Flood Estimation
Flood Estimation
Flood Estimation
Preliminary determination of
design flood
seeking protection from; in fact, fairly short recurrence intervals may be accepted
for very small dams where dam failure would have practically unnoticeable
consequences downstream; conversely for dams representing a risk for popula-
tions downstream, long recurrence intervals must be considered;
! the second difference stems from the fact that small and medium sized struc-
tures are generally located in small catchment areas which rarely have gauging
stations. In this case the quality of the available hydroclimatological information
is less good. Greater imprecision in the results of the hydrological study must be
taken into account when the type of spillway and its dimensions are chosen.
P reliminary determination of design flood
PRELIMINARIES
Today it is clearly accepted that the flood study, which serves in dimensioning spillway
and discharge structures, cannot rely simply on flow observations. The methods used
are of the hydro-meteorological type and combine rainfall data with flow data. Those
methods may rely simply on statistical concepts (GRADEX method or the AGREGEE1
model) or propose a deterministic approach to the transformation of rainfall into
flows.
In many countries (in particular in Anglo-Saxon countries), the Probable Maximum
Precipitation - Probable Maximum Flood (PMP - PMF2) method, developed by North-
American engineering, is commonly used. This method defines a probable maximum
flow for the catchment area studied which is the highest flood that can reasonably be
imagined. The risk of such flow occurring is in principle infinitely low, and in any case
cannot be quantified.
In other countries, like France, a design flood is computed along with a corresponding
risk of occurrence. Depending on the hazard this represents for the downstream area
- human fatalities, economic considerations, whether or not the dam is overtoppable,
etc. - the selected recurrence interval may be of the order of 10-2 to 10-4. In terms of
frequency, this would mean for example that a dam designed to handle a flood with
24 a frequency of occurrence of 10-3 has a risk of 1 - (0.999)100 = 9.5% of suffering the
design flood during 100 years in operation. The risk is of the order of 1% if the dam
is likely to have to discharge the 1000-year flood (in 100 years operation). The desi-
gner must therefore be aware that the dam runs a non negligible risk of being
confronted with a design flood determined this way, while bearing in mind that it
could well withstand a higher flood thanks to freeboard.
Although at first consideration all of these methods are only applicable for sites where
gauging records are available, practical considerations sometimes result in applying
them in downgraded mode. This is often the case for small catchment areas where
flow records are rarely available. The reliability of the hydrological study is however
always highly dependent on the quality of the available hydroclimatological infor-
mation.
The design flood is the flood with the longest recurrence interval considered in the
reservoir. It is taken into account to determine the Maximum Water Level (MWL) and
dimension the spillway, incorporating possibilities of flood routing. Often, the considered
design flood is the flood with the highest peak flow. It is not always certain that this
flood is the worst case in calculation of the spillway. A flood with a lower peak flow
but lasting a longer period could have worse consequences. The minimum recurrence
interval recommended for such a flood is between 100 and 10 000 years (10-2 to
10-4). The choice of a recurrence interval depends on the risk involved in dam failure.
The dam's intrinsic risk can be quantified by means of the parameter H2 V . Table 1
sets out recommendations for the choice of a design flood relative to this criterion.
However, global risk is also related to the vulnerability of the downstream area (po-
pulation density in the zone likely to be flooded in the event of a failure). The
recommendations in table 1 must be beefed up in case of serious vulnerability (for
example by going from a 500-year flood to a 1000-year flood). When the dam is of
public safety interest, the recurrence interval should never be less than 1000 years,
whatever the value of H2 V .
Once MWL has been calculated, the dam crest is set at a higher level. The difference
between these two levels is called freeboard. Freeboard is essentially intended to
avoid overtopping due to wave action but also plays an essential role in safety from 25
flooding. A method for calculating it is given in Chapter IV, p. 73.
Thanks to freeboard, a dam should be able to withstand a flood (which is known as
safety flood) higher than the design flood. This is by definition the worst case flood
that could occur at the dam without any risk of failure. In the case of an ungated
spillway at an embankment dam, the safety flood would be any flood that would
cause overspillage, provided that it did not cause overtopping at any point in the
chute that would jeopardise the fill itself. For a gravity dam, the safety flood would
also correspond to the crest of the non-overflow part. For a dam with an impermeable
core, the safety flood would be reached when the reservoir water level reaches not
the dam crest but the crest of the core.
This statistical method, developed by Electricité de France (EDF), is the standard used
in France. Its success is in particular due to its (relative) simplicity of use, which results
from an extreme simplification of the process of transforming rainfall into flow.
Rainfall is considered globally over a certain period of time, equal to the average
duration of the hydrographs.
The probability of precipitation events lasting various durations is a simple exponential
decay function. The main parameter is proportional to the standard deviation of
maximum precipitation values. It is called the exponential gradient, GRADEX.
GUMBEL's law is often applied. Its distribution function is as follows:
F(P) = EXP (-EXP (-(P-Po)/a)
The GRADEX (a) may be obtained using the event method. In this case, it is equal to 0.78
times the standard deviation. a is of course a function of the duration of the precipitation
considered.
Remarks :
26 ! When P ➔ ∞, F (P) ➔ 1 - EXP (-(P-Po)/a) and the Napierian logarithm at recurrence
interval T = (1/(1 - F (P)) is equal to (P - Po) /a. The rainfall depth varies linearly with the
logarithm of the recurrence interval, the slope (a) of this straight line being equal to the
GRADEX..
! If P and P100 indicate respectively the rainfall depth at recurrence intervals of 1 000
1 000
and 100 years, then:
P1 000 - P100 = a (ln1 000 - ln 100) = 2,3 a
(ln designating the Napierian (or natural) logarithm).
When the catchment area reaches a certain saturation level, any increase in rainfall
generates an equal volume of runoff for the same lapse of time. As a first approxima-
tion, this state is reached for recurrence intervals of ten years (impermeable catchments,
with low retention), to 50 years (permeable catchments, with high retention).
The runoff law is obtained quite simply by translating the rainfall depth law to the
point of the 10 or 50 year recurrence interval.
A physical interpretation of this process can result from observation of the graph of
runoff variations relative to the amount of rainfall (see fig. 1, p. 35). The retention
capacity of the catchment area is schematically represented by the difference between
the bisector (rainfall = runoff) and its parallel, plotted in the middle of a cloud of
points.
Figure 2 (p. 35) illustrates this principle: the adjustment of the rainfall values has the
GRADEX as slope. In this application, the recurrence interval retained for the hypothesis
concerning saturation of the catchment area is 20 years (this corresponds to a reduced
GUMBEL variable equal to 2.97). For volumes exceeding the runoff corresponding to
this recurrence interval, adjustment is made by plotting a straight line with a slope
equal to the GRADEX.
In this example the catchment area is instrumented and it is therefore possible to
make a statistical adjustment of runoff, up to the 20 year recurrence interval.
In the case of small catchment areas without flow records, this is not possible. A
regional approach based on nearby, if possible similar, catchment areas is necessary.
It is however possible to consult national analyses such as SOCOSE or CRUPEDIX1.
These methods essentially require rainfall data and give an order of magnitude for
peak flow at recurrence intervals of 10 years (10 and 20 years for SOCOSE). It
appears that even a sizeable error on a 10-year (or 20-year) flood has a relatively
weak influence on the 1000-year flood or the 10 000-year flood calculated with the
GRADEX method.
It is noteworthy that for small catchment areas with no gauging stations, the evaluation
of rare flood flows is almost exclusively based on rainfall information. Luckily this
information is generally available for most of France.
A simple ratio of affinity is used to go from runoff in the considered time-frame to peak
flow. This ratio is estimated from hydrographs; its average value is used (laws governing 27
the probabilities of ratios and average runoff may also be combined, resulting in
ratios that increase with the recurrence interval). For catchment areas with no water
level gauging stations, we may use a ratio determined for similar catchment areas.
An example of this application is given farther along in this chapter (p. 31).
1. See Bibliography, reference 1, p. 36. The SOCOSE method is derived from the work of the
American Soil Conservation Service.
P reliminary determination of design flood
precise data, the formulation of the characteristic time of the catchment area, developed
in the SOCOSE1 method, may be used. In this method, the characteristic time is
defined as the period during which the runoff is more than half of peak runoff. If no
data is available on runoff on the site, the following regionalised formula may be
used:
Log D = - 0,69 + 0,32 Log S + 2,2 (Pa/Pta)0,5
D: characteristic time (hours)
S: surface area of the catchment area (sq. km)
Pa : average annual rainfall (mm)
P: daily rainfall with a 10-year recurrence interval (mm)
ta : average yearly temperature (°C)
Note:
This method is frequently used at a daily time-step when the catchment area is of a certain
size, by virtue of a greater availability of daily information on rainfall and runoff.
The sudden rupture that affects the runoff equation at the pivot point (start of the
rainfall equation) leads to an over-estimation of flow with intermediate recurrence
intervals (50 to 500 years).
The affinity ratio to obtain the peak runoff is extremely variable. The method
recommends keeping its average value. If we have properly chosen the duration in
which the increase in runoff is equal to the increase in rainfall, it should be of the
28 order of 1.5 to 2.0.
This method does not give a design hydrograph in a form suitable for simulation of
flood attenuation. A bi-triangular shape which respects the duration, the peak runoff,
and the volume of runoff can be used. In general, these design hydrographs result in
over-estimations of attenuation capacity, as they represent only a part of the flood.
One must often take into account the base flow in the river before the flood, when it
represents a non-negligible proportion of the flood flow.
! taking into account the statistical distribution of the affinity ratio (passage from
This model makes no hypothesis on the equation of rainfall probability. The simple
exponential decay function for rainfall relative to the recurrence interval is not imposed.
Thanks to the progressive passage from the flow equation to the rainfall equation, the
model avoids the over-estimation of the discharges of intermediate recurrence intervals
(50 to 500 years). Although not very realistic, the single-frequency hydrographs
obtained are easy to use in calculation of flood attenuation.
The AGREGEE model and the GRADEX method give similar results in estimation of
extreme floods (1000 years to 10 000 years).
This method is very rarely used in France. It is based on knowledge of the Probable
Maximum Precipitation (PMP) in the catchment area and a rainfall - runoff model to
calculate Probable Maximum Flood (PMF). It results in a design hydrograph.
The PMP is defined as the highest theoretical precipitation that is physically probable 29
for a specific geographic location, over a defined period of time. Its estimation is
based on observed rainfall data and on maximising the meteorological parameters
linked to precipitation: humidity, temperature, pressure of saturating vapour in the air,
wind speed, convection, etc. Such calculation requires the skills of a meteorologist. In
order to facilitate the calculation, some countries have published regional PMP estimates.
Cemagref3 has developed a model for simulating flood scenarios called the SHYPRE model.
This approach is based on gaining maximum value from temporal information about
rainfall episodes in order to generate flood hydrographs with realistic shapes. By
coupling a stochastic model for simulation of hourly rainfall and a simple model for
transforming rainfall into discharge, the method generates a collection of flood
hydrographs over a very long period.
In this way, with no prior assumptions concerning the law of probability, the frequency
distribution, from routine to exceptional, can be constructed empirically. This goes for
both peak flows and for average flows and threshold discharges of various durations.
Such floods are available to calculate hydraulic transients, including attenuation in a
reservoir and modelling a flood plain.
In this method, the stochastic model of rainfall not only complies with temporal infor-
mation, it also gives an original approach to infinite rainfall behaviour. Processing
some 50 rainfall gauging stations along the French Mediterranean coastline has
given an idea of regional trends simply through the characteristics of daily rainfall
and has opened up the possibility of building a spatial regional model. The method
has proven to be more stable than simple statistics, which depend heavily on sampling,
and the model helps to evaluate the impact of anthropic effects.
These methods of estimating flow are extremely succinct and under no circumstance
may be a substitute for a complete hydrological study.
From observations of maximum floods over the last 2 centuries in 1400 catchment areas
throughout the world with surface areas in the 10 - 2.106 sq.km range, FRANCOU and
RODIER established an envelope curve formulated as follows:
Q/Qo = (S/So)1- k/10
• Q is the peak discharge of the flood (m3/s) for an area of the catchment S (km2).
• Qo = 106 and So = 108.
• k is a regional parameter. It varies in France in the 5.5 bracket (Mediterranean zone)
to 3.5 (oceanic zone in the north of France). At a global scale, the highest k values (and
thus in proportion the highest flow) are close to 6 like in Texas, New-Mexico and in
some of the Pacific areas affected by typhoons (Korea, Japan, Philippines, …). On the
opposite, the large African tropical streams are quite well characterised by an exponent
k value close to 2 (it is the case for the Niger and Senegal rivers).
Concerning an envelope of observed maximum floods, these estimates of discharge are
not affected by the frequency of appearance. The authors, however, consider that a good
part of the floods correspond to a recurrence period of around 100 years.
The GRADEX method has been applied by EDF for numerous French catchments of
surface areas varying between a few square kilometres to a few thousand square
kilometres. The regression established on 170 catchment areas of the peak discharge
with a 1000-year recurrence interval relative to the surface area of the catchment is:
Q = .S0,72
S is the area of the catchment in square kilometres and a parameter given in the
table hereafter for the following three zones:
! zone I : the catchment areas of tributaries to the lower Loire river (Vienne, Creuse,
etc.) located in the north of Central France, those of the Saône and Moselle rivers,
and Brittany;
! zone II : the catchment areas of the eastern and central Pyrenees, of Aude and Ariège,
of the Dordogne and the Lot rivers, the catchment areas of the Durance, the Fier and the
Arve rivers, the Dranses, and the Isère rivers;
! zone III : the catchment areas of Haute Loire, Cévennes, the Tarn river, the right-bank
tributaries of the Rhône river downstream of Lyons (Eyrieux and Ardèche rivrers, etc.),
Alpes-Maritimes, and Corsica.
This formula does not apply to catchment areas smaller than a few square kilometres. It
only gives an order of magnitude which must always be rendered more precise with a
local study. This order of magnitude is only to give a first opinion on the spillway capacity.
It should never replace a more complete study.
It is very rare to find a small catchment where we have a good understanding of the
hydrology. The subject of this study is therefore not determination of a dam's design
flood. It presents data from an experimental catchment which support the recommended
methods and justify the conclusion that the peak discharge of the design flood depends
very little on how long the considered rainfall lasts.
Located in the Massif des Maures, a Mediterranean mountain zone, this catchment
has been monitored by Cemagref since 1967, in the frame of the Réal Collobrier
Research Project. The lands adjacent are mostly compact gneiss. There is little soil
cover. The vegetation consists of scrubby briar and evergreen arbutus. The catchment
area measures 1.47 square kilometres. Mean annual rainfall (1967-1990) is 1164
mm and the corresponding runoff is 626 mm.
The largest river floods are caused by intense storms occurring mostly in September
and October. The following two events, with contrasting time histories, are an excel-
lent illustration of the different types of response of the catchment (table 2).
CATCHMENT RETENTION
Figure 1 (p. 35) shows that the retention loss was between 70 and 100 mm in the
three heaviest rainfall events, and between 55 mm and 100 mm in the three largest
flood events. Statistical analysis as described later indicates that the "pivot point" for
the GRADEX method should be taken as the 0.95 frequency (20-year recurrence
interval, reduced GUMBEL variable, u = 2.97).
Table 3 shows retention versus duration of rainfall.
Duration 1 2 4 8 12 24
(hours)
Retention 30 43 57 76 86 98
(mm)
Table 3 - Retention for a 20-year recurrence interval for different durations of rainfall.
Chapter I I
Figure 2 (p. 35) shows rainfall and runoff on the vertical scale versus the standardised
GUMBEL u.
The concave shape of the two curves is explained by the fact that events of very short
return periods are plotted. The asymptotic rainfall distribution appears if we limit
ourselves to the 27 most severe events (one event per year). It is not unusual for the
most severe event (September 13th, 1968) to lie some distance off the curve.
Since there is considerable sampling uncertainty in a 27 value sample, the computations
were run on the 150 most severe rainfall events observed for each of the durations
considered in determining the asymptote slopes, proportional to the standard deviations.
For this example, the fitting method would appear to overestimate the frequency of
the highest values. This is attributable to the sampling alone, and application of this
fitting method to samples of other rainfall durations yields estimates that are either
entirely consistent, or underestimated (points concave downwards). Results are listed
in table 4.
Duration (h) 1 2 4 8 12 24
Runoff (L) a 2.24 4.43 7.72 11.9 14.9 22.3
(mm) b 3.90 7.60 13.5 22.4 29.3 44.2
Rainfall (P) a 6.82 10.9 15.9 22.5 26.8 34.3
(mm) b 20.1 31.5 46.5 66.2 79.7 106 33
Table 4 - GUMBEL parameters a and b for rainfall or runoff: P (or L) = au + b
Duration 1 2 4 8 12 24
(hours)
Ratios between peak flow and mean flow have been calculated for all the events
considered. Their distribution makes it possible to estimate the ratios to be used in
determining peak flood flows for very long recurrence intervals.
The ratios are very variable, especially for very small flood volumes. Obviously they
increase with the duration of the event: between 1 and 2 for 1 to 2 hour events, they
may be as high as 25 for some 24 hour events.
With the ratios recorded during the three largest floods for the 1, 2, and 4 hour
durations (i.e. 1.6 - 1.74 and 2.67), the estimated 1000-year flood peaks are very
similar, i.e. 24.3 - 22.5 and 26.9 m3/s.
CONCLUSION
This application highlights the insignificant impact of the choice of duration on the
GRADEX estimates of flood peaks for very rare events.
This is similar to what was found for much larger catchments where floods were
estimated by studying one and two-day rainfalls.
34
! Flood volume of the 1000-year 24-hour value (245 mm, runoff yielding 0.36 hm ).
3
qp : peak flow
q(t) : flow at time t
D : characteristic catchment time, as defined above page 28.
A value of 2.7 meets these two requirements and gives the design hydrograph
shown in figure 3.
Chapter I I
35
BIBLIOGRAPHY
1 - Cemagref, 1980 - Synthèse nationale sur les crues des petits bassins versants ;
méthodes SOCOSE et CRUPEDIX, 3 fascicules, 120 p.
2 - Cernesson (F.), 1993 - Modèle simple de prédétermination des crues de fré-
quence courante à rare sur des petits bassins versants méditerranéens, thèse de
doctorat, Université des Sciences et Technique du Languedoc, Montpellier, 240 p. +
annexes.
3 - Cernesson (F.), Lavabre (J.), Masson (J.-M.), 1992 - Flood simulation from rainfall/
runoff modeling, Symposium dams and extreme floods, Grenade, tome I, pp. 46 -
55.
4 - CIGB - ICOLD, 1992 - Selection of design flood : current methods, bulletin 82.
5 - Duband et al., 1988 - Évaluation des crues extrêmes et de la crue de projet par
la méthode du GRADEX, 16e congrès CIGB, San Francisco, vol. IV, Q 63-R 60,
pp. 1 009 - 1 047.
6 - Francou (S.), Rodier (F.), 1967 - Essai de classification des crues maximales obser-
vées dans le monde, Cahiers de l’ORSTOM, vol. IV, n° 3, pp. 19 - 46.
7 - Guillot (P.), Duband (D.), 1968 - La méthode du GRADEX pour le calcul de la
36 probabilité des crues à partir des pluies, SHF, question 1, rapport 7, Paris, 7 p.
8 - Lavabre (J.), 1993 - Prédétermination des crues, cours à l’ENGREF, Montpellier, 30 p.
9 - Margoum (M.), 1992 - Éstimation des crues rares et extrêmes, le modèle AGRE-
GEE. Conception et premières validations, thèse de doctorat, École des Mines de
Paris, Cemagref, 252 p.
10 - Arnaud (P.), Lavabre (J.), 1995 - Simulation du fonctionnement hydrologique
d’une retenue d’eau, Research and Development in the field of dams. Swiss Natio-
nal Commitee on Large Dams, Crans Montana, pp. 641 - 652.
11 - CFGB, 1994 - Design flood : determination by the GRADEX method, 96 p.
12 - Leviandier (Th.) et al. - Hydrologie appliquée. Tome 1 : méthodes globales et
synthétiques, Cemagref éditeur, à paraître.