, due
to redundancy from the multiple solutions, is a major reason for this improvement.
Therefore, we can conclude that GNAAC solutions are not only more precise, but more
reliable than any individual contributing AC solution.

IERS STRATEGY FOR ITRF

Since its establishment in 1988, the International Earth Rotation Service (IERS)
published on an annual basis a new realization of the International Terrestrial Reference
System (ITRS). Each year, the IERS Central Bureau collected among contributing analysis
centers their solutions for Earth Rotation Parameters together with the associated
Terrestrial Reference Frame (TRF) (and Celestial if available) data. These contributions
are currently available for several space techniques: SLR, LLR, VLBI, GPS and DORIS.
Such a realization is now widely known under the label ITRFyy. Up to yy=94, this
solution was obtained by a combination of all data submitted to the IERS Central Bureau at
the begining ofyy+ 1. It was assumed that all individual analysis centers would provide by
this was their best and most complete individual solution, in particular including data for
the yearyy.
The succession of these results, from ITRF88 to ITRF94, did actually show an
improvement both in accuracy for positions of stations as well as in the geographical
coverage of the corresponding network. Nevertheless, the accuracy reached was ranging
from a few millimeters to a few centimeters, and a lot of details concerning modeling were
raised at this level (in particular relativistic and geodynamical effects).
At that time, IERS decided to establish a primary solution upon specifications of an
international working group (WG on ITRF datum). Although this is exactly a task for this
WG to recommend the precise definition of what should be a primary ITRF solution, we

12
 can agree that the main concept is to provide a consistent, homogeneous set of positions
(and time variations) at the subcentimetric level (in precision and reliability) for a well
distributed network. This work is presently in progress, in particular by performing
research (experimental and theoretical) on the geocenter, which is one of the new topics to
be investigated before establishing these general recommendations.
Considering on the other hand the numerous and increasing requirements of the users,
mostly outside the restricted IERS community, IERS decided to continue the annual series
ofITRF publications for a complete solution along the line of the series up to ITRF94.
Starting for .Y.)F96, ITRFy'y is now specified to be the best complete solution produced on
an annual basis by IERS. In particular, any data set expected to bring a useful contribution
to this goal should be included, and not only the data from the annual submission to the
Central Bureau. Specific work is currently underway for GPS and DORIS contributions,
with the possibility of producing updated realizations of ITRFyy, perhaps every month.
Both techniques are growing significantly and providing each year new stations and
improved present day position quality.

Cooperation with IGS

Close cooperation with IGS is very useful and already under expansion, with mutual
benefits:
• since the begining , the IGS organization stimulated individual analysis centers to submit
solutions in response to the IERS Central Bureau's annual call for data.
• IGS uses for operational purposes ITRFy'y solutions. IGS represents a strong user group
in favor of these annual solutions. ITRF96 will therefore be very welcome.
• conversely, the GPS contribution for ITRF96 is very strong, including results from IGS
densification pilot experiment.
• using these ITRF solutions (or maybe in the future a cleaner primary solution which is
under consideration by a ad hoc working group on ITRF datum), IGS could consider by
its ITRF densification strategy to provide an operational TRF solution similar to rapid
service for EOP.
We propose that routine monthly GNAAC solutions be submitted to the IERS as input to
a continuous realization of the ITRF. Not only would this have benefits from the standpoint
of reliability, but also non-linear station motion would also be represented by such an
approach (e.g., due to possible monument instability, the earthquake cycle, or seasonal
loading effects). This approach would also be particulary useful for recently installed
stations whose velocities are not yet very well determined.
As an IERS user group, IGS could then use the continuously updated realization ofITRF
for its products. Of course, users such as IGS may not wish to update its reference frame
every month, however the opportunity will always be there to perform a frame update
when needed. For example, IGS recently experienced a slight degradation in the network
of available primary stations used to define the frame of the orbits and Earth rotation
parameters. Although IGS considered producing its own realization of ITRF to improve

13
the situation, it was decided to wait for the imminent ITRF96. Clearly, a set of sub-annual
reference frame realizations would have been useful under these circumstances.

ORGANIZATIONAL STRUCTURE FOR ITRF DENSIFICATION

We can broaden the definition of "ITRF Densification" to mean temporal as well as

spatial densification. For temporal densification, we have already described a scheme
whereby IGS can contribute to, say, monthly realizations of the ITRF. Of course, other
space geodetic techniques (e.g., DORIS).could also be coordinated towards this goal.
For spatial densification, we have described a scheme designed on the principle of
distributed processing, which easily allows the incorporation of data from many permanent
GPS networks. We haven't explored how RNAAC's can be "recruited" to ensure a good
distribution of stations to suit a variety of scientific purposes. Since IGS is a service
organization, and not primarily in the business of scientific investigation, we should
promote the idea that RNAAC's be connected to some science group, which has its own
objectives and agenda. This should ensure that scientific needs are met, while decentralizing
the operational burden away from other IGS components.
An organizational structure for this is presented in Figure 2, which reproduces a diagram
presented at a recent workshop for monitoring sea level held at JPL in March 1997 [Blewitt
et ai., 1997]. Simple lines connecting the boxes indicate the organizational hierarchy.
Arrows indicate data flow.

Scientific Organization?

Figure 2: Chart illustrating organizational links and data flow to facilitate the activity of
tide-gauge benchmark monitoring (explained in text).

Figure 3 expands this idea specifically for the community interested in monitoring tide
gauge benchmarks. We use this here as an example of how Figure 2 might be practically
realized. Starting with the bottom right hand side, we have the goal of this organization,

14
 which is the production of a database (DB) of the coordinates and velocities of tide gauge
benchmarks available at the Permanent Service for Mean Sea Level (PSMSL), which
formally reports to the Commission of Mean Sea Level and Tides (CMSLT), under the
umbrella of the International Association for the Physical Sciences of the Oceans (lAPSO).
The structure described so far is essentially in place. What remains to be done, is to
include GPS data from tide gauge sites into the dataflow, which would be analyzed by new
RNAAC's. Figure 3 shows each RNAAC as a part of a science group which falls under the
International Association of Geodesy (lAG) through the Special Commission 8 on Sea
Level and Ice Sheet Variations (SC8). Special Commission 8's terms of reference look as
if they have been written especially for this task, since they not only mention geodetic
observing programs to investigate sea level change, but also interdisciplinary
communication betwee geodesists, geophysicists, and oceanographers. Science groups are
also connected to the CMSLT to make the collaboration with oceanographers explicit, and
for the practical necessity for expertise on tide gauge selection. It would be natural for
science groups to be regional, given that they act as RNAACs. To complete the loop, the
Science Groups access both the tide gauge records and the geodetic records from the
PSMSL for scientific interpretation.

IAPSO

ITRF

Figure 3: Chart illustrating organizational links and data flow to facilitate

the activity of tide-gauge benchmark monitoring.

CONCLUSIONS

A procedure has been outlined by which IERS can continuously realize the ITRF, using
IGS methodology which has been tested over the past 2 years during its ITRF Densification
Pilot Project. We propose that this project move out of the pilot phase, and be made
official. This will involve the resolution of some technical details and coordination between

15
IGS and IERS, however much of those details have been already resolved with the
development of the SINEX format for the exchange of geodetic solutions.
We have also pointed out that spatial densification not only requires a processing scheme,
but also requires the recruitment of interested groups to operate the networks and
participate in the scheme. We have identified scientific groups as likely candidates for
operating IGS Regional Network Associate Analysis Centers. We show, for example, that
the IGS, IERS, and the tide gauge benchmark community can be served by RNAAC's
serving as part of the lAG Special Commission 8 "Sea Level and Ice Sheets", in
collaboration with the IAPSO Commission for Mean Sea Level and Tides.
In conclusion, although there is certainly room for improvement in geodetic techniques
and reference frame definition, ultimately the user relies on products which may not
necessarily reflect the best currently achievable accuracy. What we have presented here is
one way forward to improve the ITRF as a useful, accurate, and reliable product.

REFERENCES

Blewitt, G, "GPS Data Processing Methodology: From Theory to Applications," in GPS

Geodesy, Eds. A. Kleusberg and P.IG. Teunissen, Lecture Notes on Earth Sciences,
Springer (in press, 1997)
Blewitt, G., P.B.H. Davies, T.Gregorius, RS. Kawar, and D.U. Sanli, Sustainable geodetic
monitoring of the natural environment using the IGS, in Methods for Monitoring Sea
Level, GPS and Tide Gauge Benchmark Monitoring, GPS Altimeter Calibration, Jet
Propulsion Laboratory, Pasadena, Calif. USA (in press 1997)
Blewitt, G., Y Bock, and I Kouba, "Constructing the IGS polyhedron by distributed
processing," in Proc. of the IGS Workshop, ed. by I Zumberge, IGS Central Bureau,
Pasadena, Calif., USA, p. 21-36 (1995)
Blewitt, G., Y Bock, and G. Gendt, "Regional clusters and distributed processing," in
Proc. of the IGS Analysis Center Workshop, Ed. by I Kouba, pp. 62-91, Ottawa,
Canada (1993a).
Blewitt, G., M.B. Heflin, KJ. Hurst, D.C. Jefferson, F.H. Webb, and J.F. Zumberge,
"Absolute far-field displacements from the 28 June 1992 Landers earthquake sequence,"
Nature, Vol. 361, p. 340-342 (1993b)
Blewitt, G., M.B. Heflin, F.H. Webb, U.J. Lindqwister, and R P. Malla, "Global
coordinates with centimeter accuracy in the International Terrestrial Reference Frame
using the Global Positioning System," Geophysical Research Letters, 19, pp. 853-856
(1992).
Bock. Y, Agnew, D.C., Fang, P., IF. Genrich, B.H. Hager, T.A. Herring, KW. Hudnut,
RW. King, S. Larsen, J-B. Minster, K Stark, S. Wdowinski and Wyatt, F.K,
"Detection of crustal deformation from the Landers earthquake sequence usmg
continuous geodetic measurements," Nature, 361, 337-340 (1993)
Bomford, G., Geodesy, Clarendon Press, Oxford (1980)

16
 Davies, P.B.H. and G. Blewitt, "Newcastle upon Tyne Global Network Associate Analysis
Centre Annual Report 1996," IGS Annual Report 1996, p. 237-252, IGS Central
Bureau, Jet Propulsion Lab., Pasadena, California (1997)
Davies P.B.H. and G. Blewitt, "Newcastle upon Tyne IGS Global Network Associate
Analysis Centre Annual Report 1995," IGS Annual Report 1995, p. 189-202, IGS
Central Bureau, Jet Propulsion Lab., Pasadena, California (1996)
IGS, Densijication of the IERS Terrestrial Reference Frame through Regional GPS
Networks, Eds. 1.F. Zumberge and R Liu, IGS Central Bureau, Jet Propulsion
Laboratory, Pasadena, California (1995)
Kosters, A.J.M. and 1.1. Kok, "Statistical testing and quality analyusis of aobservations and
transformation parameters in combining 3-dimensional networks," Geodetic
Computing Centre (LGR), Faculty of Geodesy, Delft Univ. of Tech., presented at
Symp. S102, lAG Congress, 3-12, Aug. 1989, Edinburgh (1989)
Minster 1.B., M. Bevis, Y. Bock, C. Boucher, o. Columbo, B. Engen, AM. Finkelstein, H.
Frey, B. Hager, T. Kato, S. Lichten, P. Morgan, W. Prescott, C. Reigber, S. Rekkedal,
B. Schutz, H. Tsuji, and V. Velikhov, "Network design considerations for the
International GPS Geodynamics Service," p.23-32, in lAG Symposia 109: Permanent
Satellite Tracking Networks for Geodesy and Geodynamics, Ed. G. L. Mader,
Springer Verlag, Berlin (1993)
Mueller, 1. and G. Beutler, "The International GPS Service for Geodynamics -
Development and current structure," in Proc. of the 6th Int. Geodetic Symp. on
Satellite Positioning, Columbus, Ohio, pp. 823-835, 1992
SINEX Working Group, "SINEX - Solution Independent EXchange format," Appendix 1
in IGS 1996 AnalYSis Centre Workshop Proceedings, Eds. RE. Neilan, P.A Van
Scoy, and 1.F. Zumberge, pp. 223-276, IGS Central Bureau, Jet Propulsion
Laboratory, Pasadena, California (1996)
Rao C.R and 1. Kleffe, Estimation of Variance Components and Applications, Elsevier,
Amsterdam (1988)
Sahin, M., P.A Cross, and P.C. Sellers, "Variance component estimation applied to
satellite laser ranging," Bulletin Geodesique, 66, 284 (1992)
Torge, W., Geodesy, Walter de Gruyter & Co., Berlin (1980)

17
 THE USE OF THE EUREF PERMANENT GPS NETWORK FOR
THE MAINTENANCE OF THE EUROPEAN TERRESTRIAL
REFERENCE FRAME

c. Bruyninx, Royal Observatory of Belgium, Av. Circulaire 3, 1180 Brussels, Bel-

gzum
J. Dousa, Geodetic Observatory Pecny, Ondrejov 2.14, 25165, Czech Republic
W. Ehrnsperger, Bay. Akademy fur die Intern. Erdmessung, Marstallpl. 8, 80539
Munchen, Germany
N. Fachbach, Austrian Academy of Sciences, Lustbuhelstr. 46, 8042 Graz, Austria
J. Johansson, Onsala Space Observatory, S-43992 Onsala, Sweden
F. Vespe, L. Ferraro, A. Nardi, CGS Centro di Geodesia Spaziale "G. Colombo",
Contrada Terlecchia, 751 00 Matera, Italy
M. Figurski, Institute of Geodesy eJ Geodetic Astronomy WUT, Pl. Politechniki 1,
00-661 Warsaw, Poland
T. Springer, Astr. Inst., University of Berne, Sidlerstr. 5, 3012 Bern, Switzerland
G. Weber, Inst. for Applied Geodesy, R. Strauss Allee 11, 60598 Frankfurt, Germany
A. Wiget, Federal Office of Topography, Seftigenstr. 264, 3084 Wabern, Switzerland

Abstract
Taking into account the growing number of permanent GPS stations in Europe,
the EUREF subcommission decided in 1995 to coordinate the activities related
to this network for the maintenance of the European Reference Frame.
The EUREF permanent GPS network, which is the European densification of the
IGS network, presently consists of more than 60 GPS stations, covering 23 coun-
tries all over Europe. The processing scheme allows for distributed processing:
10 Local Analysis Centers each analyze a part of the EUREF network. One ana-
lysis center is responsible for merging the individual subnetwork solutions into one
European solution. This solution in submitted to the International GPS Service
for Geodynamics (IGS) within the frame of the IGS Densification Project.
Weekly free-network solutions for Europe are available since April 1996. The
mean RMS values of the coordinate residuals of the combined solution with re-
spect to the individual solutions is about 2 mm for the north and east components
and 6 mm for the height component. The agreement between the solutions of the
different analysis centers is of the same quality as the week-to-week repeatabilities
of each analysis center.
In may 1997, EUREF submitted its solution to the International Earth Rotation
Service. As a result, all EUREF stations will show up in the next realization of
the International Terrestrial Reference System.
Introduction

The main task of the EUREF subcommission is the establishment, maintenance and
enhancement of a 3-dimensional European Reference Frame.
In order to have station-to-station relations fixed in the European Terrestrial Ref-
erence System (ETRS), this ETRS has been defined as fixed to the stable part of
the European plate and coinciding with the International Terrestrial Reference System
(ITRS) at epoch 1989,0 (Resolution No 1 ofthe EUREF Symposium in Florence, 1990).

A first step towards the realization of the ETRS was taken in 1989 with the obser-
vation of the EUREF89 GPS campaign which covered Western Europe (Seeger et al.,
1992). The processing of the EUREF campaigns is done following the rules set up by
the EUREF Technical Working Group. In 1993, these standards recommended the use
of the IGS sites, which have known ITRF coordinates, as fiducials for the processing of
EUREF campaigns. A year later, at the EUREF symposium in Warsaw, the use of the
European permanent stations for the maintenance of the European Reference Frame
was recognized (Gurtner, 1994) and European agencies were invited to collaborate with
EUREF for the implementation of such a network.
In May of 1995 guidelines for a EUREF permanent GPS Network were set up (Gurtner,
1995) and in October of the same year a EUREF network coordinator (C. Bruyninx,
Royal Observatory of Belgium) was designated at the EUREF Technical Working Group
meeting in Paris in order to coordinate the activities related to the European perman-
ent network.
In March 1996, EUREF responded to the IGS "CALL FOR DENSIFICATION OF THE
ITRF THROUGH REGIONAL GPS ANALYSES AS IGS NETWORK ASSOCIATE ANA-
LYSIS CENTER" and proposed to submit one official European solution to the IGS
based on the principle of distributed processing.
The EUREF proposal was officially accepted by the IGS in May 96 and the first solu-
tions for the EUREF permanent network were forwarded to the IGS.

Current Status of the EUREF Permanent Network

The CPS Tracking Network

The EUREF network consists of all the European permanent sites included in the IGS
Network, plus additional sites accepted by the EUREF Technical Working Group fol-
lowing similar rules as set up by and for the IGS.
The network consists presently of more than 60 stations, covering 23 countries. 40%
of the EUREF stations do not belong to the IGS network.

The data from most of the EUREF stations are available at the IGS Regional Data
Center at HAG (Institute for Applied Geodesy). The data of the remaining stations
can be retrieved from five local data centers which give access to a particular EUREF
subnetwork (Bruyninx, 1997a).

19
 -20 -15 -10 -5 0 5 10 15 20 25 30 35

80 80

75 75

....-
70

'L:s:
-
65 65

• ~
60
... 1 ~ 60

~
Mendele..,o

55
••
ZWenlgoro 55

{] 0 KOOIWIJk
e tmon
• Brussel _ Jozefoslaw
50 e rgem ~ -Waromme PeOlY Wrodaw 50
Wettzell •
DOUrb~be",fallenhofen. Modra-Plesok
Pfander. : • •
• Haf~ekar - Pane
Zimmerwald Graz
45 Padov. Va 45
Id

40 40

35 35

30 30

-20 -15 -10 -5 0 5 10 15 20 25 30 35

Figure 1: Stations included in the permanent EUREF GPS network (August '97). The
EUREF stations Kellyville and Thule (Greenland) are not included in the map.

Data Analysis
When EUREF took, at the end of 1995, the initiative to coordinate the activities re-
lated to the European permanent network, it was agreed that the data analysis of the
EUREF network would be divided among different European institutions. The main
goal of this distributed processing was the dissemination of the knowledge and expert-
ise to routinely process a network of GPS stations for high precision geodesy.
The CODE analysis center agreed to be responsible for merging the individual subnet-
work solutions into one European solution.

At present ten analysis centers are involved in the processing of the EUREF network:
the Italian Space Agency (Italy), the Bavarian Academy of Sciences (Germany), the
Observatory Lustbiihel (Austria), the Institute for Applied Geodesy (Germany), the
Federal Office of Topography (Switzerland), the Geodetic Observatory of Pecny ( Czech
Republic), The Nordic Geodetic Commission Analysis Center (Sweden), the Royal Ob-

20
servatory of Belgium (Belgium), the Warsaw University of Technology (Poland) and
the Center for Orbit Determination in Europe (Switzerland).
Taking into account that no specific data analysis recommendations were available to
the EUREF analysis centers until recently, the processing strategy used at most of the
analysis centers was very similar, although some important differences (for example
the elevation cut-off angle) remained.

Since mid 1996 weekly SINEX contributions of the different European subnetworks
are combined into one official weekly EUREF solution. This EUREF solution is then
submitted to the IGS where the Global Network Associate Analysis Centers combine
all global IGS solutions with several regional solutions like EUREF in the frame of the
IGS densification pilot project (Zumberge and Liu, 1994).

The EUREF Combined Solution

The Combination Scheme

The results from the EUREF analysis centers are available in the SINEX format and
are combined using the Bernese GPS Software program ADDNEQ (Brockmann, 1996)
which will generate the weekly EUREF combinations. ADDNEQ is based on the se-
quentialleast-squares adjustment. It removes first all constraints from the individual
solutions and then applies the covariance rescaling, based on the data sampling used
at each individual analysis center.
The combination is statistically correct (equivalent to a common least-squares ad-
justment using all original GPS observations in one step) assuming that there are no
correlations between the observations of each of the sequential solutions. The procedure
is not absolutely correct from the statistical point of view because the observations of
the anchor sites (sites common to different solutions) are introduced into the combined
solutions at least twice.
Two different combinations are made subsequently : one using a free network solution
and one using fiducial sites. In both solutions the same stations are used to define the
reference frame. In the free network solution the geodetic datum is defined by specify-
ing six no-net constraints (translation and rotations) for the selected reference sites.
The free solution is performed since it helps to identify problems with the combination.
The fiducial solution gives the official EUREF weekly solution which is available at the
anonymous ftp servers at AIUB (EUREF Combination Center, Switzerland), CDDIS
(IGS global data center), HAG (IGS regional data center), and ROB (EUREF Central
Bureau, Belgium).

Quality and Optimization of the Data Analysis

The consistency of the solutions from the different EUREF analysis centers is well be-
low the 10 mm level in the horizontal and around 10 mm for the vertical component.

21
 As can be seen in Figure 2, the repeatability of the coordinates from the combined
EUREF solution is very good. Only a few stations show slightly higher rms values.
The most common causes are site specific errors (like the use of old Rogue receivers
in Madrid and Wettzell) or problems with fixing the reference frame, especially for
stations at the edge of the network.

20

_15
E
E
U) 10
::!:
II: 5

20 81
35
_15
E
E
-10
~
II: 5

20

_15
E 73 49
E
U) 10 36 36 36 76
::!:
II: 5

Figure 2: Root mean square errors of the EUREF station coordinates obtained from
the weekly EUREF solutions covering the period from January 1996 to August 1997.
The numbers written above the north-east-up rms values are the number of weeks that
the station was included in the solution.

The data analysis within the EUREF network is presently being reviewed for more
optimization:

• All analysis centers follow, since recently, data analysis recommendations which
were set up at an "EUREF Analysis Workshop" organized in Brussels on 10-11
may '97 (Bruyninx, 1997);

• The observations from the stations which are processed by several analysis cen-
ters are introduced multiple times into the combined solution. The resulting
correlations are not taken into account in the combination procedure and they
distort the EUREF covariance matrix.
At the same time, the quality of weekly solution for stations which are processed

22
 by only one analysis center is rather difficult to assess.
Therefore a redistribution of the individual subnetworks has been agreed upon
between the network coordinator and the analysis centers. This redistribution
aims on one hand to equalize the number of centers analyzing each station and
on the other hand to have at least two analysis centers for each station.

• One of the main problems for the correct interpretation of the coordinate time
series are undocumented changes of the antennas or their environment at the
EUREF stations, e.g. the use of radomes. Examples are shown in Figures 3 and
4.
EUREF therefore encourages a closer contact between the analysis centers and
the station managers through the EUREF Central Bureau. This communication
runs two ways :

1. the station managers who try to document all detected site changes in the
station log files ;
2. a feedback from the analysis center to the station managers (or operational
centers) when anomalies are detected during the data analysis.
KJRU 10(03)1002

• NORTH - COIolPONENT

~
I

650 655 660 665 670 675 660 665 690 695 900
GPS WEEK

Figure 3: Coordinate time series for Kiruna (from EUREF combined solution). The
behavior of the height component can be explained by the fact that there was snow on
the antenna in November '96 (week 878-881) and January '97 (week 886-890).

1L0PI 115071L00l

- o

l:~~~~~~~~~~~~~~~
~-
I
o
C\l
I

BB5 690 B95 900 905 910

GPS WEEK

Figure 4: Coordinate time series for Modra Piesok (from EUREF combined solution).
A jump of around -2 to -3 cm is detected in the height component at GPS week 891
due to the installation of a radome on the antenna on February 3, 1997. The radome
was taken off again on June 2, 1997, before the start of the CERGOP campaign.

23
EUREF Contribution to the IERS

Early 1997 the International Earth Rotation Service (IERS) decided to produce a new
realization of the ITRS, called International Terrestrial Reference Frame (ITRF). It
was thought that due to the IGS densification project, the EUREF stations would also
show up in the new ITRF results. It turned out, however, that at the moment, due
to the "pilot" status of the IGS densification project, there is no guarantee that all
the permanent EUREF stations would show up in the new ITRF realization. Therefore
EUREF proposed to the IERS to include a separate EUREF contribution in the new
ITRF.
In May of 1997, CODE submitted the combined EUREF solution to the IERS (Springer
et aI, 1997). As a result all EUREF permanent stations will be included in the next
ITRS realization.

The EUREF Permanent Network Central Bureau

The EUREF permanent network coordination is performed at the Royal Observatory

of Belgium: a fully documented information system (including descriptions of the per-
manent GPS stations, Local Data Centers, Local Analysis Centers and their subnet-
works) is operational and maintained (ftp to ftpserver.oma. be or 193.190.249.203, cd
pub/astro/euref or the Web site: http://www.oma.be//KSB-ORB/EUREF/eurefhome .html).

The consistency between the headers of the RINEX data files and the station de-
scription files in the database are weekly checked and if necessary station responsibles
are contacted. The product availability is monitored and reports on the combined
EUREF solution are made available weekly as feedback to the contributing analysis
centers.

Conclusions

The success of the EUREF network has been demonstrated by the fact that since the
start of the activities, at the end of 1995, the number of stations, data centers and
analysis centers has more than doubled.
Weekly free-network solutions for Europe are available since April 1996. In the begin-
ning of 1997, data analysis guidelines have been set up for the EUREF analysis centers
and they are presently followed by all the analysis centers, The distribution of the
EUREF subnetworks has been revised recently and will be implemented in the autumn
of 1997.
The distributed processing is a success : the consistency between the solutions of the
different analysis centers is of the same quality as the week-to-week repeatabilities of
each analysis center.

24
 Recognizing the success of the EUREF permanent network, the EUREF Technical
Working Group will make, within the next months, the necessary calls for participation
to take the activities related to the EUREF network from an experimental stage to an
operational service.

Acknowledgments

The authors would like to express their gratitude towards the responsible agencies and
representatives at the observation sites, the data centers and analysis centers. Without
their labor and commitment this paper would never have been possible.
We would also like to recognize the work of Werner Gurtner who initiated the activit-
ies related to the EUREF permanent network. The members of the EUREF Technical
Working Group provided support and technical advice.

References

Brockmann, E. (1996), "Combination of Solutions for Geodetic and Geodynamic Ap-

plications of the Global Positioning System (GPS)", Ph.D. dissertation, Astro-
nomical Institute, University of Berne, Berne, Switzerland

Bruyninx C., W. Gurtner and A. Muls (1996), "The EUREF Permanent GPS network",
Report on the Symposium of the lAG Subcommission for Europe (EUREF) held
in Ankara 22-25 May 1996, Veroffentlichungen der Bayerischen Kommission fiir
die Internationale Erdmessung, Vol 57

Bruyninx C. (1997), "Minutes of EUREF Analysis Workshop", Brussels, April 10-11,

1997
(http://www.oma.be/ /KSB-ORB/EUREF /papers/elacwOOl/elacwOOl.html)

Bruyninx C. (1997a), "The EUREF Permanent GPS Network: Activities May '96
- May '97 and Future Plans", Report on the Symposium of the lAG Subcom-
mission for the European Reference Frame (EUREF) held in Sofia 3-7 1997,
Veroffentlichungen der Bayerischen Kommission fiir die Internationale Erdmes-
sung, (in press)

Gurtner W. (1994), "The Use of European Permanent GPS Stations for the Main-
tenance of the European Reference Frame", Report on the Symposium of the
lAG Subcommission for Europe (EUREF) held in Warsaw 8 - 11 June 1994,
Veroffentlichungen der Bayerischen Kommission fiir die Internationale Erdmes-
sung, Vol 54
Gurtner W. (1995), "Guidelines for a Permanent EUREF GPS Network", Report on
the Symposium of the lAG Subcommission for the European Reference Frame
(EUREF) held in Helsinki 3-6 May 1995, Veroffentlichungen der Bayerischen
Kommission fiir die Internationale Erdmessung, Vol 56

25
Rothacher, M. , G. Beutler, E. Brockmann, L. Mervart, S. Schaer, T.A. Springer, U.
Wild, A. Wiget, H. Seeger and C. Boucher (1996), "Annual Report 1995 of the
CODE Analysis Center of the lGS", in lGS 1995 Annual Report, edited by J.F.
Zumberge et.al., pp.151-173, lGS Central Bureau, Jet Propulsion Laboratory,
Pasadena, California, U.S.A.

Seeger, H., W. Augath, R. Bordley, C. Boucher, B. Engen, W. Gurtner, W. Schluter,

R. Sigl (1992), "Status-report on the EUREF-GPS-Campaign 1989 to the lAG
EUREF Subcommission", Veroffentlichungen der Bayerischen Kommission fur
die lnternationale Erdmessung, Vol 52

Springer, T. , W. Gurtner, M. Rothacher, S. Schaer (1997), "EUREF Activities at

the CODE Analysis Center", Report on the Symposium of the lAG Subcom-
mission for the European Reference Frame (EUREF) held in Sofia 3-7 1997,
Veroffentlichungen der Bayerischen Kommission fur die lnternationale Erdmes-
sung, (in press)

Zumberge, J.F. and R. Liu (eds) (1994), "Densification of the lERS Terrestrial Refe-
rence Frame through regional GPS Networks", Workshop Proceedings, lGS Cent-
ral Bureau, Jet propulsion Laboratory, Pasadena, California, U.S.A.

26
 THE EUROPEAN VERTICAL GPS REFERENCE NETWORK
CAMPAIGN 1997 - CONCEPT AND STATUS

J. Thdel), J. Adam2), W. Gurtne~), B. G. Harsson4), W. Schliite.-5), G. Woeppelmann6)

l)Bundesamt fUr Kartographie und Geodasie·, Au8enstelle Leipzig
Karl-Rothe-Stra8e 10-14, D-04105 Leipzig, Germany
l)Technical University of Budapest, Department of Geodesy
Mueegyetem rkp. 3.1.61, HU-1ll1 Budapest, Hungary
3)Astronomisches Institut der Universitiit Bern
Sidlerstra8e 5, CH-3012 Bern, Switzerland
4)Statens Kartverk, N-3500 Honefoss, Norway
5)Bundesamt fUr Kartographie und Geodasie*, Fundamentalstation Wettzell
Sackenrieder Stra8e 25, D-93444 Kotzting, Germany
6)lnstitut Geographique National
2. Avenue Pasteur B.P.68, F -94160 Saint-Mande, France

Abstract

The European Vertical GPS Reference Network (EUVN) is designed to contribute to the
unification of different height systems in Europe. The most important practical and scientific
aspects are

contribution to a unique European height datum

connection of European tide gauge benchmarks as a contribution to monitoring abso-
lute sea level variations
establishment of fiducial points for the European geoid determination
preparation of a European Vertical Kinematic Network.

A network of about 195 points distributed over Europe and consisting of 79 EUREF-points,
of 53 nodal points of the Levelling Networks of Eastern and Western Europe and 63 tide
gauges has been observed in the period of May 21 to 29, 1997 with GPS in order to derive
uniform ellipsoidal heights in the frame of ETRS89.

*fonnerly Institut filr Al1gewandte Geodasie (IfAG)

1. General

The existence and the establishment of geodetic networks such as

EUREF (European Reference Frame)

UELN (United European Levelling Network)
UPLN (United Precise Levelling Network)

lead to the proposal to combine these networks in the height component. Through a GPS
campaign such a combination will provide a ,,European Vertical GPS Reference Network"
(EUVN) for scientific and for practical use.

The realization of the EUVN is based on the Resolution No 3 of the EUREF Symposium in
Ankara, 22-25 May 1996.

The authors of the paper are members of the EUVN Working Group, which is instructed by
the lAG Subcommission EUREF to prepare and carry out the EUVN project.

2. Objectives

The objectives of the EUVN are

(1) to provide an integrated vertical reference frame for EUREFIETRS height values at a
few centimeter level:

For all EUVN points P three-dimensional coordinates in the ETRS89 (Xp, Yp' ZpJEl'RS
and geopotential numbers cp will be derived. The geopotential number cp = WO UEIN -
Wp is the difference between the potential of the earth gravity field in the level through
the reference tide gauge of the UELN (WoUEIN), which best approximates the geoid, and
the gravity potential in the EUVN points (Wp).

Finally in the same way the EUVN is a geometricaVphysical reference frame. In addi-
tion to the geopotential num_bers cp normal heights hn = c/Y and orthometric heights
ho =c/g will be provided (y is the mean normal gravity between the ellipsoid and the
telluroid, gis the mean gravity between the geoid and the earth surface).

(2) to connect different European height datums:

In addition to the UELN 73 for West and North Europe and the UPLN 82 for Central
and Eastern European countries national height systems exist with different kinds of
heights and different zero levels. This project will contribute to a unification of Euro-
pean height systems in the framework of UELN.

28
(3) to provide fiducial points for the European geoid determination and for future accurate
regional geoid computations:

The application of the GPS technique for practical height determination will dramati-
cally be extended if the geoid would be known precisely enough. In order to derive the
geoid a European reference geoid is required in the reference System ETRS89 of
EUREF and the reference System of UELN. So far there is no precise geoid available
for Europe with an accuracy of a few centimeters which fulfils the requirement for the
practical applications.

This proposal points out a possibility to evaluate a geoid tailored for the GPS-leveUing
methods by combining the existing reference network EUREF with the UELN95.

(4) to contribute to the realization of an European vertical datum and to connect different
sea levels of European oceans in view of work by PSMSL (Permanent Service Mean
Sea Level), also in view of anticipated accelerated sea level rise due to global warming.

At present the zero level for UELN is the tide gauge Amsterdam and for the UPLN is
the tide gauge Kronstadt. The level difference is about hAmsterdam - h Kronstadt = 0,15m.
Independent of an uniform height level for the maritime states the knowledge of the
level and, under special conditions, of the level changes of the adjacent oceans is vitally
important.

(5) to provide contribution to the determination of an absolute world height system (Balas-
ubramania, 1994).

(6) to establish a fundamental network for a further geokinematic height reference system
such as UELN 2000 under the special consideration of the Fennoscandian uplift and the
uplift in the Carpathian-Balkan region.

(7) to provide data for decoupling the land and sea level components of relative sea level
variations, as measured by tide gauges:

As tide gauges provide measurements of sea level relative to a TIde Gauge Benchmark
(TGBM), they give access to a local information which generally results from the com-
bination of the sought-after rise in sea level and the vertical movements of the land at
the tide gauge site. Therefore, global sea level studies based on tide gauge data require
to monitor the vertical crustal velocities at the tide gauge sites with respect to a geocen-
tric reference frame, in order to recover a global geocentric assessment of the sea level
variations. In this scope the use of GPS has been recommended by an international
group of experts at two occasions (Carter et al., 1989) and (Carter, 1994).

(8) to provide the basis of express the results of the regional European tide gauge GPS sur-
veys in the EUREF reference system (ETRS89):

29
 Following the Carter et al. (1989) recommendations several institutions have carried
out regional tide gauge GPS surveys around Europe. At least three common points
between two networks are needed for a rigorous coordinates combination process and
for the subsequent expression of the regional campaign results in a common geocentric
reference frame (Boucher et aI., 1994). The design of the EUVN network and its tide
gauge component has taken into account this constraint.

3. Design of EUVN

EUVN connects several kinds of heights and will integrate the vertical reference of the
European Reference Systems EUREF, UELN (and national height networks), tide gauge
sites, as well as the European Geoid.

The EUVN is designed to incorporate the already realized parts of EUREF and UELN as
well as the planned European permanent GPS station network.

European Reference Networks for positioning and for heights

EUREF realizes the ETRS89 frame for precise applications of GPS techniques for posi-
tioning in Europe with an accuracy about 1 cm. The height components refer to the refer-
ence ellipsoid GRS80.

UELN is the "United European Levelling Network" which provides physical heights with
respect to the tide gauge of Amsterdam. According to Resolution No.3 of the EUREF
Symposium 1994 in Warsaw, it is the objective to create an uniform height datum for
Europe on the basis of UELN73 (Ehrensberger and Kok 1986) which includes the enlarge-
ment also to the Central and Eastern European countries.

In the Central and Eastern European countries there already exists an United Precise Level-
ling Network (UPLN) consisting of first order levelling lines through Bulgaria, East Ger-
many, Czech Republic, Slovakia, Poland, Romania, Russia, Georgia, Estonia, Latvia,
Lithuania, Belorussia, Hungary, Ukraine and Moldavia. UPLN was observed in the 1950ies
and remeasured in the 1970ies. The readjustment was completed in 1982. It comprises
more than 350 nodal points. The lengths of lines vary between 70 km in the western part
and 200 km in the eastern part. The reference gauge is KronstadtlRussia. The adjustment
1982 was carried out using normal heights (UPLN 82).

The goal of the UELN95 is the realization of an unique European height system with an
accuracy of better than 1 decimeter. The enlargement to UELN 95 is performed in two
parts:

(1) Substitution of data of such network blocks which already are part of UELN 73 but
include new measurements.

(2) Step by step adding network blocks of the UPLN.

30
TIde gauge sites are essential to estimate a possible secular sea level rise. The tide gauge
sites will provide all the infonnation for the combination of ellipsoidal heights and physical
heights along coast lines. These results will be of great importance for the proposed cam-
paign covering the whole of Europe.

EUVN related observations

Three kinds of observation groups are necessary:

- station measurements for the documentation of situation and the detennination of

eccentricities between different markers
- GPS campaigns for the detennination of the ellipsoidal height of all defined EUVN
points
- connecting levellings between the EUVN sites and the UELN nodal points for the
detennination of the physical height of all defined EUVN points.

4. Schedule of the EUVN97 GPS Campaign

The observation period of the EUVN97 GPS campaign was:

Start time May 21, 1997, 18:00 UTC

Stop time May, 29, 1997, 6:00 UTe.

In total the campaign consists of about 195 sites: 66 EUREF- and 13 national penn anent
sites, 53 UELN and UPLN stations and 63 tide gauges (Figure 1). The EUVN is a joint
collaboration of most of the European countries.

The northernmost EUVN point is situated at Ny Alesund (78.9°, 12.0°) on Spitzbergen, the
southernmost point is the tide gauge Lamaka (34,9°, 33,6°) on Cyprus. The westernmost
point Reykjavik (64,15°, -22,0°) is situated in the North Atlantic, the easternmost point
Yozgat (39,8°,34,8°) in Turkey.

The Baltic Sea Level GPS campaign 1997 and other local GPS activities were perfonned
simultaneously to the EUVN97 GPS campaign. For EUVN and BSL stations and under
consideration that at several stations collocations between different receivers were
perfonned more than 200 teams worked in the field.

Three types of receivers were used within the EUVN GPS campaign:

- Turbo Rogue Receivers at penn anent sites (IGS standard),

- Trimble SSI or SSE,
- Ashtech Z 12.

31
 Bundesamt lOr Kartographie und Geodasie, January 26, 1998

Figure 1: Distribution of EUVN stations

• EUREF sites o GPS permanent stations - nodal points

~ GPS permanent stations - EUREF • Tide gauge sites
6. GPS permanent stations C+) GPS permanent stations - tide gauge
• UELN & UPLN nodal points ,', UELN lines

32
Collocation of the receiver and antenna types of some selected observatories is foreseen in
order to derive type dependant offsets. For the GPS campaign, the time interval was set to
30 s, the elevation mask is 5°.

5. EUVN GPS Processing

The data preprocessing performed by 9 EUVN Preprocessing Centers (PPC) is mainly a

check concerning completeness and consistency of the data and the auxiliary information.
The PPCs were requested to prepare complete access information and/or data flow
guidelines for the observing agencies before the start of the campaign.

The EUVN Data Center (DC) is established at the Bundesamt fUr Kartographie und
Geodasie, Aussenstelle Leipzig.

The task of the Analysis Center (AC) is to process the data of a subnetwork. A subdivision
of the whole EUVN Network will be done under the aspect of receiver type and regions.
9 European institutions are ready to contribute as Analysis Center.

6. Further Activities - Connection Levellings, Tide Gauge Measurements

To reach the objective of the EUVN as a European Height Reference Network, in the next
stage the connection levellings between the GPS points and the UELNIUPLN nodal points
must be delivered and the bases for the processing of the tide gauge data which were
included into the project must be created.

References

Augath, w.: Proposals for a European Task Force on Vertical Datum within the EUREF-
Subcommission. Presented at the EUREF Symposium in Warsaw, June 7-11, 1994, Dt.
Geod. Komm. Astronomisch-Geodatische Arbeiten, Heft Nr. 54, S. 171-175.
Balasubramania, N.: Definition a Realization of a Global Vertical Datum. Report No. 427,
Dep. of Geodetic Science and Surveying, Ohio State University, Columbus, Ohio 43210-
1247, 1994.

Birardi, G.: The Italian North-South GPS Traverse and a proposal for a "first order" Italian
geoidal net. Bull. Geod., Berlin 67 (1993)4, pp. 201-209.

Boucher, c., Woppelmann, G.: Proposal for a European Primary Tide Gauge Network
(EPTN), EUREF TWG Meeting, Bad Homburg, Dec. 15-16, 1994.
Brouwer, F. I. I., De Min, E. I.: On the Definition of a European Vertical Datum. Presented
at the EUREF Symposium in Warsaw, June 7-11, 1994, Dt. Geod. Komm. Astronomisch-
Geodatische Arbeiten, Heft Nr. 54, S. 171-175.

33
Carter, W. E., Aubrey, D. G., Baker, T. F., Boucher, c., Le Provost, c., Pugh, D. T., Peltier,
W. R., Zumberge, M., Rapp, R. H., Schutz, R. E., Emery, K. o. and Enfield, D. B.: Geodetic
fixing of Tide Gauge Bench Marks. Woods Hole Oceanographic Institution Technical
Report, WHOI-89-3l, CRC-89-5, August 1989.

Carter, W. E. (ed): Report of the Surrey Workshop of the IAPSO Tide Gauge Bench. Fix-
ing Committee held l3-l5 December 1993 at the Institute of Oceanographic Sciences Dea-
con Laboratory, Wormley, UK. NOAA Technical Report NOSOESOOO6, October 1994.

Ehrnsperger, w., Kok, J. J.: Status and Results of the 1986 Adjustment of the United Euro-
pean Levelling Network - UELN-73, Contributed paper to the Symposium on Height
Determination and Recent Crustal Movements in Western Europe, Federal Republic of
Germany, Sep. 15-19., 1986.

EUVN Working Group: First circular of the EUVN project, lAG Subcommission for
EUROPE (EUREF), WabemIMUnchen, March 1996.

Ihde. J., Schlater, w.: Proposal for a European Vertical GPS-Reference Network
(EUVERN). Presented at the EUREF Symposium, May 3-6, 1995, Kirkkonummi, Finland.

Ihde, J., Schlater, w., Gurtner, w., Woppelmann, G., Harsson, B. G., Adam, J.: Concept
and Status of the European Vertical GPS Reference Network (EUVN). Presented at the
EUREF Symposium in Ankara, May, 1996, Dt. Geod. Komm., Astronomisch-Geodatische
Arbeiten, Heft Nr. 57, S. 218-225.

Lang, H., Sacher, M.: Status and Results of the Adjustment and Enlargement of the United
European Levelling Network 1995. Presented at the EUREF Symposium, May 3-6, 1995,
Kirkkonummi, Finland.
Kakkuri, J.: The Baltic Sea Level Project. Allgemeine Vermessungsnachrichten, Heidel-
berg, 102 (1995), 8/9 pp. 331-336.
Seeger, H., Adam, J., Augath, w., Boucher, c., Gubler, E., Gurtner, w., Van der Marel, H.,
Zielinski, J. B. : The new European Reference System EUREF - Status Report 1995.
Invited paper at the XXIst General Assembly oflUGGIIAG, Boulder, Colorado, USA, July
2-14, 1995.

Torge, w., Basic, T., Denker, H.: Long Range Geoid Control Through the European GPS
Traverse. Dt. Geod. Komm., R. B., Nr. 290, MUnchen, 1989.

34
 FROM UELN-95 TO EVS 2000
EUROPEAN ACTIVITIES FOR A CONTINENTAL VERTICAL DATUM

W. AugathI, F. Brouwer, H. Lang3, J. van Mierl04, M. Sacher

IGeodetic Institute, Technical University of Dresden, MommsenstraBe 13,
D-01062 Dresden, Germany, email: [email protected]
2Rijkswaterstaat, Survey Department, Postbus 5023, NL-2600 GA Delft,
The Netherlands
3Bundesamt fur Kartographie und Geodasie (former Institut fur Angewandte Geodasie), D-
60598 Frankfurt, Germany
4 Geodetic Institute, University of Karlsruhe, Postfach 6380, D-76131 Karlsruhe, Germany

Abstract

In Commission X (continental networks) a lot of new scientific activities have started. For
Europe this task is overtaken since 1989 by the EUREF -subcommission in strong
cooperation with the European Survey Agencies represented by the "Comite Europeen des
Responsables de la Cartographie Officielle (CERCO)". Until now only the European
Terrestrial Reference System 1989 (ETRS 89) has already been realized which is identical to
the global realization of the International Terrestrial Reference System (lTRS) in Europe at
the epoch 1989.0.
The activities for a similar solution for the height component are much more time-
consuming. That is why the EUREF-Technical Working Group in view of the differenttime
frames of the main users decided in 1993 to divide the realization into two steps:

Step 1: United European Levelling Net 1995 (UELN-95)

A "quick" solution on the 0,1 m-level as height reference for Europeanwide geoinformation
using systems, using existing levelling data sets and a simple static adjustment model.
Step 2: European Vertical System 2000 (EVS 2000)
A realization on the highest possible level (1 cm and better) for National Survey Agencies
and scientific requirements. The activities for UELN-95 have to be carried out as usable
basis for this step 2.
The paper describes the problems of a change from national to continental datums, the status
of the first step UELN-95 and the concept of the realizationofEVS 2000.
Concept and Status ofUELN-95

The concept of UELN-95 can be summerized in one sentence: "Realization of a vertical

datum in Europe as soon as possible for the determination of transformation parameters to
the national height datums on the decimeter-level, which can also be used as a first step for a
final solution on the highest possible level (EVS 2000)". This causes the following
procedure:
- Use of the existing Westeuropean levelling data set UELN 73/86 (geopotential
differences between nodal points and connections to Westeuropean tide gauges, published
by Ernsperger and Kok (1987) and enlargementto Eastern Europe.
- Use of the Westeuropean datumpoint (tide gauge Amsterdam) and normal heights as
common height system.

Number of Sum of a posteriori

Block observations redundancies standard deviation
[kGal'mm]
Austria (A) 128 29.020 0.62
Belgium (B) 54 19.494 1.22
Switzerland (CH) 13 4.715 1.12
Germany (D) 745 271.488 0.83
Denmark (DK) 26 6.813 0.66
Spain (E) 101 27.255 1.85
France (F) 175 46.826 2.01
Italy (I) 97 32.453 1.78
Netherlands (NL) 951 164.014 1.08
Portugal (P) 22 5.857 1.77
Great Britain (GB) 60 15.000 1.72
Norway (N) 194 70.992 1.67
Finland (F) 89 20.142 0.76
Sweden (S) 122 34.865 1.74
Czech Republic (CZ) 82 28.742 1.10
Hungaria (HU) 49 12.362 0.54
Slovenia (SL) 11 1.580 0.51
Poland (PL) 179 54.863 0.99
Slovakia (SK) 74 18.518 1.39
Total 3172 865.000 1.23

Table 1: Accuracy of the national blocks derived from the variance component estimation
(UELN-95/1 0, Status 7/1997)

- Static adjustment model without additional parameters and the use of aposteriori-standard
deviations derived from variance component estimations between the national blocks.
- National computing centers and one international center for the data and the common
adjustments at Federal Office for Cartography and Geodesy (former IfAG), Germany.

Unfortunately the common levelling data set of the EasteuropeanLevelling Net including all

36
former Easteuropean socialist countries was not available. It had to be reconstructed by the
Easteuropean Survey Agencies. Missing connections between Eastern and Western Europe
were remeasured and new national data sets, remeasured after 1973, were introduced.

United European Levelling Network 1995 (UELN-95)

(Status of the Adjustment and Enlargement)

- UElN -95/10: lsollnes 01 Precision I kGol·mm I

• Reference Point: Tide Gouge Amsterdam

Picture 1: Isolines of standard deviations of the adjusted heights in relation to the datum-
point Amsterdam (Status 711997)

The adjustments had to be realized step by step:

- UELN-73/94: Repetition of the UELN-73/86-adjustment using weights for the national
blocks derived from a variance component estimation.
UELN-95/1,2: The former block D (Germany) was removed and replaced b) the new
German First Order Levelling Net "DHHN 92" (East- and Westgermany including new
connections).
UELN-95/3: The former block A (Austria) was replaced by the remeasured Austrian First
Order Levelling Net.
UELN-95/4, 5,6, 7, 8: Enlargement by the new national levelling networks of the Czech
Republic, Hungary, Slovenia, Poland and Slovakia.
UELN-95/9, 10: The former block NL (Netherlands) was replaced by the Fourth Precise
Levelling Net of the Netherlands.

37
In table 1 the accuracy of the national network blocks derived from variance component
estimations are put together. Picture 1 shows the isolines of standard deviations of the
adjusted heights (UELN-95/1 0) in relation to the datum-point Amsterdam (Lang et al. 1997).
In the near future the UELN-95-data set will be enlarged by new national blocks of Croatia,
Bulgaria and Romania. In Southeastern Europe Yugoslavia is still missing. In Northeastern
Europe a connection with the Baltic States, Russia to Scandinavia would be extremely
important. In Western Europe some further national block can be replaced (CH, DK) and the
new observations between France and the United Kingdom via the Channel-Tunnel will be
introduced as well as new hydrostatic connections between Germany and Denmark and
another one in Denmark.

Concept of European Vertical System (EVS 2000)

The actual proposal can be described as development of a Continental Vertical System on

the one-centimeter-Ievel. It includes components concerning the Vertical Datum, the
adjustment model, the network design in the field of reliability and checks against systematic
errors as they are put together in table 2.

Continental Vertical Datum

based on a selected or "mean" tide gauge and geopotentialnumbers
Kinematic adjustment model
using the national levelling epochs and actual kinematic information as repeated
GPS/Gravity/SLRlVLBI, Tide gauges
Higher reliability
by the use of denser levelling networks as done in UELN 73 (national 1sl and 2nd order
levellings)
Check against systematic errors
by independent observations (later, e. g. Geoid-traverses on the 1 to 2-cm-Ievel,
hydrostatic-levelling-traverses)
Connection to a Global Vertical Datum
on the 0,01 to 0,1 m-Ievel (e. g. by Improvement of the Global Geoid)

Table 2: Development of a Continental Vertical System on the one-cm-Ievel: EVS 2000

Continental Vertical Datum

Because of the ongoing development of a Global Vertical Datum, the European Vertical
Datum should be defined as continental datum. The reference surface can be fixed by a
selected or a "mean" tide gauge. Preliminary values of the European sea surface topography
and additional special effects will be obtained by the EUVN-GPS-campaign 1997 in
combination with the existing UELN 95-results. Until a final decision is made, the existing
UELN-definition will be used (tide gauge Amsterdam, geopotential numbers, normal
heights).

38
Kinematic Network

The fonner UELN-subcommission already proposed the upgrade from the static to a
kinematic height network (Remmer 1988). Picture 2 gives an overview over existing
levelling epochs in Europe. The kinematic infonnation of the fonner European levelling
epochs which give the relation to the last 50 - 100 years must be completed by actual
kinematic infonnation of selected points with repeated height-related observations like
GPS/Gravity/SLRlVLBI and tide gauges on the highest possible level. In this field a lot of
data are also still existing such as: the ITRS-network, IGS, EUREF pennanent sites, sites of
regional and local campaigns. For height change detennination a special focus for an
accurate height connection to the levelling network is necessary as well as a more or less
pennanent supervision of this connection. Therefore a special network which includes only
those stations should be selected (Augath 1990). A first data set could be overtaken from the
EUREF -pennanent GPS-sides (Picture 3).

Netwo~ blOck 1860 1870 1880 1810 1900 1910 19'20 1930 1940 1950 1960 '970 1980 1990 2000

I J-------"-~~~~_~_ I I
_(Pl.)
•• ••

- - --
•• ••
H~na(HIJ}

Czech 1 1 _ (Cl)

• •
Slovakia (SI()
SlOVenia (S~

"usllla (AT)
BelgIum (BE) I

- - -----
••
$WlI>o<1ond (CH)
Gonnony (DE)
De_10K)
s.o.. (ES)

•
Fnonce (FI\))
Gte" Brilaln (GB)

- - --
ItaIy(rT)
""",ay(NO)
"e"""'_ (Nl)
Ponugal (PT)
Sweden (SE)
Finland (FI)

Picture 2: Existing levelling epochs in Europe

39
 Reliability
The adjustment of UELN 73/86 used only
a reduced number of levelling
observations (large loops with 1000
kilometers circuit). The data management
of today allows the use of all first (and
second) order levelling observations in
one adjustment. This will give an upgrade
of the reliability.

Check against systematic errors

On the European level independent checks
against systematic errors are extremely
necessary. The best chance for the next
years is given by 3D/Geoid-traverses.
For this purpose the European geoid
should be upgraded to the 1-2 em-level.
I:>. QP8 per_nenl .ullon. A second possibility which is more
• TI de aaug- ell..
• QP8 per_nenl .ulloM - lid_ g"ug. expenSIve could be realized with
'V UELN II.....
hydrostatic levellings along the rivers e.
g. with the existing equipment of
Picture 3: The EUREF-Permanent-GPS-
Rijkswaterstaat (Netherlands) and
side
connections to UELN-Ievelling lines.

Outlook

Since 1993 the European countries realize a common vertical datum within the EUREF-
subcommission. It was a good decision to divide the work into two steps. The "quick"
solution UELN-95 is already available for most of the European area. Meanwhile the
timeconsuming acquisition of former levelling epochs has started and a first kinematic
adjustment project with the levelling data of Switzerland is on the way (Gubler, Wolfram
1997). Nevertheless a lot of work in the field of kinematic modelling and the integration of
different height relevant observation data and the upgrade of the quality of the height
component derived from space techniques has be done.

References

AUGATH, w. (1990): Proposal for the Establishment of an European Height Supervision

Systems (HSS-WE). GPS for Geodynamics, Volume II, pp. 135 - 140, Luxembourg
(Walferdange),1990.

AUGATH, W. (1996): UELN 2000: Possibilities, strategy, concepts. Proposals for a

European Task Force on Vertical Datums within the EUREF-Subcommission.
Veroffentlichung der Bayr. Kommission fur Intemat. Erdmessung, No. 54, pp. 170 - 175,
Beck-Verlag, Munich, 1994.

40
BOUCHER, C. (1994): Global Vertical Reference and the IERS Terrestrial Reference
System (ITRS): concepts and realizations. Proceedings International Symposium
INSMAP 94, Hannover, 1994 and Publication in LAREG, Serie Communication, CM6,
Paris, 1994.

ERNSPERGER, W., KOK, J. J. (1987): Status and Results of the 1986 Adjustment of the
United European Levelling Network - UELN 73. In Pelzer, H. and Niemeier, W. (Eds.):
Determination of Heights and Height Changes, pp. 7 - 45, Dfunmler Verlag, Bonn, 1987.

GUBLER, E., WOLFRAM, S. (1997): Personel information (unpublished), 1997.

HECK, B., RUMMEL, R. (1989): Strategies for solving the Vertical Datum Problem using
terrestrial and satellite geodetic data. lAG-Symposium 104 (Edinburgh): Sea Surface
Topography and the Geoid, Springer, 1989.

LANG, H., SACHER, M. (1996): Status and Results ofthe Adjustment and Enlargement of
the United European Levelling Network 1995 (UELN-95),pp. 163 -169, 1996.

LANG, H., STEINBERG, 1. (1993): Zur Entwicklung der Hohennetze auf dem Territorium
der neuen BundesHinder. Allgemeine Vermessungsnachrichten, Karlsruhe, 100, pp. 295-
309, 1993.

NIEMEIER, W. (1987): Observation Techniques for Height Detemination and their Relation
to usual Height Systems. In: Pelzer, H. and Niemeier, W. (Eds.): Determination of
Heights and Height Changes, pp. 85 - 108, Dfunmler Verlag, Bonn, 1987.

REMMER, O. (1987): The United European Levelling Network - Present State and Future
Plans -. In: Pelzer, H. and Niemeier, W. (Eds.): Determination of Heights and Height
Changes,pp.3 - 5, DfunmlerVerlag,Bonn, 1987.

RUMMEL, R., REUNISSEN, P. (1988): Height datum definition, Height datum connection
and the role of the Geodetic Boundary Value Problem. Bulletin Geodesique, 62, pp. 477 -
498, 1988.

SEEGER, H., AUGATH, W., BORDELEY, R., BOUCHER, C., ENGEN, B., GURTNER,
W., SCHLUTER, W., SIGL, R. (1993): Status Report on the EUREF-GPS-Campaign
1989 to the lAG EUREF Subcommission. Veroffentlichung der Bayr. Kommission fUr
Internat. Erdmessung, No. 52, pp. 26 - 34, Beck-Verlag, Munich, 1993.

TORGE, W. (1990): The Geoid in Europe: Status, Requirement, Prospects. GPS for
Geodynamics, Volume II, pp. 119 - 126, Luxembourg (Walferdange), 1990.

wDBBELMANN, H. (1993): The Adjustment of the German First Order Levelling Net
(DHHN 85). In: Arbeitsgemeinschaft der Vermessungsverwaltungen der Lander der
Bundesrepublik Deutschland (AdV): Die Wiederholungsmessungen 1980 - 1985 im
Deutschen Haupthohennetz der Bundesrepublik Deutschland, pp. 141 - 154, published by
Bayerisches Landesvermessungsamt,Munich, 1993.

41
 GLOBAL GPS NETWORKS AND THE DETERMINATION OF
EARTH ROTATION PARAMETERS

V. Ashkenazi, T. Moore, L. J. Symons, W. Chen and M. H. O. Aquino.

Institute of Engineering Surveying and Space Geodesy

University of Nottingham
Nottingham
NG72RD
United Kingdom

ABSTRACT

Recent developments of the in-house GPS processing software at the Institute of Engineering
Surveying and Space Geodesy (IESSG), University of Nottingham, has led to the capability of
processing very large or even global GPS networks. This has allowed the investigation of several
methods of constraining a network of stations and comparative results of these tests are presented
in this paper.

A precise knowledge of the Earth's position and orientation in space is required for a number of
astronomical and geophysical purposes, ranging from the tracking of interplanetary spacecraft, to
the study of the Earth's interior structure. The new software package enables the determination of
Earth Rotation Parameters (ERPs) at regular intervals from the global network of GPS data, thus
defining an Earth Rotation Parameter series and improving the final coordinate solution
accuracies. Tests performed at the IESSG are discussed and preliminary results are shown.

1 INTRODUCTION

The formation of the International GPS Service for Geodynamics (IGS) has provided the
opportunity for processing high precision, global GPS networks within a number of days after
data collection. Such easily accessible data can be used within global GPS network adjustments
to determine numerous parameters, including the orientation of the Earth.

The movement of the Earth's axis of rotation with respect to its surface, and variations of the rate
of rotation of the Earth about this axis, has been predicted for over a century. Techniques used
included optical methods such as the timing and locating of solar eclipses, plus occultation's of
stars by the moon. The precise knowledge of the Earth's position and orientation in space is
required for a multitude of astronomical and geophysical purposes, ranging from the tracking of
interplanetary spacecraft, to the study of the Earth's interior structure. These Earth Rotation
Parameters may now be determined to a high level of accuracy and resolution, using recent, new
space geodetic techniques, such as the Global Positioning System (GPS).

The IESSG has recently developed software which allows the estimation of Earth Rotation
Parameters from GPS data. A daily series of Earth Rotation Parameters has been determined and
has been compared with the corresponding series computed by lGS Analysis Centres.

2 POLAR MOTION AND UNIVERSAL TIME

The Earth rotates about an axis that has a particular orientation in space. The orientation of this
axis changes with respect to time, when viewed against an external non-rotating inertial frame,
due to motions known as precession and nutation.

More uncertain is the orientation of the solid Earth with respect to the rotation axis, or
alternatively, the location of the intersection of this rotation axis with the Earth's crust. The
displacement of this rotation axis from a conveniently defined reference point, the geographic
North pole, is termed 'polar motion'. This polar motion is primarily cyclic, due to the dynamic
response of the solid Earth to loading caused by the atmosphere, oceans and liquid core.

Polar motion is commonly described using two cartesian components, termed xp and yp, where
the xp component may be considered to be the displacement of the Earth's axis of rotation
projected on to the IERS Reference Meridian (IRM) and the yp component is the displacement
projected on to a meridian 90 degrees west of the IRM.

Not only does the position of the rotation axis vary with respect to the solid Earth, but the rate at
which the Earth spins also changes. This results in an excess length of day (d LOD) which varies
from day to day, and is of the order of 2 milliseconds. The drift in Universal time (UTI), is
commonly expressed as UTC-UTI l or UTI-TAl 2 . This variation is largely unpredictable and
can result in positioning errors of many tens of centimetres at the equator if it is not closely
monitored.

To monitor changes in the Earth's orientation in space, two precise stable reference frames must
be defined. One must rotate in sympathy with Earth (Earth fixed whose origin is the Earth's mass
centre), and one must be independent of the Earth's rotation (Inertial frame), from which changes
in orientation can be ascertained.

Similarly, if the Earth fixed (terrestrial) and inertial (celestial) frames are to be used, then
transformations between the two frames must be defined. The transformation between inertial and
Earth fixed frames involves transformations accounting for polar motion, the earth rotation,
nutation and precession and involves several intermediate reference frames. The reader is referred
to (McCarthy, 1992) for further details.

1 UTC= Co-ordinated Universal Time

2 T AI= International Atomic Time

44
3 SOLVING FOR EARTH ROTATION PARAMETERS

If GPS satellite orbits and receiver locations were sufficiently well known, the GPS data may be
used to simply monitor" changes in the Earth's orientation. When GPS data are processed in
practice, satellite orbits, station locations and Earth rotation parameters must all be estimated
simultaneously. The estimated orbital parameters are referred to a celestial reference frame,
whereas the network of estimated station locations provides the terrestrial reference frame. The
relative motion of these two frames allows the Earth orientation parameters to be estimated.

When solving for Earth orientation using GPS techniques, double difference carrier phase data are
used to obtain the receiver and satellite positions. Then the initial positions of the receiver and
satellite are used to derive observation equations which allow the estimation of changes to the
three Earth Rotation Parameter unknowns, namely LUp, Ll yp and LlUTC-UTl in a least squares
adjustment.

In fiducial GPS data processing, a certain number of stations are deemed as known 'fiducial'
reference sites. Their positions are highly constrained to their coordinate values obtained from
previous GPS, VLBI, or laser ranging measurements. Hence coordinates of the free stations in the
network may be estimated directly. Since the orientation of the Earth is of interest, these sites
should ideally span the globe, with a longitudinal spacing of no more than 90 0 and coverage in
both Northern and Southern Hemispheres. Additional sites are required to fill in the terrestrial
network redundancy.

In the free network (non-fiducial) approach no station locations are fixed, and only very loose
constraints, of the order of lO's to 1000's of metres, are applied. The scale and origin of the
network is provided by the satellite force model, and the GPS data. Consequently, only the
orientation, provided by the fiducial sites is missing, but the direction of the z-axis is further
constrained by the Earth's daily rotation.

In order to make the free (non-fiducial solution) consistent with the scale, origin and orientation of
the ITRS, a similarity transformation can be applied. The free (non-fiducial) technique can readily
sense changes in the orientation of the Earth over time but cannot establish the absolute
orientation of the Earth in space. One disadvantage of this approach is that the time series of Earth
orientation variations obtained from 2 or more non-consecutive GPS tests, cannot be combined
directly to yield a longer Earth orientation series without some transformation.

4 THE EARTH ROTATION PARAMETER SOFTWARE

An Earth Rotation Parameter estimation routine has been developed at the IESSG and
successfully integrated within the existing in-house GPS software suite, known as GAS (GPS
Analysis Software).

The routine uses initial a priori IERS (International Earth Rotation Service), xp, yp and UTC-UTI
values, taken from the Bulletin B series. The corrections to the approximate values of the three
unknown Earth Rotation Parameters, (LUp, Llyp, LlUTC-UTl), are then estimated in the least
squares process and applied to the initial a priori values, thus providing daily estimates of these
parameters. The observations input are double difference dual frequency carrier phase GPS data.

45
The orbit parameters solved for are-
Position, velocity, solar radiation/y-bias for each satellite
xp, yp and UTC-UTI values per day

The force models used are-

WGS84 gravity model (degree and order 8)
Schwiderski Ocean tide loading model.
Basic solar radiation, spherical satellites.

For the results detailed in this paper, 24 hour two session (arcs) have been processed.

5 TEST DATA AND PROCESSING STRATEGIES

To test the software and strategies, a data set comprising data from 18 IGS tracking stations,
evenly distributed across the globe, was collated. The data set covered a 5 day period from 23rd
October 1995 to 27th October 1995, (Julian Day 295-300). The selection of these sites was based
on the need for a strong network, well represented in the Northern and Southern hemispheres,
with baseline lengths in excess of 6000 km, thus providing a truly global GPS network of tracking
stations.

Two strategies were tested:

(1) Global network adjustments (no Earth Rotation Parameters estimated)

(2) Global network adjustments estimating and applying Earth Rotation Parameters.

A number of network adjustment strategies were carried out A free (non-fiducial) adjustment was
computed with very loose constraints of the order of 10 metres applied to the ITRF stations and
1000 metres to the other stations. A full fiducial network adjustment was carried out, in which
constraints of 2cm were applied to the ITRF93 coordinates of 7 stations and all other stations in
the adjustment were left free. Fixed station network adjustments were also performed, whereby 1,
4 and 8 stations were fixed rigidly. However, the fixed network strategy results are not presented
in this paper.

The aim of this analysis was to assess whether there was any significant improvement, in terms of
final coordinate accuracies, when Earth rotation parameters were estimated and applied in the
adjustment (Table 1). It should be noted that in this paper, accuracy is defined as the difference
between the assumed true value of the position of a station (ITRF93 values) and the estimated
one. This is expressed by the root mean square (RMS) coordinate differences. For the purpose of
this paper, precision is defined by the session to session coordinate repeatability, namely the RMS
of the coordinate differences from each session solution from the weighted mean coordinates of
all the session solutions.

The daily Earth Rotation Parameters obtained in the tests have also been compared with the
corresponding estimates made by four IGS Analysis Centres, over the same 5 day period. RMS

46
values have been quoted together with the mean offset. Both are expressed in terms of milli-arc-
seconds (mas) for xp and yp and milli-seconds (ms) for UTC-UTI (Table 2).

It should be noted that these are only preliminary results and more comparisons are to be made
with those Earth Rotation Parameters from a greater number of IGS Analysis Centres and the
IERS in the near future.

6 ANALYSIS OF RESULTS

Table 1 illustrates that when Earth Rotation Parameters have been estimated and applied, there
has been a significant improvement to the final coordinate solution, in terms of accuracy. This is
clearly shown in Tests 2 and 4 for example, where X coordinate differences of 2.2 and l.3cm have
been obtained. This complies with the statement, 'that the variation of station coordinates caused
by polar motion is recommended to be taken into account, since the polar displacement has a non-
zero average over any given time span, and should be considered in global solutions at the cm
level '(McCarthy, 1992).

It can be seen from Table 1, that when the Earth Rotation Parameters are not estimated and
applied in the final global network adjustment, the final coordinate solution is degraded in terms
of accuracy.

Table 1 RMS Errors of Coordinates with and without the

Estimation of Earth Rotation Parameters.

Parameter Test 1 Test 2 Test 3 Test 4

erp off erpon erp off erpon erp off erpon erp off erpon
RMS (X)cm 4.1 3.8 5.5 3.3 0.35 0.34 5.2 2.9
RMS (Y) cm 6.1 6.0 6.0 6.1 0.34 0.39 2.3 1.8
RMS (Z) cm 7.6 7.4 4.0 4.0 0.56 0.44 2.2 2.7

Test I=All Free, Test 2=1 station RCM5 constrained 2cm, Test 3= ITRF constrained to 10m
others 1000m, Test 4=Fiducial (7 stations constrained to 2cm).

Table 2 indicates that the level of agreement in the pole coordinate values with the external
results is on average, better then 0.13 mas xp, 0.10 mas yp and 0.010 ms UTC-UTI . The results
show that they fall within the similar quoted rms accuracies of various IGS bodies, for example,
0.15 mas xp and yp and 0.006 ms UTC-UT1. It should be noted that standard errors of 0.05 mas
xp, 0.15 mas yp and 0.03 ms UTC-UTI, have also been achieved. The free (non-fiducial) network
adjustment has been used to generate the results shown in Table 2, where 10m constraints were
applied to the ITRF stations and all other stations were loosely constrained to 1000m.

47
 Set No. xp (mas) yp (mas) UTC-UT1 (ms) I
Offset RMS Offset RMS Offset RMS
1 0.06 0.16 0.05 0.10 -0.005 0.0149
2 0.05 0.13 0.01 0.16 0.009 0.0112
3 0.07 0.14 0.04 0.10 -0.009 0.0124
4 0.02 0.13 -0.03 0.08 0.007 0.0101

Table 2 Comparison of 1-day Earth Rotation Parameters with External Results.

Set l=IGS (weighted mean), Set 2=Jet Propulsion Laboratory, Set 3=NOAA, Set 4=ESAIESOC

7 CONCLUSIONS

A software module has been successfully developed and implemented within the existing software
suite. The Earth Rotation Parameters obtained with the Nottingham software agree with those
obtained by the four IGS Analysis Centres at a level of 0.13 mas xp, 0.10 mas yp and 0.010 ms
UTC-UTI.

A significant improvement occurred in the accuracy of the final station coordinates when the
Earth Rotation Parameters were estimated and applied in the global network adjustments. This can
be illustrated by an improvement in the X coordinate values of 1.3cm in the fiducial solution and
2.2cm in the 1 station constrained to 2cm solution.

These preliminary test were performed with I-Day (2 session arcs). It is envisaged that future
work will include a 3-day (5 session arc 60 hrs.) smoothed series. In addition, it is anticipated to
implement the T30 solar radiation model, (Fliegel & Gallini, 1996), within the existing software
suite along with various gravity field tests. Similar tests will be carried out to ascertain if any
further improvement to the final coordinate accuracies and Earth Rotation Parameter series has
occurred.

REFERENCES

Fliegel, H. F. and Gallini, T. E, (1996), Solar Force Modelling of Block IIR Global Positioning
System satellites, Journal of Spacecraft and Rockets, Vol. 33, number 6, pp 863-866.

McCarthy, D.D., (1992), IERS Standards (1992), IERS Technical Note 13, Observatoire de Paris,
July 1992.

48
 REALIZATION OF THE ITRF-94 IN THAILAND AND MALAYSIA
BY USE OF A COMBINED NETWORK FOR GEODYNAMICS
AND NATIONAL SURVEY

M. Becker, E. Reinhart, H. Seeger

Bundesamt fur Kartographie und Geodasie, Frankfurt a. M., Germany
S. Mingsamon, C. Boonphakdee
Royal Thai Survey Department, Bangkok, Thailand
S.Abu
Department of Surveying and Mapping, Kuala Lumpur, Malaysia

ABSTRACT

In 1994 and 1996, two major GPS campaigns for geodynamics, named GEODYSSEA, were
conducted in South East Asia. The paper gives an introduction to the GEODYSSEA project
and its results. In cooperation with the national survey authorities, this occasion was used to
realize the ITRF in several countries of South East Asia. This paper describes the back-
ground and results of GEODYSSEA and the additional measurements as well as trans-
formations of the national datums of Thailand and Malaysia to the ITRF-94. By use of the
GEODYSSEA data and also by repeated occupations of the first order stations in the re-
spective countries an accuracy of 2 cm in position and about 3 cm in height was obtained for
the ITRF-94 coordinates. Depending on the accuracy of the existing local coordinates the
transformation parameters between the national datum and ITRF-94 were determined with
and accuracy of 5 to 10 cm.

1. INTRODUCTION

The determination of geocentric coordinates in a well defined reference frame is the basis for
the establishment and the maintenance of a geodetic infrastructure. It is needed for all activi-
ties in GIS, surveying and navigation in all countries. Unified national geodetic datums are
required for the integration of regional geodetic and geodynamic GPS campaigns. Up to
now, this has posed a problem in the area of South East Asia because there were no per-
manent GPS receivers of the IGS network deployed, which would have facilitated the con-
nection to the ITRF.

In the framework of the GEODYSSEA project in South East Asia, the survey authorities of
Thailand (Royal Thai Survey Department, RTSD) and of Malaysia (Department of Surveying
and Mapping, Kuala Lumpur, DSMM) and the Bundesamt fur Kartographie und Geodasie
(BKG, formerly IfAG) decided to improve the national survey networks by establishing high
precision links to the ITRF.
The objectives are:
• To establish a clearly defined and internationally accepted geodetic reference frame.
• To establish a zero-order GPS Network which would be used as the framework for the
national GPS networks.
• To improve the accuracy of the existing points in the national GPS networks by use of
the final coordinates from the project as fiducial stations in the GPS network
adjustments.
• To establish definitive transformation parameters between the current mapping datum
(e.g. INDIAN 1975) and the ITRF-94 reference frame.
• To investigate magnitudes and directions of crustal motions in the area to be able to ac-
count for the time dependent changes in the coordinates.

2. THE GEODYSSEA PROJECT

The GEODYSSEA Project was carried out under the Contract Cll *-CT93-0337 between
the European Commission and a number of ASEAN countries.

The European Partners involved in the geodetic activities are the Bundesamt fur Kartogra-
phie und Geodasie, Frankfurt am Main, observations and Analysis; Delft Inst. of Earth Ori-
ented Space Research (DEOS) , analysis; Ecole Normale Superieure(ENS), Paris, analysis;
GeoForschungsZentrum Potsdam (GFZ) , observations and analysis, coordinator; Inst. of
Physics, Univ. of Bologna, analysis; Royal Observatoire Belgique, Brussels, analysis; Univ.
of Newcastle, Dept. of Surveying, analysis.

The ASEAN Partners are BAKOSURTANAL, Cibinong, Indonesia; Dept. of Surveying and
Mapping, Kuala Lumpur, Malaysia; Bureau of Energy and Mines, Manila, Philippines; Inst.
of Geology, Nat. Center for Nat. Sciences, Hanoi, Vietnam; Royal Thai Survey Department,
Bangkok, Thailand; Dept. of Public Works, Brunei.

GEODYSSEA is an acronym for GEODYnamics of South and South-East Asia. The Objec-
tives are the ,,Determination of Plate Motions and Crustal Deformations Deduced from
Space Geodetic Measurements jor the Assessment oj Related Natural Hazards in South East
Asia". The geodetic part has as principal goal the measurement of the velocities of the con-
tinental plates and the deformations of the Earth's crust by GPS in the region of South-East
Asia. The complete GPS network covers the whole area of South East Asia, see Fig. 1.

In both Thailand and Malaysia there are two points which were selected for establishing new
permanent marks in bedrock. Field reconnaissance, monumentation and field observations
were done by BKG together with the local officers.

Simultaneous GPS observations were conducted in an area of about 4000 by 4000 km on

about 40 carefully selected and newly monumented stations. In 1994, 24 hour sessions on
days 332 to 336 (Nov. 28th till Dec. 3rd), in 1996 on days 109 to 113 (Apr. 18th to Apr.
23 rd ) were observed. In addition, data of IGS and AUSLIG sites for the connection to the
ITRF were used.

The main geodetic results are given in Tab. 1; for details on the GEODYSSEA solutions see
Ambrosius et al. (1997). Fig. 2 shows the ITRF-94 velocities estimated for the network, no
vertical velocity components were estimated.

50
Fig. 1. The GEODYSSEA Network

~QDYSSEA t 994- t 996 VelocUW$ in lTRF-94

[
20

10

I o

-10

Fig. 2. Velocities of the combined GEODYSSEA solution, ITRF-94 (m/year)

51
 Component Precision Accuraq
North East Up North East Up
Coordinates 1994 2.3 4.0 7.1 8.2 8.9 4.9 mm
Coordinates 1996 2.8 4.3 7.0 10.0 14.9 11.8 mm
Velocities 3. 5. not 13. 18. not mm/y
estim. estim.

Tab. 1. Summary of the GEODYSSEA combined solution accuracy and precision.

3. MEASUREMENTS IN THAILAND AND MALAYSIA

During the observations of the GEODYSSEA project in 1994 and in 1996, existing sites of
the first order networks of Malaysia and Thailand were observed with the same accuracy
level as specified for GEODYSSEA

20

0GEOD'YSSEA
+1tWCA
15
... MALAYSIA

'"
'tI
::t
:r:
Q
10
""'I

......
~
i
5

•
~
l
J
~
0 ~
100 105 110 115 120
Longitude

Fig. 3. Points observed in the THAICA94/96 and MALAYSIA94/96 campaigns.

The network designed for the MALAYSIA-94 GPS Campaign included two stations from
the existing sites of the first order network of Malaysia. These stations are located separately
in two networks, one in Peninsular Malaysia and one in SabahiSarawak. In 1996, the same
two and five additional points were used, see Fig. 3 and Tab. 2. In Thailand five additional
stations were observed in THAICA-94 and three of these points were repeated in THAICA-
96, see Fig. 3 and Tab. 2. These points and the connecting points used for the baseline defi-
nition are given in Tab.2 and Fig. 3. Simultaneous GPS data were recorded at all of the
stations for 24 hour observation sessions. Trimble 4000SSE GPS receivers with Trimble

52
4000SST LlIL2 GEO GPS antennas were used for the observations. In THAICA-94,
TRIMBLE SST with squaring technique had to be used on 4 points.

No Full Name ID Location Station Description Da~·s Days

1994 1996
l. Bukit Timbalai TIME Malaysia Malay GPS Network:,
Doppler 30393 station 332-336 109-113
2. Bukit Kertau KERT Malavsia Malay GPS Network
Doppler 30398 station 332-336 109-113
3. Alor Star ALOR Malavsia Malay GPS Network:,
South Base Doppler 30404 station - 109-113
4. Kuching KUCH Malaysia Malav GPS Network
West Base - 109-113
5. Tanjung Kupang KUPA Malaysia Malay GPS Network:,
Doppler 30401 station - 109-113
6. Tawau TAWA Malaysia ITRF Reference
GEODYSSEA Station 332-336 109-113
7. Kuala Terenganu KUAL Malaysia ITRF Reference
GEODYSSEA Station 332-336 109-113
8. Tanjung Baiau TABA Indonesia ITRF Reference
GEODYSSEA Station 332-336 109-113
9. Brunei BRUN Brunei ITRF Reference
GEODYSSEA Station 332-336 109-113
l. Chonburi CHON Thailand ITRF Reference 324-326
GEODYSSEA Station 332-336 109-113
2. Phuket PHUK Thailand ITRF Reference 324-326
GEODYSSEA Station 332-336 109-113
3. Uthaithani UTHA Thailand Thai Nat. GPS Net 324-326
Datum Point, Doppler, 334-336 109-113
4. Chomphon BANH Thailand Thai Nat. GPS Net 324-326 109-113
5. Srisakhet SRIS Thailand Thai Nat. GPS Net 324-326 109-113
6. Lampang OTRI Thailand Thai Nat- GPS Net 324-326 -
7. Pattani PATT Thailand Thai Nat. GPS Net 324-326 -

Tab. 2. Points of the MALAYSIA and THAICA campaigns and duration of observations

4. PROCESSING STRATEGY AND RESULTS

Processing and analysis of all campaigns have been carried out at BKG in cooperation with
DSMM and RTSD with the Bemese Software, version 4.0. The processing strategy adopted
followed the specifications of the EUREF Technical Working Group given in (Gurtner,
1993).

The strategy is summarized as:

• Use ofIGS combined orbits and ERP.
• Processing using 24 hours sessions of 15 /30 sec. data for the 1994 / 1996 campaign.
• Elevation cut -off angle 15 degrees for both campaigns.
• Phase center offsets according to Table IGS-Ol.
• Zenith delay parameters estimated once per 2-hours.
• Ionosphere-free L3 float solution for the 1994 campaign.
• Ambiguity resolution using the QIF (Quasi Ionospheric Free) strategy in 1996.

53
 • Daily solutions computed for storage of normal equations and the analysis of precision .
• Final coordinates from the combination of normal equations of GEODYSSEA and of
all sessions of the Thai/Malay campaigns, constrained to the GEODYSSEA final coor-
dinates in ITRF -94, epoch 1996.3 (Ambrosius et. ai, 1997).

Detailed descriptions of the processing and the results can be found in (Mingsamon et. ai,
1997) and (Becker et.a\., 1997). The precision obtained for the national networks is compa-
rable to that of the GEODYSSEA solution for all of the four add-on campaigns. As an ex-
ample, the repeatability of THAICA94 and THAICA96 results is given in Fig. 4. The day to
day repeatability is shown in Tab. 3 with the example of the MALAYSIA baselines.

Taking the coordinate precision of the final coordinates of 2 to 8 mm in horizontal position,

from 10 to 20 mm in height, and of about 1-2x 10. 8 for the baseline length, and adding the
overall GEODYSSEA accuracy in the ITRF-94, we arrive at an approximate absolute
coordinate accuracy of the 5 Thai and the 5 Malay stations in the ITRF-94 global reference
frame of at least better than 3cm.

RMS Repeatability Thailand 1994+1996

10

~ 5

o
CHON PHUK UTHA BANH SRIS OTRI PATT
Station

Fig 4 RMS repeatability of the combined unconstrained 1994 and 1994 THAICA solution.

BASELINE lENGlH(m) DLAT DWNG DHGT DLGT DLGTILGT

(nun) (mm) (nun) (nun) (PPM)
I. BRUN - 11MB 3~28.2761 7.9 7.0 23.4 9.3 .238
2. 11MB-TAWA 330198.1849 2.0 5.6 17.5 5.5 .017
3. KUAL-KERT 213007.6309 2.9 7.6 12.6 3.4 .016
4. KERT - KUPA 255537.5578 5.4 1.8 18.3 5.0 .020
5. ALOR-KUAL 318235.0105 3.7 6.4 13.3 7.2 .023
6. KUCH - BRUN 652150.2523 4.2 8.3 12.1 8.5 .013
7. TABA - KUCH 168033.4577 2.3 6.7 18.2 6.8 .041
8. KUPA- TABA 590482.7493 1.7 2.3 17.5 3.2 .005

Tab. 3. Repeatability of the baseline components in the MALAYSIA GPS Campaigns (Dlat,
Dlong, Dhgt are the standard deviation of each daily solution.)

54
The major geodetic results can be summarized as follows:
• GEODYSSEA shows a differential rotational motion of the Sunda Block versus the
Eurasian Plate, i.e. the area of Thailand and Malaysia may have a small deviation in
their plate velocities with respect to the main portion of the Eurasian plate.
• For the territory of Thailand a uniform velocity field (i.e. that of the Sunda Block) can
be used for all sites. Therefore, the two campaigns were combined and time-dependent
changes were not considered in the net.
• GEODYSSEA results also indicate a minimum of 12.7 mm/year crustal shortening be-
ing absorbed between Peninsular Malaysia and Borneo.
• This leads to time-dependent distortions in the National net of Malaysia. For Peninsular
Malaysia and for Sabah/Sarawak different velocities are required.
• Accuracy ofITRF-94, epoch 1996.4 coordinates is estimated to be better than 3 cm.
• These coordinates are now used to compute the Helmert Transformation parameters
for the Datum definition of the existing national GPS networks in the ITRF.
• The accuracy of these parameters depends on the quality of the national GPS
networks.
• By reason of their high precision, the national networks could be used for studies of
intra-plate deformations over the long term.

5. TRANSFORMATION PARAMETERS

The reference frame of the current mapping datum of Thailand is the INDIAN 1975 datum
(INDIAN75). At present, the absolute accuracy of the origin point of the first order hori-
zontal network (Thai National GPS Network) in the WGS84 reference frame is in the order
of 1-2 meters because the WGS84 global reference frame was realized through Doppler ob-
servations from the TRANSIT satellite system. The five common stations of the final
THAICA results are used to determine the seven parameters of a Helmert Transformation.
As can be seen from Tab. 4 the rms of the transformation is about 7 cm. For applications
where only a shift is required, an additional 3-parameter transformation is computed.

For Malaysia a datum definition was made in an earlier GPS project and computed by DMA
(Report on the Joint BruneilMalaysiai Singapore Geodetic GPS Project by 512 STRE No-
vember 1992 to June 1993, DSMM, Kuala Lumpur).

Parameters Value Std. Dev Value Std.Dev.

7-Parameter Translation Only
Translation in X 203.884 +- .978 M 205.788 +- .061 M
Translation in Y 832.904 +- .461 M 833.213 +- .061 M
Translation in Z 293.969 +- .468 M 293.720 +- .061 M
Rotation X-AXIS -.0038 +- .0152" - -
Rotation Y-AXIS .0579 +- .0151" - -
Rotation Z-AXIS .0781 +- .0316" - -
Scale Factor -.017 +- .067 mmIkm - -
RMS of Transformation .0731 M .1365M

Tab. 4. Transformation parameters from the coordinates of the Final THAICA solution,
ITRF-94,1996.3 and the coordinates of the Thai GPS Net in the INDIAN75 Datum

55
From Tab. 5 it can be seen that these coordinates are generally in very good agreement with
an Helmert transformation rms of6 cm if we consider a shift of the WGS84 versus the l1RF-
94,1996.3 of about 60 cm in Z-direction and about 20 cm in X-direction.

Parameters Value of Standard

Parameters Deviations
Translation in X: 0.198 +- .027M
Translation in Y: 0.034 +- .027M
Translation in Z: - 0.588 +- .027M
RMS ofTranstOrrnation +- .0607M

Tab. 5. Translation of Malaysian coordinates, actual results versus DMA results (WGS-84)

6. SUMMARY AND CONCLUSIONS

GPS campaigns were carried out in order to connect the Thai and Malay national networks
to the ITRF global reference frame, to establish the zero order network, to establish a new
geodetic surveying and mapping control network, to determine precise transformation
parameters, and to improve the accuracy of the existing GPS network. The final coordinates
for the 10 new ITRF stations are given in the ITRF-94 (epoch 1996.3) reference frame and
have an accuracy which meets the objectives of the GEODYSSEA project. The precision of
the final coordinates is at the order of better than 3 cm. The results so far are sufficiently
accurate to establish an ITRF-referenced datum for all practical GIS, surveying and naviga-
tion purposes. For Thailand, RTSD presently prepares a new adjustment of the national GPS
network with about 530 stations using the stations of this study as fiducials. Similar investi-
gations are underway for Malaysia.

A second repetition of the GEODYSSEA GPS campaign is planned which will then yield
improved estimates of the velocity field in the area. This will allow for the monitoring of the
time dependent changes in the ITRF coordinates. For Malaysia the transition to an active
reference system by use of a network of 13 permanently tracking GPS stations and their
analysis is planned within the next year. Further works on the complete datum definition will
also have to consider the height problem. Therefore, geoid computations will have to be
improved in order to allow the complete transformation of all types of terrestrial data.

7. References

Ambrosius, B. et. al. (1997). The Final Combined Solutionsjor the GEODYSSEA94/96
Campaigns. GEODYSSEA Final Report, Brussels, 1997 (in press)
Becker, M., P. Neumaier, E. Reinhart, Dato' Abd. Majid, Samad Abu, C. Boonphakdee
(1997) : Results oj the MALAYSIA 94/96 GPS-Campaign Processing at IfAG.
GEODYSSEA Final Report, Brussels, 1997 (in press).
Mingsamon, S., C. Boonphakdee, M. Becker, P. Neumaier, E. Reinhart, H. Seeger, (1997):
Final Results oj the THAICA 94/96 GPS Campaigns. GEODYSSEA Final Report,
Brussels, 1997 (in press)
Gurtner, W. (1993). The Use oj IGS Productsjor Densification oj Regional Local
Networks. Report on the EUREF Symposium, Budapest, May 1993.

56
 THE ITRF96 REALIZATION OF THE INTERNATIONAL
TERRESTRIAL REFERENCE SYSTEM

Claude Boucher, Zuheir Altamimi, Patrick Sillard

Institut Geographique National
ENSGILAREG
6-8 Avenue Blaise Pascal
Cite Descartes, Champs-sur-Marne
77455 Marne-la-Vallee, FRANCE
E-mail: [email protected]@[email protected]

Introduction

As part of the realizations of the International Terrestrial Reference System (ITRS), a new
complete solution (called ITRF-C 1) has been proposed to be the ITRF conventional frame
for several years, and superseding ITRF94.
The analysis of the data collected in 1996 to obtain this solution revealed some inade-
quacy to insure the robustness and the quality required for ITRF-Cl. Moreover, this fact led
to the need of more refined specifications. The IERS Working Group on the ITRF Datum was
established to develop specifications for ITRF and in particular ITRF-C 1, for which a final
conclusion has not yet been reached.
In the mean time, the current ITRF in use, namely the ITRF94, based on data up to the end
of 1994 requires updates and improvements, considering in particular that in two years, the
observations extend the time interval for determining improved velocities, particularly from
GPS and DORIS.
Consequently an ITRF96 solution was decided, combining individual solutions coming
from those received in 1997 as well as some past data.

Input Data

The solutions provided by the IERS analysis centers and selected for the ITRF96 analysis are
4 VLBI , 2 SLR , 8 GPS and 3 DORIS solutions.
As an improvement of the use of the local ties, all the eccentricities of colocated sites were
converted into a complete set of positions for each site, provided in SINEX format. Each SI-
NEX file reflects correlations between the cartesian components of the points within each site.

Data analysis

The current strategy adopted for Terrestrial Reference Frame comparison/combination ana-
lysis is twofold: simultaneous combination of positions and velocities using full variance/
covariance matrices; rigorous weighting scheme based on the analysis and estimation of the
variance components using Helmert method.
 The data analysis performed in view of the ITRF96 establishment are mainly: comparison
of the individual solutions with ITRF94, combination of the solutions within each technique
and a global combination of all the solutions together with the local ties of colocated stations.
Among those selected for the ITRF96 combination, each individual solution was com-
pared to the ITRF94 in order in one hand to estimate the transformation parameters of the
system attached to the solution with respect to ITRF94 and, on the other hand, the level of
agreement with the ITRF94 values.
In order to assess the relative quality of the individual solutions, independently from the
influence of local ties, a combination within each technique was also performed. Matrix Sca-
ling Factors have been rigorously estimated during the combined adjustment of the solutions.

The ITRF96 global combination is achieved with the following properties:

- It includes the 17 selected space

geodetic solutions provided by
the IERS analysis centers and 70
constructed SINEX files containing
positions and covariances, derived
from local ties.
, 50'
.. . .. . . . .-•
'90',~
BO~'2~
'0·:=:2~
'O·~2~
70·:=:300
~·~330
~·=O:.'!"·~30~'=:60~'~8~
O' ="2~
O· -"50':7·~'BO'

- The reference frame definition (origin, scale, orientation and time evolution) of the
combination is achieved in such a way that ITRF96 is in the same system as the ITRF94.

- Velocities are constrained to be the same for all points within each site.

Figure 1 shows the coverage of the 290 sites of the ITRF96. The position formal errors at
epoch 1993.0 plotted in Figure 2 demonstrate an improvement of the ITRF96 with respect to
ITRF94.
The final results of the ITRF96 will be distributed in fall 1997 by the Terrestrial Reference
Frame Section of the IERS Central Bureau.

Figure 2: Position formal errors

00 BO~--~---,--~----,---~--~------~---- __~
c
o
:g 60 ITRF94
Vi
(; 40
L

E 20
Q)

:J
z oL-__~~~~LW~~~~~~~~~~=D~~~~
o B 10
00 BO~--~---,--~----.---~--~------~---- __~
c
o
:g 60 ITRF96
Vi
'0 40

4 6 B 10

Formal errors at 93 .0 (em)

58
 MODERNIZING THE REFERENCE FRAMEWORK
FOR CANADA'S MARITIME PROVINCES

Roger J. Gaudet
New Brunswick Geographic Information Corporation
985 College Hill Road
Fredericton, New Brunswick, Canada E3B 5Hl

Abstract

The three Maritime Provinces have been managing a geographical reference framework
for the past 40 years. During that period a large number of spatial data sets have been
created related to this reference framework. There is no doubt that the existing reference
framework "Average Terrestrial System of 1977 (ATS77)" served its purpose well.
However, without a framework in place which is consistent with the Global Positioning
System (GPS), the technology of the next decade, it is likely that many future positioning
applications will never be tied to a consistent reference framework. The result will be
incompatibility between data sets collected by different individuals and agencies, and will
result in additional costs to the users of the data.
The three Maritime Provinces co-sponsored a study of the best way to prepare for the
reference framework technology of the future. This study's recommendations imply
profound changes for the way the provincial reference framework could be handled. This
presentation looks at the history of the provincial framework, the establishment of a GPS
based reference framework tied to the Canadian Space Reference System, the migration
from the existing framework to the new framework, the human resources and cost
implications associated with these actions and the impact on the user community.

History

The primary geodetic network in the Maritime Provinces was established between 1908
and 1927 as part of the North American primary network. Until 1976 the adjustments of
the primary geodetic network in the Maritime Provinces was done on the NAD27 datum.
In October 1977, a national readjustment of the Canadian framework to a geocentric
datum was done by the federal government and became the foundation for the Maritime
Provinces' datum identified as the Average Terrestrial System 1977 (ATS77). The
primary geodetic network in the Maritime Provinces was also part of the federal
government's national readjustment by the Geodetic Survey Division (GSD) of Natural
Resources Canada to a geocentric datum in July 1989 which was defined as the North
American Datum 1983 (NAD83). However, the Maritime Provinces never adopted
NAD83 because the difference in coordinates between the ATS77 and NAD83 was on the
average 4.5 metres and the user community did not feel that the cost of going through
another redefinition at the time was justified.
Impetus for the densification of control came as late as 1959 from the provincial
government of New Brunswick with the intent of providing a survey coordinate system,
originally for cadastral purposes and later for integrated multipurpose use [Roberts 1960,
1966]. The feasibility of undertaking control densification has been brought about by the
availability of the tellurometer and the geodimeter [Konecny & Roberts 1963].
The program which was cautiously started in 1959 saw 40,887 permanent survey
monuments established in the Maritime Provinces during the next 18 years. These
monuments were distributed throughout the Maritime Provinces in an interconnected
network of traverses, with spacing of one marker every two kilometres in rural areas and
one marker every 300 to 500 metres in urban areas. Monuments were placed along public
rights-of-ways. The coordinate values for the secondary networks were obtained by
utilizing a method of nonrigorous condition equations and computer programs developed
by A.W. McLaughlin and others at the New Brunswick Department of Natural
Resources. The adjustment of the secondary network was based on fixing the values of
the primary geodetic framework on the 1927 datum. The nominal relative accuracy of a
second order survey monument was 1:15 000.
Between 1977 and 1979 the readjustment of the maritime network was done by the
Land Registration and Information Service based on the ATS77 framework. This resulted
in a homogeneous set of positions across the Maritimes. The nominal relative accuracy
of a second order survey monument was 1:20 000.
In 1977 a maintenance program was implemented to replace and densify the secondary
networks and over 15,000 new survey monuments were added to the network between
1977 and 1997.
In 1992 the Maritime Provinces recognized the need to review the effectiveness of the
existing survey control framework. They hired A. C. Hamilton and Associates to do a
review of the existing survey control framework and to recommend a strategy for the
provision of an appropriate survey control framework which would serve the region over
a long term period, i.e., 20 years or more.
The Report of the Task Force on Control Surveys in the Maritime Provinces (Hamilton
& Doig, 1993) recommended profound changes in the way the provincial reference
framework could be handled. The control survey framework has been a sound investment
for the provinces, however, it is clear that GPS is now sufficiently "operational" to be a
cost-effective alternative in providing survey control.
The user community recognized that conversion to the new NAD83 is inevitable, but it
reminded the Task Force quite emphatically that conversion to a new reference
framework is a costly and time-consuming exercise for them and that whatever is adopted
should be designed to serve for many decades. The user community also requested that
one definitive set of transformation vectors be prepared so that everyone would be able to
make the conversion to the new framework in the same way. Regrettably, a closer
examination of the options for a new reference framework revealed that it is possible to
design a new framework that will have a definitive set of transformation vectors or to

60
design a new framework that will last for several decades. It is not possible to do both.
If the provinces opted to upgrade the existing network with a reasonable number, say
200 GPS observation points, do a readjustment, and do the conversion to the NAD83
datum, a definitive set of transformation vectors could be derived and the new framework
could be ready for adoption in two or three years. As soon as this scenario is adopted, it
is predictable that there would be complaints about its accuracy from those who will be
using GPS for positioning their surveys. These complaints would increase with time and
would continue until a new GPS-based reference framework was finally established and
adopted. On the other hand, if the provinces established a GPS-based High Precision
Network in the next couple of years, they would have a framework expected to last for
several decades but it will require several years to prepare a definitive set of
transformation vectors for it.
The Task Force recommended the latter option. Specifically, it recommended that the
provinces make a request to the GSD of Energy, Mines and Resources Canada, for the
establishment of a High Precision Network (HPN) of some 10 to 12 points as soon as
possible. A program to establish an additional 20 to 30 regional High Precision points
should be initiated either concurrent with the GSD's program or at a later time.

Maritime Provinces High Precision Network

In 1993, GSD established, as part of the Canadian Space Reference System (CSRS), an
array of 12 points in the Maritimes at a nominal spacing of200 kilometres. In addition to
the 12 federal Canadian Base Network (CBN) points, the Maritime Provinces agreed to
simultaneously establish another 32 points within the region providing a network with a
nominal spacing of 70 kilometres. These 44 points were defined as the Maritime
Provinces High Precision Network (MPHPN) and will provide the users of the provincial
framework an effective support system for today's technologies. Cooperation between
the federal and provincial governments in the establishment of these points has ensured
that the network will not only be highly accurate but also highly accessible and
completely consistent.
As part of the cooperative federal-provincial project to establish the CBN, the Maritime
Provinces agreed to perform site reconnaissance for the federal government in parallel
with reconnaissance for their own stations. In choosing the sites for both the federal and
provincial CBN stations, the Maritime Provinces followed specifications prepared in
1993 by the GSD. This document outlines seven criteria to be considered in selecting a
site for an ACP or an HPN point: obstructions and interference, future land development
and permanence, ground stability, accessibility, security, power and communications.
The next aspect considered was the stability of the monumentation. Concrete pillars
were built for all CBN points as per the GSD's specifications, including an evaluation of
the ground stability at the site. A mixture of monumentation was used for provincial
sites; in New Brunswick all sites are monumented with concrete pillars as per the federal
specifications, in Nova Scotia and Prince Edward Island both pillars and conventional
monument flush to the ground were utilized.
The observation scenarios for the GPS field collection were planned jointly by the GSD
and Maritime personnel to optimize observing and moving times for both parties. All

61
personnel participating in the federal and provincial field operations were provided with
an intensive one week training program, covering basic GPS theory, field procedures,
downloading procedures and data backup procedures.
The GSD occupied all federal CBN points for a minimum of three 24 hour sessions at a
data collection rate of 30 seconds. The Active Control Point (ACP) in Saint John's,
Newfoundland and Algonquin, Ontario were collecting data at the same rate during the
period of the campaign. Turbo-Rogue receivers were used on federal points. The
provincial stations were occupied for a minimum of three sessions of eight hours, at a
data collection rate of 15 seconds. In order to take advantage of the 24 hour federal
sessions, two eight hour provincial sessions were scheduled within each federal session.
ASHTECH Z-12 receivers were used for the observations at provincial sites.
As part of the same cooperative arrangement with GSD the GPS data reduction,
adjustment and analysis for the Maritime network was done by their staff. Bemese GPS
processing software was utilized in the data reduction task and GHOST for network
adjustment and analysis software. The objective of the Maritime Provinces was to
establish a regional high precision network with an average relative (line) accuracy of
1 ppm for the network. This objective was actually exceeded, as the average 3-D ppm
values, based on the 3-D error ellipses (at 95%) of 475 lines, from the minimum
constraint adjustment was .252 ppm.
To assist in the management of the MPHPN a set of guidelines was developed.
Guidelines for Managing a GPS Based Control System in the Maritime Provinces,
Version 1.0 (Geoplan Consultants Inc. in association with Gillis Survey Systems Inc.
1996) addresses the geodetic datum to be used in supporting the control system, accuracy
standards, GPS observation procedures, GPS data processing and adjustments, data
storage, and the use of validation networks. This document is available on the Web at
http://gauss.gge.unb.ca/gps.guidelines.html for your viewing.

Migration from Old to New Framework

During the past two years the Maritime Provinces have been densifying the MPHPN
(Figure 1) in order to develop a set of transformation tools for the conversion of huge
databases of geographic data collected during the past 20 years. The future densification
effort will also benefit the land surveying community in having closer access to the
network. The densification project was done according to the guidelines manual
previously outlined and used existing provincial control stations to determine the
relationship between the two frameworks. The nominal spacing is 30 kilometres with an
average relative (line) accuracy of 1 ppm for the network.
As part of the conversion process from NAD27 to NAD83, GSD in cooperation with
the provincial government developed the NTv2 software and a grid shift file for
converting geographic data between the two datums. The governments of the Maritime
Provinces have adopted the NTv2 software as the standard for converting all GIS
databases and are presently developing with GSD a regional grid shift file to convert
between ATS77 and NAD83 as realized through the CSRS. This will allow for
consistency and reliability in converting existing large databases of geographic data from
the present ATS network to the new MPHPN. Some of the GPS and GIS software

62
 • CBNPoInt
.. HPNPoint
8
t.
HPN DenslflClltlon Point
Proposed HPN Densiflcatlon Point

"'lj
1-'-
'@
Ii
(l)
0>
W
....

Maritime Provinces v
High Precision Network
........,,_~onc~1
I
I
distributors have already incorporated the NTv2 software as part of their products.
The official proclamation of the HPN for each province will be different. At present the
provinces of New Brunswick and Prince Edward Island have scheduled their
proclamation date for 1998 while the province of Nova Scotia will be ready after the year
2000.

Resources and Cost

During the last 10 years, the governments of the three Maritime Provinces have focused
on reducing the cost of their operations. Every government operation was reviewed with
the intent of identifying the best method and most economical way of delivering the
services to the public. One of the primary objective was to look at the present and future
technology that would allow us to deliver these services effectively and at an affordable
cost. In 1973, the Maritime Provinces were spending 70% of the budget allocated to the
maintenance of the existing survey control framework on employees' salary. All work
related to the maintenance of the framework was done in-house. Very little money was
spent on contract services and new projects. After consultation with the user community,
it was agreed to reduce our maintenance program and re-orient that portion of the reduced
budget towards the implementation of the MPHPN. It was also agreed that the
maintenance of the existing framework would stop once the new MPHPN was complete.
A program of staff reduction was implemented with the stipulation that the budget money
saved from the staff cutback would be transferred to contract services and the purchase of
GPS technology. Agreements were signed between the maritime governments to work
together in the acquisition of GPS expertise, implementation schedule, rental of GPS
equipment and purchase of GPS software. During the past four years a saving of over
$500,000 was realized by the maritime governments by working together in the
implementation of the MPHPN. The province of New Brunswick has reduced its survey
control staff from seven (7) to three (3) person year during that period and will contract
out the services to the private sector as the need arises.
The implementation of the MPHPN has been accepted well by the user communities.
Workshops and seminars are being prepared and scheduled during the coming year to
make all users aware of the new reference framework, to develop the appropriate
transformation methodology and procedures in moving from one datum to another. The
land surveyor community is presently working in the development of GPS standards for
cadastral surveying.

Conclusion

The governments of the Maritime Provinces and the user communities in the region have
always realized the importance of a survey control framework as the foundation for the
development and maintenance of spacial data sets. In this fast moving world of
technological innovation, it is imperative that the infrastructure be in place to support the
modem technology which will allow us to manage our resources and deliver programs
more effectively and economically.

64
 The Maritime Provinces have moved again in the forefront by delivering a control
survey framework which will serve the citizens of the provinces well into the next
century.

References

Geoplan Consultants Inc. in association with Gillis Survey Systems Inc. - Guidelines for
Managing a GPS Based Control System in the Maritime Provinces, Version 1.0, March
1996.
Hamilton, A C. and Doig J. F. - Report ofthe Task Force on Control Surveys in the
Maritime Provinces, prepared for the New Brunswick Geographic Information
Corporation, Fredericton, New Brunswick, March 1993.
Konecny, G. and Roberts, W. F. - Electronic Surveys in New Brunswick, Journal of the
Surveying and Mapping Division, American Society of Civil Engineers, Washington,
D.C., October 1963.
Roberts, W. F. - The Need for a Coordinate System of Survey Control and Title
Registration in New Brunswick, The Canadian Surveyor, Vol. XV, No.5, Ottawa 1960.
Integrated Surveys in New Brunswick, The Canadian Surveyor, Vol. XX, No.2,
Ottawa, 1966.

65
 PERMANENT GPS TRACKING NETWORK INCLUDING
THE MEDITERRANEAN AREA

G. Weber, M. Becker, P. Franke, W. Schliiter, H. Seeger

Bundesamt ftir Kartographie und Geodasie (BKG), Frankfurt a. M., Germany

ABSTRACT

In 1995, a proposal was developed within WEGENER to install and operate a network of
Permanent GPS Stations around the Mediterranean area and adjacent areas. Established with
the intention of determining the velocity of tectonic plate motions, the network has to cover
the region between the Azores in the west and the Caspian Sea in the east. At the BKG, a
part of this network is processed on a daily basis. Loosely constrained solutions are
generated and analyzed to investigate geodynamic effects and their influence on geodetic
reference systems in the Mediterranean area.

The time series of coordinates available today cover a period of more than two years.
Regional velocities are compared with NUVEL model velocities. The time series are
analyzed for short and medium scale variations. Details of the time series analysis are given.
Model calculations are presented and their significance for GPS data processing is
discussed. The relationship between the obtained results and tidal deformation, and other
possible effects, is being studied.

THE EUROPEAN BACKGROUND AND WEGENER

The most precise applications of GPS today make use of Permanent GPS Stations. Apart
from the global framework of the International GPS Service for Geodynamics (IGS),
additional Permanent GPS Arrays have been established over the last few years mainly
through international cooperations. The objectives are to derive e.g.
• accurate GPS satellite ephemeris and earth rotation parameters,
• site coordinates/velocities and information,
• ionospheric and tropospheric delay information.
The accuracy of the resulting products is suitable to support the realization and continuous
improvement of the International Terrestrial Reference Frame (ITRF) and the monitoring of
changes of the solid and fluid earth.

Based on intensive global cooperation between various national and international

authorities, a well coordinated world wide usage of Permanent GPS Arrays has been
established in connection with the activities of the IGS. Starting on a national level, data
from local or regional Permanent GPS Arrays are observed and processed mainly for the
establishment and monitoring of three dimensional Reference Systems. In Europe, results
are combined within the lAG Sub-Commission for the European Reference Frame (EUREF)
The BKG introduces it's southern European GPS Permanent activities as part of the
WEGENER project.

BKG's CURRENT CONTRIBUTION TO "WEGENER-PERMANENT"

Like other IGS Regional Analysis Centers, the BKG routinely processes a Permanent GPS
Array [Weber et. al. 1996]. As of today, a selection of 29 stations are included in a daily
analysis, covering a substantial part of Europe and the Mediterranean (see Fig. O. GPS
observations of these permanent stations are uploaded to Frankfurt on a daily basis.

The data analysis is carried out using the Bernese GPS Software Version 4.0, following a
processing strategy optimised to determine site movements from a continuous quasi on-line
monitoring. The estimated parameters are station coordinates and tropospheric zenith delays
only. Ambiguities are resolved using a baseline oriented Quasi-Ionosphere-Free (QIF)
strategy.

The activities started in January 1995 with ten IGS stations. Additional stations have been
introduced during the following months. The first stations of the German Permanent GPS
Network (DREF-Permanent) were included from March 1997. A total number of about
10,000 station coordinate values have been determined. Currently, daily solutions for two
and a half years are available. About 5 per cent of daily solutions are missing, mainly due to
on-site observation problems.

M illim oter

20

Jan 9S Aug 97

Fig. 2. Time Series of Coordinate Residuals, Jan 95 to Aug 97, Station Wettzell

Solutions are carried out as "loosely constrained" through the introduction of a priori
coordinates with unique standard deviations of ± 1 m. The applied HELMERT
transformations (daily solutions verso a priori solution) lead to time series of residual
coordinate components. As an example, these time series are shown in Fig. 2 and 3 for the
Permanent GPS Stations Wettzell and Ankara. The daily repeatability varies between ± 3
rom and ± 8 mm for horizontal components and ± 8 mm to ± 20 mm for height components.

67
to derive a continental wide combined solution [Bruyninx et. al. 1997]. EUREF transfers its
European solution to the IGS, which is responsible for deriving a global solution. Finally,
these GPS-based results are combined by the International Earth Rotation Service (IERS)
with data from various other techniques like Satellite Laser Ranging or Very Long Baseline
Interferometry for monitoring the earth rotation, defining the ITRF, etc.

The Bundesamt fUr Geodasie und Kartographie (BKG) currently contributes to global
networks through the operation of Permanent GPS Stations in Wettzell (Germany),
O'Higgins (Antarctica), Lhasa (Tibet), Reykjavik and Hofen (Iceland), Ankara (Turkey),
Sofia (Bulgaria) and Nicosia (Cyprus). All data are made available through the IGS
Regional Data Center (RDC) at the BKG in Frankfurt.

For regional and local applications the existing networks have to be densified through
additional Permanent GPS Stations. Regional or local GPS networks, connected to the
global system, provide more detailed information on local kinematics and sea level
fluctuations as well as atmospheric and ionospheric influences on observations. In Europe,
they will also contribute to the objectives of projects such as WEGENER (Working Group
of European Geo-Scientists for the Establishment of Networks for European Research) and
SELF (Sea Level Fluctuation in the Mediterranean) or help to satisfy national Differential
GPS (DGPS) requirements.

Fig. 1. European GPS Permanent Stations daily processed by BKG

During the WEGENER Conference held in Bologna in October 1995, it was agreed to
prepare proposals for a permanent GPS tracking network spanning the area from the Azores
along both sides of the Mediterranean Sea into the area of the Caspian Sea [SchWter 1996].

68
Millimeter

20

-20

20

-20

20

-20

Jan 95 Aug 97

Fig. 3. Time Series of Coordinate Residuals, Jan 95 to Aug 97, Station Ankara

Meter

0.4

North

o.o~~~Am~~~~~~~~~~hA~~~~~~JU~~~~~
0.4

0.4

O.O....!...-----1H¥+-------....!....---------------l------

Jan 95 Aug 97

Fig. 4. Time Series of Heimert Transformation Parameters (daily verso a priori)

69
 The time series of translation parameters as derived from daily applied HELMERT
transformations (see Fig. 4) indicate the overall movement of the barycenter of the regional
network. The linear approximation of daily station coordinate residuals, added to the linear
approximation of the barrycenters translation, permits the estimation of station velocities as
shown in Tab. 1 and Fig. 5. Due to the limited regional extension of the network and the
intentionally unconstrained procedure, it is not possible to derive full absolute site velocity
vectors (see NUVEL model in Fig. 5).

70

~
Black: BKG
Grey: NUVEL

60 ~

'"
....] ~
t;
...;

50

40
-20 0 20 40
Longitude 20 mmly

Fig. 5. Site Velocities from BKG Permanent GPS Observations verso NUVEL

Tab. 1. Relative Site Velocities and Velocity of Barycenter

North East
Site Days RmsFit vel Rms Fit vel
[mm] [mmly] [mm] [mmly]
ANKR 639 ±0.5 7.2 ±0.9 14.6
BRUS 917 ±0.3 0.1 ±0.3 1.7
ORAZ 903 ±0.3 1.4 ±OA -0.8
HERS 895 ±OA -2.5 ±0.7 0.9
KOSO 891 ±0.2 -0.6 ±0.3 1A
MADR 858 ±0.9 5.5 ± 1.2 -6.7
MATE 907 ±0.5 -3.8 ±0.6 1.2
METS 872 ±OA 1.6 ±OA -1.6
aNSA 905 ±0.3 0.1 ±0.3 OA
POTS 898 ±0.3 0.1 ±0.3 -0.6
REYK 619 ±0.6 -10.7 ±0.7 13.3
WTZR 870 ±0.2 0.6 ±0.4 0.9
ZWEN 767 ±0.6 0.4 ±0.7 -1.0
HFLK 738 ±0.4 1.4 ±0.5 2.1
KIRU 899 ±0.6 -1.5 ±0.6 -1.0
ZIMM 905 ±0.3 0.2 ±0.4 1.6

Barycenter ± 26.3 -23.6 ±26.2 37.7

70
Fig. 6 shows a Fourier Spectrum for the time series of HELMERT transformation
parameters. An interesting effect is the amplitude of about 5 mm for all stations at a wave
length of 13.6 days. The reason for this effect might be a missing ocean loading model in the
Bernese GPS Software, because the Mf wave group has exactly the same wave length. The
Software is currently being modified to account for this effect.

North HELMERT Translation Parameters

East HELMERT Translation Parameters
UP HELMERT Translation Parameters

'i.
e 13,6 days
~ 2
.=!. /
i
:1:
I .,

~
-..:

0 0.05 0. 10 0.15
Frequency [llday]

Fig. 6. Fourier Spectrum from Time Series of daily Helmert Transformation Parameters

A part of the daily solutions is the estimation of tropospheric zenith delays every two hours.
Results from these estimations (see time series for selected stations in Fig. 7) should be
strongly correlated with meteorological observation conditions. The BKG is intends to
investigate these correlations for further model improvements and/or a possible forecast of
meteorological phenomena.

Meter

2.4

2.2

Sep 96 Aug 97

Fig. 7. Time Series of estimated Tropospheric Zenith Delays, Step Width 2 hours

71
CONCLUSIONS

Developments in the field of data reduction and analysis software have proceeded so far that
automatic data handling and processing is reality today. Daily data collected from a set of
stations can be processed completely automatic, generating daily solutions. The internal
accuracy achievable for a daily solution is better than ± 0.5 cm for a horizontal position and
± 1.5 cm for height components. The precision and stability of these results encourage the
inclusion of additional sites for projects such as WEGENER.

The key to the success of permanent networks is the availability of communication links
between observation sites, operation centers, data centers and analysis centers. A high speed
daily data transfer is required. Developments in telecommunication will allow the inclusion
of more sites into the network at moderate additional cost.

In order to support the WEGENER objectives, additional stations are required in the area of
the Azores, on the North-African plate e.g. in Marokko, Tunisia, on the Arabian plate
(Dubai) and in the area east of the Caspian Sea (Ashgabat). The BGK is prepared to support
these activities through operating an increasing number of sites as long as the on-site
infrastructure will guarantee daily data transfer to the analysis center.

4. REFERENCES

Bruyninx, c., J. Dousa, W. Ehrnsperger, N. Fachbach, G. Stangl, J. Johansson, C. Ferraro,

M. Fermi, A. Nardi, C. Sciarretta, F. Vespe, M. Figurski, M. Piraszewski, J.
Rogowski, T.A. Springer, G. Beutler, M. Rothacher, S. Schaer, G. Weber and A.
Wiget (1997): The EUREF Associate Analysis Center. In: International GPS Service
for Geodynamics, Annual Report 1996, Pasadena.
Schliiter, W., H. Seeger, G. Weber (1996): Proposalfor a WEGENER Permanent GPS
Tracking Network. 7th General Assembly of the WEGENER Project, Vila Nova de
Gaia, Porto, Portugal, 3-4 June 1996.
Weber, G., W. Schluter, H. Seeger and P. Franke (1996): IfAG Contribution to a GPS
Permanent Network in Europe. In: Report on the Symposium of the lAG Sub-
Commission for the European Reference Frame (EUREF), Ankara, 22 - 25 May,
1996, BEK, Astr. Geod. Arbeiten, Heft 75.

72
 THE BRAZILIAN NETWORK FOR CONTINUOUS MONITORING
OF GPS (RBMC): OPERATION AND PRODUCTS

Luiz Paulo Souto Fortes

Roberto Teixeira Luz
Katia Duarte Pereira
Sonia Maria Alves Costa
Departamento de Geodesia
Funda-.;ao Instituto Brasileiro de Geografia e Estatistica - IBGE
Av. Brasil 15671, Parada de Lucas, Rio de Janeiro, RJ, 21241-051
Fax: 55-21-481 2747, Brazil

Denizar Blitzkow
Departamento de Engenharia de Transportes
Escola Politecnica da Universidade de Sao Paulo - EPUSP
Caixa Postal 61548, Sao Paulo, SP, 05424-970
Fax: 55-11-8185716, Brazil

ABSTRACT

The Brazilian Network for Continuous Monitoring of GPS (RBMC) is an active geodetic
reference network. In this paper its automatic operation and products are described during
the first nine months of functioning. It consists of nine permanent GPS stations established
in Brazil, in cooperation with many groups. Two of the nine stations (Brasilia and
Fortaleza) also belong to the International GPS Service for Geodynamics (IGS), while the
remaining 7 are integrated into this service by contributing the corresponding data to the
SIR IGS Regional Network Associate Analysis Center (RNAAC) located at DGFI
(Deutsches Geodaetisches Forschungsinstitut), Germany. The future activities include the
densification of the network in the Amazon region (six stations) and in Northeastern and
Southern Brazil (one station in each area).

INTRODUCTION

The Brazilian Network for Continuous Monitoring of GPS (RBMC) is one of the first
active geodetic reference networks in South America. IBGE (Funda<;:ao Instituto Brasileiro
de Geografia e Estatfstica), as the institution responsible for geodetic activities in Brazil,
presented the first proposal for the project in 1991. Since then, while IBGE was looking
for funds to establish the network, the project has experienced some changes, in order to
keep up with the state-of-art of GPS technology. A description of the status of the project
at different times can be found in FORTES (1993), FORTES (1996a) and
FORTES (1996b).
 RBMC now consists of nine permanent GPS stations, two of which are IGS stations:
Brasflia (BRAZ) and Fortaleza (FORT), established in cooperation with JPL (Jet
Propulsion Laboratory - NASA) and INPE (lnstituto Nacional de Pesquisas
Espaciais)IUSNGS (United States National Geodetic Survey), respectively. The remaining
seven stations have been established by IBGE in cooperation with EPUSP (Escola
Politecnica da Universidade de Sao Paulo) and other Brazilian institutions, and sponsored
by FNMA (Fundo Nacional do Meio Ambiente). Table 1 lists the names and the four-digit
codes of the stations, their approximate coordinates and the date of commencement of
operations. Those seven stations will be referred to in this paper as "IGS-SIR" ones, as
IBGE is contributing to IGS densification, sending the corresponding data to the SIR IGS
Regional Network Associate Analysis Center (RNAAC), located at DGFI (Deutsches
Geodaetisches Forschungsinstitut), Germany.
It is important to emphasize that, as all RBMC stations belong to the SIRGAS Reference
Network (IBGE, 1997), all GPS positioning in Brazil using them as reference will be
automatically integrated into that new system.

Figure 1 shows the current RBMC configuration, with two IGS stations, the seven IGS-
SIR ones, and future stations.

TABLE 1: Current RBMC stations

Beginning of
Station 4-digit code Latitude Longitude
Operation

Fortaleza (IGS) FORT -03°53' -38°26' 13.May.1993

Brasflia (IGS) BRAZ -15°57' -47°53' 3.Mar.1995
Curitiba PARA -25°27' -49°14' 13.Dec.1996
Presidente Prudente UEPP -22°07' -51 °25' 18.Dec.1996
Born Jesus da Lapa BOMJ -13°15' -43°25' 18.Feb.1997
Manaus MANA -03°07' -60°03' 28.Apr.1997
Vi<;osa VICO -20°46' -42°52' 22.May.1997
Cuiab:i CUIB -15°33' -56°04' 18.Jun.1997
lmperatriz IMPZ -05°30' -47°30' 18.Ju1.1997

STATION DESCRIPTION

Each station is materialized by a centering forced device, specially designed by IBGE for
RBMC, which is set up in a very stable structure of the corresponding building. In
addition, the following were the criteria for station selection: no obstructions> 10 degrees
above the horizon; no multipath surfaces close to the GPS antenna; no electromagnetic
sources in the range of the GPS frequencies (1.2 to 1.6 Ghz); a restrict public access area;
availability of an office close to the antenna location with continuous power supply and
telephone facility for data communication.

74
 The IGS-SIR stations are equipped with TRIMBLE SSI GPS receivers, each with 9 Ll
and L2 full wavelength channels, able to also measure pseudo-range on Ll and L2;
immunity to radio interference; choke ring antenna; 10Mb of memory; and backup
batteries for 16 hours of operation. In addition, there is a PC computer which is used for
local and remote operation of the station (see next item) and an uninterrupted power
supply (UPS). Detailed specifications of the station can be found in FORTES (1997).
Figure 2 shows a diagram of the station components and the connections between them.

FIGURE 1: RBMC stations. Center of a gray circle represents a RBMC station. Center of
a transparent circle represents a station proposed for future densification. Each circle has a
500 km radius

OPERATION

The IGS-SIR stations of RBMC are operated in an automatic way, based on procedures
developed at IBGE, while the IGS stations (BRAZ and FORT) follow the procedures
established by JPL and USGS, respectively.

75
 •
I I

r-----:
12v I.SA Ext.
'-
<1)
'-
<1)
(in charge)
_ __ _ _ _ _ _ __ __ , I

~ ~ ... ______ 1
0 0 Power
0.. 0..
Batteries
(in use)
CJ
c:::J 2:
o
E3 u

D
UPS Microcomputer GPS Receiver

FIGURE 2: Diagram of the station components and the connections

Each IGS-SIR station is operated according to the following tracking parameters:

• Observation rate: IS seconds;
• Elevation mask: 10 degrees;
• Session length: 23 hS9 m (from 00 hOI m UT to 24 hO~ mUT);
• Minimum number of satellites to be tracked: 1.
The automatic operation procedure for the network stations is:
1. Closing the previous observation session file and opening the new one at 00 hO~ mUT;
2. Executing a batch file on the station microcomputer at OShOOm UT (02 hOOm Rio de
Janeiro time) to accomplish the following tasks:
D to download data from the receiver memory to the computer hard disk, erasing
older files (in order to make room for new ones);
D to transform it from binary to RINEX2 format;
D to accomplish preliminary quality check, using the QC program (GURTNER,
1994);
D to compress the files using PKZIP program;
3. Downloading the ZIP file, created by the previous procedure, from the station
computer to the RBMC control center, located at IBGE Department of Geodesy's
headquarters, in Rio de Janeiro, at 06hOO mUT (03 hOO mRio de Janeiro time). This is carried
out using the communication software pcANYWHERE32 for Windows9S combined with

76
telephone lines or Internet access, when available. The computer at the control center
connects to each station, sequentially, and commands the file transfer.
With the above procedure it is expected that all data collected during a day is available at
the control center in the next morning, before working hours. The operation also includes
the remote control of the stations, from the control center, in order to allow full access to
the OPS receiver.
Concerning the lOS stations, the RINEX observation files of BRAZ and FORT are
regularly downloaded from JPL and CDDIS (Crustal Dynamics Data Information System)
databases, respectively, and made compatible with the others in terms of name and
compression utility used.

PRODUCTS

The daily xxxxdddl.zip file of each lOS-SIR station is composed of the RINEX files
xxxxddd1.yyo, and xxxxddd1.yyn, and the QC summary file xxxxddd1.yys, where xxxx
is the four-digit identification of the station; ddd is the day of the year; and yy is the year.
Once in the control center, each ZIP file is stored on an Internet server for future use,
according to the directory structure: /usr/rbmc/yyyy/xxxx. For instance, the observations
for Manaus on day 217, 1997, are located at /usr/rbmc/1997/mana/mana2171.zip. In
addition, they are also copied to compact discs for backup purposes.

PROBLEMS

The RBMC operation has been very successful, despite its recent start. The only exception
occurs at the Imperatriz station (the last one set up), which is experiencing serious
communication problems. For the remaining six lOS-SIR stations, Table 2 shows the
percentage of data loss. The proposal is to provide data collected at each station during all
days of the year and the expectation is to get closer to 0% of data loss after the first year of
operation.
Another problem is related to the automatic file transfer from the stations to the control
center. In some cases, it fails due to the poor quality of the telephone lines in Brazil.
However, this does not imply in data loss, as the file stored in the local computer and in
the receiver memory (in the later case, for some days only) is then downloaded manually
to the control center during the working hours.

PERSPECTIVES AND CONCLUDING REMARKS

The next step of the project will involve the densification of the network in the Amazon
region (6 stations), and in the Northeast and South regions (one station in each) (see
Figure 1).

77
 It is also planned to install barometers and thermometers at each station, in order to
support meteorological applications, as suggested by Duan et al. (1996).
It can be inferred from what was stated on this paper that Brazil has now a modern active
geodetic reference network based on GPS.

TABLE 2 Percentage 0 fd ata oss for SIX

. IGS -SIR statIOns 0 f the RBMC network

PARA UEPP BOMJ MANA VICO CUIB TOTAL

number of
operating days 257 252 188 120 97 69 983
(by 27.Aug.97)

number of days 0 9 17 9 0 3 38
lost

% of data lost 0 4 9 8 0 4 4

REFERENCES

DUAN, J. et al.. GPS Meteorology: Direct Estimation of the Absolute Value of

Precipitable Water. Journal of Applied Meteorology. 1996, vo1.35, Nr.6, pp 830-838.
FORTES, L.P.S .. Brazilian Network for Continuous Monitoring of the Global Positioning
System - RBMC. International Association of Geodesy Symposia 109 on "Permanent
Satellite Tracking Networks for Geodesy and Geodynamics", Spring Verlag, Berlin,
pp 95-101, 1993.
FORTES, L.P.S .. Status of the Brazilian Network for Continuous Monitoring of GPS
(RBMC). International Association of Geodesy Symposia 115: GPS Trends in Precise
Terrestrial, Airborne, and Spaceborne Applications. Spring-Verlag, Berlin, pp 85-88,
1996a.
FORTES, L.P.S .. Establishment of the Brazilian Network for Continuous Monitoring of
GPS (RBMC). End-User Applications Volume of the Trimble Surveying & Mapping
Users Conference 1996 Proceedings. Trimble Survey & Mapping Division, Sunnyvale,
pp 69-72, 1996b.
FORTES, L.P.S .. Operacionaliza~ao da Rede Brasileira de Monitoramento Continuo do
Sistema GPS (RBMC). Master of Science's Thesis. Instituto Militar de Engenharia, Rio
de Janeiro, 1997.
GURTNER, W .. IGS Data Quality Checks. IGS Electronic Mail #817,1994.
IBGE. SIRGAS Final Report, Working Groups I and II. Rio de Janeiro, 1997.

78
 TOWARDS AN AUSTRIAN GPS REFERENCE NETWORK

R. Weber (Department of Theoretical Geodesy, TU-Vienna,

Gusshausstrasse 27-29,A-I040 Vienna, Austria)

Abstract

During the past years several activities were launched in Austria to establish a state-of-the-
art GPS-network which could serve as a basis for geodynamic applications. Recently, in
June 1996, an association of engineering consultants initiated a GPS-campaign which
covered the Austrian territory with 330 sites with a mean point-to-point distance of about
25 km. The establishment of such an accurate network (+/- 0.4 cm horizontal, +/- 0.8cm
height) will be advantageous to all kinds of surveying activities in the near future and
provides a considerable contribution to the densification of the ITRF94 reference frame.

The Institute of Theoretical Geodesy and Geophysics (ITGG) in Vienna is one of the
two analyses centers, which were appointed to process the GPS-network data. This paper
summarizes the computational efforts at ITGG and puts special emphasis on the
algorithms used for reference frame fixing in the analyses. Additionally, problems of
modelling phase center variations (seven different antenna types were used in the network)
are discussed.

Introduction

Starting in 1990 a densification of the Austrian part of the EUREF-network was realized
in several subcampaigns. Due to improvements in receiver technology as well as in orbit
and troposphere modelling a substantial increase in accuracy could especially be achieved
in 1994 and 1995 ( Erker et al., 1997). At that time the network (AGREF= Austrian
Geodynamic Reference Frame) comprised of about 80 stations with an accuracy of
O"xy = ±lcm in horizontal position and O"h =±2cm in height respectively.

In 1996 an association of engineering consultants initiated a further densification in order

to decrease the mean point-to-point distance to about 25 km. The resulting network called
AREF (=Austrian Reference Frame, see Figure 1) now consists of more than 330 sites
altogether including the former AGREF points. The accompanying field campaign was
organized in cooperation with the Federal Office of Metrology and Surveying. The
Institute of Theoretical Geodesy and Geophysics (University of Technology Vienna
ITGG) as well as the Department of Satellite Geodesy (Institute for Space Research,
IWFSG) were appointed to serve as Analyses Centers.

Figure 1: AREF- overall view

The campaign objectives as well as the data-processing guidelines can be summarized as

listed below:

a) AREF should represent a precise and homogeneous state of the art reference network
with an accuracy better than u =±lcm in position and u =±2cm in height.

b) The field campaign should be limited to 14 days and can be divided in seven 24-hours
sessions.

c) The data analysis should deal with not more than 3 hierarchical levels:

- Level A: Global and regional IGS-sites (8 sites, see Figure 2)

- Level B: Campaign-specific reference sites capable of permanent tracking over a
period of 14 days (13 sites, see Figure 2)
- Level C: The remaining points in the network (about 50 sites / session)

80
d) The analysis should be carried out by means of the most recent version (V4.0) of the
Bemese GPS Software (Rothacher, Mervart, 1996)

e) Orbits (Zumberge et aI., 1996) and estimates of the tropospheric delay at the IGS-sites
provided by IGS and CODE respectively have to be used.

Out of a great variety of aspects concerning the data processing, particularly

1) the choice of an appropriate reference frame and
2) the modelling of antenna phase variations
are emphasized in the subsequent paragraphs.

Choice of Reference Frame

In order to be consistent with the IGS precise ephemerides and ERP information, site
coordinates were computed within the ITRF93 reference frame at epoch I, = 1996.45.
Only two weeks after the campaign, in July 1996, IGS carried out the transition to
ITRF94. Thus it was decided, that all results should be converted to this new reference
frame at epoch I,. Coordinates in ITRF94 might be derived by means of equation (1)

%94 (I,) =%93 (I,) + 193,94 (t s) + 143,94 (m, (i)x, {j)y, (j)z, I,) %93 (Is) (1)
T93,94(/,)=(-0.094, + 0.066, + 0.060) [m]
~3,94(/,):{j)x =-2.21 (i)y =-4.83 (j)z =+1.95 [mas]
m=0.0015 [ppm]
where ~3,94(/,) and 143,94(1,) denote a shift vector and a rotation matrix (similarity
transformation) whose components are previously determined based on ITRF sites
available in both frames (Graz, Wettzell, Zimmerwald, Padova). The conversion to epoch
10 = 1993.0 can be managed by applying velocity field information consistent with the
geophysical model NNR-NUVELIA to all stations, whose individual velocities in ITRF94
are not yet known.

Since customers are not interested to deal with continuously changing coordinates the
influence of the ITRF velocity field can be avoided by converting the whole GPS-network
to the commonly adopted ETRS89 system at epoch IE =1989.0. This step would imply the
subsequent transformations (Boucher, Altamimi, 1995).

X E(/,) = X 93 (1 s) + T93 ,E(ts ) + 143,E(m, OJ x , OJ y , OJ z , t,) X93 (t, )(t, - t E) (2. a)

T93,E(/,) =(+0.019, - 0.053, - 0.021) [m]
143,E(ts ):Wx =+0.32 OJ y =+0.78 OJz =-0.67 [mas/ year]
m=0.0 [ppm]

81
Under the assumption VE =0 (stable part of the Eurasian plate) we may conclude that

(2.b)

It has not been decided yet to which frame the AGREF/AREF -coordinate set should be
referred. Of course, ETRF89 would be one opportunity. On the other hand, also less
rigorous solutions in ITRF94 or ITRF96 are conceivable. For example, station coordinates
may always be determined with respect to a set ofITRFxx-coordinates (AREF-network),
'frozen' at a distinct epoch. The rotation of the stable part of the Eurasian plate can be
calculated for any epoch and therefore applied if necessary.

Antenna Phase Center Corrections

The large number of GPS receivers used in the AREF-survey inevitably implies a wide
variety of equipment. Tracking data stemming from 5 different receiver types and 7
different antenna types ( in alphabetical order: Ashtech LIIL2, Dome Margolin T,
Geotracer 2200 LIIL2, Leica External, Leica Internal, Micropulse Choke-Ring, Trimble
Geodetic) had to be processed. Thus especially for the height determination the correct
modelling of antenna phase center variations played an important role. Besides the
PHAS_IGS.Ol file (Rothacher et aI., 1995), whose use is recommended by IGS and which
contains basically a table of phase-center offsets and elevation dependent coefficients, a
further correction table, based on results of antenna calibrations carried out prior to AREF
was available. Eventually it was decided to use mainly the coefficients given in the IGS-file
(corrections with respect to the Dome Margolin antenna), because this procedure allows
for an understandable interpretation and comparison of the results. Differences to the IGS
numbers were only taken into account because of a misleading interpretation of the ARP-
definition of the Geotracer 2200 LIIL2 antenna and due to the impact on observations
carried out with the Leica external antenna at elevation angles above 15 degree.

Data Processing and Combination

Figure 2 highlights the tie of the campaign specific reference points to the regional IGS-
sites. Heavy constraints were put on the coordinates of the IGS-sites according to their
standard deviations which were reported in the weekly EUREF-combinations of the
regional network. Most of this campaign specific reference stations were continously
tracking over 14 days. Thus the standard deviation of their coordinate estimates tended to
become small numbers «3 mm), as expected.

Finally the huge number of regular sites were tied to the reference network. Regular
stations were connected to at least 6 stations of the superior frame, including at least 3
stations of the IGS-network. This procedure ensures a convenient tie to the ITRF, even in
the case that some of the fiducial stations shall be lost during the next years.

82
-11J .- Campaign-specific Reference Sites (Level B)
®
@
!GS-Station
Permanent- Stallon
O~Dl
o
o
:su.io:a. 2

o
SeWQD. :31
tied to o SM.IOll. 4-
o SculC1:a. &
WiEN _ the regional IGS-network (Level A) 1----:-1:=55=00=00=0-1 6.
AREf-PWlkt
AGREf-Punkt
o s.aJ.an 8
o S--IQD .,

• BOAl

0",,--===_-===-"2-..'5 km

Figure 2: AREF - Reference Sites

The combination of the network was performed at the Normal Equation (NEQ) level.
Each of the 'Session-NEQ-files' usually comprises coordinate and troposphere unknowns
of about 50 regular points. The combination of this session-NEQs leads to a normal
equation system for the whole network and of course a representation in SINEX format
which allows for a comparison with the calculations of other Analyses Centers.

Table 1 shows the standard deviation per session of the combined solution, obtained from
the results of both Analyses Centers.

Session 1 2 3 4 5 6 7

North 4 1.5 2 1.5 1 2 1.5

East 3 3 2.5 2.5 2 2 2
Up 10 6.5 7.5 5 5 4.5 5.5

Table 1: stdv. in mm

Basically the standard deviation of the horizontal components varies from 1 to 3 mm

respectively from 5 to 8 mm in the height. These numbers are quite small and encouraging
but one has to keep in mind that both solutions were based on the same data and

83
processed with the same type of software. Therefore these differences are due to different
processing schemes and do not necessary reflect the accuracy of the coordinates.

A more reliable estimate of accuracy is based on stations which were occupied several
times during the campaign with different equipment. This group comprises about 40
stations and the standard deviations of the horizontal and height components of the
combined solution are really very useful numbers to validate the quality of the network.

North: ± 3.1mm East: ± 2.1 mm Up: ±7.6mm

Future Aspects

In terms of accuracy, and due to the large number of stations incorporated, AREF
represents a major step towards a homogeneous state of the art reference network. In
addition, the combination with AGREF data should allow for the derivation of velocity
estimates with respect to the ITRF94 at about 80 sites. To preserve the benefits of AREF,
the network has to be maintained during the next years. This goal can either be achieved
by means of analysing data obtained from several planned densification campaigns which
are tied to AREF, or by re-observation of the whole network in a few years.

Further activities concentrate on the establishment of a network of GPS permanent

stations serving real-time applications. Actually there are two systems on trial. The first
one distributes RTCM 2.1 messages via Data Radio Channel over the north-eastern part of
Austria (Daller et al., 1997). The second system has been installed in a province in the
south-eastern part of Austria. It comprises 4 active reference stations with a mean point to
point distance of about 50km which transmit correction signals at 75 MHz. These stations
were connected via data lines to reduce distance dependent errors within the enclosed area
(Seeber, Bader, 1997). Which of these systems, or maybe a combination of both, will
finally become a nationwide service cannot be foreseen at the moment. But we are looking
forward to support precise real-time applications over the whole Austrian territory still
before the end ofthis century.

References

Beutler G, Mueller 1., Neilan R., Weber R. (1994): IGS - Der Internationale GPS Dienst
fur Geodynamik; Zeitschrift f. Vermessungsw., Vol. 5/94, pp. 221-232, Stuttgart.

Boucher c., Altamimi Z. (1995): Specifications for Reference Frame Fixing in the
Analyses of a EUREF GPS campaign; Bayr.Komm., Vol. 56, pp 265-269, Munich.

Boucher c., Altamimi Z., Duhem L. (1994): Results and Analysis of the ITRF93;
IERS Technical Note 18, Central Bureau ofIERS, Observatoire de Paris.

84
Boucher c., Altamimi Z., Feissel M, Sillard P. (1996): Results and Analysis of the
ITRF94; IERS Technical Note 20, Central Bureau ofIERS, Observatoire de Paris.

Doller H., Erker E., Weber R.(1997): AREF - Das GPS-Grundnetz von Osterreich als
Basis der modemen telematischen Geodasie; Paper presented at the Intergeo 97, Karlsruhe.

Pesec P., Sunkel H., Erker E., Imrek E., Stangl G. (1997): AGREF - Das Osterreichische
Geodynamische Bezugssystem, Inst. F. Weltraumforschung, Graz.

Rothacher M, Mervart L. (1996): Bernese GPS Software, Version 4.0;

Astronomical Institute, University of Berne.

Seeber G., BOder V. (1997): New Developments in the Improvement ofGPS Kinematic
Applications; this Volume.

Seeger H. (1997): The Current Status and Perspectives of EUREF, Reports on

Geodesy, Vol. 3/97, pp. 31-76, Warsaw.

Weber R., Walter G., Klotz St. (1995): GPS-relevante Koordinatensysteme und deren
Bezug zum Osterreichischen Festpunktfeld;
osterr. Zeitschrift f. Vermessung u. Geoinformation, Vol. 4/95, pp.190-200, Vienna.

Zumberge J., UrbanM, Liu R., Neilan R. (1996): International GPS Service for
Geodynamics, 1995 Annual Report;IGS Central Bureau, JPL, Pasadena.

85
 REAL-TIME DIFFERENTIAL GPS ERROR MODELLING
IN REGIONAL REFERENCE STATION NETWORKS

Lambert Wanninger
Geodetic Institute, TU Dresden
D-01062 Dresden, Germany

Abstract
Observations of several GPS reference receivers in a regional network enable two kinds
of error modelling and reduction. The spatially correlated errors (orbit, ionosphere,
troposphere) are modelled epoch-by-epoch and satellite-by-satellite by 2-D linear in-
terpolation. Furthermore, multipath effects are mitigated by averaging correspondent
observations of several receivers. As a result, observations of a virtual reference station
are created. It is assumed to be located at the rover's approximate position and its ob-
servations are used in the precise baseline positioning of the rover. This approach has
been tested with several months of observations including data disturbed by ionosphe-
ric irregularities, large tropospheric gradients and large orbit errors. It considerably
improved ambiguity resolution and reduced coordinate errors in the positioning of
rover receIvers.

Introduction
Precise (cm-accurate) GPS positioning uses carrier phase observations and requires the
resolution of the double-differenced carrier phase ambiguities. It is usually performed
in a baseline-mode with one receiver at a reference station and one rover receiver at a
station whose position is to be determined. The effects of ionospheric refraction, orbit
errors, and tropospheric refraction on the double-differenced observables grow with
increasing baseline length and also depend on the baseline direction. Furthermore,
station dependent errors like multipath show severest effects for observation sessions
of just a few minutes or less. Therefore, it becomes increasingly difficult to resolve the
ambiguities the longer the baseline length and the shorter the observation session. Fast
and on-the-fly ambiguity resolutions are often limited to baseline lengths of 5 to 15
km, depending on the actual size of the observation errors and they require continuous
observations for at least a few minutes.
In order to reduce the observation errors, GPS-measurements of a regional network
of reference stations are used to model the baseline length and direction dependent
errors (ionosphere, orbit, troposphere). As a consequence the distance dependence of
ambiguity resolution and positioning accuracy is greatly reduced. A further improve-
ment of the reference station observations is obtained through multipath mitigation
and quality control in the multistation reference network. As a result, we compute the
observations of a virtual reference station located at the rover's approximate position.

Observation errors and their reduction

The dominant error sources in differential GPS positioning and the appropriate ways
to model and reduce them in a reference station network will be briefly discussed
below.
Although the effects of ionospheric refraction can effectively be removed with dual-
frequency observations they adversely affect ambiguity resolution. The reduction of
the differential ionospheric refraction is the most important step for the improvement
of ambiguity resolution. Large-scale and medium-scale features of the differential iono-
spheric refraction can be modelled using latitude-differences and longitude-differences
as parameters of a model consisting of two coefficients (inclined plane). The models
are produced for each satellite at each epoch. A minimum number of three reference
stations is required (Wanninger, 1995).
The differential effects of broadcast orbit errors also depend on the length and
direction of the baseline. They can be corrected with a similar model to that described
above (Wanninger, 1996; Wiibbena et al., 1996). The differential effects of tropospheric
refraction often do not only depend on the length and direction of the baseline but
also on altitude differences. With small altitude differences in the reference station
network a two-dimensional-modelling should be adequate, otherwise the model has to
be extended to the third (height) dimension.
The size of multipath effects on both code and carrier phase observations depends
on the station vicinity and the receiver/antenna-system. As long as no detailed infor-
mation on the multipath occurrence at the various reference stations exists, multipath
is considered to behave stochastically and its effect is reduced by averaging the corre-
sponding observations of all reference stations.
Antenna phase centre offsets and variations should be pre-determined by calibra-
tion. No advantage is obtained by combining the observations of several reference
stations.

Virtual reference station

Our objective is to combine the observations of several reference stations in such a way
that an optimum set of observations of a virtual reference station is obtained, which
can be used to determine the position of a rover receiver in baseline-mode. Since some
of the observation errors depend on the horizontal position differences of the rover
receiver in relation to the reference stations, the rover's approximate position has to
be known. Absolute GPS-code-positioning is sufficiently accurate for this purpose. The
virtual reference station is assumed to be located at the rover's approximate position.

87
 GPS-observations
at reference stations
OBS, i=1. ..3...
distances satellite- distances satellite-
receiver receiver
antenna phase center antenna phase center
correction correction
ropospheric corr. from pospheric corr. from
standard model standard model

Figure 1:
:" 'iliTlbigUiiY 'ieSoiUiion": Creation of the
~ ......~~~.~!Tl.~y.~I....... : ~~,
:'" modeiling iii 'disi~"": observations of a
: ance and direction : virtual reference
t__ .~~p.~~~~!1f_ ~~~_. _.~ -....:=::c:.:..:...O/
corr. of distance and ""co--rr.-of"""d"-is"-ta-:Cnce'--'-an-'d--- station.
direction dependent direction dependent
errors errors

observation residuals 2
RES2 , i=1 ...3...

:'::.:'~;~~~~:::::::::
.....b_se_rv_a...:..tio=n.=..re:.asi_du_al_s.../3
_ RES3
~ +C
.
camerphase
corrections ) 0

observations of
irtual reference station

Several processing steps have to be performed in order to transform the obser-

vations of the network of real reference stations to the observations of the virtual
reference station (Fig.1). All processing steps are based on undifferenced carrier pha-
se observables. In the first step the observations are reduced by the satellite-receiver
distance computed from the broadcast ephemeris. Furthermore, antenna phase centre
corrections and standard tropospheric corrections are applied. The resulting obser-
vation residuals are used to resolve and remove the double-differenced carrier phase
ambiguities. This step is a prerequisite for all further error modelling and reduction. If
ambiguity resolution fails for any observation, it has to be excluded from further data
processing. Since the coordinates of the reference stations in the network are precisely
known, ambiguity resolution is much simpler to perform than for unknown baselines.
On the other hand, distance dependent errors limit the complete ambiguity resolu-
tion to maximum distances between reference stations of some SO km. Most of the
difficulties are encountered for low-elevated satellites. From time to time, ambiguity
resolution in the post-processing experiences difficulties with short, and typically low,
satellite passes. Real-time processing sometimes also fails for the first minutes of data
of a rising satellite.
After the successful double-differenced ambiguity removal, two linear combinati-
ons of the carrier phase observations are formed. The coefficients of the ionospheric
model are calculated from the ionospheric linear combination and the coefficients of
the geometric model, which contains the tropospheric and orbit errors, are calculated
from the ionosphere-free linear combination.
Both kinds of models are applied to the reference station observations in such a

88
 SALZ:.-_--"S""'E.....
EH

KLTZ
Figure 2: Northern part of the regio-
nal G PS reference station network in
Sachsen-Anhalt, Germany.
GENT

BURG

50 km

way that the error budget of all reference stations refers to the position of a selected
base station, e.g. at the "centre" of the network. Now, simultaneous observations of
different reference stations to a satellite should be identical with the exception of mul-
tipath, random errors, and the receiver clock error. An averaging process is performed
for code and phase on the level of single-differences between satellites, nevertheless
undifferenced observation residuals are stored. It reduces both multipath and random
errors. In the end, one set of un differenced observation residuals is obtained, which
mainly contains satellite and receiver clock errors and the errors due to ionosphere,
orbit and troposphere as they would be experienced at the selected base station.
The observations of the virtual reference station can then be constructed as shown
on the right hand side of Figure 1. With the help of the ionospheric and geometric
error models the distance dependent errors are shifted from the base station position
to the position of the virtual reference station.
The data processing can either be performed at a central computing facility or it
can be divided between central facility and users. In the first case, the rover's appro-
ximate position has to be known to the central facility. In the second case, the virtual
observations are computed for the base station and they are broadcast together with
the error models to the user. It is then his task to apply corrections to the (virtual)
base station observations in order to obtain the observations of "his" virtual reference
station.

Test data set and results

Three months of observations of a regional German GPS-network (Fig.2) with distan-
ces between adjacent reference stations of 23 to 40 km have been processed in order to
verify the concept of virtual reference stations. The receiver at station STEN has been
considered a rover receiver, all other receivers were used as reference receivers. Obser-
vations of a virtual reference station at STEN were computed and used to position
the receiver at STEN. Ionospheric and geometric error models of the complete data
set were evaluated in order to find periods with standard conditions (undisturbed)
and disturbed periods of time. Examples of error models are shown in Fig.3. For each
epoch and satellite four model coefficients are computed from the reference observati-

89
 2
o ~$i~A..~".J...~~h..~~J\A&11
~ "V v V '\7'-TV ~!111111!!!!1~"__~""""~
--....

-2
iono. model, undisturbed geom. model, undisturbed
[ppm],_----~__----~,_------------~,_------------~

o ~~lHftf'i~Afk-Wo~ ~F~~~~ ~~~~~~

-2
iono. mo el , iono. dist. geom. model, tropo.dist. geom. model, orbit errors
-E<- - - 3h --~) ~<--- 3h ---~) <E<--- 3h ---~)

Figure 3: Examples of ionospheric and geometric error models under standard conditi-
ons (undisturbed) and under disturbed conditions. Each line represents the time series
of model coefficients of a specific satellite. Each example shows either the latitude
component or the longitude component of an error model.

ons: latitude and longitude components of the ionospheric and of the geometric model.
The numbers are given in units of ppm - parts per million of the coordinate difference
for the specific component. The ionospheric model is scaled to the ionospheric effect
on Ll-observations.
Most of our data (80-90%) were free of severe disturbances and can be considered
as observed under standard conditions. Nevertheless, three kinds of disturbances were
detected in the time period from November 1996 to January 1997:

• Medium-scale travelling ionospheric disturbances (MSTIDs), which in mid-lati-

tudes mainly occur during day light hours of winter months (Wanninger, 1995),
were found on many days of the November and December data. The error models
reveal the characteristic apparent MSTID-periods of 10 to 30 minutes. Maximum
rates of change of the error model coefficients can be found in the presence of
these disturbances. They reach up to 1.5 ppm/minute.
• Severe tropospheric disturbances due to large humidity and temperature gradi-
ents occurred once in the three month period and lasted for more than 24 hours.
Observation errors exceeded 1 ppm of the baseline length and mainly affect the
height component of the baseline coordinate differences.
• Large broadcast orbit errors of 30 to 50 m were detected for about one satellite
pass per week. They produce observation errors of more than 1 ppm of the
baseline length.
Under standard conditions single epoch ambiguity resolution succeeded for more than
99% of the observations, and produced incorrect results for just 0.01 % of the data. Un-

90
Table 1: Ambiguity resolution and positioning errors with 5-minute-samples of obser-
vations (GPSurvey 2.20).

Base- # of Success Rate Positioning Errors

line samples Ambiguity RMS dN/dE/dH [cm]
Resolution Ll La
HVLB- Standard Condo 864 89% - 1.1/0.7/2.5
STEN lono. Disturb. 119 87% - 1.3/0.8/2.8
(29 km) Tropo. Disturb. 106 89% - 1.6/0.8/7.8
Large Orbit Err. 126 97% - 1.3/0.8/2.7
Virt. Ref. Standard Condo 864 98% 0.9/0.6/1.9 1.0/0.6/2.4
Station - lono. Disturb. 119 96% 1.,5/0.9/2.7 1.1/0.7/2.7
STEN Tropo. Disturb. 106 100% 1.4/0.7/1.7 1.4/0.6/2.4
Large Orbit Err. 126 99% 0.8/0.7/1.7 0.9/0.6/2.2

der disturbed conditions the success rate decreased to about 95%, with the percentage
of incorrect results increasing to about 0.5%.
Table 1 summarizes the results of positioning with 5-minute-blocks of data as
performed with the post-processing software GPSurvey 2.20. It compares the baselines
from STEN to the closest real reference station HVLB (29 km) to the positioning
of STEN with a virtual reference station. The success rate of ambiguity resolution
could be improved from about 90% to about 98%. The position estimation of a 29
km baseline is usually performed with the ionospheric-free linear combination La. For
the positioning with the virtual reference station, however, ionospheric errors have
already be considerably reduced, so that an Ll-solution would be more accurate than
an La-solution because of its lower sensitivity to multipath and random errors. And in
fact, the positioning errors are not only smaller using the virtual reference stations as
compared to the 29 km baseline, in most cases an Ll-solution is more accurate than
an La-solution. The only exception that was found was in the presence of ionospheric
disturbances. Here, the ionospheric error model removes most of the effects but it does
not perform as well as the dual-frequency ionospheric correction.

Conclusions
Combining the observations of several reference stations of a regional GPS-network to
create observations of a virtual reference station greatly improves ambiguity resolution
and reduces positioning errors. This improvement is mainly achieved by modelling the
differential errors of ionospheric refraction, broadcast orbit errors and tropospheric
refraction. The redundant data of the reference network are also used for mitigation
of multipath and random errors and for a quality control of the observations.
The main prerequisite for the calculation of virtual reference station observations
is successful ambiguity resolution for the reference station network. In general, this
requirement presents no difficulty for distances between reference stations up to about

91
50 km, even for data processing in real-time, because the reference station coordinates
are well known.

Acknowledgement. The GPS observations were made available by the state survey
department of Sachsen-Anhalt, Germany.

References
Wanninger,1. (1995): Improved ambiguity resolution by regional differential modelling
of the ionosphere, Proceedings of ION GPS-95, pp. 55-62.
Wanninger, L. (1996): Fehlermodellierung in regionalen Referenzstationsnetzen, 41.
DVW-Fortbildungsseminar: GPS-Anwendungen und Ergebnisse '96, Potsdam,
Nov. 7-8, 1996, DVW-Schriftenreihe 28/97, pp. 206-218.
Wiibbena, G., Bagge, A., Seeber, G., Bader, V., Hankemeier, P. (1996): Reducing
distance dependent errors for real-time precise DGPS applications by establishing
reference station networks, Proceedings of ION GPS-96, pp. 1845-1852.

92
 PROCESSING STRATEGIES FOR REGIONAL GPS NETWORKS

M. Rothacher, T. A. Springer, S. Schaer, G. Beutler

Astronomical Institute, University of Bern
Sidlerstrasse 5, CH-3012 Berne, Switzerland

ABSTRACT
At the Center for Orbit Determination in Europe (CODE), one of the IGS Analysis Centers,
we are routinely processing the GPS data of a dense European network. In order to improve
the height estimates and, very strongly coupled therewith, the modelling of the troposphere
for this network, we decided to include low-elevation data into the processing, something
that has been done in VLBI analyses for a long time. An appropriate mapping function,
the option to weight the observations according to the elevation angle, and the estimation
of troposphere gradients were implemented to make best use of the data at low elevations.
Problems may arise from phase center variations, multipath, and the troposphere.
143 days have been processed so far using the new processing options and data down to
5 degrees elevation. Comparing the results to a standard 15-degree elevation cut-off solution
shows, that the use of low-elevation data together with an elevation-dependent weighting of
the observations reduces the scatter in the height estimates by about 20-30 percent. The
estimation of tropospheric gradient parameters improves the repeatability in the horizontal
site coordinates by a factor of two for sites successfully tracking down to 10 degrees and
below.

INTRODUCTION
The Center for Orbit Determination in Europe (CODE), one of seven IGS Analysis Centers
(IGS = International GPS Service for Geodynamics), is a collaboration of four European
institutions and is located at the University of Berne in Switzerland [Rothacher et al., 1997J.
The main task of the CODE Analysis Center consists of the daily generation of global
products within the framework of the IGS. In addition, CODE is routinely producing regional
GPS solutions for a dense network of European stations, mainly due to the following reasons:
• The European solutions from CODE are combined with other regional European solu-
tions computed by other centers (see [Bruyninx, 1997; Springer et al., 1997]) to improve
and maintain EUREF, the EUropean REference Frame .
• Because the European solutions are based on the global products from CODE (GPS
orbits, Earth rotation parameters, etc.), these solutions provide a good check for the
quality of the daily CODE products.

International Association of Geodesy Symposia, Vol. 118

Brunner (ed.), Advances in Positioning and Reference Frames
© Springer-Verlag Berlin Heidelberg 1998
 • The network is ideal to perform tests with new processing strategies for regional net-
works.
The standard European solution routinely produced at CODE - in addition there are always
a few test solutions running - is a full network solutions including the sites shown in Figure 1.

Figure 1: Network of 41 European IGS sites processed on a daily basis by the CODE analysis
center.

The correlations between all double-difference observations (sampled at 180 seconds) are
correctly modeled [Beutler et at., 1986, 1987]. Tropospheric zenith delay parameters are
estimated once per hour for each site using the Saastamoinen mapping function. The elevation
cut-off angle is set to 15 degrees. About 85 percent of the ambiguities in the network are
resolved to integers using the QIF strategy [Mervart et at. , 1994].

NEW SOLUTION STRATEGIES

New processing strategies were developed with the idea, that low-elevation data (from the
conventional 15-degree elevation cut-off down to 5 degrees) should be included in the pro-
cessing. The inclusion of low-elevation data has two major advantages:
• The number of observations increases by about 25 percent when going from 20 degrees
to 5 degrees and about 13 percent from 15 to 5 degrees. These numbers are valid for
actual data from the European IGS network. When looking at simulated data (full
tracking down to 5 degrees) the increase would be almost 40 percent and 28 percent,
respectively.
• A better decorrelation between the estimated height and the troposphere delay pa-
rameters of the same station: the correlation coefficient changes from -0.96 to about

94
 -0.83 when going from 20 degrees down to 5 degrees elevation cut-off [Rothacher and
Beutler, 1997].
To make sure that the use of low-elevation data is not going to deteriorate the solution quality,
we decided to add three new features to the software (which will be described in more detail
below), namely: implementation of an appropriate mapping function, elevation-dependent
weighting of the observations, and the option to estimate troposphere gradients.

Niell Mapping Function

The mapping functions developed by [Niell, 1993] (one for the dry and one for the wet part of
the tropospheric delay) were implemented into our software [Rothacher and Mervart, 1996].
The Niell mapping functions were derived from radiosonde data and proved their usefulness
for processing VLBI data. To produce the results presented below, only the dry mapping
function was used to map the total delay.

Elevation- Dependent Weighting

Tropospheric refraction and multipath are responsible for a considerable increase in the
scatter of the observations at low elevations. After a few tests with different models for an
elevation-dependent weighting of the phase observations, we adopted a weighting function
w(z) where the noise increases in the same way with elevation as the tropospheric delay,
namely:
w(Z) = cos 2(z), (1)
where Z is the zenith distance of the satellite. Thus an observation at zenith is assumed to
have unit weight.

Tropospheric Gradients
When using lower elevation angles, the azimuthal variations of the tropospheric delay will
play an increasingly important role. A relatively simple model proposed by [Davis et al.,
1993; MacMillan, 1995] and generalized here for any mapping function is given by:
dmf(z) dmf(z)
~r(a, z) = mf(z) . ~rh + d . cos a . ~rn + . sina . ~re (2)
Z dz
where
z,a zenith distance and azimuth angle of the satellite
~r(a, z) troposphere delay in the direction (a, z)
~rh zenith delay parameter
~rn gradient parameter in north-south direction
~re gradient parameter in east-west direction
mf(z) mapping function
~rh is the conventional zenith delay and ~rn and ~re are the two gradient parameters of
the model. The model corresponds to a tilting of the mapping function by an angle f3 in the
direction a with

(3)

95
New Solution Types
On March 30, 1997 (day 089), four new solution types were implemented into the routine
processing of the European network based on the three software modifications discussed
above. These solution types are listed in Table 1 together with the standard solution A.

Table 1: Standard solution A and four new solution types.

Solution Mapping Elev.-Dep. Elevation Gradient

Type Function Weighting Cut-Off Estimation
A Saastamoinen No 15° No
B Niell (dry) No 15° No
C Niell (dry) Yes 15° No
D Niell (dry) Yes 5° No
E Niell (dry) Yes 5° Yes

For an easy interpretation of the results, each solution type only differs by one option from
the previous one, i.e., the new features are activated one by one. When estimating gradients
one set of parameters was set up per day (i.e., one north and east component per station
and day).
The results presented in the next section are based on the 143 daily solutions that have
been accumulated since the start of the new solution series. Each one-day solution in general
contained 30-35 European sites.

RESULTS
To get an idea of the differences between the solution types A to E we discuss two important
aspects below: the systematic effects, by looking at the parameters of Helmerl transforma-
tions between the solutions, and the repeatability of the coordinates over the series of 143 days
as a measure of the quality of the solutions (there is no ground truth with sufficient accuracy
available) .

Systematic Effects
Table 2 shows the rms errors and parameters of the Helmert transformations between coor-
dinate solutions generated by combining all 143 daily solutions for each strategy.

Table 2: 7-parameter Helmert transformations between pairs of European solutions gener-

ated from 143 days of data using different processing options (see Table 1).

Transformation RMS of Transf. (mm) Rotations (mas) Scale

between North East Up x-axis y-axis z-axis (ppb)
A and B 0.3 0.3 1.3 0.0 -0.3 0.0 1.9
Band C 0.6 0.5 3.5 0.0 0.0 0.0 1.2
C and D 0.8 0.7 2.6 -0.1 -0.1 0.0 -0.5
D and E 1.0 0.6 1.4 -0.1 0.3 0.0 0.4
A and E 1.5 0.9 2.8 -0.2 -0.1 0.0 2.9

96
 We clearly see that changing the mapping function from Saastamoinen to Niell results in a
noticeable scale change of 1.9 ± 0.1 ppb. Weighting the observations also produces a scaling
of the network in the same sense (1.2 ppb). All processing changes (A to E) sum up to give
a scale change of almost 3 ppb with respect to solution A. The rotations of the network are
relatively small.
The rms of the transformation in North, East, and Up directions indicate that the major
changes in the height coordinates occur when weighting the observations and by going to
low-elevation data, whereas the major change in the horizontal coordinates takes place when
estimating troposphere gradients. This will be confirmed in the next section.

Coordinate Repeatabilities
The site Onsala serves as an example here because its receiver performs very well and tracks
the satellites down to a few degrees. Due to its location at the shore we may expect quite
significant and rapid changes in the behaviour of the troposphere. Figure 2 shows the re-
peatability in all three components of the site Onsala over the 143 days processed. Let us
summarize the most important findings:

• The change of the mapping function did not change the repeatability rms error (in
contrast to the systematic effects seen in the previous section).

• The biggest improvement in height results from the elevation-dependent weighting of

the observations and the inclusion of low-elevation data.

• A small improvement in the horizontal position is obtained by weighting the observa-

tions, whereas the data at low elevations degrades the horizontal position.

• The estimation of troposphere gradients has definitely the largest impact on the results
by improving the horizontal positions by a factor of two (from 2 to 1 mm) compared
to the standard solution A. The small degradation of the height component visible for
solution E is due to the reduction of the number of troposphere zenith delays from 24
to 8 per session for this solution (reducing computer resources).

Many pronounced peaks in the North and East coordinate series disappear when estimating
gradients. We think it quite improbable that the gradient parameters solely absorb multi-
path effects and antenna phase center variations. In view of the fact that a permanent site
with a good performance collects almost the same amount of data from the same satellite
constellation every day (apart from a change in time by 4 minutes a day), phase center and
multipath biases in the estimated coordinates should be varying rather slowly with time
(periods of weeks to one year).
Figure 3 demonstrates, that improvements similar to those for Onsala may be achieved for
most of the sites tracking down to 10 degrees elevation or below.
Comparing the standard solution A to the best-performing solution E, we see a clear
improvement in the repeatability of both, the vertical and the horizontal components. One
has to emphasize in this context that the availability of low-elevation data of good quality
is crucial to obtain the type of results shown above. Several sites in Europe are not (yet)
delivering data below 15 degrees or are not performing as well (old receiver types, bad
environment) and no improvement can be seen for such sites.

97
 Component: North
50

E 40
§.
.!:;
30
c
a!
~
20
E
,g
10
~
.~
0 0

-10
80 90 100 110 120 130 140 150 160 170 180 190 200 210 220 230 240
Day of Year 1997

Component: East
50

E 40
§.
.!:;
30
c
a!
~
20
,gE
c 10
.2
"Iii
.~
0 0

-10
80 90 100 110 120 130 140 150 160 170 180 190 200 210 220 230 240
Day of Year 1997

Component: Height
100

E 80
§.
.!:;
60
c
a!
~
40
E
0
./:
c 20
i.~
0 0

-20
80 90 100 110 120 130 140 150 160 170 180 190 200 210 220 230 240
Day of Year 1997

Figure 2: Repeatability of the site coordinates (North, East, and Height) of Onsala for 143
daily solutions using different processing options (see Table 1). The rms error with
respect to the mean is given for each solution type.

98
 Standard Solution A
6

3
E
a:
2

o
BOR1 BRUS GOPE GRAS GRAZ KOSG LAMA ONSA PENC POTS UPAO WTZR ZIMM Station

1_ North _East Up

Solution E
6

4
E
.§.
3
E
a:
2

o
BOR1 BRUS GaPE GRAS GRAZ KOSG LAMA ONSA PENC POTS UPAO WTZR ZlMM Station

1_ North _East Up

Figure 3: Repeatability of station coordinates in Europe for the standard solution A and the
best-performing solution E (see Table 1) over a period of 143 days.

CONCLUSIONS
Several new processing options (inclusion of low-elevation data, a better mapping function,
elevation-dependent weighting of the observations, and the estimation of troposphere gradi-
ents) to improve the troposphere modelling and thus the station heights were tested using
the European IGS network.

99
 From the results we conclude that by including low-elevation data the extremely high
correlation between station heights and troposphere zenith delay parameters is considerably
reduced. Using elevation-dependent weighting of observations we account for the increased
scatter of the data at low elevations and reduce undesirable effects of multipath. The result
is a decrease of the daily height rms by about 20-30 percent. Let us mention, that low-
elevation data and elevation-dependent weighting of observations are beneficial in many other
GPS-related domains like estimation of satellite clocks with code and phase measurements,
ionospheric modelling, ambiguity resolution, etc.
The estimation of tropospheric gradients improves the horizontal position estimates by al-
most a factor of two, provided that enough low-elevation data is available and an elevation-
dependent weighting is applied. We therefore highly recommend that GPS data should be
collected down to as low an elevation angle as possible for all applications aiming at highest
precision. When selecting a location for a GPS site, this should be taken into consideration,
too. It should be emphasized, that successful ambiguity resolution is a pre-requisite to ob-
tain improvements as shown above. Whether the gradients estimated from GPS data may
indeed be interpreted as azimuthal variations in the troposphere, remains to be seen, e.g.,
by comparing the GPS results with meteorological measurements and models.

References
Beutler, G., W. Gurtner, I. Bauersima, and M. Rothacher (1986), Efficient computation of the inverse
of the covariance matrix of simultaneous GPS carrier phase difference observations, Manuscripta
Geodaetica, 11,249-255.
Beutler, G., I. Bauersima, W. Gurtner, and M. Rothacher (1987), Correlations between simultane-
ous GPS double difference carrier phase observations in the multistation mode: Implementation
considerations and first experiences, Manuscripta Geodaetica, 12, 40-44.
Bruyninx, C. et al. (1997), The Use of the EUREF Permanent GPS Network for the Maintenance of
the European Terrestrial Reference Frame, in Proceedings of the Scientific Assembly of the Inter-
national Association of Geodesy, Rio de Janeiro, Brazil, September 3-9, (this volume).
Davis, J.L., G. Elgered, A.E. Niell, and C.E.Kuehn (1993), Ground-based measurement of gradients
in the "wet" radio refractivity of air, Radio Sci., 28(6), 1003-1018.
MacMillan, D.S. (1995), Atmospheric gradients from very long baseline interferometry observations,
Geophysical Research Letters, 22(9), 1041-1044.
Mervart, L., G. Beutler, and U. Wild (1994), Ambiguity Resolution Strategies using the Results of
the International GPS Geodynamics Service (IGS), Bulletin Geodesique, 68, 29-38.
Niell, A. E. (1993), Global Mapping Functions for the Atmosphere Delay at Radio Wavelengths,
Journal of Geophysical Research, 101 (B2), 3227-3246.
Rothacher, M., and G. Beutler (1997), The Role of GPS in the Study of Global Change, submitted
to Physics and Chemistry of the Earth.
Rothacher, M., T. A. Springer, S. Schaer, G. Beutler, E. Brockmann, U. Wild, A. Wiget, C. Boucher,
S. Botton, and H. Seeger (1997), Annual Report 1996 of the CODE Processing Center of the IGS,
in IGS 1996 Annual Report, IGS Central Bureau, Jet Propulsion Laboratory, Pasadena, California
U.S.A., (in preparation).
Rothacher, Markus, and Leos Mervart (1996), The Bernese GPS Software Version 4.0, Astronomical
Institute, University of Berne, September 1996.
Springer, T. A., W. Gurtner, M. Rothacher, and S. Schaer (1997), EUREF Activities at the CODE
Analysis Center, in Proceedings of the International Seminar on GPS in Central Europe, Penc,
Hungary, May 7-9 1997, in press.

100
 A CONTINENTAL WIDE AREA DIFFERENTIAL GPS
STRATEGY

M. H. O. Aquino, V. Ashkenazi and T. Moore

Institute of Engineering Surveying and Space Geodesy
The University of Nottingham
Nottingham
NG72RD
United Kingdom

ABSTRACT
The paper introduces a new Wide Area Differential (W ADGPS) strategy, based on 'Orbit
Relaxation'. The technique is simpler to implement and less costly to maintain than the
conventional W ADGPS approach, based on a full dynamical orbit integration. The scheme
is best suited for a single large country, such as Brazil, or a continental region, such as
Western Europe. It can also be used as an extension of the American WAAS or the
European EGNOS. The orbit improvement is achieved by computing corrections to the
elements of the mean Keplerian orbit, thus generating a refined broadcast ephemeris, which
is transmitted to the users. The paper describes this new approach, in conjunction with
recent developments in modelling error components in W ADGPS. Results of tests
performed at the Institute of Engineering Surveying and Space Geodesy (IESSG), at the
University of Nottingham, with both simulated and real data, are presented and discussed.

1. INTRODUCTION
The provision of WAD (Wide Area Differential) corrections to support a Wide Area
Augmentation Service is realised through the operation of a continental or regional
W ADGPS, which is designed to account for the spatial decorrelation affecting an ordinary
DGPS scheme. A key feature of the process is the ability to isolate and model the different
error sources individually. One such error component results from the difficulty of
accurately coordinating the GPS satellites by the use of their broadcast ephemeris. The use
of Orbit Relaxation to improve the satellites orbits in a WADGPS is currently being
investigated at the Institute of Engineering Surveying and Space Geodesy (IESSG), of
Nottingham University. The operation and maintenance of such a system is simpler and less
costly than when the Orbit Integration approach is used. This new solution will work better
on a regional basis, as opposed to a global situation, for which the more complex approach
is designed.

The orbit relaxation technique has already been addressed in Nottingham with the purpose
of improving station coordinates in a fiducial solution of regional networks and short
satellites arcs (Aquino, 1990). The objective of the current study is to validate the
employment of the resulting orbital corrections in a W ADGPS. In particular, a system
where both the reference network and the user hardware are based on Ll single frequency
receivers is being investigated. Recent developments for the modelling of the ionospheric
errors are also incorporated. This however does not prevent the employment of the system
when dual frequency receivers are available.

2. IMPLEMENTATION OF ORBIT RELAXATION

The error components in a WADGPS are separated as: orbital errors, atmospheric errors
(which can be further separated into a tropospheric and an ionospheric components) and
clock related errors. One can employ the orbit relaxation technique to address the orbital
errors. It consists of solving for corrections to the broadcast orbits, therefore producing a
new "broadcast-like" ephemeris, which can be used to replace the original one.

2.1. COMPUTATION OF WADGPS CORRECTIONS.

Nottingham's current approach for a single frequency based WADGPS breaks down the
differential correction into : (a) an independent ephemeris, generated from an orbit
integration process; (b) an improved ionospheric model, given by an optimised set of
Klobuchar-like parameters; (c) an established model to deal with the troposphere and
(d) clock related errors (Ashkenazi et aI, 1997).

2.1.1 Orbit Improvement.

Ll pseudoranges are collected from a ground network and relayed to a Master station. A
time span of about 72 hours is required to allow for the application of the fiducial technique
(Ashkenazi et aI, 1989), in a least squares adjustment involving the observed data and the
numerically integrated orbit, leading to improved satellites orbits. It was envisaged that the
orbit relaxation algorithms could be implemented to replace the orbit integration. The
general approach to deal with the error components is unchanged, but for one aspect,
regarding the computation of the improved orbits. Corrections for up to six Keplerian
elements are included in the double difference observation equations and a two hour data
span is used in the least squares solution. A difficulty arises when satellites relevant to the
user are not observed at the reference stations, or whose observed arcs are too short to
allow the solution to take place. To overcome this problem, a backup approach has to be
adopted. One possibility is to rely on the original broadcast ephemeris of those particular
satellites, until sufficient data are acquired by the network. A more radical alternative is to
ignore them in the user solution. A further option is to use the navigation inversion method
for these satellites. An automatic switch between the primary and the chosen backup
approaches will be essential for the optimal performance of an operational system.
2.1.2 Ionospheric ModeUing.
In the orbit improvement least squares process, ionospheric scale factors are also estimated.
The ionospheric delay frrst approximation is given by the Klobuchar model, whose
algorithm requires eight coefficients and corrects for approximately 50 percent of the
ionospheric range delay (Klobuchar, 1977). The scale factors accommodate the deficiencies
that remain. It is possible then, in a further step, to produce more realistic delays by
applying the derived scale factors onto the original delays. Finally, an optimised set of
Klobuchar-like coefficients (Ashkenazi et al, 1997) is computed, through the input of the
improved delays in a Non-linear Static Estimator (NSE) algorithm (Kee, 1991). This set of
improved Klobuchar-like coefficients is transmitted to the user.

2.1.3 Tropospheric ModeUing.

Errors due to the troposphere are removed at the Master station by employing the so-called
Magnet Model, adopted by the Magnavox Magnet processing software (Curley,1988).

102
Residual unmodelled tropospheric errors are also absorbed by the derived ionospheric scale
factors (Chao, 1996). The derived tropospheric delays are not meant to be transmitted to
the users, who will be able to derive their own tropospheric delays by employing the same
model (Le. the Magnet model).

2.1.4 Clock Errors.

The process at the Master station ends with the calculation of the satellites clock errors. As
all other sources of errors have been properly modelled, it can be assumed that remaining
errors are clock related ones. In WADGPS the clock error is not derived from one single
reference station, since a network of stations is in place and actually used to generate the
corrections. As a consequence, the latter are contaminated by different station receiver
clock errors. Hence all reference stations clocks must be related to one master clock,
therefore making the transmitted clock terms contain the effect of a unique clock. Clock
errors are propagated round the network using satellites which are observed simultaneously
by the master clock receiver and one of the other reference station (IESSG and BAe,
1992).

2.2 CORRECTIONS UPDATE.

The system requires a cold start, which used to be of the order of 72 hours for the orbit
integration method, and now becomes two hours, with orbit relaxation. Afterwards, these
arcs are moved forwards at frequent time intervals and new least square solutions at the
Master station are performed, leading to new sets of corrections.

2.3 APPLICATION OF CORRECTIONS BY THE USER.

Once each of the error components is computed and made available to the user, they can
then be applied in the following manner: a) replace the broadcast ephemeris with the
relaxed ephemeris; b) replace the broadcast Klobuchar parameters by the set of optimised
Klobuchar-like coefficients; c) apply the Magnet model for tropospheric correction;
d) apply the pseudo range correction transmitted from the Master station, which is
represented by the composite clock correction.

3. VALIDATION TESTS

In order to assess the ability of the technique to model the orbital error component, tests
using simulated data were performed. A software designed to produce GPS observations
free of other error sources was employed. The data was created from arbitrary coordinates
of a network of stations, located at six different sites in Brazil, and from real precise
satellites ephemeris. The orbit relaxation was performed upon the corresponding (actual)
broadcast ephemeris.

A network of five peripheral stations was held fixed, leaving an inner station, Brasilia, for
user accuracy tests. Several sessions of two hours were then processed, each yielding an
improved (relaxed) broadcast ephemeris for all satellites being solved. These orbits were
then compared with the precise orbits during that session and over the following one hour.
In real time, this one hour extrapolation represents the actual situation. In the simulations,
the extrapolated relaxed orbits compare with the precise orbits within 2.5 meters on
average. Distances from station Brasilia to the reference stations vary between
approximately 900 km and 2000 km. Repeated tests showed that the total three-
dimensional user error, obtained throughout the one hour extrapolation period, was of the
order of 10 cm on average (maximum of 20 cm). It is important to note that two extra
satellites, visible from station Brasilia, had not been solved for. That is because they were

103
not observed within the session. As a consequence they could not be included in the user
solution.

4. TESTS WITH REAL DATA

Initial tests with real field data were petfonned. The data had been used to evaluate the
latest version of the Nottin~am WADGPS scheme (Ashkenazi et aI, 1997). It was
collected from the 3rd to the 5 of May, 1993 at four sites within the Racal SkyflX system,
located in Aberdeen, Cadiz, Cyprus and Hammetfest (figure 1). Observations of Ll CIA
code and Yl-Y2 cross correlated pseudoranges, along with Ll and L2 phase were
recorded by Trimble 4000 SSE receivers at 1 second intervals. The Ll CIA pseudoranges
were then used for assessment of the new proposed scheme's petformance. Only two
preliminary trials can be reported by the time this paper is being written. Cadiz, Cyprus and
Hammetfest were adopted as the reference stations. A two hours session, starting at 1 am
on the 3rd of May, was chosen for the first trial. The second trial involves a day-time
session, from 11 am to 1 pm, on the same day. The strategy outlined in section 2.1 was
followed. One ionospheric scale factor per station was estimated for the entire session.

"'> Cyprus
Figure 1 • European Network

4.1 STRATEGIES AND RESULTS / NIGHT TRIAL

The orbital corrections relate to the Keplerian elements of the broadcast ephemeris
referenced to epoch to = 2 am, the center of the session. An attempt to solve for five
Keplerian elements of each observed satellite was made (the argument of the perigee was
fixed in the solution). Solutions for eight out of 11 satellites observed were achieved.
Station Aberdeen played the user's role in the experiment.

The Klobuchar model treats the ionospheric delay as a constant term at the night time
period. Since the processed session was chosen to be mostly at night time, almost all the
derived ionospheric delays were computed by the night-time constant on its own. However,
there was the need to generate a set of improved Klobuchar-like coefficients to be
transmitted to the user, who will be navigating one hour ahead and could be observing
already during day-time. The following strategy was then devised: a) stations scale factors
are derived in the orbital solution least squares and are assumed to cover the deficiencies of
the Klobuchar model (in this case represented by the night-time constant); b) in the
following step, i.e. the process of computing the clock corrections, each of the three
stations scale factor is then applied to the night-time constant; c) in order to accomplish an

104
homogeneous treatment, stations observing partly at day time, have their day-time delays
also multiplied by their corresponding scale factors; d) the clock corrections are now in
sympathy with the combination of the orbital solutions and the scale factors; e) at the user
end a compatible approach has to be employed: a weighted average of the scale factors is
derived, using his distance to the reference stations; 0 this weighted scale factor, applied to
the Klobuchar model (if night-time, to the night-time constant) will then comprise an
optimised ionospheric correction, which will play the role of the improved coefficients. The
results of employing the procedure are shown in figure 2, where it is compared with a
solution based on the precise orbits provided by the Center for Orbit Determination in
Europe (CODE), within one hour ahead of the orbit relaxation solution session.
Coincidentally, only the eight satellites solved in the orbit relaxation adjustment were
visible by the user. The mean error for the relaxed orbits solution is 2.3 m (cr = 1.2 m),
whereas for the precise orbits solution it is 2.9 m (cr = 1.2 m).
10~-----------------------------------------------,

...
1/1
Q) 9 -6- Precise Orbits Solution
a; - Relaxed Orbits Solution
E 8
.E

ON. ~ ~ eN. ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ; ~ ~ ~ ~ ~ ~ ~ s
time - one hour ahead of session

Figure 2 - User Positoning Accuracy : Precise x Relaxed

Orbits
4.2 SIMPUFICATION OF THE METHOD / DAY-TIME TRIAL
For the day-time session the mean Keplerian elements relate to the reference epoch to = 12
am. Six out of ten satellites present in the session were solved for, along with stations scale
factors. The day-time ionospheric delays computed by the Klobuchar model at each station
were improved by the application of the corresponding derived scale factors and then used
to feed the NSE. The optimised Klobuchar coefficients, the Magnet model and the
improved orbits were then used in the process of deriving the clock corrections at the
Master station. At the user end each of the computed error component is applied, following
what is described in 2.3. The user (at station Aberdeen) mean three-dimensional positioning
error, during the one hour just ahead of the session (Le. from 1 pm to 2 pm), was found to
be of 2.0 m (cr = 0.9 m ). Two other user solutions were computed for comparison. The
first relates to a simplification of the ionospheric delays optimisation process. The use of
the NSE to produce optimised Klobuchar coefficients was replaced by the strategy devised
for the night time session. The user results are very similar when this strategy is employed
to this particular session. The third solution was derived by supposing that IS / CODE
precise ephemeris were available to the user, in a similar fashion as for the night trial
(Figure 2). An average positioning accuracy of 2.3 m (0' =0.9 m) was achieved.

105
5. CONCLUSIONS
A suitable W ADGPS scheme is being developed in order to enable the computation of
regional WAD corrections. The current W ADGPS approach in Nottingham relies on the
full dynamic orbit determination process to accomplish a proper accountancy for the orbital
error component. This paper introduces an alternative method which has an orbit relaxation
technique in its core. The investigations being carried out aim, in particular, at developing a
regional W ADGPS capable of providing the best possible accuracies based only on a single
frequency GPS hardware. This does not imply that its use in the presence of dual frequency
receivers is impossible. So far, experiments were undertaken with the use of LI
pseudoranges only. Simulations, performed to validate the orbit relaxation, demonstrated
the way it handles the orbital errors can provide fmal user accuracies within the range of
20 cm, if no other error sources are present. Real data initial experiments gave place to an
interesting and challenging situation, when a combination of the ionospheric and the orbital
error had to be dealt with. A special concern was dealing with the defmition of night-time
and day-time when the Klobuchar model is to be used in conjunction with scale factors. A
strategy was devised to overcome this problem and results of its implementation yielded
user accuracies of the order 2.3 m and 2.0 m, when a night and a day-time session were
tested respectively. It should be noted that these figures agree with average results of the
latest W ADGPS tests performed at Nottingham, based on Orbit Integration.

ACKNOWLEDGEMENTS
The GPS field data used in the real time experiments was provided by Racal Survey Ltd.
This research is part of a PhD course being undertaken by M. H. O. Aquino, sponsored by
CNPq, Conselho Nacional de Desenvolvimento Cientifico e Tecnol6gico, Brasil.

REFERENCES
Aquino, M. H. O. (1990). Improving GPS Positioning Accuracies by Orbit Relaxation.
M.Phil Thesis, University of Nottingham, UK.
Ashkenazi, V. , Chao, C. H. 1. , Chen, W. , Hill, C. 1. and Moore, T. (1997). A New High
Precision Wide Area DGPS System. Navigation, Journal of The Royal Institute of
Navigation, UK, Vol. 50, No 1.
Ashkenazi, V. , Moore, T. , Whalley, S. and Aquino, M. H. O. (1989). High Precision GPS
Positioning by Fiducial Techniques. International Association of Geodesy Symposia,
Edinburgh, Scotland.
Chao, C. H. 1. (1996). Improved Modelling of High Precision Wide Area Differential GPS.
PhD. Thesis, Uni versity of Nottingham, UK.
Curley, R. A. (1988). The Use ofTI 4100 Receivers and MAGNET Software to Determine
Height Differences. MSc. Thesis, University of Nottingham, UK.
IESSG and BAe (1992). Wide Area Differential Corrections R&D Study. Project Report,
University of Nottingham and British Aerospace (Space Systems)
Kee, C. ,Parkinson B. W. and Axelrad, P. (1991). Wide Area Differential GPS. Journal of
The Institute of Navigation, USA,Vo138. No 2.
Klobuchar, J. A. (1977). Ionospheric Time Delay Corrections for Advanced Satellite
Ranging Systems. Propagation Limitations of Navigation and Positioning Systems,
AGARD-CP-209.

106
 POSSIBILITY OF A DYNAMIC CADASTRE FOR
A DYNAMIC NATION

Graeme Blick and Don Grant

Land Information New Zealand
PO Box 5501
Wellington
New Zealand

ABSTRACT

New Zealand, lying across the Pacific/Australian plate boundary, is subject to ground
movements within the country of approximately 5 cm/yr. The existing geodetic control
network and datum, New Zealand Geodetic Datum (NZGD) 1949, has been distorted by
the effect of this ground movement. Because of these distortions, cadastral boundary
coordinates, if held fixed in a coordinate cadastre, would eventually conflict with the
position of slowly moving ground marks. However, the use of new technology such as
GPS is ideally suited to the development of a coordinate cadastre. A dynamic coordinate
cadastre is a possible outcome of recently started projects. A programme is underway to
develop a modem control network and new datum (referred to as NZGD 2000) to replace
the existing network. This could include up to 30 GPS permanent tracking stations which
would monitor the integrity of the new datum and determine rates of ground deformation.
A four dimensional (dynamic) datum with coordinates assigned velocities and changing to
reflect ground movements is an option. As a separate project, the Department is
considering the automation of the survey and land title systems. The automation of the
survey system would see the digital capture of land parcel dimensions. Through
connection to the geodetic network, geodetic coordinates would be assigned to parcel
boundary points resulting in a coordinate cadastre. As the geodetic network moves to
reflect ground movements, adjustment for this could be applied to boundary points
forming a dynamic coordinate cadastre. In this manner, coordinates of boundary points
would encapsulate available evidence of their true ground positions. This would not be a
legal coordinate cadastre, but in the absence of other evidence the coordinate could be
accepted as evidence to define parcels. To take the model further, it is conceivable that the
cadastral surveyor in performing surveys would contribute to the geodetic network and
dynamic datum. This possible model is debated.
INTRODUCTION

New technology such as the Global Positioning System (GPS) is ideally suited to take
advantage of a coordinate cadastre because of its ability to survey and coordinate points
directly in terms of geodetic control. In the near future, it may be possible to routinely
obtain centimetre level accuracy real time positioning over wide areas using Wide Area
Differential GPS (WADGPS) networks and satellite communication techniques. This, and
the possible adoption and integration of other technologies such as inertial systems, is
expected to result in the ability to routinely coordinate cadastral boundary points to
required accuracy relative to a reference network. Such a system could result in large
savings to government in maintenance of the spatial infrastructure and savings to
surveyors and their clients in undertaking cadastral surveys. It is conceivable that a highly
accurate GPS reference network and the survey methods described above, may enable the
number of geodetic survey control points in New Zealand, (over 30,000 in the geodetic
database) to be reduced dramatically to a few thousands or hundreds. The future may also
see well defined physical features in the environment (eg houses or manhole covers) being
used to densify geodetic control for cadastral surveys.

In a dynamic (kinematic) environment such as New Zealand's, several challenges must be

addressed and overcome before the development of dynamic coordinate cadastre can be
contemplated.

THE CHALLENGES

New Zealand is a Dynamic (Kinematic) Nation

New Zealand is a dynamic (kinematic) nation, lying as it does across the

Australian/Pacific tectonic plate boundary. It is subject to ground movements across the
country of approximately 5 cm/yr, disregarding the effects of large earthquakes [Walcott
1984]. This amounts to approximately 2.5m in the 50 years since the current New
Zealand Geodetic Datum 1949 (NZGD49) was observed and established. Added to this
are the catastrophic effects of large earthquakes and the more localised effects of volcanic
activity, ground creep, and subsidence due to, for example, geothermal extraction.

The earth movements have led to distortions in, and a deterioration in the accuracy of, the
current datum. The distortions increase with time causing the coordinates of the trig
stations defining NZGD49 to become increasingly incompatible with new survey
observations. Bevin and Hall [1995] provide a review of the current datum and
distortions within it. These movements also result in movement of cadastral boundary
marks, so that boundary coordinates, if they were fixed in a coordinate cadastre, would
eventually conflict with the position of the slowly moving ground marks [Grant 1995].

108
Development of a dynamic coordinate cadastre (a cadastre where coordinates of
boundary marks change to reflect their true positions) will necessitate modelling of earth
movements.

Current Spatial Definition of the Cadastre is Provided by Survey Observations

The spatial definition of the New Zealand cadastral system is currently based on physical
marks in the ground and survey observations (bearings and distances) between those
marks, rather than coordinates. Witness marks placed close to the boundary are used to
validate the location of boundary marks or to reinstate missing boundary marks. Because
the distance between witness or survey mark and boundary mark is small (tens of meters)
the current system has proved to be very robust. It accommodates distortions in the datum
and the effects of earth movement because the distortions over small areas are generally
insignificant. In the case of catastrophic events, movements cannot be ignored and
resurveys are required. In the current system, high relative spatial definition is maintained
over short distances. Absolute positional accuracy may be only of decimetre or metre
accuracy but the definition of the cadastre is not adversely affected by this.

Such a system may have limitations as we move to new technologies such as GPS with the
capability to deliver sufficient accuracy for cadastral survey over much longer distances
and the desire to achieve survey economy by operating over these greater distances, eg,
WADGPS networks. Distortions in the networks and cadastre then become significant
and new ways of accommodating these distortions must be found. Otherwise, new precise
observations will need to be distorted to fit the lower accuracy survey network.

CURRENT PROJECTS TO OVERCOME THE CHALLENGES AND HELP

DEVELOP THE CONCEPT OF A DYNAMIC COORDINATE CADASTRE

Development of Datum 2000

The New Zealand geodetic control network is the culmination of developments over the
past century [Lee 1978]. Geodetic triangulation was commenced in 1909 and completed
in 1949, forming the 1st Order triangulation of New Zealand. The coordinates of the 1st
Order trig stations were held fixed and used to define NZGD49 which is based on the
International Spheroid. The current geodetic control network has served the cadastral
system well but it now suffers from a number of limitations [Blick and Rowe 1997].

Recognising the limitations of the current datum, a programme was commenced by the
Department in 1995 to re-observe the 1st order triangulation network (293 stations) which
define NZGD49. In addition, a new network was established and surveyed to form the
basis for a new datum, tentatively referred to as New Zealand Geodetic Datum 2000
(NZGD 2000). The requirements of the new datum are described by Blick and Linnell
[1997].

109
It is proposed that the new datum will be geocentric and tied to the IERS International
Terrestrial Reference System. Three options are being considered, a 'fixed' datum, a
'dynamic' datum [Grant and Pearse 1995] or a semi-dynamic datum. As with the current
NZGD49 datum, a new 'fixed' datum is not considered appropriate as it too will degrade
through time due to the effects of ground deformation and changing user requirements.

Automation of the Survey and Titles Systems

Land Information New Zealand has been charged with the development of a strategy for
integrated automation of the survey and title systems in New Zealand [Grant et a11997].
Under this strategy survey and title transactions could be combined into a single generic
land transaction with survey and/or title components. This will enable surveyors and
solicitors to develop new relationships for creating and submitting transactions in land.

In such an automated system, it is envisaged that cadastral survey data would be

submitted in a digital format. The fundamental building block for the survey component
of an automated system is a survey accurate coordinate cadastre based on the geodetic
network that would allow efficient electronic validation of this new survey data. There
will be a requirement for a spatial reference system to underpin the integration of all
cadastral survey data into a digital database. The development of a survey accurate
coordinate cadastre would see the capture and adjustment of parcel dimensions in terms of
the new geodetic datum 2000 network. This would enable all boundary points to be
assigned geodetic coordinates.

Development of a GPS Permanent Tracking Network

Under the development of a new geodetic datum (discussed above) is the development of
a new geodetic network to support that datum [Blick and Linnell 1997]. This new
network will include a number of GPS permanent tracking stations to monitor the
integrity of the geodetic network. Currently four such stations are operating.

A plan by the New Zealand Institute of Geological and Nuclear Sciences (IGNS) proposes
that a number of organisations in New Zealand combine resources to develop a network of
approximately 40 geophysical monitoring sites, half of these sites including permanent
GPS tracking stations. Partners in such a project may include those with scientific
interests, Land Information New Zealand for monitoring the integrity of the new datum
2000, and commercial organisations who may enhance and sell the data. Under such a
system the very network that controls the dynamics of the datum may also be used by
survey practitioners to define the cadastre (eg by development of a WADGPS network).

110
IMPLEMENTATION OF A DYNAMIC COORDINATE CADASTRE

Development of a New Datum

In a dynamic country such as New Zealand, any new datum will need to account for the
effects of ground movements. Several options can be considered for the implementation
of such a datum.

Dynamic Geodetic Datum: A dynamic geodetic datum is one where the actual motion of
trig stations is accounted for in some manner. For example, all trig stations could have
defined velocities (centimetres per year) which allow their future and past coordinates to
be easily calculated. In other words, the datum definition includes a velocity model. Such
a datum carries within its definition, mechanisms for change. Earth deformation and the
increasing accuracy needs of the nation can be accommodated. Changes in coordinates are
smooth and reasonably predictable rather than being jerky and unpredictable as with a
series of static datums. Such a datum may see coordinates of control points III an
automated system changing regularly (weekly, monthly or yearly).

Semi-dynamic Geodetic Datum: It is possible to have a hybrid static / dynamic datum.

We can "freeze" the coordinates of a datum at a specified reference epoch just as NZGD49
coordinates were frozen in 1949. However we can develop and maintain a non-zero
velocity model to be used with this datum. Therefore, this could be called a "frozen"
dynamic datum which is accessed and maintained using dynamic transformations. We can
define a new datum with a reference epoch of 1 January 2000 and define a velocity model
based on actual observations of earth deformation. If we have observations made in 2005
we can convert the 2000 coordinates of existing stations to 2005, adjust the new
observations at that date and convert the coordinates of all new stations back to 2000. By
relating the observations to station coordinates at a common epoch, we can ensure that the
effects of earth deformation are accounted for. This does not stop the physical network
being distorted but it does stop the datum being distorted. We are able to monitor and
maintain the changing relationship between the physical network of ground marks and the
datum coordinates which are used in calculations.

It is probable that initially a semi-dynamic datum will be developed with a possible future
move to a fully dynamic model after the proposed automation of the survey system. A
velocity model for New Zealand is being developed by IGNS. This will enable survey
observations and coordinates of surveys marks to be managed in a way that reflects the
ground motions using either a dynamic or semi-dynamic datum.

Development of a Dynamic Coordinate Cadastre

Connection of the geodetic network to the cadastral system as proposed in the Land
Information New Zealand automation project would allow geodetic coordinates to be

111
assigned to all parcel boundaries. It is proposed that this not be a legal coordinate
cadastre but that in the absence of other evidence, such as an undisturbed boundary mark,
that the coordinate will provide another layer of evidence.

As coordinates of a dynamic control network are allowed to change due to ground

deformation, so these movements could be propagated into the coordinate cadastre. The
coordinates of boundary points would also move to encapsulate the best available
evidence of their true position and accommodate the effects of ground deformation. This
would see the development of the dynamic coordinate cadastre.

Keeping Track of the Dynamics

A GPS permanent tracking network could control the datum at the large scale. Other
control points will be required to provide origin for surveys to ensure that un-modelled
distortions have not crept into the network. While a surveyor may make GPS
observations relative to a base station many tens of kilometres from the site of their
survey, they could be required to connect to control in the vicinity of the survey to ensure
that the modelled coordinates in that area reflect the true position of the ground marks.

Taking this concept further, as surveyors make cadastral surveys and submit data in
digital format to the Department under the proposed automated system, the observations
used to establish and prove the survey origin (the geodetic component of the survey in a
dynamic coordinate cadastre model) could be used to enhance and update the datum. In
future well defined physical features in the environment (eg houses) could possibly be
used to densify the control network. In this manner, the effects of crustal deformation
over small areas could be monitored and used to improve the velocity model. Where
large changes were recorded, eg due to the effects of an earthquake, a resurvey of the area
might be required to re-definite the datum in the affected area.

Beyond GPS Permanent Tracking Stations and Survey Control Marks

Future survey technology (eg, remote sensing) may soon give the ability to map crustal
deformation in detail utilising a few survey marks for calibration and control. The use of
Interferometric Synthetic Aperture Radar (InSAR) to interpolate earthquake deformation
is an example of what is possible. Utilising such detailed deformation maps along with
the ability to easily and precisely fix well defined physical features in the environment
could result in reduced need for geodetic control points. Hardware in the environment
would take over the current function of dense, lower order control. Each house corner
may essentially become a control point. The potential of this, combined with high
resolution remote sensing, could provide a built cadastre showing both legal and physical
aspects of land ownership.

112
REFERENCES

Bevin, A.1. and J. Hall. 1995: The review and development of a modern geodetic datum.
New Zealand Survey Quarterly. Issue 1: 14 - 18.

Blick, G., Rowe, G 1997: Progress towards a new geodetic datum for New Zealand.
Journal of the New Zealand Institute of Surveyors. No. 287, pg 25-29

Blick, G.H., Linnell, G., 1997: The design of a new geodetic network and datum for New
Zealand. Paper presented to the first Trans Tasman Surveyors Conference 12-18 April,
1997, Newcastle, Australia.

Grant, D.B., 1995: A dynamic datum for a dynamic cadastre. The Trans Tasman
Surveyor, Vol 40, No 4, pages 22-28

Grant, D.B., Pearse, M. 1995: Proposal for a dynamic national geodetic datum for New
Zealand. Presented at IUGG General Assembly, 2-14 July Boulder, Colorado, USA

Grant, D., Dawidowski, T., Winmill, R., 1997: New Zealand land tenure beyond 2000:
Full integration and automation of the New Zealand survey and title systems. Paper
presented to the first Trans Tasman Surveyors Conference 12-18 April, 1997, Newcastle,
Australia.

Lee, L.P. 1978: First-order geodetic triangulation of New Zealand 1909-49 1973-74.
Department of Lands and Survey Technical Series No 1.

Walcott, R.I. 1984: The kinematics of the plate boundary zone through New Zealand: a
comparison of short- and long- term deformations. Geophysical journal of the Royal
Astronomical SOCiety 79: 613-633

113
 PRELIMINARY STABILITY TEST FOR THE
REGIONAL GPS TRACKING STATIONS IN TAIWAN

c. C. Chang and R. G. Chang

Department of Surveying and Mapping Engineering
Chung Cheng Institute of Technology
Tahsi, Taoyuan, Taiwan 335, ROC

Abstract

A regional GPS tracking network consisting of eight permanently operating GPS stations
in Taiwan and its offshore islands has worked since 1995, in order to establish a high
accuracy geodetic network purely based on GPS observations. These GPS stations aim to
act as the core stations for establishing a high accuracy three-dimensional reference frame
and motivating some other geodetic and geodynamic applications in Taiwan. The GPS
measurements of this network are to be expected to an accuracy of a few millimeters.
This is to be achieved, through the use of high accuracy GPS technique, by linking this
network of stations to several primary stations of the ITRF (IERS Terrestrial Reference
Frame). Two one-week GPS data from the preliminary stage of measurements made at
these regional GPS tracking stations were collected to compute the network adjusted
coordinates and test for their stabilities in three-dimensional components. A test of
tropospheric modelling was also carried out to investigate the effectiveness of using
associated surface measured meteorological data in high accuracy GPS network
adjustment.

Introduction
GPS has been the most recent and most important advance in space geodetic techniques
over the last 20 years. Due to its high positioning accuracy and effectiveness of
extending the working range from local, regional to continental scale, GPS has now almost
entirely surpassed terrestrial methods for high precision geodetic works. Hence, there has
been a tremendous interest over the past 5 to 10 years in using GPS to connect some
geodetic stations to a global geocentric reference frame for many geodetic applications.
Followed with its economic development and civil applications in Taiwan, a large
number of traditional geodetic control points have been found to be seriously damaged or
lost. In order to re-establish a high accuracy geodetic network purely based on the GPS
observation, a regional GPS tracking network in Taiwan area consisting of eight
permanent GPS tracking stations have been set up and operated since 1995.
The growth of continuously operating regional GPS network has been one of the most
recent trends in GPS, so this regional GPS tracking network in Taiwan was originally
designed to work as the core network for any scale of GPS observations made in this area.
These regional GPS tracking stations were also expected to provide their high accuracy
three-dimensional coordinates for some high accuracy geodetic, geodynamic, and
navigation applications. The coordinate of these tracking stations, therefore, have to be
accurately determined and constantly monitored, with respect to a global reference frame.
The network adjustments based on the two one-week GPS observations made in a short
time interval were independently carried out with a precise ephemeris calculated by the
IGS (International GPS Service for Geodynamics). This aims to measure zero movement
of these regional tracking stations between the two one-week GPS measurements, in order
to prove the potential of GPS for accurate positioning in Taiwan. The effectiveness of
using surface measured meteorological data, opposing to the software-defined standard
meteorological data, for tropospheric modelling was also tested with two GPS data sets in
the network adjustments.

Data Sets and Data Processing for GPS Network

The data sets used to test the zero movement of eight regional GPS tracking stations were
taken from the GPS observation carried out from 30 April to 6 May 1995 and from 11
June to 17 June 1995. The data for each week of GPS observation was taken from seven
consecutive days, with twenty-four hour data observed at three IGS stations nearby this
area, including USUD in Japan, SHAD in China, and TAIW in Taiwan, and also recorded
at eight regional GPS tracking stations located in Taiwan and its offshore islands (see
Figure 1). Data from these IGS stations and regional GPS tracking stations were all
obtained using Rogue SNR-8 receivers. Two tested data sets were expected to
demonstrate insignificant coordinate differences during the short time interval between the
two one-week GPS observations.

Taiwan Area

• IGS St.t/otIS • Rq/011.1 GPS Tracking St.UQfU

Figure 1 Regional GPS tracking stations in Taiwan area

The GPS network solutions were carried out using the GPS Analysis Software (GAS),
developed by the University of Nottingham. The IGS precise ephemeris was used to
provide the satellite orbit information in the tests. During the two one-week GPS
observations, the IGS precise ephemerides used were for GPS week 799 and 805,
respectively. A summary of the options used in the data processing is given in Table 1.

115
 Table 1 Data processing options for the GPS solutions

Reference Frame Definition

Stations Fixed: USUD, SHAO, TAIW

Coordinate Source: ITRF94 GPS Stations [Boucher et aI, 1996}
Tectonic Plate Model: ITRF94 Velocity Field [Boucher et aI, 1996}
.. g~f~~~.~~.~. §1?~~.h:........................... _J.22?:.??......................................................................................................................
Models Applied

Ionospheric Delay: LIIL2 Frequency Combination

Tropospheric Delay: MAGNET Model, with a Standard Atmosphere
Antenna Phase Center: Elevation Angle Dependent Model [Schupler, 1992}
. g~h.?9..4y.I~.~~.~:.......................... _. .~g~.~ . ~~~~~~~~.IM.~9.q!.!.hy!...!22.g1.. ...............................................
Adjusted Parameters

Data Sessions Defined: 14 x 12 hours / session

Stations: Position (X,Y,Z) of Regional GPS Tracking Stations
Satellites: Fixed to IGS Precise Ephemeris
Tropospheric Delay: Zenithal Scale Factor per Station, per Session
(Solved as a First Order Polynomial)
Ambi uities: Inte er Free

Results and Analysis for GPS Network Adjustments

Once GPS solutions have been obtained, e.g. 14 single-session network solutions in each
one-week GPS observation, the repeatability tests on successive single-session solutions
can be performed to test the precision, based on the root mean square deviation of each
single-session solution from the weighted mean.
The indicator of accuracy can be tested through the week-to-week agreement for all the
regional GPS tracking stations. The so-called week-to-week agreement, also seen as a
test of stability, is the discrepancy between the two one-week GPS solutions, which is
based on the assumption that the site movement at all the regional GPS tracking stations is
theoretically less than a centimeter level over the one month short time interval between
the two one-week GPS observations.
The network adjustment solutions are now detailed. The precision and accuracy for the
GPS solutions are estimated by two indicators, namely session-to-session coordinate
repeatabilities and week-to-week coordinate agreements.

Session-to-Session Coordinate Repeatabilities. The session-to-session repeatabilities,

based on the station coordinates, for each one-week GPS observation, namely GPS Week
799 and Week 805, are summarized in Table 2.

Table 2 RMS coordinate repeatabilities for the GPS solutions

Component GPS Week 799 GPS Week 805

N(mm) 6 7
E(mm) 13 19
H(mm) 16 26

116
 The session-to-session repeatabilities can show the quality of data collected during the
two one-week GPS observations. The horizontal coordinate repeatabilities are
consistently better than 10 mm in North component and 20 mm in East component. The
height repeatabilities are as expected to be worse than the horizontal coordinate
repeatabilities with a level of better than 30 mm for both sets of solutions.
It can also be seen that the Week 805 precisions are generally worse than the Week 799
precisions, by around 1 mm in North component, 6 mm in East component, and 10 mm in
height. This level of coordinate uncertainty implies that the GPS observations made at
the preliminary stage of operation for the GPS tracking stations in Taiwan were not in a
very stable condition.

Week-to- Week Coordinate Agreements. The stability test, i.e. the consistency of these
two one-week GPS observations, for the regional GPS tracking stations can be carried out
by using the week-to-week coordinate agreements. The three-dimensional coordinate
agreements between the two one-week GPS observations are given in Table 3.

Table 3 Week-to-week coordinate agreements for the GPS solutions

(values shown are Week 805 minus Week 799)

Station N(mm) E(mm) H(mm)

MZUM -2 3 39
KMNM 3 9 50
FLNM 4 -4 41
PKGM -2 1 48
TMAM 2 9 62
49 . • . . .•
RMS .·• • 1< 3 •.. <
6······· •••••••••

The week-to-week coordinate agreements represent the quality of the data from the two
one-week GPS observations, with RMS agreements in horizontal coordinates of 3 mm in
North component, 6 mm in East component, and around 50 mm in height. The week-to-
week coordinate agreements show a normal level of performance on the horizontal
coordinates, based on the network adjusted solutions using precise ephemeris. However,
the agreements in height are much worse than expected. It is, therefore, important to test
more GPS data sets to investigate the coordinate accuracy of height for the GPS tracking
stations in Taiwan.

Surface Measured Meteorological Data Applied to Tropospheric Modelling

In order to improve the GPS accuracy, a standard tropospheric model is normally used in
the GPS data processing to provide an estimate of the tropospheric zenith delay at a site.
It is believed that the standard atmospheric models provide a broad approximation of
expected tropospheric conditions, but they ignore the actual atmospheric conditions.
Alternatively, surface measured meteorological data are able to be introduced to
practically estimate the tropospheric delay errors. A test of applying such surface
measured meteorological data to the tropospheric modelling was carried out for GPS
observations made at some GPS tracking stations in Taiwan.
Data Availability. A total of four GPS stations, as a part of the continuously operating
GPS tracking stations in Taiwan, were selected to test the effect of the tropospheric delay

117
modelling. These four GPS stations, located at YMSM, PKGM, FLNM, and MZUM,
were chosen mainly based on the consideration of their geographic distributions in Taiwan
and its offshore islands.

Two GPS data sets used for the tropospheric tests were obtained from 8 January to 10
January 1996, and from 22 July to 24 July 1996, respectively. These two three-day data
sets were measured in two different seasons, i.e. winter and summer in the same year, to
present the most different atmospheric conditions. Surface measured meteorological data
were also collected at the GPS sites, provided at one hour intervals for the measurements
of temperature, pressure, and relative humidity.

Test Model and Strategy. The tests of tropospheric modelling for the two GPS data sets,
namely Winter 96 and Summer 96, were performed with a 12-hour session ofGPS data for
all the baselines, where FLNM was fixed to its GPS coordinates defined by a longer term
of observations.
The surface measured meteorological data sets were practically applied with the
Saastamoinen atmospheric model during the GPS network adjustments. The results were
then compared to those obtained by alternatively using the Bernese-defined standard
meteorological data.

Results and Analysis. The session-to-session repeatabilities based on the results of using
Saastamoinen atmospheric model with different types of meteorological data, i.e. defined
and measured data, are shown in Table 4. The so-called season-to-season coordinate
differences are also shown in Table 5 for the GPS solutions based on using two types of
meteorological data.

Table 4 RMS coordinate repeatabilities for the GPS solutions

based on two types of meteorological data

Coordinate Winter 96 Summer 96

Component Defined Data Measured Data Defined Data Measured Data
N(mm) 10 10 13 13
E(mm) 23 23 35 35
H(mm) 38 37 30 30

Table 5 Season-to-season coordinate differences for the GPS solutions

based on two types of meteorological data
(values shown are Summer 96 minus Winter 96)

It is clear that the repeatabilities are not effectively improved in all three coordinate
components when the surface measured meteorological data were used to estimate the

118
tropospheric delay, where a level of 1 mm difference is only found in height for the
Winter 96 solution. However, it can be seen from Table 5 that the season-to-season
coordinate differences are increased by 1 or 2 mm in RMS when the surface measured
meteorological data were applied.
As surface measured meteorological data are believed to be easily affected by calibration
problems and not completely representative of the atmospheric conditions dominating in
the upper troposphere, the software-defined meteorological data is somehow still effective
to be applied with a standard atmospheric model for tropospheric modelling.

Conclusions and Suggestions

This paper aims to show the preliminary tests for the GPS data observed at the regional
GPS tracking stations in Taiwan. An overall summary of the research results is given as
follows:
1. Based on the precise ephemeris network solutions carried out for the GPS tracking
stations at their early operating stage, the coordinate repeatabilities were shown to be
better than 10 mm in North component, 20 mm in East component, and 30 mm in height.
However, a 10 mm difference in height repeatability between two one-week GPS
observations was shown to be not consistent for the quality of GPS data collected.
2. The coordinate agreements between two one-week GPS observations made in a very
short time interval were found to be around 5 mm in horizontal components. However,
the agreements in height were found to be significantly biased by a level of 50 mm.
3. The use of surface measured meteorological data was not able to effectively improve
the precision of the GPS network solutions, comparing with those obtained by using the
software-defined standard meteorological data. The reason is possibly the often biased
surface measured data caused by the calibration errors of the measurement equipment,
systematic and random observation errors, radiation effects on the equipment, and ground
proximity effects [Brunner and Tregoning, 1994J.

Acknowledgments

This work has been carried out as part of the research project, entitled "High Accuracy
Positioning for the Regional GPS Tracking Stations" and funded by the National Science
Council of ROC (NSC Contract 86-2611-E-014-005-T).

References

Boucher, C, Z Altamimi, M Feissel and P Sillard (1996), Results and Analysis of the
ITRF94, IERS Technical Note 20, Observatoire de Paris.
Brunner, F K, and P Tregoning (1994), Tropospheric Propagation Effects in GPS Height
Results Using Meteorological Observations, Australian Journal of Geodesy,
Photogrammetry, and Surveying, Vol 60, pp 49-65.
McCarthy, D D (editor) (1996), IERS Conventions (1996), IERS Technical Note 21,
Observatoire de Paris.
Schupler, B R (1992), GPS Antenna Pattern Data - 11 February 1992, Personal
Communication from B R Schupler, Bendix, USA.

119
 THE USE OF HEURISTICS IN THE DESIGN OF GPS NETWORKS

Peter Dare and Hussain Saleh

School of Surveying
University of East London
Longbridge Road
Dagenham,
Essex, RM8 2AS
England

[email protected]

ABSTRACT

One aspect of GPS network design concerns the logistics of the survey. Others aspects
include location of stations, length of time for session observations. Efficient logistics
enables a survey to be carried out with a lower cost and/or shorter time. A significant
influence on the efficiency of the survey comes from the order in which the sessions are
observed (the schedule). Previous work has shown that, given a list of sessions, the
optimal schedule can be determined given the cost of moving receivers between points and
the list of sessions. The solution was obtained by transforming the problem into a Multiple
Travelling Salesman Problem enabling more than one working period to be included in the
solution. However, the method was only suitable for relatively small networks.

To enable larger networks to be solved, the use of heuristic techniques within the field of
Operational Research have been investigated. Heuristics enable optimal or near-optimal
solutions to be found for very large problems with a reasonable computation time -
optimality, however, is not guaranteed. This paper will describe one particular heuristic
(Simulated Annealing) and show how it can be applied to the logistics design of GPS
networks.
GPS NETWORK DESIGN

There are many different aspects of GPS network design to be considered. However, this
paper will be concentrating on the order of session observations.

A session is when 2 or more GPS receivers are observing satellites simultaneously. In a

survey network a number of consecutive sessions ('schedule') have to be observed. It is
usually a surveyor's intuition that decides on the sessions to be observed - there is no
automated design process to provide a list of the sessions to be observed to meet a
specified criterion. Once the sessions have been decided upon, the receiver schedule (i.e.,
the order in which the sessions will be measured) has to be determined. If S represents the
number of sessions, then the number of possible schedules is given by S! which is clearly
a very large number for some networks. This paper addresses the problem of determining
the optimal (according to some cost criteria) receiver schedule. This aspect of GPS
network design has received little attention in the past despite its importance. Instead,
previous design work has concentrated on other areas.

OPTIMAL SOLUTION TO LOGISTICS DESIGN PROBLEM

The methodology described in detail in Dare (1995), shows how, for small networks, the
optimal order of sessions can be determined whilst allowing for the inclusion of a base
(e.g., hotel, office) and permitting multiple returns to the base so allowing the network to
be observed over more than one working period. Optimality is defined as minimising
some cost criteria represented by a cost matrix containing the cost to move from each
point to all other points. The developed method is based upon an Operational Research
(OR) solution to the Travelling Salesman Problem developed by Little et al. (1963). As an
example, consider the cost matrix shown in Table 1 derived from data supplied by
Craymer (1991). The diagonals are costed as 00 to prevent these movements.

Table 1. Fundy Park survey cost matrix. Costs in minutes of travel time.

From/To 1 2 3 4 5 6
1 00 15 70 77 90 80

2 15 00 77 65 85 83

3 70 77 00 21 24 30

4 77 65 21 00 21 29

5 90 85 22 21 00 21

6 80 83 30 29 21 00

121
 The survey was to provide control for the positioning of a new road passing through
Fundy Park, New Brunswick, Canada - two receivers were to be used to coordinate six
points. The observational schedule adopted by Craymer is shown in Table 2.

Table 2. Fundy Park survey - observational schedule as used by Craymer (1991).

Day 1 Day 2

Rec.1 2 2 1 2 3 5 6 6

Rec.2 3 4 4 1 4 4 5 3

Using the data of Table 1, and the method described in Dare (1995), the optimal solution
obtained is in Table 3.

Table 3. Fundy Park survey - optimal schedule using Craymer (1991) data.

Day 1 Day 2

Rec.1 2 1 2 2 3 5 6 6

Rec.2 3 4 4 1 4 4 3 5

The total cost for the computed schedule is 287 minutes which is 13 minutes less than
the observed schedule (300 minutes) where an assumed travel time of30 minutes was used
for travel to and from the base. Comparing the observed schedule with the computed
schedule the following differences become apparent.

- During the first working day the order of sessions 1,4 and 2,4 is reversed.
- During the second working day the order of sessions 6,5 and 6,3 is reversed.

Although the computed schedule has the sessions spread more evenly throughout the
working days this is partly due to the fact that the planned schedule included an additional
session in Day 1 which was abandoned due to the inability to locate the two points.

HEURISTICS

The method described above does provide, for 2 receivers, optimal schedules which can
span more than one working period.

122
 Nowadays, however, many groups use 3 or more GPS receivers with large numbers of
points. Including additional receivers and more points will enlarge the cost matrix so
making heuristic approaches necessary.

A Heuristic, within the OR sense, is a technique to determine near-optimal solutions in a

reasonable amount of time. They are used extensively within OR to solve large problems
that cannot be solved optimally reasonably quickly (see, for example, Reeves (1993)).
Although a heuristic does not guarantee optimality, a good heuristic may provide the
optimal solution (although this will not be known) or at least close to it.

The basic strategy for a heuristic as applied to GPS schedule design is as follows:

1. Choose an initial schedule

2. Swap two of the sessions to make the schedule cheaper
3. Repeat step 2 until no improvements can be made.

The problem with the above simplistic approach is that the solution is often a local
optimum rather that the global optimum (see Figure 1). To obtain the global optimum, it
becomes necessary to 'climb out' of the local optimum to allow a later descent to the
global optimum. One method to allow this is known as Simulated Annealing.

Figure 1. Local and global optimum.

Cost

local
optimum

Iterations

SIMULATED ANNEALING

The cooling of a material in a heat bath is known as annealing. If a solid object is heated to
make it melt, the rate of subsequent cooling, and the highest temperature reached,
determine the structure of the reformed solid material. Simulated Annealing (SA) is,
obviously, a technique to simulate this process. The basic strategy for the SA heuristic as
applied to GPS schedule design is as follows:

123
 1. Choose an initial schedule
2. Swap two of the sessions to make the schedule cheaper
3. Repeat step 2 until no improvements can be made.
4. Swap two of the sessions to make the schedule more expensive.
5. Repeat step 4 for a small number of attempts
6. Go back to step 2.
7. Stop when no improvements can be made.

The application of this method to GPS will enable much larger networks (i.e., more
points, more receivers) to have optimal or near-optimal schedules designed for them. This
clearly will enable surveys to be carried out more efficiently. Further details of the SA
method are in Saleh and Dare (1997).

CONCLUSIONS

The optimal schedule for GPS networks can be solved using a modified solution to the
Travelling Salesman Problem. The solution allows for the inclusion of an office base with
multiple returns to this base when observing over more than one working period.

For larger networks, the use of heuristics is necessary. A good heuristic will provide
optimal or near-optimal solutions. A popular heuristic within the Operational Research
community is known as Simulated Annealing and the application of this to GPS schedule
design is now progressing.

Acknowledgement

This research is sponsored by the Ministry of Higher Education, Syria.

REFERENCES

Craymer, M. (1991). Personal communication.

Dare, P. (1995). Optimal Design of GPS Networks: Operational Procedures. PhD Thesis,
University of East London.
Little, J.D., K.G. Murty, D.W. Sweeney and C. Karel. (1963). An Algorithm for the
Traveling Salesman Problem. Operations Research, 11,972-989.
Reeves, C. (1993). (editor) Modern Heuristic Techniques for Combinatorial Problems.
Blackwell Scientific Publications.
Saleh, H. and P. Dare. (1997). ). "A Simulated Annealing Approach for Designing a GPS
Surveying Network". Paper presented at the 2nd Metaheuristic International Conference
(MIC'97), Sophia-Antipolis, France, July 21-24.

124
 TRANSFORMATIONS OF GPS COORDINATES
AND HEIGHTS IN TRINIDAD

Karla Edwards, Joseph Barbaste and Jarir Saleh

Department of Surveying & Land Information,
The University of the West Indies, St. Augustine, Trinidad & Tobago.

ABSTRACT
GPS/levelling undulations were computed at 20 accurately-levelled stations of a nation-
wide GPS network. The prediction accuracy of a multiple regression polynomial fitted
through these undulations ,as well as, long wavelength errors of EGM96, OSU91A and
CARIB97 over Trinidad are evaluated. Data to be used in the computation of a
gravimetric geoid is also summarized.

The 1903 triangulation of Trinidad is readjusted and compared to the 1963

triangulateration in order to decide which control points are reliable for use in horizontal
transformation computations and in order to get deformation-free transformation
parameters. In addition a multiple regression transformation and the results of a variance-
covariance simulation of similarity transformations is presented.

INTRODUCTION
Vertical Coordinate Transformation
A GPS network consisting of 91 baselines (measured with a pair of TI4100 receivers)
connecting 33 stations was adjusted in order to facilitate the computation of a GPS I
levelling geoid in the country. However, transformations of WGS84 heights to the local
height datum require the computation of a precise geoid for the island - a gravimetric
geoid. To achieve this, gravity and elevation data in Trinidad, its surrounding sea and the
neighbouring part of Venezuela are needed. As can be seen from Fig.3 , there is a large
gap of gravity data in the Northern Range of Trinidad. However, the Trinidad and Tobago
Lands and Surveys Division is currently pursuing the idea of observing gravity data in
that area. They are also currently in the process of occupying about 35 benchmarks with
GPS (hence facilitating the computation of geometric undulations at these positions) in
order to provide an independent test for the accuracy of the gravimetric geoid.
Horizontal Coordinate Transformations.
The old datum 1903 and the Naparima datum 1963 are still in use in Trinidad, hence two
horizontal transformations are necessary, namely:
1. WGS84BOId 1903 2. WGS84BNaparima 1963

In order to compute reliable transformation parameters, issues such as the size of the
island, the degree of tectonic activity, reliability of the control points, the random and
systematic errors propagated from the 1903 and 1963 networks to the transformed
coordinates and the number of transformation parameters required must be considered.
 To facilitate the computation of horizontal transformation parameters, the Trinidad and
Tobago Lands and Surveys Division has been observing GPS baselines between about
120 trig stations that belong to the 1963 and 1903 networks.

PRELIMINARY RESULTS
Vertical Positions:
• The GPS Network Adjustment.
The 33 station network (Fig.1) was adjusted
with 1 fixed point (Hom-86). This point
was observed and computed by the DMA
using precise orbits. Therefore, it is expected
to have an absolute accuracy of a few
decimetres.

Table 1: Computed
Geometric Undulations Figure 1 : Baselines of Least Squares Adjustment &
Undulation Error Ellipses of the Horizontal Positions.
Accuracy
Station from GPS&
fcm
Levelling f m
RAMONE -42.340 11.6 With the exception of 2 points, the achieved accuracy in
PENAL -43.948 10.2 the horizontal coordinates is 1cm -12cm and 10 cm - 25 cm
110C -42.677 35.9
BM 1084 -42.841 16.9
in the heights.
BM110 -42.485 13.7
BM299 -44.466 19.1
BM456 -44.497 • A Preliminary Geoid For Trinidad.
18.4
BM547 -42.149 GPSlLevelling geoid undulations were determined using
17.1
BM835 -42.860 16.5
BM737 -42.486
the orthometric heights of 20 benchmarks whose
16.9
COROZAL -41.425 corresponding ellipsoidal heights were derived from the
101.8
S 160 -43.442 17.0
SAND -45.018
Network Adjustment (Refer to Table 1). These undulations
17.0
TABLELAN -43.544 were then interpolated using Multiple Regression
13.6
TDST0028 -42.814 14.3
TDSTOO45 -42.645
Polynomial Fitting (e.g., Draper and Smith, 1977; DMA,
23.4
TRI1143 -43.008 1987) based on the model given by the Defense Mapping
19.6
HORN-86
LA FABIAN
-43.811
-44.814
Agency (1987). According to the DMA model, 21 stations
5.0
102.0
PALMISTE -42.017 (an additional benchmark was included which was not
100.8
EDAM(UWI) -42.290 originally part of the GPS network) generate a polynomial
7.8

with no more than 16 unknowns, of which 13 were found to be statistically-significant

(using partial F tests).

The resultant polynomial was found to be:

N=-43.551 +2.283u+5.659v+ 11.214u2 + 18.562uv+ 13.251v2 -18.766u2v
- 91.492v3 - 250.20u3v - 150.05u2V 2 - 648.85uv3 + 351.9u2v3 + 5607.9u3v3
where N is geoidal height , u = ~- ~mean , V = A- A mean and (~mean , A mean) represents the
mean position of 21 benchmarks.

126
The multiple regression fitting was repeated 21 times, excluding one point from the data
each time. The discrepancies between the predicted and known undulations at each
excluded point were used to evaluate the prediction error of the model, which when
computed was ±11 cm.
Long Wavelength Errors Of EGM96, CARIB97 & OSU91A
Geoid undulations were extracted from CARIB97, EGM96 and OSU91A (Rapp, 1997)
at the positions of the 33 stations of the GPS Network. Long wavelength errors were
computed using: Long Wavelength Error = NGPS&Levelling - N GPM
The weighted means of these differences were -0.678m, -0.51Om and -1.493m respectively.
Relative Geoidal Heights using EGM96
In Figure 2 the difference between the relative geoidal heights, along the baseline AB, of
EGM96 and the Preliminary Geoid were defined as:
dN = L1N EGM96 AB - L1N GPS&levellingAB
Figure 2 clearly demonstrates that the relative geoidal height difference between EGM96
and the Preliminary Geoid increases with baseline length. Furthermore, it shows that Mean
Sea Level heights, to better than 15 cm, may be obtained at the present time by a
0$ combination of relative geoid
I 0.. • undulations extracted from
2 /\
1 i 0..
04
I \ EGM96, absolute undulations
1i :~ / \ /
J \
\ of a few points on the island
I ! 0' t---4 v \ ~
and ellipsoidal heights derived
..

~
~
i 0'015

005
A

\
A
/
/
~
"'"
.I
\-oj ~
\
\
1\
\ / \ A from GPS.
\../

Figure 2: Relative Geoidal

Height Difference vs. Distance.
• Data for Proposed Gravimetric Geoid
The Digital Terrain Model (DTM) Of Trinidad
The DTM is needed to compute the terrain corrected gravity anomalies. The DTM of
Trinidad was built from 37 topographic maps of scale I :25000. The height contours were
digitized using PC ARCIINFO. However, PC/TIN was used to create a 3D model
Triangulated Interpolation Network (TIN) from which a 100m x 100m grid was extracted
using Linear Interpolation.

The accuracy of the DTM was evaluated by comparing the levelled heights, of 512
points located along a 20 km 2nd order levelling loop (accuracy ~1.8cm), with the heights
extracted from the DTM using a contour interval of 25ft and 50ft. The accuracy of the
DTM at these contour intervals was found to be 1.78m and 2.82m respectively. (Jules-
Moore, 1996)

The Gravity Data

LaCoste & Romberg gravimeters were used to collect the terrain gravity observations.
The Primary Gravity Base Station used in gravity networks in Trinidad and Tobago is
located at Piarco International Airport and refers to IGSN 1971. This data was used to

127
compute the free air anomalies. The distribution of gravity data throughout the country
proved to be substantial except in the north where the Northern Range is situated (See
Figure 3).

The gravity data positions were homogenized by referencing them to the WGS84
coordinate system while the free air anomalies were referenced to GRS80. Approximately
500 outliers were removed from the gravity anomalies data set. The resulting gravity
anomaly contour map is displayed in Figure 4.

1240000.

1200000.

,1aoooo.

11601lOO
11l1OOOQ,

1120000.
61(XX)IJ.OO 640000.00 660000,00 6WOOO.OO 700000.00 720000.00 740000.00 760000.00

820000.00 /J4OOOO.oo eeoooo.oo eaoooo.oo 700000.00 720000.00

Figure 3: Distribution of Gravity Figure 4: FreeAir Gravity Anomaly

Data in Trinidad Contour Map. Contour Interval of 10 mgal
Horizontal Positions:
• Readjustment Of The 1903 Triangulation Of Trinidad
The 1903 triangulateration network includes about 20000 directions (approximately 3500
directions after averaging) between 700 points. The sets observed from the major 135
stations were unified resulting in about 850 independent directions. 270 triangles were
closed with r.m.s. triangular misclosure of3.1".

The data was reduced to the International Ellipsoid 1924 and then to the UTM
projection plane (where the network adjustment took place). The resulting variance
component was (j02 ~ 1.4 and the standard deviations of the adjusted coordinates range
from 5 cm - 70 cm.

The adjustment revealed that:

1. stations 20, 131, 167, 274 and 328 (see Figure 5) have statistically-significant local
movement (possibly landslides) and should therefore be excluded from any
transformations computations.
2. The Los Bajos fault is not active.
3. The EI Pilar is definitely not a 1.5 cm/year right lateral transform fault! (Demets et aI.,
1990) However, more investigation is needed to determine the degree of its activity
(which is possibly not more than a few mm/year).

128
 4. The effect of more than 100 years of
oil and gas extraction can be attributed
to the movement in the southern part
of the island.

• Transformations between the 1903

&1963 Local Datums
)2.0 METERS
The transformation of GPS coordinates to
the 1903 datum is done in two steps. First,
they are transformed to the 1963 datum
and then to the 1903 datum. In order to
accurately determine the transformation
from 1963 to 1903 datums, Multiple
regression polynomial fitting through 105
common triangulation points was used.
Figure 5: Displacements of 65 Trig.Points
from 1903 - 1963 and their 95% Error Ellipses
For example, the resultant polynomial for transforming the 1903 (Cassini) to 1963
(UTM) was:
till = 260734.563 - 79884.443 u - 105.028 v + 0.557 u2 + 0.484 uv - 1.056 y2 + 0.316 u3+
0.165 u2y2 - 0.241 uv3 - 0.221 u3y2 - 0.117 U2~ + 0.108 uv4 + 0.054 U 3V3 + 0.037 U4~ -
0.003 U 3V 4 - 0.008 U 4V 4
~N = 844538.219 + 105.869 u - 79884.594 v + 0.268 u2 + 2.075 uv + 0.349 y2 - 0.726 u3
+ 1.069 u2v - 0.282 uy2 + 0.017 v3+ 0.173 u4 - 0.015 u3v - 0.859 u2y2 + 0.038 uv3- 0.045 v4
- 0.347 u4v + 0.745 u3y2 - 0.239 U2~ + 0.008 uv4 - 0.006 U4V 2 - 0.086 U 3V 3+ 0.213 U 2V 4
+ 0.069 U 4V3- 0.118 U3V 4 + 0.002 U4V4
where E and N are the easting and northing of each station on either datum and
E1963=E1903+~' NI963=NI903+~N, u=1O- 5(E - 419518.371m), v=1O- 5(N - 306731.054m).
The estimated prediction error of the polynomial is 30 em in the easting and northing
directions.
• Investigation ofSimilarity Transformations
A straight forward covariance simulation was done to identify the appropriate similarity
transformation (Rapp, 1989) for a country of the size and location of Trinidad. The
simulation used an expanding square of points where 9 points are located on its comers,
centre and middle of its sides. The assumed accuracies of the local and GPS coordinates
are 35cm and 7cm respectively. The square was centred on the datum point of Trinidad's
datum while its sides were allowed to increase on the surface of a rotational ellipsoid,
from zero to about 40°, with steps of about 200". The covariance matrix of the parameters
of various transformation models were estimated at each step. The results may be
summarized as follows:
7 Parameters Transformation
The standard deviations of the parameters of a 7-parameter similarity transformation for
an island of the size and location of Trinidad would be about 20 meters in each of the

129
three translation parameters, 0.7' in each of the three rotation angles and 2 ppm in the
scale factor. Considering that the ~ translation from WGS84 to the Naparima 1963
datum is about 0.5m and the rotations are of the order of 1", it is clear that most of the
computed parameters are statistically insignificant. In order to obtain 7 statistically
significant transformation parameters, the country should be about the size of Brazil.
However, it was also found that the correlations among the parameters are not a function
of the country's size. Furthermore, increasing the number of control points decreases the
aforementioned standard deviations by a factor of -V(nl9) which is, in the best case, three
times. Consequently, 7 parameter transformations should not be used in Trinidad and
Tobago or for islands of comparable size and location.
6 Parameters Transformation
A 6-parameter transformation is constituted by three translations, a scale factor, an
azimuthal rotation around the vertical of the datum point and a tilt rotation around an
arbitrary line of azimuth Azo (to account for a certain tilt in the geoid relative to the
WGS84 ellipsoid). This investigation revealed that the accuracy of the 3 translations
,which is about 0.1 m each, is not a function of the size of the country at all. The standard
deviations of the two rotations are 0.6 and 0.5 second and decreases as a function of
-v(nl9) where n is the number of points used for the computation of the transformation. If
120 points are used for the computation, the above errors drop to about 0.15". Thus, it is
possible to compute 6 physically meaningful, statistically significant transformation
parameters, even for a country as small as Trinidad.

ACKNOWLEDGMENTS
The first author wishes to thank the lAG for the travel award to attend IAG97. Thanks to
the Campus Committee on Graduate Studies and Research and the Research Fund
Committee ofU.W.I. who provided additional funding and software.

REFERENCES
Defense Mapping Agency (1987): Department of Defense World Geodetic System 1984:
Its Definition and Relationship with Local Geodetic Systems, DMA TR 8350.2,
Defense Mapping Agency.
DeMets C, Gordon R. G., Argus D. F, Stein S. (1990): Current plate motions, Geophys.
J. Int., 101, pp. 425-478.
Draper N R. and Smith H (1966): Applied Regression Analysis. John Wiley & Sons,
New York.
Jules-Moore, s.P. (1996): An Evaluation of the Accuracy of Elevation Information
Generated By Elevation Models, Department of Surveying and Land Information, The
University of the West Indies, St. Augustine, Trinidad.
Rapp R. H (1997): Past and future developments in geopotential modeling, IAG97 ,
Brazil.
Rapp R. H (1989): Geometric Geodesy, Part II, Department of Geodetic Science and
Surveying, The Ohio State University, Columbus, Ohio.

130
 VECTORS CONNECTING THE GEODETIC POINTS
AT METSAHOVI AND SJOKULLA

Jorma Jokela
Finnish Geodetic Institute
Geodeetinrinne 2, FIN-02430 Masala

Abstract

The Finnish Geodetic Institute has carried out geodetic operations at the Metsahovi re-
search station since the 1970s. During the last ten years a number of new observation sites
have been founded for GPS, SLR, VLBI etc. Some of these are located at Sjokulla, 3 km
NNW from the original Metsahovi. Questions concerning the positions of the points rela-
tive to each other are frequently asked. Both terrestrial and GPS measurements have been
used to determine the centering elements and connecting vectors, and measurements have
been repeated to detect possible movements of the benchmarks. A brief summary of the
observation sites and the local measurements carried out between them is now given.

Observation sites at Metsiihovi

Fundamental points.
348 Metsiihovi was the main benchmark in the triangulation era. It consists of a central
benchmark and three auxiliary benchmarks, all four are bolts in the bedrock.
388 Rogue antenna (METS 10503S011) was installed on the southern edge on the top
of a new height stabilized 20-m mast, built for the permanent GPS station in 1992. Since
then, continuous observations have been made using Rogue SNR-C and Turbo Rogue
SNR-8100 GPS receivers. The reference point of the GPS antenna (now Dome Margolin
B) is now the main GPS benchmark at Metsahovi (Ollikainen et al. 1997).
 386 old SLR (METS 10503S001, CDP 7805) and 391 new SLR (CDP 7806); the co-
ordinates refer to the intersection of horizontal and vertical telescope axes in the satellite
laser equipment.

Auxiliary markers.
384 Bilby benchmark is a bolt in the bedrock under the Bilby tower. The steel tower,
which is older than, and distinct from the GPS tower, was erected when Metsahovi was
connected to the first order triangulation network. When the continuous GPS observations
started in 1991 (with an Ashtech receiver), the antenna was first placed up on this tower.
385 Azimuth benchmark was used in astronomical observations in 1978. The astronomical
azimuth was determined between 348 and 385. The result, 14° 19' 23.96", is not used in
the present work.
It is possible to install three GPS antennas on the top of our GPS mast. The points 387
Ashtech antenna, located on the north-east edge of the mast, and 389 Trimble antenna,
located on the north-west edge of the mast have been used in some observation campaigns.
390 CIGNet benchmark ~o-operative International GPS Network) is a bolt in the rock
underneath the mast supporting the points 387 - 389.
392, an auxiliary benchmark, a bolt at the corner of the new gravity laboratory building
and 393, another auxiliary point at the corner of the new satellite laser building were used
in the last check measurements in 1997.

Observation sites at Sjokulla

Fundamental points.
349 VLBIIGPS (METS 10503M002, CDP 7601) is a steel bolt in the centre of a massive
concrete plate (Vermeer and Paunonen 1994). Like all of our bolts, this bolt has a little
hole in the top for precise centering. The plate is embedded in and lying on sedimentary
deposits. This benchmark was built for the American mobile VLBI system visiting here in
1989. This benchmark was the GPS observation site in the campaign for EUREF 89, too.
380 DORIS illoppler Orbitography and Radio Positioning Integrated by ~atellite) is the
oldest benchmark at Sjokulla, dating from 1988. The benchmark is marked during control
measurements by a special tool placed in the mounting frame, which is fixed in the
antenna base (a massive concrete block) under the DORIS antenna. The bottom plate of
the DORIS antenna (l0503S013) is 3.057 m higher than the benchmark 380.

Auxiliary markers.
In the check measurements we can monitor the possible movements of the fundamental
points. 379 is a steel bolt in the bedrock, surrounded by cultivated land. It was built in
1996 to get a reference marker which is suitable for GPS measurements, too. 381 is a steel
bolt in the bedrock in the wood. 382 is a steel bolt in a block of stone in the forest. 381
and 382 were built in 1988 to get stable reference markers for the DORIS antenna.

132
 383

Figure 1 Tie measurements between Metsahovi and Sjokulla

384

382

380 DORIS
392
gravity
laboratory

393

Figure 2 Observation sites at Metsahovi (left) and Sjokulla (right)

133
Other observation sites

As Metsahovi and Sjokulla are situated in a lowish country, existing triangulation stations
on the tops of two hills, Falkberg and Kopumaki, were used to create the measurement
contact in the terrestrial tie measurement in 1988. However, coordinates dating from the
triangulation were not used in the new computation.
27 Falkberg was selected as a triangulation point in 1919. Measurements for the first
order triangulation there were completed in 1926. The central benchmark and the four
reference markers are iron bolts built in the bedrock. To see the neighbouring stations
Metsahovi and Kopumiiki from the forested Falkberg, a new auxiliary benchmark was
necessary. 383 Kopumiiki is an old low order triangulation station with an iron bolt in the
bedrock as the central benchmark. The dilapidated wooden tower above it was done up
good enough for the measurements.

Measurements and computation

A Wild Theomat T2000 and Distomat DI2000, Kern DKM3 theodolite and Mekometer
ME3000 and Wild and Kern prisms were used in the initial measurements in 1988-1989.
Later on, Wild Theomat T2002 and Distomat DI2002 and Wild prisms have been used.
After station adjustments, we have 127 direction observations and 67 distance observa-
tions made in 1988 - 1997 in the nets presented in Figures 1 - 2.
The orthometric height of the triangulation benchmark 348 Metsahovi had been deter-
mined in 1977 with precise levelling in the Finnish height system N60, H = 54.013 m.
The levellings between Metsahovi and Sjokulla were started in 1988 to connect all the
stations to this height system. Wild N3 and Zeiss Ni002 instruments were used. If we
compare the EUREF 89 coordinates, Sjokulla (349) is 34.980 m lower than Metsahovi
(388). The difference in orthometric heights determined by precise levelling is 35.062 m.
From this we computed the ellipsoidal height differences using the FIN95 geoid (Vermeer
1995), the NKG89 and NKG96 geoids by Forsberg. The results were 34.989 m, 34.969 m
and 34.980 m, respectively. The FIN95 geoid was used in the final computation. To avoid
discrepancies between ellipsoidal and orthometric heights, the geoidal heights were
corrected with -.036 m at Metsahovi and -.027 m at Sjokulla.
GPS observations were made with Mini Rogue, Turbo Rogue and Ashtech geodetic GPS
receivers and Dome Margolin antennas, and processed with Bemese and GPPS softwares.
The vector between Metsahovi and Sjokulla was measured under fairly ideal conditions
during the DOSE campaign in August 1993. Computation with the Bemese 3.4. software
gave (Vermeer and Paunonen 1994):
~ = -1918.178 m, ~y =-1 548.001 m, ~ = +1 324.664 m,

134
from Metsahovi to Sjokulla; ±1 mm accuracy was obtained. This vector and the
EUREF 89 coordinates of Sjokulla (349 VLBIIGPS),
q> = 60° 14' 31.06834", A =24° 23' 03.00378", h = 59.563 m,
have been used to compute the EUREF 89 coordinates for Metsahovi (388 Rogue
antenna),
q> = 60° 13' 02.89097", A = 24° 23' 43.13725", h = 94.543 m.
To make the orientation of the terrestrial nets more reliable, GPS measurements were
recently made at every point where it is possible, i.e. at points 348 and 379, too.
Now the EUREF 89 coordinates of benchmarks 349 and 388 have been fixed in the
processing of both GPS observations and angle and distance measurements. The first angle
and distance observations of the Metsahovi-Sjokulla traverse in 1988-1989 were adjusted
in connection with the final adjustment of the Finnish first-order triangulation (Jokela
1994, pp. 119-121). The Intemational1924 ellipsoid and ED87-FIN coordinates were used
then. In consequence of troublesome measurements and unfavourable geometry, no better
accuracy than ±1 cm was obtained. Now the terrestrial observations form a loose
connection between the two places, but the crucial constraints in the adjustment corne
through the fixed GPS vector. All the observations are reduced now on the GRS80
ellipsoid. A new version of the original adjustment program described by Jokela et al.
1994 was used.
The computation gives the positions of all observation sites relative to the two fixed
points. In the control measurements submillimeter accuracy has been obtained. No
horizontal movement has been found. At Sjokulla there are suspicions of a slight
subsidence.

Eccentricities expressed in coordinate differences in EUREF 89

The results of the adjustment presented in Table 1 are in EUREF 89 coordinate differences
with respect to the Metsahovi GPS station (Rogue antenna, 388). The vector to Sjokulla
(VLBIIGPS point, 349) is kept fixed. Errors in terrestrial measurements cause maximum
discrepancies of about 3 mm both at Metsahovi and at Sjokulla, and standard errors of
about 2 mm. For the whole project such a good accuracy can not be guaranteed, because of
the inaccuracies in geoidal heights. More control measurements will be carried out during
September 1997. This may cause slight differences in the results. A more exhaustive report
of the work presented here will be published by the FGI in the near future.

135
 Table 1
Coordinate differences in EUREF 89. For the point 388,
X =2 892 571.072, Y = 1 311 843.306, Z =5512633.939.

!!.y

27 Falkberg - 532.698 - 3 494.985 + 1 120.716

348 Metsahovi main triang bm + 9.785 -46.090 - 19.140
349 Sjokulla VLBIIGPS -1918.178 -1548.001 + 1324.664
379 Sjokulla DORIS GPS ref bm - 1 928.900 - 1 716.441 + 1380.006
380 Sjokulla DORIS bm - 1931.170 - 1 533.411 + 1328.240
381 Sjokulla DORIS ref bm I -1 794.600 - 1659.527 + 1 301.256
382 Sjokulla DORIS ref bm II -1999.033 - 1 655.915 + 1403.844
383 Kopumiiki -3115.144 - 605.160 + 1 768.131
384 Metsahovi Bilby bm - 22.364 - 37.109 - 3.342
385 Metsahovi azimuth bm -23.126 - 50.596 +0.104
386 Metsahovi old SLR + 24.474 - 35.526 - 23.276
387 Metsahovi Ashtech antenna - 0.947 + 0.201 + 0.397
388 Metsahovi Rogue antenna 0.000 0.000 0.000
389 Metsahovi Trimble antenna - 0.370 - 1.102 + 0.440
390 Metsahovi eIGNet bm - 9.762 - 4.505 - 17.701
391 Metsahovi new SLR + 36.219 - 30.251 -35.392
392 Metsahovi gray lab aux bm + 4.231 + 10.259 - 27.521
393 Metsahovi aux p + 42.391 -17.132 -44.328

References

Jokela J (1994) The 1993 adjustment of the Finnish First-Order Terrestrial Triangulation.
Publ. of the FGI 119.
Jokela J, Poutanen M, Konttinen R (1994) The program for the 1993 adjustment of the
Finnish First-Order Terrestrial Triangulation. Rep. of the FGI 94:3.
Ollikainen M, Koivula H, Poutanen M, Chen R (1997) Suomen kiinteiden GPS-asemien
verkko. Geodeettisen laitoksen tie dote 16.
Vermeer M (1995) Two new geoids determined at the FGI. Rep. of the FGI 95:5.
Vermeer M, Paunonen M (1994) The vector connecting the fundamental points Metsa-
hovi and Sjokulla. In: Gubler E, Hornik H (ed.) Report on the Symposium of the lAG
Subcommission for the European Reference Frame (EUREF) held in Warsaw 8 - 11 June
1994. VerOffentlichungen der Bayerischen Kommission fUr die Internationale Erdmessung
54, pp. 210-215.

136
 USE OF THE FINNISH PERMANENT GPS NETWORK (FINNNET)
IN REGIONAL GPS CAMPAIGNS

Hannu Koivula, Matti Ollikainen and Markku Poutanen

Finnish Geodetic Institute

Geodeetinrinne 2, FIN-02430 Masala, Finland
E-mail: FinnNet@fgiJi

Abstract

The Finnish Geodetic Institute has established a network of 12 permanent GPS-stations in

Finland (FinnNet). The planning of the network started during winter 92/93. The first
stations were installed in 1994 and the last one was built in the autumn of 1996. The
stations are equipped with geodetic Ashtech Z-12 receivers and Dome Margolin type
antennas. All the stations are working fully automatically. The network is used as a 1st
order reference frame in Finland for GPS users, for crustal dynamics studies and for
providing RTCM corrections in real time applications.
FinnNet is used as a reference frame or as a source of data for numerous national and
international GPS campaigns and projects like DOSE (Dynamics of the Solid Earth), BSL
(Baltic Sea Level), BIFROST (Baseline Inferences for Fennoscandian Rebound Obser-
vations, Sea Level and Tectonics), EUVN (European Vertical GPS Reference Network),
and EUREF permanent GPS Network.
We will introduce FinnNet and its function in Finland, present the first results of the
EUREF-FIN campaign and give an overview of the projects FinnNet has taken part in. We
will discuss in general how permanent networks can be used in regional GPS measure-
ments and call into question the function and need of large episodic GPS field campaigns
in the future.

Introduction

The Fennoscandian Regional Permanent GPS network was established by the Nordic
Geodetic Commission in response to an initiative by the directors of the Nordic Mapping
Institutes (KAKKURI et al. 1995). The primary scientific goals of the project are to use the
GPS determinations to estimate the post-glacial deformation of the Earth's crust in
Fennoscandia, and to correct the extensive Fennoscandian tide gauge record for the
influence of vertical motions (BIFROST PROJECT 1996, MITROVICA et al. 1997).
 The planning of the Finnish part of the Fennoscandian Regional Permanent GPS
network was started in the Finnish Geodetic Institute (FGI) during the winter 92/93.
Eleven stations were then chosen for future reconnaissance. The selection of the stations
was made according to the following principles (which initially were not very explicit):
• The network should give a good coverage over the whole country in such a way that the
maximum land uplift differences and horizontal motions between various parts of the
crust can be detected.
• The site should be established on the bedrock.
• The absolute gravity at, or near, the station should be measured.
• The site should be connected to the precise levelling network.

The establishment of the stations

In 1993 steel grid masts of 2.5 m height were erected at five sites. The construction of the
steel mast is shown in Fig 3. The height changes due to the thermal expansion of the steel
tower, appro iO.8 mm during the course of the year, were considered insignificant and will
average out anyway during the yearly cycle. The eccentricities of the towers from the
reference bolts were measured during Sept., 1993, with rapid static GPS in 1994, and with
tachymeter in Sept. 1997.
Three sites, i.e. Olkiluoto, Konginkangas, and Kuhmo were built up in co-operation
with the company Posiva Oy. The mount of the antenna is a steel enforced concrete pillar
which is more stable than the standard mast. The mounts are a part of high precision local
GPS control networks which are periodically remeasured (CHEN and KAKKURI, 1994).
5'
Fig. 1 Permanent GPS stations
in Nordic countries. Circles
denote points used in NKG
combined solution which is an
EUREF Local Network,
squares are other permanent
stations.

55'

15' 20'

138
 An existing building was used for the GPS equipment at four stations, i.e. in Tuorla, at
the Astronomical Observatory of Turku University, and at Sodankylii Geophysical Ob-
servatory. ill Oulu a 8.5 m high existing mast has been converted for GPS use and the
existing plywood observatory building was also available. At Kevo the antenna mast is 5
m tall due to trees around the point; otherwise it is similar to the "standard" setup. The
existing building of Kevo Arctic Research Station is used. Separate instrument cabins
were built in Virolahti, Joensuu, Vaasa, Kuusamo, Kivetty, Romuvaara and Olkiluoto.

Receivers, data downloading and current status of the network

Five TurboRogue SNR-8100 GPS receivers were purchased in 1994. The first obser-
vations with the new receivers were made during the DOSE'94 Campaign (DOSE =
Dynamics Of the Solid Earth), in Aug. 15-26, 1994. The data was collected via telephone
line from four sites. i.e. Tuorla, Virolahti, Vaasa, and Sodankylii. Since the campaign two
sites, i.e. Virolahti, and Sodankylii have been operating continuously.
Due to frequent malfunctioning of many of the TurboRogues, it was decided to change
them to Ashtech Z-12 receivers in 1995. The only exception is the Metsiihovi station
where a TurboRogue receiver is still being used. The Dome Margolin type choke ring
antennas are retained for compatibility and because choke ring antennas have superior
multi-path properties compared to other antenna types (ROCKEN et al. 1994).
20' 30'

Fig. 2 (left) The Finnish Permanent CPS Net-

work - FinnNet, First order triangulation
network, and EUREF densification sites.
EUVN Tide gauges are also marked.

Fig, 3 (below) Construction of the FinnNet

antenna mast.

Top view.
1--=------I-:..9-];L::.:...,.~f.\__--T 65'
~l'"

2.5 m

Steel grid
mast

60 '
Concrete

20'
25' 30'

139
 Continuous observations were started in 1994 at Virolahti, Tuorla, SodankyHi, and
Olkiluoto, and in addition in 1995 at Vaasa, Joensuu, and Oulu. During 1996 four stations,
e.g. Konginkangas, Kuhmo, Kevo and Kuusamo, were taken into operation. Altogether 12
stations are working today.
Data are downloaded automatically every 24 h via an ordinary modem-equipped
telephone line. The control is made with an in-house made program running on a Pentium
PC under DOS 6.22. For the actual downloading, the Ashtech program Remote is used.
Data are stored on hard disk and a backup is made on another computer. Moreover, data
are also sent by ftp to Onsala for BIFROST and NKG EUREF subnetwork computation.
CD-R and DAT are used for archiving both original data and RINEX files.

Use of FinnNet data for international GPS campaigns

The base station in the permanent GPS network is Mets3hovi Space Geodetic station
which has been collaborating e.g. in international SLR observation programs during two
decades. Today it belongs also to the IGS network producing GPS data for IGS orbit
determinations.
The first observations in FinnNet were made during the DOSE'94 campaign. (see also
OLLIKAINEN and POUTANEN 1995). Since then the network has been used in consecutive
DOSE campaigns; e.g. for DOSE'94, DOSE'95 and DOSE'96, occupying also several
temporary points near the land uplift maximum.
The data have been used also in the BIFROST (Baseline Inferences for Fennoscandian
Rebound Observations, Sea Level and Tectonics) project (BIFROST PROJECT 1996,
MITROVICA et at. 1997). One of the BIFROST goals is to obtain three dimensional vectors
from GPS observations for determination of mantle viscosity. First results on land uplift
rate already look very promising.
The third BSL (Baltic Sea Level) campaign was organized in May 1997, simultaneously
with the EUVN GPS campaign. The Baltic Sea Level Project was initiated already in
1989. The purpose of the project is to interconnect the vertical datums of the countries
around the Baltic Sea. For this, a series of GPS campaigns has been arranged, the first one
in 1990, the second one in 1993 and the third one in 1997. More than 30 tide gauges are
included in the campaign. (POUTANEN 1995)
During the second BSL GPS campaign there were no permanent stations in the Nordic
countries, except the IGS network points Mets3hovi, Onsala and Troms!lS which were used
as fiducial points in computation. During the third campaign, the situation was completely
changed because a good network of permanent stations existed in the area. This will be
used as a background skeleton for computation.
EUVN (European Vertical GPS Reference Network) utilizes also the permanent
network. The campaign was arranged in May 1997 for producing backgroud data for
unification of European vertical networks. Because there were several sites common with
the BSL network, the third BSL GPS campaign was decided to be arranged simultaneously
with the EUVN. From FinnNet, Mets3hovi, Vaasa, Joensuu, Kuusamo and SodankyHi
belong also to the EUVN network.

140
First results of the EUREF -FIN campaign.

During the years 96 and 97 a GPS campaign called EUREF-FIN was performed where the
FGI measured a network of 100 points over Finland. The permanent network is a part of
this. Coordinates will be given in the EUREF-FIN reference frame, which is a subset of
the original EUREF89 reference frame. The whole network was measured using Ashtech
Z-12 receivers equipped with Dome Margolin type choke ring antennas. The nominal
observation time at each point was 48 hours.
The first part of the network was measured in 1996 and the second part of the campaign
in Aug.-Sept. 1997. The network was computed with Bernese v.4.0 software. The first
solution was computed keeping only Metsahovi fixed. This free network solution is then
combined with network of fiducial stations, taken from the EUREF network.
In Fig. 4 we show an example of residuals, obtained during the free network solution.
One can see that network shape will have an affect on the results. However, the overall
repeatability is on the sub-cm level for FinnNet points and about the same also for EUREF
densification points. The whole network is shown in Fig. 2.

Need of large GPS campaigns in the future

Today in Northern Europe, e.g. in Finland, Norway and Sweden, there are approximately
43 continuously working GPS stations. The coordinates of these stations may be computed
with centimetre accuracy at any time in whichever global reference frame, thus forming a
good local reference frame for any kind of geodetic positioning or geodynamic research.
Using the GPS data recorded at these stations it is possible to determine the position of
a GPS receiver with cm-accuracy where-ever at any time in the area of the Nordic
R•• Idu.le ~ combined aotuUon, TuorJa

~ ~--------------------------------~

,,-lj- - - - - - - - - - - - -- - - - - - - - - - - - - - - - - -- -- l

Er.,
N

D U

.".«------------------------ - - - - - - --1
.~ .JI-"'I""" _ _ _......_ _ -,.....,......,....""I""'"..,...."""!""'_......-'
2 , , 50 e 7 • a 10 11 '2 " I. ,~ lao
Dey

R•• IdwI'.o'comblnodoolullon. K.VD Fig. 4 Residuals of combined

~~-------------------~ solution in EUREF densification. In
,.-lj- - - - - - --f1-- - - -----:;;_ r--- -- - - - -- -...1 this example one can see some de-
pendence of site position in the
network. Tuorla is situated in the

[;]"
• E
DU
middle of the measurement area,
Kevo is the most distant site, far
from other points. In all cases, the
Nand E component residuals stays
.",J.!- ----j.J---- - - - - - - - - - - - - - - - - - -- - --Ll-- l well below 0.5 cm, the height comp-
onent below 1.5 cm.
I ;2 3 4 5 8 7 • a '0 " 12 13 ,.. 1$ '.
Dey

141
Countries. The position may be computed in whichever global reference frame desired.
The question arises: Do we really need to organize a large international GPS
observation campaign to achieve the goal which may be obtained using a single GPS
receiver together with data recorded at a few permanent GPS stations?
Because the network of permanent stations offer a proper and stable connection to the
global reference frames the need for simultaneous observations is diminishing. One has
only to assure that when measuring new points,that those are well tied to the old ones. In
this way the uniformity of the whole new network is preserved even if measured in small
pieces and during a moderate time span. Such a strategy has already been used in EUREF
campaigns. The original big European-wide GPS measurement in 1989 was followed by
several smaller campaigns.
In the current situation one is tempted to ask if large field campaigns are needed
anymore. There could be some special needs for simultaneous observations but in most
cases regional measurement with a good tie to the existing reference frame will be
sufficient. Also, using data from permanent stations will save one or more GPS receivers
during the campaign for measuring the new points. The scheme sounds first trivial one but
inertia in changing old habits seems to be quite big. The use of permanent networks in
everyone's own measurements could be much more than what it is today.

References

BIFROST PROJECT (BENNET RA., T.R. CARLSSON, T.M. CARLSSON, G. ELGERED, RT.K.
JALDEHAG, P.O.J. JARLEMARK, J.M. JOHANSSON, B.1. NILSSON, B.O. RONNANG, H.-G.
SCHERNECK, R CHEN, J. KAKKURI. H. KOIVULA, M. OLLIKAINEN, M. PAUNONEN, M.
POUTANEN, M. VERMEER, J.L. DAVIS, P. ELOSEGUI, 1.1. SHAPIRO, M. EKMAN, G.
HEDLING, B. JONSSON, J.x. MITROVICA and R.N. PYSKLYWEC) (1996). GPS measure-
ments to constrain geodynamic processes in Fennoscandia. EOS 77, No 35, 337.
CHEN, Rand J. KAKKURI (1994). Feasibility study and technical proposal for long-term
observations of bedrock stability with GPS. Report YJT-94-02. Nuclear Waste
Commission of Finnish Power Companies. Helsinki. 33 pp.
KAKKURI, J., H. KOIVULA, M. OLLIKAINEN, M. PAUNONEN, M. POUTANEN AND M.
VERMEER (1995). The Finnish GPS Array FinnNet: Current Status. Invited paper, IGS
Workshop, Potsdam, 15.-17.5. 1995.
MITROVICA J.x., DAVIS J.L., SCHERNECK H.-G., AND JOHANSSON J.M. (1997). BIFROST
Project: GPS measurements to constrain geodynamic processes in Fennoscandia.
European Geophysical Society, XXII General Assembly, Vienna, Austria 21-25 April,
1997. Annales Geophysicae, 15, Supplement I, C8t.
OLLIKAINEN, M. and M. POUTANEN (1995). GPS-geodesy. In Geodetic Operations in
Finland 1992-1995. (Ed. J. Kakkuri). FGI. Helsinki.
POUTANEN M (1995). A combined solution of the Second Baltic Sea Level GPS campaign.
In Final Results of the Baltic Sea Level 1993 GPS Campaign (Ed. J. Kakkuri). Reports
of the Finnish Geodetic Institute 95:2, 115-123.
ROCKEN, C. J. JOHNSON, J. BRAUN, C. MEERTENS, S. PERRY (1994). UNAVCO facility
GPS receiver tests. Draft 7/21/94, July 1994.

142
 THE NEW SWISS NATIONAL HEIGHT SYSTEM LHN95

Urs Marti and Dieter Schneider

Federal Office of Topography, Seftigenstr. 264
CH-3084 Wabern, Switzerland

Abstract

The existing old height system of Switzerland (LN02) consists of purely levelled heights
without considering gravity measurements. This causes difficulties in the application of
modern techniques such as GPS for height determination and in the data exchange with
neighboring countries. Therefore, a new height system LHN95, based on geopotential
numbers, is being determined at the Federal Office of Topography. LHN95 is treated as a
kinematic network with consideration of the Alpine uplift. The corresponding relative
vertical movements are determined by repeated observations of the first and second order
levelling lines. For LHN95, it was decided to use orthometric heights. This has some
advantages when combining levelling with GPS measurements in mountainous regions.

Introduction

The Swiss Federal Office of Topography (SWISSTOPO) is responsible for the

establishment, the revision and renewal of national geodetic networks. Between 1989 and
1995 SWISSTOPO established a new geodetic reference system based on a GPS network
(LV95) covering the whole country and being well connected to ETRS. Whereas LV95 is
a purely geometric network, there is also a growing demand for a new, well-defined height
system and an adequate height frame. This paper intends to describe the concept and the
procedure for establishing the new height system LHN95 for Switzerland.
The existing National levelling network (LN02) was established at the beginning of this
century. It contains basic elements or observations which even go back to the 19th century
and is still in use for cadastral and most other surveying and mapping applications. It is
based on "observed heights" rather than values corrected with respect to the gravity field.
The heights are static, although it is known from systematic remeasurements that there are
significant relative height changes in the order of 1 to 2 mm/year caused by tectonic
movements of the uppermost crust in the Alps [Gubler et al. 1981].

Concept of the Swiss National Height System LHN95

Project LV95 / LHN95.

In the last few years, SWISSTOPO has been establishing a concept for a new geodetic
control (LV95). The backbone of the concept is the new national GPS network (also called
LV95), which is planned to replace the old triangulation (L V03) from the beginning of this
century. A particularity of LV95 is that it includes 2 reference systems. The first
(CH1903+/L V95) is chosen as close as possible to LV03 and will be used for cadastral
surveying and as the base for the national map projections. The second frame
(CHTRS95/CHTRF95) is as close as possible to ETRS89 and is used in geodesy for
national and international projects [Schneider et al. 1995].
Besides this, a new national height network (LHN95) is part of the new concept. It will be
a modem kinematic network computed rigorously in a combined adjustment using all
existing levelling and gravimetric data as well as the ellipsoidal heights from the GPS
network and the new geoid of Switzerland. The new reference frames are designed to
fulfill not only the requirements of cadastral surveying but also those of geodetic control
for large engineering projects and international data transfer.

The new geoid ofSwitzerland.

An important prerequisite for the new height system is an accurate geoid and quasigeoid in
order to take advantage of the economic and fast GPS techniques in height determination
as well. Therefore, during the last several years the aim of a joint project of the Institute of
Geodesy and Photogrammetry (ETH Zurich) and SWISSTOPO was to generate a national
geoid and quasigeoid with an accuracy of a few cm over the whole country [Marti, 1997].
About 600 astronomically observed deflections of the vertical and 70 connections between
GPS stations and levelled bench marks were used for the computations.
Furthermore, a 25-meter DTM and density models for crustal structures were introduced.
After reducing the measurements from the effects of topographic masses, the residuals
were interpolated by means of least square collocation. The final geoid was calculated by
re-adding the effects of the mass models (remove-restore technique). The result revealed
an accuracy of 2 to 3 cm in the flatter areas and of about 5 cm in the Alps.

The new National Height Network LHN95

Why a new height system?

The maintenance of vertical densification networks for cadastral surveying in the Alps
using levelling techniques is extremely expensive. Using GPS to determine height would
be a much more economic technique for this task. There is a large demand for combining
or replacing low order levelling networks with GPS networks in mountainous regions.
However, this is not possible with the existing height system LN02, which consists of
levelled heights with no consideration of the gravity field and therefore it has no simple
relation with the geoid or quasigeoid.
Furthermore, LN02 contains discrepancies of up to 20 cm stemming from errors in the old
measurements and adjustment processes and because the effects of recent crustal
movements were neglected. Therefore, for high-precision applications the official height
systems in general cannot be used.
Another reason for introducing a new height system is the non-compatibility of LN02 with
the systems of neighboring countries, making the exchange of data, as needed for several
projects, rather difficult. All these considerations lead to the decision to establish the new
Swiss National height system LHN95.

144
Definition of the new vertical datum ofLHN95.
The existing Swiss vertical datum (LN02) is based on the height of the reference point
"Repere Pierre du Niton (RPN)" in Geneva (373.600 m above sea level) which was
introduced in 1902 and derived from a levelling connection to the tide gauge in Marseilles.
On the other hand, the European vertical datum (UELN) is based on the tide gauge in
Amsterdam. A comparison of the UELN results with a provisional rigorous adjustment of
the Swiss first order levelling network revealed an offset of about -13 cm between the two
systems (huELN - hLNo2 )·
Two vertical datums (analogous to the dual definition of the Swiss terrestrial reference
system) are defined for LHN95: For both datums, the fundamental point in Zimmerwald is
used as reference point. For the local vertical datum definition (CHI903+) the reference
value of Zimmerwald is derived from the old (orthometric) height of the RPN, whereas for
the global datum (CHTRS95) this value is derived from the geopotential number of a node
(Olten) of the UELN.
The basic values of the bench mark heights of LHN95 are geopotential numbers derived
from the kinematic network adjustment with the values of the fundamental station in
Zimmerwald held fixed. For practical use, these geopotential numbers have to be
transformed into either normal or orthometric heights. An orthometric height system was
chosen for the official "local" system CH1903+ for cadastral surveying. This is in
accordance with the new height system in Austria but in discordance with those of France,
Germany and other European countries where normal heights were introduced as the
official system. The discussion which lead to this decision is summarized as follows:
• The advantages of normal heights are that they are very easy to calculate and that it is
not necessary to make any assumptions on the density distribution in the interior of the
earth. Furthermore, the normal heights are in better accordance with levelled heights
than orthometric heights.
• The advantages of orthometric heights are only evident in mountainous regions such as
the Alps. There the geoid shows a much smoother behavior than the quasigeoid, which
is advantageous for the interpolation of geoid undulations. Therefore, it is also easier to
derive precise orthometric heights from GPS heights.

Differences between the old (LN02) and new (LHN95) reference frame.
In order to make the transition from the existing levelled heights of LN02 to the
orthometric heights of LHN95 the differences between the 2 frames have to be known
with a high accuracy. First test computations revealed that these differences reach several
decimeters and are strongly correlated with height. Therefore, the first functional approach
for the transition from LN02 to LHN95 is a linear function in height. In this way, the
residuals can be reduced to less than 5 cm for most bench marks. The residual height
differences which are caused by influences such as distortions of the old levelling network,
kinematics or the influence of gravity can be modelled by a two-dimensional interpolation.
With this model the differences between LN02 and LHN95 can be predicted with an
accuracy of 1 to 2 cm, which is enough for low order levelling bench marks and
triangulation points.

145
Combination of levelling networks with GPs.
An important option of LHN95 is the integration of selected GPS-derived heights into the
adjustments with the aim of stabilizing the levelling network over long distances. The
main problem with this combination until now has been the limited accuracy of the height
components of relative GPS positions due to systematic errors caused by tropospheric
refraction etc. To contribute a significant improvement of the solution, the accuracy of the
height component should be in the order of I to 2 cm over a distance of 100 km. But the
present solution of the GPS network LV95 shows a relative height accuracy of only 2 to 5
cm, especially in the mountainous part of the country with height differences of up to
almost 3000 m between adjacent points. Further improvements are expected from water
vapor radiometer measurements and from a better modeling of the troposphere.
Another problem of introducing GPS heights into the adjustment of levelling networks is
the fact that orthometric heights or potential differences must be introduced instead of the
observed ellipsoidal heights. Therefore, the geoid also has to be known with the same
accuracy as the GPS heights.

Kinematic adjustment of LHN95.

The adjustment of all first order levelling data measured during this century is performed
in a kinematic way. This means that not only the last measurements are of interest but also
the older observations have to be taken into account. Because of the large amount of
observations, the adjustment process cannot be performed in a single step but is done in
the following sequence:
1. Separate calculation of 'observed' geopotential differences for each levelling line by
integrating the product of the observed height differences and the observed or
interpolated gravity values.
2. A kinematic adjustment of the 'main points' of the whole network. These are stable
markers which have been observed more than once and therefore represent the regional
vertical kinematics. This step leads to geopotential numbers and their time derivatives
("velocities") as well as to the variance-covariance information for these main points.
3. The results of step 2 are kept fixed, and the potentials and - where possible -
"velocities" of the other stations are now determined by further interpolation.
4. Finally, the orthometric or normal heights are obtained by dividing the geopotential
numbers by the mean gravity (or the mean normal gravity) along the plumb line. The
mean gravity itself is determined from the observed gravity using the same models for
the density distribution of the earth's crust that were used for computing the geoid.
This stepwise determination of heights does not affect the results significantly since the
obtained heights differ less than 1 mm from a rigorous solution, as could be confirmed
with test calculations. The main problems with this procedure are caused by unstable
markers which were observed more than twice and therefore are not consistent with the
mathematical model of linear movement in time. And since the degree of freedom of
levelling networks usually is very small, the a posteriori variance factor of the least
squares solution increases too much if these constraints are not eliminated from the
adjustment.

146
Test computations in the central Alps

A first test of this concept of evaluation was made in an area of the central Alps. This
region is most appropriate for testing because of the large variations in the gravity field
and the relatively large changes in the vertical velocities (I dv z I > 1 mm/year) of the
uppermost crust. The data set used consists of 628 km of levelling lines with a height
ranging from 220 m to 2400 m above sea level. All lines were observed at least twice.
More than 5000 bench marks are included in the evaluation but only 1200 of those still
exist. Out of these, only 780 bench marks were observed more than once, thus allowing
the determination of relative vertical velocities.
Until now GPS observations have not been included in the adjustment, however, a
comparison of the results between levelled orthometric heights and heights derived from
GPS and the new geoid was possible for 15 stations. This comparison showed average
differences of ± 2 cm and a maximum value of 7.5 cm for one station.
The main results obtained are potentials, orthometric heights and normal heights in the
chosen reference system with their standard deviations as well as vertical velocities
(relative to a hypothetical reference station in Aarburg). These results are shown in Fig. 1.

II 1 mm/year I 0\

',r' 'rt1
. \"'l . ~ '

.- ..........
Fig. 1: Vertical movements relative to a marker in Aarburg
The vertical movements show a rather smooth distribution except for unstable markers. In
the flatter areas the calculated movements are of no significance. On the other hand, the
velocities in the Alpine regions with a maximum of 1.6 mmlyear (relative to Aarburg) are
highly significant on the 95% level of confidence.

Comparison (~f obtained results

The results obtained on a part of the test network (the Grimsel line between Balmhof and
Gletsch) are now being discussed in more detail. This is a levelling line with a length of 60
km and height differences of up to 1600 meters (Fig. 2). It was observed twice, once in
1920 and again in 1985.

147
 0 .• 0 0."0

--
- - .- - - - - - - ,
- orthometrtc: heights minus normal heights
orthometric heights minus levelled heights
E
.....
-;;; 0.30 --_. orthometric helghls mlnu, constrained levanad heights
)
~
0 ,30

..=
c
0 .20
- ~

0.20
~
:;;
.c.2'J 0 ,10
..
.c ... - - - - - -- - - _... ----- ........ - ----- ... - ......
0.10

-,.,,-
------,--' ......... - ...
0 .00
--------- -------- -----_ ....
------ 0.00

.
0.0 10 .0 20.0 30 ,0 .0.0 50.0

] <
..,'0 ~
c e c<
N
.~
~
E l!
2500
'C
'0 ~ i!, 0 J! 1.0
~ :t
"
"
~
~
'i:'
2000
:
'"
. .. •
~

.•
.c E
0.. 1~OQ
E
•
.•
!! 0.5 ';;
co 0

•
0
a.. 1000 • 0 ~
....0 'u
0
G

, ·•
500 0 >
0.0

0.0
•
10.0 20.0 30.0 40.0 50.0
• Distance

Fig. 2: Comparison of different solutions relative to the orthometric heights

In addition to the topography, the lower part of Fig. 2 depicts the calculated velocities as
dots. These velocities increase from about 0.35 mm/year in Balmhofto about 0.7 mm/year
in Gletsch. They show a rather smooth behavior except for some unstable points and a
whole region near Meiringen which do not follow the general trend. In the top solid curve
of Fig. 2, the differences between the determined orthometric and normal heights are
obvious. This curve shows a rather high correlation with the topography and reaches
maximal values of almost 40 cm. The second, dashed curve shows the differences between
orthometric heights and the old levelled heights of LN02. This curve is also correlated
with the topography and is more or less parallel to the first curve, which indicates that
levelled heights are much closer to normal heights than to orthometric heights. This curve,
however, contains not only the systematic differences between orthometric heights and
normal heights but also the kinematic modeling and the distortions of the old system. The
third curve finally shows the differences between the rigorous orthometric system and a
result as it would probably be obtained by a local surveyor, regardless of gravity
measurements: The starting and ending stations are kept fixed with their orthometric
heights and the resulting discrepancy is distributed linearly. This leads to differences from
-4 to +6 cm on this line with its extreme topography.

References

Gubler E. , Kahle H.-G., Klingele E., MUller St. and Olivier R. 1981: Recent Crustal
Movements in Switzerland and their Geophysical Interpretation. Tectonophysics Vo1.71.
Marti U. 1997: Integrierte Geoidbestimmung in der Schweiz: Diss. ETH ZUrich 12015.
Schneider D., Gubler E., Marti U. and Gurtner W. 1995 : "Aufbau der neuen
Landesvermessung der Schweiz 'L V95': Teil 3: Terrestrische Bezugssysteme und
Bezugsrahmen" . Berichte aus der L+T Nr. 8, Wabern.

148
 THE NON-FIDUCIAL APPROACH APPLIED TO GPS NETWORKS

Joao Francisco Galera Monico

FCTIUNESP CEP: 19060-900 Pres. Prudente, SP, Brazil. e-mail: [email protected]

Edvaldo Simoes da Fonseca Junior

Denizar Blitzkow
EPUSP-PTR - Caixa Postal 61548 CEP: 05424-970 Sao Paulo, SP, Brazil

ABSTRACT

The basic fundaments of the fiducial and the non-fiducial approaches are presented. The
.main advantage of the non-fiducial approach over the fiducial one is the simplicity of
'computation when the choice of the known stations change. In the traditional fiducial
technique all main steps of the processing must be repeated. An example involving the Sao
Paulo State Network, which is composed of 24 stations (Plus 3 fiducial stations) is
presented. Results obtained from both approaches are quite similar.

INTRODUCTION

The adjustment of a GPS network can either be carried out by using one of the available
ephemerides, or by estimating the satellite orbit (state vector) during the processing. At
both cases, it is usual to constrain the coordinates of a number of stations to their known
values. This approach is referred to as fiducial GPS. In the non-fiducial one, all station
coordinates are estimated instead, as well as the satellites state vectors. Such approach has
been applied to global GPS experiments, in which fiducial stations are not necessary to
provide an origin and scale. This approach may also be suitable for regional networks, but
in such a case, instead of estimating the satellites state vectors, one can use an IGS
(International GPS Geodynamics Service) precise ephemerides. In order to make the non-
fiducial approach solution consistent with that of the fiducial one (both using precise
ephemeris), some extra computation should be carried out. The main advantage of the non-
fiducial approach over the fiducial one is the simplicity of computation when the choice of
the known stations change. In the traditional fiducial technique all main steps of the
processing must be repeated, while in the non-fiducial approach only a transformation is
required. Therefore, if both approaches provide similar results, it is advisable to make use
of the non-fiducial one.
In this paper, the basic fundaments of both approaches will be presented, followed by a
description of the software used and an example involving the Sao Paulo State Network,
which is composed of 24 stations (Plus 3 fiducial stations). This network was integrated to
the ITRF-93 (International Terrestrial Reference Frame 1993) by using the fiducial
technique in conjunction with the IGS precise ephemerides (Fonseca, 1996). The results of
both approaches will be compared.

THE FIDUCIAL AND NON-FIDUCIAL APPROCHES

The Fiducial Technique

The fiducial GPS technique involves the observation of simultaneous (dual frequency)
carrier phase measurements at a network of fiducial stations, whose coordinates are known
to a very high order accuracy in a global reference frame, and a number of new stations
 whose coordinates are required. Full details of the method are given by Ashkenazi et a1
(1990), and only a brief outline is given here.
The fiducial process involves the computation of a theoretical (integrated) orbit for each
of the satellites, based on the forces acting on it. The forces modeled include gravitational
attraction (of the moon, sun and the planets, and earth and ocean tides), surface forces (e.g.
solar radiation) and other perturbing influences. The theoretical orbit for each satellite is
computed by numerical integration of the acceleration vector, starting from a initial state
vector (3-d position and velocities) obtained from a previous broadcast or precise
ephemeris. These theoretical orbits are improved during the network adjustment. The
complete set of carrier phase measurements from all the stations is adjusted by least
squares, holding the 3-d coordinates of a minimum of three fiducial stations fixed, and
solving for the 3-d coordinates of the new stations, as well as for the corrections to the
satellite orbits and other bias parameters (ambiguities, tropospheric scale factors, y-bias,
etc.). The resulting adjusted network of ground stations and satellite orbits is positioned,
scaled and oriented to the reference frame defined by the adopted coordinates of the
fiducial stations held fixed. The high accuracy of the coordinates of the fiducial stations are
then transferred to the new stations, via the satellite orbits.
With the establishment of the IGS in 1990 by the International Association of Geodesy
(IAG) , high quality GPS orbits and Earth rotation parameters of high resolution became
available to the scientific community, allowing to expand geographically the current
International Terrestrial Reference Frame maintained by the International Earth Rotation
Service (IERS) (Mueller, 1993). The precise ephemerides produced by IGS may be used in
the fiducial GPS technique, simplifying the approach. Instead of estimating the satellite
orbits, one can use the IGS ones and only the coordinates of the new stations have to be
estimated. Additionally, IGS can effectively provide fiducial stations coordinates at the
epoch of a GPS campaign. For example, such coordinates may be obtained from the most
recent realization of ITRF. In the context of this paper, the satellite orbits are not
estimated in the GPS fiducial technique. Instead of, the IGS precise ephemerides are used.

The Non - Fiducial Technique

The main difference between the non-fiducial and the fiducial technique is that in the former
one, besides the satellites state vectors and other bias, all station coordinates are estimated
in the network processing. The non-fiducial technique has been applied to global GPS
experiments, in which fiducial stations are not necessary to provide an origin and scale
(Heflin et ai, 1992; Blewitt et ai, 1992). They are provided by the satellite force model,
radio propagation model, and GPS data. This approach may also be suitable for regional
networks, but in such a case, instead of estimating the satellites state vectors, it is advisable
to use an IGS precise ephemerides. The adjusted network will be referenced to the datum
realised by the precise ephemerides. .
In order to make the results consistent with those of the fiducial approach, a
transformation to the reference frame of the stations used as fiducial has to be carried out.
It involves a similarity transformation as follows:

[X 1 -~'l[~]
-0 z
[YX] =Y [tx] + [ 0s
+t s
Z N_F Z ITRF t: -~ y ox s Z ITRF
(01)

where x, y and z are the Cartesian coordinates derived from the non-fiducial (N-F)
network;
X, Y and Z are ITRF Cartesian coordinates of the fiducial stations mapped
to the time of the campaign;
tx , ty and t z represent the offsets in origin;
s represents a difference in scale and
ex ,9 y and 9 z represent differences in orientation.

150
 In order to apply the transfonnation given by Equation (01), only a minimum of seven
coordinates values (i.e. minimum of three fiducial stations) has to be available. There are
also options of estimating less than seven parameters. Usually, only the translation
parameters (tx, ty and tz) are estimated.
Once the transfonnation parameters have been estimated, those stations not included in
the transfonnation may be transformed to the required reference frame. The transformation
is the last step involved in the non-fiducial network approach. It means that only this step
needs be carried out if the choice of the fiducial stations changes. This is an advantage over
the fiducial technique, where all the main steps must be recomputed.

PROCESSING SOFTWARES
GPS Analysis Software (GAS)
The data was processed using GPS Analysis Software (GAS) developed at the University
of Nottingham (Stewart et aI, 1994). GAS process either double-difference carrier phase
measurements or pseudo-ranges, formed with different base satellites for each baseline
processed. This provides a capability of processing large GPS network, even a global one.
The GPS network processing by GAS can be carried out either with the full fiducial
technique (Ashkenazi et aI, 1990) or by using the broadcast or the precise ephemerides for
the satellites. In all cases, corrections for the effects of the Earth body tide (EBT) and
Ocean tide loading (OTL) can be included in the processing. For the former case, the
model recommended in lAG standards was used (McCarthy, 1992). The effects of the
ionosphere can be greatly reduced by using the ionospheric free observable (Ffoulkes-
Jones, 1990). For the tropospheric delay, GAS provides a series of modeling options. For
example, one could define zenithal scale factors as unknown parameters. GAS also
provides the option of 'fixing' the double-difference carrier phase integer ambiguities to
their integer values. This is achieved by one of the two different techniques, namely the
ambiguity search technique and the sequential bias fixing (Ffoulkes-Jones, 1990).
The non-fiducial technique can easily be applied by using GAS. At such a case, no
fiducial stations are selected and all stations coordinates are estimated in the adjustment.
As the model has full rank, a solution will be obtained using conventional algebra.

Cartesian Network Adjustment Program (CARNET)

CARNET enables the integration of several types of conventional observations with GPS
and other space geodetic techniques. These observations can be adjusted together,
providing a 3-dimensional solution (Lowe, 1994). Several options can be invoked during
the network adjustment, e.g. the estimation of transfonnation parameters or their
application once they have been detennined.
The Cartesian coordinates or Cartesian coordinate differences estim.lted in the network
adjustment using GAS from several sessions or days can be combined together to produce
a final solution using CARNET. The final result is a set of weighted mean values. The main
output of CARNET is a list of Cartesian and geodetic coordinates with the respective
estimated standard deviations as well as infonnation related to the quality of the processing.
CARNET was used in this work to combine all daily GPS solutions.

DATA SET AND PROCESSING STRATEGIES

Sao Paulo State GPS Network
The data set used refers to the Sao Paulo State Network, which is composed of 24 stations,
plus station Chua, the origin of the Brazilian Geodetic System. The data was collected
during two main campaigns. The first one was carried out during the period of 7 to 18
March 1994 , using 7 dual frequency receivers. Each session was 8 hours long (8:00 a.m.
to 17:00 p.m. - local time) with 15 seconds sample rate. A total of 18 stations was
occupied. The second campaign took place during the period of 21 November to 3
December of 1994, using 5 (dual frequency) receivers. The duration of each session of this

151
campaign was reduced to 6 hours (8:30 a.m. to 14:30 p.m.), also with 15 seconds sample
rate. The number of occupied stations was 18.
In order to connect the network to a global reference frame, the data of 3 IGS stations
was transferred via Internet and introduced in the processing. They are the stations
KOURU, SANTIGO AND FORTALEZA, which are equipped with IGS standard
receivers. Figure 1 illustrates the Sao Paulo State network with the 3 fiducial stations.

PA.~

UEl'P
TAW +
+
'Got.
\ PE'JA -

...
.J..

Figure 1: Sao Paulo State Network with Connection to the IGS Stations

Table 1 summarises the data set used in the processing. Full details can be found in
Fonseca (1996).
Table 1: Information of the Data Set Involved in the Processing

~
' ~ 06S 07C 07 072 073 074
Day of Year (994)
07: 076 327 328 329 332 333 334 33~ 337
CHUA XX XX XX XX
AGUA XX XX XX
PRET XX XXi XX
JABO XX ~ ,~
PIRA XX XX XX XX
LIMO XX XX XX
FRAN XX XX XX
FERN XX XX XX
AVAN )a )a )a
IBIT )a XX XX XX XX
UEPP XX XX XX XX XX XX XX XX XX
TAOU XX XX
GRAN XX XX XX XX
S0I1 XX XX
SAOP XX )a XX XX
BELA XX XX XX
BUNA XX XX XX
VALl XX XX XX XX XX
PAUl. XX XX
REGI XX
INGA XX XX
PEVA XX XX
BOru XX XX
MARl XX XX XX
PANO XX XX
FORT(*' XX XX XX XX XX XX XX XX XX XX XX XX XX XX
KOUR(*) XX XX XX XX XX XX XX XX XX XX XX XX XX XX
SANT(* ~ XX XX XX XX XX XX
(*) lGS StatIons Wlth 24 hours of data (30 seconds sample rate)

152
Processing Strategies
Several processing strategies were carried out using the Sao Paulo State Network data set.
Details of each processing can be found in Fonseca, 1996. For the purpose of this paper,
the fiducial GPS network processing will be taken into account. In this processing, IGS
precise ephemerides were used and the coordinates of the three IGS stations were held fix
to the ITRF93. Earth Body Tide corrections based on the IERS standards were applied to
the observations (MaCarthy, 1992). The basic observable was the double difference
ionospheric free without solution (integer) to the ambiguities and a scale factor was
estimated in order to improve the tropospheric refraction model.
For the non-fiducial solution, the same models described above was used, unless of the
station coordinates, which were all estimated in the adjustment. In order to make the results
consistent, a transformation to the reference frame defined by the three fiducial stations
used in the fiducial approach was carried out. Only the 3 translations parameters were
estimated.

RESULTS AND DISCUSSIONS

The results obtained are illustrated in terms of repeatabilities, i.e. nns (root mean square) of
.the differences from the mean. The mean was obtained from all daily solutions, using
CARNET software. The repeatability shows the internal consistency of the results, i.e. the
precision.
Figure 2 illustrates the repeatabilities of the fiducial and the non-fiducial solutions. In the
both cases, the precision in Latitude, Longitude and Height are of the order of 1,3.5 and 5
cm respectively.

Figure 2: Precision of the GPS Fiducial and Non-Fiducial Solutions

As expected, the Height components provided the worse results, whilst the Latitude
components the best ones. From these results one can conclude that both approaches
provide similar results in terms of precision. In addition, the stations of Sao Paulo State
network were positioned with resulting precision better than 8 cm (1 sigma).
Figure 3 illustrates the discrepancies between both solutions for the X, Y, Z and h
(Height) coordinates components.
The maximum value reached 3.5 cm, in the Y component of station BOTU; almost three
times the average discrepancies of the other stations. There is not any clear explanation for
such discrepancy at this station. However, this station also provided the worse results in
tenns of precision. Taking into account the precision of the Y component at both solutions
(approximately 5 cm), this discrepancy is within an expected range. On the average, the
discrepancies for each of the components are smaller than 1 cm, and the precision smaller
than 5 cm.

153
 0.035
0.03
0.025
!: 0.02
I iXl
~
0.015
"c
.
;:;

~
0.01
Q.
0.005
"
U C h '
~
0
is
·0.005
·0 .0 1
·0 .015

Figure 3: Discrepancies in the X, Y, Z and h components

CONCLUSIONS

The discrepancies between the results of the fiducial and the non-fiducial GPS solutions can
be considered statistically insignificant, if one take into account the precision of both
solutions, which agree quite welL Therefore, both approaches can be applied in the
adjustment of GPS networks. However, the non-fiducial approach provides a more
advantageous way of integrating GPS networks. In this method, if one wants to change
fiducial stations in order to perform some tests, only a transformation needs be carried out.
Therefore, in such cases it is advisable to apply the non-fiducial approach.

REFERENCES

Ashkenazi, V., Moore, T., Ffoulkes-Jones, G.H., Whalley, S. and Aquino M. (1990). High
Precision GPS Positioning by Fiducial Technique. In: Bock Y., Leppard N. (eds): Global
Positioning System: An Overview, Springer, New York, Berlin, Heidelberg, Lond,
Paris, Tokyo, Hong Kong, pp 195-202, [Mueller I.I. (ed): IA G Symposia 102J.
Blewitt G., Heflin M. B., Webb F. H., Lindqwister U. 1. and Malla R. P. (1992). Global
Coordinates with Centimeter Accuracy in the International Terrestrial Reference Frame
Using GPS, Geophysical Research Letters, VoL 19, No.9, pp 853-856.
Ffoukes-Jones G.H. (1990) High Precision GPS by Fiducial Techniques, PhD Thesis,
University of Nottingham.
Fonseca E. S. Junior (1996) Estudo e Avaliarao MetodolOgica da Rede GPS do Estado
de Sao Paulo, MSc Thesis, Universidade de Sao Paulo, Escola Politecnica.
Heflin M., Bertiger W., Blewitt G., Freedman A, Hurst K., Lichten S., Lindqwister V.,
Vigue Y, Weeb F., Yurck T. and Zumberge 1. (1992b) Global Geodesy Using GPS
without Fiducial Sites, Geophysical Research Letters, VoL 19, NO.2, pp 131-134,
January, 24-1992.
Lowe D. P. (1994) CARNET User Guide. Version 3.13. IESSG Publication, University of
Nottingham.
McCarthy D. D. (1992) IERS Standards (1992), IERS Technical Note 13, Central Bureau
of IERS- Observatoire de Paris
Mueller 1. 1. (1993) The International GPS Service for Geodynamics: An Introduction.
Proceedings of the 1993 IGS Workshop, pp 1-2, Ed. by G.Beutler and E. Brockmann,
Univ. of Berne, 1993.
Stewart M. P., Ffoulkes-Jones G. H. and Ochieng W. Y (1994) GPS Analysis Software
(GAS) Version 2.2 User Manual, IESSG Publication, University of Nottingham, UK.
ACKNOWLEDGEMENTS

The first author has financial support of CNPq, the Brazilian Research Council, and
FAPESP, the Sao Paulo State Research Foundation.

154
 THE STATEFIX WEST AUSTRALIAN GPS NETWORK

M P Stewart
School of Spatial Sciences, Curtin University of Technology, Perth, Western Australia

H Houghton
Department of Land Administration, Western Australia

X Ding
Department of Land Surveying and Geoinfomatics, Hong Kong Polytechnic University

ABSTRACT

The STATEFIX project, commissioned and managed by the West Australian Department
of Land Administration, represents the densification of the Australian National Network
in West Australia to an average spacing of approximately 200km. Some 230 dual
frequency baselines (mean baseline length 202km) were observed between March 1996
and November 1996. Data processing and network adjustment were undertaken at Curtin
University of Technology. Final results indicate the network is accurate (relative to the
ANN) to better than 2cm in the horizontal component (95% confidence), and 6cm in the
vertical component (95% confidence). The STATEFIX project has been particularly
challenging in terms of the vast area surveyed (over 2,500,OOOkm 2) and the isolated nature
of many control points, located in uninhabited desert areas. This paper will outline
STATEFIX observation, processing and network adjustment strategies, and describe the
precision and accuracy analysis performed for the network.

INTRODUCTION

West Australia is one of the largest yet most sparsely populated states in the world and the
establishment a state geodetic control network has been an enormous challenge to
surveyors since settlement by Europeans in the early nineteenth century. The STATEFIX
project was conceived by the West Australian Department of Land Administration
(DOLA) ten years ago as part of an Australia-wide policy to improve geodetic control
using satellite-based positioning technology and aid the transition from the existing
Australian Geodetic Datum (AGD84) to a geocentric datum, the Geocentric Datum of
Australia 1994 (GDA 1994 0).
STATEFIX uses existing points of the Australian National Network (ANN) as a control
framework. The ANN itself represents a densification of the Australian Fiducial Network
(AFN) which consists of eight permanent, continuously operating, Rogue GPS receivers
on the Australian mainland and Tasmania (Manning and Harvey, 1992). The network was
initially observed during the International GPS Service for Geodynamics (IGS) Epoch '92
campaign, July to August, 1992. The AFN, in conjunction with six additional sites
beyond the Australian mainland, forms the Australian Regional GPS Network (ARGN).
AFN station co-ordinates are based on the ITRF92 at epoch 1994.0, and are estimated to
have a precision of a few centimetres (2-4 parts in 108).

The Australian National Network (ANN) consists of 78 GPS campaign points spaced at
approximately 500 km intervals across Australia. This network was observed between
1992 and 1994; the first GPS observation campaign being conducted during the IGS
Epoch '92 campaign and followed by a further nine days of observations during August-
September, 1993. It is estimated that the horizontal and vertical precisions of the co-
ordinates at the 95 percent confidence level are better than three centimetres and five
centimetres respectively (Morgan et ai, 1996). Figure 1 highlights the location of AFN
and AFN stations within Australia.

'~
120' 130' 100' ISO'

Figure 1 Australian Fiducial Network (AFN) and Australian National Network (ANN)

The STATEFIX project was conceived ten years ago but only executed in 1996. Several
reasons were used explain this delay. By 1996 technological advances in GPSreceiver
hardware and software, and greater availability of equipment and expertise, were
anticipated to ensure the objectives of the project could be easily achieved. The
specification set by DOLA (DOLA, 1996) is an absolute accuracy of Scm (95%
confidence) in X, Y, Z and latitude, longitude and height. Furthermore, minimum
ionospheric disturbance due to the low in 1996 of the eleven year sunspot cycle was
deemed important to ensure success when processing the long baselines required by
necessity for this project.

156
This paper outlines STATEFIX observation, processing and network adjustment
strategies, discusses precision and accuracy analysis performed for the network, and offers
recommendations for the future densification of regional networks

STATEFIX NETWORK DEFINITION AND DATA COLLECTION

The full STATEFIX network is illustrated in figure 2a. Baseline lengths range from 45km
to distances longer than 450km. In total, 228 baselines have been observed to 119 stations.
The mean baseline length is 200km.

• ST ATEFIX site
.& ANN site
r BR

-.....L~---- 4734

lOOOkm

Figure 2a The STATEFIX Network Figure 2b Diagram showing division of

STATEFIX Network into 9 cells

For the purposes of observation and processing, the STATEFIX network was divided into
9 individual cells (figure 2b). Because of the size of the STATEFIX network, cells were
observed by different private contractors over different time periods in 1996. Baselines of
less than 250km in length were observed for a minimum of 12 hours, baselines longer
than 250km in length were observed for a minimum of 16 hours. The observation interval
was 30 seconds for all baselines. A minimum of three receivers were in use for each cell
simultaneously. In total, 198 baselines were observed, with 30 of these being measured
twice as they constituted cell boundaries. Several other repeat baselines were observed
within individual cells.

157
All cells, with the exception of cell 3, were observed using Trimble 4000 SSI dual
frequency GPS receivers. Cell 3 was observed with Ashtech Z-12 dual frequency GPS
receivers. Cells 3 and 5 also incorporated data from the IGS permanent Rogue SNR-8
receiver at Yaragadee (YARl), whilst cells 5 and 6 incorporated data from a similar
receiver (though different antenna) at Karratha (KARR), which is part of the Australian
Fiducial Network (AFN).

DATA PROCESSING AND ADJUSTMENT

Baselines from each cell were processed at Curtin University using the in-house SWAG
(S.outh West Australian GPS software) GPS processing software suite. The following data
processing procedures/models were applied:

a) IGS Precise Orbits:

IGS precise orbits were held fixed in the processing of all cells.
b) Earth Body Tide Modelling:
The Wahr two-step Earth body tide model was applied, as per IERS standards
(McCarthy 1992) in the processing of all cells.
c) Antenna phase centre modelling:
The IGS PCV_01 antenna phase centre models (Rothacher and Mader 1996),
referenced to the Dome Margolin T chamber test values of Schupler (1994), were
applied.
d) Atmospheric Modelling:
Processing was undertaken using the ionosphere free linear combination (<PIF =
2.546<PL1 - 1.984<PL2)' Corrections to the zenithal delay tropospheric model parameters
in the form of scaling and second order rate of change parameters were estimated as
part of the baseline solutions.
e) Ambiguity Resolution:
Ambiguity resolution was attempted on all baselines using a combination of the wide
lane observable and the ionospheric free observable. Ambiguities were successfully
resolved on all STATEFIX baselines with the exception of one baseline in cell 8.
f) Length of Data Span:
The smaller cells, 1, 2 and 3, were processed by dividing the observation time spans
into two independent sessions and computing two solutions for each baseline. Cells 4 -
9 were processed using the full 16 hour session.
g) Rigorous Adjustment of Simultaneously Observed Baselines:
All simultaneously observed baselines were rigorously adjusted taking baseline
correlations into account. The resultant baseline vectors and full multi-baseline
covariance matrix for each day were then adjusted into the single cell solutions or full
STATEFIX network solution.
h) Individual Cell Adjustment
One ANN point was held fixed in each cell and coordinates of other ANN stations
within the cell were computed and compared with their ANN coordinates as an
accuracy estimation. Internal cell precision was also studied. Full results and analysis
from the individual cell adjustments can be found Stewart et at (1997).

158
i) Full Network Adjustment Strategy:
The full STATEFIX network was adjusted using GEOLAB version 2.4, constraining
all ANN station coordinates to 5cm in height and 3cm in plan (95% confidence).

4. RESULTS

i) Relative Errors
Figures 3a and 3b illustrate the relative errors from the STATEFIX network solution at
a 95% confidence level. Both the relative horizontal and height errors exhibit a baseline
length dependency with error of 0.0 1 - 0.04 ppm in the horizontal and approximately
0.2ppm in height. Relative horizontal errors are all better than lcm. Relative height
errors are better than 6cm with the exception of baselines connected to site R084 in cell
8, which demonstrate uncertainties of between 7 and 8 cm.

ii) ANN coordinate recovery

In addition to the full solution, a series of further adjustments were conducted which
were identical to the full solution except that one ANN point was left free, allowing a
direct comparison of STATEFIX coordinates with ANN 'ground truth' coordinates.
The results from the coordinate recovery tests are presented in table 1 overleaf. The
ANN site coordinates from the additional network adjustments are recovered to better
than 2.5cm (±l-2cm) in all components. The largest error is 5.5cm, in the X component
at eARN and the overall recovery of DEAK.
0.09
0,09
0.08

0.07
0,08 •
~
0,07
.
:[006
~
:[
0 ~ 0.06
~ 0.05 l:
~ ! 0.05
.
-a.
~

.::
0.04
'"
~ 0.D4
j 0.03 ~
0,03
f '"
~

0.02 f 0.02
0.01
0.01
0
SOOOOO 1000000 1500000 200000O 2SOOOOO 0
0 SOOOOO t 00000o 1500000 2000000 2500000
baseline length (m)
baseline length (m)

Figure 3a Relative Plan Error for Figure 3b Relative Height Error for
ST ATEFIX Network Solution STATEFIX Network Solution

SUMMARY

The precision of the adjusted STATEFIX station coordinates within the ANN is better
than lcm in the horizontal and 6cm in height (95% confidence). ANN station recovery
tests retrieved ANN coordinates with an accuracy of better than 2.5cm ± l-2cm. The
relative precision of the STATEFIX network is better than 0.04ppm in the horizontal and
O.2ppm in height. Given that the published formal error on the ANN, the reference

159
framework for STATEFIX, is 3cm in the horizontal and 5cm in height (95% confidence),
the accuracy of the STATEFIX network can be said to be of a similar order of magnitude
to the ANN.

The STATEFIX network has illustrated that given adequate planning, high precision GPS
control networks can be established over large, remote regions. The success of the project
can be mainly attributed to the existing geodetic control infrastructure offered by the
Australian National Network and the long observation spans which ensured adequate data
were available to resolve ambiguities, even on long baselines.

Table 1 Recovery of ANN Coordinates from STATEFIX Network Solutions

ANN station recovery X (m) recovery Y (m) recovery Z (m)

BATE 0.005 0.015 0.019
CAIG 0.007 0.037 0.002
CARN 0.055 0.031 0.013
COOL 0.010 0.026 0.007
DEAK 0.056 0.040 0.060
ESPE 0.002 0.020 0.020
KILl 0.010 0.023 0.025
PER2 0.001 0.007 0.013
PIVT 0.012 0.010 0.034
RATH 0.020 0.023 0.005
YARI 0.025 0.038 0.015
mean 0.018 0.020 0.019
standard deviation 0.019 0.011 0.016

References

McCarthy, D D (ed.), 1996. IERS Conventions. IERS Technical Note 21, Central Bureau
of IERS - Observatoire de Paris. France.
Morgan P, Y Bock, R Coleman, P Feng, D Garrard, G Johnston, G Luton, B McDowall,
M Pearse, C Rizos and R Tiesler, 1996. A Zero Order GPS Network for the Australian
Region. Report ISE TR 96160, University of Canberra. 183pp
Rothacher, M, and G Mader, 1996. Combination of Antenna Phase Centre Offsets and
Variations. Antenna Calibration Set: IGS_Ol. IGS Central Bureau
(http://igscb.jpl.nasa.gov)
Schupler, B R, R L Allshouse and T A Clark, 1994. Signal characteristics of GPS user
antennas, Navigation, 41(3), pp277-295.
Stewart, M P, X Ding, M Tsakiri and W E Featherstone, 1997. 1996 STATEFIX project,
final report. West Australian Department of Land Administration Internal Report. 47pp
(available on request).

160
 POLREF-96
THE NEW GEODETIC REFERENCE FRAME
FOR POLAND

J.B. Zielinski, A.Lyszkowicz, L.Jaworski, A.Swi~tek, R.Zdunek

Space Research Centre, PAS, Department of Planetary Geodesy
00-716 Warsaw, Bartycka 18A, Poland

S.Gelo
Head Office of Geodesy and Cartography
00-926 Warsaw, Wspolna 2, Poland

ABSTRACT. The principles of the new National System of Geodetic Coordinates are the
result of analyses carried out by the study group of the Section of Geodetic Networks and
the Section of Cartography Geodetic Committee of the Polish Academy of Sciences. As
the basis of the new National System of Geodetic Coordinates the European Terrestrial
Reference System (ETRF'89) and ellipsoid GRS-80 were adopted. Practical realisation of
the new reference system was carried out in the Department of Planetary Geodesy, Space
Research Centre using GPS technique. In 1992 precise zero-order network EUREF-
POL'92 containing eleven stations was established. In 1994 and 1995 its further
densification (356 points) provided POLREF network which is practical realisation of
ETRF'89 Reference System in Poland. The paper gives a survey of all works connected
with the establishment of the new National Geodetic System in Poland.

OVERVIEW OF CLASSICAL TRIANGULATION NETWORKS IN POLAND

The process of creating the horizontal control networks in Poland has a very long and very
complicated history. That history is a reflection of Poland's fate as an independent state.
With the reference to historical events, the development of the Polish primary horizontal
control net may be divided into the following steps: the period before 1918, the twenty
year long inter-war period (1918-1939) and the period after 1945 (Baran,1990).
After 1945 the basic horizontal networks in Poland and in the whole Eastern Europe
were related to the Krassovski reference ellipsoid with the initial point in Pulkovo
Observatory. In relation to this geodetic datum two adjustments of the Polish networks
were made in 1957-58 and in 1983 (fig.1). National reference system based on Krassovski
ellipsoid is called ,,1942". It has been modified in 1983 as System "JSAG-1983".
Finally the Polish national classical horizontal network consists of the asto-geodetic
network having about 55 stations and of the filling network having about 6000 stations.
The astro-geodetic network was created on the basis of fragments of different networks
which were established on the territory of Poland during over eighty years (1899 - 1981).
Densification of the network was carried out in 1955-1970. The results of the adjustment
(Gaidzicki et aI.,1983) show that the average value of a mean relative errors (md ) of
distances are as follows:
• in the case ofastro-geodetic network md is of the order of2.5 - 0.65 ppm,
• in the case of the filling network md is ofthe order of 0.45 - 0.31 ppm.
It indicates that network is rather quite good. Further comparisons done twenty five years
later showed that the classical network has significant deformations.
Further development of the horizontal geodetic control networks in Poland is connected
with the rapid development of satellite observations techniques. In 1983 first experiments
with incorporation satellite Doppler technique to strengthen classical network in Poland
were carried out. Finally in 1988 the National Doppler Network consisting of ten station
was established in Poland. The GPS technique introduced to geodesy in early eighties
amounted true revolution in technology of establishing control geodetic networks.

Fig. 1 Polish primary horizontal control network as in 1955 and as in 1983

162
EUREF-POL'92 GPS OBSERVATION CAMPAIGN AND DATA PROCESSING

Introduction. First concept of the extension of the EUREF network over Poland's territory
was presented in 1991 (Baran et aI., 1991). It contained eleven stations. Some of them were
located in geodetic observatories (Borowiec, Borowa G6ra, J6zefoslaw, Lamk6wko).
Other points of the EUREF-POL network coincided with the National Doppler Network
1986 (Baran,1992). An official presentation of the Polish part of the EUREF network took
place in March in Bern, during the EUREF Symposium (Baran, Zielinski, 1992). Also in
Bern the term and technical conditions of the EUREF-POL'92 campaign were set up.
Observation campaign. The campaign was successfully performed in the period of time
from July 4 to 8, 1992 with a one day calibration session in Borowiec. Measurements were
made with Trimble 4000 ST receivers, during four days with 10.5 hours of observations
each day and one day with four hours session. The observations were carried out with an
elevation cut-off angle of 10 degree and a sampling rate of 15 seconds. Among 19 foreign
stations there were 15 reference points and 4 newly-positioned points; two of them on
German territory (Karlsburg and Kirschberg) and two on the territory of Lithuania
(Akmeniskiai and Meskonys). Additionally not planned point Predni Pricka in Czech
Republic was included in EUREF-POL'92 network. The Institut fur Angewandte Geodasie
(IfAG) had to include 3 additional stations Schonfeld and Kaminke and the station Simeis
(Crimea) on the request of Russian colleagues.
Data processing. The data were processed by the team from the Space Research Centre
(SRC) and simultaneously by the team from the IfAG in Leipzig. EUREF-POL'92 GPS
network was computed with Bemese Software Version 3.4. Strategy of data processing,
coordinates of the fixed stations, precision of processing and differences between SRC and
IfAG solutions were accounted for in details in (Zielinski et aI.,1994).
Final solution. The final solution of EUREF-POL'92 campaign is a combined solution
obtained from two sets of coordinates (SRC and IfAG solutions) and their covariance
matrices. It was expressed in ETRS89 coordinate system. The rotation to the epoch 1989.0
was computed in ITRF91 with individual velocities as far as available or with velocities of
the European plate motion model NNR-NUVELl. Then the transformation from ITRF91
epoch 1989.0 to ETRF89 was done with a seven parameter Helmert transformation. The
final coordinates in ETRF89 at epoch 1989.0 were given in (ibid. table 8). The precision
obtained from the daily solution repeatability was: 0.007m for north, 0.012 m for east and
0.022 m for height components.

DENSIFICATION OF THE EUREF-POL'92 NETWORK

Introduction. Just after the establishment of the EUREF-POL'92 network, a project of a

new primary geodetic network POLREF was launched. The project assumed that 356
points of the new network should be positioned using GPS technique. Relative accuracy of
position should do not exceed 1 - 2 centimetres. Points of the classical astro-geodetic
network, where it is possible, should be included to the new network. Points of the
POLREF network were monumented to ensure proper conditions for carrying out GPS,
levelling and gravimetric observations.

163
 Observation campaign. Network was observed in three campaigns from July 1994 to
May 1995. In summer and autumn 1994 Space Research Centre organised two GPS
campaigns. As a result of these campaigns 178 points were observed and about 50 % of
the territory of Poland was covered with the new network. In spring 1995 the Institute of
Geodesy and Cartography carried out the third GPS campaign and observed 178 new
points. Common rules were adopted in all three campaigns, namely, length of session 220
minutes and at least two sessions at each POLREF point. GPS measurements were carried
out with Turbo Rogue and Wild-Leica 200 receivers. The Turbo Rogue receivers were
used for permanent observations at five reference points of EUREF-POL'92 network. The
Wild Leica 200 receivers were used at the new points of POLREF network.
Data processing and analysis. Originally, each sub-network was separately adjusted
using Trimvec and Bernese software. In the next stage, the east part was solved as one sub-
network and the final solution was obtained by linking the eastern and western part with
the Bernese Compar routine. Accuracy of the POLREF network meets the demands of the
EUREF densification network. The standard deviation of a single observation is 0.39 cm,
standard deviation of dB, dL, dh is 0.5 - 1.0 cm, 0.5-1.0 cm, 1.0 - 1.5 cm respectively. The
scattering of the errors is rather regular.

TRANSFORMATION PARAMETERS BETWEEN "JSAG-1983" AND EUREF-89

The classical horizontal network in Poland is related to Krassovki ellipsoid and geodetic
datum Pulkovo. Any strengthening of the classical network by GPS requires that the
possible distorsions in classical network should be investigated beforehand. An obvious
way of studying the relative distorsions is to compare the geodetic coordinates of common
points in two networks. It requires the knowledge of the transformation parameters
between two datum. The are several datum transformation formulas for accomplishing
that.

Fig. 2. Residual vectors after 7-parameter transformation of JSAG-1983 to EUREF -89

164
The most common is 7-parameters transformation which requires the knowledge of the
geoid/ellipsoid separations. In 1993 the first gravimetric geoid computed in Poland was
fulfilled in the Department of Planetary Geodesy. Since that time the number of available
gravity data increased considerably and the last geoid solution by the spherical FFT
methods comprises more than 130 000 mean/point gravity anomalies.
Last modification of the Polish National Reference System "JSAG-1983" has been
compared with the new satellite network. Because almost all stations of POLREF network
are identical with the classical triangulation points thus it was possible to make a
comparison of the networks by 7-parameter transformation.". The final coordinates of 339
POLREF points were used to determine the transformation parameters between networks.
Classical network was of course two dimensional and for mutual transformation normal
heights of the classical network had to be determined. It was done by spirit levelling. The
quasigeoidal heights used for converting the normal heights of the triangulation stations to
ellipsoidal heights were taken from (Lyszkowicz and Forsberg,1995). The residuals after
the transformation of the reference system "JSAG-1983" into EUREF-89 are displayed in
fig.2. Distribution of the differences are very regular and only in the north-west part small
deformations can be observed. The maximum residuals in this case are 32 cm in horizontal
position and 20 cm in height. It means that classical observations have a good quality and
could be used in the future.

ACKNOWLEDGEMENTS

The final support provided by the Committee of Scientific Research grant PBZ-08-07 is
gratefully acknowledged.

REFERENCES

Baran, W. (1990), History of creating and improvement of triangulation network in

Poland, Proceedings of XVIX Congress of International Federation of Surveyors in
Helsinki, pp.101.2/3-101/.2/14
Baran, W. (1992), GPS Network for Poland - Status and future plans, Veroffentlichungen
der Bayerischen Kommission fiir die Internationale Erdmesssung der Bayerischen
Akademie der Wissenschaften, Astronomisch-Geodatisch Arbeiten, Heft Nr 52,
~tinchen,pp. 117-123
Baran, W. (1994), New National System of Geodetic coordinates in Poland, Geodezja i
Kartografia (Geodesy and Cartography), vol. XLIII, No.1, pp.51-55
Baran, W., Sledzinski, 1., Barlik, ~., Czarnecki, K., Dobrzycka, M., Gajderowicz, 1.,
Oszczak, St., Pachelski, W., Rogowski, 1. Cisak 1., (1991), Raport 0 potrzebie i zasadach
zalozenia podstawowej zintegrowanej sieci geodezyjnej Polski w ukladzie EUREF,
Raport Komitetu Geodezji PAN, Sekcji Sieci Geodezyjnych, Warszawa
Baran, W., Zielinski, J.B. (1992), Design and preparation of the EUREF-POL'92 GPS
Campaign in Poland, VerOffentlichungen der Bayerischen Kommission flir die

165
 Internationale Erdmesssung der Bayerischen Akademie der Wissenschaften,
Astronomisch-Geodatisch Arbeiten, Heft Nr 52, Miinchen, pp. 233-235
Baran, W., Zielinski, J.B. (1993), Status Report on participation of Poland in creation of
European Reference System _ EUREF, Veroffentlichungen der Bayerischen Kommission
fur die Internationale Erdmesssung der Bayerischen Akademie der Wissenschaften,
Astronomisch-Geodatisch Arbeiten, Heft Nr 53, Miinchen, pp. 160-163
Baran, W., Zielinski, J.B. (1994), The Use of EUREF-POL Network for Scientific and
Practical Purposes, Veroffentlichungen der Bayerischen Kommission fur die
Internationale Erdmesssung der Bayerischen Akademie der Wissenschaften,
Astronomisch-Geodatisch Arbeiten, Heft Nr 54, Miinchen, pp. 327-330
Baran, W., Zielinski, J.B. (1995), Realisation of the GPS Primary Network for Poland, -
Status Report, VerOffentlichungen der Bayerischen Kommission fur die Internationale
Erdmesssung der Bayerischen Akademie der Wissenschaften, Astronomisch-Geodatisch
Arbeiten, Heft Nr 56, Miinchen, pp. 184-186
Baran, W., Zielinski, J.B. (1996): Polish National Report on EUREF Activities in 1995-
1996, Veroffentlichungen der Bayerischen Kommission fur die Internationale
Erdmesssung der Bayerischen Akademie der Wissenschaften, Astronomisch-Geodatisch
Arbeiten, Heft Nr 57, Munchen, pp. 288-289
Gaidzicki, J., Bokun, J., Derylo-St~pniak, J., Gelo, St. (1983), Present Status of the
Primary Horizontal Control Network in Poland (in Polish), Proceedings of the 3rd
Geodetic Symposium on Actual Problems of Primary Control Networks, Warsaw
Jaworski, L., Lyszkowicz, A., Gelo, St. (1996), Comparison of the l-st order national
network POLREF with the classic triangulation, Paper presented at the symposium of
lAG Subcommission for the European Reference Frame (EUREF) in Ankara, Turkey
Jaworski, L., Gelo, St., SwiCltek, A., Zdunek, R. Zielinski, J.B. (1996), Adjustment results
of the new 1st order national network POLREF, Paper presented at the symposium of
lAG Subcommission for the European Reference Frame (EUREF) in Ankara, Turkey
Lyszkowicz, A., Forsberg, R. (1995), Gravimetric Geoid for Poland Area Using Spherical
FFT, lAG Bulletin d'Information N.77, IGES Bulletin N.4, Special Issue, Milano,
pp.l53-161
Zielinski, J.B. , Jaworski, L., Zdunek, R., Engelhardt, G., Seeger, H., Toppe, F. (1993),
EUREF-POL 1992 GPS Campaign and data processing, Veroffentlichungen der
Bayerischen Kommission fur die Internationale Erdmesssung der Bayerischen Akademie
der Wissenschaften, Astronomisch-Geodatisch Arbeiten, Heft Nr 53, Miinchen, pp. 92-
102
Zielinski, J.B., Jaworski, L., Zdunek, R., Engelhardt, G., Seeger, H., Toppe, F., Luthardt,
l (1994), Final Report about EUREF-POL 1992 Campaign, Veroffentlichungen der
Bayerischen Kommission fur die Internationale Erdmesssung der Bayerischen Akademie
der Wissenschaften, Astronomisch-Geodatisch Arbeiten, Heft Nr 54, Miinchen, pp. 92-
99
Zielinski, lB., Jaworski, L. (1995), Preliminary Comparison of the Classical and GPS
Reference Network in Eastern Poland, GIPSY-OASIS II Course, 17-21 July, Pasadena
(USA)

166
 AN OVERVIEW OF THE SIRGAS PROJECT

Luiz Paulo Souto Fortes

President of SIRGAS Committee
Departamento de Geodesia
Funda~ao Instituto Brasileiro de Geografia e Estatistica - IBGE
Av. Brasil 15671, Parada de Lucas, Rio de Janeiro, RJ, Brazil, 21241-051
Fax: 55-21-4812747; e-mail: [email protected]

ABSTRACT

A general description of the South American Geocentric Reference System (SIRGAS)

Project is presented. Since its establishment, in October 1993, the two working groups
("Reference System" and "Geocentric Datum") of the project have carried out the tasks
towards defining a new geocentric reference system and an associated Datum for South
America. The achievement of the project objectives were scheduled to be reached on 1997
and the corresponding final reports presented at this lAG Scientific Assembly. This was
only possible due to the very high cooperation of many groups, as: the South American
countries; lAG, IPGH and NIMA, the sponsors; and many others from North America and
Europe. The development of the project was based on a continental GPS network,
established during a very successful campaign in 1995. With its conclusion, South
America has now an unified reference system, with an accuracy compatible with the up to
date positioning techniques, what is of a great benefit to all countries in the continent.

NOTE

This paper is published in "IBGE. SIRGAS Final Report, Working Groups I and II.
Rio de Janeiro, 1997".
 THE DEFINITION AND REALIZATION OF THE REFERENCE
SYSTEM IN THE SIRGAS PROJECT

Hoyer M '.; Arciniegas S.2; Pereira K.J; Fagard H.4; Maturana R.!; Torchetti R.6;
Drewes H.'; Kumar M.8; Seeber G."

SIRGAS Project Working Group I

ABSTRACT

This paper describes the different stages executed to establish the Reference System to be
used in the definition of the geocentric datum in South America, in the frame of the
SIRGAS project.
The Working Group I developed the activities to plane, to prepare and to coordinate the
GPS measuring campaign, besides the processing organization to be executed by the
calculating centers. The analysis and decision about the final solution, represented by the
geocentric coordinates of the GPS network, are the last phase of the work.
The GPS observations were carried out from May 26 to June 04, 1995. 58 principal
stations were observed in 11 countries. The campaign was possible due to the collaboration
of around 30 participant institutions from South, North America and Europe. DGFI
(Germany) and NIMA (USA) were the processing centers.

INTRODUCTION

The objective of this paper is to describe the procedure employed by Working Group I of
the SIRGAS Project to define the new geodetic reference system for South America and to

1 Melvin Hoyer; Universidad del Zulia, Apartado Postal 10311; Maracaibo, Venezuela; Fax 58-61-512197; e-mail:
[email protected]
2 Susana R. Arcienagas; Instituto Geografico Militar Eldorado, Edificio del IGM; Quito Ecuador
3 Katia Pereira; IBGEIDEGED; Av. Brasil 15671, Parada Lucas; Rio de Janeiro, RJ, Brasil, CEP 21241-051
4 Herve Fagard; Institute Geographique National SGN; BP 68; 94160 Saint Mande, France
5 Rodrigo Maturana; Instituto Geografico Militar; Nueva Santa Isabel 1640; Santiago, Chile
6 Ricardo A Torchetti; Instituto Geografico Militar; Cabildo 381; 1426 Buenos Aires, Argentina
7 Hermann Drewes; DGFI,ABT.I; Marstallplatz 8; D-80539; Muenchen, Germany

8 Muneendra Kumar; Defense Mapping AgencyIIOG; 4600 Sangamore Road; Bethesda, MD 20816, USA
9 Gunter Seeber; Institut fur Erdmessung, University of Hannover; Schneiderberg 50; D-30167 Hannonver, Germany
establish the corresponding reference frame.
At the first workshop of the project in Asuncion, Paraguay (October 1993), it was agreed
that the reference system should coincide with that of the IERS -International Earth
Rotation Service - and that the reference frame should be realized by means of the
observation of a highly precise network of GPS stations.
The GPS observation campaign was carried out according to the schedule during ten days
from May 26 to June 4, 1995. A total of 58 stations were observed. After the collection
and preparation of the observation data, which was mainly done at DGFIII in Munich,
DGFI (Deutsches Geodaetisches Forschungsinstitut), IBGE (Instituto Brasileiro de
Geografia e Estatistica), and NIMA (National Imagery and Mapping Agency) started with
the data processing.
At the workshop in Santiago (August 1996), the preliminary results of these three
processing centers were presented and discussed. Important decisions were then made with
respect to the pending computations and how they should be completed.
DGFI and NIMA presented their final results in Margarita (Venezuela) in April 1997. At
this workshop the procedure for the final, unique solution was defined, and the
corresponding computations were done during the workshop.
This paper is dealing with the above mentioned activities and results of the Working
Group I. It includes the most important aspects of the GPS observation campaign, some
aspects of the processing and the procedure to obtain the definitive results.
The evaluation of the results and the quality of adjusted station coordinates allows the
conclusion that the Working Group I accomplished completely the planned objectives.
South America has at its disposal one of the most precise continental networks which will
serve as a basis for the establishment of a geocentric datum. The immediate task is now to
provide the adequate maintenance.

GPS OBSERVATION CAMPAIGN OF THE SmGAS REFERENCE FRAME

The idea to observe a continental GPS station network and the planification of its
configuration started from the beginning of the project at the Asuncion meeting in October
1994.
The assumed initial criterion for the selection of the stations was to include in the network
all the existing LASER, VLBI, DORIS, and GPS observatories in South America. In
addition one had to look for a homogeneous continental coverage of the net, to guarantee
the facile access to the sites and the possibility to perform the GPS measurements. Finally
there had to be some overlapping with the official geodetic network of each country.
SIRGAS Working Group I, with the valuable collaboration of the scientific consultants,
presented the technical specifications necessary for the GPS measurements in the SIRGAS
campaign. These were elaborated taking into account all the aspects which, according the
specialist's opinion, had to be included in a project like this.
Before the observations, discussions were carried out and decisions were made with
respect to the compatibility and availability of equipment to be used. The idea was to
guarantee homogeneity in the technical generation of the instruments and to include as
many geodetic GPS receivers as possible from South American institutions. On the other

169
hand, only highly precise receivers should be employed. Four receiver types were selected
for this purpose: Ashtech Z12,Leica 200, Rogue/Turbo Rogue, and Trimble SSE (SSI).
This selection includes the receivers operating as part of the global IGS network in the
regIOn.

A principal difference between receiver types with regard to precise positioning is the
relative location of antenna phase centers. The exact location of the phase centers has to
be known in the data processing in order to reduce the coordinates to the station's
monument marker. If only one type of receiver and antenna is used in a campaign, the
relative location of phase centers is identical in all stations and cancels in a relative
positioning (with respect to a reference station or in a coordinate difference). If different
receiver types are used, the variation of phase center locations between their antennae has
to be corrected, i.e. the phase center corrections have to be known for each receiver type
antenna.
Since the antenna phase center corrections of the employed receivers were not known in
an internationally accepted way (nowadays an IGS recommendation is available), it was
decided to co-locate different types of the selected receivers at several sites and to
determine their exact three-dimensional distances by local tie measurements. For that
reason, nine sites were occupied by two or three different receiver types.
In addition to this, the handling of the observation data was organized. A data center was
selected in each country, and two global data centers for the collection of the observations
from all countries were established.
The observations were carried out as planned from May 26, 0:00 UT until June 4, 24:00
UT, 1995, i.e., over 10 days in total. Finally, 58 principal stations and 9 eccenters in 11
countries were observed:

Country No. of stations Country No. of stations

Argentina 10 French Guiana 1

Bolivia 6 Paraguay 2
Brazil 11 Peru 4
Chile 8 Uruguay 3
Colombia 5 Venezuela 5
Ecuador 3 -----------------
Total 58

The location of all sites occupied during the SIRGAS campaign is displayed in figure 1

FINAL COORDINATES OF THE SIRGAS 1995.4 REFERENCE FRAME

Independent coordinate solutions of the SIRGAS GPS campaign were computed by DGFI
(DGFI, 1997) and by NIMA (NIMA, 1997). DGFI used the Bernese software version 3.4
with some modifications implemented at DGFI, and NIMA used the GIPSY-OASIS II
software. The station coordinates computed in the first step by DGFI and by NIMA are

170
 300' 320'
20' 20'
D ',
#
,,

O '~-------+----~------~~--~-----+----~~----+-----~ O '

- ---1. -20·
EIRO

• ISLA DE PASCUA

-4 0 '~----l--------1----j'-f----f-+-------+------I -40 '

... A liTEC ..

• LEICA
• ROC E
• TR IMBLE

- 60 ·~----l--------1------+----~--+-----Jl - 60 ·

260' 280' 320·

Fig 1: Stations of the SIRGAS reference frame and receiver types of the occupation during
the GPS campaign May/June 1995

171
nearly unconstrained solutions and refer to the IERS (International Earth Rotation Service)
Terrestrial Reference Frame 1993 (ITRF93) epoch 1995.4 by fixing the satellite orbits in
this frame. DGFI used the International GPS Service (IGS) precise combined orbits and
clock offsets while the NIMA results were computed with the JPL orbits and clock
parameters.
After applying a seven parameter similarity transformation (Helmert transformation), the
maximum difference between the DGFI and NIMA coordinates was 3.5 cm. The rms
differences between the solutions were ± 1.0 cm in X, ± 1.4 cm in Y and ± 0.7 cm in Z.
DGFI's solution includes a correction for the elevation angle dependent phase center
variation of each type of used receiver antennae as proposed by IGS (model IGS_01). This
correction is not included in the NIMA solution. The error caused by not including this
correction reduces in principal to a constant coordinate (mainly height) offset for each
antenna type because of the (nearly) 24 hours permanent observations and consequently
full coverage of the satellite passes under all elevation angles. It was therefore decided to
combine the two solutions in the following manner:

1. Compute a separate seven parameter similarity transformation between the DGFI

solution and each of the five receiver antenna type sub-networks (Ashtech P 700228,
Ashtech P 700718, Leica SR299, Trimble 4000ST/TR GP, Dome Margolin) of the
NIMA solution. This will account for the small differences in the antenna corrections as
well as the small reference frame differences.

2. Apply the transformation to the five antenna type subsets of the NIMA coordinates. The
rms differences between the DGFI coordinates and the transformed NIMA coordinates
reduce then to ± 0.7 cm in X, ± 0.9 cm in Y and ± 0.6 cm in Z.

3. Compute the mean of the DGFI coordinates and the transformed NIMA coordinates to
produce a combined set of coordinates in the ITRF93.

The final coordinates should be given in the ITRF94 (the most recent realization of the
ITRF up to date) according to a previous decision of the SIRGAS Committee. In order to
accomplish this, a subset of SIRGAS stations identical with global stations independently
determined in the ITRF94 is required. Using these stations, transformation parameters can
be computed to convert the SIRGAS coordinates from ITRF93 to ITRF94.
Only four stations (Arequipa, Fortaleza, Kourou and Santiago) on the mainland of South
America are included in the official ITRF94 solution of the IERS with station coordinates
at epoch 1993.0 and velocities to extrapolate to epoch 1995.4. This was considered too
few to derive a good seven parameter transformation. Two nearby SIRGAS stations with
IERS-determined ITRF94 coordinates and velocities (Easter Island and O'Higgins) outside
the mainland may be added to increase the number of stations identical with the SIRGAS
reference frame. Doing so, however, considerable distortions were found in the network.
The International GPS Service (IGS) is computing station coordinate and velocity
solutions in the ITRF more frequently than the IERS. At the time of the final SIRGAS
computations, two 1996 solutions of the IGS Analysis Centers CODE (Bern) and JPL
(Pasadena), respectively, were available, both referring to the ITRF94 and including three

172
additional SIRGAS stations on the South American mainland (Bogota, Brasilia and La
Plata). The CODE and JPL solutions are computed with the Bernese and GIPSY software,
respectively. Since this software is almost identical with that used by DGFI and NIMA,
respectively, it was decided to use the average CODE and JPL station coordinates
(referring to ITRF94 and extrapolated to epoch 1995.4) as fiducial coordinates. The
maximum component differences between the CODE and JPL coordinates for the nine
stations are 3.6 cm in Bogota and Easter Island and 3.8 cm in Santiago.
The final SIRGAS solution was computed by a seven parameter similarity transformation
between the mean CODE/JPL coordinates in ITRF94 at epoch 1995.4 and the combined
DGFIINIMA coordinates using the nine IGS stations. The transformation parameters were
then applied to the combined DGFIINIMA coordinates to produce the final SIRGAS
coordinates in ITRF94 at epoch 1995.4. The result is given in Table 2.
The final SIRGAS coordinates and their corresponding accuracies are presented in the
Final Report of SIRGAS Project (SIRGAS Final Report, 1997). The coordinates of the
network have r.m.s. errors with an average of± 4 mm in each component.

CONCLUSION

A cooperative effort of many institutions and countries permitted the definition of a modern
geodetic reference system for South America coinciding with that of the IERS. The
reference frame of this system is a highly precise network of 58 GPS stations. The
coordinates of the network have r.m.S. errors with an average of ± 4 mm in each
component.

REFERENCES

NIMA, Publication GIMGB-97-1603, Fundamental Station Coordinates for the SIRGAS

Continental Geodetic Control Networks, 1997.
DGFI, Abt.l, Report on the Processing of the SIRGAS 95 GPS Network at DGFI, 1997.
SIRGAS Final Report, September 1997

173
 TIME EVOLUTION OF THE SIRGAS
REFERENCE FRAME

Hermann Drewes
Deutsches Geodatisches Forschungsinstitut, Abt. I (DGFI/I)
Marstallplatz 8, D-80539 Munchen, Germany

ABSTRACT

The geodetic reference system for South America established by the SIRGAS project and
adopted by the South American countries provides a set of some 60 stations coordinates over
all the continent for the epoch of its observation in May/June 1995 (1995.4). To preserve the
consistency with the ITRF, station coordinates have to change in time according to the ITRF
kinematic reference frame. The motion of the South American continent cannot completely
be modelled by a rigid plate rotation after current models but has to account for extended
continuous deformations, in particular along the Andes mountain range. It is proposed to
determine individual station velocities of all SIRGAS sites and to derive a continuous
velocity field from a combined rigid plate and regional deformation model for the
transformation of new stations into the SIRGAS reference frame.

INTRODUCTION

The SIRGAS (SIstema de Referencia Geocentrico para America del Sur) reference system
was adopted by the official representatives of the South American countries in the SIRGAS
project during a workshop held in Isla de Margarita (Venezuela) in April 1997. It is defined
according to the International Earth Rotation Service (IERS) conventions (McCarthy 1996)
using for its realization the IERS Terrestrial Reference Frame (ITRF94, Boucher et al. 1996).
The realization of the SIR GAS reference frame is done by a total of 67 sets of station
coordinates at 58 sites for the epoch of the GPS observation campaign in May/June 1995
(epoch 1995.4). The materialized stations at the Earth's surface, however, are moving due to
recent crustal movements. In the ITRF which is also used for precise satellite orbit
parameters, e.g., those of the International GPS Service for Geodynamics (IGS), the station
motions are taken into account by providing individual station velocities in addition to the
epoch station coordinates.
 Therefore, the maintenance of the SIRGAS reference frame has to include - besides the
physical maintenance of the monumented sites - the time evolution of the coordinates too, in
order to guarantee the consistency between the terrestrial (SIRGAS) and the satellite
reference system. For this purpose, station velocities V (Le., coordinate changes dX/dt,
dY/dt, dZ/dt) are required for each station. These velocities may either be derived from
repeated observations and coordinate determinations, or from crustal deformation models. To
include additional stations observed at other than the SIR GAS epoch, we need a continuous
velocity model to transform their coordinates to the SIRGAS epoch.

STATION VELOCITIES FROM REPEATED POSITIONING

The stations of the International GPS Service for Geodynamics (IGS) included in the
SIRGAS network are equipped with permanently operating GPS receivers. The observed
data sets are routinely evaluated providing station coordinates on a weekly basis. The
organization of the IGS distinguishes two levels of stations (IGS 1997): IGS global stations
and IGS regional stations. The observations of the global stations are processed by seven
Analysis Centers (AC) and combined to one common solution by three Global Network
Associate Analysis Centers (GNAAC); observations of the regional stations are processed by
six Regional Network Associate Analysis Centers (RNAAC). All the sets of data, global and
regional, are combined to a so-called polyhedron solution (P-SINEX) by two GNAACs
including presently more than 120 stations.

For South America, the RNAAC SIR (Le. SIRGAS) operated by Deutsches Geodaetisches
Forschungsinstitut (DGFI) is processing all the available data from the existing permanently
observing stations in this region. These include presently (September 1997) sixteen SIRGAS
stations (Seemiiller and Drewes 1997). The result is then combined with other solutions by
the GNAACs to the global polyhedron. In this way we get weekly coordinates for all the
included permanently observing SIRGAS stations in the ITRF and can thus derive station
velocities. It is strongly recommended to join this procedure by installing as many permanent
GPS receivers as possible at SIRGAS sites. More information may be requested from IGS or
the RNAAC SIR.

To derive the velocities of stations not equipped with permanent GPS receivers we need
at least one (better more than one) repetition measurement (GPS campaign) at time t; for
coordinate determination. From coordinate changes 6X (6X,6 Y,6Z) = X (t) - X (to) divided
by the time interval6t we get then the velocities V (Vx' Vy , Vz ):

(1)

To improve the accuracy of velocity estimations, the time interval between the campaigns
should not be too short. Five years seems to be a reasonable time span. It is important to do
the adjustment of the repetitive campaigns in the same reference frame as the first SIRGAS
campaign, Le., ITRF94 with its time propagated station coordinates at the observation epoch.

175
STATION VELOCITIES FROM LITHOSPHERIC PLATE MOTIONS

As long as there has been no repetitive coordinate determination, station velocities cannot be
estimated empirically. In order to propagate the coordinate changes with time, approximation
models have to be used. A very familiar model is that of plate tectonics. The global crustal
deformations are in a first approximation described by the motions of rigid plates (Le.,
spherical segments of the globe). These motions can be expressed for each point of a plate by
the rotation on a sphere with the geocentric rotation vectors 0 (Ox, Oy, Oz). The velocity V
(Vx' Vy, V z) for a station at the Earth's surface with the geocentric coordinates X, Y, Z is
then

Vx = Oy . Z - Oz . Y
Vy = Oz . X - Ox . Z (2)
Vz = Ox . Y - Oy . X .

The IERS adopted the geophysically determined kinematic plate model NNR NUVEL-1A
for all the stations for which the velocity has not yet been determined in the ITRF (McCarthy
1996). The South American mainland is covered by two plates in this model: South America
and Caribbean plates. The Pacific islands (SIRGAS stations Galapagos and Isla de Pascua)
are situated on the Nazca plate and O'Higgins on the Antarctic plate. The rotation parameters
of these plates are given in table 1.

Table 1: NNR NUVEL-1A rotation vectors for SIRGAS (from McCarthy, 1996)

Plate Name Ox[ rad/10 9a] Oy[rad/109a] Oz[rad/10 9a]

South America -1.038 -1.515 -0.870

Caribbean -0.178 -3.385 1.581

Nazca -1.532 -8.577 9.609
Antarctica -0.821 -1.701 3.706

STATION VELOCITIES FROM CONTINUOUS DEFORMATION MODELS

It has to be stated that the hypothesis of rigid plates is only an approximation for the
modelling of recent crustal movements, Le., the station velocities computed by this model
may differ from the real station motions. This holds in particular in the plate boundary zones
where we have considerable regional deformations, i.e., the plate boundaries cannot be
modelled as rigid bodies. Figure 1 shows the South American region with the known plate
boundary zones. We clearly identify all the western part of the South American continent
along the Andes mountain range as an extended deformation zone. In all this region one
cannot use the rotation pole as given for the South American plate in table 1.

176
 A better approximation is a regional rotation vector derived from space geodetic
observations (Drewes 1993).

The difference between computed station velocities in the central Andes using this
rotation vector instead of the rotation vector for South America (table 1) is 2 cm/a, i.e., in
five years we get a coordinate difference of 10 cm. It is therefore necessary to determine the
real deformations from a dense network of stations with observed velocities (see above), and
to derive a continuous deformation model for the plate boundary zones. This can be done by
a geophysical model, e.g., using the finite element method (Drewes 1993) or by a least
squares prediction approach (Drewes 1997).

Fig. 1: Lithospheric plates and plate boundary deformation zones (from Gordon 1995)

Besides the continuous deformations we also have to consider discontinuous motions of

the Earth's crust, e.g., produced by earthquakes. An example is the July 30, 1995 Antofagasta
(Chile) earthquake (two months after the SIRGAS GPS campaign) where horizontal
displacements in the order of 70 to 80 cm have been observed (Ruegg et aI., 1996, Klotz et
al. 1996). The SIRGAS coordinates have to be corrected because of such motions before
using them as a reference frame. Such discontinuous motions have also to be reduced when
computing a crustal deformation model as described above.

177
TIME EVOLUTION OF THE SIRGAS REFERENCE FRAME

In the following we shall summarize the requirements for the use of SIR GAS coordinates as
fiducial stations in precise geodetic positioning in South America using geodetic space
techniques (e.g. GPS). It is assumed that an observation campaign has been performed at a
time tj occupying new stations and SIRGAS stations simultaneously. The result of the data
processing ought to be coordinates of the new stations in the SIR GAS reference system, e.g.,
for national densification of the SIRGAS (= ITRF) reference frame.

The following steps have to be done when using SIRGAS coordinates in the space
geodetic (e.g. GPS) data processing:

1. Propagate the SIRGAS coordinates of stations S used as fiducials (connecting points)

from epoch 1995.4 (defined epoch for SIRGAS) to the new observation epoch tj using the
station velocities Ys derived either from repeated observations or from crustal
deformation models:

K.,(t) =K.,(1995.4) + Ys . (tj - 1995.4).

2. Perform the coordinate adjustment of new stations N using the observations at epoch tj in
connection with the SIRGAS coordinates K.,(tj) .

3. Transform the coordinates of the new stations N from observation epoch tj back to the
SIRGAS reference epoch to = 1995.4 in order to get a homogenous coordinate set in the
SIRGAS system:

~(1995.4) =~(tj) - ~(tj - 1995.4).

As the velocities ~ of new stations are normally not known, they have to be derived
from crustal deformation models (rigid plate motions and/or continuous deformations) as
discussed above.

In particular the last mentioned problem clearly demonstrates the necessity of the
inclusion of a continuous crustal deformation model into the SIR GAS reference system. If
we don't reduce new coordinates (valid for their observation epoch tj) to the reference epoch
to' we will get an inhomogenous set of coordinates referring to different epochs. The
difference is 1 to 2 cm per year time interval from the reference epoch 1995.4. To perform
the reduction we need a continuous deformation model for the entire continent. A rigid plate
kinematic model alone is not adequate for the plate boundary zones. It is strongly
recommended to support the geodynamic projects for monitoring crustal deformations in
South America and to define an approach (finite element or least squares vector prediction)
to compute a continuous velocity field.

178
REFERENCES

Boucher, c., Z. Altamimi, M. Feissel, P. Sillard: Results and analysis of the ITRF94. IERS
Paris, Technical Note No. 20, 1996.
Drewes, H.: A deformation model of the Mediterranean from space geodetic observations and
geophysical predictions. Springer, lAG Symposia, No. 112, 373-378, 1993.
Drewes, H.: Sistema de referencia cinematico incluyendo un modelo del movimiento actual
de las placas tect6nicas. IV Congreso Internacional de Ciencias de la Tierra, p. 230,
Santiago de Chile, 1996.
Drewes, H.: Combination of VLBI, SLR and GPS determined station velocities for actual
plate kinematic and crustal deformation models. lAG Scientific Meeting, Rio de
Janeiro, 1997.
Gordon, R.G.: Present plate motions and plate boundaries. In: T.J. Ahrens (ed.): Global Earth
Physics, A Handbook of Physical Constants. AGU Reference shelf 1,66-87, 1995.
International GPS Service for Geodynamics (IGS): 1996 Annual Report. IGS Central Bureau,
JPL Pasadena, 1997.
Klotz, J., J. Reinking, D. Angermann: Die Vermessung der Deformation der Erdober- fHiche.
Geowissenschaften (14) 389-394, 1996.
McCarthy, D.D. (ed.): IERS Conventions (1996). IERS Paris, Technical Note No. 21, 1996.
Ruegg, J.C., and 14 others: The M=8.1 Antofagasta (N. Chile) earthquake of July 30, 1995:
First results from teleseismic and geodetic data. Geoph. Res. Lett. (23), 917-920, 1996.
Seemiiller, W., H. Drewes: The IGS regional network associate analysis center for South
America at DGFIII. lAG Scientific Assembly, Rio de Janeiro, 1997.

179
 RESULTS OF THE SIRGAS 95 GPS NETWORK PROCESSING
ATDGFI/I.

Klaus Kaniuth, Hermann Drewes, Klaus Stuber, Herbert Tremel

Deutsches Geodatisches Forschungsinstitut, Abt. I (DGFI/I)
Marstallplatz 8, D-80539 Munchen, Germany

Juan Moirano
Facultad de Ciencias Astronomicas y Geof'isicas
Universidad La Plata, La Plata, Argentina

ABSTRACT

For establishing an accurate and geocentric reference system for South America, Sistema de
Referencia Geocentrico para America del Sur (SIRGAS), a continental GPS campaign
including more than 60 stations has been carried out in May/June 1995. DGFIII was acting
as one of the analysis centers processing the entire data set. The paper describes the applied
processing strategy and discusses the achieved results.

INTRODUCTION

The project for the establishment of an accurate geocentric reference system for the South
American continent, Sistema de Referencia Geocentrico para America del Sur (SIRGAS),
has been initiated in 1993. The first goal of SIRGAS was the realization of a reference
network of some 60 stations equally distributed over the continent by performing a GPS
campaign. Besides acting as a data center for the SIRGAS project, the Deutsches
Geodatisches Forschungsinstitut, Dept. I (DGFIII) processed the entire data set acquired
during this campaign and computed a unique network solution. This paper discusses the
strategy applied for processing the large amount of data, the models and software used, and
summarizes the achieved results. The processing has been performed at DGFIII in
collaboration with the University of La Plata, Argentina.
DATASET

The SIRGAS GPS campaign was successfully carried out from May 26 to June 04, 1995.
Following the issued guidelines the majority of stations operated continuously during the ten
days period 23 hours per day. Figure 1 displays the sites occupied during the campaign.

300' 320·
20'f==---~-T~~=T~----~==~~-----F=====-----~======~20'

.,,
C)'.

0'~----------1------;~-------*~~~~~------+-------'~~----~--------~0'
GAlAPAGOS "'IATACU GA

p}~~ lourr:r
ZAMORAf .~
M
.....0. I~~~ ~ "~TOIIZ
I)
FORTALEZA

1\
UMA~
~ eRIBERALTA
CLA RAl I ..
..
BOMJESUSIAPA
~RASILIA II
_20' ~__________1-_________
AREQUIPA. e
-+_ _ __H-\U~IC1-H_URA
e e1"\ CUIABI\
_ _YA- "" SJ.DE CillO ITOS r ------~ -20'
OLl.AGUEe C MIR!.I ~ PRES.~RUDENTE" \'? fOSA
• . ~ESTIGARRIBIA . . . . RIO DE JANEIRO
ANTOFAGASTA r-'
e P'" 1'4 .4.
CAC!l0EIRA

• I SALTA ASUNC~;; MW CURJTIBA

'" I
I
?~
PUERTO IG AZU
ISlA DE PASCUA CIIAMO ATE VIU.A ROBLES

) •
CIUCYT
... MORRO
I YACARE

SANTIAGO I CERRO VIGIA

, IA PIA!A MONTEVIDEO
CARRIELSUR .. LOTE!4

-40·+---------~----------~r_--~----_,~------------_+------~ -40·

.. ASIITECH

• LEICA

• ROGUE

• TRIMBLE

-60· +-------~r_----------~------------4-----------_4------~_60·
..
O'HIGGI 'S

260' 280' 300' 320'

Fig. 1 Location of the sites occupied during the SIR GAS 95 GPS campaign

181
 Figure 1 also indicates that different GPS receiver systems were involved in the SIRGAS
campaign. The distribution of receivers reflects the availability of equipment at the
participating South American organizations and at the contributing European and North
American institutions. In early 1995 the phase center characteristics of the involved GPS
antennas were not yet known precisely enough for combining different antenna types when
applying techniques differencing between stations. Therefore, co-locations of different
receiver/antenna systems were done (figure 1) enabling the combination of receiver-specific
sub-networks, provided local ties are available from independent measurements.

PROCESSING STRATEGY

During the SIR GAS campaign extending over ten days and including some sixty stations a
total of about 2 . 107 phase observations were collected. Processing such a data set with the
Bernese software (Rothacher et al. 1993) which models double differences would require to
solve for some 2.5 . 104 parameters. Therefore, we have applied a strategy of distributed
processing resulting under the assumption that no correlations are neglected in a solution
identical to a one step adjustment. This procedure can be outlined as follows:

- As at the time of the campaign the phase center characteristics of the involved antenna
types were not accurately determined, no mixed baselines including different antennas
were defined in order to avoid any accuracy degradation. Nevertheless, since the
availability of a first set of phase center variation models (Rothacher and Mader 1996)
these have been applied.

- Considering also the number of units of each of the receiver types involved in SIRGAS,
receiver specific networks including the measurements acquired during a one day session
were created.

- Solutions of these daily sub-networks provide a means for assessing the day by day
consistency of the results and allow the identification of outliers and processing errors; in
addition, local parameters such as ambiguities can be pre-eliminated in these receiver-
specific adjustments, and reduced normal equations are then saved for external
accumulation with all other subsets.

- These adjustments for setting up the normal equations were performed in the ITRF refe-
rence system realized by the International GPS Service for Geodynamics (IGS) combined
orbits and compatible Earth orientation parameter series without applying any constraints;
consequently adding fiducial information such as local ties between co-located antennas
did not require any reprocessing.

- The accumulation and solution of all normal equation systems is done using an external
program ACCSOL developed at DGFIII. Before this final stage no datum realizations or
additional fiducial informations are applied.

182
 This approach provides high flexibility, e.g. for analyzing the sensitivity of the network
solution to reference frame variations. The preprocessing has been done with the original
Bernese software, and the adjustment with a modified version of the Bernese program
GPSEST. Among others, the modifications relevant to the SIRGAS project are

- the ability to define constraints for troposphere estimates at co-location sites,

- the availability of advanced troposphere prediction models and mapping functions,
- the capability to store normal equations for the accumulation with ACCSOL.

As already mentioned, the solution is based on the IGS combined orbits and compatible
Earth orientation parameter series. A data sampling rate of 30 seconds was used in all
adjustments although a number of stations generated observations at a 15 seconds rate. The
elevation angle cutoff was set to 15°, because the SIRGAS network includes several sites
providing free satellite visibility only above 15°. Considering the observation session length
of 23 hours/day and the investigations by Mervart (1995) we have not attempted to resolve
the phase ambiguities.

For the a priori calibration of the tropospheric delays the Saastamoinen (1973) zenith
delay model and the mapping function proposed by Davis et al. (1985) have been used;
residual zenith delays were estimated in the adjustment for 4 hours intervals. Decreasing
these intervals to e.g. 2 hours did not result in significantly better daily coordinates
repeatabilities.The Bernese software models these zenith delays as a step function which can
be constrained and smoothed by assigning standard deviations to the a priori calibration. We
have applied ± 1 m for the first absolute parameter and ± 5 cm for the subsequent relative
ones; these numbers let the troposphere estimates practically unconstrained.

COMBINATION OF NORMAL EQUATIONS

As described in the previous chapter receiver-specific daily normal equation systems were
generated which may be considered to represent unconstrained networks in the IGS orbits'
system. The accumulation and solution was done using the program ACCSOL
(ACCumulate and SOLve) which provides among others the following options:

- To introduce coordinates of fiducial points as additional observation equations assigning

individual weights;

- to set up condition equations constraining certain pairs of troposphere estimates to

equality, e.g. at co-location sites;

- to apply local eccentricities information at fiducial stations or for co-located receivers as

stochastic information.

The general structure of combining the SIRGAS normal equations in ACCSOL is dis-
played in figure 2.

183
 Day 146 NEQ
I ,
..-
I AsI1tech
I Flduciallntormation such a.
latest ITRF coordinate.
!
~
Leica (high welght Identity constralnta tor pairs
ob&ervatlon equations) at tropo.phere parametel'l

/
at coloeatlon sites or
I Trtmble
! subnetworX j~ctlon points

I I
( )
Turbo Rogue

·· ACCSOL

Day 155 NEQ

•
_. .__.__. _. J_. __._.
V
i

~
Local eccentricities lit
I A.htech
Mix.cd baselines
colocation .ites
(high weight

I Leica including elevation

dependent phase center
corrections
observation equations)

I Trtmble
I
I Turbo Rogue
I
Fig. 2 Combination of SIRGAS normal equations and additional information in ACCSOL

RESULTS

A comparison of the results for the single day receiver-specific network is a means to detect
not only data and processing errors but also to assess the achieved accuracies. Therefore, we
summarize in table 1 for each receiver type the average root mean square (rms) agreement
between the single day adjustments and the ten days solution. The quoted numbers result
from similarity transformations which account for the small datum shifts between the
unconstrained networks.

Table 1 Consistency of daily solutions: average rms differences in north (N), east (E) and
height (H) between single day adjustments and the 10 days solution for each
receiver-specific sub-network [mm]

Network Stations N E H

Ashtech Z-XII 23 3.7 4.8 7.8

Leica 200 15 6.0 7.8 14.3
Trimble SSE 19 5.3 6.9 11.7
Rogueffurbo Rogue 11 8.6 12.5 14.0

184
 From table 1 it is evident that the Ashtech Z-XII demonstrated the best performance
during the SIRGAS campaign. However, the slightly worse consistency for the other
receivers does not necessarily reflect a worse tracking performance. Instead the distribution
of the stations is obviously of a certain relevance. In case of the RoguefTurbo Rogue
network the processing of the data acquired by the IGS stations close to the equator Bogota
(BOGT), Fortaleza (FORT) and Kourou (KOUR) required particular effort. This holds also
for some of the Leica stations in Colombia, Peru and Venezuela which due to lack of
sufficient memory operated also at a 30 seconds sampling rate. We suppose these problems
to be due to rapid ionospheric delay variations which could not sufficiently be met by those
stations in the equatorial region operating at a 30 seconds observation rate.

Another documentation of the agreement of the daily adjustments with the final 10 days
solution is presented in figure 3. This figure displays for one baseline per receiver type the
departures of the daily baseline length estimates from the mean. The examples shown are
selected to be available on each of the 10 days and to represent a wide range of baseline
lengths. Again the Ashtech baseline of 1630 km length shows the smallest scatter, but also
for the long Leica and Turbo Rogue baselines the daily repeatability is in the order of
3 . 10- 9 •

After combining all daily normal equation systems with ACCSOL the unconstrained 10
days solution was transformed to ITRF 94 using seven fiducial points including Richmond
(RCM5). We do not present a list of coordinates here, because the final set of SIRGAS
coordinates at epoch 1995.4 is a combination of our solution and the results provided by the
National Imagery and Mapping Agency (NIMA). The final set of station coordinates is
included in (SIRGAS 1997).

Clara - Rlberalta 617 km RMS = 3.9 mm Asuncion - Lote 24 1630 km RMS =2.7 mm

~F· ····~±~• +~
Trimble SSE AshIed1 Z - XII
20 • .•.. •••.•.. .•.•.•.•.•..••••• •... • •.• . ..•.•.•...•.•.• . .. .. •.... . •.• .
15 •.•.•.••.....•.••....•............•.. .....•.. .......................
10 •.•..............................•.. . ... .. .. .. .................... ..

_~ :~::~::~::~::m::mi:::~:::~:::t¥j:::·::·:
i··· .... ··· .... ··············· ...... ·· ...... ·.. ·.. ··········· ...... ..
· 10
-15 •1s
.~~r-.--.--.--r-,r-.--.--.--.- ·20 I I I I I I I i I

Puerto Iguazu -Inlrida 3517 km RMS =9.6 mm Forfaleza - Rio Grande 5907 km RMS =9.2 mm

·~I·i• ~.·.·1~·.• .~. .•. .~. .•. .~. .•.~.·.~

. .. . . . .•.·~·. . .. :;l: ~ • ~.·~ •••••7- •1
Leice200 Turbo Rogue

• 7·.~• 7• ~.
20 ....• ...•......... ... ....... . . ......... ..•.. ... ..... ... ........ . ....

-1& .••••••••.•.••••
-20 ' -'-'-'T---'- ·20 - ---,---r-···r···---r··-··.,.-··-,---·,-'-,-;-

Fig. 3 Differences of daily baseline length estimates from the mean for various baseline
lengths and receivers [mm]

185
REFERENCES

Davis, J.e., T.A. Herring, 1.1. Shapiro, A.E.E. Rogers and G. Elgered (1985): Geodesy by
Radio Interferometry: Effects of Atmospheric Modelling Errors on Estimates of Baseline
Length. Radio Science 20, 1593-1607.
Mervart, L. (1995): Ambiguity Resolution Techniques in Geodetic and Geodynamic
Applications of the Global Positioning System. Geodiitisch-geophysikalische Arbeiten in
der Schweiz, Bd. 53.
Rothacher, M., G. Beutler, W. Gurtner, E. Brockmann and L. Mervart (1993): The Bernese
GPS Software Version 3.4. Astronomical Institute, University of Berne.
Rothacher, M. and G. Mader (1996): Combination of Antenna Phase Center Offsets and
Variations; Antenna Calibration Set IGS-Ol. International GPS Service for Geodynamics
(IGS).
Saastamoinen, J. (1973): Contributions to the Theory of Atmospheric Refraction, Part II:
Refraction Corrections in Satellite Geodesy. Bull. Geod. 107,13-34.
SIRGAS (1997): SIRGAS Final Report, Working Groups I and II. Edited by Instituto
Brasileiro de Geografia e Estatfstica (IBOE), Rio de Janeiro, 1997.

186
 THE INTEGRATION OF BRAZILIAN GEODETIC
NETWORK INTO SIRGAS
Costa, Sonia Maria Alves
Pereira, Katia Duarte
e-mail: [email protected]
Brazilian Institute of Geography and Statistics
Av. Brasil 15671- Parada de Lucas
Rio de Janeiro - RJ - CEP : 21241-051, Brazil

Beattie, Don
e-mail: [email protected]
Geomatics Canada
615 Booth Street - Ottawa, Ontario - K1A OE9, Canada

Abstract

Cooperating as part of South American Geocentric Reference System (SIRGAS), whose

purpose is to establish a geocentric reference system for the whole of South America,
Brazil will take the next step to continue this change.
The integration of Brazilian Geodetic Network to SIRGAS will be carried out through a
simultaneous adjustment of the network using the original data and the software Geodetic
Adjustment using Helmert Blocking of Space and Terrestrial data (GHOST). In the
adjustment will be adopted the GRS80 elipsoid and the Horizontal Network will be
constrained to fit the coordinate values of the eleven SIRGAS stations in the national area
The installation and operation of Brazilian Network for Continuous Monitoring of GPS
(RBMC), concluded in 1997, will provide a continuous link to SIRGAS for future users.

Introduction

Being responsible for geodesy and mapping in Brazil, the Brazilian Institute of Geography
and Statistics (IBGE) started the establishment of the Brazilian geodetic network in the
1940's. Since this time, the methods of surveying and processing surffered through great
transformations mainly on the respect of accuracy. Focusing on homogeneous results, the
geodetic net had various adjustment in differents horizontal datums. Due to computational
system limitations at the moment, only partial adjustment of the net were carried out.
Definition and establishing of the South American Datum 1969 (SAD69), targeted the
last partial adjustment made. From that time, a continental adjustment software was
looking for, and efforts were put on a first global and simultaneous adjustment of the
Brazilian geodetic net.
Next step to be performed till the end of 1997, its the integration of the Brazilian net
into SIRGAS System. With this aim the 11 SIRGAS stations in the country were
connected to existing geodetic stations of the net. With 493 stations, the Brazilian GPS
Network is already adjusted and constrained into SIRGAS stations, but more campaigns
have to be included. The combination of GPS and classical geodetic network step, have
5225 stations already simultaneously adjusted through the Helmert Blocking method.
This paper present a historical abstract of all adjustments made and geodetic systems
used in Brazil, also the efforts in preparation and processing of both classical and GPS
nets, in order to integrate them into SIRGAS system.

History of Horizontal Geodetic Control in Brazil - Establishment, Adjustments and

Geodetic Reference Systems

A triangulation project in the south of the country in the 1940' s, started the establishing of
the high accuracy geodetic network. Since that time, the triangulation network spread over
meridians and paralels, allowing to use the polygonal method for further densification.
In the 1970' s the availability and accuracy of the TRANSIT satellite geodetic
positioning, transported the horizontal net to areas where conventional methods were
difficult to use, as Amazonic region.
The densification method by GPS process was adopted when four TRIMBLE GPS,
4000 sst model came in 1991, marking a new tridimensional geodetic network era.
Establishing the Active Control GPS Network, named Brazilian Network for Continuous
Monitoring of GPS - RBMC, with 9 stations (same as SIRGAS Project's ones occupied)
was a very important step for the geodesy in Brazil. At the moment, the geodetic net have
6318 stations being of them, 3498 triangulation stations, 26 trilateration stations, 1158
traverses stations, 1143 Doppler stations and 493 GPS stations.
It is importat to pointed out the fact that coordinates were adjusted using various
methods and geodetics systems during this time. Brazilian network had three main
adjustments.
First adjustment was made with the condition equations method and adopting the
C6rrego Alegre Geodetic System as reference.
Second adjustment had two computer phases, first one carried out by DMAIIAGS (Inter
American Geodetic Survey), adopting SAD69 as horizontal datum and using the computer
system HAVOC (Horizontal Adjustment by Variation of Coordinates). The next phase
performed at IBGE, adjusted the remained network using USHER program (Users System
for Horizontal Evaluation and Reduction). During the 20 years application of the
adjustment in parts, unavoidable net distortions were introduced. It was felt the needed for
a computational system, able to perform a simultaneous adjustment of great amount of
geodetic data and at the same time using a tridimensional model in order to include
classical and GPS observations. At the end of the 1980's GHOST (Geodetic adjustment
using Helmert blocking Of Space and Terrestrial data) system for DOS and UNIX
platforms was installed at IBGE. The whole classical observations were passed through a
validation and magnetic storage process.
It was then in the middle of 1996, when the third geodetic adjustment was finished, but
this time, was a global and simultaneous one. Adjustment was made on the SAD69
geodetic system because it is the official one in the country. Adoption of geocentric
reference system is another step to be carried out. With this aim, Brazil has an active

188
participation in the SIRGAS Project, through technical cooperation, being one of the
project data center and applying the projects' recomendations made, one of them refering
the integration of geodetic network into SIRGAS geodetic system.

Terrestrial Data

The classical network observations used in the global adjustment of 1996 were: 16913
horizontal directions, 1534· geodetic distances, 389 astronomic azimuts and 379
astronomic coordinates. Same observations will be used in the adjustment into SIRGAS
system. The validation and storage observations steps were development in three different
projects.
A first project was the horizontal directions treatment, being the longest due to the
amount of data to be analysed.
The second one was the reprocessing of astronomic observation (latitude, longitude and
azimut) in FK5 system was made in the project "Pro-Astro".
The third project aimed the reprocessing of geodetic bases.
All the three projects were carried out with computational systems developed at IBGE.
The data of 144 traverses were through the validation and reprocessing phase as well.
Once the project was finished, it was started an analysis of the combination of data
adjusted. The classical network was divided in 13 areas, separately adjusted and during a
later step the classical network was adjusted using the Helmert Blocking method.

Spatial Data

During the period 1973-1991, IBGE made use of the satellite's TRANSIT geodetic
positioning system, mainly in difficult areas for classical field works, e.g. Amazon
region. The more used observational method for stations determinations, was point
positioning. For a more accurate treatment of DOPPLER observations, a precise
ephemeris postprocessing with program GEODOP V was made. During the adjustment
only 179 DOPPLER stations were used, those which are coincident with classical ones.
GPS technology was first used at IBGE, with the purchase of four geodetic receivers in
1991. Till 1994 a total of 187 GPS stations, were processed with TRIMVEC Plus, D
version and using broadcast ephemeris. Later on data was processed with Bemese version
4.0 and using now, precise ephemeris. In both cases, the station coordinates and its own
variance-covariance matrix were generated in the final processing solution and later on
used in the GPS network adjustment. The mathematical correlations were handled
correctly for the data processed with Bemese software. In order to prepare the adjustment,
the data of each individual campaign were put in GHOST format and a minimun constrain
adjustment was performed. The variance factor obtained on adjustment will be used to
scale the matrixes. For each campaign this procedure is repeated and in a second step it is
included in the final adjustment file.
For SIRGAS integration we have participating a total of 1565 GPS baselines, as can be
seen in figure 1.

189
 5:~ __ ~ ____ ~ ____ ~ ____- L_ _ _ _- L_ _ _ _- L_ _ _ _- L_ _ _ _- L_ _ _ _ ~

-15

-2

-25

-3 Il SIRGAS stations
C GPS stations

-70 -65 -60 -55 -50 -45 -40 -35 -30

FIGURE 1 - Brazilian GPS Network -1997

Adjustment Methodology - SIR GAS Integration

For the adjustment of large geodetic nets the numerical procedure of partitioning the total
net in several blocks in a very useful tool. The subdivision into blocks can be brought
about according to diferent points of view: with respect to regional aspects and special
kind of observations. It is not only useful for the partitioning of a geodetic net, moreover
plays an important role for the rigorous fusion of triangulation nets with other geodetic
systems. This is why GHOST software was chosen to carry the brazilian network
adjustment out.
GHOST software is composed by a set of programs using a tridimensional modeled
system for least squares adjustment. Its main features is to employ the Helmert Blocking
method for partitioning a continental size geodetic network as the brazilian one is.
The defined strategy for block partitioning is based on a minimun number of junction
station between blocks, minimizing in that way the computational effort in the adjustment.
As seen in figure 2, the geodetic network was splited into 8 blocks, each of them with
three different levels. The adjustment task was divided into three tasks. After finished all
data set validation of classical network, it was performed a Helmert blocking adjustment.
The second task was perform the adjustment with GPS data The third and last task was

190
performed an adjustment combining GPS and classical observations. The eleven SIRGAS
station complement the integration with the introduction of the sigmas' weighted
coordinates.

> ..

BNll
-I

BN12

-1

BN2J

-2
AS22

-2 AS12

., Triangulation
CI Traverse
-3 ( Doppler
(. GPS ASll

-70 -65 -60 -55 -50 -45 -40 -35 -30

FIGURE 2 - Brazilian Geodetic Network 1997 and Helmert Blocking division

Relation of SAD·69 to SIRGAS

Established on the end of 1960's, the SAD-69 was officially used in Brazil in the 1970's.
With the satellite positioning techniques adopted, it was inferred the necessity for solving
transformation parameters between various geocentric referece systems, as NWL-lOD,
NSWC-9Z2 and WGS-84, aiming the agreement with SAD-69.
From this eleven SIRGAS station, only five already had SAD-69 coordinates known,
generated from the simultaneous and global adjustment finished in 1996. In a first test,
there were determined a set of seven Helmert transformation parameters. During the
results analysis was find the rotations as one hundred part of a second.
In the second test with translations only, high Fortaleza station residuals were found as
compared with the remained stations. Taking this into account and aiming the most
accurate results, Fortaleza station was removed for the computation step, leaving the
following ones: Curitiba, Cachoeira Paulista, Brasilia, Presidente Prudente, and having
this final results.

191
Number of Parameters: 3
Number of Coordinates: 12
rms of transformation: 0.0716 M
Parameters:
Translation in x: 67.327 +- 0.036 M
Translation in y: -3.899 +- 0.036 M
Translation in z: 38.292 +- 0.036 M

NAME RESIDUALS IN METERS

X Y Z
Brasilia 0.0064 -0.0468 -0.0977
Pres. Prudente 0.0148 -0.0123 0.0842
Cachoeira Paulista -0.0299 0.0017 0.1125
Curitiba 0.0145 0.0558 -0.0998
RMS / COMPONENT 0.0213 0.0427 0.1144

Conclusion

Considering the worldwide tendency in adopting geocentric reference systems, bustled by

the bigger necessity of geodetic network integration and the technique accessibility
provided by GPS, its was noted the importance of the Brazilian Geodetic Network
integration into SIRGAS system, judged at the moment one of the most accurate
geocentric reference system worldwide.
According to the development project timetable, it is scheduled for next december the
activities conclusion, with some still processing GPS campaign final integration. The new
SIRGAS system referred coordinates will be available to the comunity at the begining of
1998.
Some studies for adoption of SIRGAS system for mapping are already being made. The
first step was the determination of transformation parameters in order to become
automatic the SIRGAS system refered digital maps elaboration. It is known that is this
process is slow because great part of the maps sets is not in a digital way, but efforts have
been made to conclude this step as soon as possible.

Bibliography

- Beattie, D.S. - Progmm GHOST User Documentation, Geodetic Survey of Canada,

Ottawa, 1987.
- Cadess, H. et aI -Inter Americam Geodetic Integmtion - PAlGR, 1991.
- Costa, S.M. & Fortes L.P.S. - Resultados Preliminares do Ajustamento da Rede
Planimetrica do Sistema Geodisico Brasileiro, Rio de Janeiro, 1993.
- NGS, North American Datum of 1983, NOAA Professional Paper NOS2, National
Geodetic Information Center, NOAA, 1983.
- Rothacher M.& Mervart L. - Bernese GPS Software Version 4.0 - Astronomical
Institute University of Berne - Berne - september 1996

192
 ADJUSTMENT OF THE NEW VENEZUELAN NATIONAL GPS
NETWORK WITHIN THE SIRGAS REFERENCE FRAME

Hermann Drewes and Herbert Tremel

Deutsches Geodatisches Forschungsinstitut, Abt. I
Marstallplatz 8, D·80539 Munchen/Germany

Jose Napoleon Hermindez

Servicio Autonomo de Geografia y Cartografia Nacional
Edif. Camejo, CSB, Caracas lOlO/Venezuela

ABSTRACT

The new Venezuelan national geodetic network (Red Geodesica Venezolana, REGVEN)
was observed nearly simultaneously with the GPS campaign for the establishment of a
geocentric reference system for South America, Sistema de Referencia Geocentrico para
America del Sur (SIRGAS). This paper describes the REGVEN data, its processing and the
connection to the SIR GAS reference frame. The accuracy of the geocentric station coordina-
tes is in the order of ±2 cm.

INTRODUCTION

The GPS campaign for establishing an accurate geocentric geodetic reference system for
South America, SIRGAS, has been carried out in May/June 1995. The Deutsches
Geodatisches Forschungsinstitut, Abt.I, Munchen (DGFI/I) was acting as a data center and
as an analysis center processing the complete data set of this project. Besides the national
institutions of South America, several institutions in Europe and North America participated
in the SIR GAS observation campaign.
All South American countries expressed their intention to connect their national reference
systems to SIR GAS, or to establish new national reference systems within the SIR GAS
frame by GPS, respectively. The Venezuelan Servicio Autonomo de Geograffa y Cartograffa
Nacional (SAGECAN) installed a completely new geodetic network (Red Geodesica
Venezolana, REGVEN). It consists of 67 stations covering the entire territory of Venezuela,
except the sparsely settled southern area (Territorio Federal Amazonas) and was observed
quasi simultaneously with the SIRGAS network. REGVEN is connected to SIRGAS by the
five Venezuelan SIRGAS stations Maracaibo, Junquito, Agua Linda, La Canoa and Kama.
As SIRGAS refers to the ITRF94 at epoch 1995.4, REGVEN is thus defined in the same
reference system.

OBSERVATIONS

The REGVEN GPS campaign was carried out during the period May 20 - June 16, 1995.
The observation period covers thus 28 days. Four of the five SIRGAS reference stations in
Venezuela were attended and supported by DGFI/I, all the other stations, including the
SIRGAS reference station Junquito were observed by SAGECAN. The GPS systems
Trimble SSE / SST and Leica 200 (only during the SIRGAS campaign at station Kama)
have been used for the measurements.
The daily occupation of stations was as follows:
- May 20 - 25, 1995 10 stations,
- May 26 - June 4, 1995 13 stations and
- June 5 - 16, 1995 5 to 7 stations.
Apart from a few exceptions all stations observed at least two or three different days (ses-
sions), operating six hours per day. All the SIRGAS reference stations recorded 23 or 24
hours daily during the SIRGAS observation campaign. Outside the SIRGAS campaign at
least three SIRGAS reference stations were occupied simultaneously. Generally the sam-
pling rate was 15 seconds. The observations included data down to an elevation cut off angle
of 10°. The location of all sites occupied during the REGVEN campaign is shown in figure
1. The names of stations correspond to the conventions of SAGECAN.

DATA PROCESSING

For the processing of the REGVEN data mainly two program systems were used:
- the Bemese GPS software system version 3.4 (BV3.4) (Rothacher et aI., 1993) with some
modifications implemented at DGFI/I;
- the DGFI/I program ACCSOL (ACCumulate and SOLve) for the accumulation and
solution of normal equations.

The processing started with the RINEX (Receiver INdependent EXchange format) files
containing the observation data, which had to be verified and edited in a first step. This
included in particular the checking of the compatibility of file headers with the RINEX
format and the reduction of the antenna phase centers to the defined reference points.

A generalized scheme of the complete GPS data processing with BV3.4 and ACCSOL is
given in the description of the DGFI processing of the SIR GAS '95 GPS network in the
official SIR GAS Report (SIRGAS, 1997).

194
 290' 295' 300'

I ,...J~ ~ ~ tJ
. - J ' CAST(04) AMUA(O~ lJ
• -- TOOO<''',
~ .P'
c:::)
/ ' \ ./ I -- "
jUNQ{J<,01>GU{3" 0 ANTO{", 'c 11'0'
MARA • • ,
~'CA{'" °C O G O ( ' ~
~
\
.1 (03) QU,,,,,, RORA{", °N'R",'" CAMA{'" ELHA(SZ)CRU<{55, • CCOL(63)

0 0 0 60)
10' .R ECR\;{,m 0 'ARA{>O, o H'GT,(51) 0 • : )TUCU(6S)
r·: L~GU{'" I •
TA"A{3~ MA~
CEIB(l2)
~ CARA{'~ 0PIRI(22) . . "'"'C(2) BVIS(4S) 0 POCA{,,' NEGR{
I CAPI(08)
•OBSE(19) I BAUL(2S) . HIN(SO) . f"'ANO(S9)
o oC .
CD
~RIG(
•
O I)
I\IIRA(09) ~HIP(ZI) ~UAN(26) • ARIS(3J)
01
•CHIG(lI)
• . PERI(38) • SRIT(37) • GUAR(66)
•SANT{ZO) NIO(3IA) • MAPI(S7)
•QUIN(27) •SAI\IA(30) • • MATA(S8)
•SANA(IOA) •SURI(IO) • PAYA(39) • SIPA(46) • MAZA(68)
• CHAV(44)
•GUAS(28) GUAC(40)
•ELOR(Z9) LAPI(43) • FORT(67)
•
• PIJI(48) • DORA(70)
• CANA(69)
•AGUA(49)

KAMA(7I)
•
5' I U5"
290" 295' 300'

Figure 1: Location of sites of the new Venezuelan reference network (REGVEN)

Similar to the SIRGAS processing daily solutions for the 28 days of REGVEN were per-
formed. These daily solutions provide a means for assessing the day by day consistency,
allow the identification of outliers, and detect processing errors; in addition local parameters
(e.g. ambiguities) can be pre-eliminated. The reduced and unconstrained normal equations
are then saved for accumulation. The accumulation and solution of all the 28 unconstrained
normal equation systems is done by the external program ACCSOL. The main characteri-
stics of the modelling (observations, orbits) and the parameter estimation are described in
the sequel:

- basic observables are the carrier phases Ll and L2;

- code data are only used for clock modelling;
- the sampling rate is 15 seconds;
- the elevation cut off angle is tOO in preprocessing and 15° in adjustment;
- modelled observables are double differences, ionosphere free linear combination L3;
- the ionosphere is not modelled because first order effects are eliminated by the linear
combinations L3;
- tropospheric delay modelling: Saastamoinen's zenith delay prediction and Niell's map-
ping function (Niell, 1996); estimation of zenith delay corrections every two hours;
- the offsets of mean antenna phase centers with respect to the defined antenna reference
points and elevation dependent range corrections are according to the model IGS-Ol,
issued on June 30, 1996;
- ambiguities are resolved as far as possible, but always pre-eliminated in the daily solu-
tions of the reduced normal equations;
- the definitive combined IGS orbits and clock parameter series are used, according to the
files IGS08016.SP3 to IGS08056.SP3;
- the earth orientation is adopted from IGS solutions compatible to the satellite orbits,
given in files IGS08017.ERP to IGS08057.ERP;
- GEM T3 gravity field truncated to 12 x 12 is used.

RESULTS

The final result of the SIRGAS reference network adopted by the South American countries'
representatives at the SIRGAS workshop held at Margarita Island, Venezuela, April 8 - 11,
1997, is the basis of the final adjustment of REGVEN. The unconstrained station coordinate
solution as described in the previous section was transformed to the five Venezuelan stations
of the SIRGAS reference frame (epoche 1995.4) by a seven parameter similarity trans-
formation. The r.m.s. fit to the SIRGAS coordinates is ±14 mm in X, ±11 mm in Y, and ±15
mm in Z components. For the final REGVEN coordinate set, the SIR GAS coordinates of
these stations are kept.

196
The overall accuracy of the REGVEN adjustment may be estimated from the daily solutions
and the 28 days final solution. For their comparision the numerical rank: defects are removed
by performing similarity transformations. The r.m.s. differences between the daily adjust-
ments and the final solution range from ±2.9 mm to ±23.6 mm in the horizontal components
and ±4.2 mm to ±28.8 mm in the vertical component. Estimating the external accuracy one
has to consider, that some stations have been occupied only one day. As the stations are
situated close to the equator, the observations are strongly influenced by high fluctuations of
the ionosphere.

A further possibility to control parts of the REGVEN coordinates indepentently, are the
results of the CASA (Central and South America) project (Kaniuth et aI, 1997). Eight
stations of the CASA network are identical with REGVEN stations. In table 1 the residuals
after a similarity transformation between the two networks at epoch 1995.4 are given. They
are nearly all less then two centimeters.

As a summary one may estimate the accuracy of the station coordinates of the REGVEN
network to be in the level of ±2 cm or better for all components. This is a very good basis to
serve as a national reference frame for the future. The maintenance of the network will
include repeated observation campaigns allowing to derive station coordinate changes which
have to be expected due to the tectonically active region along the Caribbean - South
American plate boundary crossing the northern part of Venezuela (see also Kaniuth et al.
1997).

Table 1: Residuals (mm) in North (N), East (E) and Height (H) between identical
stations of REGVEN and CASA after similarity transformation and num-
ber of sessions in the two independent campaigns

.Station No. of Sessions Residuals [mm]

...
:.' REGVEN CASA N E H

AMUA(5) 2 4 16 -11 -1
NIRG(23) 4 4 -9 22 -11
ENBC(32) 3 4 4 16 -13
KAMA(71) 21 13 3 -8 0
MARA(3) 16 13 -2 -7 -4
JUNQ(34) 28 13 -7 -3 17
CANO(59) 27 13 -2 -1 -1
OBSE(19) 2 4 -3 -9 13

197
REFERENCES

Kaniuth K., H. Drewes, K. Stuber, H. Tremel, N. Hernandez, M. Hoyer, E.

Wildermann,H.G. Kahle, Ch. Straub (1997): Crustal Deformation along the Caribbean-
South American Plate Boundary Derived from the CASA GPS Project. lAG Scientific
Assembly, Rio de Janeiro, Sept. 3-9, 1997.
Niell A.E. (1996): Global mapping functions for the atmospheric delay at radio wavelength.
J. Geophys. Res., 101, B2, 3227-3246.
Rothacher M., G. Beutler, W. Gurtner, E. Brockmann, L. Mervart (1993): Bernese GPS
Software Version 3.4 - Documentation. Astronomical Institute, University Bern, 1993.
SIRGAS (1997): SIRGAS Final Report, working groups I and II. Edited by Instituto Brasi-
leiro de Geografia e Estatfstica (IBGE), Rio de Janeiro, 1997.

198
 REALIZATION OF A GEOCENTRIC REFERENCE SYSTEM IN
ARGENTINA IN CONNECTION WITH SIRGAS

J. Moirano 1 and C. Brunini

Fac. de Cs. Astron6micas y Geofisicas (FCAG), UNLP.
Paseo del bosque sin, 1900 La Plata, Argentina

H. Drewes and K. Kaniuth

Deutsches Geodatisches Forschungsinstitut Abt. 1 (DGFII1)
Marstallplatz 8, D-80539 Munich, Germany

ABSTRACT

Between 1993 and 1994 the POSiciones Geodesicas ARgentinas (POSGAR) GPS network
was observed, and early in 1995 the POSGAR'94 reference frame was established. It
consists of 127 evenly distributed stations providing a relative precision better than 1 ppm
in the WGS84 reference system.
Since 1995, the DGFIIl and the FCAG are working together to obtain a Geocentric
Reference Frame for Argentina, consistent with the one realized by the SIstema de
Referencia Geocentrico para America del Sur (SIRGAS) network.
This work, which involved a cooperation in the SIRGAS reference network computation
and then the re-processing of Argentina's reference network is scheduled to be finished in
early 1998.

INTRODUCTION

In 1993 there was a growing need in Argentina of a national reference frame in accordance
with the positioning accuracy of the new space techniques. In this context, a GPS geodetic
network with a relative precision of 1 ppm was proposed.
Between 1993 and 1994, POSGAR was observed mainly by the IGM (Instituto
Geognifico Militar) with the collaboration of the UNAVCO consortium through the
Central Andes geodynamic Project (CAP). The network was computed by the FCAG

I Fellow of the Consejo Nacional de Investigaciones Cientificas y Tecnicas

(Facultad de Ciencias Astron6micas y Geofisicas) on contract with the 'Programa de
desarrollo econ6mico y saneamiento financiero de las provincias argentinas', a development
project managed by the Ministry of the Interior and partially funded by the World Bank.
The observations were processed with a commercial software and the reference frame
was realized through 19 CAP stations and one DORIS point all of them transformed to
WGS84.
The POSGAR94 reference frame was delivered in early 1995 and published by IGM some
months later [IGM, 1995]. It consists of 127 evenly distributed points forming a grid of
approximately 200 kIn cell size covering the whole country.
Meanwhile, in 1993, the STRGAS project had begun, its main goals being the
establishment of a high precision geocentric reference frame and the integration of all the
South American countries to this frame. Within this project, DGFIII and FCAG have been
working in cooperation, first on the measurement and computation of the reference
network, and now on the integration of Argentina to the SIRGAS frame. This integration is
being done following the SIRGAS WGII recommendations delivered to the countries by
the end of 1996 [SIRGAS, 1996][SIRGAS WGII, 1997].
Finally, in 1997, FCAG and IGM signed an agreement to cooperate in the maintenance of
the national reference system. This constitutes an important step towards a future adoption
of SIRGAS as the official zero level reference frame for Argentina.
The POSGAR94 frame is briefly described here from the observations to the final
adjustment. Indications of the relative and geocentric quality of the network are included.
The status and strategy of the integration of POSGAR into SIRGAS are also described.
Finally, some results showing the difference between POSGAR94 and its recomputation
are also included.

THE OBSERVATIONS

The measurements of POSGAR stem from three different GPS campaigns: POSGAR'93
and CAP, carried out simultaneously, and POSGAR'94, observed in the following year.
The POSGAR'93 observations were made between 10th February to 12th April 1993,
whereas the CAP observations were done between lIth February and lIth March 1993.
The POSGAR'93 sessions last for 6 hours and the CAP ones for 22 hours.
The receivers used were TRIMBLE 4000 SST and TOPCON both with double frequency
and squaring L2 recovery. The number of receivers varies from 5 for the pure POSGAR'93
sessions up to 9 for the ones when POSGAR and CAP were measured simultaneously.
The POSGAR'94 campaign was carried out between 8th March and 5th May 1994. The
sessions lasted for 6 hours and 3 TOPCON receivers were used.

COMPUTATIONS AND INTRODUCTION OF THE REFERENCE SYSTEM

All the possible vectors between the simultaneously measuring stations on each session
were computed by means of the program GPPS V5.1 from Ashtech.

200
 The adjustment was carried out by means of the program oalpnet, developed in FCAG
[Usandivaras and Brunini, 1992]. It treats the vector components as observations and
allows for the introduction of a priori coordinates and weights for some stations.
The reference system was introduced by means of the coordinates of 20 stations:
19 CAP stations located on the north-west and mid-west of the country.
The EARG (Estaci6n Astron6mica de Rio Grande) station, on Tierra del Fuego, whose
coordinates came from the DORIS beacon which operates on the site.
The CAP coordinates were obtained through a personal communication from Robert
Smalley, from the Centre for Earthquake Research and Information at Memphis State
University. In both cases the original coordinates were transformed to WGS84 and then
introduced into the final adjustment with an a priori error of 3 cm [Usandivaras and
Brunini; 1995].

THE POSGAR'94 FRAME

Figure 1 shows the number of ties for each point as seen by the adjustment software, the
source being the 660 vectors included in the final adjustment in 1994. However, it should
be kept in mind that this number of ties does not reflect reality in the sense that the inter-
baseline correlations were not taken into account.
A better idea of the redundancy within the network would be given by the number of
times each station was occupied. This information is shown on figure 2.

140 140r--- - - - - - -- - - - ,
127 127

:I
.~
' 20 ,,, l!
.k
120t---------
106
!o 100 r-- ! 'oot-- - -- - - - - - -
o ~
j 60
r-- jE 6 O t - - - - - - - - -
E

1 .0+ - - - - - - - - _
~
o 11
l 00
r-- 60
:;
E 40 r-- E

..g
~

~
"" r-- .. . ot----~=_
o ;lo
2i
ro > 20 > 10 > b > 50 ""
~

4. ;:. 3 .. ~ •1
~

No. ollie.

Figure 1: Number of ties per point Figure 2: Number of sessions per point

In 80% of all cases the errors of the final coordinates are below 30 cm at a confidence
level of 95%. This is shown in figure 3.
From the errors in the final coordinates, the relative errors were computed for the 660
baselines that were included in the adjustment. The result is shown in figure 4, where it can
be seen that for a confidence level of 64% more than 75% of the baselines have errors of
less than 1 ppm and more than 90% lie below 2 ppm.

201
 100 ,-------------;=-------..,...----,
85+-- - --1
100
90 o 90 +---rn.-~~~

..
~ eo
C
70
60
~ 85+---
~ ~80 +--­
·0
Cl. 50 ~ ~ 75
'0 40 t~po
.,;
Z
30 a 65
20 ~ SO
10
0 55
C>
v
C>
N
v ::7
C>
'J" :v '-'
v
50
Oil 112 '113 3/4 415 5/10
S .. ml-mayor axis of .. lIipse (em) Standard de>'l8aon [ppm)

Figure 3: Errors in absolute coordinates Figure 4: Std. Errors in computed baselines.

(95% significance ellipses).

As an external check, POSGAR'94 and SAGA (South American Geodynamic Activities)

networks where compared in four points in northern Patagonia. The comparison is
summarized in table 1.
Another comparison, for one point in Buenos Aires province was made between
POSGAR'94 and SAGA [Del Cogliano et aI., 1996]. This is shown in table 2.

Table 1 Table 2
SA GA-POSGAR94 Mean (m) RMS(m) SAGA -POSGAR94 Di erence (m)
Latitude -0320 0.024 Latitude -0.065 m
Longitude -0.125 0.017 Longitude -0.264 m
Hei ht 0.422 0.089 Hei ht 1.016 m

POSGAR '94 and the SIRGAS network. In April 1997, during the SIRGAS meeting in
Margarita, the final solution of the SIRGAS network was presented. This final solution has
a geocentric precision of better than 1 cm and realizes the ITRF94 reference system for the
epoch 1995.4 [SIRGAS, 1997].
There are 6 identical points in the SIRGAS and POSGAR'94 frames. They were
compared directly (no transformations involved) to have an overall external assessment of
geocentric accuracy and relative precision of the POSGAR'94 frame. The results are
summarized in tables 3 and 4, which show the relative precision of the network to be better
than 1 ppm and the geocentric accuracy to be better than 1 m.

THE NEW COMPUTATION AND INTEGRATION OF POSGAR INTO SIRGAS

During the first half of 1996 some processing tests showed that a re-computation of the
POSGAR network with the Bernese software would significantly improve the quality of the
national reference frame. This improvement would allow a better definition of the
geocentric reference frame when connecting POSGAR to SIRGAS.

202
 Since August 1996, the FCAG is working with the DGFVI in the re-computation of the
POSGAR network as a first step before the connection to SIRGAS is made. At this time,
adding four CAP stations from the Chile campaign plus the Santiago IGS (also SIRGAS)
station were judged useful.

Table 3 Geocentric coordinates Table 4 Baselines

POINT
LAT LON H BASELINE LENGTH DIF.
[cm} [cm} [cm} LOTEIOB to [KM [PPM}
LOTE lOB 48 53 -14 900 0.04
LOTE24
LOTE24 49 14 -17 -0.01
ELMAITEN 498
ELMAITEN 40 26 34 0.04
MORRO 1438
MORRO 50 24 41 2040 -0.01
V.ROBLES
V. ROBLES 52 23 4
862 0.31
RIO GRANDE
RIO GRANDE 83 33 -128

For the data processing, the Bernese software V3.5 is used [Rothacher and Mervart;
1996]. During the quite lengthy phase pre-processing, a careful data editing was necessary
to eliminate noisy data that would otherwise be misinterpreted as series of cycle slips. This
implies a first processing of each vector alone to make finding data problems easier. CODE
orbits and rotation parameters were used for all the POSGAR campaigns since there are no
IGS precise orbits available for the 1993 campaigns. The elevation mask in processing was
set to 15 degrees. The tropospheric delay is computed with the Saastamoinen prediction
model using standard atmospheric data plus a zenith delay correction obtained from the
GPS data with a 1/cos(z) mapping function. These parameters are set for each station to be
valid for at most 5 hours.
After cleaning all vectors, each session was processed, solving for coordinates,
tropospheric parameters, and ambiguities. In these runs, ambiguities were eliminated after
inversion of the normal equations which were also saved. No a priori weights for the
coordinates were applied, leaving then the definition of the reference system to the orbits.
Finally, the normal equations from all sessions were added and the complete network was
adjusted. Firstly, only a free adjustment was computed in order to perform quality testing.
In the near future, the reference frame will be introduced through the SIRGAS stations.
The epoch for the fiducial coordinates will be set to the middle of the POSGAR
campaigns, which is 1993.8. The best choice for the velocity model to be used for the
transformation will be analyzed. The final POSGAR coordinates will be transformed to
epoch 1995.4 so as them to be in the SIRGAS system.
The link between POSGAR and SlRGAS. The SIRGAS points in Argentina not included
in POSGAR have been integrated in three cases: LPGS (La Plata, IGS), CRIC (Cricyt) and
IGUA (lguazu). They were connected during 1995 and 1996 with at least one POSGAR
station by at least two sessions of more than 6 hours. The computations will be finished
soon this year. The coordinates will be obtained in ITRF94, epoch of the observations, and
then transformed as stated above to be used as fiducials in the final adjustment.

203
 POSGAR '94 and POSGAR '98. Here we show the comparisons between the SIRGAS
results and the POSGAR coordinates for both POSGAR'94 and the latest new
computation. Table 5 shows a summary of the residuals from similarity transformations
computed between the frames for 5 out of the 6 points common to the three frames.

Table 5 POSGAR'94-SIRGAS POSGAR'98-SIRGAS

RMS 7 parameter 3-D shift. 7 parameter 3-D shift. Ratio 7 P. Ratio 3 P.
North(m) 0.030 0.026 0.008 0.010 3.8 2.6
East (m) 0.042 0.086 0.017 0.014 2.5 6.1
Up (m) 0.176 0.289 0.007 0.027 25.1 10.70
General 0.130 0.175 0.014 0.018 9.3 9.72

It is clear from the ratios between the transformations· that the new computation is about
one order of magnitude more consistent with SIRGAS than POSGAR'94. It is also clear
that most of the improvement is achieved in the heights, the greatest weakness of
POSGAR'94 as can be readily seen from tables 1,2 and 3.

CONCLUSION

POSGAR'94 has a relative precision of 1 ppm and an accuracy of roughly 1 m. The new
computation being carried out by the FCAG in cooperation with the DGFIII would
improve the quality of the frame by a factor of ten and at the same time integrate Argentina
to the continent through the SIRGAS frame. This work will be finished early in 1998.

REFERENCES

SIRGAS, 1997. Final Report of the SIRGAS project, Working Groups I and II. Presented
at the lAG Scientific Assembly, Rio de Janeiro, Sept 3-9, 1997.
SIRGAS WG II, 1997. Recommendations for the integration of national geodetic
networks into SIRGAS; Personal communication.
SIRGAS, 1996. Boletin informativo n° 4.
Rothacher M. and Mervart L., 1996. Bernese GPS Software Version 4.0; Astronomical
Institute, University of Bern.
Del Cogliano D., Perdomo R, Oi Croche N. and Napal E., 1996. Control de referencias
con GPS en Bahia Blanca, FCAG; La Plata.
Usandivaras J., Brunini c., Canosa D., Mondinalli c., Gende M., Moirano J. and Alvarez J.
1995. Calculo de la red geodesica nacional argentina, estrategia y resultados; Proceedings
of the XVII Congreso Brasileno de Cartografia, Salvador, Bahia.
IGM, Instituto Geografico Militar, 1995. La Red POSGAR; IGM, Buenos Aires.
Usandivaras J. and Brunini C. 1992. Programa general de compensacion por minimos
cuadrados; 17th meeting of the AAGG; Buenos Aires.

204
 SOME CONSIDERATIONS RELATED TO THE NEW
REALIZA TION OF SAD-69 IN BRAZIL

Leonardo C. Oliveira I ; Joao F. G. Monico 2 ; Marcelo C. Santos3 ; Denizar Blitzkow4

lIME - Departamento de Engenharia Cartografica - Pra~a General Tiburcio, 80

Rio dc .J'lIIciro - IU - Unl:l.il - 22290-270 - ( leOlulfl[[email protected])

2UNESP - Departamento de Cartografia - Rua Roberto Simonsen, 305

Presidente Prudente - SP - Brazil - ] 9060-900 - ([email protected] )

3UFPR - Departamento de Geoeiencias - Caixa Postal 19011

Curitiba - PR - Brazil- 81531-970 - ( [email protected]/pr.br)

4 USP - Departamento de Transportes - Av. Prof. Almeida Peado, sin

Sao Paulo - SI) - Brazil- 05508-900 - ([email protected])

ABSTRACT

IBGE - the institution responsible for the Brazilian Geodetic System (BGS) worked for
more than ten years in order to readjust the obsetvables belonging to this system. This
work was finished at the end of 1996. The main reason for this paper is that there is the
need for complementary analysis of the results. A few aspects related to the new
realization of the SAD-69 in Brazil will be shown, taking into account its accuracy and
aspects related to the conversion between the new and the old realizations. For the last
case, a studied was carried out by using either the seven or three parameters
transformation model. A new proposal for the transfonnation of coordinates is presented.
Further investigations should still be carried out. The analysis of the accuracy was based
on 34 GPS stations, which are coincident with the conventional network, being their
coordinates considered as true values. A total of 34 adjustments were carried out, in such
way that the GPS baseline vectors of each one of the stations were removed in the process.
The estimated coordinates were compared with the true ones. -the results shown an
improvement of the new realization. For a more conclusive analysis, further GPS stations
would be required.
Introduction

The establishment of the South American Datum 69 (SAD-69) had as its major objective
to provide the whole continent with a uniform geodetic system, capable of controlling
both surveying and cartographic activities (Fisher, 1973). In Brazil SAD-69 was adopted
late 1970's. Since then all geodetic positioning has been refered to it. Its first adjustment
was carried out by Defense Mapping Agency (DMA), by sequentially adjusting
triangulation blocks, being this the first realization of SAO-69 in Brazil. Hereafter this
realization is simply referred to as SAD-69.

The readjustment of SAO-69 took part of a ten years long projetc called REPLAN,
which was concluded in 1996. This task was carried out using the software GHOST -
Geodetic adjustment using Helmert blocking Of ,S.pace and Ierrestrial data, involving
4,939 geodetic points. The estimated variance factor was equal to 1.493, refering to the
adjustment of 16,91 J horizontal directions, 389 astronomical azimuths, 378 astronomical
stations, 257 geodetic baselines (triangulation), 1,277 geodetic baselines (traverses), 179
doppler coordinates (point positioning), and 1,198 GPS baselines. In order to better deal
with network deformations, II auxiliary parameters were included, namely 7 scale
parameters for the classical network, 1 orientation parameter for the astronomical
azimuths and 3 translation parametres for the doppler stations (Costa, 1996). This
realization is hereafter simply referred to as SAD-69/96.

The origin point of the network - Chua, did not have its coordinates, nor its vertical
components and nor its associated geoidal undulation altered. Also unchanged remained
the origin azimuth Chua-Uberaba. In this way, the difference between SAD-69 and SAD-
69/96 coordinates represent the improvement as a consequence of (a) the insertion ofGPS
control points into the geodetic network; (b) the new methodology of computation; (c) the
simultaneous adjustment of all observations. This paper deals with the new realization of
the South American Datum in Brazil.

Assessment of SAD-69/96 accuracy

An experiment was carried out in order to assess the accuracy of SAD-69/96. For this
purpose the IBGE Brazilian GPS network was taken as a reference. This network is
composed of 187 GPS stations, of which 49 coincident with the classical network. A
number of ] 5 GPS stations were not used due to connecting problems with the classical
network.

This experiment involved the computation of geodetic distance and difference in

orthometric height between station positions as given by the adjustment of the GPS
network and its corresponding ones derived from a number of 34 "global adjustments". In
each one of these global adjustments the GPS observations of the station under scrutiny
were removed, meaning that this sation was been considered as a classical station only.

206
The largest planimetric difference is equal to 2.889 m; the smallest equal to 0.015 m; with
an average value equal to 0.319 ± 0.550 m. The largest altimetric difference is equal
to -4.377 m; the smallest equal to -0.132 m; with an average absolute value equal to 1.239
± 1.075 m. As far as Chua is concerned, the planimetric and altimetric differences are
equal to 0.396 m and 0.619 m, respectively.

By inspecting standard deviations with respect to the avarage values, it can be noted that
there are values with significant magnitude within the sample, indicating a non-
homogeneity of the data. These results seem to indicated a non-parallelism between
SAD-69/96 and WGS-84, even though dealing with a small sample size. Probable causes
for that may be:
• the non-uniform distribution of the GPS stations used for controlling the classical
network;
• the use of difTerent processing strategies in the computation of GPS stations as far as
modelling is concerned. For example, the usc of triple difference /()r processing long
baselines.

Transformation between SAD-69 and SAD-69/96

Two geometric transformation have been tested, one using only three translation
parameters - like the officially adopted in Brazil by IBGE for the transformation between
SAD-69 and WGS-84, and the other one the seven-parameter similarity transformation,
indicated by most of the literature elsewhere. For the three-parameters model (translations
AX, AY, AZ along the X-, Y-, Z-axis), the following results were obtained:

AX= - 0,248 m
AY= 1,202 m
AZ = - 1,908 m.

Taking the significance level a equal ] 00% for the Chi-square test, the "confidence
interval" for the estimated variance factor has to be equal to 1.016. Taking as hypothesis
in the adjustment that all coordinates are of same precision, the standard deviation value
for the acceptance of the test on the estimated variance factor for the estipulated
significance level is equal to 2.769 m. Table 1 shows the values (in metres) of the largest
and smallest residual (V), and its rms value for the X, Y and Z coordinates:

Ymax Vmin rms

X 10.680 -7.689 2.062
Y 9.138 -13.009 1.830
Z 17.314 -30.385 3.949

Table 1 - Residual and rms values for the three-parameters model.

207
For the seven-parameters model (translations L1X, L1 Y, L1Z along the X-, Y-, Z-axis, scale
K,and rotations Rx, Ry,Rz along the X-, Y-, Z-axis) follows the results:

L1X = -]2,84] m
L1 Y = 14,366 m
L1Z = 4,990 m
K = 0.00000308
Rx= -0,1991"
Ry = 0,] 353"
Rz = 0,0670".

Applying the same procedure as before,the standard deviation IS equal to 1.979 m.

Likewise the three-parameter case, table 2 can be build:

Vmax Vmin rms

X 8.200 - 9.730 1.281
Y 8.631 -12.083 ].474
Z 19.643 -25.630 2.833

Table 2 - Residual and r111S values for the seven-parameters model.

Obviously, the seven-parameter model allows a better mapping of the coordinate

transformation from SAO-69/96 to SAD-69, such as the results seem to indicate.
However, it is evident that the use of this model is not enough to handle the situation
treated in this paper because it cannot absorb all deformations existing in the network.

A new proposal for the transformation between SAD-69 and SAD-69/96

The new proposal consists in dividing the network in regions, applying geometric
transformation to the set of neighbouring stations, as opposed to the several solutions
found elsewhere in the literature (Junkins, 1990; Nakiboglu et el., 1994; Brunini et aI.,
1996; Sillard & Boucher, 1996; Vanicek & Steeves, 1996). The transformation should
allow the evaluation of all elements which define the relative positioning of the coordinate
systems (translations, rotations, non-orthogonality and scales). This transformation is the
Afinne Transformation in which every axis has a specific treatment.

The metodology applied for the regionalization of the network was the Delaunay
Triangulation, since it maximizes the inner angles of the tetrahedra fonned (in the case of
a 3D space). In this case, the best geometry the region can be used. For every station of the
network there is a number of tetrahedra associated with it. The region of interest for this
particular station is defined by both its neighbours and by itself There is a huge number of
material concerning Delaunay Triangulation available electronically (Dubois, 1997).

208
 The Affinne Transformation is used only after defining the regions. In the first step, to
generate preliminary coordinates; in the second step, a new transformation is then applied
in order to map with sufficient precision the coordinates from SAD-69/96 to SAD-69, or
vice-versa. In the second step, instead of transforming coordinates what is really being
done is transforming corrections. For that, it is necessary to identify the tetrahedron which
contains the point of interest. The solution given for this is to work with homogeneous
coordinates, since in this case the coordinates of the point belonging to the tetrahedron
have values always positives and never larger than one.

The methodology was tested on the set of stations for the transfonnation from SAD-
69/96 to SAD-69. The larger difference found in absolue value is equal to 4* 10-9. This
behaviour was already expected since the determination of the corrections imply a
deterministic solution, because the degrees of freedom is equal to zero. Preliminary testes
with non-coincident stations or the nctwork havc shown good results, either with the
direct and the inverse transformation.

Due to the coordinate difference between SAD-69 e SAD-69/96, the building of the
Delaunay Triangulation yielded different (triangulation) regions for both systems. For
example, the station number 10022 belongs to 28 tetrahedra in the SAD-69/96 and to 12
tetrahedra in the SAD-69. In this way, the parameters derived from the Affinne
Transformation related to the mapping of coordinates and corrections may be different for
certain regions of the network.

It is important to stress that this methodology provides good results probably due to the
large number of stations involved, all belonging to the same geodetic system - in the
definition level, for two different realizations in distinct time. Another important point to
mention is that the methodology can be directly applied for points within the tetraedra. For
points outside them, an extrapolation should not be used. Other approach still to be
worked on could be applied

Conclusions

The use of geometric transformations alone cannot absorb all the deformations existing in
the geodetic network. This fact can be degrade for those applications which require to
map coordinates from one system into another. The methodology under consideration is
based on the alternative that the transfonnation can be done with respect to a certain
region of the network associated with the possibility that all degrees of freedom in space
are quantified, namely, 3 translations, 3 scale factors, 3 rotations and 3 parameters due to
the relative orientation of the axis (non-orthogonality). This can yield more expressive
results as far as modelling network deformations is concerned.

209
 It can be shown the improvement brought by inserting GPS baselines into the classical
networks. For the case presented in this paper, it is obviuos that more GPS stations would
be desirable. However, the densification with GPS stations should follow a carefull study.

Finally, the transformation between geodetic systems should be regarded as a very

important topic due to its implications into other sciences, such as Cartography, specially
in the Brazilian case in which all mapping activities remain attached to the South
American Datum.

References

Brunini, c.; Olondriz, .I.; Rodriguez, R.C. Determinacion de Parametros de

Transformacion entre los Sistemas de Campo Inchauspe 1969 y WGS-84 en Argentina.
Presented to IV Congresso Intcrnac.ional de Ciencias de la Tierra. Santiago. August, 5-9,
1996.

Costa, S.M.A. Ajustamento da Rede Planimetrica do Sistema Geodesico

Brasileiro. Rio de Janeiro, Fundacao Instituto Brasileiro de Geografia e Estatistica.
Technical Report. July, 1996. Not published.

Dubois, G. The AI-GEOSTATS WEB SITE. URL: http://java.eijre.iti

rem/gregoire. August, 1997.

Fischer, 1. The Basic Framework of the South American Datum of 1969. Presented
to XII Pan American Cosultation on Cartography. Panama, 1973

Junkins, D. The National Transformation for Converting Between NAD27 and

NAD83 in Canada. D. C. Barnes ( Ed. ). In Moving to NAD'83, The Canadian Institute of
Surveying and mapping. Ottawa, 1990.

Nakiboglu, S.M.; Eren, K. & Shedayed, A.M. Analysis of distortions in the

National Geodetic Network of Saudi Arabia. Bulletin Geodesique (1994) 68:220-229.

Sillard, P. & Bouch~r, C. Improvement of the transformation between ITRF and

Doppler-Realized WGS-84. Journal of Geodesy (1996) 70:768-780.

Vanicek, P. & Steeves, R.R. Transformation of coordinates between two

horizontal geodetic datums. Journal of Geodesy (1996) 70:740-745.

210
THE IGS REGIONAL NETWORK ASSOCIATE ANALYSIS CENTER
FOR SOUTH AMERICA AT DGFIII

Wolfgang Seemiiller and Hermann Drewes

Deutsches Geodatisches Forschungsinstitut, Aht.I
Marstallplatz 8, D-80539 MiinchenlGermany

ABSTRACT

In response to IGS's call for participation in the pilot project for the densification of the
ITRF, DGFIII is acting on behalf of the SIRGAS project as a RNAAC for South America.
The analysis results of the first year of activity are presented here in terms of the internal
stability of the station coordinates and the discrepancies with respect to global solutions. The
r.m.s. deviations for both internal and external comparisons are in the centimeter level.

INTRODUCTION

Early in 1996 the International GPS Service for Geodynamics (IGS) called for participation
in a pilot project to densify the IERS Terrestrial Reference Frame (ITRF) by regional
networks and a distributed data processing by Regional Network Associate Analysis Centers
(RNAAC). DGFI/I is acting on behalf of the SIRGAS project (Sistema de Referencia
Geocentrico para America del Sur) as a RNAAC for South America. Beginning in mid of
1996 all available data of permanently observing GPS stations in the mainland of South
America and surrounding regions are processed routinely and forewarded as SINEX files to
the IGS Global Data Centers.

RNAAC-SIR NETWORK

At the beginning of the pilot project the SIR network consisted of 12 stations; due to the
sparse coverage of South America with permanent GPS stations, all these stations were also
processed by at least one of the global analysis centers. Recently the data of 22 stations are
processed (Figure 1). Stations exclusively included in the RNAAC SIR network are the new
Brazilean permanent stations Born Jesus Lapa, Cuiaba, Curitiba, Imperatriz, Manaus,
Presidente Prudente and Vi90sa.
 260' 280' 300' 320' 340'
40' r---""'t'-----=-h;r---+---4;....~-......:.t:_-_,. 40·

20' -j--~-+_T__7_+_-p..~~--t-----t----+_'~_4 20'

-~
T
" ..00 ··-··'-+----1-0.
o·

C
CALAPAGOS ; , : , y i FORTAlEZA

rANAUS IMPERAjZ

"'"1~
ASCEHSjON

BOM JESUS lAPA

AREiQlNPA CUIASA IBRASIUA
·20' -j----+----t=='rt-----.r4>i~--...,.d;=---+_-_4·20·

EASTER ISLAND

·40' - j - - - - + - - - - t - - H ' - - ' " 7 " ' 1 ' - t - - - - - t - - - - - ! - - - l . 4 0·

-60' - j - - - - + - - - - t - - - - - + - - - - - + - - - - - ! - - - l . 6 0 ·

260' 280' 300' 320' 340'

Figure 1: Regional Network processed by RNAAC-SIR

DATA PROCESSING

The RNAAC-SIR coordinate solutions are generated weekly using the automated Bernese
software version 4.0, the so-called Bernese Processing Engine (BPE, Rothacher and
Mervart, 1996). Up to the end of 1996 aHP workstation was used, since early 1997 the soft-
ware is running on an IBM AIX workstation. Characterizing features of the performed
solutions are:

Final IGS orbits and Earth orientation parameters applied;

Cut off elevation angle 15 0 , sampling rate 2 minutes for single day adjustments;
Residual tropospheric zenith delays every four hours, a priori sigmas applied with
respect to the prediction model: first parameter ± 5 m absolute,
following parameters ± 10 cm relative;
Ambiguities partly resolved according to statistical tests,
remaining ones estimated as real numbers;
Station coordinates estimated in IGS orbit frame, applying a priori sigmas of ± 1 m.

212
The weekly solutions are forwarded as R-SINEX (Software INdependent EXchange
Format) files to the IGS Global Data Center. They are then combined by the IGS Global
Associate Analysis Centers (GNAAC) with the global solutions (G-SINEX) to the
polyhedron solution (P-SINEX).

RESULTS

Figures 2, 3 and 4 demonstrate the stability of the processing results. Figure 2 shows the
internal stability of the weekly RNAAC SIR solutions for one regional and one global
station. The coordinate variations are in the centimeter level. Some deviations - similar in
all stations - are obviously due to reference frame effects.

STATION CURITIBA
• NORTH COMPONENT .a. EAST COMPONENT • UP COMPONENT
3
2 _-1_
1 -
Ea
u
-1
-2 ...
-3

860 865 870 875 880 885 890 895 900 905 910

STATION FORTALEZA ~
• NORTH COMPONENT .a. EAST COMPONENT • UP COMPONENT
2

a
E
U
-1
-2
-3
-4

860 865 870 875 880 885 890 895 900 905 910
June 30, 1996 Week June 28, 1997

Figure 2: Internal stability ofRNAAC results: Coordinate variations for a regional (Curitiba)
and a global (Fortaleza) station

Figure 3 shows the deviation of the weekly RNAAC solution with respect to the combined
global polyhedron solution of the GNAACs at MIT (USA) and Newcastle University (GB),
respectively. It is obvious that the discrepancy of the regional station, which is only
processed by RNAAC-SIR, is smaller than the discrepancy of the global station, which is
also processed by the GNAACs independently.

213
 STATION CURITIBA
MIT (P-SINEX) - SIR (R-SINEX) NORTH COMPONENT NCL (P-SINEX) - SIR (R.SINEX)

§:w: 1~~~=~ : j EAST COMPONENT

:! ' i- ~~~
MIT (P-SINEX) - SIR (A-SINEX)
•• ••

UP COMPONENT
.-_.-. -.-.
MIT (P-SINEX) - SIR (R-SINEX) NCL (P-SINEX) • SIR (R-SINEX\
AA ... '
2
1
E 0
U -1 1 -------I-
I
-2 -,--- - - I

860 865 870 875 880 885 890 895 900 905 910
Week

STATION FORTALEZA
MIT (P-SINEX) - SIR (R-SINEX) NORTH COMPONENT NCL (P-SINEX) - SIR (R-SINEX)
.. ... .l.

EAST COMPONENT
MIT (P-SINEX) - SIR (R-SINEX) C:... ( - ,,'-
• • •

UP COMPONENT
MIT (P-SINEX) - SIR (R-SINEX)
Jt-------.-II ... ... ....
NCL (P-S!NEX) - SIR (R-SINEX)

860 865 870 875 880 885 890 895 900 905 910
Week

Figure 3: Differences ofRNAAC-SIR coordinates with respect to GNAAC global

polyhedron solutions after seven parameter (Helmert) transformation

214
Figure 4 gives the variations of the weekly MIT solutions for two selected SIR stations. This
is to demonstrate the external stability of station coordinates.

STATION CURITIBA
• NORTH COMPONENT .t. EAST COMPONENT • UP COMPONENT
4
3 -- - - ~I--_--r-_----'---:;::"------~-I­
2 . -.1--_ .. _-
Ea
1 -----~----l--~--~----
u
·1 I
·2 L 1-----~-
-3 . _ - - . - - - - - , - - - : - .--"""'---.----;----,,- - - -
-4~-~==~==~==~~~~~==~~~=-~
--~-
==-~~
i ==~~~=t
860 865 870 875 880 885 890 895 900 905 910

STATION FORTALEZA
• NORTH COMPONENT

860 865 870 875 880 885 890 895 900 905 910
June 30, 1996 Week June 28, 1997

Figure 4: Consistency of GNAAC MIT weekly polyhedron solutions

Table 1 shows a comparison of coordinates of selected stations on the South American

mainland from independent solutions. Considering the station velocities (V) we find an
agreement in the centimeter level.

CONCLUSION

The first analysis of RNAAC SIR processing demonstrates satisfying results. The South
American network has grown rapidly in a short time interval. The number of stations will
further increase in the near future and may serve as a permanent control net in South
America. Thanks are due to the contributing countries, in particular to the Instituto
Brasileiro de Geograffa e Estatistica (IBGEIDEGED) for its efforts in providing the data of
the Brazilean network with highest reliability.

215
 Table 1: Comparison of independent coordinate solutions

Station IGS P-SINEX SIRGAS ITRF94 V (IGS)

Name (alv 1997) (1995.4) (1993.0) (mmJa)

Bogota X 1744399.119 .102 .138 0

Y - 6116037.797 .852 .863 10
Z 512731.622 .597 .575 25
Kourou X 3839591.431 .438 .447 - 1
Y - 5059567.536 .532 .538 - 6
Z 579956.925 .906 .867 15
Fortaleza X 4985386.629 .661 .643 1
Y - 3954998.564 .585 .560 -10
Z - 428426.490 .515 .550 14
Brasilia X 4115014.088 .106 - - 1
Y - 4550641.509 .513 - -14
Z - 1741444.065 .086 - 12
Presidente X 3687624.320 .310 - ( 0)
Prudente Y - 4620818.562 .571 - ( -6)
Z - 2386880.388 .407 - (10)
Curitiba X 3763751.647 .639 - ( 0)
y - 4365113.753 .768 - ( -6)
Z - 2724404.735 .755 - (10)
Arequipa X 1942826.731 .727 .693 14
Y - 5804070.240 .249 .243 -10
Z - 1796893.971 .994 4.035 16
Santiago X 1769693.350 .314 .258 22
Y - 5044574.135 .146 .114 -17
Z - 3468321.050 .077 .112 11
La Plata X 2780102.976 .979 - 4
Y - 4437418.854 .845 - -10
Z - 3629404.586 .604 - 14

REFERENCE

Rothacher, M. and Mervart, L. (eds): Bernese GPS software version 4.0. Astronomical
Institute, University of Berne, 1996.

216
THE REDEFINITION OF THE GEODETIC REFERENCE SYSTEM
OF URUGUAY INTO SIRGAS FRAME

Subiza, Walter H.
Servicio Geografico Militar
8 de Octubre 3255, Montevideo 11600, Uruguay

Perez Rodino, R., Barbato, F.

Instituto de Agrimensura, Facultad de Ingenieria
Universidad de la Republica, Julio H. Reissig 565
Montevideo 11600, Uruguay

Alves Costa, Sonia M.

Instituto Brasileiro de Geografia e Estatistica
Av. Brasil 15671, Parada Lucas, 21241·051
Rio de Janeiro, RJ, Brasil

Abstract

In 1908, a systematic geodetic surveys started with cadastral and mapping objetives.
Classical operations were carried out along main paralel and meridian chains till 1960's
when the Fundamental Geodetic Network was finished.
A common adjustment using the variation coordinates method was performed in 1963,
by the US Army Map Service. The whole 1st • Order data available was evaluated giving a
mean direction error of 0".71 and 0".89 misclosure triangle. The data included 15 bases,
12 astronomical stations and 248 geodetic stations. Hayford Ellipsoid was used and the
Yacare Datum was established.
After that, normal geodetic densification of the 1st • Order was performed. The
introduction of the EDM in the 1960's, allowed for electronic polygons, getting field
operations faster than classical method.
 During 1992 the GPS stroke on in our country and quickly remained as the principal tool
to get positioning. The 1150.000 scale cartography was rapidly finished with a couple of
submeter receivers.
Two years later, the SGM bought three Ashtech Z-XII receivers, and the Facultad de
Ingenieria (Universidad de la Republica) got Wild-Leica ones. Later on others institutions
went into GPS world. The fact allowed Uruguay to participate in SIRGAS Project, and
densify the network.
With all this information, a new redefinition of the National Geodetic Reference System
was felt needed, in order to adopt a worldwide geodetic and cartography system. GHOST
(Geodetic adjustment using Helmert Blocking of Space and Terrestrial data) software
from Geomatics of Canada was used to carry the task out in the frame given by SIRGAS
Project. Major steps taken and some results are presented here.

Historical development of the Uruguayan Geodetic Network

The geodesy as a cartographic and positioning tool was born in 1908 with the first
systematic surveys in the center of the country. Bamberg and Huetz theodolites and
directions methods were used.
In 1914 a triangulation network plan was designed, including four meridian chains and
five paralel ones. Clarke ellipsoid was adopted as reference surface.
The Laguna Merim International Chain with Brasil was measured in 1912-1918. In
1917-1919 a survey of the capital of the country was made. The International Rio Uruguay
Chain with Argentina was observed between 1918 and 1937. In 1946 the first
triangulation adjustment was calculated, using Least Squares Method. The complete
boundary chain with Brasil first, and the argentinian counterpart later were calculated
using 1930 International Ellipsoid as surface reference. In the 1940's a complete Rio
Negro Valley survey, in the middle of the country, was carried out, with approximately
400 km long. Densification of the geodetic net continues, using better equipments for
directions and distances (Wild T4 and T3 theodolites, invar bases and electronics distance
meters), until 1961 when thtefundamental network planned in 1914 was finished.
First major adjustment was performed in USA in 1965, using coordinates variation
method, involving 248 geodetic station and using the 1930 International Reference
Ellipsoid. In 1969 the main geodetic network was included in the South American Datum
69, adding at this time 4 new astronomics stations and 7 geodetic bases in order to
improve the net. In 1990 approximately 420 stations have been survey and calculated for
the first order. The satellite technology for positioning began in 1993, and Sirgas Project
put the net at regional and global level.

SIRGAS Project

Three main geodetic stations (LaPlace stations) were measured with Ashtech ZXII, during
a ten days-long campaign in 1995 processing (95/147, 148 and 149 Jdays). At the same
time five more stations were observed during three days long with Wild-Leica receivers.

218
 In 1997 a two days campaign was performed (97/116 and 117 Jdays), adding three more
stations. Montevideo and Bella Vista stations were used as a link with Sirgas campaign.
(Figure 1)

Satellite data processing

Software used for processing GPS data was Bemese version 4.0, developed by Berna
University.
GPS data to process comprised two campaigns, observed with both double
frecuency/code, Ashtech Z-XII and Leica receivers. 1995 GPS campaign was
simultaneous with Sirgas one, using same observation parameters and having eigth
stations (three Sirgas), it was processed with 15 0 elevation and 30 seconds recording time.
1997 campaign, included five stations measured in two days long, processed at 15 0
elevation and 60 seconds recording time.
As baselines distances were ranging from 100-250 lan, the pre-processing and
processing was developed on modeled observable L3 (ionosphere free linear
combination), the files that contain local ionosphere parameters was not necessary be
generated. As a manual suggestion was used Saastamoinen troposfere model without
meteorological values observed. The receiver phase center correction was applied
according file IGS_PHAS.Ol .
The strategy to create single diferences files was shortest in both cases.(Figure 2) It means
that only number of stations - 1 independent single differences were formed, choosing the
shortest paths. The cycle slips automatic detection and correction was made in triple
diference, The ambiguities resolution was developed in double diference, solving for 90%
of ambiguities in a linear combination L5 (wide-lane). The solved ambiguities were
introduced as known values in the final solution.
The final combination of 95 and 97 results were made using coordinates and variance-
covariance matrix of coordinates from session/day processings, showing 4 mm maximun
difference on the baselines and sigmas less than 3 mm on final cartesian coordinates
Montevideo was used as fixed station. These informations were converted in Ghost input
format to integrate in the same adjustment extraterrestrial and terrestrial data. A seven
parameters Helmert transformation between Bemese and Ghost GPS software results was
made. Residual less than 3 mm and a RMS of 3 mm was obtained, showing a good
agreement between campaigns/software. An scaled adjustment of both 95 & 97 campaigns
in Ghost result on a 1.017 variance factor.

Geodetic Network Adjustment Results and comments

The final adjustment was made scaling GPS campaign according the preprocessing
variance factor obtained (100.7 for 1995 and 108.8 for 1997), resulting on a l.655
variance factor. Sirgas stations were included as weighted observed positions with
Sirgas's sigmas.
From this first approximation, a computer program was adapted in order to get geoidal
altitudes for each station and improve the Z component. Regional geoidal model

219
 URUGUAYAN OEQDETICNBTWORK. GPS LlNBS MOUNTIID FOR. PREPR.OCESSING
'WITH GEOIDAL MODEL WGS84 95 &; 97 Campaigns

)
)'

FiglD"e 1
Figure 2

HORIZONTAL VBC1'ORS OF DISTORTIONS CONFIDENCBBLlPSBERRORS

YACARB DATIJM - GRS 80 GPS NETWORK. STATIONS

3""-
~-J . .

Figure 3

220
developed for Uruguay in 1993, with a thousand of gravimetric stations and GEMT2
model as reference, was used.
Geoidal altitudes were added on the adjustment file later.

Adjustment data is at follows:

* 417 geodetic stations with geoidal altitudes (including 11 GPS stations)
* 2337 horizontal directions
* 66 distances
* 11 astronomic azimuths
*4 parameters to determine (3 electronic distances and 1 azimuth)
* 1341 normal equations generated.
* 1162 system degrees of freedom
*0 fixed station
*3 SIRGAS weighted position equations
* 65 position difference equations
After adjustment a 1.65 variance factor was reached and the maximum horizontal
residual was ± 4".
A total of 181 ( 43.4 %) geodetic stations were classified as First Order, 232 (55.6 %) as
Second, 4 as Third and Four Order (1 %).
The weighted contributions of different kind of observations were: directions 92.31 %,
distances 3.12 %, azimuths 1.07 %, position difference equations 3.49 %, weighted
positions 0.01 %.
Raw azimuths without any reductions, had a maximum residual of ± I"., mostly of them
below of their a priori sigmas.
Distances were classified in 5 groups according the ppm showed (30 up to 3 ppm, 21 3-
10 ppm and 15 more than 10-35 ppm). The old invar bases were in first group, meanwhile
some electronic ones had to be eliminated due to high residuals, maybe by lack of correct
equipment calibration.
Horizontal distortions vectors of the network (Figure' 3) due to change of reference
system, showed a general direction of 2600 and a mean scalar value of 2 m in latitude and
-37 m in longitude, with a maximum of~8 m in the last one. It is clear the readjustment in
longitude caused by GPS observations.
Latitude-longitude standard deviation have values from 0.07-0.44 m, and confidence
elipse errors shown a value ranging from 0.2 to 1.2 m for classical observations and some
tens of mm for GPS stations (Graphic 4).

Conclusions

A major readjustment of the geodetic net was succesfully performed in the Sirgas Frame
last July/97. Combination of tridimensional and clasical geodetic observations were
possible through the Ghost software use.

221
 A new set of coordinates in ITRF/WGS84, is available to local or international
institutions/researches, and mainly for positioning or cartographic use. Clasical stations
showed confidence errors between 0.2-1.2 m and GPS ones, some cm.

Local cartographic system of reference will change to standard UTM 21 & 22 zones, with
WGS84 ellipsoid and coordinates

Further steps will be the development of a more accurate geoidal model introducing
some one thousand gravimetric stations and GPS data observed over the levelling net.

Acknowledgements

Special thanks to Don Beattie from Geomatics Canada for fully and patient support. Also
to Sirgas Project and Geomatics of CanadalIBGE, Brazil & Bern University, Switzerland
for kindly allow to use the adjust and GPS processing software.

References

Geodetic Survey Data Dictionary. 1987, Ghost documentation. Geomatics of Canada.

Bernese Manual Reference 4.0. 1997 Bern University, Switzerland.

222
GPS AMBIGUITY RESOLUTION FOR NAVIGATION, RAPID STATIC
SURVEYING, AND REGIONAL NETWORKS

Paul J. de Jonge
Delft University of Technology, Faculty of Geodetic Engineering
Thijsseweg 11, 2629 JA Delft, The Netherlands
E-mail: [email protected]

ABSTRACT

In this contribution an overview is given of the capabilities of methods for ambiguity

resolution in use nowadays. This is done in the context of SSG 1.157, "Ambiguity
Resolution and Validation", although the overview is not restricted to the algorithms used
within this group.
Methods from the fields of navigation, rapid static surveying, and regional networks are
compared, and similarities and differences are indicated. It is shown that the subdivision of
the algorithms by their applications is a rather arbitrary one, and that some of the
ambiguity resolution algorithms can be applied in all three fields.
The capabilities of state of the art algorithms for ambiguity resolution are indicated in
terms of maximum distances over which the ambiguities can be resolved and validated. In
particular the role of the observational time span and the set of observables used in the
estimation (single versus dual frequency, phase-only versus phase and pseudo range) is
addressed. As the ionospheric delay is an important reason for being unable to resolve the
ambiguities, the consequences of the next solar activity maximum at the end of the century
are briefly discussed.

INTRODUCTION

Traditionally, algorithms for the estimation of the integer GPS double difference (DD)
ambiguities have been developed for two different fields. On the one hand methods have
been devised for applications where a multiple of stations are occupied for several hours
until several days, and maximum inter-station distance can be of the order of thousands of
kilometers, (Dong and Bock 1989), (Blewitt 1989), (Mervart 1995). On the other hand
methods have been developed for rapid-static and navigation applications, where usually
only two stations are involved, the maximum distance is some tens of kilometers, and time
of occupation is of the order of seconds to minutes, or the receiver is moving: AFM
(Counselman and Gourevitch 1981), LSAST (Hatch 1990), FARA (Frei 1991), optimized
search using Cholesky (Euler and Landau 1992), LAMBDA (Teunissen 1993), GASP
(Corbett 1994), null-space method (Martin-Neira et. al. 1995).
At first sight it seems that ambiguity estimation for large regional networks is treated as
an entirely different problem, although conceptually there is no real difference with the
short baselines as far as the estimation of the integer ambiguities is concerned. Of course,
for longer inter-station distances more adequate mathematical models for e.g. tropospheric
and ionospheric delays, as well for the orbital parameters have to be employed.
If one looks more carefully, one may observe similarities between some of the
algorithms applied to the regional networks, and some of those applied to the short
baselines. The methods described in Dong and Bock (1989), Blewitt (1989) as well as the
'sigma' and QIF (Quasi-Ionosphere Free) method in the Bernese software (Rothacher et.
al. 1996) all use a sequential conditioning (fixing) of ambiguitie~. The conditioning makes
that subsequent ambiguities become more precise, and usually pushes the values of the
ambiguities conditioned on the previous ambiguities towards integer values. The schemes
differ in the criteria that are used for selecting the next ambiguity to be fixed, and in the
way parameters are modeled and constrained in the preceding float solution.
This principle of conditioning of ambiguities one also finds with (Talbot 1991) and in
the sequential conditional least-squares adjustment which is part of the LAMBDA method.
The latter differs from the methods above in the sense that if the resolution or adjustment
is concluded with a complete or partial vector of integer ambiguities a, it is guaranteed
that that this vector minimizes the integer least-squares criterion (a - a)' Qil (a - a), with
a the vector of float ambiguities, and Qa its variance-covariance matrix.
Due to the hybrid character of regional networks with inter-station distances from some
kilometers to several thousands of kilometers, and with some receiver-satellite
combinations observed less well than others, the chance to find a valid integer solution for
all ambiguities together is decreased, and thus one has to resort to a proper subset for
which such a solution exists.
Another difference between the regional network and the short baseline algorithms, is
that in the former no validation step is involved, other than comparing several solutions in
time. The baseline algorithms that are used in rapid static and navigation applications have
some sort of validation step which is based on the data itself, and thus can be computed
and evaluated as soon as the data are collected and processed.

Decorrelation.
In Teunissen (1993) it was shown that the DD ambiguities are strongly correlated,
especially when the observational time span is short, due to the small change in the
receiver-satellite geometry. Since this correlation makes the estimation of the integer
ambiguities via a sequential conditioning of the ambiguities far from efficient, a method to
construct a decorrelating transformation for the ambiguities was proposed in ibid.
Alternative methods can be found in Li and Gao (1997), and in Han and Rizos (1995),
where it is also shown that decorrelating the ambiguities often has a favorable influence on
the efficiency of other resolution methods too.

224
 The ambiguities of the one day solutions for the large regional networks are less
correlated, but still can be improved upon. This most likely will produce more ambiguities
that will pass the criteria that allow them to be fixed, thereby further improving the
eventual fixed solution. Possibly it could also reduce the time span for which a fixed
solution can be computed. For monitoring coseismic and postseismic deformation with
magnitudes up to several decimeters, and several mm/day, respectively, shortening these
time spans would be very welcome.

MEASURING PERFORMANCE OF AMBIGUITY RESOLUTION SCHEMES

Not all ambiguity resolution algorithms have been described in sufficient detail, such
that they can be implemented and tested, and for only a few algorithms the source code is
available. A complicating factor is that often 'improved' and 'combined' versions of
existing algorithms are proposed.
The algorithms that are used inside the large scientific GPS softwares like Bernese,
GAMIT and GIPSY, for resolution in regional networks are well described, and source
code is in principle available. Other methods that are described in detail, and for which
source code is available are the FARA (implemented in Bernese s/w) and the LAMBDA
method (de Jonge and Tiberius, 1996). The reason for this lack of willingness of making
algorithms available in the public domain, usually has its roots in the protection of the
commercial interests of the proprietor of the software. Needless to say that this makes a
full comparison of methods not very feasible. Despite these difficulties, Han and Rizos
(1995), and Hein and Werner (1995) show a comparison of some methods.
In the absence of having an accurate description of the various methods one can still
make bilateral comparisons of methods on the basis of an exchange of e.g. the float vector
of ambiguities and its variance-covariance matrix. Exchange of e.g. RINEX data itself has
to be discouraged since often one will end up comparing ways to compute the float
solution, instead of comparing the integer ambiguity estimation process itself.

Ground truth.
When evaluating the performance of ambiguity estimation one has to have some sort of
reliable (accurate) 'ground truth' available. This ground truth usually refers to the
coordinates of the points surveyed, or functions thereof. For stationary points one can look
at the repeatability of discrepancies of the coordinates with respect to a well surveyed
point.
For moving points, the discrepancies between the coordinates computed with respect to a
well surveyed point, and those that have another well surveyed point as reference, can be
used. Another approach is to look at the discrepancies between the computed length
between two moving points, and the a priori known length. For example the case of two or
more antennas on a moving vehicle.
As far as a threshold for the discrepancies is concerned, one should take care to stay at
the conservative side. If the ground truth is established with a sufficient accuracy level, the
ambiguities are constrained to the correct integers, and (in case of longer distances) the
troposphere and ionosphere are modeled correctly, the discrepancies should be at the few

225
centimeter level, and not at the sub-cycle length level as sometimes is proposed. Fixing
one of a set of ambiguities at a value wrong by one cycle does not necessarily cause a shift
in coordinates of the level of one cycle.
One can also inspect the continuity of the ambiguities over a not too short time span; the
time span should not be too short since the ambiguities often show a remarkable
persistence in sticking to the wrong integer values. So preferably this approach should be
used in combination with one of the strategies above. Some ambiguity resolution
algorithms use the continuity of the ambiguities as a measure for validation; i.e. if the
ambiguities stay constant for a predefined period of time after convergence, they are
believed to be the correct ones. A disadvantage of this validation approach is that there is
no clear link with probability theory, and that the minimum time for a reliable fix can
never be less than the time needed for convergence of the ambiguities.
Things get more complicated once the distance between the reference point and the point
of which the coordinates are to be established, becomes larger. To get a sufficient good
accuracy for the ground truth of the new point it will have to be occupied for a longer
period, but in setting up an experiment one is usually inclined to put most effort and time
in the actual experiment while neglecting the establishment of a solid ground truth. This is
often also the case for data sets that are not especially collected for the purpose of testing a
resolution scheme.
For the regional networks, one usually looks at the repeatability of the coordinates over
the days, and at the a posteriori variance of the observables after fixing the ambiguities.

How well can we resolve the ambiguities?

The answer to this question depends on to whom you ask it. The answers from people in
practice that are engaged in making 'ambiguity resolution engines' which will give with a
minimum of user interference a near 100 percent correct answer, will be far more
pessimistic than the answer from a research person. For instantaneous ambiguity
resolution, i.e. using dual frequency phase and pseudo range data of one epoch, answers
vary between a maximum baseline distance of 5 km using at least 7 satellites and a
maximum baseline distance of 10 to 15 kilometer with at least 6 satellites and a validated
solution for 98-100 percent of the time.
For kinematic solutions successful resolution is reported for single frequency phase and
pseudo range over distances of 20 to 30 kilometer for time spans of some minutes using at
least 6 satellites, while for distances less than 20 kilometers using dual frequency phase
and pseudo range data, time spans can be as short as 1 minute.
Results that were reported for longer baselines (but that only could be validated using
external information) were e.g. correct resolution over a baseline of 88 km using dual
frequency phase and pseudorange on Ll, for a time span of 1 minute for 20 out of 120
cases, and correct resolution over a baseline of 70 km using 15 minutes of dual frequency
phase data. These are rather individual cases, but they indicate where the frontiers of
ambiguity resolution lie.
As far as the large regional networks are concerned, correct ambiguity resolution over
baselines up to 2000 kilometer using a time span of one day using dual frequency phase
observables is routinely performed for 85-95 percent of the baselines. In the case where the

226
ambiguities are resolved in a network approach, a similar percentage of the ambiguities
can be resolved .

... and when is the ambiguity resolution becoming difficult?

Signal multipath decreases the success rate of ambiguity resolution, and can render
resolution impossible even over short (several tens of meters) baselines. Just before and
after data gaps, the success rate is also decreased.
In areas with a large ionospheric activity, as e.g. near the geomagnetic poles, or near the
geomagnetic equator, the chances of successful ambiguity resolution are also greatly
reduced. As we are heading again towards a maximum sunspot number which is predicted
to occur in late 1999 or early 2000 (Joselyn et. al. 1997) resolving the ambiguities will
probably become more difficult. During the period from 1999 through 2005, severe
geomagnetic storms are likely to occur. A comparison with the last sunspot maximum is
difficult, because since then the capabilities of receivers and algorithms have seen large
improvements, and at present a whole new generation of GPS satellites (Block IIR) is
being deployed.

ACTIVITIES OF THE SPECIAL STUDY GROUP AND OUTLOOK

A start has been made in exchanging data in the form of float vectors and their variance-
covariance matrices. As far as the exchange of source code is concerned, until now only
the LAMBDA method has been made general available, and some bilateral exchange of
prototype software has been established. A more general exchange of source code will be
further encouraged in the future.
The theory of ambiguity resolution and validation is still not finished; for validation
purposes more research into e.g. the discrete probability distribution of the integer
ambiguities is needed. Research is ongoing into ambiguity resolution for GNSS2 with
possibly a third frequency and for GLONASS possibly in combination with GPS.
Papers produced by the members of SSG 1.157 over the last two years, can be found at
the web page of the SSG: http://www.geo.tudelft.nl/mgp/people/paullssg_1157.html. Also
at these pages a still growing bibliography of more than 200 papers concerned with
ambiguity resolution and validation can be found: http://www.geo.tudelft.nl/mgp/people/
paul/amb.html
The contributions of Simon Corbett, Mattia Crespi, Hans-Jtirgen Euler, Shaowei Han,
Hans-Jorg Kutterer, Herbert Landau, Zuofa Li, Manuel Martin-Neira, Joao Galera Monico,
Daniela Morujao, Benjamin Remondi, Stefan Schaer, Peter Teunissen, Christian Tiberius,
Ming Yang, and Wolfgang Werner are acknowledged.

REFERENCES

Blewitt, G. (1989): Carrier phase ambiguity resolution for the Global Positioning System
applied to baselines up to 2000 km, Journal of Geophysical Research, Vol. 94, No. B8,
pp. 10187-10203.

227
Counselman, c.c., and S.A Gourevitch (1981): Miniature interferometer terminals for
earth surveying: ambiguity and multi path with Global Positioning System, IEEE Trans.
on Geoscience and Remote Sensing, Vol. 19, No.4, pp. 244-252.
Corbett, SJ. (1994): GPS single epoch ambiguity resolution for airborne positioning and
orientation, PhD thesis, Dept. of Surveying, University of Newcastle upon Tyne, UK.
Dong, D.-N., and Y. Bock (1989): Global Positioning System network analysis with phase
ambiguity resolution applied to crustal deformation studies in California, Journal of
Geophysical Research, Vol. 94, No. B4, pp. 3949-3966.
Euler, H.-l, and H. Landau (1992): Fast GPS ambiguity resolution on-the-fly for real-time
applications, Proceedings of Sixth International Geodetic Symposium on Satellite
Positioning, Columbus, OH, March 17-20, pp. 650-659.
Frei, E. (1991): Rapid differential positioning with the Global Positioning System, PhD
thesis, University of Berne, Switzerland, Geodetic and Geophysical Studies in
Switzerland, Vol. 44.
Han, S., and C. Rizos (1995): A new method of constructing multi-satellite ambiguity
combinations for improved ambiguity resolution, Proceedings of ION GPS-95, Palm
Springs, 12-15 September, pp. 1145-1153.
Hatch, R. (1990): Instantaneous ambiguity resolution, Proceedings KIS90, Banff, Canada,
September 10-13, Springer Verlag, pp. 299-308
Hein, G.W., and W. Werner (1995): Comparison of different on-the-fly ambiguity
resolution techniques, Proceedings ION GPS-95, Palm Springs, 12-15 September, pp.
1137-1144.
Jonge, P.J. de, and C.CJ.M. Tiberius (1996): The LAMBDA method for integer
estimation: implementation aspects, Delft Geodetic Computer Centre LGR series, No.
12.
Joselyn, lA, J.B. Anderson, H. Coffey, K. Harvey, D. Hathaway, G. Heckman, E.
Hildner, W. Mende, K. Schatten, R. Thompson, AW.P. Thomsom, and O.R. White
(1997): Panel achieves consensus prediction of solar cycle 23, Eos Trans. AGU, Vol. 78,
No. 20, pp. 205,211-212.
Li, Z., and Y. Gao (1997): Construction of high dimensional ambiguity transformation
for the LAMBDA method, Proceedings KIS97, Banff, Canada, 3-6 June, pp. 409-416.
Martin-Neira, M., M. Toledo, and A Pelaez (1995): The null space method for GPS
ambiguity resolution. Proceedings of DSNS'95, Bergen, April 24-28, Paper No. 31, 8 pp.
Mervart, L. (1995): Ambiguity resolution techniques in geodetic and geodynamic
applications of the Global Positioning System, PhD thesis, Astronomical Institute of the
University of Berne, Switzerland.
Rothacher, M., G. Beutler, E. Brockmann, S. Fankhauser, W. Gurtner, J. Johnson, L.
Mervart, S. Schaer, T. Springer, and R. Weber (1996): The Bernese GPS software
Version 4.0, edited by M. Rothacher and L. Mervart, Astronomical Institute, University
of Berne, Switzerland.
Talbot, N.C. (1991): Sequential phase ambiguity resolution for real-time static differential
GPS positioning, Manuscripta Geodaetica, Vol. 16, pp. 274-282.
Teunissen, PJ.G. (1993): Least-squares estimation of the integer GPS ambiguities, Invited
lecture, Section IV Theory and Methodology, lAG General Meeting, Beijing, August,
also in Delft Geodetic Computing Centre LGR series, No.6, 16 pp.

228
 GPS ANTENNA AND SITE EFFECTS

Jan M. Johansson

Onsala Space Observatory, Chalmers University of Technology, S-43992 Onsala, Sweden

Abstract
The improvement in precision obtained from GPS observations over recent
years has revealed problems related to the local conditions at the GPS sites. In
order to further improve high precision GPS positioning, orbit determination, and
estimation of atmospheric parameters, investigations of site dependent effects are
required. Concerns have been raised regarding the antennas and the monuments
used and the long- and short-term mechanical and electromagnetic stability of
the sites. Here, we review the problems associated with site-specific errors and
present recommendations on how to eliminate or minimize these effects.

Introduction
During the last few years, an increasing number of permanent GPS sites have been es-
tablished. The demonstrated repeatability of horizontal position estimates for regional
networks is currently of the order of 2 mm and typically a factor of 3-5 greater for
the vertical component. There are many advantages to continuously operating GPS
networks. Stable pillars with fixed antennas eliminate errors associated with variations
in the measurement of the local vector from the reference marker to the phase reference
point of the antenna. For fixed pillars in a continuously operating network, the refer-
ence marker is usually a fixed, well-defined point on the antenna. In addition, denser
position estimates (spatially and temporally) decrease the statistical uncertainty of
the results. Continuously operating networks may also serve as a global or regional
reference frame for different types of regional and local surveys. Another essential ad-
vantage is the increased ability to study and eliminate unmodeled systematic effects
on daily estimates of site positions, both short- and long-term effects.
To be able to constrain the common mode of motion, sometimes in the submillimeter
range, in a regional or local network a strong reference network is needed. The origin of
the reference frame must be maintained with a high degree of robustness. In addition,
orbits must be compatible with the reference frame. For this purpose data from the
IGS network and other permanent network need to be regularly examined in detail.
Site-specific errors at permanent stations may introduce errors in the determination
of satellite orbit parameters and in the estimate of site positions. In the following we
adress the problem of site-specific errors and present some recommendations on how
to handle these errors.

International Association of Geodesy Symposia, Vol. 118

Brunner (ed.), Advances in Positioning and Reference Frames
© Springer-Verlag Berlin Heidelberg 1998
 Site Specific Errors
We have chosen to divide the site specific error sources into three subgroups. The
first group consists of problems associated with the receiver, antenna, radome, and
the signal. These are effects that will not, in general, change on a day-to-day basis.
However, they might introduce biases in the solution. As long as nothing changes the
effect stays the same. If something changes, such as the satellite constellation or the
elevation cut-off angle, the results will be affected. The influence is especially obvious
looking at the estimates of the vertical components and precipitable water vapour
where bias terms can be introduced (e.g. [Elosegui et al., 1995]; [Niell et al., 1994];
[Niell, 1996]). Such a bias could seriously affect the interpretation of the GPS data.
The second group represents areas of site effects that will vary, but will only periodi-
cally affect the measurements. Precipitation, multipath, atmospheric pressure loading,
and atmospheric gradients are probably the most important of these, but others may
be discovered.
Finally, the third group consists of errors that might affect the long-term stability
of a site such as the location of the site, ground, and the monument. Most of the
material related to this specific group of site errors are rather new. These errors may
seriously affect the reference frame and the geodynamical projects.

GPS Antennas
It has been found that antenna-to-antenna phase differences can introduce range biases
at the several centimeter level, which may limit the precision of the measurements
[Rocken, 1992]. Differential phase errors due to GPS antennas will not only affect the
precision in GPS networks with different types of antennas, but also in networks using
identical antennas if the network covers a large spatial area (baseline lengths ~1000
km) [Schupler and Clark, 1991]; [Schupler et al., 1994]. Differential phase errors in
regional networks (baseline lengths ~1000 km) using identical antennas are dependent
on the electromagnetic environment around each individual antenna.
The problem of antenna mixing was addressed at the IGS Analysis Center Workshop
in Silver Spring, 1996. Two sets of phase calibration corrections (PCC) tables have been
put together based on material presented by Mader and MacKay [1996], Rothacher and
Schar [1996], and Meertens et al. [1996a] to be used by the IGS Analysis Centers and
others in the GPS community: (1) a set of "mean" phase center offsets and (2) a set
of elevation-dependent PCC and offsets relative to the Dorne Margolin T antenna.
Since the PCC values are all relative to the Dorne Margolin T antenna some effects
of antenna mixing still remain. Even with the same type of antenna the variation in
the apparent phase center as a function of elevation angle will influence the results
on longer baselines. Therefore the task of getting absolute calibration of the antennas
through, e.g., chamber measurements or simulation software may be essential for some
applications even though these calibration values most likely will change when the
antenna is deployed in the field.
Effects like these can of course be reduced by utilizing antennas less sensitive to
scattering from external structures. One way to achieve this is to reduce the side- and

230
back-lobe levels of the amplitude patterns by means of well designed ground-planes.
For this purpose new antenna designs have been proposed (see e.g., [Alber, 1996]; [Ware
et al., 1997]; [Jaldehag, 1995]; and [Clark et al., 1996]). Futhermore, several groups are
currently developing methods to perform absolute field calibration of antennas (see
e.g., [Wiibbena et al., 1996]) and insitu calibration of antenna/pillar systems.

Antenna-Pillar System and the Signal

Here we concentrate on the site-dependent error associated with the electromagnetic
coupling between the antenna and its nearby environment (e.g., [Tranquilla, 1986];
[Tranquilla and Colpitts, 1988]). The total electromagnetic field of an antenna which
radiates a signal in the presence of conducting structures may be expressed as a super-
position of the transmitted field and the fields scattered (i.e., reflected and diffracted)
by the structures. By reciprocity, the same is true for a receiving antenna. The signifi-
cance of the scattered field depends on the degree of electromagnetic coupling between
the antenna and the scatterer, that is, the distance to the scatterer and the size and
reflectivity of the scatterer. Signal scattering affects both the amplitude and phase of
the received GPS signal, presumably independently at each site in a network. This
independence creates differential phase errors.
Scattering from structures in the vicinity of the antenna effectively changes the an-
tenna phase pattern, and, thus, affects the precision of the carrier phase measurements
of the GPS signal. In studies by Elosegui et al. [1995] and Jaldehag et al. [1996a] it
was shown that estimates of the vertical component of baselines formed between sites
using identical antennas were dependent on the minimum elevation angle of the data
processed. Both studies found that the elevation-angle- dependent systematic effect
was associated with non-identical pillar arrangements, causing differential phase er-
rors due to scattering from structures associated with the mounting of the antenna to
the pillar, and with the pillar itself. Even the most perfectly calibrated antenna the
antenna phase pattern will change when attached to a pillar.
Jaldehag et al. [1996a] demonstrate that estimates of the vertical component of
many baselines strongly depend on the minimum elevation angle (elevation cutoff an-
gle) of the data analyzed. A significant part was found to be due to differential phase
errors caused by scattering from structures associated with the mounting of the an-
tenna to the pillar and with the pillar itself. As the precision and accuracy of GPS
measurements improve in general, antenna phase pattern variations due to different
pillars and antenna mounts could be the major error source in just a few years, if not
now. Modeling of the scattering effect, or rather the complete phase response of the
antenna system, including the pillar, is an important issue for future improvements of
the GPS technique.

Radomes - Protective Covers

At several permanent GPS sites located in areas with periodically severe environmental
conditions (snow, rain) radomes have been employed. Until recently, most radomes in
use have had a conical shape.

231
 All materials have some effect on a electromagnetic wave. Radomes appear to delay
and refract the GPS-signal in a similar way as snow [Jaldehag et al., 1996bj. Several
groups have recently been investigating effects due to the excess signal path delay
through the radome. Different radomes have been tested in anechoic chambers [Clark
et al., 1996]; [Meertens et al., 1996b] as well as in field tests [Meertens et al., 1996b];
[Jaldehag et al., 1996c]. All tests show that a conical cover may cause cm-level vertical
errors when the tropospheric delay parameter is estimated. The recently employed
hemispheric radomes seems to show much less elevation dependence. The influence
on the tropospheric wet delay estimates and subsequently, the vertical component will
only be on the 1-2 mm level. We can conclude that all radomes effect the GPS signal
at some level and in form of an excess signal path delay which will map into other
parameters in the GPS software. The effect of the protective covers can most likely be
misinterpreted as a tropospheric effect in a similar way as snow. The effect is more or
less constant and may be calibrated or modeled.

Precipitation
Signal propagation delay during snow storms has been investigated by, e.g., Tranquilla
and Al-Rizzo [1993] and Tranquilla and Al-Rizzo [1994] who demonstrated that due
to the localized nature of many snow storms differential effects may cause systematic
variations at the centimeter level in estimates of the vertical coordinate of site position.
Systematic variations introduced by snow storms may, however, if short-lived (minutes
to hours), be reduced to a high degree by data averaging. A potentially more serious
effect of heavy snow precipitation is the accumulation of snow on the top of the GPS
antenna and on its surroundings, such as on the top of the GPS pillar or, when present,
on the radome covering the antenna. This accumulation may last for days, weeks, or
months. Webb et al. [1995] reported variations on the order of 0.4 m in estimates
of the vertical coordinate of site position. The variations were correlated with the
accumulation of snow over the antenna. Variations at the several centimeter level in
estimates of the vertical coordinate of site position strongly correlated with changes in
the accumulation of snow on top of GPS antennas have also been observed by others
[Jaldehag et al., 1996b]; [BIFROST project members, 1996]; [Meertens et al., 1996a].
The results indicate that the variations in the vertical coordinate of site position can
be fully explained by reasonable accumulations of snow which retard the GPS signals
and enhance signal scattering effects.

Horizontal Atmospheric Gradients and Air Pressure Loading Effects

In the data processing the atmosphere is normally considered to be spherically strati-
fied. We assume that one equivalent zenith wet delay value determines the wet delay
in any direction, given a certain elevation angle. More advanced models, using more
parameters to describe the atmosphere, have been proposed as alternatives to this very
simplified model (e.g., [Davis et al., 1993]; [MacMillan, 1995]). Several groups are now
implementing possibilities to estimate horizontal gradients in the software [Bar-Sever
and Kroger, 1996]; [Chen and Herring, 1996].

232
 The lack of pressure data available during the GPS analysis can be the reason for
different errors. During the entire G PS processing we have to model many external
and internal effects on the crust of the earth. One effect currently not modeled is the
pressure loading. The vertical position of the GPS receiver changes due to different
atmospheric pressure loading the Earth [vanDam and Herring, 1994J. Extreme values
could affect the vertical component of the GPS estimates on the cm level. These
effects are of course related more to the general presssure field in the region rather
than to a specific site. To properly model this effect a grid of pressure data has to be
available. Unfortunately, it is very difficult to isolate these effects from other elevation-
angle-dependent effects (multipath, scattering, snow/ice, etc.). Small variations in
the vertical component are also caused by these other errors. We are thus not in the
position of being able to correct for horizontal atmospheric gradients and loading errors
optimally. At this point, theoretical studies are needed to quantify these effects, and
to understand how we can best deal with these problems.

Local Stability and Monumentation

As G PS measurements have become more precise and are more frequently acquired, the
issue of monumentation and site stability has become more important. The long-term
contribution to the maintainance and densification of the global reference frame could
be seriously affected by instable sites. The IGS network consists of a large variety
of monuments established on top of everything from solid bedrock to buildings. The
long-term stability of the reference frame and products associated with it, such as
the orbits, are at issue here. Much attention is currently focused towards motions
of geodetic monuments. These motions have been found by some researchers to be
random-walk-like (e.g. [Johnson and Agnew, 1995J; [Johnson et al., 1996]) while others
find no evidence for random walk behavior (e.g. [Mao et al., 1996J; [Davis et al., 1996]).
An ideal G PS monument would move in response only to the tectonic motion of the
Earth. However, location, ground, and the environment at ground surface can have
dramtic impact on the long-term stability of a site. The implication of this type of
power-law noise is serious if the data are used to estimate low-frequency characterisitics
of a time series such as the slope (deformation rate) . Mao et al. [1996J and Davis et al.
[1996J found no tendency of a random-walk like behavior possibly because the records
where not long enough to see a random walk component above the noise in the low
portion of the signal. It is also quite possible that monument motion may depend
critically on the monument design and the site locations.
Nevertheless these investigations will continue and are most effectively addressed
using continuous G PS measurements gathering data in a large variety of local con-
ditions and GPS satellite configurations. There are design techniques which can be
employed to mitigate this unwanted influence, most of which involve anchoring the
monument to several points at depth and isolating the monument from surface mate-
rial. Detailed spectral analyses and examination of the long time series available for
some of the global sites. Monument and local stability problems could also manifest
themselves with a periodic behavior, and be correlated with atmospheric conditions

233
and precipitation.

Conclusion
Site-specific errors cannot be separated out when data from the global IGS sites are be-
ing used to determine orbits and reference frame. To be able to constrain the common
mode of motion, sometimes in the submillimeter range, in a regional or local network
a strong reference network is needed. The origin of the reference frame must be main-
tained with a high degree of robustness. In addition, orbits must be compatible with
the reference frame. For this purpose the IGS sites need to be better examined. We
especially found that the problems associated with the antenna-pillar system and the
signal distortions have to be addressed. The effect of the antenna and signal related
errors are constant from day-to-day but are biasing products like the orbit determina-
tion, station time series, and precipitable water vapor time series. Any changes either
at a station or in the GPS-data analysis strategy might change this bias and thereby
affect the daily products and the reference frame. The other important issue that needs
attention is the long-term stability of the sites and the monuments used in the IGS
network. This is especially important bearing in mind that local and regional continu-
ously operating GPS networks are now used to detect motion at the level of 1 mm/yr
or less.

Acknowledgements
Many thanks to all members of lAG SSG 1.158. Special thanks to Jim Zumberge, Jim
Davis and Chuck Meertens for valuable contributions to the manuscript. This research
was in part supported by the EC Environment and Climate Research Programme, the
Swedish Natural Research Council, and the Swedish National Space Board.

References
Alber, C., Millimeter Precision GPS Surveying and GPS Sensing of Slant Path Water Vapor,
Ph.D.Thesis, Univ. Colorado, Dec. 1996.

BIFROST project members, GPS Measurements to Constrain Geodynamic Processes III

Fennoscandia, EOS, Transactions, American Geophysical Union, 35, 1996.

Bar-Sever, Y., and P. M. Kroger,Strategies for GPS-Based Estimates of Troposphere Delay,

Proc. of the 9th International Technical Meeting of the Satellite Division of the Institute
of Navigation, vol. 1, 615-623, 1996.

Chen, G., and Herring, T.A., Effects of Atmospheric Azimuth Asymmetry in the Analysis of
Space Geodetic Data, Eos Trans. AGU, 77(46), Fall Meet. Suppl., F453, 1996.

Clark, T.A., B. Schupler, C. Kodak, and R. Allshouse, GPS Antennas: Towards Improving
Our Understanding of Factors Affecting Geodetic Performance, Eos Trans. AGU, 77, Fall
Met. Suppl. 1996

234
Davis, J. L., G. Elgered, A. E. Niell, and C. E. Kuehn, Ground-based measurement of gradi-
ents in the "wet" radio refractivity of air, Radio Bci.,28, 1003-1018, 1993.

Davis, J.L., RA. Bennet, J.M. Johansson, H.G. Scherneck, and J.X Mitrovica, Spectral
Analysis of Site Position Variations from Project BIFROST, Eos Trans. AGU, 77(46),
Fall Meet. Suppl. F453, 1996.

Elosegui, P., J.L. Davis, RT.K Jaldehag, J.M. Johansson, and A.E. Niell, Effects of Signal
Multipath on GPS Estimates of the Atmospheric Propagation Delay, Eos Trans. AGU,
15(44), p. 173, Fall Meeting Suppl., 1994.

Elosegui, P., J. L. Davis, R T. K Jaldehag, J. M. Johansson, A. E. Niell, and 1. 1. Shapiro,

Geodesy Using the Global Positioning System: The Effects of Signal Scattering on Esti-
mates of Site Position, J. Geophys. Res., 100, 9921-9934, 1995.

Jaldehag, R T. K, Space Geodesy Techniques: An Experimental and Theoretical Study of

Antenna Related Error Sources, Ph. D. Thesis, Tech. Rept. No. 276, School of Electrical
and Coputer Engineering, Chalmers University of Technology, Gothenburg, Sweden, 1995.

Jaldehag, R T. K, J. M. Johansson, P. Elosegui, J. L. Davis, A. E. Niell, B. O. Ronnang,

and 1. 1. Shapiro, Geodesy Using the Swedish Permanent GPS Network: Effects of Signal
Scattering on Estimates of Relative Site Positions, J. Geophys. Res., 101, 17814-17860,
1996a.

Jaldehag, R T. K, J.M. Johansson, J.L. Davis, and P. Elosegui, Geodesy using the Swedish
Permanent GPS Network: Effects of snow accumulation on estimates of site positions,
Geophys. Res. Lett., 23, 1601-1604, 1996b.

Jaldehag, RT.K, J.M. Johansson, and J.L. Davis, Environmental Effects on the Swedish
Permanent Network, Eos Trans. AGU, 77(46), Fall Meet. Suppl., F453, 1996c.

Johnson H.O.and D.C.Agnew, Monument Motion and Measurements Crustal Velocities, Geo-
phys. Res. Lett., 22, 3533-3536, 1995.

Johnson H.O., F.K Wyatt, D.C.Agnew, and J. Langbein, Evidence for Power-Law Behavior
in Geodetic Time Series: Should we care, and if so, Why?, Eos Trans. AGU, 11(46), Fall
Meet. Suppl. F453, 1996.

Mao A., C.G.S. Harison, and T. Dixon, Monument Stability of the Permanent GPS stations
and Tide Gauges, Eos Trans. AGU, .77(46), Fall Meet. Suppl., F453, 1996.

MacMillan, D. S., Atmospheric gradients from very long baseline interferometry observations,
Geophys. Res. Letters, 22(9), 1041-1044, 1995.

Mader, G.L., and MacKay, J.R, Calibration of GPS Antennas, Proc. fGB Analysis Center
Workshop, Silver Sping, MD, 1996.

Meertens C., Rocken, C., Braun, J., Exner, M., Stephens, B., and Ware, R, Field and
Anechoic Chamber Tests of GPS Antennas, Proc. fGB Analysis Center Workshop, Silver
Sping, MD, 1996a.

235
Meertens C., Braun, J., Alber C., Johnson J., Rocken C., Van Hove, T., Stephens, B., and
Ware, R, Antenna and Antenna Mounting Effects in GPS Surveying and Atmospheric
Sensing, EOS Trans. AGU, 77(46), Fall Meet. Suppl., F453, 1996b.

Niell, A. E., P. Elosegui, J.L. Davis, 1.1. Shapiro, RT.K. Jaldehag, and J.M. Johansson,
Reduction of Signal Multipath Effects on GPS Estimates of Site Position (abstract), Eos
Trans. AGU, 75(44), p. 171 Fall Meeting Suppl., 1994.

Niell A.E., Reducing elvation-dependent errors for ground-based measurements, EOS Trans.
AGU, 77(46), Fall Meet. Suppl., F453, 1996.

Rocken, C., GPS antenna mixing problems, UNAVCO Memo, UNAVCO, Boulder, Colorado,
1992.

Rot hacher , M., and Schar, S., Antenna Phase Center Offsets and Variations Estimated from
GPS Data, Pmc. IGS Analysis Center Workshop, Silver Sping, MD, 1996.

Schupler, B. Rand T. A. Clark, How different antennas affect the GPS observable, GPS
World, 32-36, Nov./Dec., 1991.

Schupler, B. R, R L. Allshouse, and T. A. Clark, Signal characteristics of GPS user antennas,

Navigation, 41, No.3, 277-295, 1994.

Tranquilla, J.M., Multipath and imaging problems in GPS receiver antennas, paper presented
at Fourth International Symposium of Satellite Positioning, DMA, Austin, Tex., 1986.

Tranquilla, J.M., and B.G. Colpitts, GPS antenna design characteristics for high precision
applications, paper presented at ASCE Specialty Conference GPS-88: Engineering Appli-
cations of GPS Satellite Surveying Technology, Am. Soc. Civ. Eng., Nashville, Tenn., May
11-14, 1988.

Tranquilla, J.M. and H.M. AI-Rizzo, Theoretical and experimental evaluation of precise rela-
tive positioning during periods of snowfall precipitation using the global positioning system,
Manuscripta Geodaetica, 18, 362-379, 1993.

Tranquilla, J.M. and H.M. AI-Rizzo, Range errors in Global Positioning System during ice
cloud and snowfall periods, IEEE Trans. Antennas and Pmpagat., 42, 157-165, 1994.

vanDam, T. M., and T. A. Herring, Detection of Atmospheric Pressure loading using Very
Long Baseline Interferometry Measurements, J. Geophys. Res., 99, 4505-4517, 1994.

Ware, R, C. Alber, C. Rocken, and F. Solheim, Sensing integrated water vapor along GPS
ray paths, Geophys. Res. Lett 24, pp 417-420, 1997.

Webb, F.H., M. Bursik, T. Dixon, F. Farina, G. Marshall, and RS. Stein, Inflation of Long
Valley Caldera from one year of continuous GPS observations, Geophy. Res. Lett., 22,
195-198, 1995.

Wiibbena G., Menge F., Schmitz, M., Seeber, G., and Volksen, C., A New Approach for Field
Calibrations of Absolute Antenna Phase Center Variations, Pmc. ION GPS-96, 1996

236
 PRECISE GPS POSITIONING IMPROVEMENTS
BY REDUCING ANTENNA AND SITE DEPENDENT EFFECTS

Gunter Seeber, Falko Menge, ChristofVOlksen

Institut fur Erdmessung, Universitiit Hannover, D-30167 Hannover, Germany

Gerhard Wubbena, Martin Schmitz

Geo++, D-30S27 Garbsen, Germany

ABSTRACT

In order to estimate the characteristics of GPS antennas this paper describes a new
approach for the estimation of absolute phase center variations (peV) in a field calibration.
The main objective of this approach is the elimination of mUltipath from the GPS
observable and the elimination of any influence of the reference antenna. This is achieved
by forming the so called mean sidereal day time difference between observations of
successive days. Since the satellite geometry of GPS satellites repeats every mean sidereal
day multipath does as well. Therefore, by differentiating observations of two different days
multipath can be eliminated. To gain information about the pev special rotations have to
be applied to the antenna of interest. The paper shows the results for three different antenna
types.
A second application of the mean sidereal day time difference shows the estimation of
small position changes between consecutive days. The mean sidereal day time difference is
formed in the same manner as for antenna calibration, however, without any rotation.
Therefore multipath and pev are eliminated from the observable. We achieved with
observation periods of 10 minutes a horizontal position resolution of only 1.2 mm which is
better than results from standard data processing.

INTRODUCTION

Precise positioning with GPS has now reached a level of accuracy where the remaining
error sources are identified as effects caused by phase center variations (peV) of the
antenna and site effects like multipath. Phase center corrections are generally important for
GPS observations with mixed antenna design to take into account the different phase
patterns of each antenna type. In addition it is not easy to separate between tropospheric
errors and phase center biases. Multipath can have a significant influence on precise
positioning depending on the site. Therefore the impact of multi path on the position
estimation should be considered carefully.
 So far the pev are estimated by two different methods. In an absolute sense it is possible
to determine the pev in an anechoic chamber [1] and in a relative sense by a field
procedure [2] with a known antenna as a reference, whose pev are estimated based on the
first method. The main disadvantage of the relative field calibration is the influence of the
used sites. One can never assume that the field calibration is free of multipath. Therefore it
is very likely that the determined pev highly depend on the multipath effects and are
correlated with the selected site. With a new method [3] we have shown a field procedure
for the calibration of antennas in an absolute sense. This new method is based on the
elimination of multipath by using observation data of different days.

ELIMINATION OF MULTIPATH

A site which is totally unaffected by multi path does not exist. Thus, field procedures for an
antenna calibration as well as short time GPS observations are disturbed by this effect. The
results are incorrectly estimated pev values which are site dependent and a decreased
resolution for the point positioning with short observation times.
A short summary of the basic concept for the method to eliminate multipath is given in
this paragraph. It is based on the repeatability of the satellite geometry and therefore of the
multipath signals after one mean sidereal day [4, 5]. New investigations concerning the
exact period for the repeatability of the satellite constellation during a four days test in
1997 revealed slightly different values for each satellite in a range between 240 s and 256 s
instead of the generally assumed value of 3 min 56 s (236 s). The period for the
constellation repeatability was estimated by three methods. First of all, double differences
of two successive days were cross-correlated in the time domain. The maxima show the
time lags and clearly indicate the significant periodical appearance of multi path after a mean
sidereal day (see [3]).
Orbiting Times for Single Satellites Furthermore the periods are
-240 1 2 3 4 0 7 II 10 14 15 10 17 18 111 21 22 23 24 25 2!1 27 211 30 31
I) calculated from the correlation
of elevation! azimuth time
series and from individual
I>< -245
ephemeris. The results of these
three methods are in good
+
agreement. Figure! shows
the orbiting times for different
-250
satellites computed from the
ephemeris of four days. The
periodical appearance is used
to greatly reduce the effect of
-2515 multipath by subtracting the
un-differenced GPS
observable of two successive
Figure 1: The orbiting times for different satellites.
days, taking into account the
evaluated time lag for the
difference between a mean solar and a mean sidereal day, which should be calculated from
the data sets. The following is a simplified linearized notation of the phase observation

238
equations 1<1> in meters containing the design matrix subvector a; the receiver coordinate
corrections x; the receiver and satellite clock error dt. and dT, respectively, scaled to meters
by the speed of light co; the ambiguity N scaled to meters by the wavelength A.; the error
terms d for ionosphere (ION), troposphere (TROP), multipath (MP), and phase center
variations (peV); and the noise of the phase ccp:

(1)

The subscript i and superscript j stand for receivers and satellites, respectively. Building a
mean sidereal day time difference 8SID eliminates the multi path, phase center variation and
the complete geometric information. The following observation equation does not contain
any information about geometry, since the design elements a are almost identical on two
successive days:

The remaining terms comprise the mean sidereal time differences of every component,
which are small for a short baseline (i.e. atmospheric errors) and/or are correctly modeled
in the GPS processing package GEONAP [6] (i.e. clock errors). The noise level of the
observable changed due to error propagation to 8SIDccp. In our paper [3] we demonstrated
the elimination of multipath applying the mean sidereal day time difference 8SID • Together
with the dramatically reduced multipath the resolution of the phase measurements improved
by a factor of 1.5.

MODELLING AND ESTIMATION OF ABSOLUTE PHASE CENTER

VARIATIONS (PCV)

Although the mean sidereal day time difference eliminates site dependent errors, this
approach can be used for the determination of absolute phase center variations.
Observations are carried out on two days. One antenna is identically orientated on both
days, whereas the antenna to be calibrated performs vertical and horizontal rotations on the
second day only. Now, in extension to equation (2), the linearized observation equation (3)
for the time difference 8SID of the rotated antenna contains pev of two different
orientations:

8SlD1<l>; =co . (8 SlD dt i -8 SlD dTj)-A.·8 slD N! -8 SlD d ION ; +8SlDdTROP;

(3)
(ao,zo)j _
+dPCV d(ao+~a,zo+.1z)j + s:: SID ..
i PCV i U C <I> •

The observable for the estimation of the absolute pev is the difference in the pev of two
antenna orientations to an identical satellite:

(4)

239
 c
o

l
iii

30 eo eo 120 ISO 110 210 240 270 300 330 38(1

Azimuth [01
a)

eo
e...
c
0
~
>

.
CIl
iii

30 60 DO 1:20 150 lao 210 240 210 300 330 .)10

Azimuth [01
b)

c
.2
<0
>
CIl
iii

30 80 go 120 1500 110 210 240 270 300 330 360

Azimuth [01
e)

Fig. 2: LI-PCV [rnm], a) Ashteeh 700228 Rev. B, b) Trimble 4000 ST LII2 Geod,
e) Trimble Dome Margolin Choke Ring.

240
In equation (4) ao und Zo are the orientation parameter of the first day, while ~a and ~z
represent the applied orientation changes on the second day.
As noted before, relative observables are used for the generation of the absolute phase
pattern. Thus, only the topology of the pattern can be described. The absolute size is not
known, but will be treated as a constant clock error or a hardware delay in the GPS
position solution. The term "absolute antenna calibration", however, is still valid for the
approach, because the phase center variations are determined independently from the phase
pattern of a reference antenna. For the estimation of the elevation (and azimuth) dependent
PCV serves a spherical harmonic function [3]:

nrnax n
d pcv (a,z) = LL(Anm cosma + Bnm sin ma)Pnm(cos z). (5)
n=Om=O

Prun are normalized associated Legendre functions. Azimuth a and zenith angle z refer to
the position of a particular satellite in the antenna coordinate system. To enable horizontal
rotations and vertical tilts of the GPS antenna, a special mount must be used. Through the
tilts a reception of undisturbed satellite signals from higher elevations contribute to a
reliable PCV determination at low elevations in the antenna coordinate system, even at
elevation zero.
Using this approach three different antenna types have been calibrated so far (Ashtech
700228 Rev. B Notches, Trimble 4000 ST LIIL2 Geod, Trimble Dome Margolin Choke
Ring). The Ll- and L2-PCV are determined by a spherical harmonic function of degree 10
and order 5. Only the LI-PCV of these antenna types are shown in figure 2. The calibration
for the Ashtech and the Trimble Choke Ring antenna was verified in a second independent
experiment showing good agreement. Clearly, elevation and azimuth dependent variations
are visible. All antennas have very large variations at low elevations in common, especially
for the Trimble Choke Ring reaching up to several cm. The performance of the antennas
seems to be rather poor at very low elevations.

DETECTION OF SMALL DEFORMATIONS

Another application of the proposed observable 8sm is feasible for the determination of
small position changes. A baseline is observed on two different days. The linearisation,
identical on both days for each station, is performed in the same manner as shown in
equation (1). Using the reduced observable, i.e. the mean sidereal day time difference, one
should yield the baseline components with a higher precision due to the improved phase
measurement resolution. In an experiment for the verification of this approach a short
baseline was observed on two different days. One of the antennas was mounted to a special
mechanical sledge (figure 3) which allows shifts in two horizontal directions normal to each
other. These can be performed with high precision (better than 0.1 mm). The observations
on the first day were carried out without changing the position of the sledge. Small position
changes at a rate of 2 mm were applied during the second day. The sledge was shifted
every 10 minutes, first to the north and then to the east. This procedure was repeated
several times. The dashed lines in figure 4 and 5 show the performed shifts.

241
 Fig. 3: Sledge

Figure 4 shows the displacements computed from the observations of the second day
only. The positions were estimated with a simultaneous adjustment of Ll and L2, called
LX. It is clearly visible that the agreement with the known shifts is rather poor. In figure 5
one can see the estimated positions after applying the mean sidereal day time difference
with the GEONAP software module GNSDIF. The agreement between the computed
displacements and the actual displacements is very good. Therefore the shifts of the sledge
can easily be monitored by the mean siderial day time difference. The accuracy of the
horizontal position estimation is in the order of 1.2 mm for an observation windows of 10
minutes only. Hence, with this method it is possible to monitor rather small deformations at
the Imm-Ievel, even with very short observation times. Although observations on two days
with an identical setup (mount, antenna etc.) are required, a number of applications are
conceivable (e.g. dam monitoring).

CONCLUSIONS

Multipath is still the most limiting factor for precise posItioning. With the proposed
approach, the mean siderial day time difference, it is possible to eliminate multipath from
the GPS observables. This derived observable can be used to observe small position
changes from one day to another or estimate absolute pev of an antenna.

242
The main advantages of the absolute antenna calibration are that it is a field procedure, free
of multipath, and independent from a reference antenna's phase pattern. Through the
rotations and tilts of the antenna areas without any observations are avoided, i.e. northern
hole or low elevations, and the antenna is completely covered. Still, further examinations
are requisite, e.g. errors introduced by the mount (errors in the rotation), influence of the

Horizontal Positions
Data Blocks of 10 Minutes (LX)

0.020

E 0.010

0.000

-0.0~00.L01-0---0-'.0'-00-~--O"""
.O-
10----0.-'-
02- 0------.J
1m]

Fig. 4: True positions (dashed line) versus estimated positions (solid line) without BSID •

Horizontal Positions (after GNSDIF)

Data Blocks of 10 Minutes (LX)

0.020

E 0.010

0.000

- 0.010 '--_~_-"-_~_-L_ _~_-'--_~_---'

-0.010 0.000 0.010 0.020
1m]

Fig. 5: True positions (dashed line) versus estimated positions (solid line) using BSID •

243
mount on the mUltipath environment, and comparison with other sources by applying the
PCV in operational GPS evaluations. A future goal is the automatic calibration with a very
precise robot to avoid instrumental erros of the mount and save observation time.
The first experiments with the mechanical sledge for the determination of small
deformations are already promising. The resolution of the positioning is clearly improved.
Further experiments have to be carried out in order to improve this method for different
applications, e.g. deformation monitoring in permanent GPS arrays.

Acknowledgment: Parts of this work are funded by the German Bundesministerium fUr
Bildung, Wissenschaft, Forschung und Technologie (BMBF, No. 03PL022B).

REFERENCES
1. Schupler, B.R., T.A. Clark, R.L. Allshouse: Characterizations of GPS User Antennas:
Reanalysis and New Results, In: Beutler, G. et al. (Eds.). GPS Trends in Precise
Terrestrial, Airborne, and Spaceborne Applications. lAG Symposium, No. 113, Boulder,
Colorado, USA, 1995.
2. Rothacher, M., S. Schaer, L. Mervart, G. Beutler: Determination of Antenna Phase
Center Variations Using GPS Data, Workshop Proceedings, IGS Workshop Special
Topics And New Directions, May 15-18, Potsdam, Germany, 1995.
3. Wiibbena, G., F. Menge, M. Schmitz, G. Seeber, C. VOlksen: A New Approach for
Field Calibration of Absolute Antenna Phase Center Variations, Proceedings of ION
GPS-96, 9th International Technical Meeting, September 10-17, Kansas City, Missouri,
USA,1996.
4. Bock, Y.: Continuous Monitoring of Crustal Deformation, GPS World, June 1991, pp.
40-47, 1991.
5. Genrich, J.F.,Y. Bock: Rapid Resolution of Crustal Motion at Short Ranges With the
Global Positioning System, Journal of Geophysical Research, Yo. 97, No. B3, pp.
3261-3269, 1992.
6. Wiibbena, G.: The GPS Adjustment Software Package -GEONAP- Concepts and
Models, Proceedings of the Fifth International Symposium on Satellite Positioning, Las
Cruces, New Mexico, pp. 452-461,1989.

244
IGS ORBIT, CLOCK AND EOP COMBINED PRODUCTS: AN UPDATE

Jan Kouba and Yves Mireault

Geodetic Survey Division
Natural Resources Canada
Ottawa, Ont., Canada

Abstract

Since January 1994 when the International GPS Service for Geodynamics (lGS) became
an lAG service, several classes oflGS products have been generated on daily and weekly
basis. IGS orbits, clocks and EOPs, based on weighted averages of the seven IGS
Analysis Center (AC) daily solutions, are among these products. Throughout 1996 and
1997, a number of improvements, both in quality and product delivery times were
realized resulting in enhanced quality of IGS combined products. The Final and Rapid
orbit, clock and EOP delays were reduced significantly on June 30, 1996. Delays were
reduced from one month to eleven days for the Final products and from eleven days to 24
hours for the Rapid products. New improved modeling and conventions, such as subdaily
EOP modeling, ITRF94 and EOP rate solutions, were introduced on June 30, 1996 by all
seven IGS ACs. These enhancements, along with the improved geometry of the IGS
tracking network, resulted in steady orbit, clock and EOP solution improvements. The
IGS combined orbit, clock and EOP products are currently approaching the 5 cm, 0.5 ns
and 0.1 mas precision level respectively. Since March 1997, a new IGS orbit prediction
product, also based on a weighted average, has been introduced. The new IGS orbit
prediction is available in real-time and is significantly better than the broadcast GPS
orbits. Typically, the IGS orbit prediction, when compared to the IGS Rapid orbits, is at
or below the 1 metre precision level, while broadcast orbits compare at the 3-5m RMS
level. Also, in March 1997, a new IGS LODIUTI combination based on a weighted
average of AC LOD solutions was officially introduced. Encouraging results were
obtained and are presented hereafter.
Introduction

Since 1994, the International GPS Service for Geodynamics (lGS) has been producing
combined products of GPS orbits/clocks and Earth Orientation Parameters (EOP) (Kouba
et aI., 1995, Kouba and Mireault, 1996; 1997). On June 30, 1996 (GPS Week 860), a
number of significant changes and improvements have been implemented. The former
IGS Rapid orbiticlocklEOP combination (lGR), based on EOP(lGS) and produced within
11 days after the last observation, became the IGS Final combination (lGS) replacing the
former IGS Final combination (Bulletin B (lTRF93)) produced about two months after
the last observation. New IGR orbiticlocklEOP products, produced within 24 hours from
the last observation, were introduced on GPS Week 860 (lTRF94) replacing the former
IGS Preliminary combinations (lTRF93) run on a trial basis only. The new IGR
combination is generated daily as opposed to a weekly cycle for the IGS Final products.
In addition, two new IGS combined products have been introduced: the IGS 2-day orbit
prediction (lGP) and the IGS LODIUTI combination. All the changes that have occurred
since January 1994 (GPS Week 734) are summarized in Table 1, including the product
names, orbitlEOP reference frame, orientation and availability after last observation.

Table 1: History of the IGS Rapid and Final Products

GPS Short IGS Name Reference/Orientation Availability Status Cycle

Weeks Name Frame

730-781 IGS IGS Final ITRF92/IERS Bu1l.B-ITRF92 2 months Official Weekly

IGR IGS Rapid ITRF92/IERS Bu11.A 14 days Official Weekly

782-802 IGS IGS Final ITRF93/IERS Bu11.B-ITRF93 2 months Official Weekly

IGR IGS Rapid ITRF93/IERS Bull.A 14 days Official Weekly

803-859 IGS IGS Final ITRF93/IERS Bull.B-ITRF93 2 months Official Weekly

IGR IGS Rapid ITRF93/ITRF93 11 days Official Weekly
834-859 IGP IGS Prelim. ITRF93/ITRF93 38 hours pilot Daily

860- ... IGS IGS Final ITRF94/ITRF94 11 days Official Weekly

860-901 IGR IGS Rapid ITRF94/ITRF94 24:00 UT Official Daily
902- ... IGR IGS Rapid ITRF94/ITRF94 22:00 UT Official Daily

895- ... IGP IGS Pred. ITRF94/IERS Bull. A -23:30 UT Official Daily

Orbit and Clock Evaluations

The long arc orbit evaluation was implemented for the IGS Final orbits to detect
problems that could affect the daily weighted average combinations and to assess the
consistency of individual AC solutions, including IGS combined orbits, over a one week
period. Ephemerides are analyzed for individual AC independently from the combination
process. The Long arc (La) orbit evaluation is described in more detail in the IGS 1994
Annual Report (Kouba et aI., 1995). LaRMS are summarized in Table 2.

246
 Starting with GPS Week 834 (Dec. 31, 1995), the IGS FinallRapid combined
orbits/clocks as well as all AC solutions which contain both the orbit and clock data, are
further evaluated by an independent single point positioning program (navigation mode)
developed at NRCan. This is done to verify clock solution precision and orbit/clock
consistency. Pseudorange data from three stations are used daily and their corresponding
position RMS (with respect to ITRF93 prior to GPS Week 860 and to ITRF94 since GPS
Week 860) are summarized in the weekly/daily IGS FinallRapid combination summary
reports. The three stations are Brussels in Belgium (BRUS), Usuda in Japan (USUD) and
Williams Lake in Canada (WILL). On average, horizontal and vertical point positioning
RMS are below the 50 cm and the 100 cm level respectively.

1996-97 IGS Rapid and Final Combination Results

AC orbits, clocks and EOPs are routinely combined into IGS official products on a daily
(Rapid products) and weekly cycles (Final products). Orbit, clock and EOP combinations
are described in more detail in the IGS 1994, 1995 and 1996 Analysis Coordinator
Reports (Kouba et al.,1995; Kouba and Mireault, 1996; 1997) and will not be described
here. Since GPS Week 803 (May 28, 1995), Polar Motion (PM) x and y coordinates and
since GPS Week 857, PM x and y rates have also been combined as weighted averages
from available AC PM values using orbit weights. Orbit, clock and PM overall RMS
with respect to the Final combination results for the period covering June 30, 1996 to
August 9, 1997 (GPS weeks 860 to 917) are summarized in Table 2.
Three types of orbit RMS are included in Table 2: the weighted combination RMS
(WRMS), the combination RMS, and the Long arc evaluation RMS (LaRMS). ACs used
in the Final clock combination are COD, EMR, ESA, GFZ, and JPL. NGS provides
broadcast clock corrections and SIO provides no clock corrections at all. ACs used in the
Rapid clock combination are EMR, ESA, GFZ and JPL. Starting on GPS Week 902/day
3 (April 23, 1997), USNO's Rapid products (USN) have been used in the IGS Rapid
combinations. NGS and SIO are excluded for the reasons mentioned above and COD, as
NGS, is excluded because it provides broadcast clock corrections in its Rapid
submissions. However, the clock information these ACs provide is still compared with
the combination results. The IGR orbit, clock and EOP results are included in the Final
combination for comparisons only.

Bad satellite orbit and/or clock solutions are excluded from the combination if they bias
the IGS combined solution but are included in the RMS computations. All exclusions are
reported in the IGS weekly/daily summary reports. For the IGS Final combination, the
best clock RMS have now reached the 0.5 ns level for more than one AC and the best
orbit position RMS have been approaching the 5 cm level (see Table 2). For the Rapid
combination (IGR), the best clock RMS have also reached the 0.5 ns level and the best
orbit position RMS vary between 5 and 10 cm which is quite remarkable considering that
these are available within a day after the last observation.

247
Table 2: Final IGS Orbit/ClockIPM Combination RMS
GPS Weeks: 860-917; WRMS: Weighted Rms; LaRMS: Long arc RMS
cm: centimetres; ns: nanoseconds; mas: milliarc-seconds

AC/Produets Orbit Clock PM RMS PM rates RMS

RMS WRMS LaRMS RMS x y x y
(em) (em) (em) (ns) (mas) (mas) (mas) (mas)

IGS 9
COD 7 5 9 0.9 .14 .32 .21 .21
EMR 10 10 13 0.5 .20 .33 .73 .64
ESA 10 7 11 1.7 .19 .27 .51 .48
GFZ 7 7 11 0.5 .12 .13 .23 .24
JPL 8 6 11 0.5 .13 .18 .41 .31
NGS 15 15 17 .47 .52 .91 1. 08
SIO 9 8 11 .74 .70 .44 .52
IGR 8 6 11 0.7 .12 .22 .37 .33

New IGS Products

Following recommendations of the 1996 AC Workshop (Neilan et al., 1996), the lOS 2-
day predicted orbit (lOP) and the lOS LODIUTl combinations were introduced on March
2197 (OPS Week 895). It took about a year to implement and the quality of both products
is still under evaluation.

IGS 2-Day Orbit Prediction Combination (IGP)

Work on the generation of the lOS Predicted orbit combinations (lOP) started in the
Summer of 1996. ACs were encouraged to take part, to develop and test their long arc
orbit prediction scheme. After several months of testing, it became apparent that a
combination of AC predictions would also be more reliable and in most cases more
precise than the best individual AC orbit predictions. Since March 2, 1997, daily AC orbit
predictions, available by 23 :OOUT, are combined and made available to the lOS DCs and
lOS CB no latter than 23:30UT, so that the lOP orbits are available for real time
applications for the following day. Extrapolated broadcast clocks are also included in lOP
orbit files. Within 22 hours after the end of the day, the lOP and the corresponding
broadcast (BRD) orbits are compared with the lOR orbits. The statistics and
transformation parameters are also included in the daily lOR report files to provide timely
quality evaluation of lOP and BRD orbits and the transformations to ITRF94. Figure 1
shows the daily WRMS, RMS and median RMS (with respect to lOR after a 7-parameter
Helmert transformation) for the broadcast orbits (BRD) and the predicted orbit
combinations (lOP) since OPS Week 895. With the exception of occasional high RMS
for certain satellites usually affecting both types of orbits, the lOP precision is much
better than the broadcast (~ 50 cm median RMS for lOP compared to ~ 250-300 cm for
BRD). All but one AC are contributing to lOP.

248
 brd
500

E
~ 400
C/)
~
0::: 300
c
o
;I
'iii 200
o
...
0..

:c... 100
o
o
895 897 899 901 903 905 907 909 911 913 915 917 919
GPS Weeks
IO Comb. WRMS A Comb. RMS - Comb. RMS Median I

igp
500

E
~ 400
(J)
~
0::: 300
c
o
;I
'iii 200
o
...
0..

:c... 100
o
o•• HB
895 897 899 901 903 905 907 909 911 913 915 917 919
GPS Weeks
D Comb.WRMS AComb.RMS -Comb.RMSMedian

Figure 1. Daily Mean Orbit Prediction Position RMS for BRD and IGP
(GPS weeks 895-919)

IGS LOD/UTi Combinations

The need for IGS LODIUTI combinations was discussed at the 1996 AC Workshop
(Ray, 1996b; Kouba, 1996). By the end of 1996, an approach proposed by Ray (1996a)
has been implemented with some modifications and improvements. It is based on a
weighted average of AC LOD solutions, which are calibrated with respect to 21 days of
Bulletin A non predicted UTI values. This calibrated LOD series is then integrated into
IGS UTI. Tests from November 1996 to March 1997 indicated that the 2-6 day Bulletin

249
A UTI predictions (used in IGR) with sigmas of about 600 I-lS could be significantly
reduced to about 170 I-lS using the LODIUTI combinations.
The IGS combined LOD and integrated UTI series were officially adopted for the IGS
Rapid (EOP(IGS)96P02) and Final (EOP(IGS)95P02) combinations on GPS Week 895
(MJD 50509) and GPS Week 894 (MJD 50502) respectively. Since GPS Week 908/day
3 (MJD 50603) for IGR and GPS Week 907 (MJD 50593) for IGS, the IGS combined
LOD and integrated UTI series are initiated with the five day old non predicted UTI
values of Bulletin A. This reduces the effects of higher noise levels so characteristic for
the latest Bulletin A non predicted UTI values usually based on small number of
observations. As in the IGS PM and PM rate combinations, the AC orbit weights are also
used for the LOD combinations. Well determined Bulletin A UTI values are usually
available at the time of the IGS Final combinations. Significant UTI improvements are
expected for IGR LODIUTI. As of the GPS Week 895/894 both EOP(IGS) series (95P02
and 96P02) include consistent PM, LOD and UTI combinations.

Figure 2 shows comparisons of two UTI series used for IGR combinations since March
2, 1997 with respect to Bulletin A's August 21, 1997 UTI update. The one labeled "old"
represents Bulletin A UTI values available at the time of the IGR combination including
Bulletin A predictions. The second one, labeled "new", is the IGR UTI series obtained
by integrating IGR combined LODs. The "new" LODIUTI combination strategy shows
noticeable improvement.

2000

1500

«
'5 . 1000

-..
Dl
2- 500
i
CD
U
C
'"
01
::I
i!! ....
.,. 0
CD
~
C -500
0

-1000

-1500 +----+----+----+-----+----+-----+-----.,I----t
50509 . 5 50530 . 5 50551.5 50572.5 50593 . 5 50614 . 5 50635 . 5 50656 . 5 50677 . 5
MJD

1-0 old : rms=485 ......... new : rms=@

Figure 2. Comparison Between Bulletin A Predictions ("Old Strategy"),

IGR Integrated UTI ("New strategy") and Bulletin A's 97Aug21 Update.

250
Acknowledgments

The authors gratefully acknowledge and thank all IGS AC colleagues for their continuous
assistance and cooperation without which the IGS product combinations and
enhancements would not be possible.

References

Kouba, J, 1996, IGS combination of GPS Earth orientation Parameters (EOP), IGS AC
Workshop Proceedings, Silver Spring, Md., March 19-21, 1996. pp. 33-43.
Kouba, J, Y Mireault and F. Lahaye, 1995, 1994 IGS Orbit/Clock Combination and
Evaluation, Appendix of the Analysis Coordinator Report, International GPS Service
for Geodynamics (IGS) 1994 Annual Report, pp. 70-94.
Kouba, J and Y Mireault, 1996, Analysis Coordinator Report, International GPS Service
for Geodynamics (IGS) 1995 Annual Report, pp. 45-76.
Kouba, J and Y Mireault, 1997, Analysis Coordinator Report, International GPS Service
for Geodynamics (IGS) 1996 Annual Report, pp. 55-100.
Neilan, R.E., P.A. Van Scoy and JF. Zumberge (ed.), 1996, Proceeding of the IGS AC
Workshop Proceedings, Silver Spring, Md., March 19-21, 1996.
Ray, J, 1996a, Measurements of Length-of-Day using the Global Positioning System,
Journal of Geophysical Research (JGR), Vol. 102, No. B9, 20141-20149.
Ray, J, 1996b, GPS Measurements of Length-of-Day: Comparisons with VLBI and
Consequences for UTI, IGS AC Workshop Proceedings, Silver Spring, Md., March 19-
21, 1996,pp.43-60.

251
 THE USE OF GPS FOR MONITORING
OF THE IONOSPHERIC DISTURBANCES

L.W.Baran
Institute of Geodesy, Olsztyn University
of Agriculture and Technology
Oczapowski St.1, 10-957 Olsztyn, Poland

1.1. Shagimuratov
West Department of the IZMIRAN
Pobeda St.41, 236017 Kaliningrad, Russia

ABSTRACT

Since January 1995 GPS observations performed by Polish IGS network stations
have been used for monitoring the ionosphere. During January-May 1995 we revealed
a number of violations of the ionospheric structure from regularity. The disturbances
resulted from magnetic storms with sudden onsets. The storms caused an increase of
ionospheric delay by a factor of 1.5-3.0. During disturbances gradients and TID's effects
are also increased. We found that such severe ionospheric conditions have an effect on the
determination of GPS equipment biases. The biases determined from code data for
disturbed days are different from those for quiet days. The differences can reach up to 0.3-
0.5 m in differential delay. It can be expected that the effects will be even more
appreciable during the next solar maximum.

INTRODUCTION

GPS observations performed at stations of the IGS network may be used for
monitoring of the ionosphere on both time and space scales. GPS observations have
provided the possibility of regular studies of ionospheric total electron content (TEe)
variations on a diurnal, seasonal and solar activity cycle basis. These investigations
provide the basis for developing, calibrating and improving ionospheric models for
different applications.
Precise GPS positioning requires the resolution of the double differenced phase
ambiguities. In such a case, the geometry-free linear combination is analyzed, which
contains the ionospheric information. Ionospheric disturbances cause severe conditions for

International Association of Geodesy Symposia, Vol. 118

Brunner (ed.), Advances in Positioning and Reference Frames
© Springer-Verlag Berlin Heidelberg 1998
GPS measurements and complicate the ambiguity resolution process. The large-scale
horizontal gradients in the ionosphere produce errors in precise relative GPS positioning
(Wanninger, 1995).
In this paper the storm-time effects in the ionosphere at mid latitudes during solar
minimum activity are presented. The analysis of the results of determination of GPS
equipment biases during disturbances are also discussed. For the study, GPS
measurements recorded at Polish stations of the IGS network: Lamkowko (53.9N, 20.7E)
and Borowiec (52. IN, l7.lE), have been used.

ALGORITHM USED FOR DETERMINATION OF TEC FROM GPS OBSERVABLES.

To determine the ionospheric TEC, a geometry-free linear combination is used. It

contains the ionospheric delay and the ambiguities for phase measurements and equipment
biases for code measurements.
The relationship between ionospheric delay and the TEC, and the difference
between dual-frequency code (P) and phase (<I» measurements may be written as (Baran et
aI., 1997):

AI> [m] =M·TEC/cos z + Ap + cp, (la)

L\<I> [m] =M·TEC/cos z + Act> + Ect>. (lb)

Here TEC is the vertical electron content, M is a scale factor, cP, Ect> are noise terms,
Ap and Act> are equipment biases (Act> contains the phase ambiguity), z is the zenith angle of
the ray at the sub-ionospheric point. In the algorithm, the following ionospheric model of
the diurnal variation ofTEC is used (Georgiadiou, 1994):

6
TEC = Lan cos (ns) + bn sin (ns) + az L\q> + as L\q> s, s = 1t(LT-14)/12. (2)
n=O

Here LT is the local solar time and L\q> is the latitudinal difference between the
coordinates of the receiver and the sub-ionospheric point. The TEC values are estimated
simultaneously with equipment biases over a 24 hour period using a least squares
procedure. We have employed such an algorithm where the code and phase measurements
are treated as independent of one another. The TEC obtained from both observables should
coincide. The difference is less than 0.1 - 0.2 of the TEC unit. The following estimates are
presented in the form of diurnal variations of TEC.

STORM-TIME VARIATIONS OF TEC

For the period of January - May 1995, some moderate but sudden ionospheric
disturbances occurred. For all storms, the geomagnetic conditions were nearly similar. In

253
Fig.1 storm-time variations in the ionosphere for the period 28 - 31 January, 1995 are
presented.
TECxE16e11rn' TECxE16811m' TECxE16e11rn' TECxE16811m 2
20 20 20 20

16 01-28-95 Kp = 6 16 16 01~ Kp=36 16 01-31-96 Kp = 31

12 12 12 12

8 8 8

4 4 4 4

0 0
0 6 12 18 24UT 0 6 12 18 24UT 6 12 18 24 UT 6 12 18 24 UT
(IoF2)'/3 (MHz)' (foF2)'/3 (MHz)' (IoF21' /3 (MHz)'
20 20 20

16 01-28-96 Kp = 8 16 16 01-30-95 Kp = 36 16 01-31-95 Kp = 31

12 12 12 12

8 8 8 8

4 4 4 4

6 12 18 24UT 6 12 18 24UT 6 12 18 24UT 6 12 18 24UT

Fig.1: Storm-time effects in the ionosphere. Top panel: variations of TEe from GPS
measurements (solid), TEe of the IRI model (dotted), the monthly mean of GPS TEe
(dashed). Bottom panel: variations of (foF2i (solid), monthly mean of (foF2i
at Warsaw (dashed).

On the top panel, the day by day ionosphere TEe vanatlOns are shown. For
comparison, the average daily GPS TEe and IRI model derived TEe are shown. Model
TEe was estimated using the International Reference Ionosphere (IRI) model with
correction by the monthly mean frequency foF2 from ionosonde measurements near
Warsaw. For winter the IRI model typically overestimates values of TEe compared with
GPS estimates.
The day that preceded the storm (28 January) was very quiet, with a summary
index of magnetic disturbances (Kp) was equal to 6. In the first day of the storm, on 29
January (Kp=33) TEe had increased by a factor of 1.5 - 2.6 relative to the previous,
undisturbed day. In the diurnal variation two humps appear with a pronounced maximum
after noon. During the following days, the negative phase of the storm began. The
ionosphere TEe dropped to a level lower than the undisturbed one. The negative phase of
the storm is less marked in the diurnal variation.
The ionospheric TEe values show a good correlation with the electron density of
the F2 ionospheric layer. The maximum density (NmF2) may be obtained from ionosonde
measurements: NmF2 = kx(foF2)2, where k is a constant. On the bottom panel the
variation of (foF2)2 at Warsaw is plotted. For comparison, the monthly mean is also given.
The increasing value (foF2i, the same as for TEe, occurred on 29 January. Here also are
two pronounced humps. The abnormal variation of foF2 may be seen simultaneously
at Kaliningrad (54.7 N, 20.4 E) and Warsaw (52.1 N, 21.1 E) (Fig.2).

254
 TECxE16 el/m2 (foF2)2 (MHz)2
20 20

16 16

12 12

\
8 8 \

4 4
01-29-95 01-29-95
O+---~----~----~----~ O+-----.---~----_r----~·
o 6 12 18 24 UT 0 6 12 18 24 UT

Fig. 2: The latitude behaviour of TEe and (foF2)2 for the disturbed day of 29 January
1995. The TEe at Lamkowko (dashed) and Borowiec (solid). The foF2 at Kaliningrad
(dashed) and Warsaw (solid).

For the more southerly Warsaw station, the first hump has a pronounced character.
The same as for foF2, the ionosphere TEe value is higher at Borowiec, located to the
south of Lamkowko. Some time delay may be seen between Warsaw and Kalinigrad
events (Fig. 2). The delay amounts to about 30 min. On assuming that the disturbance
moves from north to south, we may estimate the speed of the disturbance. It is about 700
m1sec.
The differences of TEe on quiet and disturbed days can be revealed by studying
the behaviour of the electron content for individual satellite passes. In Fig. 3 the vertical
TEe variation for satellite passes of PRN 5 and 2, on January 28 and 29, 1995
respectively, are given.
LAMA, sat. 2 lAMA, sat. 5
3,5 . , . . . - - - ----=....:::..::....::...:.=..:::...=---- - - - - , - 4,0 3,5 .-----------'-----~----r 4 , 0

I
'C 'C
2,0 ~ ~
N-;::. 2,5 ~ N 25
8 § ,
2.0

~
i
.S!. ~ .S!.
.... 0,0 ;,
0,0 ~
~ i :;
~
•
()
1,5 1,5
i
I!! -2.0! ~ -2,02:
8
0,5
10 12 14 16 14 16 18 20
II 18,6 45,5 68,0 59,4 34,5 10,3 II
19,1 44,4 71,1 91 ,2 53,9 27,7 3,3
lat 46,3 SO ,9 52,9 54,2 55,3 55.7 101 49,2 52,5 53,6 53,7 53,0 SO,7 41.9
long 21,8 21 .7 22,1 23,5 27,2 39,2 long 11,4 16,1 19,0 21 ,5 24,0 27 ,6 35,9

Fig. 3: Variation of vertical TEe and ionospheric Doppler shifts during satellite passes of
PRN # 5 and PRN # 2 on quiet and disturbed days. The TEe of 28 January 1995 c .... ·.. ·),
TEe of 29 January 1995 (---). Doppler shift of 28 January 1995 ("0- ..0- ..0-.), and
29 January 1995( ~ ).

255
 For PRN 5 the absolute TEe is determined with code measurements corrected for
equipment biases. For PRN 2 we used phase measurements corrected by the initial phase
ambiguities. The code data includes multipath effects. In Fig.3 the ionospheric Doppler
shifts obtained from phase data are also presented.
Storm-time Doppler shifts indicate increasing TID's effects. Ionospheric Doppler
shifts characterize the gradients of TEe along the satellite trajectory. The large-scale
gradients on a disturbed day are distinguished from quiet ones. In disturbed periods, the
Doppler shift may even be of opposite sign compared to that of the quiet time (see Fig. 3).
The horizontal large-scale gradients affect the determination of instrumental biases and
ambiguity resolution. The pronounced difference in Doppler shifts for two days (Fig.3)
seems to be caused by the variations of the biases during disturbances.

THE NEGATIVE PHASE OF STORM

On the second day of the storm, the ionosphere again reaches almost its ordinary
state. In the diurnal variation the negative phase is poorly distinguished.

20
20 TEC-E1 8~; 20 20
!~:\
~'!~r
16 16 16 16

12 12 12 12
:...).",\:"\

I,' \',
: I ""'-
6 8 6 8

4 //' " 4 4
.........
8 12 18 24lIT 6 12 18 24 lIT 6 12 18 24UT

TEC- E18 oIIm' lECxE18ollm2

20 20 20 20

16 16 16

12 12 12
-, ,'..~":" ''':'

,,,:, . ~..,.
6 \ .. 8 ' ~.
\
\
, ., I .:
\

.... ~ ... .......{ ...

o ~--~~--~--~ o~--~~--~--~
6 12 18 24 UT 0 6 12 16 24 UT 6 12 18 24 lIT 0 6 12 18 24lIT

TECxe'8.vm' TEC-E18.vm'
20 20 20 20

16 16 16 ()[email protected] Kp - 39 16 O~·95 Kp - 3o\

12 12 12 12

8 8 8

4 4 4

0 ---6.---~
0 ..... 12---1~8--2~
4-lIT 0""0---6.---~
12---1~8--2~
4 ·UT 00 6 12 18 24lIT 00 8 12 18 24 UT

Fig.4: Storm-time variations in the ionosphere on February, March and May 1995. The
GPS TEe at Larnkowko (solid), the (foF2)2 at Warsaw (dotted) and the monthly mean of
TEe (dashed).

256
 For the analysis of behaviour of the TEC, we used the TEC variations on individual
satellites. In that case, the code data were corrected for equipment biases obtained during
quiet days. The procedure enabled the determination of the changes in TEC more
accurately. The analyses show that for most passes the total electron content on the second
day is lower than that for the quiet day. The decrease was about 20 - 30 % in comparison
to the undisturbed level. For other satellite passes at the same time the changes are not
marked relative to the quiet day. So the response of ionosphere to storm recovery stage is
ambiguous.
The response of the ionospheric electron content during storms at other periods
may be seen in Fig. 4. The general behaviour ofTEC for storms with a sudden onset is still
there. Always there is a pronounced increase in the TEC on the first days and recovery to
previous level following one-two days.

VARIATIONS OF EQUIPMENT BIAS V ALVES

Differential biases (Ap) have been computed day after day simultaneously with the
ionospheric TEC from full 24 hours of measurements. Results of the determination of
satellite/receiver biases for some satellites over the period of 28-31 January 1995 are
presented in Table 1. The data refer to the Lamkowko receiver.

Table 1· Satellite/receiver biases

Quiet days, Ap m) Disturbed days, Ap (m)
PRN 25.01 26.01 28.01 Mean RMS 29.01 30.01 31.01
12 0.47 0.48 0.37 0.44 0.04 0.44 0.33 0.53
14 -0.59 -0.49 -0.61 -0.56 0.05 -0.75 -0.79 -0.62
17 -0.39 -0.44 -0.49 -0.44 0.03 -0.59 -0.48 -0.44
18 -1.05 -1.03 -1.05 -1.04 0.01 -1.16 -1.27 -1.13
25 -1.54 -1.55 -1.50 -1.53 0.02 -1.71 -1.70 -1.58
29 -1.49 -1.46 -1.45 -1.46 0.01 -1.57 -1.69 -1.55
31 -1.43 -1,36 -1.32 -1.37 0.04 -1.83 -1.61 -1.39

LKp 11 6 6 33 36 31
Ap/L -0.84 -0.82 -0.87 -1.01 -0.91 -0.79
Ap/B -0.25 -0.28 -0.37 -0.50 -0.37 -0.39

The biases are expressed in units of meters. In Table 1 there are listed the biases averaged
over quiet days (Mean) and RMS respectively. In the second part of Table 1, the summary
index of magnetic disturbances (LKp) and the biases averaged for all satellites observed
over 24 hours (Ap) are given. It can be seen that for the highly ionosphere disturbed day of
29 January, 1995, the Ap are different from those for the quiet days. The difference of the
mean biases amounts to about 0.2 m (2TEC). For individual satellites the differenc~ may
reach 0.3 - 0.5m. In the last line on the same period the mean biases for Borowiec (Ap/B)
are shown. The behaviour of Ap day by day for Borowiec is similar to that for Lamkowko

257
(Ap/L). The equipment biases, as for the ionospheric TEC, are a maximum on the first day
of the storm. In Table 2, the results for variation of biases are summarised for all
disturbances over the January-May 1995 period.

T able 2 Companson 0 fb·lases or qUIet and d·lSturb ed days

Quiet days Disturbed days
January -0.82 -0.87 -1.01 -0.91 -0.79
February -0.65 -0.65 -0.60 -0.87 -0.85
March -0.86 -0.82 -1.11 -1.34 -1.28
April -1.17 -1.12 -1.34 -1.23 -1.11
May -1.08 -1.11 -1.11 -1.34 -1.10

Here the biases are presented for two days before the disturbances and three days after the
onset of the storm. The behaviour of biases on all storm-time days is similar.
The large - scale horizontal gradients substantially influence the determination of
the equipment biases as well as the phase ambiguities. In Fig. 3 in the records of Doppler
shifts it is seen how the gradients may be markedly distinguished between quiet and
disturbed conditions. In addition to this, during storm-time the TID's effects are increased.
In turn, the errors in determining biases are increased. We note that the TEC obtained
independently from code and phase data are similar.

CONCLUSIONS

During ionospheric disturbances at solar mmlmum, at mid latitudes, the

ionospheric electron content may be increased by a factor of 1.5 - 3.0 relative to quiet
conditions. During storm-time, the abnormal large-medium-scale horizontal gradients in
the ionosphere appear and the TID's effects are also increased. Storm ionospheric
conditions cause the equipment biases determined from code measurements on disturbed
days to be different from those obtained for quiet periods. Variations in the ionosphere
may give rise to errors in the ambiguity resolution when using the geometry-free
combination. The ionospheric effects will increase in the next ten years. Thus studies of
storm-time effects that cause errors in precise GPS positioning must be continued.

REFERENCES

Baran L.W., I.I.Shagimuratov, N.J.Tepenitzina, (1997). The use of GPS for ionospheric
studies. Artificial Satellites. Journal of Planetary Geodesy. V. 32, N 1, pp 49-60.
Georgiadiou J., (1994). Modelling the Ionosphere for an Active Control Networks of
GPS Stations. In LGP Series, Publications of the Delft Geodetic Computing Centre,
N07.
Wanninger L. (1995). Monitoring ionospheric disturbances using the IGS network. In
Proceedings of the 1995 IGS Workshop, May 1995, Potsdam, Germany.

258
 AN INTEGRATED GPS MONITORING SYSTEM FOR SITE
INVESTIGATION OF NUCLEAR WASTE DISPOSAL

Ruizhi Chen and Juhani Kakkuri

The Finnish Geodetic Institute,
Geodeetinrinne 2 ,
02430, MASALA,
FINLAND

Abstract

An integrated GPS monitoring system has been in operation in the about 1O-km2 study
area at Olkiluoto, Finland since October, 1994. The system includes a permanent GPS
station and a local GPS monitoring network. The permanent station is used to investigate
regional crustal deformations in Finnish territory, while the local monitoring network is
used to monitor relative deformations within the study area. The analysis of 2.8 years of
continuous GPS tracking data shows that the horizontal components of the regional
crustal deformations are about 2 mm/yr, while the vertical component is about 8 mm/yr.
The local GPS monitoring network has been measured six times at half-yearly intervals.
The horizontal components of the local movements obtained from the repeated
measurements are less than 1.0 mm/yr. The GPS result indicates that the investigation
area at Olkiluoto is rather stable.

1. Introduction

According to a decision by the Finnish government, the Finnish Posiva Oy is responsible for
preparations for the permanent disposal of spent nuclear fuel in Finland. A special
programme including studies of geology and geophysics has been carried out since 1983.
According to the generic studies, three areas, namely Olkiluoto on the west coast, Kivetty
in Central Finland and Romuvaara in northeast Finland, were selected in 1992 for detailed
investigation. The detailed studies were begun in 1993, and will be continued until the
year 2000. One of the areas will then be selected as the location for permanent disposal of
the spent fuel. Complementary investigations will be continued in the area selected until
the year 20 I o. In co-operation with other geophysical studies, the Finnish Geodetic In-
stitute is now taking responsibility for operating a GPS monitoring system to monitor the
 @ IGS Sta tion
• Permanent
StaHons
re'I Investigation
~ areas

• PeflNlnent:$tation
• l.oc8I&.tiona
, / FradL".
.. BafthoI••

Fig. 2. The local GPS monitoring network at

Fig. 1. The Finnish permanent GPS array and Olkiluoto. Locations of the fractures are
the investigation areas for nuclear obtained from Anttila et al. 1992.
waste disposal in Finland.

crustal deformations at the study areas (Chen and Kakkuri, 1993; 1994; 1996; 1997). The
system includes:
• three permanent GPS stations which operate continuously, and
• three high-precision local GPS monitoring networks at the study areas.
The permanent stations belong to the Finnish permanent GPS array which consists of
12 stations as shown in Fig. 1. The network is connected to the IGS network through
Metsahovi. Each station is installed with an Ashtech z-xn receiver and a choke-ring
antenna.
The first local GPS monitoring network was established at the Olkiluoto study area in
October, 1994 as shown in Fig. 2, while the other two networks for the Kivetty and Romu-
vaara study areas were established in September, 1995. Each network includes a
permanent station such that the local GPS monitoring networks are in fact connected to the
IGS network.
The reinforced pillars need about half a year to settle completely, therefore the first
measurement was performed in May, 1995 for Olkiluoto, while those for the other two
investigation areas were performed in May, 1996. For this reason, we have observations
for only about one year from the networks at Kivetty and Romuvaara. This is too short a
time to make any deformation analysis for these two areas, therefore we will show only the
results from the Olkiluoto investigation area. The results are based on 2.8 years of
continuous GPS observations for the permanent station, and six repeated measurements of
the local monitoring network.

2. Regional deformations

Regional deformations are relative deformations between the permanent GPS stations and
the IGS station at Metsahovi. GPS data collected from the permanent stations are
processed with the Bemese GPS software by a standard procedure. A set of batch

260
programs has been developed for the automatic processing of daily observations. Only a
single DOS command, as simple as "C:\AUTOSESS\ATP", is required to process all the
data stored on a CD-ROM. The system can identify the days of OPS data on the CD-ROM
and create corresponding preprocessing batch files. The daily normal equation is stored for
the combination adjustments which are performed for every five days. The OPS data are
processed with the lOS precise orbits. The lOS station at Metsahovi is used as the fixed
station in the daily solutions. Fig. 3 shows the velocity components obtained from 2.5
years of continuous OPS observations.
The horizontal components indicate that Olkiluoto is moving by 2 mmlyr in a
southeasterly direction relative to the lOS station at Metsahovi. The compression direction
agrees very well with global tectonic models (DeMets et ai. 1990, 1994). The borehole

mm Var iaII ons 0 fB ase II ne lengl h between Metsahov and Olk luoto

-..
20
•• ....
. ... .. .
Vs = -1 .8 ± 0.4 mm/yr
10
A . . -4la"-.
,- .......
--
~~,
.~

o
•
.... :" ,~
~
-~-
."A~
-10
•
-20
01995.0 0,5 Year 1.5 2 2.5 3

.. ,.4......". ...........
• .. •
Vn = -1.6 ± 0.3 mm/yr

.
~ #1 ....,. .- .... ••
~ ... ~ • " ~5}~ ~
• •
•
~ .;
... ... .~

•
°1995.0 D,S 1 Year 1,5 2 2,5 3

mm Variations of the Easl Component Between Metsahovi and Olklluolo

20 ~--------------------------------------------------------~

..~~~~~~~~~.~
10 +-----------------~--~--~------------------------------~

o+---~~~~~~~--~~
-10 +-------~------------------------------------------------~

~O~------~----------~--~----------------------~--~----~
Ve =1.2 ± 0.4 mm/yr
01995.0 0,5 Year 1,5 2 2,5 3

mm Variation of the Height Component Between Metsahovi and Olkiluoto

30 ~~~--~~~---=~--~~~------~--~~~=-~--~~----~
Vh = 7.8 ± 0.5 mml r
20 +-----~----~~~~~~---------------~~--------~~--~~

1~t=====~iS~~~~~~~;;~~~~~~~~~~~::~~~==~
-10 +-----~~--------~
~..=·~-------·~--------------------------~
-20 ~--------------------------------------------------------~
o 1995.0 0,5 Year 1,5 2 2,5 3

Fig. 3. Velocity components of the GPS permanent station at Olkiluoto. The result is obtained from
five-day solutions. The zero-epoch in the figure is 1995.0.

261
stress measurements performed in the investigation area indicate that the maximum stress
orientation is in the east-west direction (Anttila, et al. 1992). The discrepancy can be
explained in such a manner that the GPS result represents regional stress orientation for
the zone between Metsahovi and Olkiluoto, while the borehole measurements represent
the local stress orientation at Olkiluoto.

.- .
mm GPS6 (East) 1,5mm GPS6 (North)
, -_ _ _--"'::.....::..::....>:...:..::..:..:;c.:.L-_ _ _- . ,
1,5

0,5 •- ...
Ve = -0.3 ± 0.2 mm/yr
0,5 +--_= •
Vn = -0.4 ± 0.2 mm/yr
_- - - - - -- - - - - - i

..- • •
- ~

+-----------=-~..._t_I
•
-0,5 -0,5

·1 ,5 -1,5 .L-_ _ _ _ _ __ _ _ _ _.-J

O)'=' O)'=' O)'¢

!..
O)'¢ ~ ~ O)'='
~~. 0 0.' '?-~ 0 0.' ~~. 0 0.' ~~.

Fig. 4. Result obtained from five measurements at the local GPS network at Olkiluoto. The measurements are
repeated at a half-yearly interval beginning in May, 1995. The velocities are relative to the permanent
station GPSl, which is considered as the fixed station, as shown in Fig. 2. Similar pictures can be
found for other stations.

For the vertical component, the uplift rate obtained from GPS is 7.8 mm/yr, which is
larger than the uplift value of 4.1 mm/yr obtained from precise levelling (Kiiiiriiinen,
1966). By taking the uplift of the geoid into account, the uplift rate of the ellipsoidal height
obtained from precise levelling becomes 4.2 mm/yr as the difference of the uplift of the
geoid between Metsahovi and Olkiluoto is about 0.1 mm/yr (Kakkuri, 1997; Kakkuri and
Chen, 1997). Although the GPS result is rather scattered, it indicates the correct trend,
therefore a more reliable result can be obtained with GPS over a longer observation time
span.

3. Local deformations

The permanent stations in the investigation areas are used as reference stations for the
corresponding local monitoring network. Therefore, local movements are referred to the
corresponding permanent station.
The local GPS monitoring network at Olkiluoto has been observed for six times since
1995 at half-yearly intervals. The stations were occupied for about 10 hours with Ashtech
Z-XII receivers and choke-ring antennae. The GPS data are analyzed carefully with a
processing procedure which has been studied comprehensively in our earlier study (Chen
and Kakkuri, 1993). In an optimal observation environment, we are able to achieve an
accuracy of less than 1 mm for the horizontal components (Chen and Kakkuri, 1993). Fig.
4 shows the result obtained from six repeated measurements for station GPS 6.
The rotation errors of the network should not be significant because
• the IGS precise orbits are used in processing the GPS data, and

262
 KR3
... .... GPS1
KRl
V
"'KR2

--=>-",.... KR7
I~

~ O.Smmlyr
• Permanent Station
• Local Stations
... Boreholes
Fractures

Fig. 5. Local movements in the surroundings of the nuclear power stations at Olkiluoto. The
error ellipses of the displacement vectors are at confidence level of 95%.

• the fixed station is a permanent station, therefore the accuracy of the fixed
coordinates is high.
It is therefore reasonable to depict the relative deformations with displacement vectors
though it is theoretically not strict. The displacement vectors give a more direct view of
the deformation pattern than the strain parameters.
Fig. 5 shows the relative velocity vectors at the stations of the local monitoring
network. According to geological and geophysical investigations (Anttila et aI., 1992), the
investigation area is one of the most stable areas in Finland. It is therefore not surprising
that the relative velocities obtained inside the about 1O-km2 investigation area are less than
1.0 mmlyr.
We can conclude conservatively that the relative movements among the GPS stations in
the investigation area are less than 1.0 mmlyr. As the movements are so small, a longer
period is required to detect such small movements.

4. Conclusions

An integrated GPS monitoring system has been in operation since October, 1994 for site
investigation of nuclear waste disposal in Finland. Based on the analysis of 2.8 years of
continuous GPS observations, regional deformations for the investigation area were
obtained. It has been shown that the permanent station in the Olkiluoto investigation area
is moving in a southeasterly direction by 2 mmlyr relative to the IGS station at Metsahovi.

263
This indicates that maximum compression occurs in the southeast direction. It agrees very
well with the global tectonic model. For the vertical component, the result obtained from
GPS measurements is 7.8 mm/yr, while that from precise levelling is 4.2 mm/yr. The GPS
result is scattered, therefore a longer time span is required to obtained a more reliable
result.
Results obtained from six measurements of the local monitoring network show that the
local movements for the stations inside the investigation area are generally less than 0.5
mm/yr. The GPS result indicates that the investigation area is rather stable. This agrees
with the result from geological and geophysical investigations.

s. References
Anttila P., S. Paulamaki, A. Lindberg, M. Paananen, T. Koistinen, K. Front and P. Pitkanen
(1992). Geology of the Olkiluoto Areas, Summary Report. Report YJT-92-28. Voimayhtibiden
Ydinjatetoimikunta (Nuclear Waste Commission of Finnish Power Companies), Helsinki.
Chen Rand J. Kakkuri (1993). Capability of GPS Technique for Local Crustal Deformation
Detection. Proceedings of the Eighth International Symposium on Recent Crustal Movements.
Pp. 209-213. Kobe, Japan, Dec. 6 -11, 1993.
Chen Rand J. Kakkuri (1994). Feasibility Study and Technical Proposal for Long-term
Observations of Bedrock Stability with GPS. Report YJT-94-02. Nuclear Waste Commission
of Finnish Power Companies. Helsinki.
Chen R. and Kakkuri (1996). GPS Operations at Olkiluoto, Kivetty and Romuvaara for 1995.
Project report, PATU -96-07e, POSIVA OY, 1996, Helsinki.
Chen Rand Kakkuri (1997). GPS Operations at Olkiluoto, Kivetty and Romuvaara for 1996.
Project report, PATU -96-65e, POSIV A OY, 1996, Helsinki.
DeMets C., RG. Gordon, D.F. Argus, S. Stein, 1990. Current plate motions. Geophys. 1. Int., 101,
425-478.
DeMets C., R.G. Gordon, D.F. Argus, S. Stein, 1994. Effect of recent revisions to the
geomagnetic reversal time scale on estimates of current plate motions. Geophys. Res. Lett., 21,
2191-2194.
Kakkuri J, 1997. Postglacial deformation of the Fennoscandian crust. Geophysica 33, pp. 99-109.
Kakkuri J. and R . Chen, 1997. Postglacial deformation of the Fennoscandian crust. Latest results
from the geodetic measurements. Proceedings of the IAG Regional Symposium. Deformations
and Crustal Movement Investigations using Geodetic Techniques. Pp. 8-15. 31 Aug. - 05 Sept.
1996, Szekesfehervar, Hungury. 1997.
Kaariainen E. 1966. The second levelling of Finland in 1935-1955. Publication of the Finnish
Geodetic Institute. Nr. 61. Helsinki.

264
 DETERMINATION OF STOCHASTIC MODELS OF
GPS BASELINES IN GPS NETWORK ADJUSTMENT

Xiaoli Ding
Department of Land Surveying and Geo-Informatics
Hong Kong Polytechnic University, Kowloon, Hong Kong

Michael Stewart
School of Surveying and Land Information, Curtin University of Technology
GPO Box U1987, Perth, WA 6001, Australia

Jason Chao
Department of Land Surveying and Geo-Informatics
Hong Kong Polytechnic University, Kowloon, Hong Kong

Abstract

It is well known that the variance covariance matrices of GPS baselines (vectors) as obtained
from the processing of GPS baseline observations are in general too optimistic. Therefore, it
is a common practice, in the adjustment of GPS networks, to scale up the variance covariance
matrices by for example multiplying the matrices with a constant factor. Though the method
can always manage to get an adjustment to pass the necessary statistical tests, the method is
not rigorous and often arbitrary.

This paper uses the newly completed Western Australian state wide GPS network
STATEFIX as an example to look at the effects of the different ways of determining the scale
factors for the variance covariance matrices of GPS baselines. The method of variance
component estimation is used to assist the analysis. It is found that the adjustment results can
vary significantly when different methods are used to determine the scale factors. For most
GPS networks, it is recommended that a scale factor be estimated for each independent GPS
observation session using the technique of variance component estimation.

1. Introduction

The variance covariance matrices of GPS baselines (vectors) can be determined either
using certain functional models formulated according to the error properties of GPS
surveys (e.g., Ananga et al., 1994) or directly from GPS baseline processing.
 In least squares adjustment of GPS networks, when using the variance covariance
matrices of baselines obtained from baseline processing, it is a common practice to scale
up the matrices using certain scale factors as they are usually too optimistic. This paper
looks at, using the data from the newly completed Western Australian state wide GPS
network STATEFIX as an example, the effects of different ways of determining the scale
factors when using the method of variance component estimation.

2. The STATEFIX Network and the Initial Adjustment

The STATEFIX network is a GPS network that densifies the ANN (Australian Fiducial
Network) in the State of Western Australia. The STATEFIX network consists of over 200
observed GPS baselines and 80 stations, with 17 of these stations being the ANN stations
(Figure la). The baseline lengths range from as short as 45km to distances of longer than
450km. The mean baseline length is about 2ookm.

2.1 Baseline Processing

The STATEFIX network was divided into 9 interconnected individual cells (sub-
networks) (Figure 1b). Each cell contains a number of GPS baselines observed in different
GPS observation sessions.

Cells 1, 2 and 3 were processed by dividing the observation time spans into two
independent sessions and computing two solutions for each baseline. Consequently, two
baseline vectors were generated for each baseline in the three cells. This was done mainly
for the purpose of quality control. The results thus obtained for the baselines are very
close to those obtained by processing all the data together. Besides, some baselines in the
network were observed twice as they constituted the cell boundaries.

All data in a session were processed together. Therefore, each session produced one or
more baselines depending on the number of receivers used for that session. 232 baseline
vectors in total and their variance covariance matrices were produced from the processing
of 141 independent sessions (see Table 1). The data processing strategies and further
details about the STATEFIX network can be found in Stewart et al. (1997a; 1997b).

2.2 Cell Adjustment

All the baseline vectors within a cell were combined in a least squares adjustment. A
minimum constraint datum was used in the adjustment by fixing one of the points in the
cell. The adjustment results for each cell were checked for any problems in baseline
processing. The variance covariance (VCV) matrices of all the baseline vectors in a cell
was scaled up, in the cell adjustment, using a scale factor to make the estimated variance
factor to be close to 1. The scale factors used in the adjustment are listed in Table 1. The
GEOLAB (version 2.4) software was used for all the adjustment computations.

266
 Table 1: Scale factor for the variance covariance matrices
Determined from cell adjustment

Cell Sessions Baselines Scale Factor for Cell Sessions Baselines Scale Factor for
VCVMatrix VCVMatrix
1 24 68 45.6 6 8 18 168.7
2 18 50 68.3 7 13 26 150.1
3 34 66 46.6 8 10 20 83.9
4 10 26 110.7 9 10 28 192.5
5 14 30 83.7

• STATEFIX site
' ANN site
r BR'-'--_DIV1I~ ,.......,.

lOOOkm

Figure la The STATEFIX Network Figure Ib Diagram showing division

of STATEFIX Network into 9 cells

2.3 Full Network Adjustment

The baseline vectors and their variance covariance matrices from all the 9 cells were
combined in the final full network adjustment. The coordinates of all the 17 ANN points,
except the height of point DEAK due to the problems with its height value, were used as
controls in the adjustment. The published uncertainty values of 3 cm in horizontal

267
directions and 5 cm in height (at 95% confidence level) were used for all the ANN
coordinates.

The estimated variance factor from the combined adjustment passed the standard Chi-
square test (at 95% confidence level), though all the estimated scale factors listed in Table
1 were scaled further up by 10% to make the variance factor almost exactly equal 1.

3. Variance Component Estimation

Some further tests were carried out to assess the appropriateness of the stochastic models
used in the above full adjustment and to examine the effects of using different scale factors
for the variance covariance matrices. The methods for variance component estimation
(e.g., Grafarend, et al., 1980; Welsch, 1981; Sabin et al., 1992) can be used for
determining the scale factors for the variance covariance matrices. Since the adjustment
computation was carried out using the GEOLAB software and the variance component
estimation was done based on the output from the software, the following approximate
method was used for estimating the variance components,

(1)

where S~g is the estimated variance component for the gth group of baselines; and
vg , Pg , Qv g ' and i g are the least squares residuals of the baseline vectors from the last
epoch of adjustment, the a priori weight matrix, the cofactor matrix of the residuals and
the redundancy in the corresponding group. A few iterations were usually required for
computation.

3.1 Grouping Baseline Vectors by Cells

All the baseline vectors in a cell were first considered to belong to the same observation
group. The ANN coordinates were also considered as a group of observations in the
adjustment and variance component estimation. There were therefore totally ten
observations groups formed, one for each of the 9 cells in the network and one for the
ANN coordinates. A variance component was estimated for each of the ten groups. The
estimated values are given in Table 2.

It can be seen from the results that the estimated variance components (scale factors)
for the the GPS baselines are fairly close to those obtained from the individual cell
adjustments. In fact the values would be even closer if the scale factors determined from
the cell adjustments have not been scaled up by 10%. This indicates that the scale factors
used in the full network adjustment discussed in Section 2.3 should not have been
increased. Instead, the uncertainty values of the ANN coordinates should be scaled up.

268
 When the new scale factors were used in the adjustment, the estimated variance factor
from the adjustment still remained to be equal to 1. The estimated coordinates and their
variance covariance matrix differ to a certain extent from the results obtained in the initial
adjustment. The changes in the coordinates are summarised in Table 3. The sizes of all the
absolute error ellipses for the horizontal point positions and of the error bars for the height
positions increased noticeably due to the scale factor of 1.6 for the variance covariance matrix
of the ANN coordinates.

Table 2: The estimated variance components when observations are grouped by cells

Observation Original Scale Factor Estimated Variance New Scale Factor

Group for VCV Matrix Component for VCV Matrix
CellI 50.6 0.86 43.5
Cell 2 75.9 0.84 63.8
Cell 3 51.8 1.00 51.8
Cell 4 123.0 0.77 94.7
Cell 5 93.2 1.60 149.1
Cell 6 187.4 1.10 206.1
Cell 7 166.8 1.00 166.8
Cell 8 93.2 0.86 80.2
Cell 9 213.9 1.00 213.9
ANN
Coordinates 1.0 1.60 1.6

3.2 Grouping Baseline Vectors by Observation Sessions

The baseline vectors were next grouped by the observation sessions for the purpose of
variance component estimation. A variance component was estimated for each observation
session on the basis of the results from Section 2.3. The 141 variance components for the
GPS baselines vary from 0.01 to 9.34 with a mean value of 1.039 and a standard deviation of
± 1.328.
The estimated coordinates from the adjustment using the new variance factors also
changed fairly significantly from those obtained in the initial adjustment as seen in Table 3.

Table 3: Chan~es in coordinates when new variance components were used (unit = m)

Grouping baselines by cells Grouping baselines by observation sessions

Spatial N/S EIW UID Spatial N/S EIW UID
Min 0.00034 0.00003 0.00001 0.00000 0.00080 0.00000 0.00000 0.00025
Max 0.00500 0.00105 0.00067 0.00488 0.02049 0.00312 0.00437 0.02032
RMS 0.00013 0.00005 0.00003 0.00011 0.00092 0.00012 0.00016 0.00090

Grouping GPS baselines by observation sessions allows in general a better determination

of the stochastic models of the GPS baselines than grouping the baselines into larger sub-
networks such as the network cells in the STATEFIX. However small observation groups

269
may also mean lower reliability in the estimated variance components as the estimation would
be easily affected by the errors in the observations. This may be responsible for some of the
very small and very large scale factors obtained. Besides, as a general condition for reliable
variance component estimation, the network should have sufficient and well distributed
redundancy.

The histogram diagram of the standardised residuals of all the GPS baselines and the ANN
coordinates, generated from the adjustment using the new variance components, looks to
follow the normal distribution more closely than the residuals obtained from the initial
adjustment discussed in Section 2.3.

4. Conclusions

The variance covariance matrices of GPS baselines as obtained from GPS baseline processing
are usually too optimistic. Therefore the variance covariance matrices need to be scaled up
using certain scale factors. This paper has looked at, using the STATEFIX GPS network as
an example, the method of determining the scale factors using the technique of variance
component estimation. The effects of different ways of grouping the observed GPS baseline
vectors in the estimation of the variance components have also been demonstrated using the
example.

It has been seen that the method of dividing a large network into smaller sub-networks can
lead to reasonably good estimation of the scale factors. The method of grouping observations
according to the observation sessions in the estimation of the variance components is
preferred though care should be taken as the estimation is more easily affected by observation
errors.

Acknowledgements
The work is partly supported by a research grant from the Hong Kong Polytechnic University
(Grant No. 3511614).

References
Ananga, N., Coleman, R. and Rizos, C. (1994) Variance-covariance estimation of GPS
networks. Bull. Geod. 68:77-87.
Grafarend, E., Kleusberg, A. and Schaffrin, B. (1980) An introduction to the variance
covariance components estimation of Helmert type, zfv 105:129-137.
Sahin, M., Cross, P.A. and Sellers, P.C. (1992) Variance component estimation applied to
satellite laser ranging. Bull. Geod. 66:284-295.
Stewart, M.P., Ding, X., Tsakiri, M. and Featherstone, W.E .. (1997a) 1996 STATEFIX
project, final report. Report Submitted to West Australian Department of Land
Administration. School of Surveying and Land Information. Curtin University of
Technology, 47pp.
Stewart, M.P., Houghton, H. and Ding, X. (1997b) The STATEFIX Western Australian GPS
network. Proc. of IAG97.
Welsch, W. (1981) Estimation of variances and covariances of geodetic observations. Aust. J.
Geod. Photo. Surv., No. 34. Pp.l-14.

270
 THE ASS lSI LANDSLIDE GPS NETWORK

D. Dominici (*), F.Radicioni (**), S. Selli (**), A. Stoppini (**)

*DISTART - University of Bologna

Viale Risorgimento 2, 40136 Bologna - Italy
** Istituto di Ingegneria Ambientale - University of Perugia
Via Duranti 1,06125 Perugia - Italy

Abstract. The surface movements of a landslide involving an urban area of the Assisi
town, in central Italy, have been measured since 1995 using the GPS technique. This
paper presents some design aspects of the GPS network, and a statistical analysis of three
annual survey campaigns.

Introduction

The control networks are set up to monitor natural and anthropic phenomena such as
landslides, subsidence and geotectonic movements. The GPS monitoring networks have
certain advantages compared with the traditional geodetic networks: visibility between
points not required, rapid and weather-independent surveying, etc.
A GPS network for the monitoring of the Assisi landslide, in central Italy, was
established in 1995. Since then, three measurement campaigns have been carried out of
this network. The monitoring zone is an urban area of about 2.5 km2 . The movements of
the landslide are relatively slow (a few centimetres per year); therefore great accuracy is
necessary for the three-dimensional positioning of the control points. All measurements
have been made by GPS double-frequency receivers, in static mode, using a forced-
centring system for all points (Dominici et al., 1996).
This paper presents an analysis of the results obtained so far. The quality of each
solution is discussed and a comparison is made between different computation modes
(single baseline vs. multibaseline, etc.). Some design aspects are also considered and a
comparison is made between simulated and real results.
Finally, a two-epoch statistical analysis is performed, in order to: a) verify the stability
of the "fiducial" markers; b) highlight the likely ground surface movements and evaluate
their significance.

International Association of Geodesy Symposia, Vol. 118

Brunner (ed.), Advances in Positioning and Reference Frames
© Springer-Verlag Berlin Heidelberg 1998
GPS control network at Assisi

A recently builded area of the Assisi town, in central Italy, is located on a slope where a
landslide has caused, in the last 20-25 years, consistent damage to several buildings. The
area has been monitored since the early '80s using geotechnical instruments
(inclinometers, piezometers, etc.), and by means of a traditional topographic survey of the
surface movements for some control points. About the end of the 180s, the topographic
survey was interrupted for economical reasons. Meanwhile, some stabilisation measures
had been applied, so that the landslide motion has slowed down.
In 1995, a GPS network was monumented by the University of Perugia with the
financial support of CNR (Italian National Research Council), aiming to: a) newly
monitoring the possible ground surface movements in the landslide area; b) testing the
effective accuracy in the three-dimensional positioning achievable with a local GPS
network.
The network (fig. 1) consists of six reference or "fiducial" points located in geologically
stable sites, and fourteen control points situated in the landslide area. The monumentation
of the vertices was done using stainless steel centring devices on concrete foundations.
The network was set up in spring 1995, and three measurement campaigns were carried
out in June 1995, July 1996 and May 1997. Most of the observations were made using
Trimble 4000 SSE receivers, with geodetic LIIL2 antennas. Only in the 1997 campaign,
an Ashtech Z-Surveyor receiver was used in some points having radio noise troubles.

S3

mrn Landslide Area

S1

100 m
....-t

S5

Fig. 1: The Assisi GPS network.

272
 The survey was organised in sessions, in order to obtain all the independent baselines
for the fiducial points network, and adequate connections between those and the control
points. The solution for each annual campaign was calculated using the Geotracer
Ver. 2.25 processing software, adopting the "single-baseline" procedure and Ll +L2
frequency (achieving fixed ambiguities solutions), and then performing a minimum
constraint adjustment for the whole network (with the S2 reference station kept fixed).
To control the quality of the results, further solutions were obtained for each campaign
with the Bernese Ver. 4.0 software, both in single and multi-baseline mode. Table 1 shows
a comparison between the adjusted coordinates of the 1996 survey obtained with two
different methods and softwares.

Points Vx Vy Vz Vn Ve Vh
(mm) (mm) (mm) (mm) (mm) (mm)
SI 0.0 -2.2 -1.7 -0.9 -2.2 -1.S
S3 -1.S -2.1 -2.7 -0.7 -1.8 -3.3
S4 O.S 2.9 -0.3 -1.0 2.7 0.7
SS -1.1 1.4 2.2 2.1 1.6 0.9
S6 2.0 0.1 2.S O.S -0.3 3.2

Table 1: Fiducial network; campaign 1996.

Residuals ofa 6-parameter transformations between:
a) Geotracer single baseline solution; b) Bernese multibaseline solution

Another solution for each campaign was computed by means of a robust method in the
adjustment phase, adopting a suitable influence function. The differences on the adjusted
coordinates with respect to the previous solutions resulted to be less than 2 millimeters.

Design aspects

To investigate a priori the quality of the network in the design phase, a simulation was
done, using a simplified approach (Dominici et al., 1994) which takes into account only
the network configuration on the ground, referring to "standard" observation conditions.
Considering a baseline expressed in the local system by NEH components, a simple a
priori stochastic model is adopted, assuming that:
• the correlations between the components N, E, H are negligible;
• the variances of the planimetric components Nand E are equal and given by:
cr/ = crn2 = cre2 = a2 + (bLl (1)

• the variance of the height component H is higher than those of the planimetric ones,
and given by:
(2)
In the above expressions, L is the baseline length, while a, b , c ,d are coefficients
whose values take into account the observation conditions, and are assigned on the basis
of previous experience. The error ellipsoids obtained with this simulation model have the

273
major axis in the H direction and the N-E section degenerates into a circle.
In Table 2, the results of the simulation (assuming a= 2 mm; b= 1 ppm; c= 3 mm; d=
1.5 ppm) are summarised and compared with those of the real solution. The comparison
shows the simulation values being slightly "optimistic", especially for the height
component.

Planimetric components Height component

(s) (r) (s) (r)
Fiducial points 3.3 + 3.5 2.0+ 3.7 4.0 +4.2 3.9+4.9
Control points 5.5 +6.0 3.7+7.1 6.6 +7.2 8.2 + 13.3

Table 2: Results of the simulation compared with those of the real solution.
2D (planimetric) and 1D (height comp.) confidence regions at 95% level; values in mm
(s) = Simulation; (r) = using real observations (campaign 1995)

Reference frame

For a local control network, like the one presented here, it does not seem necessary to
operate with a very accurate definition of the geocentric reference frame, since the results
are substantially relative positions in a small area. The problem of the reference system
stability, indeed, is more involved with the GPS networks of great extension.
Nevertheless, due to the high accuracy requirements, it is always recommended to
perform the computations in a "correct" reference frame, which practically involves:
• the connection of the network with one or more stations with known ITRF co-
ordinates (such as the IGS permanent stations);
• the use of precise ephemerides (such as the IGS ones).
Thus, the Assisi network data were processed using precise IGS ephemerides and
taking "good" ITRF coordinates from the Perugia University roof marker. This marker is
scheduled to host a GPS permanent station, and was connected to IGS permanent stations
previously.
For the analysis of the possible movements of the control points, a "local" reference
system - at a few millimetre level accuracy - is then defined by the coordinates of the
fiducial points, which must be tested if they can be assumed as fixed by means of the
congruency test between the results of each campaign (see next paragraph).

Stability of the fiducial network

The solutions obtained for the fiducial points from each measurement campaign (1995-
1996-1997) have been first compared using a 6-parameter Helmert transformation (1996
on 1995 and 1997 on 1995). As expected, the residuals were a few millimeters in
magnitude only.

274
 A more detailed analysis was then performed by means of a congruency test between
two epochs (Caspary, 1987). The aim of the congruency test is to check whether the
reference points have remained stable.
The null hypothesis Ho is represented by the equation:

Ho: E[x,]= E[xj]=x => H· x = 0; H = [I -II, x= [:;J (3)

From a practical point of view, a suitable test statistic is expressed by:

(4)

(5)

Where f,1. represents the independent rows of the Ho system, f represents the degree
of freedom of the two-epoch system and T is distributed as a central Fisher distribution if
Ho is true.
The results of the test between the 1995 and 1997 epochs are summarised in Table 3.
For the congruency test between the 1995 and 1996 solutions, the reader is referred to a
previous paper (Dominici et al., 1996).
Adj. WGS84 Displacements Test value
Point Coord. coordinates 1997-1995 t30(95%)=1.697 Test results
1997 sol. (mm) t30 (99%)=2.457
4> 43°03' 40.95537" -1.7 0.715 not
SI A. 12°38' 45.89612" -1.3 0.769 not
h 803.415 m -8.7 1.884 ?
4> 43°04 '24.04899" 1.8 0.322 not
S3 A. 12°36' 51. 97882" 4.1 0.713 not
h 548.001 m -0.4 0.067 not
4> 43°03'56.75781" 2.8 0.51 not
S4 A. 12°37'08.08233" 4.4 0.807 not
h 431.219 m 0.5 0.095 not
4> 43°03 '24.18942" -6.8 0.807 not
S5 A. 12°36'55.36042" 8.6 1.010 not
h 281.804 m 8.8 1.045 not
4> 43°04'19.48823" 8.5 1.384 not
S6 A. 12°37'23.43907" 4.3 0.698 not
h 486.317 m -11.1 1.800 ?

Table 3: Summary a/the results a/the congruency test between 1995 and 1997 solutions
(N.B.: Point S2 was fixed in the minimal constraint adjustment)

275
Testing movements of the control points

Finally, for those points that are expected to move (the control points), it is necessary to
compute vector displacements and related confidence regions, and to verify the
significance of the movements.
This analysis was performed comparing the 1996 and 1997 solutions with the initial
1995 positions. The movements found in the 2-year period (1-;.-2 cm values) result to be "at
the border" of the significance. However, a general trend could be seen: planimetric
movements towards south-west - directed downhill - and lower height components for
most of points, indicating a downhill motion.
Obviously, the analyses will be performed also for the next campaigns scheduled from
1998 to 2000. Some movements, indeed, are reasonably expected during the five-year
period 1995-2000, although some stabilisation work has already been carried out and will
be continued during the next few years.

Conclusions

The results shown have substantially confirmed the possibility of using the GPS technique
for monitoring surface movements at the cm accuracy level, and the correctness of the
statistical testing procedures adopted.
Future research should explore the application of robust estimation methods to the
deformation analysis.

References

Ayan T., Hekimoglu S., Ozliidemir M T., IIAnalysis oflandslide deformation measurements
by robust estimation methods ll , 1st Turkish Intern. Symposium on Deformation (1994).
Caspary w'F., IIConcepts and deformation analysis ll , Monograph 11 School of Surveying,
The University of New South Wales, Kensington, Australia (1987).
Chrzanowski A., Chen Y.Q., Secord J.M, IIGeometrical analysis of deformation surveysll
Deformation Measurements Workshop, MIT (1986).
Chrzanowski A., IIFIG Commission 6 activities in deformation monitoring and analysis ll ,
1st Turkish International Symposium on Deformation (1994).
Dominici D., Radicioni F., Stoppini A., Unguendoli M, IITesting on the redundancy effect
in GPS networks: some examples ll , 1st Turkish International Symposium on
Deformation (1994).
Dominici D., Radicioni F., Selli S., IIStatistical analysis of GPS monitoring network II ,
Proc. lAG Regional Symposium, SzekesfeMrvar, Hungary (1996).
Gruendig L., IIAdjustment and design methodology II, Deformation Measurements
Workshop, MIT (1986).
Kleusberg A.,Teunissen P.J.G.(Eds.), IIGPS for Geodesy II , Lecture Notes in Earth
Sciences, 60, Springer-Verlag, Berlin, Heidelberg (1996).
Schaffrin B., II Aspects of network designll, Optimization and design of geodetic networks,
Springer-Verlag (1985).

276
 USE OF A REGIONAL IONOSPHERIC MODEL IN
GPS GEODETIC APPLICATIONS

M. Gendel , C. Bruninil and A. Kleusberg2

1 Facultad
de Ciencias Astronomicas y Geofisicas, Universidad Nacional de La Plata,
Argentina
2 Department of Geodesy and Geomatic Engineering, University of New Brunswick, Canada

Abstract

A regional ionospheric model was developed using data from dual-frequency GPS receivers
aiming to correct single frequency observations. The data analyzed correspond to an
intermediate southern latitude band in a period of stable ionospheric conditions. The
efficiency of the ionospheric model was tested by comparing L3, Ll plus the ionospheric
model and L 1 solutions. This was done using different numbers of GPS stations, and different
session lengths were used. These comparisons were performed by similarity transformations,
paying particular attention to the scale factor. Advantages and limitations of the model are
discussed.

Introduction

This study analyzes the ability of a regional ionospheric model to correct the residual
ionospheric error. This error produces a contraction in the baseline length by up to 0.06 PPM
per TECU of ionospheric electron content (Santerre, 1991).
Data obtained from the survey of a 1st order geodetic network covering most of the Chubut
Province, Argentina, was used. This province is located in an intermediate latitude band (42°
to 46° south) which is usually a region with stable ionospheric conditions. The period

lFellow of the CONICET Consejo Nacional de Investigaciones Cientificas y Tecnicas. Argentina

analyzed (27-28 April 1995) corresponds to a period of low solar activity. The network was
measured with six dual-frequency GPS receivers. The distances between stations ranged from
200 to 400 Km. The observations were collected during a period of 15 hours. These
observations were used to adjust a regional ionospheric model that was incorporated within
the Bemese GPS, V 3.5, software (Rothacher et aI., 1993). Two sets of network coordinates
were calculated with this software using Ll observations only, with and without ionospheric
corrections.

The Geometry-Free Observable

In order to preserve the ionospheric delay information, the geometry-free observable P4

(difference between P code pseudorange in Ll and L2) was computed. Since P4 eliminates all
frequency-independent biases (selective availability, clock errors, ephemerides errors and
tropospheric delay) this linear combination was used to adjust the ionospheric models
(Beutler, 1989):

(1)

where TEC (total electron content) is the total amount of electrons contained in a cross
section of 1 m2 of a cylinder, having as its axis the signal path between the satellite and the
receiver. When the TEC is expressed in TECU (TEC Unit; 1 TECU= 10 16 electrons/ m2), the
constant k takes the value -0.105 rnI TECU; c is the speed of light, Ll'tR and Ll't s represent the
bias due to differential hardware delay in the receiver (Ll'tR) and in the satellites (Ll'ts)
respectively; v is the combination of the noise of measurement and of the multipath in PI and
P2.

The Ionospheric Model

In this model it has been assumed that the temporary variations of the vertical electron
content, VEC (Vertical Electron Content) are slow when they are described in a coordinate
system in which the sun remains approximately fixed (Brunini and Kleusberg, 1995), being a
geocentric system whose Z-axis points to the north pole and the system rotates about the Z-
axis, maintaining the X-axis on the meridian that contains the sun.
In order to relate the TEC to the VEC, the mapping function M=TEC/VEC has been
introduced. To model this function a simplification was made: the ionosphere was represented

278
as a thin spherical layer located at a height of 400 kilometers (the height which approximately
corresponds to the FI layer).
Figure 1 represents the basic geometry of the model: the incoming signal from a satellite S
goes through the thin layer at the point P and arrives at the station E at a zenith distance z.
The projection of the point P on to the terrestrial surface is the subionospheric point Q. The
spherical coordinates of the subionospheric point in the sun-fixed system are the latitude ~
and the hour angle h.
We adopted a geometric mapping function equal to the ratio between the oblique and
vertical distance, to the H height of the thin ionospheric layer:

TEC
M( z') = - - ~ cosec(z') (2)
VEC

where z' is the zenith distance of the satellite to the height of the ionospheric layer. This
approach ignores the contribution of the horizontal gradients in the distribution of free
electrons.

S: satellite, P: layer point, E: station, X: points to the sun

Figure 1: Basic Geometry for a single layer model in a sun-fixed system

279
Parametrization and Adjustment of VEe

A polynomial equation was used to describe the spatial variations of the VEe in the sun-fixed
system:

L M I
VEC(h,~)= L Lalm(h-ho) (~_~o)m (3)
I=Om=O

where hand <l> are the coordinates of the subionospheric point in the sun-fixed system. Using
equation (2) and (3), equation (1) becomes:

L M I S
P4=Kcosec(z')L Lalm(h-ho) (~_~o)m+C(L'l-rR +L'l-r )+v (4)
I=Om=O

Equation (4) is the basic observation equation of the problem. The unknown aim and L1'tR +
~'ts are determined by a least squares adjustment of the P4 observations. Figure 2 shows an
ionospheric delay map in meters obtained between 22.5 hours (local time) on 27 April 1995
and 8.5 hours on 28 April 1995 .

• 11.7) . 12.e:I ·12.14 ·11.:011 .1e.:?II ·9.&1 · 6.117 .7. 1Il .•. " • •• 14 · b.3ol · 4.40 ·1.";>

·31.1?':l . -'&!?J

·~.TJ ·~73

."1,2") .41 . !?J

IU

.a
~
· -€' .TJ ·1F.73
'-p
~
~ · 44 ,e-) ·44. f3

.-£, 7] · 1~.7]

• ..(l',?) . .P.e=3

Hour Angle

Figure 2. Regional ionospheric map

280
Computing the Model

We computed the ionospheric model using different numbers of GPS stations finding that it
does not change significantly with the number of stations used.
Due to the North-South dimension of the network a development of 2nd degree in latitude
was chosen; the longest sessions had a development of2 in hour angle, and the shorter ones of
1 in hour angle. Other developments were tested by increasing the number of terms in the
series, but the results got worse due to the Runge effect.
The session lengths were shortened, from 16 to 3 hours. The limiting factor for reducing the
session length was the deterioration of the results as a consequence of the worsening of the
parameters (ambiguities and coordinates) estimated from the shorter observation sessions.
Three hours was the optimum session length for our baseline solutions.

Evaluation of Model

The following procedure has been used in order to evaluate the results obtained after applying
the ionospheric corrections. The observations of all the stations of the network were processed
using the ionospheric-free combination (L3). This linear combination will reduce the first
order ionospheric effect, leaving only about 2% of it (Kleusberg et aI., 1996). Since L3 gives
the most precise set of coordinates (considering the dimensions of our network) we will
designate these as the "true coordinate set".
L1 frequency observations only were processed, in which case the ionospheric delay was
not corrected at all, and from this solution a different set of coordinates was obtained. Finally
we processed the L1 frequency observations applying our ionospheric model. Now the effect
of the ionosphere was partially absorbed although it was not entirely removed. A third set of
coordinates was obtained. We evaluated the efficiency of our ionospheric model by
comparing different similarity transformations. The first one was between the "true
coordinates" and those obtained using the L 1 data only; the second one was between the "true
coordinates" and the coordinates calculated with the L1 data corrected by the ionospheric
model. In this way two sets of 7 parameters were obtained and it was possible to compare the
influences that the ionospheric model produced on the network stations. This procedure was
repeated for the four sessions of different length.

281
 Session length Scale without RMS without Scale with RMS with
hrs ionospheric ionospheric ionospheric ionospheric
model model model model.

16 0.147 ± 0.073 0.0407 0.016 ± 0.073 0.0407

8 0.257 ± 0.119 0.0663 0.084 ± 0.l16 0.0646
4 0.241 ± 0.210 0.1174 0.031 ± 0.214 0.1195
3 0.287±0.181 0.1012 0.016 ± 0.190 0.1064

Table 1: Results for 7 parameter transformation. Residuals in Local System (North, East, Up).
Taylor series development: 2 latitude terms, 2 hour angle terms and 2 mixed terms.

Conclusion

The results in Table 1 show the utility of the model in minimizing the scale factor of the 7
parameter transformation.
Once Ll was processed without using the ionospheric model, a scale factor similar to the
theoretical value suggested by Santerre (1991) was obtained, while the factor diminished by at
least one order of magnitude when the model was applied. This fact demonstrates the success
of the model that we have used, consistent with the previously mentioned fact that the
ionosphere biases result in a baseline contraction. Appreciable differences in the other
parameters do not arise from the present results.
Residuals (RMS) do not change with the application of the model, which indicates that only
low frequency information could be estimated. Residuals also grow as the session length
decreases.

Bibliography

Beutler G., Gurtner W., Rothacher M., Wild U. and Frei E. (1989), Relative static positioning
with the Global Positioning System: basic technical considerations. In: The lAG General
Meeting, Edinburgh
Brunini C. and Kleusberg A (1995), Mapas globales de retardo ionosferico vertical a partir de
observaciones GPS. In: Actas de V Congreso Internacional Ciencias de la Tierra, Santiago,
Chile.
Georgiadou Y. and Kleusberg A (1988), On the effect of ionospheric delay on geodetic
relative GPS positioning. Manuscripta Geodaetica, Vol. 13. 1-8.
Kleusberg A, Teunissen P. (Eds.) (1996), GPS for Geodesy. Springer. 175 -218.
Rothacher M., Beutler G., Gurtner W., Schildknecht T. and Wild U. (1993), In: Bernese GPS
Software Version 3.4. Documentation, Astronomical Institute, University of Berne.
Santerre R. (1991), Impact ofGPS satellite sky distribution. In: Manuscripta Geodaetica, Vol.
16, 28-53.

282
 INSTANTANEOUS AMBIGUITY RESOLUTION FOR MEDIUM-
RANGE GPS KINEMATIC POSITIONING USING
MULTIPLE REFERENCE STATIONS

Shaowei Han & Chris Rizos

School of Geomatic Engineering
The University of New South Wales
Sydney NSW 2052, Australia

ABSTRACT

A linear combination functional model, formed from the single-differenced functional

equation for baselines from a mobile receiver to three or more reference receivers, is
proposed, in which the orbit bias and ionospheric delay can be eliminated, and, in addition,
the tropospheric delay, multipath and observation noise can be significantly reduced. As a
result, the ambiguity resolution technique that can be employed for medium-range GPS
kinematic positioning is similar to that used for the short-range case. Kinematic tests have
been carried out in Sydney, Australia, with separations from the nearest reference stations
greater than 30km. The carrier phase ambiguities can be resolved for every epoch and the
success rate (correct identification of the integer ambiguities) was 100%. This technique
is well suited to real-time precise GPS kinematic positioning.

INTRODUCTION

On-the-fly ambiguity resolution for short-range kinematic positioning assumes that the
orbit bias and differential ionospheric delay can be ignored, and the integer ambiguities
can be resolved easily (Han, 1997a). For medium-range GPS kinematic positioning, the
residual biases after double-differencing the data have been investigated over the last few
years. It was shown that using more than three reference stations with known coordinates,
the orbit bias can be eliminated for medium-range applications (less than 100 km to the
nearest reference station) through the use of a linear combinations of single-differenced
observations (Han & Rizos, 1996; Wu, 1994). The ionospheric delay relative to one
reference station can also be interpolated if the relative ionospheric delay for three or more
reference stations are known (Han & Rizos, 1996; Wanninger, 1995; Webster &
Kleusberg, 1992). In this paper a linear combination method has been used to account for
orbit bias and ionospheric delay. The tropospheric delay, multipath, and observation noise
will also be mitigated using this method. After the distance-dependent biases are
eliminated or mitigated, an ambiguity resolution method incorporating a three-step quality
control procedure is used, which has been succesfully used for short-range GPS kinematic
positioning (Han, 1997a).
LINEAR COMBINATION MODEL

Figure 1 shows a three reference station network with one roving receiver. It is preferable
that one of the reference stations is connected to the IGS network in order to obtain precise
positions in the global ITRF frame. Data processing for the reference stations is necessary
in order to determine the double-differenced integer ambiguities between them.

Figure 1. Configuration of the reference stations and the roving receiver

The single-differenced carrier phase observation can be written (Han, 1997b):

(1)

where ~(.)i = (.)U - (\; i indicates the reference station i, and u the user station; l/J: the
carrier phase observation in unit of metres; p: = IIX Xii,
s - xs is the satellite position
vector, X is the station position vector; dp: the effect of ephemeris errors, including SfA
effects; dT: the receiver clock error with respect to GPS time; d ion : the ionospheric delay;
d trop : the tropospheric delay after model correction; d~: the multipath on the carrier
phase; erp: the carrier phase observation noise; Il: the wavelength of the carrier phase;
and, N: the integer ambiguity.
A set of parameters ai can be determined, based on the conditions given in Han & Rizos
(1996):

3
and Lai(X -Xi)=O
U (2)
i=l

284
where Xu and Xi (i=1,2,3) are the position in the Gauss plane coordinate system. If
3
more than 3 reference stations are used, L a/ = min should be introduced in roder to
i=1
reduce the observation noise (see equation (8)).
The linear combination of the single-differenced observations can be formed as:

3 3 3 3 3
Lq ·~lR ~ La ·~Pi + La ·~dpi -c· La ·~d1j +,1,. La
i=1 i=1
i
i=1
i
i=1
i
i=1
i . &Vi -

3 3 3
-Lq .Mi(Xl,i + Lq .MtrqJ,i + Lq .M!r,i +£
i=1 i=1 i=1
3
I,~'~
(3)
1=1

The orbit bias term has been proved to be (Han, 1997b):

3
La ·~dpi ",,0
i (4)
i=1

The epoch-by-epoch and satellite-by-satellite ionospheric model can be applied to

estimate the single-differenced ionospheric delay relative to reference receiver 3. It has
been proved that the interpolation of the single-differenced TEC between receivers, or
double-differenced TEC based on an ionosphere layer can be simplified to an the
interpolation dependent on the receiver position in the Gauss plane coordinate system
(Han, 1997b). The ionospheric delay term can be deduced as (ibid, 1997b):

~ai
3
. Mion,i =dion,u - dion ,3 -
[ a1
a ]T[d -d ]=0
d.
ion,l ion,3
-d. (5)
1=1 2 IOn,2 IOn,3

When the distance between the reference stations increases, the residual error will become
greater due to the ionospheric delay interpolation.
If it can be assumed that the tropospheric delay can be interpolated from the residual
tropospheric delay at the reference stations, the residual tropospheric delay can be
represented as:

L
3
ai . ~dtrop,i = dtrop,u - d trop ,3 - [xu
Yl ]-1 [d troP ,1 - d troP ,3]
(6)
i=1 Y2 d trop ,2 - d trop ,3

where (xu,yJ, (X 1'Yl) and (X 2 'Y2) are North and East coordinates relative to reference
station 3 in Gauss plan coordinate system. How closely it should be to zero depends on
the spatial correction of tropospheric delay. The residual tropospheric delay is mostly
contributed to by the wet component of the troposphere, which shows strong variation

285
 Ia ·l1d
3
with height, time and location. But it can only be expected the term i trop ,i to be
i=l
mitigated to some extent dependent on the distance between stations.

The multi path term can be rewritten as:

(7)

3
The last term La; .d!p,;
;=1
on the right hand side of equation (7) is the weighted mean value

of the multi path values at the three reference stations for this satellite. Due to the random
nature of multipath at the different stations, the weighted mean value will be significantly
reduced if all a i (i=I,2,3) are positive and less than 1, although the weight a i is not
derived from its standard deviation. On the other hand, the multipath at the roving station
will become a high frequency bias, and mostly will be close to random noise (Zhang &
Schwarz, 1996). Therefore, the multipath term has been significantly reduced and will be
ignored in the functional model.
The standard deviation of the one-way carrier phase observation can be approximated as
a function of the elevation angle. Because all stations are located within a region of about
100km radius, the elevation of a satellite is approximately the same. The standard
deviation of the one-way carrier phase observation can also be approximated by (]j and
then the standard deviation of the linear combination of single-differenced observations
£3 can be expressed as:
Lar~l/!i
i=1

(8)

Compared to the standard deviation of the single-differenced carrier phase observation

.J2. (]j , the standard deviation will become smaller if the roving station is located within
the triangle formed by the reference stations (Han, 1997b).
The single-differenced carrier phase observation functional model can be simplified to:

3 3 3 3
"£.JI
a· ·11'/'.
'rl = " a· ·l1dTI + 11.,. "£.JI
a· .l1p.I - c· "£.JI
£.JI a· . MI·I + £ 3 (9)
i=l i=l i=l i=l .La;-l:J.l/!i
,=1

DOUBLE-DIFFERENCED FUNCTIONAL MODEL

Equation (14) can be written as:

286
 3
~¢u.3 - [a
l . ~<A.3 + a 2 . ~¢Z.3 ] = ~Pu.3 - [a
l . ~P1,3 + a 2 . ~P2.3 ] - C· La
i=I
i • ~d1j +
+A,·/1Nu.3 -[a l ·/1NI.3 +a2 ·/1N2.3]+C3
'i,a;,ArfJ;
(10)
;=1

Define the residual vectors:

(11)

V2,3 = V~¢Z,3 - V/1N 2,3 - V~P2,3 (12)

The double-differenced observation model can then be written as:

(13)

Using the known coordinates and known integer ambiguities, the correction vectors VI3 ,
V23 . For real-time applications, the correction vectors, together with the carrier phase and
pseudo-range data at reference station 3, can be sent to the roving receiver in real-time.

EXPERIMENTS

The experiment was carried out on 14 December, 1996, using four Ashtech Z12 GPS
receivers. Three reference receivers were set up with separation of 40.0 km, 65.6 km and
70.7 km, and a roving receiver was mounted on a car from a site 31.44 km, 34.11 km and
46.5 km distant from the reference receivers. The roving receiver started to move a
distance of about 7 km and back to the start point, during a period of about 32 minutes.
The data rate was 1 Hz and a total of 1903 epochs were collected.
The coordinates have been precisely determined at the three reference stations, and the
correction terms [a ·VI,3 +a2 ,V2,3]
l in equation (13) for the Ll and L2 carrier phase
observations were derived and, as an example, plotted in Figure 2 for satellite pair PRNs 9
& 24. Using the proposed functional model and an ambiguity resolution method with
three-step improvements suggested in Han (1997a), integer ambiguities have been
resolved using single-epoch data with 100% success rate.

CONCLUDING REMARKS

A linear combination functional model is proposed, formed from the single-differenced

functional equation for baselines, from the user roving receiver to three or more reference
stations, in which the orbit biases and ionospheric delay terms can be eliminated and, in
addition, the tropospheric delay, multipath and the observation noise can be reduced.

287
Using this model, the integrated method with three-step quality control procedure can
resolve integer ambiguity instantaneously. The medium-range experiment described here
indicates a 100% success rate. This experiment has demonstrated the feasibility of the
proposed technique for medium-range kinematic positioning. Greater separation between
the roving receiver and reference stations will be tested in the near future. This algorithm
has been designed for real-time applications.

8~~--~~--~~~--~~--~~~

E t
I~fL~__~~__~~~__~~__~~~
1600 1000 2000 2200 2400 2600 2800 3000 3200 3400 3600
Seconds Counted from 11 :OOam on 14 Dec., 1996 (Local)
8

.!:!o6

!:
E f
"---'------'----'------'-------1.--'----'----'------'-----'------'
1600 1aJO 2000 2200 2400 2600 2800 3000 3200 3400 3600
Seconds Counted from 11:ooam on 14 Dec., 1996 (Local)

Figure 2. Example of the corrections for double-differenced carrier phase observations

for satellite pair PRNs 24 & 9

REFERENCES
Han, S. (1997a). Quality Control Issues Relating to Ambiguity Resolution for Real-Time
GPS Kinematic Positioning. Journal of Geodesy, 71(6):351-361.
Han, S. (1997b). Carrier Phase-Based Long-Range GPS Kinematic Positioning. PhD
Dissertation, UNISURV S-49, School of Geomatic Engineering, The University of New
South Wales, Sydney, Australia, 185pp.
Han, S. and C. Rizos (1996). GPS Network Design and Error Mitigation for Real-Time
Continuous Array Monitoring Systems. Proc. 9th Int. Tech. Meeting of The Sat. Div. of
The U.S. Inst. of Navigation, Kansas City, Missouri, 17-20 Sept., 1827-1836.
Wanninger, L. (1995). Improved Ambiguity Resolution by Regional Differential
Modelling of the Ionosphere. Proc. 8th Int. Tech. Meeting of The Sat. Div. of The U.S.
Inst. of Navigation, Palm Springs, California, 12-15 Sept., 55-62.
Webster, I.,and A. Kleusberg (1992). Regional Modelling of the Ionosphere for Single
Frequency Users of the Global Positioning System. Proc. 6th Int. Geodetic Symp. on
Satellite Positioning, Columbus, Ohio, 17-20 March, 230-239.
Wu, J. T. (1994). Weighted Differential GPS Method for Reducing Ephemeris Error.
Manuscripta Geodaetica, 20, 1-7.
Zhang, Q. J. and Schwarz, K. P. (1996). Estimating Double Difference GPS Multipath
under Kinematic Conditions. Proc. IEEE Position Location & Navigation Symp.
PLANS'96, Atlanta, Georgia, 22-26 April, 285-291.

288
 EXPLOITING THE SIRGAS COLOCATIONS FOR DETERMINING
ELEVATION DEPENDENT PHASE CENTER VARIATIONS
OF GEODETIC GPS ANTENNAS

Klaus Kaniuth
Deutsches Geodatisches Forschungsinstitut, Aht. I (DGFI/I)
Marstallplatz 8, D-80539 Munchen, Germany

ABSTRACT

During the SIRGAS 95 GPS campaign several sites were continuously occupied during ten
days by two different receiver systems. From the data acquired at all those colocation sites
providing sufficiently accurate ground truth, either from terrestrial geodetic observations or
from SIRGAS independent GPS measurements, the elevation dependent phase center
differences between the involved antennas have been determined. The modelling approach
and the achieved results are presented.

INTRODUCTION

Besides multipath and tropospheric effects the antenna phase center variations are probably
the most important error contribution when aiming at few millimeters accuracies in GPS
applications. Moreover, parameters accounting for phase center biases and for tropospheric
delays are highly correlated, and both errors will affect the height estimates. For determining
phase center variations two different approaches have been applied so far:

Absolute calibrations of antenna phase patterns by anechoic chamber measurements

(Schupler et al. 1995);
Determination of relative phase center variations from field calibration experiments
(e.g. Breuer et al. 1995, Rothacher et al. 1995).

Within the International GPS Service for Geodynamics (IGS) frame a compilation combi-
ning all available calibration results has recently been realized and suggested for application
(Rothacher and Mader 1996).
 Field calibration experiments in local networks extending over some tens of meters only
generally require that

the eccentricity vectors in the network are very accurately established or can be
determined by properly interchanging the antennas during the experiment,
any systematic effects such as orbit errors, satellite clock biases and tropospheric errors
cancel over such short distances.

The elevation dependent phase center variations presented in this analysis do not result
from a dedicated calibration experiment, instead they are derived from colocation data
acquired during a GPS field campaign.

DATASET

In May I June 1995 a GPS campaign for establishing a geocentric reference system for the
South American continent, SIstema de Referencia Geocentrico para America del Sur
(SIRGAS), has been carried out. Consisting of some sixty stations occupied during ten days,
different precise GPS receiver systems were involved. As at the planning stage of that project
no commonly accepted antenna phase center models were available, different receivers were
operated in colocation at a number of sites allowing to process receiver specific subnetworks
and to combine these at the normal equation level by including independently established
local ties. Table 1 displays all colocations providing at least four days of observations, the
majority of sites were occupied during the whole campaign.

Table 1 Colocations of different receiver systems during the SIRGAS 95 GPS campaign

Site (Country) Involved Receiver I Antenna Systems

Bogota (Colombia) - TurboRogue I Dome Margolin T
Leica 200 I Internal
Curitiba (Brazil) - Trimble SSE 14000ST LIIL2 Geo
Ashtech Z-XII3/Geodetic LIIL2 700228
El Maiten (Argentina) - Trimble SSE I TR Geod LIIL2 GP
Leica 200 I Internal
Maracaibo (Venezuela) - Trimble SSE I TR Geod LIIL2 GP
Leica 200 I Internal
Montevideo (Uruguay) - Leica 200 I Internal
Ashtech Z-XII3 I Geodetic LIIL2 700 718
Puerto Iguazu (Argentina) - Ashtech Z-XII3 I Geodetic LIIL2 700228
Leica 200 I Internal
Rio Grande (Argentina) - TurboRogue I Dome Margolin T
Leica 200 I Internal

290
 The selection of these colocation sites could not be adequately based on theoretical consi-
derations but was mainly depending on equipment resources, infrastructure and logistic
aspects. This is the reason for the overrepresentation of the Leica 200 Internal compared to
all the other systems.

The ground truth required for estimating the phase center differences is based on GPS
measurements performed independently of the SIRGAS campaign with identical receiver/-
antenna systems. At some of the sites also levelled height differences between the GPS
markers are available. Unfortunately, compared to the large amount of colocation data the
local ties cannot be considered equally accurate.

ANALYSIS

This analysis is restricted to elevation dependent phase center variations because due to the
objective of the performed colocations the antennas were not rotated and at some sites the
vertical ground truth was more accurate than the horizontal. For estimating phase center
variations the following approach has been implemented in the Bernese GPS software
version 3.4 (Rothacher et al. 1993) and in DGFI's program ACCSOL which accumulates and
solves normal equations and allows to introduce a variety of fiducial informations and
constraints:

Considering that the main part of offsets with respect to the applied a priori phase
center locations (table 2) is expected in the vertical and that some software systems are
not capable of applying elevation dependent phase center corrections, we firstly
estimate a mean vertical phase center offset between the involved antennas; this esti-
mate is the least squares fit to all included data down to the selected cutoff angle and
maps to any elevation angle E with sine E.

Relative to these mean vertical phase center offsets elevation dependent range
corrections are estimated in a second step; either one polynomial of arbitrary degree is
fitted to the entire elevation range, or piecewise linear functions for elevation bins
constrained to continuity at the bin boundaries are estimated. The results to be
presented are based on the latter approach.

All fiducial information such as the local baseline vectors or height differences are
introduced in program ACCSOL as individually weighted observation equations in
both adjustment steps; in particular, the elevation dependent corrections relative to the
mean vertical phase center are solved applying the minimum condition
II y - Axl12 + All xp 112 = minimum,
where y is the observation vector, A the design matrix, Xs the vector of station
coordinates, ~ the vector of phase center parameters, xT =(xs' ~), and A a relative
weighting factor which was set to one in this analysis.

291
Table 2 Applied a priori phase center offsets in L 1 and L2 [mm] with respect to antenna
reference point (ARP); BPA =Bottom of preamplifier, TOP =Top of pole (350
mm above height hook)

Antenna ARP L1 L2

TurboRogue Dome Margolin T BPA 110.0 128.0

Leica Internal TOP 91.0 91.0
Trimble 4000ST L11L2 Geo BPA 70.0 68.0
TR Geod LIIL2 GP BPA 62.5 62.5
Ashtech Geodetic L11L2 700 228 BPA 64.0 64.0
Geodetic L11L2 700 718 BPA 72.0 72.0

RESULTS

The discussed approach for estimating mean vertical antenna phase center differences as well
as elevation dependent phase center variation differences has been applied to the SIRGAS
colocation data. From all the performed computations the following results are presented:

The mean vertical phase center differences between the colocated antennas with
respect to the a priori effects (table 3);
An example of the consistency of the elevation dependent range correction estimates
(figure 2);
The resulting elevation dependent range corrections in L 1 and L2 for each antenna
combination (figures 3 and 4).

It should be noted that the piecewise linear functions are represented by cubic splines in
figures 2, 3 and 4.

Table 3 Resulting mean vertical phase center offset differences [mm] between colocated
antennas relative to the applied a priori offsets

Antenna I - Antenna II Ll L2
Leica Internal - TurboRogue Dome Margolin T 21.7 25.9
Leica Internal - Ashtech Geodetic LIIL2 700 228 11.6 11.4
Leica Internal - Ashtech Geodetic LIIL2 700 718 7.1 18.4
Leica Internal - Trimble TR Geod L11L2 GP 13.6 16.1
Trimble 4000ST L11L2 Geo - Ashtech Geodetic LIIL2 700 228 -19.3 -12.5

292
 30,-----------------------------~

SITE: RIO GRANDE

..-.
~ 28 mL 1 §I L2
'-'

z
o
2
~
U
LLI
IX
IX
o 0
u

43
I I I150
DAY
152 20 ~ 80
ELEVATION ANGLE (DEG)
80

Fig. 1 Consistency of daily estimates of Fig. 2 Consistency of daily estimates of

the mean vertical phase center the elevation dependent range cor-
offset difference Leica Internal - rection differences Leica Internal -
TurboRogue Dome Margolin T for Trimble SSE TR Geod LIIL2 GP
both frequencies L 1 and L2 forLI

to,------------------------------.
to

..-.
;::!;
;::!;
'-' '-'
0
z z
0 o
~ I- a
U U
LLI
LLI
IX
--' IX
IX IX
0
U
8 -5

LLI -to LLI

(!) (!)
z Z

"'
IX
-t5
•
o
INTERNAL - DORNE MARGOLIN T
INTERNAL - TR GEOD Ll/L2 GP
~ - to •
o
INTERN... L - DORNE MARGOL I N 1
I NTERN... L - TR GEOD L 1/LZ GP
o INTERNAL - GEODETIC Ll/L2 700718 o INTERNAL - GEODETIC LI/LZ 700718
• INTERNAL - GEODETIC Ll/L2 700228 • I NTERN... L - GEODEl I C L 1/L2 700228
• 40005T Ll /L2 GEO - GEODETIC L1/L2 700228 - 15 • 40005T LI/L2 GEO - GEODEl l C 1I/L2 700228
-20
~ ~ ~ ~ ~ ~ ~ ~

ELEVATION ANGLE (DEG) ELEVATION ANGLE (DEG)

Fig. 3 Resulting elevation dependent range Fig.4 Resulting elevation dependent range
correction differences for all analy- correction differences for all analy-
zed antenna pairs (L 1) zed antenna pairs (L2)

293
CONCLUSION

Data acquired during the SIRGAS 95 GPS campaign by colocated receivers have been
exploited for determining the differences between the elevation dependent phase center
variations of the involved antennas. The modelling approach includes

the fitting of a mean vertical phase center offset to the whole elevation range, and
a subsequent estimation of elevation dependent range corrections applying a minimum
condition.

The consistency of daily estimates of both the mean phase center as well as the elevation
dependent correction terms is on the one millimeter level. The accuracy of the resulting
variations depends on the reliability of the available ground truth, and may therefore in some
cases be slightly worse.

ACKNOWLEDGEMENT

The support given by several institutions providing local tie measurements at the SIRGAS
colocation sites is gratefully acknowledged.

REFERENCES

Breuer, B., 1. Campbell, B. Garres, 1. Hawig and R. Wohlleben (1995): Kalibrierung von
GPS-Antennen fUr hochgenaue geodatische Anwendungen. Journal of Satellite Based
Positioning, Navigation and Communication (SPN), 2/1995, 49-59.
Rothacher, M., G. Beutler, W. Gurtner, E. Brockmann, and L. Mervart (1993): The Bernese
GPS Software Version 3.4. Astronomical Institute, University of Berne.
Rothacher, M., S. Schaer, L. Mervart and G. Beutler (1995): Determination of Antenna
Phase Center Variations Using GPS Data. In Gendt, G. and G. Dick (Eds): Proceedings of
the IGS Workshop on Special Topics and New Directions, GeoForschungsZentrum
Potsdam, May 15-18, 1995,205-220.
Rothacher, M. and G. Mader (1996): Combination of Antenna Phase Center Offsets and
Variations; Antenna Calibration Set IGS-Ol. International GPS Service for Geodynamics
(IGS).
Schupler, B.R., T.A. Clark and R.C. Allshouse (1995): Characterization of GPS User
Antennas: Reanalysis and New Results. In Beutler, G. et al. (Eds): GPS Trends in Precise
Terrestrial, Airborne and Spaceborne Applications, lAG Symp. No 115, July
3-4, 1995, Boulder, Co., 328-332.

294
REAL-TIME FAILURE DETECTION AND REPAIR IN IONOSPHERIC
DELAY ESTIMATION USING GPS BY ROBUST AND
CONVENTIONAL KALMAN FILTER STATE ESTIMATES

Lao-Sheng Lin & Chris Rizos

School of Geomatic Engineering, The University of New South Wales
Sydney NSW 2052, Australia
Stelios Mertikas
Department of Mineral Resources Engineering
Technical University of Crete
GR-73 100 Chania, Crete, Greece

ABSTRACT
The ionospheric delay is one of the main error sources for precise GPS positioning and
navigation. Dual-frequency GPS receivers can be used to eliminate the ionospheric delay
(to the first order) through a linear combination of Ll and L2 observations (Hofmann-
Wellenhof et aI., 1994). Taking advantage of this physical dispersive principle, one or
more dual-frequency GPS receivers can be used to determine a model of the ionospheric
delays over a region of interest and, if implemented in real-time, can support single-
frequency GPS positioning and navigation applications.
There are, however, several factors which may affect the accuracy of real-time ionospheric
delay estimation. One of the dominant factors, the failure in the carrier phase observations,
such as the occurrence of cycle slips (for example, Engler et aI., 1995), is the topic of this
paper. In order to improve the real-time ionospheric delay estimation performance a novel
approach to addressing this problem is presented here.
The concept and methodology of the proposed algorithm is introduced. Preliminary test
results are presented to indicate the performance of this approach for real-time ionospheric
delay estimation.

REAL-TIME FAILURE DETECTION AND REPAIR

The Kalman filter, referred to as the conventional Kalman filter here, is a recursive
estimation procedure that solves the state vector estimation problem for a linear dynamic
system with the linear observation model (Gelb, 1974). For a linear discrete-time stochastic
system with Gaussian white noises the conventional Kalman filter produces the minimum
variance estimate of the states. However, when the noise processes are non-Gaussian the
filter performance is very sensitive and the state estimates will be easily biased, yielding
poor accuracy of its estimates. In order to obtain a robust estimate of states a robust Kalman
filter of the Masreliez-Martin type (Mertikas & Rizos, 1998) is adopted in this paper.
Further details concerning robust Kalman filters can be found in, for example, Wang et al.
(1993), Mertikas & Rizos (1998), and other Kalman filtering textbooks

Concept of Real-Time Failure Detection and Repair

The differences between the conventional (non-robust) and the robust Kalman filter
estimates of the states are considered as being "failure-indicating signals" (Ibid, 1998).
These signals are assumed to follow a normal distribution with mean value Jl and variance
0'2 , provided the assumptions underpinning the Kalman filter are satisfied. Measurement
departures from a conforming class (process in-control) will be indicated as shifts in the
mean value Jl and the standard deviation O'of the process. Further, when a gross error is
present in the measurements this can be detected by readily implementing an outlier
screening procedure in a sequential form. The following steps describe the method of
detecting failure of Total Electron Content (TEC) estimates using GPS measurements in
real-time (Lin, 1997).
The failure-indicating signals X are defined as:
i =1,2,,,, (1)

where TECj and TECj* are the TEC states estimated by the conventional and robust Kalman
filter at time i respectively. _
The sample mean value Xr and the sample variance S~ of the first r observations
{X1, X 2 ,···, X r} can be obtained sequentially using the following expressions:

- 1 -
Xr = -[(r -1)Xr-1 +X r]
r (2)
2 r-2t"2 1 - 2
Sr =--"r-1 +-(Xr -Xr-1)
r-1 r
In order to downgrade the influence of outliers on the estimation of the mean and standard
deviation of the process, the Winsorization process is used to robustify the estimation
procedure. In this case the raw measurements X r have been replaced by pseudo-
observations X~ such that:

. IX r -Xrl
Xr If <c
Sr
X*= 'f Xr - Xr (3)
r X r +cSr I >c
Sr
'f X r - X r
Xr -cSr I <-c
Sr

These truncated data values X * are then used to produce estimates of the mean X; and the
standard deviation S~ of the process. The constant c regulates the amount of robustness in
the Winsorization process (in this paper c=;=3, a value derived empirically). To test the
compatibility of the new random variable X r with its predecessors {X~, X 2 , .. ·, X~_1} , the
standardised variate t, which is a good approximate to the Student's t-distribution, is defined
as:

296
 (4)

The new variables are then transformed to a more convenient standard normal variable
u( t r) using the following approximation:

u(tr) = sigr(X; - -Xr-1)

* [8( r - 2) + 1]
8(r-2)+ 3
{e
(r - 2) 10 1+ r
(r-2)
J- N(O,1) (5)

To screen the outliers in this sequential data set, a discordance test using the well-known
three-sigma standard deviation identification rule is used:

Declare X r as failure if lu(t r)1 > 3 (6)

A disagreement between the conventional state estimates and its robust ones indicates that a
failure has been detected. The following example illustrates the operation of the failure
detection and repair algorithm.

Detection and Identification Step

Two Kalman filters, one conventional and one robust, are used to estimate the TEC for each
tracked GPS satellite (in this example, only PRNI satellite is considered). If the lu(tr)1
value of PRNI is greater than 3, then it is assumed that a carrier phase failure has occurred.

Repair Step
Let TS be the phase-derived TEC, D the linear combination of the integer ambiguity
parameters for Ll and L2 carrier phase observations 0"1, A2 being the wavelengths), L1TS
the failure correction to TS, TS the corrected TS, and CN the cycle slip number (see Lin,
1997, for more details). L1TS and CN are set to zero at the start of tracking of PRNI. The
relationship between TS and TS is:

TS= TS+L1TS (7)

Let Inno(TS) be the innovation value of TS, which can be computed from the
conventional Kalman filter as:

Inno(TS) = TS- [TEC(-) + 0(-) ] (8)

where TEC( -) and 0(-) are estimates of TEC and D before measurement update.
Let dN1 and dN2 be the carrier phase cycle slips on the Ll and L2 signals respectively (in
units of cycles). The combined effects of dN1 and dN2 on TS are:

(9)

Considering the fact that the minimum possible value of dTS is ±O.24 TECU (TEC units),
when dN1=-5 and dN 2=-4, or dN1=5 and dN2=4, the critical value of 0.25 TECU was

297
selected as the criteria for invoking the repair action. For example, if the failure occurred at
epoch k, then the innovation Inno(TS) is computed by equation (8) and the failure
correction ~TS is determined according to the steps listed in Table 1. The predicted states
of epoch k are adopted as estimated states. From epoch k+ Ion, the TS is corrected by
equation (7).

Table 1. Main steps of the failure detection and repair algorithm.

CN=O, and ~TS=O

while PRN 1 is tracked:
read data
compute TR, TS
TS=TS+~TS
estimate TEC, D, etc.
if lu(tr )1> 3 if
Ilnno(TS)1 ~ U25fECU
ifCN=O
~TS=O
~TS=~TS-lnno(TS)
CN=CN+ 1
endif
~TS=~TS-lnno(TS)
CN=CN+ 1
X(+)=X(-)
endif
endif
endwhile

TEST RESULTS AND DISCUSSION

Data Description
The UNSW data set was originally collected from 28 April, 1995 (day 118) to 1 May, 1995
(day 121) on the roof of the Geography & Surveying Building, The University of New
South Wales, using Leica System 200 GPS receivers. Four astronomical pillars, EC02,
EC05, EC14 and ECI9, were selected as station sites. Each GPS receiver collected 2 hours
of data on four consecutive days. The data rate was 1 second and the elevation cut-off angle
was set to 20 degrees.
From earlier tests using this data it was found that there were measurement failures (that is,
cycle slips) in the data collected from pillar EC05. Therefore, the data sets from this site
were chosen to test the above algorithm.

Cycle Slip Detection and Repair

Post-processing mode
In order to verify the performance of the proposed procedure, a carrier phase data post-
processing algorithm involving widelane and cI> -79 phase combinations (Han, 1995) was
used to detect and repair the cycle slips. The data; after application of the cycle slip repair
procedure, was processed using the carrier phase levelling process (see, for example, Lin et

298
aI., 1996). The carrier phase levelled TEC derived from the cycle slip free data (line B in
Figure 1) was used to verify the performance of the procedure.
Real-time mode
A real-time TEC estimation software package, known as REALTEC, was also developed.
REALTEC was used to process the raw data from day 121, 1995. The test results are
shown in Figure 1. The real-time TEC estimates without applying the multipath template
(Lin, 1997) are represented by line A. The post-processed TEC estimates with cycle slip
free data are represented by line B. Line C represents the real-time TEC estimates after
applying the multipath template generated from the current day's data. Line D indicates the
post-processed TEC estimates from the raw data. The range of elevation angle of the
satellite (PRN21) is also shown. (Note that line C has been shifted by +0.5 TECU from
line B for clarity, otherwise there appears to be no difference in the cycle slip estimates
whether made in post-processed mode or in real-time.) From these test results it is
concluded that: (1) REALTEC is capable of detecting and repairing the carrier phase cycle
slips in real-time, and (2) the achieved TEC accuracy from REALTEC is identical to that
obtained from post-processing if the multipath template is used.

PRN 21
-5~----~----~~----~----~------~----~ 60

c
o
.~ 40
>
Q)

w
30
1000 2000 3000 4000 5000 6000
Epoch (sec)

Figure 1. Post-processed versus real-time TEC estimates for PRN21 , for day 121, 1995.

Another example is taken from data from day 119, 1995. The test results are shown in
Figure 2. The post..;processing result indicates that there are three cycle slips on PRN21 , as
indicated by line C. The post-processed TEC estimates with cycle slip free data are are
shown as line A. The real-time TEC estimates using REALTEC with raw data are indicated
by line B. (As before, line B has been shifted from line A by +0.5 TECU for the purposes
of clarity.) The current day's multipath template is applied. Again, it can be concluded that
REALTEC is able to generate good quality TEC estimates (when compared to post-
processed TEC estimates) in real-time, even though there are multiple cycle slips in the raw
data.

CONCLUDING REMARKS
In order to improve the accuracy of real-time ionospheric delay estimation an integrated
approach to the processing of phase and pseudo-range data was developed. On the basis of
several tests, it appears feasible that carrier phase derived TEC "failure", caused by cycle
slips, can be detected and repaired in real-time. Further tests to refine the algorithm are
necessary.

299
 PRN 21
-2
-3
"

-4

-5 I
5'
u -6
w
I
c.
u
I
-7
w
l-
I
e
('II
-8 I
C?i ~I
-9
el/

-12~----~----~----~----~----~----~--~
o 1000 2000 3000 4000 5000 6000
Epoch (sec)

Figure 2. Post-processed versus real-time TEC estimates for PRN21 , for day 119, 1995.

ACKNOWLEDGEMENTS
The post-processing cycle slip detection and repair software was kindly provided by Dr.
Han (UNSW). The first author would also like to acknowledge the financial support of the
Government of Taiwan, Republic of China.

REFERENCES
Engler, E., Sardon, E., Jakowski, N., Jungstand, A. & D. Klahn, 1995. Real-time monitoring of the
ionosphere, Proc. 1995 IGS Workshop, Postdam, Germany, May 15-18,67-76.
Gelb, A., 1974. Applied Optimal Estimation., The M.I.T. Press, Cambridge, Massachusetts, 374 pp.
Han, S., 1995. Ambiguity recovery for GPS long range kinematic positioning, 8th Int. Tech. Meeting of
the Sat. Div. of the U.S. Inst. of Navigation, Palm Springs, California, Sept. 12-15, 349-360.
Hofmann-Wellenhof, B., Lichtenegger, H. & J. Collins, 1994. Global Positioning System: Theory and
Practice, 3rd edition, Springer-Verlag Wien, New York, 355 pp.
Lin, L.S., 1997. A novel approach to improving the accuracy of real-time ionospheric delay estimation
using GPS, 10th Int. Tech. Meeting of the Sat. Div. of the U.S. Inst. of Navigation, Kansas City,
Missouri, Sept. 16-19, 169-178.
Lin, L.S., Rizos, C. & YJ. Wang, 1996. Real-time estimation of ionospheric delays using GPS, Proc.
1996 Int. Conf. on GPS, Taipei, Taiwan, June 12-13, 117-127.
Mertikas, S.P. & C. Rizos, 1998. Real-time failure detection in the carrier phase measurements of GPS by
robust and conventional Kalman state estimates, Marine Geodesy, Vo1.21,41-65.
Wang, Y.J. & K.K. Kubik, 1993. Robust Kalman filter and its geodetic applications, Manuscripta
Geodaetica, 18,349 - 354.

300
GPS LEVELLING RESULTS FROM TWO TEST AREAS IN FINLAND

Matti Ollikainen
Finnish Geodetic Institute
Geodeetinrinne 2, FIN-02430 Masala, Finland
E-mail: [email protected]

Abstract

GPS levelling was used in two test areas in Finland in order to investigate the capability of
GPS measurements for orthometric height determinations. Both test areas are surrounded
by the spirit levelling loops of the Precise Levelling network. Besides, the test areas are
covered by a dense network of lower order levelling lines that are connected to the Precise
Levelling network. The GPS networks measured in both areas consisted of 51 and 45 GPS
sites that were located at spirit levelling benchmarks. The average side length was 15 km.
Four different geoid models, two global models (OSU91A, EGM96) and two local mod-
els (NKG89, FIN95), were used to convert the GPS ellipsoidal heights to orthometric
heights, after which the results were compared to the spirit levelled heights. In all cases the
height differences showed a clear tilt which was removed by fitting first and second order
surfaces to the differences. When the global geoid models were used for the conversion,
the RMS of the height differences varied from ±40 mm to ±80 mm. The RMS of the
differences was reduced to appro ±15 mm when the local geoid models were used. When
the errors caused by the spirit levelling in the computed height differences were removed,
the estimation of the average accuracy of the GPS levelling on the test areas gave the result
±12mm.

Research area

Two areas limited by 2 levelling loops of the Precise Levelling Network were chosen as
research areas. The size of the first area, observed with GPS in 1994, is about 9000 km 2,
and its location is between longitudes 24.8°E and 27.00E and latitudes 60.0oN and 61.00N.
The second area, where GPS observations were performed in 1995, is located on the
southwestern coast of Finland, between longitudes 21.8°E and 24.0oE, and latitudes
60AoNand 61.5°N; the size of the area is about 10000 km2 .
 The height differences in both areas are small;
the heights of the GPS sites vary from 0 to 120 m
in Area I and from 7 to 122 m in Area II.
We tried to choose the research areas such that
the shape of the geoid would be different in the
two areas. The shape of the geoid is reasonably
flat throughout the country; the steepest slopes are
located in northern Finland, but in these areas the
levelling network is sparsest. The areas chosen
for research are surrounded by Precise Levelling
loops, while a dense network of lower-order
levelling lines occurs inside the loops. In Area I N
varies 15 - 19 m running east to west, but in Area
II 18.8 - 19.6 m.

22 24 26 28 30
Spirit Levelling lines
Fig. 1 Location of the research areas
in Finland.
The Precise Levelling loops limiting the areas
discussed have been levelled three times. In Area Area I levellin lines
I, which is surrounded by the levelling Loop ill, 6760

the epochs were 1899.7, 1937.7, 1979.3 for the

First, Second and Third Precise Levelling, respec- 670110

tively. In the second research area, Area II, 6720

surrounded by the levelling Loop VI, the

respective levelling epochs were: 1902.6, 1946.6 6700

and 1981.4. According to the three precise Gseo

levellings the postglacial land uplift can be
380 420 440 460 '80 500
computed and the height of the levelling
benchmarks can be reduced to any epoch, e.g. to
the epoch of the GPS observations discussed here.
The accuracy of precise levelling was evaluated,
according to misc10sures of the levelling loops in
southern Finland, to be ±0.73 mml..Jkm in the
First, ±O.42 mml..Jkm in the Second and
±0.88mml.Jkm in the Third Precise Levelling,
respectively (KAKKURI and VERMEER 1985). In
addition to the Precise Levelling Network, several
lower-order levelling lines occur inside the
Precise Levelling loops. These lines have been
levelled by the National Land Survey (NLS). The
levelling accuracy performed by NLS is according Fig. 2 Levelling lines in Area I and II.
to (OKSANEN 1991) ±0.5-±0.8 mml.Jkm in Class o Precise Levelling line
Ib and ±1.5 mml.Jkm in Class II levellings. The + Class Ib and II levelling lines
levelling lines are shown in Fig. 2. o GPS site

302
Geoidal heights

A total of four different geoid models were used to

convert the GPS ellipsoidal heights to orthometric
heights e.g. OSU91A (RAPP et al. 1991), EGM96,
NKG89 (FORSBERG 1990) and FIN95 (VERMEER 1995).
As an example of the geoidal height in the research
areas, the geoidal heights according to the FIN95 geoid
are shown in Fig. 3.
4.2 4.4 4.6 4.8 5

GPS observations

The GPS sites observed in Areas I and IT are shown in

Fig. 2. The number of sites observed in Area I was 51,
and the area of the network was approx. 9000 km 2 or
one GPS site per 158 km2; the average distance between
stations is thus 13 km. The size of Area IT was approx.
10000 km2, and the number of sites was 45, or one GPS
site per 222 km2; the average distance between sites was
15 km.
The GPS observations were made using 6 Ashtech Z-
12TM type receivers. In Area I the observations were per- Fig. 3 Geoidal heights of the FIN95
formed during a period of 6 days. The lengths of the geoid in Area I and Are II Contour
observation sessions were two hours only. In Area IT, the interval: 0.1 m.
distances between observation sites were slightly greater
than in the first area, which is why the observation
sessions were prolonged to three hours. In Area IT the
observations were performed during a period of 8 days.

Solution of the GPS observations

Table 1. The strategy used in GPS-
The GPS observations were processed using two
solutions with the Bemese software.
software packages: GPPS Ver 5.0 software (ASHTECH
1993) and Bemese Ver 3.5 software (ROTIIACHER et al. o L3, ionospheric-free solution of
the baselines.
1993). 8 Evaluation of the ionospheric
The final solution with the GPPS software was models.
obtained using Precise Ephemeris obtained from «) Wide-lane solution ofL5

IGS/Center for Orbit Determination in Europe (1994- ambiguities.

95), the Lc option (ionospheric-free solution), and o L1&L2 solution of L1 and L2
ambiguities
widelane solution. All possible baselines were solved in e Final L3 solution introducing
each session, and finally all baselines with fixed the ambiguities from the
ambiguities were adjusted by the FILLNET program previous step.
(ASHTECH 1992). The strategy used in GPS solutions o Troposphere zenith corrections
with the Bemese software is presented in Table 1. were not estimated.

303
Comparison of the GPS solutions

Solutions with GPPS were undertaken to obtain preliminary coordinates for the stations.
To determine the real accuracy obtainable with a commercial software package, the GPPS
coordinates were compared with the Bemese solutions. The largest height differences be-
tween the GPPS and Bemese solutions are -20 mm. The RMS of the differences is ± 4 mm
in Area I, and ± 5 mm in Area II.

Conversion of ellipsoidal height differences to orthometric height differences

The orthometric height differences between the benchmarks were computed according to
the following formula:

(1)
where llh GPS is the ellipsoidal height difference observed by GPS and MY the geoidal
height difference. To determine the orthometric heights obtained with GPS levelling
(HGPs ) in the same reference frame as the levelled orthometric heights (HLev) the height
differences were added to the orthometric height of the initial benchmark, Ho:
H GPS = Ho +llHGPS (2)
The height differences obtained with GPS were reduced for the land uplift from the epochs
of the observations to the epoch of the spirit levelling (1960.0) using the following
formula:
dH u = (1960 - T) . A (3)
where T is the epoch of GPS observations and Athe land uplift (mgpu/year). Avalues were
obtained from the adjustment of the consecutive Precise Levellings of the loops surrouding
the research areas.

Comparison of GPS-Ievelled heights with spirit-levelled heights

The GPS levelling results obtained with four different geoids were compared with the
spirit levelled heights. The height differences computed showed in all cases a clear tilt,
which is why the systematic part of the differences was removed by fitting first- and
second-order surfaces to the differences. The following conclusions may be drawn
according to the RMS of the residuals of the plane fit:
The RMS values of the residuals clearly describe the errors in the geoidal heights used.
In both areas the best result was achieved by using the PIN95 geoidal heights. When the
FIN95 geoid was used, the RMS of the residuals was ± 17 mm in Area I, and ± 14 mm in
Area II. The result obtained with the NKG89 geoid was very good, « ± 20 mm on both
areas), but the global models resulted in RMS values that varied from ± 40 mm to ± 80
mm. The RMS values of the residuals are shown in Figs. 4-1 and 4-2.

304
Discussion of the accuracy
Order
- - - - - - - - 1 of
Height differences (Mf) between GPS levelling - - - - - - - - 1 poly.
and spmt levelling results contain errors !7-o. - - -- =- _ l nomlal
fil
originating from three main sources: GPS .0
determinations ((J h)' spirit levelling ((J H) and 01s1
o 2nd
geoidal heights ((J N ). A minor error ((J u) is
OSIl91A EGM 96 NKGa9 F1N9S
caused by the land uplift reduction which is
included in the height differences. We estimated Fig 4-1 RMS of the height differences in mm
the different error components according to the between GPS and spirit levelling in Area I.
standard deviations of different adjustments. (J h is
slightly dependent on the distance from the initial
Ord",
point of the network. With linear regression we 11:::: - - - - - - - - - 1 of
obtained the following formula: _ - - - - - - - - - 1 nomial
poly.

(J h[mm] = ± (S + 0.02· d [km]) (4) . _- - - - - - . , fit

For estimation of (JH we performed an common

adjustment of the Precise Levelling and lower 0SU91 A EGM 96 NKGa9 Fl'l9S
order levelling data in both research areas. These
adjustments resulted in the common value ± Fig 4-2 RMS of the height differences in mm
O.S mnI.JkID for both research areas. The between GPS and spirit levelling in Area II .
adjustment of the Precise Levelling loops gave for
(J u the value ± 0.02 rrrrV,Jki;vy. 18 ~------~

16
(J N was estimatd according to the rule of thumb
~
~ 14
derived by VERMEER (1995) for the FIN95 geoid: -= 12
• GPS
--Levelling
~ 10
(J N = 0.4· d . (J ~g .Jln (D / d) (5) ..
'2 8
~ 6
-+- Geoid
- x - Uplijt
- Total
where (J ~g is the accuracy of prediction of gravity iii 4
2
anomalies [2 mGal], d is gravity point spacing [5 o#--~--~-~
so 100 150
km] and D is the spacing of the GPS sites used in DIstance trom the lritial poinl [km]

the fitting of the FIN95 geoid [-100 km].

Fig. 5-1 Error budget of height differences
The error components may be summarized as
HGps·HLev
follows:
(6) 14 r--------~

E 12
The errors caused by spirit levelling do not, £ 10 t~~:-~~------
g
., a
however, affect the actual GPS levelling and may '2 6
not be considered in the error budget, i.e. ~ 4
'"
iii 2
estimation of the GPS levelling errors ((J~( GPS) )
may be done according to the formula:
O~-~--_+--_l

o 50 100 150
Distance from the Initial point (km]
2
(J H(GPS) = (J h2 + (J 2N (7)
Fig. 5-2 Error budget of GPS levelling.

305
Conclusions

According to the error model the following rule of thumb was derived:
• The accuracy of the GPS levelling over moderate distances in Finland is approx. ± (9
mm + 0.03 mmlkm).
Comparison of the GPS levelling accuracy with the spirit levelling accuracy gives some
clear answers to the following questions:
• The Precise Levelling cannot be replaced by GPS levelling.
• The lower order levelling (accuracy - ± 1.5 mml.Jkm) could be replaced by GPS level-
ling on lines that are longer than 50 km.
• If the spirit levelling accuracy is lower than ± 2 mm1.Jkm , better results are probably
obtained by GPS levelling.

Acknowledgements

I wish to express my sincere thanks to Mssrs. SEPPO OKSANEN and MATTI MUSTO, Na-
tional Land Survey, for the spirit levelling data observed by NLS and to Professor Dr.
TEUVO PARM, Helsinki University of Technology for loaning two of the GPS receivers
used in the fieldwork.

References

ASHTECH, Inc. (1992): FILLNET, GPS-adjustment program, version 3.1. Sunnyvale, CA,
USA.
ASHTECH, Inc. (1993): ASHTECH XII GPPSTM, The GPS Post-Processing System Man-
ual. Software version 5.0.00, Document Number 600196, Rev. A. Sunnyvale, CA, USA.
FORSBERG, R. (1990): A new high resolution geoid of the Nordic Area. Proc. lAG Sym-
posium 106: "Determination of the Geoid, Present and Future". (Eds. R. Rapp, F.
Sanso). Springer Verlag. New York.
KAKKURI, J. and M. VERMEER (1985): The study of land uplift using the Third Precise
Levelling of Finland. Rep. of the FGI, No. 85: 1. Helsinki.
OKSANEN, S. (1991): Geodetic Operations of the National Board of Survey. In Geodetic
Operations in Finland 1987-1991. (Ed. J. Kakkuri). FGI. Helsinki.
RApp, H.R., Y.M. WANG and N.K. PAVLIS (1991): The Ohio State 1991 Geopotential and
Sea Surface Topography Harmonic Coefficient Model. Report No. 410, Department of
Geodetic Science and Surveying, The Ohio State University, Columbus, Ohio.
ROTIIACHER, M., G. BEU1LER, W. GURTNER, E. BROCKMANN and L. MERVART. (1993):
Bernese GPS software, version 3.4, documentation May 1993. Bern.
VERMEER, M. (1995): Two New Geoids Determined at the FGI. Rep. of the FGI, No.
95:5. Helsinki.

306
 QUALITY CONTROL ALGORITHMS FOR PERMANENT GPS
RECEIVER APPLICATIONS

Chris Rizos
Lao-Sheng Lin
ShaoweiHan
School of Geomatic Engineering, The University of New South Wales
Sydney NSW 2052, Australia
Lienhart Troyer
Department of Engineering Geodesy, Technical University Graz
8010 Graz, Austria
Stelios Mertikas
Department of Mineral Resources Engineering, Technical University of Crete
GR-73 100 Chania, Crete, Greece

ABSTRACT

There has been a burgeoning of activity involving the establishment and operation of
permanent GPS stations. The variety of applications is broad, from geodetic reference
sites in support of IGS initiatives, to national base stations for real-time differential GPS,
"integrity monitoring" and datum maintenance, and even local GPS base stations
providing data to surveyors and other users. The level of sophistication of such permanent
stations also varies, from systems capable of automatic data collection, transmission and
sometimes processing, to relatively simple configurations. All applications, however,
share a concern for GPS data "quality". The detection of cycle slips, data "spikes",
multipath disturbances, and bad data sequences of GPS measurements is a major problem
for many applications, even more so in the case of real-time applications.
Some quality control (QC) strategies may be implemented either on a single receiver
basis or using GPS network data. This paper discusses the issue of QC for permanent
GPS receivers, and comments on aspects of QC which require investigation.

QC AND PERMANENT GPS RECEIVERS

Concerns about GPS "quality" are shared by all users, from those engaged in the most
precise geodetic applications through to the casual navigator. The quality of GPS
positioning is dependent on a number of factors. Experience with precise geodetic
applications of GPS has shown that sophisticated mathematical modelling, careful field
procedures and top-of-the-line GPS hardware are all necessary prerequisites, however great
care still has to be applied to ensure that data quality is uniformly high. "Quality control"
(QC) must be applied to both receivers, the one located at the "reference" site and the
second GPS receiver, the one the user is generally interested in. Increasingly the former
may be a permanent GPS receiver operated by an agency or organisation on behalf of a
wide variety of users, while the latter may be in motion and hence experiencing very
different conditions. Although QC for such a "roving" receiver is a significant challenge,
the quality of the data from the permanent reference receiver must also be assured. This
paper deals exclusively with this issue and hence is concerned with single (reference)
receiver QC.

Quality Management Issues

To date the procedures of data screening, computations, result evaluation and quality
assurance have generally been an off-line (as well as iterative) process. However, with the
development of precise "on-the-fly" GPS positioning, and other real-time techniques, it is
no longer possible to analyse (and re-analyse) GPS data in post-mission mode until GPS
result quality is assured. The challenge therefore is to develop quality control (QC) and
quality assurance (QA) procedures that can also be applied in real-time (or "near-real-
time") GPS applications during or shortly after the measurement process.
Quality management, in the context of GPS applications, is concerned with assuring an
agreed level of accuracy and reliability for GPS results. The focus is therefore on
procedures for defining, measuring and verifying quality. The language of "quality
management" is replete with terms such as QC and QA. Although the two terms are often
considered synonymous, a good working definition is that: (a) QA refers to the set of
practices and procedures which are intended to maximise the chances that the product or
service will satisfy the client's requirements, at a reasonable cost, while (b) QC refers to
the procedures used to verify the level of quality achieved and, if it is unacceptable, to
detect the source of the problem and, if possible, remedy it. The QC/QA practices and
procedures include:
• Procedures for hardware procurement -- if the GPS instrumentation is inappropriate
or somehow faulty, how can the data be expected to be of consistently high quality?
• Procedures for software verification -- does the software generate "correct" results?
• Algorithm definition and testing -- the data processing algorithm plays a significant
role in ensuring high quality results, as it may be, e.g., more or less "robust" in
relation to varying quality input measurements.
• Procedures for site selection -- local environmental conditions may cause multi path
and signal interruption problems.
• Procedures for instrumental installation -- e.g. quality of groundmark, ruggedness
of cabling, efficiency of power sub-systems, etc.
• Check procedures -- how is the overall system quality being monitored (are there
backup systems, check surveys, quality indicators, etc. ?), is it being carried out on a
continuous or periodic basis?

lAG Initiatives
The International Association of Geodesy (lAG) established the Special Study Group
(SSG) 1.154 1 on "Quality Issues in Real-Time GPS Positioning" in 1995 to identify
practical procedures, as well as mathematical techniques, that can be applied to assure the
quality of positioning results obtained from real-time GPS applications. There are several
characteristics that distinguish these applications from others:
• the communication of data from GPS receivers to a computing site where may occur
transmission interruption, either recovering by duplicated encoding procedure or re-
transmitting,

1 The first author is Chairman of this SSG, while the third and fifth authors are members.
Web site URL for this SSG is http://www.gmat.unsw.edu.au/ssg_RTQC/.

308
 • data latency due to the transmission, where it is processed with no, or minimum,
delay,
• make use of carrier phase data, in addition to pseudo-range data,
• rely on data processing on an epoch-by-epoch basis, or at the very least small
"batches" of GPS data,
• do not permit extensive data "pre-processing" or the review of data and results in
iterative procedures, and
• may involve kinematic or static positioning.

Poor quality GPS measurements makes the task of assuring high quality GPS results very
difficult, and perhaps impossible. The data from a permanent reference receiver must
therefore be assured before it is used in any computations.

QC/QA OF GPS MEASUREMENTS

QC/QA procedures must contend with those factors and influences that affect
measurements. Restricting our interest to permanent GPS receivers we have the following
problems to contend with:
• Measurement "failure" requiring isolated measurements to be discarded.
• Measurement "failure" that causes cycle slips in carrier phase observations, which
can be repaired.
• Variable data quality arising from multipath disturbances, which must be mitigated
at the single receiver level if at all possible.
• Antenna phase centre variation, which is an antenna-dependent error, which should
be corrected for if use is to be made of different kind of GPS receivers,
• Variable data quality due to unusual atmospheric effects such as ionospheric
scintillation.

Many procedures and processing strategies have been developed to account for such
effects, both mathematically and empirically, however we will only mention several that
can be applied directly to the raw data from a single receiver capable of making dual-
frequency carrier phase (<1>1, <1>2 in radians) and pseudo-range (R 1, R 2 in metres)
measurements, as is typical of modem GPS reference station receivers.

Failure in GPS Measurements

By "failure" is meant changes in the geometric (or range) information contained within the
measurements that are: (a) of unknown magnitude, and (b) occur at unknown instants of
time. A failure in GPS measurements includes cycle slips in phase data or outliers in
pseudo-range date. Such failures have usually been detected during processing, or pre-
processing, of double-differenced (i.e. baseline) data. A repair procedure may then be
applied, or the data simply flagged "bad" during residual testing, and the amended dataset
is ready for processing. The procedure is applied iteratively until the process converges.
A procedure for scanning the raw observation data based on using a Kalman filter to
model the behaviour of phase and phase-rate measurements (and their changes) in discrete
time is described in, for example, Cannon et al. (1992), Mertikas & Rizos (1998). This
procedure can be applied to each of the data types (<I> 1, <1>2, R 1, R2) separately. The fact
that the GPS receiver is stationary is an important constraint, making it a much easier task
than if the receiver were moving. Following the detection step, a subsequent procedure
may be applied to repair cycle slips in the phase data. Outliers in the pseudo-range data
cannot be repaired and must be eliminated from the dataset. Such an algorithm's
application to the problem of determining Total Electron Content (TEC) from single

309
receiver measurements which may be contaminated with cycle slips is reported in, for
example, Jungstand et aI. (1995), Lin (1997a), and others.
Dual-frequency data offers especially rich opportunities to construct combinations of
observables (phase-only, pseudo-range-only, as well as phase and pseudo-range) (e.g.,
Hofmann-Wellenhof et aI., 1994; Rizos, 1997) which may be screened using a number of
procedures based on Kalman filters of various types, trend-following polynomials, digital
filters, and so on. The following sequences are typically used:
• One-way residual sequences, including receiver clock bias, satellite clock bias, SA
effect, atmospheric biases, multi path and observation noise.
• The difference sequences (1..1. <1>1 - 1..2. <1>2) between L 1 and L2 carrier phase data in
units of metres, which includes ionospheric delay and ionospheric scintillation, and
cycle slips.
N.. =' 9240(i+ j)+289i R 9240(i+ j)+289j R
• I,j I.
<I> . <I>
1+ J. 2
-
2329.1..1 . 1+ 2329.1.. 2 . 2
especially the wide-lane ambiguity sequences (i=1 , j=-1), where 1..1 and 1..2 are the
wavelength of the Ll and L2 carrier phase, i and j are integers. These sequences are
affected by ionospheric scintillation, cycle slips and pseudo-range biases.
• Multipath on pseudo-range data can also be estimated, though the estimates are
affected by cycle slips, ionospheric scintillation and pseudo-range biases.
• Single-differencing (between satellites) may be applied to the above sequences, so
as to eliminate the receiver clock bias.

Multipath at Permanent Receiver Sites

Multipath is a signal disturbance arising from the fact that the signal entering the GPS
antenna, in addition to containing the direct satellite-receiver component, also includes
reflections from buildings, water surfaces and the ground. Multipath and diffraction
effects cannot be easily accounted for during data processing. Fortunately the mUltipath
error on carrier phase observations is significantly less than that experienced on pseudo-
range data (of the order of several centimetres, compared with metre level disturbance on
pseudo-ranges). Furthermore, its effect tends to average out for static baseline
determinations with observation sessions of the order of an hour or more. Nevertheless,
for the highest precision static and kinematic applications the effect of multipath
disturbance must be addressed.
The multipath component in the Ll and L2 pseudo-range data can be estimated (e.g.,
Hofmann-Wellenhof et aI., 1994; Rizos, 1997). If the Ll and L2 carrier phase is
measured, the multi path in the pseudo-range data can be estimated. Obviously this
approach cannot be used with single-frequency data.
The multipath error in carrier phase cannot be estimated from the raw measurements on
a single receiver basis, but may be estimable on a baseline basis from the double-
differenced residuals after baseline processing.
In the case of a permanent receiver, the geometry of the satellites, relative to the receiver
and surrounding reflective objects, is almost exactly the same after one sidereal day.
Hence the multi path disturbance tends to exhibit a daily signature, both in the raw
measurements and in the baseline residuals. So-called "multipath-templates" can be
constructed to correct pseudo-range measurements, from an analysis of the past one or
more day's data. This technique is applied to precise pseudo-range data at permanent GPS
sites used, for example, to estimate atmospheric parameters such as TEe (Lin, 1997b).
In the case of carrier phase, multipath elimination could be achieved by using a filter
tuned to the appropriate frequency/amplitude characteristics of the disturbance. Because

310
of the daily multipath signature at permanent GPS receivers the filter is easier to "tune"
than under other circumstances. A Finite Impulse Response (FIR) lowpass filter has been
proposed by Han & Rizos (1997).
However, apart from mathematical procedures there are several strategies for
overcoming the problem of multipatb in the observations at permanent GPS receivers:
• Careful selection of site in order to minimise the multi path environment.
• Use multipath resistant antennas.
• Use special receivers that contain some form of "multipath elimination technology".
• If it is possible, multipath should be "corrected" at each receiver.

Ionospheric Scintillation
The ionospheric delay on GPS measurements are subject to spatial and temporal
variations. The spatial variations are usually low in frequency and generally correspond to
the various ionospheric latitude zones: tropical, mid-latitude and auroral. The temporal
variations can have high frequencies (of the order of minutes to hours), medium frequency
(diurnal and seasonal effects), and low frequency (the 11 year solar cycle). TEC is a
maximum at low latitudes (the tropical zone) and at the poles (the auroral zone), and is a
minimum at mid-latitudes. At night the ionospheric delay is approximately five to ten
times less than for day time observations. The diurnal cycle for TEC is such that the
maximum occurs two hours after solar noon, and is a minimum before dawn. Ionospheric
disturbances, which can occur suddenly and be very severe, affect the amplitude and phase
of GPS signals (Wanninger, 1993; Knight & Finn, 1996). One of the phenomena
responsible for these are "travelling ionospheric disturbances", another is due to
irregularities in the ionosphere causing "scintillations" (especially in the tropical and
auroral zones). Under such conditions the ionosphere is so perturbed that single frequency
operations may become impossible because the GPS receiver loses lock on the satellite
signals. Where tracking is possible, the likelihood of cycle slips and interrupted tracking
is increased, both of which, for example, make ambiguity resolution a more difficult and
unreliable task.
Knight & Finn (1996) describes an algorithm for determining the so-called S4
"scintillation index". Empirical filtering techniques will need to be developed to cope
with such effects in real-time, particularly when the next solar cycle maximum occurs at
the turn of the century.

Other Biases Dependent on Receivers and Antennas

There is an offset bias which is dependent on the antenna used due to phase centre offset
and variation. The electrical phase centre of the antenna is the point to which the radio
signal measurement is referred and generally is not coincident with the physical antenna
centre. The offset depends on the elevation, azimuth, and intensity of the satellite signal,
and is different for Ll and L2. The antenna phase centre offsets for different kinds of GPS
receivers relative to the Turbo Rogue receiver with the Dorne Margolin T antenna have
been determined by the International GPS Service for Geodynamics (IGS report available
from http://igscb.jpl.nasa.gov/, 1996). The offset components in the north and east
direction can reach 5mm. The antenna component in height can reach 3cm, and changes
significantly with elevation angle.
If the reference GPS receivers are generating data for different kinds of "mobile" GPS
receivers, the antenna offsets should be considered carefully when seeking to achieve
millimetre accuracy. Therefore, GPS antenna offset determination and correction will
become necessary for precise GPS positioning.
The bias which is dependent on the type of GPS receiver is due to the manner in which
different receivers internally process the carrier phase signals. A zero baseline experiment

311
has been performed by Braun & Rocken (1995) using the same antenna (Trimble 4000ST
L lIL2 Geod antenna) and two different types of GPS receivers from the same
manufacturer (Trimble 4000SST and Trimble 4000SSE). The results indicate that vertical
height error can be up to five centimetres, even when using one hour observation spans.
Similar results were reported by Brunner & Tregoning (1994). More research is necessary
if millimetre accuracy positioning using mixed GPS receivers/antennas is to become
routine.

CONCLUDING REMARKS

Ideally what is required is an algorithm, or "cocktail" of algorithms, capable of automatic,

real-time operation, on a single site basis. They must be able to delete obviously "bad
data", repair failures such as carrier phase cycle slips, and mitigate multipath in pseudo-
range and phase data.
However, the quality of the data either before or after QC procedures are applied cannot
be characterised in a simplistic "black and white" fashion. Even after the application of
such an algorithm (or algorithms) the data quality will likely be on a "grey scale", ranging
from "bad" (but not so "bad" that it is considered an outlier) to "good" inferring a phase
(or pseudo-range) noise (typically measured as a measurement variance) that is not
constant. Perhaps what is required is a measure of data quality, such as an index between
1 and 5, that ensures that the measurement stochastic model is realistic.
There is still considerable R&D necessary to address this issue, even for the relatively
benign QC/QA circumstances of permanent, stationary GPS reference receivers.

REFERENCES

BRAUN,1. & C. ROC KEN, 1995. Vertical height errors when mixing Trimble 4000SST and Trimble
4000SSE observations, UNAVCO report at http://www.unavco.ucar.edu.
BRUNNER, F.K. & P. TREGONING, 1994. Investigation og height repeatability from GPS measurements,
Aust. J. Geod.Photo.Surv. , 60, 33-48.
CANNON, M.E., SCHWARZ, K.P., WEI, M. & D. DELIKARAOGLOU, 1992. A consistency test of
airborne GPS using multiple monitor station, Bull. Geod., 66, 2-11.
HAN, S. & c. RIZOS, 1997. Multipath effects on GPS in mine environments, Xth Int. Congress of the Int.
Soc. for Mine Surveying, Fremantle, Australia, 2-6 November, 447-457.
KNIGHT, M. & A. FINN, 1996. The impact of ionospheric scintillations on GPS performance, 9th Int.
Tech. Meeting of the Sat. Div. of the U.S. Inst. of Navigation, Kansas City, Missouri, Sept. 17-20,555-
564.
HOFMANN-WELLENHOF, B., LICHTENEGGER, H. & COLLINS, J., 1994. GPS Theory and Practice,
Springer-Verlag, Vienna New York, 3rd ed., 355pp.
JUNGSTAND, A., ENGLER, E., SARDON, E. & KLAHN, D., 1995. Error separation concept in
experimental TEC monitoring network, U.S. Inst. of Navigation National Tech. Meeing, Washington,
D.C., 323-335.
LIN, L.S., 1997a. A novel approach to improving the accuracy of real-time ionospheric delay estimation
using GPS, 10th Int. Tech. Meeting of the Sat. Div. of the U.S. Inst. of Navigation, Kansas City,
Missouri, Sept. 16-19, 169-178.
LIN, L.S., 1997b. Real-time estimation of ionospheric delay using GPS measurements, PhD dissertation,
School of Geomatic Engineering, The University of New South Wales, Sydney, Australia.
MERTIKAS, S.P. & c. RIZOS, 1998. Real-time failure detection in the carrier phase measurements of GPS
by robust and conventional Kalman filtering, Journal of Marine Geodesy, VoL21, 41-65.
RIZOS, c., 1997. Principles and practice of GPS surveying, Monograph 17, School of Geomatic
Engineering, The University of New South Wales, ISBN 0-85839--071-X, approx. 56Opp.
WANNINGER, L., 1993. Effects of the equatorial ionosphere on GPS, GPS World, 4(7), 48-54.

312
 FAST AMBIGUITY RESOLUTION IN NETWORK MODE

P.J.G. Teunissen, P.J. de Jonge, D. Odijk, C.C.J.M. Tiberius

Delft University of Technology, Faculty of Geodetic Engineering
Thijsseweg 11,2629 JA Delft, The Netherlands

Abstract

This contribution presents and discusses practical results of network-based instantaneous

ambiguity resolution. It is shown that high-dimensional integer solutions can be computed
efficiently with current ambiguity resolution methods. For that purpose the least-squares
ambiguity decorrelation method has been integrated in the network analysis. We also
introduce a new ambiguity dilution of precision measure to express the intrinsic precision
of the ambiguities and show why it is preferred over the more traditional measures.

Measurements and model

The experiment on which our data analysis is based, uses a small network of four points.
The experiment took place on December 22nd , 1996, about 80 km North-East from Delft.
Four dual frequency receivers were used (Trimble 4000 SSI Geodetic Surveyor). Two of
the receivers were placed stationary (points 15 and 28), about 12.7 km apart. The antennas
of the third and fourth receiver were mounted on the roof of a van (points 38 and 39).
The data were collected at a one second sampling rate, during several sessions; each
session took about 50 minutes. In the following we will show the results of only one
session. These results are representative for all sessions. To get an impression of the
dynamics involved, figure 1 shows a typical example of the velocity curve.
With the exception of some interruptions, the same 7 satellites (PRN's 02, OS, 07, 09,
21, 23 and 26) were tracked by three of the four receivers. The fourth receiver (point 39)
failed to track satellite PRN 26. All satellites were tracked at an elevation angle larger than
10 degrees. The corresponding skyplot is shown in Figure 2.
The linearized system of DD observation equation:~ of epoch i can be cast in the matrix-
vector form as y(i) = Aa + B(i)b(i) + e(i), i = 1, ... , k, where the vector y(i) consists of the
observed minus computed dual frequency phase and code data of epoch i, a is the time-
invariant vector of unknown integer DD ambiguities, b(i) is the unknown increment vector
of the remaining parameters (such as network coordinates and possibly ionospheric delays)
at epoch i, e(i) is the noise vector and A and B(i) are the appropriate design matrices.
Our analysis is based on using dual frequency phase data and single frequency C/A code
data. For the undifferenced observables, the a priori standard deviations were set at 30 cm
for the code data and at 3mm for the phase data. Time correlation was assumed absent.
Due to the relatively short baselines and small height differences between the points, the
tropospheric and ionospheric delays were initially assumed absent as unknown parameters.
A priori corrections for the tropospheric delays were applied on the basis of the
Saastamoinen model.

..,. >000

Figure 1: Velocity curve of van Figure 2: Skyplot 12-22-96 <p = 52°26'N A = 5°14'E 14:30 UTe

The data of the four receivers were directly adjusted in a network-mode, instead of using
a baseline-by-baseline approach. An example of a small-scale network with rapid static
GPS for crustal deformation analysis can be found in Genrich et al. (1997). The vector b(i)
consists of the coordinate increments of all receivers, but relative to the reference point 15.
The vector a consists then of the DD ambiguities belonging to all observed DD carrier
phases. A computationally efficient method enables fast and strict integer least-squares
estimation of the ambiguities. In the following, when we speak of ambiguity resolution,
we refer to solutions which were obtained by solving all ambiguities of the complete
network in one step. For our small network of four points this implies that the total
number of ambiguities equals 34 (=2*[7+7+6]-2*3). This number is based on using dual-
frequency data with three receivers tracking 7 satellites and the fourth receiver tracking
one satellite less.

Ambiguity resolution

The process of ambiguity resolution can conveniently be separated into two parts: (1) the
ambiguity estimation problem; (2) the ambiguity validation problem. The first part
addresses the problem of finding the least-squares estimates of the integer ambiguities.
The second part is concerned with the question whether one is willing to accept this
integer least-squares solution. Conceptually one can break the first part into three different
steps. In the first step one simply disregards the integer constraints. Hence the problem has

314
now become a standard least-squares solution. As a result one obtains the (real-valued)
estimates of a and b(i) (the so-called float solution), together with their variance matrices.
In the second step, the 'float' ambiguity vector and its variance matrix are used to solve for
the integer least-squares ambiguities. These integer ambiguities are then finally used in the
third step to obtain the 'fixed' solution of b(i).
The difficult part in the above procedure is the computation of the integer least-squares
ambiguities. The LAMBDA method provides a very efficient way of solving this problem,
see e.g. Teunissen (1993), de longe and Tiberius (1996) or Kleusberg and Teunissen
(1996). In short it consists of the following steps. First a local search region, the ambiguity
search space, is defined: (li - a)T Qil(li - a) ::; x2,a E Zn , with Ii the (real-valued) DD least-
t
squares ambiguity vector and where is an appropriate chosen constant. Before the actual
search is performed, the DD ambiguities and their search space are first transformed by
means of a decorrelating transformation. As a result one obtains transformed ambiguities
that are largely decorrelated, more precise than the original ambiguities and that have a
search space which is closer to a hyper-sphere. These properties make it possible to
t
perform the search in an efficient manner. For setting the value of one can make also a
t
good use of the high precision of the transformed ambiguities. The value should namely
be not too small but also not too large. A too large value would imply a search space with
an abundance of unnecessary grid points, while a too small value could result in an empty
search space. Once the size and shape of the search space are set, the actual search can
commence. It is based on a sequential conditional least-squares adjustment which results
in (sharp) bounds on the individual (transformed) ambiguities. This search will produce
the integer least-squares ambiguities and if needed for validation purposes, the second best
solution as well.
In our present application we were interested in the performance of instantaneous
ambiguity resolution using the network mode of operation. Hence for the time span of
about 50 minutes this amounts to about 3000 single-epoch solutions in which each
solution was based on a search within a 34 dimensional search space. Due to the efficiency
of the current integer least-squares procedures, this does not pose any difficulties from a
computational point of view. The computational cycle times for such single epoch
solutions are typically below the 0.2 sec level. This includes all steps after formation of the
normal equations. Hence it includes: (a) solving the normal equations; (b) construction and
application of the decorrelating ambiguity transformation; (c) the actual search; and (d) the
computation of the 'fixed' network parameters. Timing was done on a PC486-66Mhz.

Ambiguity dilution of precision.

For the validation of the ambiguities, we would like to get an impression whether
successful ambiguity resolution is to be expected. In many applications of GPS it is
already common practice to make use of various dilution of precision (DOP) measures.
These measures are based on applying the trace operator to relevant parts of the variance-
covariance matrices. For the ambiguities however, we believe that the trace cannot be
considered adequate. There are two main reasons for this: (1) the trace only operates on the
diagonal entries of the variance matrix; (2) the trace lacks some important properties of
invariance. The fact that the trace operates only on the diagonal entries of the variance
matrix would not be too harmful if the correlations were small to moderate and if they

315
only played a secondary role in the process of ambiguity estimation and validation. This is
not the case however. The DD ambiguities are usually highly correlated and these
correlations have an important impact in shaping the ambiguity search space (Teunissen,
1997a).
Also the lack of invariance of the trace, when applying members from the class of
admissible ambiguity transformations, makes the trace unsuitable. The trace of the
variance matrix remains namely only invariant when orthogonal transformation are
applied. But the admissible ambiguity transformations are not orthogonal. They only share
the volume preserving property with the orthogonal transformations.
Based on these considerations we make use of the determinant of the ambiguity variance
matrix. The ambiguity dilution of precision (ADOP) is therefore defined as
ADOP = 2-\jdetQa (cycle), where n is the order of the ambiguity variance matrix Qa (in
our case n=34). The ADOP is invariant for the class of admissible ambiguity
transformations. Thus the same value is obtained, irrespective of which satellite is chosen
as reference in the DD definition. The same value is also obtained when one uses the
variance matrix of the transformed ambiguities instead of the variance matrix of the
original DD ambiguities. Thus the ADOP better measures the intrinsic precision
characteristics of the ambiguities. It can be shown to approximate the (geometric) average
of the precision of the least-squares ambiguities.
The computation of the ADOP can be incorporated efficiently into the ambiguity
resolution process. The various options available together with the class of GPS models
for which closed form expressions exist, are given and discussed in (Teunissen, 1997a,b).
Figure 3 shows the ADOPs for the single epoch solutions over the complete time span of
about 50 minutes. With the exception of the five 'interruptions', the value stays constant at
the level of about 0.08 cycle. The five 'interruptions' are due to partial loss of lock of the
receivers on the van or due to short occasional loss of the L2 signal (obstacles). The small
values of the ADOP indicate that successful single epoch ambiguity resolution should be
feasible with the present set up.
O.16
l Validation.
01< ~ Various tests were executed in order to get an
indication of how well the single-epoch solutions fit
the model. These tests consisted of checks on the
internal and external consistency of the results. The
checks on the external consistency were possible for
the stationary baseline 15-28 (using given
coordinates) and to a somewhat lesser extent for the
0.06
1i periods that the van was stationary (using
L--~~C-::'00=--0~~-"-;2:::::000;:-----~~3000
epoch (III)
stationarity). For the internal consistency, we relied
on the one hand on the available redundancy of the
Figure 3: The single epoch based ADOPs
for the complete observation time span
single epoch solutions and on the other hand on how
these single epoch solutions compared to the overall
solution based on the data of the complete time span.
The overall picture that emerged from these comparisons is as follows: (1) the float
solutions, based on the model without the ionospheric delays included, performed

316
according to expectation; (2) again without the ionospheric delays included, almost all of
the computed integer ambiguities could be validated successfully; the success rate was
99.6% (of over 3000 single-epoch solutions); (3) the fixed solutions performed rather
poorly when the ionospheric delays were excluded, but improved considerably when the
ionospheric delays were included.
As a typical example of these results, the histograms of the single-epoch-based overall
model tests statistics are shown in Figure 4. The overall model test statistic is defined as
the ratio of the a posteriori variance factor and its a priori counterpart. It has a central F-
distribution F(r, 00, 0) under the null hypothesis, where r denotes the redundancy (note: for
the fixed solutions one has to make the additional assumption that the integer ambiguities
are non-stochastic). Two features can be seen from these histograms. First note the
'outlying' character of the second histogram. As pointed out above, its rather
discontinuous behavior is thought to be caused by the erroneous assumption that the
ionospheric delays were sufficiently small. This behavior is absent in the first and third
histogram. This indicates an interesting feature of integer least-squares estimation. As far
as the integer estimation process is concerned, the data are permitted to be corrupted with
some bias. That is, despite some biases in the float solution, the integer least-squares
process will still pull the float solution to the correct integer value. But once the integer
ambiguities are assumed fixed and are treated as such, the bias, still present in the data,
becomes predominantly visible in the fixed solution. The fact that these relative small
biases are not noticable in the float solution is due to the fact that, in the single epoch case,
the phase data fail to contribute. Hence, these solutions are dominated by the relatively
poor precision of the code data.
The second feature concerns the expectation of the distribution. Under the null
hypothesis it should to be equal to one. The two remaining histograms however show a
mean which is much smaller than one. It is our belief, which is also based on past
experience, that this is caused by the still too pessimistic assumptions about the stochastic
model.

... notlt tol\1111on - IOIlOlp"'.r. uclud4c1

000
fi lled _llIlIon .ono.plar~ I!u:ludt)(!

800 ~
rliudl IOhnllitn - ~l:Io~hu. nch.lal'd:

lOG
... 000 -

... .... •.. r

200

0
0
200

0
0
. ~.

3 .
200 -

0
0
Jbrr._ .
1 2 3

Figure 4: Histograms of the single epoch based overall model tests statistics

To conclude a typical example of the repeatability in the position domain is shown in

Figure 5. It shows the histograms of the about 3000 single epoch based coordinate
solutions of baseline 15-28. The top row refers to the float solution assuming the
ionospheric delays to be absent, while the bottom row refers to the fixed solution assuming

317
the ionospheric delays present. The float solution performs quite well with only a slight
bias in the East component. Note the well known GPS phenomenon that the height
component is determined worst, while - for mid latitude regions (<p=52°) - the East
component is determined better than the North component. The same can be seen for the
fixed solution. This solution shows however less homogeneity than the float solution
despite the fact that the ionospheric delays were solved for. This indicates that there are
still some small, unidentified biases left in the data. Nevertheless note the tremendous
improvement in dispersion when compared with the float solution. The improvement,
which is solely due to the integer ambiguity resolution, is about a factor of one hundred.

Float I East (r'rl] 200 [

Flo..1Up ]
t I[en] Figure 5: North, East, Up
'O .. t I North
200Fr n [m] 200
histograms of baseline 15-28: (top
1:;;0 150 150
100
row) float solution without solving
100 100
so SO 50
for the ionospheric delays; (bottom
o o _2 o
row) fixed solution with the
-2 0 2 o 2 _2 0 2
ionospheric delays solved for.
~~~[]J
ix~
ed I North [m] 200FU
lxed I East
J [m] 200 [
Fixed
j I UP
J I""']
150 150
100 100 100
SO SO SO
0_ 002 0 002 0 _002 0 0 .02 0 _002 0 002

Acknowledgements.
The contribution of the third author was done under contract with the Triangulation
Department of the Dutch Cadastre, and the fourth with the Survey Department of the
Rijkswaterstaat. Also the computational assistance of Peter loosten is acknowledged.

References
Genrich, 1.F., Y. Bock and R.G. Mason (1997): Crustal deformation across the Imperial
Fault: results from kinematic GPS surveys and trilateration of a densely spaced, small-
aperture network. Journal of Geophysical Research (102): 4985-5004.
longe de, P.I., C.C.I.M. Tiberius (1996): The LAMBDA method for integer ambiguity
estimation: implementation aspects. LGR Series, No. 12, Delft Geodetic Computing
Centre. Also available through the internet: http://www.geo.tudelft.nl/mgp/
Kleusberg, A, P.J.G. Teunissen (Eds.)(1996): GPS for Geodesy, Lecture Notes in Earth
Sciences, Vol. 60. Springer Verlag.
Teunissen, P.I.G. (1993): Least-squares estimation of the integer GPS ambiguities. Invited
Lecture, Section IV Theory and Methodology, IAG General Meeting, Beijing, China.
Also in: LGR Series No.6, Delft Geodetic Computing Centre.
Teunissen, P.I.G. (1997a): A Canonical Theory for Short GPS Baselines; Part IV:
Precision versus Reliability. Journal of Geodesy (71): 513-525.
Teunissen, P.J.G. (1997b): Closed Form Expressions for the Volume of the GPS
Ambiguity Search Spaces. Artificial Satellites, Vol. 32, No.1, pp. 5-20.

318
 HIGH PRECISION GPS KINEMATIC POSITIONING:
PROGRESS AND OUTLOOK

Chris Rizos & Shaowei Han

School of Geomatic Engineering, The University of New South Wales
Sydney NSW 2052, Australia

ABSTRACT
High precision GPS kinematic positioning in the post-processed or in the real-time mode
is now increasingly used for many surveying and navigation applications on land, at sea
and in the air. The distance from the mobile receiver to the nearest reference receiver may
range from a few kilometres to hundreds of kilometres. As the receiver separation
increases, the problems of accounting for distance-dependent biases grow and, as a
consequence, reliable ambiguity resolution becomes an even greater challenge. In this
paper, the challenges, progress and outlook for high precision GPS kinematic positioning
for the short-range, medium-range and long-range cases, in both the post-processing and
real-time modes, will be presented.

INTRODUCTION
The standard mode of precise differential positioning is for one reference receiver to be
located at a station whose coordinates are known, while the second receiver's coordinates
are determined relative to this reference receiver. In addition, carrier phase measurements
must be used to assure high positioning accuracy. However, the use of carrier phase data
comes at a cost in terms of overall system complexity because the measurements are
ambiguous, requiring that ambiguity resolution (AR) algorithms be incorporated as an
integral part of the data processing software.
Such high accuracy techniques are the result of progressive R&D innovations, which
have been subsequently implemented by the GPS manufacturers in their top-of-the-line
"GPS surveying" products. Over the last half decade or so several significant
developments have resulted in this high accuracy performance also being available in
"real-time" -- that is, in the field, immediately following the making of measurements, and
after the data from the reference receiver has been transmitted to the (second) field
receiver for processing. Real-time precise positioning is even possible when the GPS
receiver is in motion. These systems are commonly referred to as RTK systems ("real-
time-kinematic"), and make feasible the use of GPS-RTK for many time-critical
applications such as machine control, GPS-guided earthworks/excavations, automated
haul truck operations, and other autonomous robotic navigation applications.
CHALLENGES IN PRECISE GPS KINEMATIC POSITIONING
If GPS signals were tracked and loss-of-Iock never occurred, the integer ambiguities
resolved at the beginning of a survey could be kept for the whole GPS kinematic
positioning span. However, the GPS satellite signals are occasionally shaded (e.g. due to
buildings in "urban canyon" environments), or momentarily blocked (e.g. when the
receiver passes under a bridge or through a tunnel), and in most cases the integer
ambiguity values are "lost" and must be redetermined. This process can take from a few
tens of seconds up to several minutes with present commercial GPS systems for short-
range applications. During this "re-initialisation" period the GPS carrier-range data
cannot be obtained, and hence there is "dead" time until sufficient data has been collected
to resolve the ambiguities. If interruptions to the GPS signals occur repeatedly, ambiguity
re-initialisation is, at the very least, an irritation, and at worse a significant weakness of
commercial GPS-RTK positioning systems. In addition, the longer the period of tracking
required to ensure reliable ambiguity resolution on the fly (OTF-AR), the greater the risk
that cycle slips will occur during this crucial (re-)initialisation period. These
shortcomings are also present in any system based on data post-processing as well.
The goal of all GPS manufacturers is therefore to develop the ideal real-time precise
GPS positioning system, able to deliver positioning results, on demand, in as easy and
transparent a manner as is presently the case using pseudo-range-based differential GPS
(DGPS) techniques, which typically deliver positioning accuracies of 1-10 metres. For
example, the DGPS technique is robust, implemented in real-time via the transmission of
correction data, and there is negligible delay in obtaining results. However, there are
significant challenges for the developers of a similarly reliable "plug-and-play"
positioning system that is capable of sub-decimetre accuracy:
• Residual biases or errors after double-differencing can only be neglected for AR
purposes when the distance between two receivers is less than 15-20km. For medium-
range or long-range precise GPS kinematic positioning, the distance-dependent biases,
such as orbit bias, ionospheric delay and tropospheric delay, will become significant
problems.
• Determining how long the observation span should be for reliable AR is a challenge for
real-time GPS kinematic positioning. The longer the observation span is required, the
longer the "dead" time during which precise positioning is not possible. This can
happen at the ambiguity initialisation step if the GPS survey is just starting, or at the
ambiguity re-initialisation step if the GPS signals are blocked, such as cycle slips or
data interruptions.
• AR techniques normally require five or more visible satellites and expensive dual-
frequency GPS instrumentation in which the geometric constraints and combination of
dual-frequency observations make AR easier.
• Data latency is a challenge for many time-critical applications. The data latency is
normally caused by the data transmission and the data processing, both of which cannot
be avoided. Even if the data latency is only of the order of a few tenths of seconds, it
may restrict many applications. If the receivers at the reference and mobile stations
were outputting raw GPS observation data at 38400 baud rate, the data link rate was
9600 baud, and both differential carrier phase and kinematic records were output at the
9600 baud rate, the kinematic position output latency could be 0.9-1.4 seconds
(Lapucha et al., 1995).
• Quality control of the GPS kinematic positioning results is a critical issue and is
necessary during all processes: data collection, data processing and data transmission.
Quality control procedures are not only applied for carrier phase-based GPS kinematic
positioning, but also for pseudo-range-based DGPS positioning. However, the quality
control or validation criteria for AR, for precise GPS kinematic positioning, is a
significant challenge.

PROGRESS IN PRECISE GPS KINEMATIC POSITIONING

Over the last few years several important developments have occurred that appear to have
overcome some of these constraints:
(a) Under certain conditions decimetre level positioning accuracy has been possible
even when the baseline lengths have been up to hundreds of kilometres in length.

320
(b) Reliable OTF-AR in the shortest period of time possible, following just one
measurement epoch, has been demonstrated.
(c) Given very short periods of time-to-AR the notion of cycle slips, or having to re-
initialise the ambiguities, has no meaning because so-called "instantaneous" OTF
(lOTF) is the normal mode of kinematic positioning for all epochs.
(d) Improved multipath mitigation within the GPS receivers themselves.
(e) For certain applications single-frequency GPS instrumentation can be used.
(f) The release of several commercial integrated GPS-GLONASS receivers.
The two most significant algorithm improvements therefore have been in: (a) overcoming
the baseline length constraint, and (b) shortening the "time-to-AR" to just one epoch of
data. However, advances in receiver hardware have had to be made in concert.
Short-range «15km) IOTF-AR has been reported by Han (1997) and others.
Developments in fast ambiguity resolution algorithms and validation criteria procedures,
together with improvements in stochastic modelling and the application of careful quality
control procedures, have generally been responsible for this increased level of
performance.
Carrier phase-based medium-range GPS kinematic positioning has been reported for
baselines several tens of kilometres in length (Wanninger, 1995; Wubbena, et aI., 1996).
IOTF-AR has also been reported for medium-range GPS kinematic positioning (Han &
Rizos, 1997). Such medium-range performance requires the use of multiple reference
stations in order to mitigate the orbit bias, as well as the ionospheric and tropospheric
biases. These are exciting developments that will require testing and implementation in
operational positioning systems.
In the case of long-range kinematic positioning several innovative concepts have been
reported. Colombo & Rizos (1996) report results of decimetre accuracy navigation over
baselines up to a thousand kilometres in length. Although it is not yet possible to resolve
ambiguities OTF for baselines of several hundreds of kilometres in length, ambiguity re-
initialisation or ambiguity recovery is achievable (Han, 1995). In other words, if loss-of-
lock occurs, the AR algorithm can recover the ambiguities as long as any data "gap" is
less than a minute or so. Initial AR must be carried out using traditional techniques,
including static initialisation. A new long-range precise positioning technique that does
not require AR has also been suggested by Han & Rizos (1996b). The technique is best
described as "GPS traversing", in which the relative positions of successive GPS stations
are determined to high accuracy, not the positions in relation to a distant reference
receiver.
The development of integrated GPS-GLONASS receivers offers special challenges, not
the least being that the signals to the different GLONASS satellites are of different
frequency making the standard GPS data processing strategies based on double-
differencing inappropriate. However, the extra satellites that can be tracked should make
precise positioning a more robust procedure. During the last few years, several research
groups has been trying to develop optimal GLONASS data processing techniques. One
technique is to determine GLONASS carrier phase ambiguity "float" solution and to then
try to "fix" the GPS carrier phase ambiguities (Rossbach & Hein, 1996; Landau &
Vollath, 1996). Another method is determine the GLONASS carrier phase ambiguities by
either correcting the receiver clock bias, or estimating the small component of the
ambiguities (e.g. Raby & Daly, 1993; Leica et aI., 1995).
Data latency problems can be resolved in either of the following two ways: (a)
synchronise reference data and mobile receiver data (which gives the maximum precision
but a substantial delay), or (b) use the latest reference data and extrapolate them to the
time of the mobile receiver data (which will cause some additional error). The former is
better for the carrier phase AR process, as all errors have to be minimised for maximum
reliability and performance. However, the kinematic position will suffer due to a time
delay of up to 1-2 seconds (which may be crucial for some real-time applications). The

321
latter solution will introduce additional errors due to observation extrapolation.
Experimental results show that the linear extrapolation model will introduce an additional
double-differenced error of about 2cm for a 1 second delay and about 8cm for a 2 second
delay. A quadratic extrapolation model will introduce an additional double-differenced
error of about 4cm for a 2 second delay (Landau et aI., 1995).

OUTLOOK FOR IMPROVEMENTS IN PRECISE KINEMATIC GPS

The dramatic improvements in precise GPS kinematic positioning are the result of
improvements and innovations in several areas. Some of these are highlighted below.
Improvements in AR Techniques
Several ambiguity search procedures for OTF-AR have been suggested during the last five
years, including the FARA, FASF, Cholesky, Hatch, and U-D decomposition methods
(Han & Rizos, 1996d). The U-D decomposition method was found to be the fastest.
However, the most optimal procedure uses the LAMBDA transformation (Teunissen,
1994; Han & Rizos, 1995b) in combination with the U-D decomposition search
procedure. Although these are all search techniques in the estimated ambiguity domain,
when combined with search procedures in the measurement and coordinate domain,
single-epoch IOTF-AR is possible. Suggestions on how to improve the Ambiguity
Function Method (AFM) of AR (generally referred to as searching in the coordinate
domain) were made by Han & Rizos (1996a). Although new search algorithms will
continue to be developed, the most significant improvements to AR must come from
increasing the reliability of AR as well as the speed of the AR algorithm, or minimising
the time-to-AR. This requires careful attention to quality issues such as statistical testing
and other AR validation procedures.
Improvements in Quality Assurance and Validation Techniques
At the heart of reliable IOTF-AR are several innovations in quality assurance, stochastic
modelling and result validation. For example, an online stochastic model determination
method and an adaptive procedure are described in Han (1997). A factor to scale the over-
optimised variance-covariance matrix, which will make the ambiguity search region more
reliable, has been suggested by Han & Rizos (1995a). The validation procedure for AR in
the ambiguity search domain was proposed in Han (1997), and the validation procedure
for the AFM was described in Han & Mok (1997). The application of the quality
assurance and validation procedures should ensure correct results, and at the same time the
correct results should not be rejected. Further research should be carried out on the
refinement of the stochastic model and an adaptive procedure which can automatically
adjust with changing environment.
Bias Mitigation and the Use of Multiple Reference Stations
Medium-range kinematic positioning based on OTF-AR requires that baseline length
dependent biases be mitigated. The most important of these are satellite orbit, ionospheric
and tropospheric biases. Multiple reference stations surrounding the area of survey serve
to generate empirical correction terms for the moving GPS receiver (Wanninger, 1995;
Wtibbena, et aI., 1996). A linear combination model was proposed (Han & Rizos, 1997)
for instantaneous ambiguity resolution, which can eliminate orbit bias, ionospheric delay,
as well as mitigate tropospheric delay, multi path and measurement noise across the area of
survey. Raquet & Lachapelle (1997) also report the use of reference receiver networks for
long-range kinematic positioning. Further research needs to be undertaken in order to
determine how closely spaced the network of reference stations should be in order to
derive bias corrections with sufficient accuracy to resolve integer ambiguities, and also to
perform carrier phase-based GPS kinematic positioning using W ADGPS or W AAS
differential corrections.

322
Instrumentation Issues
GPS equipment has undergone rapid improvement, and full wavelength L2 carrier phase
and precise pseudo-range data can now be obtained from the new generation of GPS
receivers, such as Ashtech Z12, Leica SR 399, Trimble 4000Ssi and NovAtel Millennium.
The CIA pseudo-range accuracy can be derived at the lOcm level (Fenton et aI., 1991).
Multipath can also be reduced through the use of new antenna and improved receiver
tracking loop design, such as Multipath Elimination Technology (MET) and the Multipath
Elimination Delay Lock Loop (Townsend & Fenton, 1994). With respect to the GPS
system itself, an additional civilian frequency could be transmitted by the Block IIF
satellites, and Selective Availability (SA) may be turned off. The additional civilian
signal will significantly improve the reliability of AR. Without SA, double-differenced
carrier phase extrapolation will be possible to a higher accuracy, and the extrapolation
period may be much longer than the current 1-2 seconds. GPS and GLONASS receivers
such as the Ashtech GG24 will increase satellite availability, improve integrity and
accuracy. Further research is needed to improve the GPS receiver technology in order to
reduce multi path, to access the GLONASS L2 signals, and to develop optimal algorithms
for processing GPS/GLONASS data.
On the other hand, improvement can also be made in operations. The most dramatic
improvements in AR have been reported when using the latest generation of dual-
frequency GPS receivers, capable of both precise pseudo-range and carrier phase
measurements on both Ll and L2. These instruments permit IOTF-AR for short and
medium length baselines, and ambiguity recovery in the case of long baselines. However,
single-frequency instrumentation has a role to play in the "GPS traversing" technique
(Han & Rizos, 1996b) and for GPS-based attitude determination (Han et aI., 1997).
Further research may develop different operational modes that avoid the need for AR.
Real-Time Implementation
Real-time GPS kinematic positioning products (RTK) has been developed by several GPS
manufacturers, with a few decimetres accuracy during integer ambiguity initialisation.
The use of IOTF-AR when employed within such RTK systems can improve
performance, as is evident from Figure 1. The general issues relating to real-time GPS
kinematic positioning and the requirements for AR have been discussed in Han & Rizos
(1996c), and Rizos et al. (1997). Further research needs to be undertaken in the general
area of quality control, and to investigate the operational requirements of potential
applications such as machine control and guidance, robotics, precision farming, etc.
CONCLUDING REMARKS
The future of precise GPS kinematic positioning is dependent on a number of factors,
including developments in receiver hardware, changes in official GPS policy, and also the
augmentation of GPS with pseudolites or inertial navigation systems, W AAS system, and
the combination of GPS with GLONASS. All these will significantly improve the
reliability, integrity, and accuracy of the position results. Algorithm development will
play an important role and significant performance improvements can be expected in the
near future.

REFERENCES
Colombo, O.L. & C. Rizos, 1996. Testing high accuracy long range carrier phase DGPS in
Australasia. Proc. lAG Symp. liS, "GPS Trends in Precise Terrestrial, Airborne, and
Spaceborne Applications", pub. Springer, 226-230.
Fenton, P.C., W.H. Falkenberg, TJ. Ford, K.K. Ng & AJ. Van Dierendock, 1991. NovAtel's
GPS receiver - the high performance OEM sensor of the future. Proc.ION GPS-9J,

323
 Albuquerque, New Mexico, 11-13 Sept., 49-58.
Han, S., 1995. Ambiguity recovery for GPS long range kinematic positioning. Proc. ION GPS-
95, Palm Springs, California, 12-15 Sept., 349-360.
Han, S., 1997. Quality control issues relating to ambiguity resolution for real-time GPS
kinematic positioning. J. of Geodesy, 71(6), 351-361.
Han, S. & E. Mok, 1997. Validation criteria and accuracy estimation of the ambiguity function
method. Geomatics Research Australasia, 67, 67-82.
Han, S. & C. Rizos , 1995a. Standardization of the variance-covariance matrix for GPS rapid
static positioning. Geomatics Research Australasia, 62, 37-54.
Han, S. & C. Rizos, 1995b. A new method of constructing multi-satellite ambiguity
combinations for improved ambiguity resolution. Proc. ION GPS-95, Palm Springs,
California, 12-15 Sept., 1145-1153.
Han, S. & C. Rizos, 1996a. Improving the computational efficiency of the ambiguity function
algorithm. J. of Geodesy, 70(6), 330-341.
Han, S. & c. Rizos, 1996b. Centimeter GPS kinematic or rapid static survey without ambiguity
resolution. Surveying and Land Information System, Journal of the American Congress on
Surveying & Mapping, 56(3), 143-148.
Han, S. & C. Rizos, 1996c. Pro~ress and constraints of real-time carrier phase-based marine GPS
positioning. Gravity, Geoid and Marine Geodesy, Springer-Verlag, 1997,712-719.
Han, S. & C. Rizos, 1996d. Comparison of GPS ambiguity resolution techniques. Proc. Int.
Symp. on Global Positioning Systems, Digital Photogrammetry Systems, Remote Sensing &
Geographical Systems (Geo-Informatics'96), Wuhan, P.R.China, 16-19 October, 136-146.
Han, S. & c. Rizos, 1997. An instantaneous ambiguity resolution techniques for medium-range
GPS kinematic positioning. Proc. ION GPS-97, Kansas City, Missouri, 16-19 Sept., 1789-
1800.
Han, S., K. Wong & C. Rizos, 1997. Instantaneous ambiguity resolution for real-time attitude
determination. Proc. Int. Symp. on Kinematic System in Geodesy, Geomatics & Navigation,
Banff, Canada, 3-6 June, 409-416.
Landau, H. & U. Vollath, 1996. Carrier phase ambiguity resolution using GPS and GLONASS
signals. Proc. ION GPS-96, Kansas City, Missouri, 17-20 Sept., 917-923.
Lapucha, D., R. Barker & Z. Liu, 1995. High-rate precise real-time positioning using differential
carrier phase. Proc. ION GPS-95, Palm Springs, California, 12-15 Sept., 1443-1449.
Leica, A., J. Li, J. Beser & G. Mader, 1995. Processing GLONASS carrier phase observations-
theory and first experience. Proc. ION GPS-95, Palm Springs, California, 12-15 Sept., 1041-
1047.
Raby, P. & P. Daly, 1993. Using the GLONASS system for geodetic survey. Proc. ION GPS-93,
Salt Lake City, Utah, 22-24 Sept., 1129-1138.
Raquet, J. & G. Lachapelle, 1997. Long-distance kinematic carrier-phase ambiguity resolution
using a reference receiver network. Proc. ION GPS-97, Kansas City, Missouri, 16-19 Sept.,
1747-1756.
Rizos, C., S. Han & B. Hirsch, 1997. A high precision real-time GPS surveying system based on
the implementation of a single-epoch ambiguity resolution algorithm. Proc. 38th Australian
Surveyors Congress, Newcastle, Australia, 12-18 April, 20.1-20.10.
Teunissen, P.J.G., 1994. A new method for fast carrier phase ambiguity estimation. Proc. IEEE
Position Location & Navigation Symp. PLANS'94, Las Vegas, Nevada, 11-15 April, 562-573.
Townsend, B. & R. Fenton, 1994. A practical approach to the reduction of pseudo-range
multipath errors in an L1 GPS receiver. Proc. ION GPS-94, Salt Lake City, Utah, 20-23 Sept.,
143-148.
Wanninger, L., 1995. Improved ambiguity resolution by regional differential modelling of the
ionosphere. Proc. ION GPS-95, Palm Springs, California, 12-15 Sept., 55-62.
Wiibbena, G., A. Bagge, G. Seeber, V. Boder & P. Hankemeier, 1996. Reducing distance
dependent errors for real-time precise DGPS applications by establishing reference station
networks. Proc. ION GPS-96, Kansas City, Missouri, 17-20 Sept., 1845-1852.

324
 KINEMATIC POSITIONING USING ADAPTIVE FILTERS AND
MULTIPLE DGPS RECEIVER CONFIGURATIONS

A. M. Bruton, Department of Geomatics Engineering,

The University of Calgary.
2500 University Drive, NW,
Calgary, Alberta, Canada.

Abstract

The accuracy of trajectory determination by differential GPS (DGPS) is limited mainly

by receiver noise, atmospheric errors, orbital errors and multipath. Independent estimates
of a trajectory can be obtained if several reference stations are used to generate
differential corrections. Using these estimates and an adaptive filtering system it is
possible to estimate, characterize and remove some of the errors mentioned above.
The objective of this paper is to introduce the adaptive filter as a means of estimating
and reducing the errors that influence the accuracy of a trajectory determined by DGPS.
An analysis of trajectory estimation for an aircraft using a multiple DGPS receiver
configuration and an adaptive filtering system is described. Data collected during an
airborne survey carried out over the Canadian Rocky Mountains is used in the analysis.
Four ground-based reference stations were spread out over approximately 100 km,
providing independent estimates of the trajectory of the aircraft.
It is briefly reviewed how an adaptive filter can be used to estimate the correlated part
of two discrete-time signals. Since the true trajectory of the aircraft is the underlying
signal in each of the four estimates, the adaptive filtering system proves to be a useful
tool for the determination of that trajectory and the effect of the errors that influence it.
Examples from the analysis are used to demonstrate the effectiveness of the technique.

Introduction

An adaptive filtering system provides a means of estimating and reducing the errors in
estimates of the trajectory of a vehicle in motion. Acting as an interference canceler, an
adaptive system can isolate the correlated part of two sequences. This concept is very
useful if the two sequences represent an underlying signal of interest and each is
contaminated by different interference.
 An airborne survey was carried out over the Rocky Mountains in September, 1996.
Two east-west flight lines, each approximately 100 km in length, were used in this study.
The data was collected using PC computers at four ground-based reference stations
located as shown in Figure 1. Double-differenced differential processing was performed
in post-mission using Ll data collected at each of the ground stations, providing four
independent estimates of the trajectory of the aircraft. Taking a simplified approach, the
accuracy of each trajectory is limited mainly by receiver noise, atmospheric errors and
multipath.

.' . .. . .... ......

~

0 ••••

,.-:'

;~ ',', .

. ,.: .. .., .: .. .. .
. :. .······ ·····:.l... ·········· :..: .. ~~.~f;~··~~d·2····l······ . . ....... t....... .... . :
1400
g . ; < / > ;.. ......... · 2;i.,,. •••••••
-5, 1100 :.' 51.4
'jj
.s:.
800 .... . <fl.:,Iti~~rn:a~re
-117 . . . . . ..
. .".:

longitude(deg) latitude(deg)
-114

Figure 1: Flight Profile and Relative Positions of Ground Stations

An effective way to use an adaptive filter to estimate the errors present in multiple
estimates of a trajectory is to form differences between the estimates and use these as
input to the system. The difference between two estimates is a measure of the influence
of the errors at both reference stations. If two such sequences are formed with one station
in common, however, then the correlated part corresponds to the influence of the errors at
that station. These can then be used to correct the estimate made using that station.
The basic adaptive system is briefly introduced below. It is shown that when used as
an interference canceler, the output is the correlated part of two input sequences. This
concept was implemented and examples are given showing that estimates of the
trajectory of the aircraft are significantly improved. An adaptive system implemented in
this way can also be used to isolate individual sources of error at a reference station. By
taking advantage of the fact that multipath errors are different at each station, it is
demonstrated that station multipath can be isolated by careful selection of the location of
those stations.

326
Using the Basic Adaptive System as an Interference Canceler

Figure 2 shows how the basic adaptive system is configured, where x k and Yk are the input
and output of the system respectively, dk is the desired sequence and ek = dk - Yk is the
error sequence. An adaptive filter is a time-variant system that is designed to work for
cases of non-stationary input signals. With noisy data as input, the transfer function Hiz)
will adapt to minimize the effects of noise at the output at time f k •
The applications of an adaptive system vary depending on the source of the desired
sequence. When the input and desired sequences are derived independently, the system
acts as an interference canceler. Conceptually this means that noise (or more generally,
interference) present in both signals is estimated and removed.
Consider the following general form of the input and desired sequences:
dk = Sk + n! k and x k = S/ + n2k , (1)

where S and s' are correlated with each other and nJ and nz are not correlated with each
other or with sand s'. In this case, the output sequence Y will be an estimate of s, the
correlated part of x and d. The error sequence will therefore be an estimate of nJ , the part
of d not correlated to any part of x. In summary:
A A

(2)

For further details concerning adaptive filters, the reader is referred to HAYES (1996).

r------------------------ If

dk where E[S.S' •.)=Pk Vj,n

E[s.n! •.)=0 Vj,n
E[S. n 2 • .) = 0 Vj,n
E[n!. n 2 • .) = 0 Vj,n

and Pk is generally an unknown cross-

correlation function, then
A A
~------------------------ Yk = S k and £k = nl t·

Figure 2: Using the Basic Adaptive System as an Interference Canceler

Implementation of the Adaptive System

Differences formed between the four independent estimates of the trajectory are used as
input to the adaptive filter. Table 1 summarizes the components of each estimate and the
errors present in some of the differences that were formed, ignoring the effect of GPS
receiver noise.

327
 By using the differences (labeled 5-10) as input to the adaptive filter, it is possible to
isolate the influence of the errors at individual stations. These can then be used to correct
the estimates (labeled 1-4). Table 2 shows the input and output of the adaptive filter in
several cases according to equation (2).

Table 1: Summary of the components of

estimates of the trajectory and some differences.
(Station abbreviations: B-Banff, I-Invermere, C-Calgary, P-plane)

m unmodeled
o atmospheric
t errors oetween
1 plane and: multipath at station:
Trajectory o
Estimates: n B I C Bl B2 I C P

~
1. Bl +
2. B2 +
3.
4.
I
C
+
+
I+ I+ I+ I+ I I
Differences:
5. BI-1 + - + -
6. BI-C + - + -
7. BI-B2 + -
8. I-C + - + -
9. I-B2 - + - +
10. B2-C + - + -

Table 2: Summary of the input and output of the adaptive filter in several cases.
d I x y I
11. 5. 6. +
12. 8. 9. +
13. 5. 7.
14. 5. 10. +

The differences are typically of very low frequency. For this reason and to facilitate
the adaptive process by removing unnecessary parts of the signal, the differences were
band-limited to 0.055 Hz before being used as input to the filter. The normalized Least-
Mean-Squares (NLMS) algorithm was implemented, as described in STEARNS and DAVID
(1996).

Examples of Correcting Estimates of Trajectories

Consider the output y corresponding to row 11 in Table 2. This sequence is an

approximation of the influence that the errors at station B 1 have on the estimate of the
trajectory and can be used to improve the estimate made from station B 1. In the same
way, the output y in row 12 can be used to correct the estimate made from station I.

328
 Figure 3a shows the difference between the estimates of aircraft height made using
stations Bland I before and after correction using the adaptively estimated errors. The
dotted line represents the differences before correction and the dotted line represents
them after correction. It is clear that there is considerable improvement. The standard
deviation of the disagreement is reduced from 9.3 cm to 3.2 cm over the whole time
period. Notice that the level of improvement increases steadily in the period from 2:00 to
2: 15 during which the filters were converging. Also notice the small jump in the
agreement at around 2:30. This time corresponds roughly to the start of the second flight
profile and might be due to large changes in the signal statistics. Figure 3b shows the
frequency spectrum of the differences before and after correction. It is clear by
inspection that the band where the improvement is greatest is below 0.005 Hz. This
corresponds to long-term multipath and unmodeled atmospheric effects.

An Example of Separating the Sources of Error at a Reference Station

Because stations Bland B2 in the Banff area were separated by several metres, the
effects of atmospheric errors are the same at both. Because multipath errors are different
at each ground-based reference station, their influence can be recognized and partially
removed. Using an adaptive filter it is possible to estimate the effects of multipath at
each of Bland B2 and the effects of unmodeled atmospheric errors at the Banff stations.
Consider row 13 of Table 2, which is a direct estimate of the effects of multipath at B 1.
The difference BI-B2 in row 7 can then be used to obtain an estimate of the multipath at
B2. The influence of the unmodeled atmospheric error at the Banff stations is given by
row 14 of Table 2. The solid line in Figure 4a shows the estimate of the influence of
multipath at station B 1. The corresponding frequency spectrum is shown in Figure 4b.
The multipath at station B 1 can also be estimated indirectly using the difference between
the output in rows 11 and 14 of Table 2. It is clear by inspection of Table 1 that these
two ways of obtaining the estimate of multipath are not independent and should yield the
same result. The dotted line in Figure 4a shows this second estimate and demonstrates a
high level of consistency of the method and that the scale of the solutions is correct.

t:~~li'~~1
')\".'20
0200
375320
02:15
t~ (OP8, loe_l)
376220
02:30
317120
02:45
')\".'20
0200
375320
02:15
t1.ae (OPs. loc.l)
376'220
02:30
377120
02:45

10 (b) Prequency BpeC't.~ of ttl.. 1J:Kw. 'Dlffar-.c.. 10 (b) PT~ . . oy S'pectr\lal of the Kultipath at &1
: . .
h····· .........:.............
~

i 8 "f:' ................. 1... ~ . ............................. . .... --. - .............. .......................

~

....
•
:: :
6 :'7" " .............;........ .. ., ·r·········

1:~t_Fr· o 0.005 0 .01 0.015 0.02

....... --., ... ,." .......... ,., ..

°0~~~~0.~OO~S~==~0~.0~1~~~0~.OI~5----~0.02
~ ........ .. , •.••••.....
- ~ .... .

fr-.qu.IlCY (8a) t: r.qu.ncy (Hz)

Figure 3: Agreement of Two Estimates Figure 4: The Estimates of Multipath at

Before and After Correction Station Bl

329
Notes on Accuracy

By comparing the output of the filter in a given case to the output if the input and desired
sequences are reversed, a measure of the absolute accuracy of the technique is obtained.
Obviously, if the algorithm is successfully estimating the correlated part of these signals,
then the two outputs should agree closely. In all cases considered in this analysis, this
agreement was at the 2-3 cm level (I-a). This measure of the absolute accuracy agrees
with the 3.2 cm level agreement (I-a) between the corrected trajectories in Figure 3a
Several simplifications about the sources of error in an estimate of the aircraft
trajectory were made in this study. For example, possible errors due to the float
ambiguity solution and orbital errors were ignored. It is important to understand that an
adaptive filter will not correct for any error present in all estimates. With this in mind, it
has been demonstrated that the absolute accuracy of the corrections made to a trajectory
is at the 3 cm level, while the accuracy of the corrected trajectory is still subject to the
limitations of the DGPS estimation.

Conclusions

The adaptive filter was introduced as a means of estimating the correlated part of two
discrete-time signals. It was discussed how it can be used as a tool to estimate, analyze
and reduce the errors in a trajectory estimated using DGPS. Examples were drawn from
an analysis, showing that the errors are significantly reduced by the technique. Using
multipath as an example, it was also shown that individual sources of error can be
isolated. It was seen that including the convergence period, the absolute accuracy with
which the influence of a single error source can be estimated is at the 3 cm level. The
accuracy of corrected trajectories is therefore at this level and subject to the limitations of
the DGPS estimation. It was seen that for the data set used in this analysis, the errors
influencing the trajectory are of very low frequency, corresponding to atmospheric errors
and long-term multipath. Future research goals include evaluating the performance of
different adaptive filter algorithms and combining corrected estimates to improve
accuracy. Overall accuracy might be further improved by averaging the corrected
trajectories or by using them as input to a second level of adaptive filters.

Acknowledgments

The author expresses his thanks to Dr. K.P. Schwarz for his support through each stage of
the work presented in this paper.

References

HAYES, M. [1996] Statistical Digital Signal Processing and Modeling. John Wiley &
Sons, Inc., pp. 493-553.

STEARNS, S.D. and R.A. DAVID [1996] Signal Processing Algorithms in Matlab.
Prentice-Hall Signal Processing Series, pp. 290-315.

330
 ESTIMATING THE RESIDUAL TROPOSPHERIC DELAY
FOR AIRBORNE DIFFERENTIAL GPS POSITIONING
(A SUMMARY)

J. Paul Collins and Richard B. Langley

Geodetic Research Laboratory
Department of Geodesy and Geomatics Engineering
University of New Brunswick, Fredericton, N.B. E3B SA3 Canada

ABSTRACT

When post-processing dual frequency carrier phase data, the residual tropospheric delay
can easily be the largest remaining error source. This error can contribute a bias in height
of several centimetres even if simultaneously recorded meteorological data are used. This
is primarily due to the poor representation of the water vapour profile in the tropospheric
delay models. In addition, a lack of real-time meteorological data would force the scaling
of either surface values or standard atmosphere values; these are also unlikely to
accurately represent the ambient atmosphere.

To obtain the highest precision in kinematic GPS some advantage may be obtained by
estimating this error source along with the position solution. The simple tests reported in
this paper removed biases of several centimetres in height when estimating the residual
tropospheric delay from GPS data recorded at an aircraft in flight. However, important
limitations exist in the geometry of the satellite coverage which must be considered before
the full reliability of the technique can be quantified.

INTRODUCTION

This paper provides a brief summary of our investigations into estimating the residual
tropospheric propagation delay from GPS signals. This parameter is the remaining part of
the tropospheric delay not predicted by empirical models. In post-processed dual
frequency carrier phase data, it can easily be the largest remaining error source. Unlike
most applications of the technique, we have used data recorded at an aircraft in flight.
This idea was motivated by the fact that highly accurate aircraft positions are required for
gravimetric, altimetric and photogrammetric surveying purposes. Increasingly, GPS is
being used to provide the decimetre-Ievel accuracy required for some of these techniques.
This level of precision can be achieved using carrier phase observables, but we will show
that unmodelled tropospheric effects could potentially contribute a bias of a similar
magnitude.

When processing GPS observations, a value for the tropospheric delay is predicted using
empirical models which must be provided with meteorological values of the ambient
temperature, pressure and relative humidity. Unfortunately, even with accurate values,
these models rarely predict the true delay with a high degree of accuracy. In theory the
hydrostatic component of the delay can be predicted in the zenith to the millimetre level,
however the highly variable nature of atmospheric water vapour means that the accuracy
of the non-hydrostatic delay is at the centimetre, or even decimetre level.

In addition, when recording GPS data at an aircraft, it is often the case that no
meteorological information is recorded at the same time. When processing this data,
assumed meteorological values must be used, and in addition to the poorly modelled wet
component, there could also be a bias contributed by the hydrostatic component.

The results presented here are a subset of those presented in Collins and Langley
[1997b] in which results using a wider set of models are presented.

MODELLING CONSIDERATIONS AND DATA PROCESSING

A least-squares positioning model using double-differenced, dual-frequency, GPS carrier

phase observations is implemented in the KARS processing software [Mader, 1996]. The
code has been modified at UNB to allow for the estimation of the tropospheric delay as a
scale factor at either the roving receiver or at both the rover and the reference receiver.

Unfortunately, there is a problem with using differenced data to estimate the residual
tropospheric delay over short baselines. For this situation, there exists a strong
mathematical correlation between the partial derivatives of the tropospheric delay. For
baseline lengths up to several hundred kilometres the elevation angles to a particular
satellite will be similar and hence so will the partial derivatives (but with opposing signs).
Even if the meteorological conditions are drastically different at the ends of such a
baseline, it is difficult for a least -squares model to separate the two contributions. The
usual technique to overcome this problem is to fix the tropospheric delay at the reference
station and to estimate the relative delay at the secondary station. We have used real-time
meteorological data at the reference station to help minimise the error in the estimated
residual delay.

The data set used in this paper consists of dual-frequency GPS data recorded at a two
second interval at a reference station and an aircraft in flight. The flying time was approx-
imately 103 minutes up to a maximum distance of 210 kilometres from the reference
station at St. John's, Newfoundland (see Figure 1). A set of fixed integers for all satellites
on both frequencies was derived. This was done by processing the flight data at various
elevation cut-off angles while resolving the ambiguities "on-the-fly". Comparing the
ambiguities from these solutions with ambiguities computed for the short static period
before the flight, has enabled stable sets of integers to be selected. While confident that
these are the correct values, without actually estimating these values in flight, we can only
confirm this by examining the residuals of the positions solution to see if they diverge over
time.

332
 Of the remammg error sources, the
primary one is the satellite position
error. To minimise it as much as
possible, International GPS Service for
5000 Geodynamics (IGS) precise orbits were
used. This leaves multipath and noise
which should be of the order of
centimetres or less for the carrier phase
observable.

The solution was computed estimat-

ing the three-dimensional positions of
the aircraft along with the residual
L
ongitude (d
-51 tropospheric delay as a scale factor. No
eg)
filtering was applied and no a-priori
Figure 1. Flight path of aircraft for data used constraints were placed on any of the
in this study. parameters. Each epoch provided an
independent solution.

RESULTS

The results presented here used one tropospheric zenith delay and mapping function
combination at the reference station and the aircraft. These were the Saastamoinen
[1973] zenith delays using simultaneously recorded meteorological data and the mapping
functions of Nieli [1996] which only require position and day-of-year information. This
model is denoted as SAANf in this paper.

Solution Residuals

Considering first of all the root-mean-square (rms) of the double-differenced carrier

phase residuals after the least-squares adjustment, examination of Figure 2 shows the
general improvement gained by estimating a residual tropospheric delay parameter.
Overall a small improvement has been made, indicating that estimation of the residual
tropospheric delay has reduced the impact of the errors in this model on the solution.

Residual Delay Estimates

Turning to the actual residual delays estimated, Figure 3 shows the residual delay
estimated over the flight. The plot can be considered in two halves - before and after the
45 minute epoch. Consideration of the elevation angle trace shows that before this point
in the flight there are no satellites at low elevation angles « 10 degrees). As is well
known in GPS, this limits the potential for adequately estimating the tropospheric delay.
The wide variation in the first half of Figure 3, coupled with the large negative magnitude
could mean that the residual estimates for this time span are unreliable.

333
 0.05 Tr===============:::;----- -- - - - - i
- SAANf - no estimation
I 0.04 - SAANf - scale factor

§
C/l 0.03
(ij
::J
'0
'iii 0.02
~
.c
g
a. 0.Q1
W

o 10 20 30 40 50 60 70 80 90 100

Time (min)

Figure 2. The rms double-difference carrier phase residuals with and without residual
tropospheric delay estimation with real-time meteorological data.

0.04 ~-----------------------. 22

20
0.02

18
0.00
m
eg'
16 <D
... -0.02
o - Scale factor estimate
14
~ \ - Scale factor uncertainty

/
-0.04 III
Q)
Lowest elevation angle / ::J
12 co
~
(f) -0.06 I / <D
"0:
I I" I 10
S
(\)

-0.08
I /\ / I 8

-0.10
V \J l/~
-0.12 ..l...-~--r----r---.--.--"---'--------''---''----'---'--'- 4
6

o 10 20 30 40 50 60 70 80 90 100

Time (min)

Figure 3. Comparison of residual tropospheric delay estimates and their uncertainty

with the lowest elevation angle used in the solution.

Position Differences

Without residual delay estimation, we would consider the kinematic solution using
SAANf to be the "best" obtainable because of its realistically-modelled zenith delays and
mapping functions driven by real-time meteorological data. By estimating the residual
delay we would hope to model any deviations from the average atmospheric structure
implied by these models. Figure 4 shows the difference in the position components for
solutions computed with and without residual tropospheric delay estimation_ The

334
difference in the height component is considerable: of the two sets of statistics for this
data, even when considering only the "good" estimates after the 45 minute epoch, there is
a mean bias in height of -5 cm with an rms of -9 cm.

0.4...,......- - - - - -- - - - - - , - - - - - - - - - - - - - - - - ,

0.3
I<I> 0.2
U
~ 0.1
iii
:t:: 0.0
'6
§ -0.1
:;:::
'iii -0.2 - Latitude
o
a... - - Longitude Heieht Difference Statistics
-0.3 - - Height All data: mean =0.078 m, rms =0, 115 m
After 1=45: mean := 0,045 m, rms = 0.085 m
-0.4 -"-.-- --,--- --.--- --.--- ---,-----'-----.-- - - - , ---,- -,--,.---..---'
o 10 20 30 40 50 60 70 80 90 100

Time (min)

Figure 4. Difference in position solutions with and without residual delay estimation
from predictions with real-time meteorological data.

It is possible under other conditions, for example using standard atmosphere

meteorological values with the same zenith delay and mapping function combination, for
the biases and differences shown in Figure 3 and Figure 4 to be much larger. However
estimating the residual tropospheric delay appears to remove the impact of such biases.
This is shown in Figure 5, where position differences between a solution computed with
the "composite" tropospheric delay model UNB4 at the aircraft and the SAANf model
solution are shown. (Note: UNB4 supplies meteorological data based on the 1966 U.S.
Standard Atmosphere Supplements to the Saastamoinen and Nie11 algorithms. For more
details see Collins and Langley [1997a]).

E 1.0 - Latitude Height difference rms =0.4 cm

.£. - - Longitude
~ 0.5
c - Height
<I>
iii
:t:: 0.0
'6
§ -0.5
:;:::
'iii
~ -1 .0

o 10 20 30 40 50 60 70 80 90 100

Time (min)

Figure 5. Position differences between the UNB4 solution and the SAANf solution.
(Residual delay estimated in both solutions.)

335
CONCLUSIONS

We have presented in this paper a brief summary of our investigations into the effects of
implementing residual tropospheric delay estimation from GPS data recorded at an aircraft
in flight. The aim was to remove any unmodel1ed effects of the troposphere that cannot be
predicted by empirical models, even when using meteorological measurements of the
ambient atmosphere.

Estimating the residual delay appeared to almost wholly remove the impact of errors in
the tropospheric delay model. However, the impact of the satellite geometry is important.
It appears crucial that there exists data at low elevation angles (less than 10 degrees) for
the tropospheric residual estimate to be meaningful. If the highest possible precision is
required for aircraft positioning then estimation of a residual delay should be considered,
otherwise biases of up to ten centimetres may be present in the solution.

This has been only a preliminary study and further work is required to study the
condition of the normal equations of the least-squares adjustment and the reliability of the
technique. New investigations could include the impact of antenna phase centre
corrections, as the data is particularly sensitive to these at low elevation angles.
Additionally, the implementation of a Kalman or other type of constraining least-squares
filter could significantly enhance the technique by providing some a-priori constraints to
the estimates.

REFERENCES

Collins, J.P. and R.B. Langley (1997a). A Tropospheric Delay Model for the User of the
Wide Area Augmentation System. Final contract report prepared for Nav Canada,
Department of Geodesy and Geomatics Engineering Technical Report No. 187,
University of New Brunswick, Fredericton, N.B., Canada.
Collins, J.P. and R.B. Langley (1997b). "Estimating the residual tropospheric delay for
airborne differential GPS positioning." Proceedings of ION GPS-97, Kansas City, Mo.,
16-19 September, pp. 1197-1206.
Mader, G.L. (1996). "Kinematic and rapid static (KARS) GPS positioning: Techniques
and recent experiences". lAG Symposia No. 115, Eds. G. Beutler, G.W. Hein, W.G.
Melbourne and G. Seeber. IUGGIIAG, Boulder, Colo., 3-4 July. Springer-Verlag,
Berlin, pp. 170-174.
Niell, AE. (1996). "Global mapping functions for the atmosphere delay at radio
wavelengths." Journal of Geophysical Research, Vol. 101, No. B2, pp. 3227-3246.
Saastamoinen,1. (1973). "Contributions to the theory of atmospheric refraction." In three
parts. Bulletin Geodesique, No. 105, pp. 279-298; No. 106, pp. 383-397; No. 107, pp.
13-34.

336
 HIGH-ACCURACY AIRBORNE INTEGRATED MAPPING SYSTEM
Dorota A. Grejner-Brzezinska

The Ohio State University Center for Mapping

1216 Kinnear Rd., Columbus, OH 43212
e-mail: dorota@c(m.ohio-state.edu

Abstract

A fully digital Airborne Integrated Mapping System (AIMS) for large-scale mapping and other precise
positioning applications is currently under development at The Ohio State University Center for
Mapping. AIMS is designed to be installed in an aerial platform and incorporates state-of-the-art
positioning (differential GPS integrated with INS) and imaging (CCD) technologies. Preliminary test
results show that platform position and orientation can be obtained with an estimated accuracy better
than 7 cm and 10 arcsec, respectively, over long baselines.

Introduction

Two primary components of the AIMS system - the Positioning Module and an Imaging Module are
combined and installed on an aerial platform. Tightly coupled Global Positioning System (GPS) and
Inertial Navigation System (INS) comprise the Positioning Module. This module provides direct
platform orientation with an accuracy of 4-7 cm in position (over long baselines) and better than 10
arcsec in orientation. Such accuracies support the extraction of geographically referenced information
from the imaging component of AIMS without the need for ground control. The performance of the
Positioning Module has been repeatedly tested, and has proven its high accuracy and reliability. This
paper provides an overview of the Positioning Module of the integrated system, with special emphasis
on a discussion of the airborne test results.

AIMS: Concept and Description

The architecture of AIMS, as presented in Figure 1, emphasizes flexibility that enables augmentation of
a variety of sensors beyond the high-resolution CCD (Charge-Coupled Device) cameras, including
Synthetic Aperture Radar (SAR) and radar or laser ranging devices. The GPS time is used to
synchronize position information with measurements from the other sensors. Currently AIMS operates
in a post-processing mode, however, the ultimate goal is to build a real- or near real-time system. It is
anticipated that the system will require no ground control, except for the base station that enables
differential GPS positioning. Although both technologies, GPS and inertial navigation, are well known
and have been used for precise positioning for a number of years (Abdullah, 1997; Kerr, 1994; Lapucha,
1990; Schwarz, 1981, 1990, 1995), it should be pointed out that application of digital cameras in the
aerial mapping field is a relatively new component in the mapping market.

The hardware components currently implemented in AIMS are two dual-frequency Trimble 4000SSI
GPS receivers, a medium-accuracy and high-reliability strapdown Litton LN-100 inertial navigation

337
system, and digital camera based on 4,096 by 4,096 CCD with a 60 mm by 60 mm imaging area (15-
micron pixel size), manufactured by Lockheed Martin Fairchild Semiconductors (implementation under
development). The LN-lOO internal software was modified to provide raw IMU measurements, i.e., the
velocity and angular rates measured in the IMU coordinate system. The main reason for this
modification was the fact that the INS vertical channel needs to be externally stabilized, and any errors
in external aiding result in a corruption of the inertial navigation solution. This problem is avoided by
using an external strapdown solution based on the raw IMU data, and IMU calibration results estimated
by the integrated filter. Updating INS with accurate GPS information at 1 Hz rate (or better) will assure
platform orientation at the ~5 arcsec level. The tight integration scheme implemented in AIMS allows
continuous precise positioning even if the number of tracked satellites drops below four.

N.,...,_
w.....
. _...
"",...a,.~
V..... IY . .. A'"

Figure 1. The structure of AIMS. Figure 2. GPS/INS integration structure.

GPSIINS Integration and Kalman Filtering.

In AIMS a tight GPS/INS integration is implemented, as shown in Figure 2, where a single Kalman
filter is used to process the GPS double-differenced phases, combined with the inertial solution, to
optimally estimate the state unknowns (Da, 1997; Grejner-Brzezinska, 1997). Tight integration supports
GPS cycle slip fixing and implementation of closed-loop INS error calibration. The inertial strapdown
algorithm is implemented for processing the raw IMU data, providing the navigation solution. The state
unknowns are errors in position, velocity, and orientation, three biases and three scale factors for the
accelerometers, three gyro drifts, two deflections of the vertical and the gravity anomaly. In addition,
GPS ionospheric delay is estimated for every satellite in the solution. Also, lever arm errors can be
optionally included in the state vector. These are related to the unknown offsets of the GPS antenna
from the IMU center. The INS error model adopted in our solution follows the approach specified by
Bar-Itzhack and Berman (1988). The initial conditions for the navigation error parameters follow the
recommendations of Litton Systems, Inc., for a high accuracy navigation instrument.

On-The-Fly (OTF) Ambiguity Resolutionfor the Integrated System.

A robust GPS/INS integration, as implemented in AIMS, provides fast and reliable correction of cycle
slips and losses of lock affecting GPS measurements, and allows for fast OTF ambiguity recovery over
long baselines. Initial L 1 and L2 double-differenced phase ambiguities, and precise platform location,
are provided by the real-time GPS software, as starting conditions for the integrated Kalman filter

338
(Grejner-Brzezinska, 1997). In addition, the ambiguities of the satellites whose tracking starts during the
flight are resolved OTF, and cycle slips are detected and repaired based on the filter predictions. In the
case of a total loss of lock, the four highest satellites are selected, and the search loop is activated. The
INS error growth in AIMS is about 10 cm for the horizontal components, and about 20 cm for the
vertical one, after a 50-second loss of GPS lock, as shown in Figures 3 and 4, which still enables
instantaneous ambiguity recovery after the GPS signal is recovered. The OTF search loop was tested for
speed and reliability of ambiguity recovery as a function of the quality of the INS prediction. Our tests
showed that all cycle slips could be fixed instantly if positioning quality was not worse that 1 m per
coordinate. In all of the tests performed, the system recovered the correct ambiguties.

0.2
0.15 0.08

0.06
Ien 0.1
8c 0.05
~
~ 0 =
Q
c -0.05
0
~
en tJ. - difference in north coordi nate ·0.04
0 -0.1
a. tJ. E - difference in east coordinate
-0.06
-0.15 tJ.o - difference in vertical coordi nate
-0.08
-0.2 0
10 20 30 40 50) -O b'-------:-:IO----:20~----:307---40~-~50
Time Lsecl Tlme,loeel

Figure 3. Differences between filter predictions Figure 4. Differences between filter-predicted L 1

and GPS positions; no GPS data in the filter. DO and GPS DO; no GPS data in the filter.

Preliminary Test Results

The AIMS system performance has been tested under different conditions with varying base-rover
separation. This paper presents the airborne test results of the flights performed on March 6, 1997 near
St. Louis, Missouri, with the OMNI Solutions International Ltd. Beachcraft airplane, and on July 2,
1997 near Berkeley, California with the Hammon, Jensen, Wallen & Associates, Inc., two-engine
Cessna 31 OP aircraft. In these tests the IMU data update rate was 256 Hz; however, a rate of 400 Hz was
applied for data acquisition to guarantee no loss of data. Differential GPS observations were collected
by a Trimble 4000 SSE for the St. Louis test and a Trimble 4000 SSI for the California test. A Zeiss
RMK TOP aerial camera was used in the California test.

The level of double-differenced (DO) residuals and standard deviations (square roots of the diagonal
terms from the covariance matrix) per coordinate for position, velocity and orientation angles represent
the positioning quality in our tests. Since the standard deviations for the parameters behave very
similarly for all the tests, the results from the 10-km flight test are plotted in Figures 5-7 as an example.
Figures 8 and 9 present the double-differenced residuals for both flight tests. Figure 3 shows the
differences between filter predictions without using GPS data in the filter, and GPS positions determined
from an independent GPS positioning package. Figure 4 displays differences between filter-predicted
double-differences and GPS double-differences, when no GPS data are applied in the filter. These
results show that the average error growth rate is about 2 mmls for the horizontal components, and about

339
 4mm1s for the vertical, for the first 50 s of GPS total loss of lock, which indicates that most correlated
IMU errors have been correctly estimated by the Kalman filter, and removed from the IMU
measurements by the feedback update loop. Thus, the system is able to provide accurate velocity and
position information even when GPS observations are not used for a short period of time. When GPS
observations are used to update the filter at 1Hz rate, residuals range from 0 to 4 cm, after the removal of
the outliers (Fig. 8 and 9). In Figure 8 only one satellite, PRN 7, has somewhat larger residuals, reaching
4 cm at the largest base-rover separation. PRN 7 also shows the largest discrepancy between the
ionospheric estimates from the filter, and from the geometry-free combination (Figure 10). These
indicate that both residuals and the ionospheric estimates are contaminated by an unmodeled effect,
most probably differential troposphere that increases with the distance and altitude separation. Another
example of the DD ionosphere estimates is presented in Figure 11.

," 'OI
,.. ''''''
'ODO
\---...., _____ (I ...

._------------'\ """
00$1 "-
_ Go.
~ ',.
,o. 1 """ i
I
1 '''''
La> \
\............
i 000<1
i
l. 1\

1:: I~~_=_=__=__~__=__~.~~~~/~~~~.. =_~..~~..~.~.__=__.

:. oem

......-- . -~---- .. --------

',~~,=~--~~~~-=---~,~~--=~~~~=---~~
TirM [__ I

Figure 5. Position standard deviations. Figure 6. Velocity standard deviations.

...".
....
. ____ _ CI' . .

-'~

I .,
1
i .,
t",L-----:::;----:;.~~~~
-~~.~-.__,._. .__. . _
..
, ~ ~ =
~(-=)
~ ~ ~ ~ ~. r---~~~---,~~--~,~~--~--~~~--~
n-I"cl

Figure 7. Orientation standard deviations. Figure 8. DD residuals, 180 km baseline, low

level of GPS phase observation noise.

The comparison of the results obtained from GPS/INS, and independently from the aerotriangulation
based on the California test, is currently being analyzed. Figure 12 presents the trajectory for the
California flight test where the imaging data was collected. Even though we experienced some problems
with the image time tags, we were able to compare horizontal coordinates of the four control points
marked with an arrow in Figure 12. Preliminary results indicate that difference between GPS/INS and
photograrnmetric solutions ranges between 1-3 cm per coordinate component for three of the points, and

340
 reaches -10 cm for the fourth one. More flight tests, where independent photogrammetric solution can
be obtained, are needed to obtain a detailed performance assessment of our navigation system.

0.04 r----..----~--~----r--__._--_, 0.15 r------------~----___,

0.Q3 lono!Jphere (rom

0.10 g.eometf)' ~ free
0.02 combination
I E.$timated
IO.Q1 ~ 0.05 ionosphere

VI
c;;
'"
"0
.~ -0.01
0::
0-0.02
o
-0.03

-0.04
-O . 15 '---~--~--~--~--~---'
o 500 1000 1500 2000 2500 3000
200 400 600 800 1000 1200 Time l,ocl
GPS Week Time (338800 - 340000 5)
Figure 9. DD residuals, 25 km baseline, high Figure 10. DD ionosphere estimates,
level of GPS phase observation noise. 180 km baseline.

0 .15 .--_--_--~--~--_-_,
37.9
6
lonosphrra ( rom
0.10 PRN I' ItLtom etry.rtee 37.94
Ele\,a'lon 11- · 22 - comblnltlon
E S11Mililed
Oi 37.92
'"
IOnosphillre
:2-
~ 37.9

~'" 37.88
-0.05
37.86

·0.10 37.84 * Position where image

was taken
-0.15 '---~--~--~--~--~----'
o 500 1000 1500 2000 2500 3000 37: ¥22 .32 -122 .31 -122 .3 -122 .29 -122.28 -122.27 -122.26 -122.25
Time [s.c) Longitude [deg )

Figure 11. DD ionosphere estimates, Figure 12. Aircraft trajectory, California flight test.
180 km baseline.

Conclusions

The results presented in this paper verify the high quality and reliability of the positioning component of
the Airborne Integrated Mapping System that is currently under development at the Ohio State
University Center for Mapping. The level of double-differenced residuals and the low rate of INS error
growth when GPS data are not available demonstrate that the positioning system is well designed and
calibrated, and most of the systematic effects are properly accommodated. However, there is still room
for further improvements, especially in the modeling of differential atmospheric effects, and the attitude
component of the system - a crucial factor that enables elimination of the ground control and traditional
photogrammetric processing. Therefore, we plan to use 1'xl' (2'x2' optionally) grid values of
deflections of the vertical in the strapdown navigation algorithm, and also to implement stochastic

341
modeling of the tropospheric effect into the Kalman filter. Proper accommodation of the differential
tropospheric effects might improve the estimation of the vertical positioning component of the platform.

One of the major goals of this system was to provide position and orientation of the aerial platform and
the digital imagery with an accuracy that would permit the elimination of the ground control and the
need for aerotriangulation. The results presented here look very promising, however, final conclusions
about AIMS absolute accuracy can only be drawn following extensive testing against ground truth.
Table 1 presents results of our preliminary simulations of the impact of the errors in the camera exterior
orientation on the position accuracy of the ground points. Image size of 4kx4k, 10llx 1Oil pixel size,
focal length of 30 cm, and flight altitude of 1000 m were assumed; average errors in the initial
position/orientation O"XoYoZo, O"roO(POKO are given.

Table 1. Impact of the inaccurate exterior orientation on the ground positioning: simulation results.

O"XoYoZo,O"rompoKo Mean [m] Standard Dev. [m] Median [m]

5 cm, 5" 0.130 0.077 0.113
10 cm, 10" 0.260 0.153 0.226
20 cm, 20" 0.505 0.290 0.429
30 cm, 30" 0.768 0.481 0.651

Acknowledgments

This work is supported by the NASA Stennis Space Center, MS, Commercial Remote Sensing Program
grant #NAGI3-42, and a grant from Litton Systems, Inc. Trimble Navigation, OMNI International, Ltd.
and Hammon, Jensen, Wallen & Associates, Inc. are acknowledged for providing GPS receivers and
airplane time.

References

Abdullah, Q., (1997): Evaluation of GPS-Inertial Navigation System for Airborne Photogrammetry, Applanix, Applied
Analytical Corporation.
Bar-Itzhak, I. Y., Berman, N. (1988): Control Theoretic Approach to Inertial Navigation Systems, AIAA Journal of
Guidance, Control, and DynamiCS, Vol. II, No.3, pp. 237-245.
Da, R., (1997): Investigation of Low-Cost and High-Accuracy GPS/IMU System,
ION National Technical Meeting, Santa Monica, pp. 955-963.
Grejner-Brzezinska, D. A., (1997): Airborne Integrated Mapping System: Positioning Component, ION Annual Meeting,
June 31 - July 2, Albuquerque, NM.
Kerr III, T. (1994): Use of GPS/INS in the Design of Airborne Multisensor Data Collection Mission (for Tuning NN-based
ATR Algorithms), Proc. ION GPS, Salt Lake City, Utah, pp.1173-1188.
Lapucha, D. (1990): GPS/INS Trajectory Determination for Highway Surveying, Kinematic Systems in Geodesy, Surveying
and Remote Sensing, Springer-Verlag, pp. 372-381.
Schwarz, K. P., (1981): A Comparison of Models in Inertial Surveying, pp. Bulletin Geodesique, Vol. 55, No.4, pp.300-
314.
Schwarz, K. P. (1990): Kinematic Modelling - Progress and Problems, Kinematic Systems in Geodesy, Surveying and
Remote Sensing, Springer-Verlag, pp. 3-16.
Schwarz, K. P. (1995): INS/GPS as a Georeferencing Tool for Multi-Sensor Systems, presented at the Mobile Mapping
Symposium, May 24-26, Columbus, OH.

342
 AIRBORNE GPS PERFORMANCE DURING A
PHOTOGRAMMETRIC PROJECT

Paolo Baldi *, Maria Marsella **, Luca Vittuari ***

* Dip.
Fisica - Universita di Bologna - E-mail: [email protected]
** DITS - Universita. di Roma "La Sapienza .. -E-mail:[email protected]
*** DIS TART-Un ivers ita di Bologna - E-mail:[email protected]

Abstract

Preliminary results obtained from the analysis of GPS data collected during an airborne
photogrammetric project conducted in 1996 over the Island of Vulcano are described. The
objective of this study is to determine an optimal processing procedure for obtaining high
accuraacy projection center positions for large scale aerial photography applications. GPS
measurements were collected using a multi antenna configuration both on the aircraft and
on the ground. The performance of standard software for OTF differential kinematic
processing was evaluated. In order to use redundant information both rigorous combination
of independently computed solutions and simultaneous adjustment of GPS observation
from multiple reference stations and mobile receiver were performed. GPS-derived camera
projection centers are compared to the results from a aerotriangulation adjustment results.
The results indicate that the GPS data can be helpful to preserve the accuracy level when
there is a lack of ground control points.

Introduction

In order to acquire a detailed knowledge of the deformations related to the eruptive

dynamics of the Vulcano Island, the use of a GPS-based aerial photogrammetry technique
was investigated. Large scale photographic images were acquired in order to achieve
sufficient accuracy of the mapping products and thus detect, by comparison with earlier
products, large deformation patterns (Achilli et aI., 1994). The collection of images and
GPS data was carried out in September 1996 using three GPS reference ground stations
operating in the project area plus an additional station located about 70 km away. On the
aircraft a photogrammetric camera was interfaced with two GPS receivers connected
through a splitter to the same antenna. The flight was performed under regular operational

International Association of Geodesy Symposia, Vol. 118

Brunner (ed.), Advances in Positioning and Reference Frames
© Springer-Verlag Berlin Heidelberg 1998
conditions with no particular intention to avoid signal outages during aircraft turns. GPS
kinematic solutions were obtained using both commercial and scientific software with OTF
ambiguity resolution capabilities. Different data processing strategies were used in order to
select the most appropriate and effective in satisfying the stringent accuracy requirements
« 20 cm). In particular, the performance of different On-The-Fly ambiguity resolution
algorithms was studied. Performance of different GPS receiver models was also compared.
In order to use all the redundant data, solutions obtained using the three reference stations
were combined into a rigorous adjustment of kinematic vectors which connect the aircraft
antenna to each reference station assumed as fiducial points with standard deviations of a
few millimeters.
Software capable of multi-station adjustment of undifferenced GPS observable, was used in
order to combine the correlated redundant data and perform a simultaneous estimation of
ambiguities and relevant parameters, such as the coordinates of the rover station,
atmospheric correction parameters, etc. GPS-derived position of the aircraft at the instant
of image exposures was compared to those obtained independently by means of an
aerotriangulation adjustment.

Airborne GPS Project

In September 1996 two GPS-photogrammetric flight mission were performed over the
Vulcano Island. The GPS data were collected at the altitudes of 1000 and 1800 meters in
order to acquire images at 1: 5 000 and 1: 10000 scale. Distances between the aircraft and
the reference stations on the island (YUOO, YU04, YULC) did not exceed a few
kilometers; but an additional reference station (RCFX) was operated at the airport located
about 70 kilometers from the project area. The aircraft was equipped with a WILD RC20
camera connected to a dual GPS configuration consisting of a Ashtech ZXII and Trimble
SSI receivers logging data respectively at 1 Hz and 0.2 Hz. Both receivers were connected
through a splitter to the same antenna installed on the fuselage. The spatial offset between
the GPS antenna and the camera projection center was about 1.5 meters in the vertical and
a few centimeters in the horizontal components.
An accurate GPS control network was established in the project area and specially
designed targets placed. The network included the three reference stations used for the
differential kinematic solutions where GPS receivers were continuously operated during the
flight mission, collecting observations every 1 second. The network was computed using
the Bernese software and tied to ITRF through the Italian permanent station of Matera.
Results of the network adjustment showed standard deviation of few millimeters both in the
horizontal and vertical components. Table 1 summarizes the main characteristic of the GPS
airborne missions and Figure 1 shows the horizontal aircraft trajectory on day 273.

Kinematic GPS Data Processing

GPS data collected on the aircraft and at the reference ground stations were processed by
different software developed for handling kinematic GPS data, and witch included On-The-

344
Fly ambiguity resolution algorithms: PNAV (Precise Differential GPS NAVigation and
Survey Program) (Ashtech, 1994), GEOTRACER GPS (TerraSat, 1994) and GEONAP-K
(Wubben a, 1989).

Table 1 : Main characteristics of the Vulcano GPS airborne project

Airborne GPS Mission

Period September 1996
Location Vulcano Island, South Italy
Aircraft Navajo
Camera Wild RC 20
Number of missions 2 (Day 2721 Day 273)
Hours flown 2 (Day 272) 1 3 (Day 273)
FligM heights 1000 m (1:5000) 1 1800 m (1:10000)
GPS receiver model Ashtech Z-XII 1 Trimble 4000SSI
GPS acquisition rate Isec/5sec
Ground stations 2 Ashtech Z-XII 1 2 Trimble 4000SSI
Cut-off angle 10°
Number of visible satellites 6/7

PNAV, based on a Kalman filter

and a smoother for combining
backward and forward solution,
can process all combination of GPS
observables; and was used for a
first analysis of the whole data set
in order to access the overall
quality and detect faulty
observations. Results obtained
from PNAV are summarized in
Table 2, where the RMS of height
differences are listed for each pair
Fig. 1 Horizontal trajectory of aircraft on Day 273 of reference stations. This level of
repeatability can be achieved on
most of the aircraft positions using standard processing procedures. During certain critical
situations, due to unexpected disturbances on GPS signal acquired from one ground
receiver (VU04), the standard procedure was not successful; and manual interaction to
introduce waypoints, independently computed, to constrain the solution or to eliminate the
affected observations was necessary.
A I-hour dataset (from epoch 33000 to 37000), during which the photographic images
were taken, was selected for further analysis with GEOTRACER GPS and GEONAP-K
software. Comparing the results with PNAV, it was noted that the OTF algorithm
implemented in GEOTRACER is more finely tuned resulting in a higher stability and
slightly better RMSs of differences. It should be noted, though, that the estimated solutions

345
resulted in a larger number of rejected positions. The main differences between the two
software were observed during changes in the satellite configuration, as indicated in Figure
2. In order to correctly combine the redundant observations, the GEONAP-K software,

Table 2 RMS (cm) of the differences between trajectory solutions from two ground
stations

stat.1-stat.2 RCFX-VUOO RCFX-VU04 RCFX-YULC

float/fixed solution 16.9 9.7 10.8
stat.1-stat.2 VUOO-VU04 VUOO-YULC VU04-YULC
fixed solution 8.9 9.1 3.8
float solution 12.2 13.5 7.2

capable of multi-station
adjustment of undifferenced GPS
0.2 - r - - - - -- - - - - - , - ----.
observables, was used. Using the
0, 15
three reference stations on the
]; Ii island as fiducial points, and
t: o.J E adopting a rigorous simultaneous
:; =
.c
0,05
" dual frequency processing
procedure (Seeber et aI., 1995),
o 4 high accuracy aircraft positions
33000 33500 34000 34500 35000 35500 36000 36500
GPS We~k (sec) were estimated. Results from a
single-station configuration
1- gcotracer-PNAV ..- num, saL
showed differences with the
multi-station solution at the 2-3
Fig. 2 Height differences between GEOTRACER centimeter level RMS on the 3-D
and PNAV solutions coordinates (Figure 3).

Comparison Between GPS and AT Projection Centers

As the GPS antenna and the camera system were spatially and temporarily connected, the
coordinates of the camera projection center can be derived by interpolating the computed
coordinates of GPS antenna position at the time of exposures and applying the measured
spatial offset from the camera. GPS-derived projection centers can act as aerial control
within the image processing procedure (Colomina, 1993), especially where the control
network cannot be very dense due to operational constraints (Hothem, 1994).
In order to take advantage of the GPS aerial control and improve or at least maintain the
quality of the relative mapping products, the GPS positions should fulfill specific accuracy
requirements, that in case of 1:5000 photo scale are less than 20 cm. At the same time,
when a conventional control network is established on the ground it is possible to estimate
the complete set of image external parameters, that is position and orientation of the
camera, by performing an aerotriangulation bundle adjustment. The accuracy of the

346
estimated projection centers positions can be assessed from the standard deviation of the
residuals on the control and photographically measured pass and tie points. For this test the
AT adjustment was performed over a 5-strip photogrammetric block formed by 23 selected
images, using 12 ground control points, from witch was obtained mean standard deviation
values ranging between 5 and 25 centimeters (Table 3). It should be mentioned though,
that the higher standard deviations for STP 2 and STP 5 can be explained by the lack of
ground control points in the corresponding images.
The triangulated projection centers of the
remaining strips can be used as an
0.08 independent check of the GPS positions.
The GPS projection centers to be
g U.04
'"
compared with the AT results can be
~ 0) ~""t\1~~. derived both by the rigorous adjustment
~'5 ~.04 of the three PNAV solutions (one for
-0.08 - 1 - - - - - - _ _ _ < _ - _ - + -_ _ _ each reference station on the island) or
33uoo 34000 35000 36000 37(\()() by the multi-station GEONAP-K
CPS W";E K (srr)
solution. The comparison between the
two solutions shows the presence of
small biases with different behaviour
Fig. 3 Differences in XYZ components (Figure 4). The differences are about 5
between single and multi-stations solutions centimeters for the horizontal
(GEONAP-K) components and 10 centimeters for the
vertical component.
This indicates that the quality of airborne
Table 3 Mean value of standard GPS data is high, so that we can obtain
deviations of the projection centers from OTF position with higher accuracy both
AT adjustment using a standard procedure and a more
rigorous approach. The centimetric
Standard deviations differences can be attributed to the
strip DE(m) DN(m) Dh(m) different error models applied in the
STP2 0.207 0.186 0.067 various software.
STP3 0.104 0.149 0.054 It should be noted that in case of a GPS-
STP4 0.104 0.119 0.051 supported bundle block triangulation all
STP5 0.248 0.225 0.151 the systematic discrepancies can be
STP8 0.111 0.128 0.046 modeled and removed using the "shift and
drift" approach.
An affine transformation was applied strip by strip between GPS and AT derived projection
centers yielding residuals shown in Figure 5. The large residuals on strips 2 and 5 can be
attributed to the lack of ground control points. In this case the use of high accuracy GPS
control could be beneficial.

Conclusions
Different GPS kinematic processing were attempted in order to obtain high accuracy
projection centers in the framework of a GPS-supported aerial photogrammetry project.

347
The use of multiple reference stations was advantageous both for controlling and
constraining the kinematic solution. The use of accuracy GPS-derived camera centers can
be helpful to preserve the accuracy level when there was lack of ground control points, as
often occurs in remote and inaccessible areas.

Cl r-~--~----~----------- 03 ,--------:----~~---------:
sq,
2: sip 3 . Ip 4 .Ip ~ Alp S ;
02 ..• __ .. _-_ .. _- .... -.... -. 02 ........... ..... .... .......... .
.;
~.... -

f
0' .. ,,' .

(~;1~ nr. .
01 : ",'""."".""."',, ......
E Il 11 11~';'1
1= '1
~,: 'iL
' '. il
"'.{1 -r- .<,., 'If '
, : '1

.. 1ln 11111T11'1Tf~
; o . , . :j I; ,j < :

'6 i
·0'
'. r
"1'..... "" :. '
i

-02 . "
-02

Photo Number
·0 l =-•.=-..
L:.,.-:"".."=,..:-:c"":-::_:-::,,~::-:::_::':••::-.,,:::-:~.=-" ...-.v."..=-=_:-::...=-=...--...---'
_=_'::':'~,-= _
Photo Number

.. __ ...................................... -----------_._--- ................................... -. :

Figure 4 Comparison between adjusted Figure 5 Comparison between GPS

PNAV and multi-station GEONAP-K and AT derived projection centers.
solutions

Acknowledgment

This work was supported by CNR - Gruppo Nazionale di Vulcanologia (Italy).

References

Achilli, v., Baldi, P., Baratin, L., Mulargia, F. (1994). First results in applying
aerophotogrammetry to Volcanology. Cahiers du Centre Europeen de Geodynamique et
Seismologique, 8, 183-189.
Ashtech. (1994) - Precise Differential GPS Navigation and Surveying (PNAV) Software
User's Guide, Pre-release version 2, Ashtech Inc., Sunnyvale, California, 316 p.
Colomina I. (1993) - A Note on the Analytic of Aerial Triangulation with GPS Aerial
Control, Photogrammetric Engineering & Remote Sensing, Vol.59 , No.l1, November
1993, 1619-1624.
Hothem, L., K.J. Craun, and M.A.Marsella (1994). 110perational Experience with Airborne
GPS-Controlled Photography at the U.S. Geological SurveY,11 Proc. of ASPRS and
MAPPS Conference, Washington, D.C., pp. 51-63.
TerraSat (1994). 11TOPAS Turbo 3.3 User's Manual,11 Hohenkirchen, Germany, 96p.
Seeber G., Boder W., Goldan H, Schmitz M., Wtibbena G. (1995) Precise DGPS
Positioning in Marine and Airborne Applications. In GPS Trends in Precise Terrestrial,
Airborne and Spaceborne Applications, lAG Symposia 115, Springer, 202-21.
Wtibbena G. (1989) The GPS adjustment software package GEONAP - Concepts and
Models, Proceedings of 5th International Geod. Symp. On Satellite Positioning, Las
Cruces, NM, 452-461.

348
 SIGNAL DISTORTION IN HIGH PRECISION GPS SURVEYS

F .K. Brunner and H. Hartinger

Department of Engineering Surveying and Metrology
Technical University Graz
Engineering Surveying and Metrology
Steyrergasse 30
A-8010 Graz

Abstract

The aim of this paper is to show that GPS signal distortions can significantly influence the
accuracy of the results of a GPS survey. Signal distortions may occur if the satellite
signals propagate through a diffuse obstacle such as a bush and the distorted carrier phases
arrive at the GPS antenna. As a result errors in the horizontal position of up to 2 cm and in
the vertical position of up to 5 cm can occur. The results of several experiments will be
shown. Different GPS choke ring antennas and receivers as well as different GPS
processing software packages were used in order to prove that this effect is hardware and
software independent. The signal distortion was evident in all experiments. The detailed
investigation of the double difference phase data allowed the removal of the affected data
which subsequently yielded the correct GPS results.

Kinematic GPS result of the railway track survey result in mm against arc length [m] of
the track:

GPS result of railway track survey with GPS result of railway track survey after
signal distortion. signal distortion has been removed

Submitted to Survey Review

 TESTING GPS APPROACHES FOR CIVIL AVIATION
IN THE SWISS ALPS

A. Geiger, M. Scaramuzza, M. Cocard, H.-G. Kahle (Geodesy and Geodynamics

Lab, Swiss Federal Institute of Technology, [email protected])
H. Lang, R. Aebersold (Swiss Federal Office of Civil Aviation)
B. Meier, A. Dose (Crossair, Switzerland)

Abstract

To analyse and test the feasibility of GPS-aided approaches for civil aviation in the Swiss
Alps, an internationally co-operating group, leaded by the Swiss Federal Office of Civil
Aviation (FOCA) has been set up. During several test flights and dedicated ground
missions various problems, such as satellite visibility, multipathing, GPS signal
interference and the influence of the topography on the navigation and the GPS system
were addressed. The tests showed the potential of satellite based systems for approaches,
in particular for use in rugged terrain.

The Project

The Swiss Federal Office of Civil Aviation (FOCA) has initiated a test program to gain
insight into the use of GPS for approach and landing in rugged terrain, to gather data for
reliable statistical analysis, and to draw conclusions on certification aspects. To reach this
goal a group consisting of FOCA, Crossair, Swisscontrol, and the Institute of Geodesy and
Photograrnrnetry of the Swiss Federal Institute of Technology (ETH) has been set up. The
project is structured as a joint investigation with the co-operation of additional parties,
such as the Technical University of Braunschweig, Telematica (Germany) and
EUROCONTROL. The airport of Lugano-Agno in the southern part of the Swiss Alps
was chosen as the test area because of its rugged terrain surrounding, thereby representing
a demanding flight technical environment. The flights are primarily analyzed in view of
the Required Navigation Performance (RNP) parameters, the accuracy, integrity,
availability and continuity (Geiger et aI., 1996).
Test Setup

In order to analyze the flights for the Required Navigation Performance (RNP)
parameters, the accuracy, integrity, availability, and continuity a special data acquisition
unit has been designed by the Swiss Aircraft Factory. This Airborne Data Acquisition and
Recording System (ADARS, Fig. 1), which collects more than 100 different flight
parameters (about 80 different ARINC labels, including inertial data, FMS, avionics-GPS
data) and GPS data from an additional GPS-measurement unit has been installed on a
SAAB 2000 aircraft (Fig. 2). At the same time data is acquired at the ground reference

Fig. 1: The data acquisition unit has been especially designed by the Swiss Aircraft Factory
for the use in civil air carriers.

Fig. 2: Crossair's SAAB 2000 test carrier on the approach to the airport.

352
station. The concept of collecting range and phase data from an additional GPS equipment
allows the comparison of different positioning methods. The additional GPS data is treated
in a post-processing mode reaching an accuracy below the 10 cm level, and is, therefore,
considered to provide the true path of the aircraft. By comparing the true path with the
desired course the Total System Error (TSE) can be estimated. Since also the operational
navigation solution is acquired the Navigation System Error (NSE) and the Flight
Technical Error (FfE) can be determined (Fig. 3). These errors are put in relation to
tunnel width (area around the desired course in which the aircraft path has to stay) , which
defines the accuracy requested for specific segment of the approach. To certify the
navigation system the NSE will be of most interest while TSE is of interest for the
qualification of the total approach.

ACFT true position

tunnel width

Fig.3: Definition of the nomenclature of errors (NSE = Navigation System Error; FTE =
Flight Technical Error; TSE =Total System Error).

Special investigations

During the tests several problems have been encountered. Operational constraints and
flight technical aspects have to be considered as well as electrical and software
engineering solutions have to be found. Three of the geodetic ally interesting special
investigations are described in the following.

Terrain effects

Terrain can cause problems to avionics even if the planned track follows a save altitude
and is feasible for (D)GPS guidance. The ground proximity warning is controlled not only
by the height above terrain (clearance) but also by the clearance change rate. Figure 4
shows a track with enough clearance but too high rates. This approach will therefore not
be feasible for operational use. Detailed pre-calculations can be done by using high
resolution DTM (digital terrain model), such as the DTM of the Swiss Topographic office
(25m).

353
 Fig. 4: Approach path over rough terrain causing problems to avionics because of steep
slopes.

Mu/tipathing

Multipath causes erroneous pseudodistance measurements. The error, if not technically

mitigated, may reach several meters (code measurement). The advantage of using
multifrequency systems could also be studied (Cocard and Geiger, 1995). Since DGPS
will not help in reducing the effect, multipathing may cause major errors in position.

4m

3m

2m

1m

.<::
'&
."5
'"
Om

E
· 1m

-2m

-measured
·3m
-modeled

·4m+--r~--~~~--~~~--r-.-~~r-~-r-.
3.0 rad 3.5 rad 4.0 rad 4.5 rad
angular length

Fig. 5: Residuals of GPS CIA-Code measurements (SV 21) affected by multipath. The
smooth curve represents the result of the modeling with multiple reflectors. (X-axis:
topocentric along track angle of satellite)

354
First attempts to model the impact of multipath on pseudodistance measurements have
been made. Multipath residuals are calculated by combining carrier and code phases. By
modeling of the correlation process and by assuming a given reflector configuration it is
possible to approximately reconstruct the multipath residuals.

Interference

GPS is vulnerable to disturbing and interfering radio frequencies, which can considerably
decrease the signal to noise ratio of the GPS receiver. In extreme cases, to the GPS
satellite signals may completely be lost. Regions of interference (low signal to noise ratio)
have been detected by measurement flights carrying spectrum analyzer and GPS-receivers
(Scaramuzza and Geiger, 1995). In figure 6 the measured signal to noise strength is
depicted. The value is represented in dependence of the elevation and the azimuth with
respect to the location of the interfering source. By these measurements two distinct
jamming radiation beams have clearly been detected (regions of low sin, azimuth 25
degrees and 100 degrees, respectively).

Fig. 6: GPS signal to noise ratio around an interfering source. The data has been acquired
by Swisscontrol measurement flights in southern Switzerland (Schulte, 1995).

355
Conclusions

The described project shows the possibilities to implement geodetic methods into non-
classical fields of geodesy. Precise kinematic GPS positioning, parameter estimation and
modeling techniques are applied to flight technical problems of operational relevance. The
first tests showed in principle the feasibility of the methods, however, different
problematic areas have been detected, such as operational tasks, interference or
topographic effects. These problems have to be solved before the definitive feasibility can
be shown. The tests are now on redesign status and shall be continued for further
statistical investigations in the near future.

References

Cocard, M., A. Geiger (1995): The importance of multifrequency systems for future
GNSS. ISPA95, International symposium on precision approach and automatic landing.
Deutsche Gesellschaft fur Ortung und Navigation. p. 475-481.
Geiger, A., Scaramuzza, M. and Cocard, M., 1996. GPS guides civil aviation in Swiss
Alps. Royal Institute of Navigation, Navigation News November/December 1996: 13 - 14,
London.
Scaramuzza M., Geiger A. (1995): Use of Digital Terrain Models for Detection of
Potentially Interfering Zones, Proceedings to 'GPS Interference - is it a Problem', London,
12.-13.10.1995, p. 145 - 150.
Schulte M. (1995): GPS Interference Search around the Swiss Lugano/Agno Airport,
Proceedings to 'GPS Interference - is it a Problem', London, 12.-13.10.1995, p. 151-165.
Scaramuzza, M., Geiger, A., Lang, H., Aebersold, R., Dose, A., Meier, B., Kummer, H.
and Huwiler, M., 1997: Lugano-Trials: Experiences and First Results. EUGIN, DGON,
GNSS 97, Proceedings.

www.geod.ethz.ch/GGL

356
 STATIC AND KINEMATIC POSITIONING WITH GPS
FOR THE CONSTRUCTION AND MAINTENANCE
OF HIGH SPEED RAILWAY LINES

Heribert Kahmen
Dept. of Engineering Geodesy
Technische Universitat Wien
GuBhausstraBe 27 - 29
1040 Vienna, Austria
Fax: +43 1 5042721 E-mail: [email protected]

Abstract
In Europe the densification and improvement of the high speed railway network has been
made apriority, as it is generally accepted that railroads are most efficient for transporting
people. In the near future high speed trains will connect all major towns with a speed of
about 200 kmIh. The Dept. of Engineering Geodesy, Technical University in Vienna, was
involved in the development of fundamental concepts concerning geodetic networks and
rail track alignment methodes.

The investigations have shown that the fundamental technical geodetic networks should
be established in four steps. For a most economic organization of the measurement
sessions simulation calculations have to be performed in advance. Approximate coordi-
nates taken from a map and IGS orbit data were used.

For the railway track alignment a carriage was constructed carrying the rover of a DGPS
system and other electronic sensors. The alignment calculations can be performed on-line,
using algorithms based on Wiener filtering procedures.

Design of the networks

The fundamental concept of the network was applied to the design of a special 45 km long
section of the new high speed railway line Cologne - Frankfurt in Germany (Kahmen, et
aI, 1998).

A maximum speed of 300 kmIh shall be possible. That means that the networks have to
satisfy the highest requirements concerning accuracy and reliability. In summary, the
following were the major points of the network design:

- A fundamental network should provide points in the proximity of the railway line,
whereby the spacing between them should be about 1 km.
- The relative accuracy should not exceed 1 cm.
- The fundamental network should be densified by traverses implemented on both sides of
the railway line, whereby the spacing of the points should be about 200 m and the
relative accuracy 1 cm.
- The network of tunnels and bridges should be integrated into the fundamental network.

It was decided to implement the network in four steps. The fundamental network should
be established by GPS surveys in two steps, the densification with traverse nets by
terrestrial surveys in a third step and finally by integration of the nets for tunnels and
bridges in a fourth step. The main reasons for the hierarchical concept were:

- The connection with the national datum should be performed using a homogeneous
network.
- The transformation parameters between the GPS network and the national network
should be derived from a network of the highest accuracy.
- The points along the railway line should be of a high relative accuracy.

The fundamental network (step 1) consists of 12 points (Fig. 1).

Ml Control Points of National Network

• Modular Points

-Session I
- - • Session II
_ .- Session III
••• • Session IV

Fig. 1: Fundamental network (1st order)

Six points belong to the national network. The other points were established - referred to
here as modular points - along the railway line with an average spacing of 7.5 km. These
points could then be considered control points for the 2nd order network.
For the GPS surveys 6 receivers were used. The planning of the surveys was based on
simulation calculations (see section 3). Four sessions were required each lasting 5 hours.
For the least squares adjustment of the network (step 1) the vectors between the modular
points, being measured during step 2, were used in addition. Thus they could be
considered linearly independent.
The linear networks of step 2 were tied to the modular points of step 1. Every of those
networks consists of 8 additional points with a spacing of 1 km (Fig. 2). For economical
reasons the planning of the surveys was also based on simulation calculations (see section
3).

358
 Receiver 3 Receiver 5
M3
---+--- - _
_ - _.. ....l!!!!.... 4_ _ _ _ M4
• Modular Points
Ref. Recv . 1 4 .. Ref. Recv . 2 km-Points
Receiver 4 Receiver 6

-Session I
- - - Session II
- . - Session III
---- Session IV
Session I - IV

Fig. 2: Network (2nd order)

Six receivers were used, two of them occupied modular points (temporary reference
stations) and the rest of them visited the unknown points. Four sessions were required each
lasting 30 minutes.

The design of the densification network (Fig. 3) was also based on simulation
calculations. Two traverses connect the I-km-points of the fundamental network.
Simulation calculations showed that it was necessary to add diagonal connections to
increase the reliability of the network.

Borders of Project Area

P001 P002 P003 P004

7.~~:====Z!---~
·-------"'KO~ km-Points
P005 pooe P007 P008
• Cr Traverse Points
Fig. 3: Densification network (3rd order)

Simulation calculations

The accuracy and reliability analysis of the network, which had to be performed in
advance, to organize the sessions in a most economical way, was based on a simulated
least squares adjustment. For these calculations the geometry of the network and the
stochastic parameters were needed. One part of the geometry was given by coordinates of
fixed points of the national network. The rest of the information was taken from
topographical maps showing the new points. While the a-priori standard deviations of the
measurements were known from the operating characteristics of the GPS receivers the
covariances had to be derived from simulated observations.
At first the coordinates of the network (step 1) available from the national network and
from a 1 : 200,000 map were transformed into the "ITRF-System". Then, for the
simulation of the observations, the GPS-Software Bemese 4.0 was used. Ephemeries and

359
earth rotation parameters were provided from the CODE-Centre Bern. The reference time
was chosen in such a way that the configuration of the satellites was comparable with the
configuration during the measurements.

The simulations showed, that after an observation time of 5 hours, baselines up to 23 Ian
in length long would have an accuracy of ± 3.7 mm (1 0'). Within an observation time of
only 3 hours the accuracy decreased towards 6 mm. This accuracy was considered too
small, as for the calculations based on real measurements, unresolved ambiguities and
cycle slips had to be taken into account. Observation times longer than 5 hours did not
really improve the results. Consequently an observation time of 5 hours was prescribed for
the sessions.

The relative accuracy of the modular points is described by relative coinfidence

ellipsoids; their horizontal components did not exceeding 3 mm. Consequently we could
see hat enough clearance was available for unforeseen difficulties during the measurement
procedures.
The simulated least squares adjustment of the network (step 2) was performed in a
similar way. In that case, however, the observation time of one session was estimated a-
priori with 30 minutes, based on past GPS experience in the Department.

The results obtained after the practical measurements confirmed, that GPS surveys can
be planned with simulated least squares adjustments.

Integration of the fundamental network with nets for tunnels and bridges
For high speed railway lines a large number of tunnels and bridges have to be built in
mountainous areas.

A tunnel net normally consists of at least 10 control points (Fig. 4a):

- one control point at each portal end section

- three control points about I Ian away from the portal points, to define the orientation of
the tunnel traverse
- two control points on both sides of the tunnel.

The only condition for the design of the network was that the standard deviation of the
cut through error should not exceed 1 cm.
,
,
,

-~~:::. ....,::::...
,
-~:~,.~~ . . ...?
.
-.-,_..: ..:~~~~rf:·~:·~·;;::;,,[j
, _\':'_~:~,:,~~~~---~';'-~:~._.-~~;~_~--O~> \. '
-";~-'~ :~;__ .~ ' '' ' ~
...........
or

.o .;"-:, _ ... . - ••.• -•.; "'~~... \ '

~ ~"
•• ••• //" _ s.slkN\ I

.... , /" --· Set;alonll

--- Session II
'., .... / - , - Sliuion III
_ . - Session III
••• - Seu.lon rv
--- - Session rv

Fig_ 4a: Design of a tunnel net and the Fig. 4b: Design of a bridge net and the
planning of the measurement planning of the measurement

360
 The control points at each portal end section and on both sides of the tunnel are needed
to transform the local tunnel net into the national network. For the points at the portal,
modular points or I-lan-points can be chosen. The two points on both sides of the tunnel
have to be measured together with the GPS surveys ofthe network (step 2).
For the construction sites of bridges networks of high accuracy are also needed. The
relative accuracy of the network (Fig. 4b) should not exceed 0.5 mm. The number of
points depends on the length of the bridges and the topography. The network normally
consists of squares with diagonals. Four control points are needed to transform the local
network into the national network. Two ofthem can be modular points or I-km-points, the
other two are positioned on the left and right hand side of the bridge. A six parameter
transformation is chosen, the scale factor is normally kept fixed.

The GPS-surveys of the tunnel and bridge-network follow in a fourth step. Examples of
the observation plans are depicted in Fig. 4a and b.

Mathematical models for the maintenance of high speed railway tracks

For track-surveying special railway vehicles will be used in the near future, carrying a
multi sensor system for data acquisition. A research project of the Department of
Engineering Surveys ofthe University of Technology in Vienna showed, that GPS surveys
can be used for the positioning of the measurement vehicle if Wiener filters are applied for
the data evaluation procedures.

The measured data and the railway track can be described by signals, whereby three
signal components have to be distinguished (Fig. 5). Mathematically the signals can be

y
actual alignment
a
~ _ _ _'-o
o a

o
-------
deSigned track ____ - - - -

------ ----
-----
~-------­

x
Fig. 5: Model of the Wiener filter evaluation procedure

described by the fundamental equation of the Wiener filter:

!=Ax+~+!! (1)
where! is the observation vector, Ax Is the function of the best fitting alignment, s
describes the differences between the actual alignement and the best fitting alignment ana
n is a noise vector.

361
 The function comprises elements, normally used for the alignment, such as straight lines,
circles or clotho ids. It must not conform to the trackdesign, however maximum or
minimum alignment parameters and certain shifts of the rails per pendicular to the railway
axis are not allowed to exceed specified limits. There should not be any shifts at the
beginning and end of the surveyed track. This is met by increasing the weighting of the
measurements at the beginning and the end of the track section in the Wiener filter
evaluation procedure.

The evaluation process consists of 3 steps (Retscher, 1996).

In step 1 single track elements are computed. We get parameter vector ~ from:
~= (AT Qu- 1 ~-l AT Qu- 1 ! (2)

where Qu = Qu + Qs is the cofactor matrix. The cofactor matrix Qu of the measurements

and Qs of the signal can be estimated using the correlation functIOn:
(~lr
C(~l) = Coe a (3)

a can be estimated from a ~ (2/ 5)A, where A is a mean value of the wavelength of sand 1
the length of the elements. - -

The elements are connected in a 2 step procedure. Then certain conditions must be met

Xi (1) = Xi+! (1) = X

Yi (1) = Yi+l (1) = Y
Zi (1) = ~+l (1) = Z (4)
T ll) =T i+l(1) =T
K (1) =K i+l(1) =K

where T is the azimuth of the tangents, K the curvature.

In step 3 finally the signal ~ is computed:

~ = QsQu- 1 (! - Ax) (5)

It contains the corrections (shifts) for the rail straightening machine. Practical results
confirm that the high relative accuracy of 1. .. 2 mmllO m for the construction and
maintenance of high speed railway tracks can be met (Retscher 1996).

References

Kahmen, H. and Wunderlich Th., Retscher, G., Kuhn, M., Plach, H., Teferle, F. N.,
Wieser, A.: Ein modemes Konzept zur Absteckung von Hochgeschwindigkeitstrassen.
Allgemeine Vermessungs-Nachrichten 1998.

Retscher, G.; 3D-Gleiserfassung mit einem Multisensorsystem und linearen Filterver-

fahren. Geowissenschaftliche Mitteilungen. Heft 44. Institut fUr Landesvermessung und
Ingenieurgeodasie, TU Wien, 1996.

362
 KINEMATIC GPS FOR ICE SHEET SURVEYS IN GREENLAND

K. Keller, R. Forsberg and C.S. Nielsen

National Survey and Cadastre (KMS)
Rentemestervej 8; DK-2400 Copenhagen NV; Denmark
([email protected])

Abstract

This article presents results of repeated GPS surveys on the Greenland Ice Sheet for
studying ice elevation changes, and for providing ground truth data for evaluation of
remote sensing methods (e.g. satellite altimetry and SAR interferometry). The surveys
have been carried out at several central sites of the Ice Sheet, as well as on local marginal
ice caps, which are more sensitive to coastal climate changes. Kinematic GPS techniques
have been used for surface profiling using snowmobile traverses and airborne laser
altimetry, providing good agreements at the sub-metre level. Static GPS repeated surveys
yield sinking and strain rates which are in good agreement with glacilogical mass balance
models, indicating no major shrinking or growth of the Ice Sheet.

Introduction

This paper presents some of the data from the first year of the Danish project called
ECOGIS (Elevation Changes Of the Greenland Ice Sheet), undertaken by:
1) National Survey and Cadastre - Denmark.
2) University of Copenhagen, Geophysical Dept.
3) Technical University of Denmark, Danish Center for Remote Sensing (DCRS).
In this project we will study the elevations and their changes over the Greenland Ice
Sheet on a multi-disciplinary basis, employing geodetic field measurements, satellite radar
altimetry, and synthetic aperture radar interferometry (InSAR) (Nielsen et aI., 1997). The
project focus will be on evaluating the accuracy of the emerging remote sensing height
measurement techniques, and to provide base measurements of the Greenland Ice Sheet
topography for future early detection of height changes. We will continue and verify
earlier change measurements at the Greenland Ice Core Project (GRIP) drilling site at
Summit. Comparisons of previous and current satellite altimetry results will also be done.
 The field work during 1996 was
carried out at 5 sites III Greenland
(figure 1):
Site 1): Airborne laser altimetry and
ground truth ski-doo traverses were
done at the camp of the North
Greenland Ice Core Project
(NGRIP) placed on the North
Central part of the Ice Sheet
(75N;42W; elev. 2920 m a.s.l.).
At the following two ice cap stations
airborne laser altimetry were carried
out:
Site 2): Sukkertoppen Ice Cap, a small
(25 krn by 25 krn) ice cap located at
West Coast (66N;52W) some 150
km west of Kangerlussuaq (S!1\ndre
Str!1\mfjord).
Site 3): The Geikie Plateau, also a
small (25 krn by 75 krn) irregular
ice cap with a very undulated
topography located at the East Figure 1. Map showing the 5 sites of fieldwork.
Coast (70N; 26W).
At Site 4), Saddle North, which is the
topographic saddle point between the main Ice Sheet and the south ice cap kinematic
GPS data were collected on short traverses. These data will not be treated here.
Site 5): The deep ice core drilling site GRIP at the Summit of the Greenland Ice Sheet
(72.6N ;37 .6W).

Survey methodology

The primary goal of the geodetic surveys was to provide precise GPS ties between
reference poles and coastal benchmarks, in order to monitor overall ice sheet changes. In
addition local surface elevations and strain rates were measured to monitor ice volume
changes and flow, and to provide the necessary height information within ERS-l satellite
altimetry footprints to gain an understanding of the return waveforms.
At the GRIP site at the center of the Ice Sheet a fiberglass pole has been frozen into the
ice sheet at 80 m depth, and monitored by GPS on a yearly basis since 1992. Table 1
shows the height of the top of this pole, and the corresponding height above the snow
surface. The pole on the average sinks by 22.4 crn/year, but this sinking is offset by the
accumulation of new snow, so that the overall ice sheet at this point is in balance. This is
also confirmed by glaciological modelling of the ice flow based on the GPS strain net
established in the area of the GRIP site (Hvidberg et al; 1997), cf. Fig. 2.

364
Table 1. Movement of the reference marker at
r-----I----:+----l------il1'!2S
Summit (1992-1993-1994-1996). 1)~ n

Annual change in position in cm.

Year Latitude Longitude Height
92-93 + 7.4 cmN +24.0cmW - 28.7 cm
93-94 + 12.1 cmN + 16.7 cm W - 22.0 cm
94-96 +9.2cmN +25.7 cm W - 19.5 cm
Mean + 9.5 cmN +23.0cm W - 22.4 cm

The kinematic survey of the ice elevations

u~oo "==---:
."6,oo---.-;
,,.6:
...- --.-;;l,):=,._ _ =' '2~0Ct
were carried out by ski-doo traverses, where
a GPS receiver is mounted on a sled, with a
typical traverse speed around 25-30 km/h. Figure 2. Strain rates and surface elevation
contours at the GRIP site, summit of the
Surveys were done relative to a central Greenland Ice Sheet, as determined by repeated
reference receiver using Trimble 4000 SSI GPS surveys and surface surveys. Contour
receivers, processed by the GPSurvey or interval 5 m, arrows show annual ice movements.
Geotracer software, and using IGS orbits.
Results are tied to ITRF94 through the
coastal GPS links.
To augment the surface surveys, additional airborne surveys were carried out by laser
altimetry. A Greenlandair Twin-Otter was mounted with an Optech laser altimeter and a
Trimble 4000 SSI unit, and kinematic GPS surveys were carried out over different areas
with baselines up to 150 km. Since we did not have a proper INS at our disposal, a rough
first order attitude correction of the laser was carried out using a dual-axis inclinometer
(HP Planartechnik NS-45). By combining the horizontal accelerations measured by the
inclinometer with the GPS-derived horizontal accelerations, a first-order estimate of roll
and pitch can be obtained, and the GPS co-ordinates can be transferred to the footprint of
the laser, provided an assumption is made on the heading (assuming zero yaw). The
inclinometer corrections work reasonably well when flying low, but with larger elevations
over rough ice these are not sufficient, as outlined below. We are therefore currently
preparing a new custom-made INS unit based on Litef fibre-optic gyros and Schaewitz
accelerometers. All data are logged on a laptop PC, using an in-house logging program
designed by Jon Olsen of KMS.

The 1996 survey at NGRIP

Numerous geodetic surveys were carried out at the NGRIP camp during the 1996 field
season. For ice modelling and mass balance studies a double 8 pole strain net was
established at distances of 10 to 50 km from the camp. This was put up partly using Twin-
Otter support, and partly by daily ski-doo traverses having distances from 50 km to 170
km. A second purpose of the traverses was to collect topographic ground truth data for Ice

365
 Sheet monitoring. Furthermore gravity
measurements were made each 10 km
along the traverses. These height data were
collected using kinematic GPS mounted on
a sledge carrying all kinds of equipment,
including a complete polar survival kit
(figure 3.)
The GPS was processed using the
Figure 3. Ski-doo GPS survey setup ("strap-down" software Geotracer which gave quite good
GPS). results. The overall accuracy of the GPS
solutions is estimated to be approx. 5 cm,
but due to small bumps and the antennas
mounting on top of a sledge no more than a precision of 10 cm can be expected in the final
heights measurements. Cross-over analysis shows that the 10 cm level has been achieved
at the ski-doo traverses. The data from airborne laser altimetry is consistent with these
measurements and all together the data are combined in a height model of the area around
the NorthGRIP camp, as indicated in figure 4.

-43.0 -42.5 -42.0 -41.5 -41.0

75.2 ~-----4-----j 75.2

75.1 75.1

75.0 75.0

74.9

Figure 4. NGRIP height contours with data tracks, c.i. 2.5 ffi.

The surveys at Geikie Plateau

On July 16th and 17th 1996 two surveys over the Geikie Plateau were flown using
Constable Pynt as base. Both survey tracks are shown on the map (Figure 5). The first one
was mainly a reconnaissance flight which also was used to put out a depot containing radar

366
reflectors and fuel for the operation in
August. The second flight was a survey
flight only and lasted for almost four
hours. As reference for the survey a
permanent pole was put in the gable of one
of the buildings at the airport and the point
was tied to the ITRF94 system. When we
returned in August we found that the pole
had been adopted as a windbag for the
airport and a new reference pole had to be
put up!
I ,

lit _ 1:.' fiCIOQCI ~'kIIIi " '\JQl..Jjlft'l ID4

On August 18th and 19th an other flight Figure 5. Flight tracks over the Geikie Plateau ice
was made to the Geikie Ice Cap for cap.
assembling and posItIoning radar
reflectors, to provide ground truth data for
an airborne synthetic aperture radar (SAR) campaign, originally planned for late August.
The SAR interferometry survey was going to provide digital elevation models of the
region using the DCRS EMISAR system, mounted on a Royal Danish Air Force (RDAF)
Gulfstream G-3 jet. Due to a tragic air crash in the Faroe Islands the RDAF postponed all
scientific Greenland flights of the remaining G-3's, so the reflectors were never actually
used as SAR targets, and may now be considered lost due to excessive snow accumulation
in the area (estimated from the snow at the depot to be 4 m/year). Besides the depot and
the four reflector sites three other sites were marked and positioned, among them a site
which were considered as the top of the ice cap at an elevation of 2272 m located at 69 0

56' 11" N; 25 0 31' 47" W.

With the absence of airborne SAR, studies were done in the region on comparing ERS-
112 tandem mission satellite SAR interferometry (Nielsen et aI., 1997). The airborne SAR
flight took place August 1997 as well as repeated laser measurements, and employed 4
new radar reflectors.

Comparisons of the surface and airborne GPS surveys

As an evaluation of the collected data a cross-over analysis was performed using different
kinds of observations. The results of this
analysis can be seen in table 2. At NGRIP Table 2. Results of the cross-over analysis.
the ski-doo measurements fit together very
nicely. The overall fit is biased by 7 cm and IX-over analysis ~ean diff. Istd. Dev.
the std. dev. is 10 cm which is very good NGRIP GPS vs. GPS 0.07m O.lOm
considering that the data were collected in
NGRIP GPS vs. Laser -0.38 m 0.33 m
two weeks, where snowfall had occurred
during the middle of the period. Further Geikie laser vs. Laser -0.02 m 1.95 m
more was the sledge moving somewhat up Runway GPS vs. Laser 0.05 m 0.40m
and down due to small sastrugas on the
surface, which also cause small irregularities in the data. These surface measurements fit

367
quite well with the airborne observations, even though there is an unexplained offset of 38
cm. The std. dev. is 33 cm.
At Constable Pynt the runway was used for validation of the laser altimeter heights and
compared with a kinematic GPS survey using one of the airports trucks. It gave an overall
difference of 5 cm and a std. dev. of 40 cm, which is satisfactory. The analysis of the data
over the Geikie PI. shows not as good results as on the runway test. The measurements
have an overall difference of 2 cm but a std. dev. of 1.95 m which is not satisfactory. The
highest discrepancies are seen in outer regions of the plateau where the slope is very large.
It should be noted that the wind was very strong during the flight and caused a lot of
turbulence.

Conclusions and future plans

On-The-Fly kinematic GPS processed using standard software is working very well both
in an environment of low dynamics (ski-doo), and in a high dynamic environment such as
an aircraft even on longer distances up to 200 km. The overall accuracy is believed to be
within 10 cm. Results might be improved by using software capable of processing data
from multiple reference stations. Integration of GPS and INS will also further increase
precision, and this will be implemented in future surveys.
The airborne laser altimetry shows satisfying results over smooth areas such as the
Greenland Ice Sheet, but the results from the Geikie Plateau indicate that INS data is
required to resolve pitch, roll, and yaw over high undulating surfaces in order to determine
the footprint of the laser measurements.
The purchased INS will also permit automatic GPS time-tagging of all the
measurements, something which was not done in the current survey hence introducing
another uncertainty, even though the laser altimeter data were correlated with the GPS data
to check for any leap seconds.
The future plan is to repeat and extend the surveys of the sites shown at figure 1. Results
from these surveys will then give input to a first estimate of the mass balance at the
NorthGRIP area and other sites where e.g. height models will be improved due to further
development in airborne GPS and InSAR height measurements.

References
C.S. Hvidberg, K. Keller, N.S. Gundestrup, c.c. Tscherning and R. Forsberg, 1997: Mass
Balance and Surface Movement of the Greenland Ice Sheet at Summit, Central Greenland,
Geophysical Research Letters, Vol. 24, No. 18,2307-2310.
C.S. Nielsen, R. Forsberg, S. Ekholm, 1.1. Mohr, 1997: Merging of Elevations from SAR
Interferometry, Satellite Altimetry, GPS and Laser Altimetry in Greenland, ESA-SP-394,
Proc. of 3rd ERS Scientific Symposium, in press.

368
A REWEIGHTED FILTERING ALGORITHM AND ITS APPLICATION
TO OPEN PIT DEFORMATION MONITORING

Minghai Jia, Maria Tsakiri and Mike Stewart

School of Surveying and Land Information

Curtin University of Technology
GPO Box U1987, Perth 6845
Western Australia.

Abstract

The Global Positioning System (GPS) has been proved to be a capable surveying tool for
monitoring the stability of steep walls in open pit cut mines. However, due to restrictions
on satellite geometry and severe environment-dependent errors, as well as the uncertainty
of the a priori knowledge of system noise variance, sub-centimetric accuracies using
existing algorithms cannot always be achieved. This paper proposes a reweighted filtering
algorithm which treats GPS deformation monitoring in an open pit environment as a
kinematic surveying system. Estimating techniques of a posteriori weights of predicted
states and observations are developed to match the weights appropriately and obtain
reweighted filter solutions. Two different weight functions to reweight the predicted states
and observations are established based on the state residuals and observation residuals.
The proposed algorithm can automatically assign the larger weights to the more accurate
predicted states and observations and the smaller weights to the less accurate predicted
states and observations. Results from simulated GPS deformation monitoring data are
shown using the proposed algorithm. Comparisons with the traditional Kalman filtering
method indicate high levels of filter stability and accuracy for the proposed algorithm.

1. Introduction

In the open pit environment, GPS can offer several benefits over conventional surveying
techniques. No 'line of sight' measurements between reference and objective points are
required. In addition, the baseline lengths required in open pit deformation monitoring do
not practically exceed lkm and the 'relative' data processing techniques ensure that GPS
solutions remain unaffected by local atmospheric conditions for these baseline lengths.
The sub-centimetric level accuracy required to detect precursor movements to pit wall
failures necessitates the use of carrier phase GPS processing techniques. Given the slow
movement rates of open pit walls, the static nature of the deformation problem can be
treated with an alternative approach, using GPS navigation techniques. In this way, repeat
rapid GPS surveys of few seconds provide data which are integrated with data from
previous surveys. This integration of all available data can be achieved by using optimal
Kalman filtering estimation techniques (Tsakiri et al 1996) and the problem can be
described by a kinematic surveying system (KSS). As open pit deformation networks are
usually established by static control surveys, precise a priori knowledge of the coordinates
of the points is not a problem. As a result, ambiguities can be resolved almost
instantaneously in a process identical to the initialisation phase of a kinematic GPS carrier
phase solution. A subsequent ambiguity fixed solution solving for coordinate changes
produces the deformation solution. Provided sites are re-observed at a controlled time
interval, based on real velocities of the monitoring points, motion and observation error
should be less than one carrier phase half wavelength and will not affect ambiguity
resolution. Using this approach, repeat surveys can produce ambiguity fixed solutions,
theoretically good to better than 4mm (the noise of a raw double difference carrier phase
observations).

Open pits remain one of the most challenging environments for the utilisation of GPS high
precision technology. The main problem, affecting reliable and accurate solutions for
deformation monitoring, is restricted satellite visibility, as monitoring points situated in pits
can be up to several hundred metres deep and may be unable to observe the full available
constellation. The small number of satellites and the resulting poor geometry can severely
degrade an ambiguity fixed carrier phase solution. In addition, environment-dependent
errors, such as multi path, can be the dominant noise source for monitored short baselines.
Because of the small motions of the pit walls, attaining the optimum GPS solution is of
fundamental importance.

This paper develops a new numerical-based method, namely, one-step-reweighted filter

method, to deal with the residual errors in double difference carrier phase observations and
access the a posteriori weights of the observations.

2. Kalman Filter for Kinematic Surveying Systems

The very slowly moving GPS deformation monitoring problem can be treated as a
kinematic surveying system (KSS), which is composed of a state model with the stochastic
disturbances and an observation model with the measurement noises. The description of
the state and the observation model and their associated stochastic models at any epoch k is
well documented (eg Bryson and Ho 1969). Kalman filter equations for KSS can be
derived based on least squares principles (Bryson and Ho 1969). The Kalman filter optimal
estimators (ibid 1969) are,

x= x + J(l- Ax)
Q x =Qx -JDr
J = Q"ATD- 1 (1)

370
d=l-Ax
D= AQx AT +Qe

x
where, is the optimal estimator of the state vector; x is the predicted state vector; I is
the nx1 vector of observations; A is the n x t measurement matrix; J is the gain matrix; d
and D are the innovation vector and its cofactor matrix respectively; Q x , Q e are the cofactor
matrices for x and the vector e of observation noises respectively.
Based on equation 1 a reweighted filtered GPS carrier phase solutions can be derived.

3. One-step Reweighted Method

As mentioned previously, open pit environment-dependent errors and especially multi path
are the major error sources in the monitoring data. These errors make the observation data
from different satellites, with different directions and elevations, present significant
heteroscedastic error properties. For example, the error effect is heavily dependent on the
direction of the arriving signals from a specific satellite to the monitored point, which is
usually very close to the open pit wall. The quantitative modelling of these type of errors is
complex and not within the scope of this paper. However, the magnitude of these
unmodelled errors in the observation data is sufficient to submerge the deformation
information desired for sub-centimetric accuracy. These errors are reflected in the GPS
double difference carrier phase filtered residuals and, based on these residuals, an
appropriate weight can be given to observations obtained from satellites with different
errors. A one-step-reweighted filter algorithm is developed in this section based on the
evaluation of the filter residuals which substantially improve the solution at any epoch.

The GPS data processing technique, implemented in the software developed at Curtin
University, comprises the use of two filter runs, as described in equations (1). The
implementation of the first filter employs identical weights for all observations from
different satellites due to lack of a priori quantitative knowledge of the environment-
dependent errors, even though the observation errors are heteroscedastic. After the
completion of the first filter, the filter residuals, which reflect the properties of the
predicted states and the observation errors, are used to assess the a posteriori weights of
the observations. Then, the filter is implemented again using the new weights to obtain the
final results. In this filter the magnitude ofthe residuals define the assigned weight, in other
words, observations with large residuals are assigned low weights and greater weights are
given to the observations with small residuals. The a posteriori weights of the observations
are expressed as a function of the their a priori weights and filter residuals (Yang 1994, Jia
et alI997),

where, Pij (v i , v) is the a posteriori weight, w ij (v i , V j) is a weight factor function, and

Vi' Vj include both measurement and state vector residuals. The weight factor functions can
be directly defined based on robust estimation (eg, Hampel et al 1986, Yang 1994) and

371
robust filter (Jia et a11997) techniques. These should be continuous, even and non-negative
functions of the filter residual.

Based on the above properties two known weight factor functions, Andrew's, and Tukey's
(Andrew 1974, Beaton and Tukey 1974) are modified

(1) Modified Andrew's Weight Factor Function

wij(v;, Vj) = cos(V; / coO'o)cos(Vj / coO'o)
where 0' 0 is a scale parameter and Co = 0.51t .

(2) Modified Tukey's Weight Factor Function

wij(v;, Vj) = [1-lv;vjIl0'~C2]2
where Co = 4.658 or Co = 6.0 with known or unknown 0'0'

The adopted values for the so-called tuning constant Co are computed based on Hampel et
al (1986). The chosen values for Co result in the 95% asymptotic efficiency on the
standard normal distribution assuming the errors are Gaussian. This is important because in
the one-step-reweighted filter, the values of constant Co should make the reweighted
results on the homoscedatic observation error distribution to be as close to the
unreweighted (ie Gauss distributed) results as possible.

The a posteriori weight matrices Pii and Ii; of the filtered states and the measurements
respectively, based on the above weight functions, can be calculated as,

Px(Vx-.,V xJ.)
I
= Px *W = (Px. w;J.(vx.,v x.))
v_
x U I J

PI (VI.' VI.) = PI * WV1 = (PI.U w;J.(v l ., VI.))

I J I J

where, * denotes the Hadamard product of two matrices with the same dimensions (Rao
1973). Using the a posteriori weight matrices, and based on equation (1), the reweighted
filter solutions are obtained.

4. Numerical Example

The effectiveness of the proposed one-step-reweighted filter algorithm is illustrated with

results from simulated GPS monitoring examples in a situation of poor satellite geometry.
In-house software for GPS simulation was used to generate data for a 30-second data span
with 4 satellites and introduced error on the data from one satellite. The error was
introduced after the 5 seconds to allow the filter to become stable.

372
 0.04
0.04
e 0.035
§ 0.035 'i 0.03
: 0.03 .~ 0.025
""
~ 0.025
~ 0.02
~ 0.02 >
0.015 ...... .
.: 0.Q15
.5
~ ........ .......
~ 0.01
~ 0.01 t
t ~ 0.005
~ 0.005
is 0
is 0
·0.005 5 10 15 20 25 30
.{).005 5 10 15 20 25 30 Epoch (•• c)
Epoch (HC)

0.04
0.035 - - Andrew's (unction
g 0.03 ......... .. Tukcyts function
3" 0.025 - - Unn:wcighlcd mcillod
..,~ 0.02
N
.s 0.Q15
g 0.01
5
~ 0.005
is 0 ·-n"j"·· "j"." ""j"" .• j " ' " •• :

·0.005 5 10 20 25- 30
Epoch (.... )

Figure 1. Comparison between reweighted methods and standard Kalman filter in X, Y, Z

directions for observations from 4 GPS satellites

The GPS data are simulating a pit wall environment with an introduced masking of
elevation angle from 0-55 degrees and azimuth from 90-135 degrees. This masking results
in a data set of observations with only four satellites and hence, three double difference
observations and a DOP (dilution of precision) value of 5.8. An environment-dependent
error of 0.4 of a cycle (approximately 7.6 cm) is introduced to a random satellite (number
27) in every epoch. The differences between the 'truth' and the obtained results using a
standard Kalman filter and the proposed reweighted method are shown in the three graphs
of figure 1. All results refer to solutions in which ambiguities have been fixed. In all three
graphs, the unreweighted standard Kalman filter produces the greatest differences of all,
which reach almost 3.5 cm. The solution from the reweighted method is superior to the
standard Kalman filter (unreweighted) solution for both functions. For the reweighted
solutions the greatest errors are in the Y direction, which reach up to 1.5 cm, and the
smallest errors are in the Z direction which are in the order of 0.1 cm. The differences
from the 'truth' in X and Y direction are the greatest, reflecting a satellite direction effect
resulted by the introduced error at the satellite. A comparison between the two reweigh ted
functions indicate some differences with the Andrew's function tending to produce the
smallest errors. In addition, there is an evident drifting pattern at the errors, mainly in Y
direction, suggesting the influence of the simulated error contaminating the data from one
satellite when observations from only 4 satellites are available. This means that although
there are 3 double difference observations, in fact only 2 are unaffected and can be used
reliably. Overall, however, there is an improvement in all three directions using the
reweigh ted method, two times with Tukey's function and three times with Andrew's
function. When more than 4 satellites are available and an error to the data of one satellite

373
has been introduced, the reweighted method still offers improved solutions despite the
adequate number of double difference observations (more than 3).

It must be mentioned here, that when no environment-dependent errors are included in the
data set, as expected, the reweighted method produces almost identical results to the
unreweighted standard Kalman filter.

5. Concluding Remarks

A filter method, namely a one step reweighted filter, has been developed to improve the
accuracy of the GPS carrier phase solutions for deformation monitoring in open pits,
where environment-dependent error contamination of the observations is likely to occur.
The reweigh ted method reduces the influence of large errors by assigning appropriate
weights depending on the GPS double difference residuals and implementing known
weight functions which have been modified appropriately. The use of simulated tests have
shown that the proposed method is capable of producing the required sub-centimetric
accuracies with 4 satellites despite the existence of some large environment-dependent
errors in the observations. In addition, the use of the reweighted filter suggests its
applicability in restricted satellite geometry with results at the required level. A
comparison between the weighted functions indicates an improvement using the modified
Andrew's function which provides position errors of less than 5 millimetres. Further
testing with real data will enable verification of the obtained results.

6. References

Andrew D F (1974). A Robust Method for Multiple Linear Regression. Technometrics,

Vol. 16, 4:523-53l.
Beaton A and J W Tukey (1974). The Fitting of Power Series, Meaning Polynomials,
Illustrated on Band-Spectroscopic Data. Technometrics, Vol. 16,2:147-157.
Bryson A E and Y H Ho (1969). Applied Optimal Control. Wiley, Inc., New York. ISBN
0470114819.
Hampel F R, E Ronchetti, E M Rousseeuvu and W Stahel (1986). Robust Statistics.
WileySeries in Probability and Mathematical Statistics, ISBN 0-471-82921-8.
Jia M, M Tsakiri and X Ding and (1997). A new robust algorithm for kinematic surveying
systems. International Symposium on Kinematic Systems in Geodesy, KIS97, June 3-6,
Banff, Canada.
Rao CR (1973). Linear Statistical Inference and Its Applications, Second Edition. John
Wiley & Sonc, New York.
Tsakiri, M, M Stewart and X Ding (1996). Robust Kalman Filter Architecture for GPS
Deformation Monitoring. Proc. of the 8th FIG International Symposium on
Deformation Measurements, June 25-28, Hong Kong, pp 275-285.
Yang Y (1994). Robust estimation for dependent observations, Manuscripta Geodaetica,
19: 10-17.

374
 PERMANENT AUTOMATIC GPS DEFORMATION MONITORING
SYSTEMS: A REVIEW OF SYSTEM ARCHITECTURE AND
DATA PROCESSING STRATEGIES

Craig Roberts
Chris Rizos
School of Geomatic Engineering, The University of New South Wales
Sydney NSW 2052, Australia

ABSTRACT
Ground deformation due to volcanic magma intrusion, crustal motion, ground subsidence,
etc., are phenomenon ideally suited for study using GPS. The change in length, height
difference and orientation of baselines connecting GPS receivers in a carefully
monumented ground network can be monitored. This is done by repeatedly measuring the
same baseline components to an accuracy commensurate to, but preferably much higher
than, the expected baseline component changes. Such GPS techniques are based on the
"campaign" principle: the periodic (often annual) re-survey of a network of control points.
However, over the last half decade or so, there has been a growing interest in the
deployment of permanent, continuous GPS monitoring networks. The factors responsible
for this trend include the declaration a few years ago of Full Operational Capability of the
GPS system, and the steady decrease in price, size and power requirements of GPS
receivers. Important additional factors have been the high cost of annual GPS surveys
(manpower, travel, logistics, etc.), as well as the fact that geo-scientific research can be
furthered because of the continuous measurement of a deformation phenomenon, rather
than its periodic measurement.
The GSI network in Japan, the SCIGN network in California, and the SWEPOS network
in Sweden are examples of large scale, continuous GPS networks for near-real-time
crustal motion monitoring. However, smaller scale GPS arrays such as those on the
Augustine volcano (Alaska), the Popacatepetl volcano (Mexico), the Kilauea volcano
(Hawaii), and the Rabaul volcano (Papua New Guinea) reflect a growing interest in local
continuous GPS volcano monitoring systems. GPS is also increasingly used to monitor
engineering structures such as dams, bridges, offshore drill platforms, etc.
Developing an automatic GPS array system for such small scale monitoring applications
is an engineering and software challenge. A network of permanent GPS receivers need to
be deployed, often in an inhospitable environment in which they must operate reliably on
a continuous basis. The GPS observations must be telemetered to a central computing
facility where data processing occurs with minimum delay. Analysis of the time series of
baseline results then takes place in order to detect any baseline component change
between successive solutions which may be a precursor to failure or eruption. What are
the bases of small scale monitoring systems? Often the solution has been simply to
purchase commercial "off-the-shelf" real-time-kinematic (RTK) GPS systems. This is the
high cost option, yet there are several hundred active volcanoes in the world, many located
in the less developed countries, and the cost of GPS monitoring systems must be
significantly reduced if the technology is to contribute to volcano hazard mitigation. An
alternative approach is to develop a system based on single-frequency GPS receivers,
integrated with communications and in-field computer sub-systems.
This paper discusses the characteristics of GPS monitoring networks and considers such
issues as monument design, network design, GPS hardware, communication links, power
supply, and data processing strategies.

INTRODUCTION -- GPS-BASED DEFORMATION MONITORING SYSTEMS

The first applications of "GPS geodesy" have been to measure the relative motion of
tectonic plates and the deformation of active continental margins. These applications have
grown to such an extent that there are now very few parts of the world which have not been
studied (or are not at present under study) using the GPS technology. However, there is an
increasing interest in GPS-based ground deformation measurement on a continuous basis,
operating over small regional extents, using a variety of system configurations, which are
optimised for specific monitoring applications. There are therefore several bases for GPS-
based deformation monitoring systems.

Repeat GPS Surveys

In essence, GPS field observation campaigns are mounted to determine the relative
position of groundmarks at a certain epoch. The data collection typically occurs over a
period of several days or weeks, as teams of GPS technicians occupy the groundmarks for
up to several days at a time, before moving on to other locations. The data is then
collected, archived and prepared for processing back at the office of the organisation
performing the survey. By repeating the survey campaign a number of times, any change
in the baseline components between successive field campaigns is interpreted as a measure
of the deformation. In the context of geodynamical studies, GPS receivers are relatively
inexpensive and highly portable (compared to traditional space geodetic techniques), and
this campaign-mode measurement strategy has been used to good effect for over a decade.
Many other deformation applications are also being addressed using repeat GPS survey
methods. These include the measurement of ground subsidence due to fluid extraction or
underground mining, and the monitoring of engineering structures or volcanos.
Permanent Geodetic Arrays
The large number of geodetic GPS receivers being purchased by organisations around the
world, together with the escalating labour costs of GPS campaigns, has led to an increasing
number of receivers being simply deployed at fixed sites on a permanent basis. Although
many of these stations have been identified with the International GPS Service for
Geodynamics (IGS), many permanent GPS receiver stations are being established in the
form of sub-nets to support deformation monitoring. Hence patterns of deformation
occurring over very short time scales, from a matter of a few minutes or hours, to several
weeks, can be detected and studied. The Southern California Integrated GPS Network
(SCIGN) in southern California (Bock et aI., 1997) and the permanent Geographical
Survey Institute (GSI) network in Japan (Tsuji et aI., 1995) fall into this category. In
addition to establishing the hardware arrays, data processing software has been
implemented to ensure that the continuous stream of GPS data can be "digested". The
software system consists largely of standard scientific software modified to run in "batch"
mode with automatic data file handling and pre-processing capabilities. Further, these
arrays are established in areas where the ground infrastructure is well developed, for
example a situation that is generally atypical of applications such as the monitoring of

376
volcano flanks, micro-faults, ground subsidence due to underground mining or fluid
extraction, slope stability, and engineering structures such as dams, bridges, etc.
The authors contend that permanent GPS-based monitoring systems will be increasingly
deployed across small areas, for a variety of deformation applications.

Deployment of Systems in Emergencies

In the case of volcano monitoring applications, there may be little or no activity over many
years and the permanent deployment of a GPS array may not be justified. However, as
soon as volcanic activity is detected (by whatever means), a range of instrumentation
(including GPS) may be rapidly deployed by the responsible agencies. These systems
must be highly portable and robust in order to function anywhere in the world. On the
other hand, given that their deployment normally takes place after the commencement of
inflation or ground uplift on a volcano's flanks, the sensitivity of the monitoring system
need not be at the sub-centimetre level as the deformations are likely to be significantly
greater than this.

Commercial GPS Surveying Systems

An example of an "off-the-shelf" system are the commercial "real-time kinematic" (RTK)
products developed by several GPS manufacturers. Although marketed for tasks such as
engineering surveys, etc., they can also be adapted for deformation applications. The
following comments can be made:
• The integration of hardware, software, and telemetry sub-systems is well engineered.
• Due to limitations of the telemetry sub-system (generally UHFNHF), and the
"standard" double-differenced observation model that is used in such systems, the
baseline length constraint is of the order of a few tens of kilometres at best.
• The coordinates are derived by the "rover" receiver(s), and hence must be transmitted
back to a central monitoring facility if they are to be used in (near-)real-time.
• Typically there is no software that checks the baseline solutions to determine whether
the observed variability is really a deformation signal or just "noise".
• Many deformational phenomena cannot be characterised as "kinematic" (except
under extreme conditions just before a failure or eruption), but could be considered
to be "static" over a period of one hour or so.

Certainly for many deformation monitoring applications that span relatively small areas,
the RTK solution is the one of "first choice" as there are so few other options. However,
because of its high cost it is likely to be only implemented in a small number of instances.

CHARACTERISTICS OF PERMANENT GPS·BASED SYSTEMS

We will restrict ourselves to permanent systems with the following characteristics:
• A network of well-monumented benchmarks upon each of which a GPS receiver and
various auxiliary instrumentation is installed.
• One station within the network will be designated the Master Control Station
(MCS), operating as the node of the network and where the computations are made.
• The areal extent of the network is of the order of several tens of kilometres.
• Capable of continuous autonomous operation.

Such permanent GPS-based monitoring systems can be categorised according to the

following characteristics: (1) monument design; (2) network design; (3) GPS hardware;
(4) communications links; (5) power supply; and (6) data processing strategy.

377
 The characteristics of individual systems will tend to reflect, on the one hand, the nature
of the deformational phenomenon being monitored, and on the other hand, the capabilities
of the organisations responsible for installing and operating the systems, hence the
following issues have also to be considered:
• Whether the deformation is mostly characterised by continuous, relatively small
changes in the network geometry, or long periods of no deformation followed by
sudden, violent changes in the position of the network benchmarks.
• Whether the deformational signal is mostly horizontal, or in the vertical.
• The physical environment in which the system will be deployed.
• The level of performance required of the monitoring system, e.g. the accuracy, the
tolerable delay in production of the results, and so on.
• The level of GPS expertise within the organisation.
• The budget for equipment, and network construction, maintenance and operation.

Monument Design
The monuments must be stable. However, monuments designed to have stabilities at the
millimetre level are very expensive and restricted to geodynamic networks such as
SCIGN's. For many other applications the primary criteria may be that they are easy to
construct (given the difficult environmental conditions that may be encountered).

Network Design
The siting of receivers is a critical issue, requmng careful reconnaissance and an
understanding of the deformational phenomenon being monitored. Apart from the need to
have stable, well defined benchmark monuments, the following concerns and constraints
must also be addressed:
• The locations must have a clear view of the sky and at the same time facilitate line-
of-sight communications as well as power generation.
• As the system will have to be deployed for long periods, security issues are
significant and the involvement of local organisations is desirable.
• Multipath disturbance from nearby structures should be a minimum.
• If there is already instrumentation installed at the station, advantage may be taken of
these facilities as they may have power, security and communications infrastructure.
• In order to "check" the deformation monitoring system, regular calibration may have
to be conducted by means of periodic precise GPS surveys of the monuments.

GPS Hardware
To date deformation networks have consisted of "top-of-the-line" dual-frequency GPS
receivers. These are also used in commercial RTK systems adapted for deformation use
(Lowry & MacLeod, 1997). However, to significantly lower the hardware costs of future
deformation systems (e.g. by a factor of ten) simpler GPS hardware must be considered.

Communication Links
Data telemetry is possibly the weakest "link" of any system. Without telemetry the data
from the various network stations cannot be transferred to the MCS for processing. Where
the infrastructure is available, telephone lines are, of course, the best data link option.
However, in remote area locations such an option may not be available. The use of
satellite communications technology would be preferable because of its versatility (e.g. the
GPS array would not need to be deployed in such a way as to ensure line-of-sight

378
communications), however it is generally too expensive. What is therefore often used is
VHFIUHF radio technology.

Power Supply
Power is needed to operate the GPS receiver, the computer and the telemetry sub-systems.
However, if the sites are not able to be serviced by mains power, a configuration based on
batteries and a solar recharger will typically be used. Power management will therefore
always be a significant challenge.

Data Processing Strategy

The heart of any GPS system is its software. In terms of carrier phase observation
modelling we may distinguish between "GPS geodesy" and "GPS surveying" type
processing. In the former many additional parameters are estimated, including satellite
orbital elements, tropospheric scale parameters, etc. Such a processing strategy
distinguishes the SCIGN and GSI networks that stretch across hundreds of kilometres. In
the case of small scale systems, such as those based on "RTK-GPS surveying" modelling,
the data processing software generally has the following characteristics:
• "On-the-fly" ambiguity resolution (AR) techniques as developed for use with dual-
frequency receivers.
• Epoch-by-epoch "kinematic" data processing mode.
• The broadcast ephemerides contained within the Navigation Message is used.

However, there are several aspects of the deformation monitoring application that deserve
comment as they do impact on the design of optimised data processing software:
• Because the receivers are more or less stationary, then the baseline components are
already known (certainly to within 10cm), making AR more reliable.
• Several quality control algorithms will need to be implemented to detect bad data.
• Multi-baseline capability is desirable, so that the system can be easily scaled up to
cater for a large number of GPS receivers.

Some Options for Permanent GPS-Based Deformation Monitoring Systems

Some of the defining characteristics and general features of three GPS-based deformation
monitoring systems are summarised in Table 1. The deformation monitoring system may
be custom designed, or based on "off-the-shelf" hardware and software components. If
they are custom designed, then the following features may be incorporated:
(a) True continuous monitoring, or sampled monitoring every hour, or every day, etc.
(b) Single baseline mode, or based on a large network of receivers.
(c) Real-time processing at the base, or "near-real-time" (e.g. on a daily basis).
(d) Employ a kinematic baseline model, or a static baseline model, or selectable.
(e) Full duplex, half duplex or single duplex communications link between stations.

In Table 1, "Type 1" systems refers to the geodetic arrays such as SCIGN's and GSI's;
"Type 2" refers to the commercial RTK-based system such as described in Lowry &
MacLeod (1997); and "Type 3" refers to such custom designed systems as described in
Hein & Riedl (1995), Han & Rizos (1996), Rizos et al. (1997).
Different hardware configurations, operational procedures as well as data processing
strategies will be necessary to address the various deformation monitoring applications.
However, the ideal system must be of relatively low cost, sensitive to the expected
deformational signal and able to accommodate a large number of GPS receivers. All other
design issues are minor in comparison to these. There is a role for GPS-based

379
deformation monitoring systems, but it will only be significant if the major constraints
identified in this paper can be overcome.

ACKNOWLEDGMENTS
The Australian Research Council is supporting a project to "Develop, Test and Deploy a
GPS Array System for Continuous, Automatic Monitoring of Earth Deformations Arising
From Volcanic Activity" (1996-1998). A follow-on project (for 1998-2000) has recently
been approved. The first author is supported by a Trimble Navigation NZ scholarship.

REFERENCES
BOCK, Y., and others, 1997. Southern California Pennanent GPS Geodetic Array: continuous
measurements of regional crustal defonnation between the 1992 Landers and 1994 Northridge earthquakes.
J.Geophys.Res., Vol.l02, No.B8, 18013-18033.
HAN, S. & C. RIZOS, 1996. GPS network design and error mitigation for real-time continuous array
monitoring systems. 9th Int. Tech. Meeting of the Sat. Viv. of the U.S. Inst. of Navigation , Kansas City,
Missouri, Sept. 17-20, 1827-1836.
HEIN, G.W. & B. RIEDL, 1995. First results using the new DGPS real-time defonnation monitoring system
"DREAMS". 8th Int. Tech. Meeting of the Sat. Div. of the U.S. Inst. of Navigation , Palm Springs,
California, Sept. 12-15, 1467-1475.
LOWRY, A. & R. MacLEOD, 1997. PMoS - a real time precise DGPS continuous defonnation monitoring
system. 10th Int. Tech. Meeting of the Sat. Div. of the U.S. Inst. of Navigation , Kansas City, Missouri,
Sept. 16-19,923-927.
RIZOS, C., S. HAN & c. ROBERTS, 1997. Pennanent automatic low-cost GPS defonnation monitoring
systems: error mitigation strategies and system architecture. 10th Int. Tech. Meeting of the Sat. Viv. of the
U.S. Inst. of Navigation , Kansas City, Missouri, Sept. 16-19,909-917.
TSUJI, H., Y. HATANAKA, T. SAGIYA, & M. HASHIMOTO, 1995. Coseismic crustal defonnation from
the 1994 Hokkaido-Toho-Oki earthquake monitored by a nationwide continuous GPS array in Japan.
Geophys. Res. Let., Vo1..22, No. 13, 1669 -1672.

Table 1: General features of three GPS-based deformation monitoring systems.

CHARACTERISTIC TYPEl TYPE 2 TYPE 3
Relative accuracy 10- 8 10-6 10-6
Dual-frequency D-F D-F mixed
Corns link telephone UHFNHF various
Modelling geodesy surveying surveying
sophistication (baseline) (multi-baselin~
Kinematic or static kinematic variable
static model
Solution daily real-time near-real-time
strategy solutions (epoch-by-epoch) (session-by-session)
Centralised yes no, solutions yes
processing sent back to base
Limit to receivers no unknown no
Operational expertise required high low to moderate low
Areal extent > lOOkm 10- 20km 10-100km
Portable no partly yes
Cost/site > US$40K >US$20K US$2-3K

380
 MODIFIED GPS-OTF ALGORITHMS FOR BRIDGE MONITORING:
APPLICATION TO THE PIERRE-LAPORTE SUSPENSION BRIDGE
IN QUEBEC CITY

Rock Santerre and Luc Lamoureux

Centre de Recherche en Geomatique
Universite Laval
Quebec, Canada G lK 7P4

Abstract

Algorithms for On-The-Fly ambiguity resolution have been modified for the deformation
monitoring of a suspension bridge. Instantaneous relative positioning of a deformation
network at a precision of about ±5 mm horizontally and ± 10 mm vertically has been
achieved using GPS LI phase observations. Particular attention has been paid to the
modeling of relative tropospheric delay and to the phase center calibration between
antennas of different types. Moving averages on station coordinates have also been
applied to reduce multipath effects.
The Pierre-Laporte bridge is a 6-lane, 1040-m-Iong suspension bridge which crosses the
St. Lawrence river in Quebec City. Three 48-hour GPS sessions have been conducted
during the months of July and October 1996 and February 1997. For each session, 5 GPS
receivers were observing at a data rate of 2 seconds.
Daily variation in the vertical position of the antenna located at the deck center shows
clear correlation with respect to temperature and vehicle loading. Transverse movement
of the deck center has been monitored and correlated with (transverse) wind speed.
Seasonal variation in temperature caused a vertical displacement of the deck center, a
contraction of the towers as well as a displacement of the towers towards the river banks.

Modified OTF algorithms

In this study, a suspension bridge is seen as a "kinematic" deforming structure but with
limited movement amplitude. The OTF (On-The-Fly) algorithms used to resolve GPS
phase ambiguity, even if the receiver is in motion, have been modified for the
particularity of a suspension bridge. The algorithms have been designed to work with
single frequency receivers. The steps of the evaluation and the validation of Ll phase
ambiguities and least-squares solutions are schematically presented in Figure 1 and
described in the next paragraphs.
To reduce the number of ambiguity combinations, good a priori coordinates must be
available. Unlike dam deformation GPS surveys, one cannot use the coordinates
determined from a previous session because the deformation of a suspension bridge (even
the towers) can easily exceed 10 cm (half of a Ll wavelength) between 2 sessions. The
coordinates obtained from filtered code solutions or from double difference float
solutions, as used in pure kinematic mode, give too many ambiguity sets to be tested. A
solution is to start with a static-like processing of a short data span. The time span must
be long enough to provide a positioning precision of ±5 cm (one quarter of a Ll
wavelength), but within this time period the amplitude of the movement of the bridge
should not exceed 5 cm. For bridge towers this criteria is always respected, but attention
must be paid to the deck
center movement when
there are rapid fluctuations
of temperature, wind speed
or traffic jam during this
"static" session. For the
other epochs, the
coordinates of the previous
epoch are used as a priori
coordinates. With the a
priori coordinates, double
difference ambiguity is
calculated, for each pair of
satellites, starting from the
first epoch of the
observation session, with
the following equation:
Figure 1: Data flow for ambiguity resolution and LSA solution.

V L\N = V L\<I> - V L\p - V L\dtrop

A (1)
If the decimal part of equation (1) departs by more than 0.25 cycle from the closest
integer value (this can be due to multipath or observation noise) two integer ambiguity
values associated to this pair of satellites will be tested. If not, only the closest integer
value will be retained.
Note that the ambiguities are not parameters to be estimated in the least-squares
adjustment (LSA). In fact, different potential ambiguity combinations are tried and tested
to isolate the correct ambiguity set among all possible candidates. Least-squares
adjustment is performed with all ambiguity combinations and 3 criteria are used to
identify the correct ambiguity set:
- the correction to the a priori coordinates must be smaller than 10 cm (in absolute
value);
- the residuals must be smaller than 10 cm (in absolute value);
- the ratio of the second smallest and the smallest a posteriori variance factors must be
larger than 2.
A test is also done to detect potential cycle slips between epochs. In this step, Doppler
data at epochs i and i+ 1 are used to predict phase variation between epochs i and i+ 1, as
follows:
(<1>;+1 - <1>; ) - ( D; +2D;+l) (7;+1 -7; ) ~ A (2)

If the difference (in absolute value) is larger than one wavelength, the observation at
epoch i+ 1 is flagged, a new ambiguity value is calculated with equation (1), and the
validation steps are again checked. Equation (2) is very efficient when the data rate is
high and the receiver is static or moving slowly between the observation epochs.
To avoid mathematical correlation in double difference observations, the Hatch
approach [Hatch, 1990] has been used. With this approach the observation weight matrix
is kept diagonal and the residuals are associated to each satellite, instead of each pair of
satellites. This last feature is very important for a detailed analysis of remaining
unmodeled errors. At every epoch (at each 2 seconds in our case), four unknown
parameters are solved in a least-squares adjustment.

382
GPS error sources and modeling
The main GPS errors to deal with are relative tropospheric delay, multipath and antenna
phase center variation. The effect of other errors such as ionospheric refraction,
ephemerides and Selective Availability (SA) can be assumed negligible for baselines as
short as 1 km. The readers are referred to [Santerre, 1991] for a detailed analysis of GPS
error propagation in GPS network.
Tropospheric refraction: Because some baselines of the monitoring network have
height differences of about 60 m, attention has been paid to the modeling of relative
tropospheric delay. Meteorological data collected at the level of the bridge deck were
extrapolated in altitude for the GPS stations located on the 2 bridge towers. The height
extrapolation algorithms for temperature, pressure and relative humidity are described in
[Rothacher et aI., 1986] and used as input for the Hopfield's tropospheric model.
Multipath: To avoid multipath as much as possible, antennas have been setup above
any reflective objects at the reference stations and on the top of the 2 towers. At the deck
center, the antenna was installed on a 4-m-high beam to avoid multipath (and
obstructions) from the vehicles. However, the towers and suspension cables could be a
source of multipath. A method to filter (short period) multi path effect is the use of the
moving average of instantaneous coordinate results. The moving average window must
be carefully selected, i.e., wide enough to reduce multipath effect but not too wide to
avoid the loss of information about real bridge movement. Unfortunately, no chokering
antennas were available to us during the observation sessions.
Antenna phase center variation: Three types of antennas have been used, namely
Ashtech Z-XII and LD-XII and NovAtel 501. The relative phase centers of the antennas
have been calibrated using a I-m calibration beam. This aluminum beam, with precisely
known length, was setup and leveled on a geodetic point and oriented (with an optical
device) towards the direction of another geodetic point [Bourassa, 1994]. The comparison
of the known baseline components of the calibration beam with the baseline components
obtained from GPS observation sessions allows to determine relative phase center
variation between pairs of GPS antennas. Four 24-hour observation sessions have been
conducted for the relative phase center calibration of the 5 GPS antennas. The relative
phase center offset never exceeded 2 mm in the 3 baseline components. These measured
offsets are within the manufacturers technical specifications and are not significant,
considering the precision associated to the calibration process. Accordingly, no further
correction has been applied for the phase center variation.
Before surveying the Pierre-Laporte bridge, the methodology has been tested on known
baselines with a device which mimics realistic bridge displacements. The length of the
baselines and the height differences were typical of the bridge monitoring network. For
independent and precise comparisons, the baseline height differences have been
determined by geometrical leveling (relative geoid undulations were also taken into
account). This analysis showed that instantaneous GPS relative positioning has typically a
precision of about ±5 mm, horizontally and about ±1 cm, vertically, when PDOP factor
does not exceed a value of 6.

Deformation monitoring network

Figure 2 shows the stations of the deformation monitoring network. The baseline length
(D) and the height difference (i1h) between the stations are given in Table 1. Two
reference stations (RINI and RIN2) have been setup in bedrock, on the north river bank
close to the bridge. Stations TON and TOS are located on the top of the North and South
towers, respectively. Station TACE is located on the deck center of the bridge.

383
 To verify the stability of the 2
reference stations, they have been
connected, by GPS observations and
precise geometric leveling, to station
UL2005 UL2005 located on the Laval
University campus. This concrete
pillar, located 4 km away from the
bridge, is part of the GPS calibration
network in Quebec City area,
established by the Geodetic Surveys
of Canada. Between the 3 sessions,
the coordinates of stations RIN 1 and
RIN2, relati ve to UL2005, did not
change significantly. So, the same set
of coordinates for the reference
stations was used for all 3 sessions.
1---1 The Pierre-Laporte bridge is a 6-
lane, 1040-m-Iong suspension bridge
100 m which crosses the St. Lawrence river
in Quebec City. The center span is
670-m-Iong supported by 2 110-m-
high towers. The bridge structure, of
Figure 2: Deformation monitoring network of the a total weight of 18,000 tons, is
Pierre-Laporte bridge. suspended at 2 main cables of 62 cm
in diameter. Such an engineering
T abIe 1: Baserme en~t h an d helgl
. ht don
1 erence" structure is affected by strong
Baseline a b c d e f constraints caused by winds, traffic
D(km) 0.3 0.7 1.0 0.1 3.5 3.5 loading and temperature variations.
M1 (m) 59 -4 59 -2 17 19

Analysis of the movement of station T ACE (deck center)

Figure 3 illustrates the vertical displacement of the deck center (station TACE) and the
temperature as a function of time for the July 1996 session. The epochs where the
temperature is the highest roughly correspond to the lowest vertical positions of the deck
center. A variation of temperature of 8°C has caused a vertical displacement of about 12
cm. Thermal dilatation of the suspension cables explains this movement of the deck
center. The vibrations caused by vehicle traffic on the bridge are more apparent in Figure
4 which is an enlargement of Figure 3, between 13:00 and 14:30 (EDT). Moreover, the
influence of vehicle loading is clearly detected in Figure 4. In fact, one can see the impact
of a car accident leading to the entire blockage of 3 lanes of the bridge, for about 30
minutes. The center deck of the bridge, half-full with stopped vehicles, has been bent by
about 40 em during this time period.
Figure 5 shows the transverse displacement of station T ACE caused by the transverse
wind speed. The wind speed was recorded on the bridge deck with an anemometer. The
high correlation between these 2 quantities are clearly visible. For example, a transverse
wind speed of 20 kmlh (from the East direction) has caused a transverse displacement of
the deck center of about 5 em, in the West direction.
In Figures 3 to 5, the gray line, superimposed on the raw displacement results, is the
moving average applied onto the coordinates to reduce the impact of observation noise
and multi path. The length of the moving average window is 60 seconds, in Figures 3 and
4, and 120 seconds in Figure 5.

384
 0.5rrOT~~~,,~~,,~~rr~~rrOT~~~OT~rrrT,,~rr"~30 0
~
25 ~
E ~
~0.25~~--------~~~--------~~----------------~~~----~~20 ~
m ~
E E
~ 15 ~
~ 0
en
'5
n;
u
.~ -0.251----W----+"..u..------------------------'----=-+-+--'+-t---------------=----
>

Figure 3: Vertical displacement of station TACE as a function of temperature (July 1996).

Eastern Daylight Time

13:15 13:30 13:45 14:00 14:15
0.5

I
C 0.25
m
E
m
u 30 min.
co 0
Ci
en
'5
n;
.-eum -0.25
>

-O· ~:OO 3:15 3:30 3:45 4:00 4:15 4:30

Elapsed Time (hours)
Figure 4: Vertical displacement of station TACE due to traffic loading during a car accident.

2"
"""-
0.2 50 E
:::..

-
I 't "0
m
East m
c: ~
en
m 0.1 0
E "0
c:
m
u .~
co
a. m
~
(J)
'5
0 Q)

Q)
>
en
~ c
co
Q)
> t=
c -0.1
Cf)

co
t= West
+
-0 .2 0 48

Figure 5: Transverse displacement of station TACE as a function of wind speed (October 1996).

385
Analysis of the seasonal displacements of the bridge (towers and deck center)

Table 2: Seasonal variations of the average

n Ige statIOns.
coord'mates 0 f the b'd . Table 2 presents the seasonal variations
variations between (between July 1996 and February 1997) of the
July and February average coordinates for each station. The
analysis of these results shows that the 2 towers
sessions
moved away from each other of 8 cm along the
L: 4.0cm longitudinal axis of the bridge. The vertical
station TON T: -1.0 cm variation of -6 cm at stations TON and TOS is a
V: -5.5 cm consequence of the thermal contraction of the
L: -3.8 cm 1l0-m-high towers. In fact, between the 2
sessions the average temperatures dropped by
station TOS T: -0.6cm 25°C. The deck center (station T ACE) rose of
V: -6.1 cm 46 cm between the 2 seasons. The thermal
L: 0.2cm contraction of the suspension cables explains
station T ACE T: -4.5 cm this movement. Finally, the average transverse
V: 45.5 cm displacement of station TACE (5 cm towards
the West direction) is caused by the variation of
temperature -25 D C the average transverse wind speed of 21 kmlh
between the 2 sessions.
transverse 21 kmlh
wind speed
L: longItuomal, 1: transverse, V: vertIcal

Conclusions and further research

A methodology, algorithms and software have been developed for bridge deformation
monitoring in post-processing. The precision of instantaneous relative positioning is
about ±5 mm horizontally and ±1O mm vertically, when PDOP values are smaller than 6.
The analysis of the displacements shows: i) a transverse movement of the deck center due
to a transverse wind speed of about: -2 mm / 1 kmlh; ii) a vertical displacement of the
deck center with temperature variation of about: -2 cm / 1DC; iii) a vertical variation of
the deck center correlated with traffic flow; iv) an elongation of the tower with respect to
temperature variation of about: 2 mm / 1DC; and v) a displacement of the towers towards
the river banks of about: -1.5 mm / 1DC.
Other GPS sessions (with a higher GPS data rate) for studying the high frequency
movements of the bridge are planned. Provisions are made to add more receivers with
chokering antennas. The methodology will also be implemented for real-time deformation
monitoring and applied to other engineering structures such as towers, skyscrapers and
dams.
Acknowledgments: We would like to thank the Quebec Transportation Department for their
helps during the GPS campaigns, and the Natural Sciences and Engineering Research Council of
Canada and the Laval University Faculty of Forestry and Geomatics for their financial supports.
References
Bourassa, M. (1994). "Etude des effets de la variation des centres de phase des antennes GPS."
Memoire de maitrise, Departement des sciences geomatiques, Universite Laval, Quebec, 109 p.
Hatch, R. (1990). "Instantaneous Ambiguity Resolution." lAG Symposium No. 107, Kinematic
Systems in Geodesy, Surveying, and Remote Sensing, Banff, Alberta, Canada, pp. 299-308.
Rothacher, M. et al. (1986). "The Swiss 1985 GPS campaign." Proceedings of the Fourth
International Geodetic Symposium on Satellite Positioning, Austin, Texas, pp. 979-991.
Santerre, R. (1991). "Impact of GPS satellite sky distribution." Manuscripta Geodaetica, 16, pp.
28-53.

386
 THE GPS COMPONENT OF THE PROJECT FOR DIGITAL
MAPPING OF THE KARST AQUIFER SYSTEM
NEAR CURITIBA, BRAZIL t

Marcelo C. Santos, Antonio J. B. Vieira & Henrique Firkowski

Universidade Federal do Parana

Curso de Pos-Gradua~io em Ciencias Geodesicas
Caixa Postal 19011
81531-990 Curitiba PR Brazil
[email protected], [email protected], [email protected]

Abstract

The correct management of water supply, both surface and ground water, has
tremendous environmental implications, directly affecting the quality of life of the
population in general, with economical and political consequences. Very recently, part of
the Karst aquifer system, which passes under part of the Brazilian Southern Region, has
become under scrutiny for solving part of the problem of water supply for Curitiba and
surroundings. The project for its digital mapping aims to correctly understand the processes
governing this aquifer allowing for its use without depleting it.
This paper describes the GPS component of this project, applied to a test area of
around 40 km2 . This component is basically composed of the surveying of the artesian wells
and the definition of a local geoid. For the latter, the strategy adopted was to reposition
two diferent geoidal maps, one given by the program MAPGEO from ffiGE, and the other
presented by Sa & Molina with a far greater resolution. Bench marks of the fundamental
vertical network were used as constraints in the adjustments carried out for both geoidal
maps and as control points ("ground truth"). Results indicate that a better local geoid was
obtained based on the second geoidal map.
This project presents an application of Geodesy to sustainable development.

t A more detailed description of the work summarized here will be published in GPS World magazine.
 KINEMATIC GPS POSITIONING WITH ADAPTIVE KALMAN
FILTERING TECHNIQUES

Jinling Wang, Mike Stewart and Maria Tsakiri

School of Surveying and Land Information
Curtin University of Technology
GPO Box, U 1987, WA 6845
Perth, Australia

Abstract

In a kinematic GPS positioning system, state parameters are usually estimated using the
Kalman filtering method. It has been noted that reliable ambiguity resolution, and
estimation of other state parameters are highly dependent on the correct stochastic models
for the GPS double difference measurements. In this paper, a real-time statistical
procedure to estimate the covariance matrix of GPS double differenced measurements is
presented. Test results indicate that with the proposed procedure, the reliability of
ambiguity resolution and kinematic positioning results can be improved.

Introduction

In a kinematic GPS positioning system, the determination of the state parameters, which
may include the position, velocity and acceleration of a moving platform and other
parameters of interest, such as unknown carrier phase ambiguities, is achieved by
continuously tracking GPS satellites in view and taking a series of measurements (code
pseudo-ranges and carrier phases) to the satellites at constant intervals or epochs. The
mathematical models involve the dynamic model, measurement model and stochastic
models. The real-time estimates of the state parameters can be performed using the
Kalmanfiltering technique (eg. Gao et aI, 1996; Landau and Euler, 1992; Qin et al; 1992).
With the Kalman filtering technique, the state parameters can be not only estimated
recursively in real-time mode, but also have the statistically defined optimal properties. It
is well known, however, that the optimum filtering results are highly dependent on the
correctness of the adopted stochastic models.
 In GPS kinematic positioning, the precise carrier phase measurements with code pseudo-
ranges can generate accurate positioning results. In order to avoid the effect of the errors in
process noise covariance matrix on the state estimation, the filter can be set up to operate
only on the measurement noise. However, stochastic modelling for GPS double
differenced measurements still is a difficult task in real-time data positioning.
The widely used approach to construct the covariance matrices of GPS double
differenced measurements is based on the variance-covariance propagation law, and it is
always assumed that all the one way code or carrier phase measurements are independent
and they have the same accuracy. Unfortunately, these assumptions are not realistic. The
accuracy of the GPS code and carrier phase measurements may change with different
measuring conditions. Adopting incorrect stochastic models for GPS measurements in data
processing will inevitably result in unreliable statistics for ambiguity resolutions and
biased positioning results (eg. Hatch and Euler, 1994).
In the literature, the procedures for the on-line estimation of the process and
measurement noise matrices are termed Adaptive filtering techniques (e.g. Mehra, 1972,
Chin, 1979). In this paper, one of the commonly used adaptive kalman filtering techniques
will be reviewed. Based on the filtering residuals, a new adaptive filtering procedure will
be presented. The performance of the new procedure will be tested with a GPS data set.

Kalman Filtering as Least Squares

The functional models used in GPS kinematic positioning can be expressed as

the dynamic model: Xk =<l>k,k-Ixk-l +'tk , (1)

and the measurement model: Zk =Hkxk +Ek , (2)

The corresponding stochastic models are further assumed as:

T)
E('tk't l' = {Qk i =k , T
E(EkEI' ) ={Rk i =k
, and (3)
o i'# k o i '# k

where Xk is the state parameter vector which may contain the time-dependent position and
velocity parameters and time-independent carrier phase ambiguity parameters, if they have
not been resolved; <I> k,k-l is the state transition; 't k is the random error vector; Zk is the
measurement vector; Hk is the corresponding design matrix; and Ek is the measurement
noises; Qk is the so-called process noise covariance matrix; and Rk is the measurement
noise matrix; k and i are the time indices.
For convenience in the following discussion, the Kalman filtering estimation formulae
are derived using least squares method. The predicted state values xk can be obtained by:

(4)

390
where Xk-l is the optimal estimator of the state parameters at the previous epoch (k - 1 ).
The covariance matrix of xk is expressed as:

(5)

By integrating the measurements Yk and the predicted values of state parameters xk' the
optimal estimators of the state parameters Xk can be obtained using the least squares
technique (e.g. Pelzer, 1985; Cross 1983). The Gauss-Markov models are:

(6)

(7)

where lk =[~:J. Vk =[:;:lAk =[~kl

In equation (6), V Zk is the residual vector of the measurements Zk; V Xk is the residual
vector of the pseudo-measurements xk; E is the identity matrix; elk is the covariance
matrix of the measurement vector lk; P is the weight matrix. With the least squares
principle, the optimal estimators of the state parameters and their covariance matrix are
formulated as (e.g. Pelzer, 1985):

Xk =(AI PkAk )-1 Al Pkl k =Xk + Gkd k , (8)

Q Xk =(Ai PkAk )-1 = Q Xk -GkQdk Gi ' (9)

where Gk is the gain matrix; dk is the innovation vector and Qdk is its covariance matrix,
which are described as:

(10)

dk=Zk-HkXk' (11)
T
Qd k = Rk + HkQxk Hk , (12)

respectively. The innovation vector and its covariance matrix are important statistics used
in the existing adaptive filtering techniques.

Adaptive Kalman Filtering Techniques

Various adaptive filtering techniques are divided into three categories: Bayesian,
maximum likelihood (ML) and innovation approaches (Mehra, 1972). The Bayesian and

391
ML methods are computationally intensive and cannot be realistically used for real-time
data processing (e.g. Chin, 1979). One of the well-known innovation approaches is the so-
called covariance-matching. For simplicity, the process noise matrix is assumed known,
and the identification of the measurement noise covariance matrix Rk only, is discussed.

Covariance Matching Procedure Based on Innovations

The basic concept of this procedure is to make the actual innovation covariance matrix
consistent with their theoretical values (Mehra, 1972). The actual covariance matrix of the
innovations sequence d k is calculated approximately by its sample covariance:

(13)

where m is chosen empirically to give some statistical smoothing. By matching the above
estimated covariance matrix with its theoretical form presented by (12), the measurement
noise covariance matrix is estimated as:

(14)

Many numerical tests indicate, however, that the innovations sequences are very sensitive
to the approximate values used in the linearization of the GPS measurements equations.
More importantly, with this procedure, it cannot be guaranteed that the resulting matrix Rk
is positive definite.

A New Procedure Based on Filtering Residuals

In order to obtain a realistic estimator of the measurement noise covariance matrix, instead
of the innovations sequence, the more precise estimator of the measurement noise level
should be used. From equation (6), we obtain the measurement filtering residuals:

(15)

which, obviously, is the best estimator of the measurement noise level because the
estimated values xk (not the predicted values Xk) of the state parameters are used in their
computations.
Similarly to equation (13), an estimator of the measurement noise covariance matrix is
derived from the filtering residuals as:

(16)

After epoch m, Rk is updated with the incoming filtering residuals. m is called the width
of the moving window. It is noted that, compared with the computation of the innovation

392
vector, which is generated by the standard Kalman filtering process, the computation of
the filtering residuals will involve some extra computations. Fortunately, the number of
added calculations is very small.

Test Results and Analysis

A GPS kinematic data set was used to test the performance of the adaptive filtering
technique. The data were collected on July 18, 1997, at the Fremantle port in Perth,
Australia, using two Trimble 4000SSE dual frequency receivers. The rover receiver
antenna was mounted on a boat, and was moving around the offshore test area (about
5.5krn away from the reference station). During the 10 minutes of the experiment, 7
satellites were tracked. The data collection rate was 1Hz.
In the data processing, carrier phase and pseudo-range measurements from both L1 and
L2 frequencies were used and double differenced measurements were formed. The true
ambiguity values were recovered with the whole data set. The first epoch solution with
fixed ambiguities was used as the initial values for the Kalman filtering. The initial
standard deviations for L1 and L2 carrier phase and pseudo-range data are 1.0m, 1.0m,
O.OlOm, 0,012m, respectively. The width of the moving window in the adaptive filtering
procedure was set to 200 epochs.
After 200 epochs, ambiguity resolutions were performed on-the-fly for each period of 10
epochs. A total of 40 batches were formed. In each batch, the solutions with and without
using the new adaptive filtering (AF) procedure expressed by equation (16) have been
obtained (in this test, the Covariance Matching Procedure did not generate a positive
definite matrix Rk)' Some of the results are shown in Fig. 1 and Fig. 2.

I -- -- --No-AF I
......... .
- - - - - - No-AF AF AF
60~================~

40
<J)
g:~
co :- ...... , # _ _ "'~. ..

> 20
~:~
o 5 10 15 20 25 30 35 40 o 5 10 15 20 25 30 35 40
Batch Number Batch Number

Fig. 1 : Ambiguity search space volume Fig. 2 : F-ratio values

Fig.1 shows the changes of the ambiguity search space volume Vs, which is defined by
the determinant (times 10 8 ) of the ambiguity states covariance matrix. In all the cases, the
ambiguity search spaces with the adaptive filtering procedure are smaller than those
without using it.
Fig. 2 compares the F-ratio values used in the ambiguity resolutions (eg. Landau and
Euler, 1992). In almost all the cases, the F-ratio values in the solutions with the adaptive
filtering procedure are bigger than those without using it. More importantly, without using
the adaptive filtering procedure, in up to 10% cases, the wrong ambiguities have been

393
identified as the best ambiguity combinations. Therefore, the reliability of the ambiguity
resolution is improved by using the adaptive filtering procedure.
Test results have also demonstrated the accuracy of the kinematic positioning results can
be improved using the adaptive filtering procedure. For example, at epoch 200 when the
filtering process was using the preset standard deviations for measurements, the estimated
standard deviations of (x, y, z) coordinates were 0.009m, 0.021m and 0.0 10m,
respectively. At epoch 201 when the filtering process began to use the estimated
measurement noise covariance matrix with the adaptive procedure, the standard deviations
of (x, y, z) coordinates are down to 0.002m, 0.OO6m and 0.002m, respectively.

Concluding Remarks

In GPS kinematic positioning, an incorrect measurement noise covariance matrix can

result in the wrong ambiguity resolution and thus biased positioning results. The currently
used method to construct such a covariance matrix is not reliable in some practical
applications.
Based on the filtering residuals, a new adaptive filtering procedure has been proposed.
With such an adaptive procedure, the measurement noise covariance matrix can be
estimated in real time data processing. Test results have shown that with the new adaptive
procedure, the reliability of the ambiguity resolutions and the accuracy of kinematic
positioning can be significantly improved.

References

Chin L. (1979) Advances in Adaptive Filtering, In Leondes C.T. (ed.): Advances in

Control Systems Theory and Applications, Academic Press, No. 15, pp.277-356.
Cross P. A. (1983) Advanced Least Squares Applied to Positioning Fixing, working paper
No.6, Department of Land Surveying, Polytechnic of East London.
Gao Y., J. F. McLellan and J.B. Schleppe (1996) An Optimized Fast Ambiguity Search
Method for Ambiguity Resolution on the fly, Proceedings of IEEE PLANS, Atlanta,
Georgeia, pp.246-253.
Hatch R. and H.-J. Euler (1994) Comparison of several AROF kinematic techniques.
Proceedings of ION-GPS-94, Salt Lake City, Utah, pp.363-370.
Landau H. and H-J Euler (1992) On-the-Fly Ambiguity Resolution for Precise Differential
Positioning, Proceedings of ION GPS-92, Albuquerque, New Mexico, pp. 607-613.
Mehra R. K. (1972) Approaches to Adaptive Filtering, IEEE Transactions on Automatic
Control, AC-17, Oct., pp. 693-698.
Pelzer, H (1985) Statische, Kinematische und Dynamische Punktfelder, In Pelzer (Ed.):
Geodiitische Netze in Landes- und Ingenieurvermessung II, Stuttgart, Wittwer.
Qin X., Gourevitch S., and M. Kuhl (1992) Very Precise Differential GPS: Developemnt
status and results, Proceedings of ION GPS-92, Albuquerque, New Mexico, pp. 615-
624.

394
 AUTHOR INDEX

Abu, S. 49 Dousa, J. 18
Adam, J. 27 Drewes,H. 168,174,180,193,199,211
Aebersold, R. 351
Altamini, Z. 57 Edwards, K. 125
Aquino, M.H.O. 43, 101 Ehrnsperger, W. 18
Arciniegas, S. 168
Ashkenazi, V. 43, 101 Fachbach, N. 18
Augath, W. 35 Fagard, H. 168
Ferraro, L. 18
Baldi, P. 343 Figurski, M. 18
Baran, L.W. 252 Firkowski, H. 387
Barbaste, J. 125 Forsberg, R. 363
Barbato, F. 217 Fortes, L.P.S. 73, 167
Beattie, D. 187 Franke, P. 66
Becker, M. 49, 66
Beutler, G. 93 Gaudet, RJ. 59
Blewitt, G. 8 Geiger, A. 351
Blick, G. 107 Gelo, S. 161
Blitzkow, D. 73, 205 Gende, M. 277
Boonphakdee,C. 49 Grant, D. 107
Boucher, C. 8, 57 Grejner-Brzezinska, D.A. 337
Brouwer, F. 35 Gurtner, W. 27
Brunini, C. 199,277
Brunner, F.K. 349 Han,S. 283,307,319
Bruton, A.M. 325 Harsson, B.G. 27
Bruyninx, C. 18 Hartinger, H. 349
Heflin, M.B. 8
Chang, C.C. 114 Hernandez, J.N. 193
Chang, R.G. 114 Herring, T.A. 8
Chao, J. 265 Houghton, H. 155
Chen, R. 259 Hoyer, M. 168
Chen, W. 43
Cocard, M. 351 Ihde, J. 27
Collins, J.P. 331
Costa, S.M.A. 73,187,217 Jaworski, L. 161
Jia, M. 369
Da Fonseca Junior, E.S. 149 Johansson, J.M. 18,229
Dare, P. 120 Jokela, J. 131
Davies, P.B.H. 8
De Jonge, P.J. 223,313 Kahle, H.-G. 351
Ding, X. 155,265 Kahmen, H. 357
Dominici, D. 271 Kakkuri, J. 259
Dose, A. 351 Kaniuth, K. 180,199,289
Keller, K. 363 Santerre, R. 381
Kleusberg, A. 277 Santos, M.C. 205,387
Koivula, H. 13 7 Scaramuzza, M. 351
Kouba, J. 8, 245 Schaer, S. 93
Kumar, M. 168 Schliiter, W. 27,66
Schmitz, M. 237
Lamoureux, L. 381 Schneider, D. 143
Lang, H. 35, 351 Seeber, G. 168,237
Langley, R.B. 331 Seeger, H. 49, 66
Lin, L.-S. 295, 307 Seemiiller, W. 211
Luz, RT. 73 Selli, S. 271
Lyszkowicz, A. 161 Shagimuratov, 1.1. 252
Sillard, P. 57
Malys, S. 1 Slater, J.A. 1
Marsella, M. 343 Springer, T.A. 18,93
Marti, U. 143 Stewart, M.P. 155,265,369,389
Maturana, R 168 Stoppini, A. 271
Meier, B. 351 Stuber, K. 180
Menge, F. 237 Subiza, W.H. 217
Mertikas, S. 295, 307 Swiatek, A. 161
Mingsamon, S. 49 Symons, L.J. 43
Mireault, Y. 245
Moirano, J. 180, 199 Teunissen, PJ.G. 313
Monico, J.F.G. 149,205 Tiberius, C.CJ.M. 313
Moore, T. 43, 101 Torchetti, R. 168
Tremel, H. 180, 193
Nardi, A. 18 Troyer, L. 307
Nielsen, C.S. 363 Tsakiri, M. 369, 389

Odijk, D. 313 Van Mierlo, J. 35

Oliveira, L.C. 205 Vespe, F. 18
o llikainen, M. 13 7, 301 Vieira, A.J.B. 387
Vittuari, L. 343
Pereira, K.D. 73, 168, 187 V6lksen, C. 237
Perez, R.R 217
Poutanen, M. 13 7 Wang, J. 389
Wanninger, L. 86
Radicioni, F. 271 Weber, G. 18,66
Reinhart, E. 49 Weber, R 79
Rizos, C. 283,295,307,319,375 Wiget, A. 18
Roberts, C. 375 Woeppelmann, G. 27
Rothacher, M. 93 Wiibbena, G. 237

Sacher, M. 35 Zdunek, R 161

Saleh, H. 120 Zielinski, J.B.
Saleh, J. 125

396
 Springer
and the
environment
At Springer we firmly believe that an
international science publisher has a
special obligation to the environment,
and our corporate policies consistently
reflect this conviction.
We also expect our business partners -
paper mills, printers, packaging
manufacturers, etc. - to commit
themselves to using materials and
production processes that do not harm
the environment. The paper in this
book is made from low- or no-chlorine
pulp and is acid free, in conformance
with international standards for paper
permanency.

Springer