WORLD METEOROLOGICAL ORGANIZATION

Operational Hydrology
Report No. 15

SELECTION OF DISTRIBUTION TYPES

FOR EXTREMES OF PRECIPITATION
by B. Sevruk aud H. Geiger

U.D.C. 551.577.2; 556.12

I WMO - No. 560 I

Secretariat of the World Meteorological Organization • Geneva - Switzerland

1981
 © 1981, World Meteorological Organization
ISBN 92-63-10560-X

NOTE

The designations employed and the presentation of material in this publication do not
imply the expression of any opinion whatsoever on the part of the Secretariat of the World
Meteorological Organization concerning the legal status of any country, territory, city or area~
or of,· its authcrities,' or concerning the delimitation of its frontiers or boundaries. .
 CON TEN T S

Foreword ••..•..•...••.•.••.......•••..•..•••...•......•................... V

Abbreviations used in the report .....•......•........................•... VI

Summary (English, French, Russian, Spanish) •..••••••••.••••.•••••..•.•••. VII

1. INTRODUCTION •••.•..••.•.•.••....•..•••..••.•••..•••.••.••.••.•... 1

2. GENERAL •••.•••••••.••••.••••••••••••••••••••.••••.....••...•••••• 1

3. ASSUMPTIONS FOR FREQUENCY ANALYSIS •...•...••..•..•••...•..••••••• 2

3.1 Systematic error of precipitation measurement •......••........... 2

3.2 Consistency, homogeneity and stationarity of data series •....•••. 3

3.3 Independency of extreme precipitation values 3

3.4 Length af record •••.•••.••••.••.•....•....•..••..•••.•••.•••••••• 3

3.4.1 Outliers ••••••••...•.•••..••..••••.•••.•••.••••.•••.••.•.••.•••.. 4

4. METHODS OF FREQUENCY ANALYSIS OF EXTREME PRECIPITATION .. 4

4.1 Annual maxima series .....................•................ 0·0 ••••• 5

4.1.1 Selection of distribution type 5

[Link] Effect of climate and seasonal pattern on the distribution-type

selection 6

[Link] Effect of geographical pattern on the distribution-type selection. 7

[Link] Effect of precipitation duration on the distribution-type

selection 7

4.1.2 Parameter estimation of the distribution . 7

[Link] Confidence limits 8

[Link] Plotting positions ......•...••.................................. ~ 9

4.2 Partial duration series . 9

IV CONTENTS

Page

5. CONCLUSIONS •••••••••••••••••••••••••••••••••••••••••••••••••••••• 10

6. RECOMMENDATIONS •••••••••••••••••••••••••••••••••••••••••••••••••• 10

References 'O . 11

Figures and Tables . 32

 FOREWORD

The Commission for Hydrology ot its fifth session (in 1976) pointed out thot
although several \</MO publications contained examples of probobilistic approaches or
statistical procedures for the computation of extremes of precipitation, they all
lacked comprehensive information on the advantages and disadvantages of each method.
Bearing this in mind, and the need for guidance on this subject, the Secretariat l 1n
consultation with the president of the Commission for Hydrology, arranged for the
preparation of the present report written by Mr. B. Sevruk of the Laboratory of
Hydraulics, Hydrology ond Glaciology, Federal Institute of Technology, Zurich,
Switzerland, and by Mr. H. Geiger of the Swiss Federal Institute of Forestry
Research, Birmensdorf.

I should like to express the gratitude of \</MO to Messrs. B. Sevruk and

H. Geiger for the time and effort they have put into the preparation of this
excellent report on a very complex subject.

A.C. Wiin-Nielsen
Secretary~General
 ABBREVIATIONS USED IN THE REPORT

D Distribution{s) (type, function, law, etc. as used by different authors

in almost the same contextj also theoretical Df statistical DJ probability
Dr frequency Dr etcJ

EP Extremes of precipitation defined according to Figure 1.

 SUMMARY

The report reviews methods of selecting distribution types for the anolysis
of extremes of precipitation, and results obtoined. The main emphosis is on the
selection of distribution types bearing in mind the effect of climate, seasonal
and geographical pattern and duration of precipitation. A practical example of
computation of extremes of precipitation based on Swiss experience is presented in
the annex.

RESUME

Le rapport passe en revue lesmethodes utili sees pour selectionner les types
de distribution en vue de I'analyse des valeurs extremes desprecipitations, ainsi
que des resultats obtenus. Les auteurs mettent 9urtout llaccentsur le choix des
types de distribution compte tenu des effets du climat, de la repartition saisonniere
et geographique et de la duree des precipitations. Un exemple protique de col cuI
des valeurs extremes de precipitations, fonds sur des travaux suisses, figure en
annexe.

PE3lOME

B OT~eTe paccMaTpllBaroTcH MeTO~H BH60pa THTIOB pacnpe~eneHHfl

;n;[Link] aHaJIHSa :3KCTpeMaJIbHI:iIX [Link] o Ca,TI,KOTI-, a TaK)Ke nOJIy-qeHHhIe pe-:3yJJb-

~aTH. OCHOBHoe BHHMaHlle y~e~HeTcH B~6opy TRnOB pacnpeAeneHHR C y~eToM

B~HflHHH K~HMaTa, ce30HHOH H reorpa~H~ecKoM cxeMN, a TaK~8 apo~o~~HTeRb­

HOCTH oca~KOB. TIpaKTH~ecKHM npHMep no~c~eTa 3KCTpeMaTIbH~X 3Ha~eHllH

oca~KOB Ha OCHoBe OIIhITa IIIB8Hn;apllH CO~8P)f(llTCjJ B llpUJImK8HllM.

RESUMEN

En el presente informe se examinan 105 metodos utilizados en la selecci6n

de tipos de distribuci6n para el analisis de 105 valares extremas de la precipita-
c~on, as! coma de 105 resultadas abtenidas. Se subraya en particular la importan-
cia de la selecci6n de 105 tipos de distribuci6n, habida cuenta de 105 efectos del
clima, de las caracterIsticas estacionales y geograficas y de la duraci6n de las
precipitaciones. En el anexo a1 presente in forme, se expone un ejemplo practico del
calculo de 105 valores extremos de las precipitaciones basado en trabajos suizos.
 SELECTION OF DISTRIBUTION TYPES
FOR EXTREMES OF PRECIPITATION

1. INTRODUCTION

The aim of frequency analysis is to select and adjust (or fit) D to observed
data. Thus values needed by hydrologists and engineers may be extrapolated beyond
the range of the observational period. The accuracy of EP estimates depends on
various assumptions, which are discussed below.

The aim of the present report is to review the methods and results availobm
concerning annual maxima series and partial duration series of point rainfall
extremes (Figure l)~ The main emphasis is on the possibilities of selecting a
correct Do Alternative methods to EPestimation on an empirical or semi-empirical
basis are not dealt with, nor are probable maximum precipitotion (PM?) methods, the
regional analysis of EP and the modelling of time series. In the annex, an example
from Switzerland is given in detail for the computation of EP using the first and
second extremel D with the choice between them made by the test according to
Van Montfort. The aim of this annex is not only to give the reader an insight
into the problem, but olso to provide a practical example.

More than 400 references, most of which are in English or German, are listed.
Here, only brief comments could be given and the reader should consult the works
for details. The literature from eastern Europe and the U.S.S.R. has not been fully
reviewed, due to time limitations. For the fundamentals see Gumbel (1958), Chow
(1964), Remenieras (1965), Draschoff (1972), Rozhdestvensky and Chebotarev (1974),
NERC (1975), and Essenwanger (1976); for applications, see Jackson and Aron (1971),
Thiess (1975, 1980) and Platte (1979); and for examples of EP estimation see Thom
(19660), Jenkinson (1955, 1969) and the WMO Guide to hydrological practices (1974).

Theoretical derivations of 0 can be found in standard textbooks on statistics.

An abbreviated chronology of the subject is given in Table I. A serious weakness of
the past literature, however, is the use of different, very involved, and even mlS-
leading (Thiess, 1980) terminology, as well as the use of an individual's name for
his modification of some procedure.

2. GENERAL

At present there are several D used to estimate EP but the merits of their
applicability to different types of data and for different purposes has not been
established. The D used vary between countries and even within the same county, the
reasons being mostly subjective or historical. Nevertheless, since Fuller (1914),
much work has been done on selecting a relativly suitable D. As a consequence J new
D, methods for choosing between them,as well as methods of parameter estimation
have been developed. So far there are D for all practical possibilities and at least
one D can be found that fits the observations quite well or better than other D.
For floods the log PearsonlfID was recommended for general use in some countries, but
 - 2 -

there is no such recommendation concerning EP. Consequently, the selection of correct

D is done in most cases by using statistical tests or by comparing the observed data
with theoretical D using analytical and graphical methods. As stated by
Alexander (1963), such an a posteriori examination is applicable within the range of
observations but is less pertinent where extrapolation is contemplated. Thus there
is a need for objective criteria to test the applicability of different D to EP.
Bearing this in mind, the WMO Commission for Hydrology, at its fifth session,
appointed a Rapporteur on Extremes of Precipitation to undertak~ amongst other
tasks, the preparation of a report dealing with probability approaches and statistical
procedures for the computation of EP. It was intended that this report should include
exomples of objective tests of apolicability of D to different types of data and for
different purposes, and homogeneity tests for the available time-serie5 records and
recommendations for their selection.

As the rapporteur appointed by the Commission met with difficulties in

carrying out his work, in 1978 the authorities requested Mr. B. Sevruk, Ropporteur
on Accuracy of Point Precipitation Measurement, to assist in preparing the above-
mentioned report. Mr. B. Sevruk agreed to carry Qut this task in co-operation with
Mr. H. Geiger. Due to time limitations, this report only deals with the items
mentioned in the Introduction.

3. ASSUMPTIONS FOR A FREQUENCY ANALYSIS

The basic assumptions far a frequency analysis of EP are defined in Table 11.
The data series may be tested for basic assumptions using the corresponding tests,
bu~ however good a test may be, it can only indicate the occurrence and not the
cause of the disturbance. The latter is a matter of interpretation. However, in
most cases, the test together with a hydrological or meteorological interpretation of
its results presents a satisfactory solution.

If an assumption is proved to be invalid, serious errors are to be expected

and theoretically no analysis should be made. The error due to insufficient length
of record ond the errors arising from the incorrect selection of both D and the
method of parameter estimation (Table 11) are practically unknown and may be assessed
by the comparison of estimates by different D with observed data.

3.1 Systematic error of precipitation measurement

The fact that the precipitation measurement commonly used is inaccurate due
to wind-field deformation, evaporation and wetting losses, must not be given great
weight in connection with EP for short time intervals, because these losses are
generally small for heavy rainfall with large drops (Golubev et al., 1965). The
instrument error of recording gauges can be more important (height intensities are
registered with losses). According to Braslavsky et 01. (1975), the losses due to
wind-field deformation amount to up to three per cent for the medium-exposed
Tretyakov gauge wit~la daily amount of rain of 30 mm, if the wind speed at a height
of 10 m is 15 m sec • For the Hellmann gauge, with the same exposition and a wind
speed at a height 10 m of 15-20 m s-l it is five to seven per cent if the rain-
foIl intensity equals 10 mm h- l (Alleiup and Madsen, 1979). The wetting and
evaporation losses can be disregarded for daily precipitation sums. Nonetheless, in
 - 3 -

all the papers reviewed here no correction of measured precipitation values for the
above loss have been taken into consideration. Thus there is still some uncertainty
as to what extent the measurement error affects the results of frequency analysis.

3.2 Consistency, homogeneity and stationarity of data series

Inconsistency and non-homogeneity (as defined in Toble 11) are related to

the physical causes of disturbing effects and have the same impoct on the dota.
Thus it is not always possible to differentiate between thelTI if the station's
records are missing. Statistically, a precipitation data series is consistent and
homogeneous if all data come from the some population. Since EP vary to 0 great
extent, the inconsistency and/or non-homogeneity is difficult to decipher (Miller,
1972). However, because of their cumulative effects, they can be found by testing
the combined depth records (Former and Fletcher, 1972), or the yeorly precipitation
sums. As a consequence, some adjustment of data may be necessary (Table 11). This
does not pose a problem for monthly or yearly sums but is controversial for EP of
shorter time intervals. Nevertheless, the consistent and/or homogeneous data series
may still be non-stotionery due, for example, to climotic fluctuotions (Toble 11).

3.3 Independency of extreme precipitation values

According to Flohn (1957), it is difficult to detect climotic voriotions if

the observational period is less than 200 years. Temporal variations in heavy
rainfall have been noted but are less substantially documented (O'Mahony, 1965;
Tomlinson, 1978).

The references to possible dependencies and cycles of EP are sparse (HEC,

1975; Jenkinson, 1975; Frederic et al., 1975). Thus the elements of annual series
of short-duration rainfall may, in practice, be assumed to be independent (Hershfield
and Kohler, 1960; Guillot and Duband, 1967). However, this cannot be true for
partial duration series with comparatively more elements (Table 11).

3.4 Length of record

Comparisons of EP estimates using various lengths of record show that an

observational period of 15-25 years is not sufficient in many cases to define a
"stable" D. Geiger (1975) found differences of up to ! 15 per cent by estimoting
daily EP over 100 years using excessive 3D-year periods in central Switzerland with
o moderately humid climate. According to Kuznetsovo (1964), the length of record
needed to define D is reloted to the generol humidity of the region ond to its
physiographic conditions determining the variability of the daily precipitotion sum.
Thus the estimates using the 25-year periods only differ from 12-15 per cent in the
humid northern U.S.S.R. (Table Ill: Leningrod, Solechard) but differ from an
amount to 25-30 per cent in the central part (Table Ill: Kiev, Surgut) and as much
os ~OO ~er cent in southern [Link] (1964) concluded thot even the 50-year
per10d 15 not adequate in regions with a distinct periodic fluctuation of precipita-
tion (Table Ill: Bornaul, Nerchinsky Zovad) becouse the difference may still amount
to 40-50 per cent. However, in general, a record of 40-50 years is satisfactory in
most cases (Kuznetsova, 1964; Lambor, 1967; Miller, 1972).
 - 4 -

3.4.1 Outliers

An outlier is an extremely rare event which disrupts the pattern established

by other observations (Dickinson, 1977). Thus, its return period may be considerably
longer than the period of record, or the Qutlier may represent quite q separate popu-
lation from those represented by the other EP. An outlier is due to the shortness
of record ond/or due to weather situations quite unlike those of other EP in the
sample (Tomlinson, 1978). It can be identified by using the ratio of lOO-to 50-year
EP estimates from nearby stations and, if needed, replaced by the next largest EP in
the same year. However, if the weather situation is not very different from that of
other events, the outlier should not be separated from the sample and its return
period should be assessed to be very long, i.e. considerably in excess of 100 years.
The literature concerning the problem of outliers in general or in normal distributed
samples is numerous (Anscambe, 1960; Dixon, 1950, 1953, 1960, 1968; Ferguson, 1961;
Grubbs, 1950, 1969; Guttman, 1969, 1971, 1972, 19730, 1973b; Tietien et 01. 1973:
Tukey, 1962). An analysis of outliers of 300 different skewed samples ot the
PearsonIIID for floods was made by Eggers (1979). He showed that the coefficient
of skewness changes very quickly due to outliers and that it is possible to detect
and define outliers and to deal with them appropriately. The references concerning
outliers in EP are sparce (Table 11). Tomlinson (1978) presented an interesting
study on outliers of EP on the I extremal D paper.

4. METHODS OF FREQUENCY ANALYSIS OF EXTREME PRECIPITATION

There are three possible means of analysing EPa At one extreme there is the
frequency analysis of annual maxima series; at the other, modelling of time series;
and in between there are methods of analysis of partial duration series. In the
latter category are the annual exceedance series which consist of the N largest
values in the record, where N is the number of years record.

Only annual maxima and partial duration series are dealt with in this report
(Figure 1; Figure 2, models I and 11). The procedures of analysis are similar.
In contrast, the modelling of time series is based on other theories (Chow, 1964).
In recent years, more sophisticated methods based on theories of stochastic processes
have been developed for the analysis of partial duration series (Todorovic et al.,
1968, 1975; NERC, 1975; Taesombut and Yevievich, 1978). These are not dealt with
further here, nor are models III and IV which are shown in Figure 2.

Hydrologists and engineers are interested in the occurrence of a certain

event, and generally not in the occurrence of this event as a yearly maximum. So
the results obtained from a portial duration series analysis would normally be
appropriate, but because the difference between the results of frequency analysis of
both series are small for return periods greater than ten years (Langbein, 1949;
Chow, 1953; Kuznetsova, 1964; Bruce, 1968; Draschoff, 1977; Cunnane, 1973;
Rosbjerg, 1977), the annual maxima series are more frequently used for reasons of
convenience and better statistical background. For return periods of less than ten
years, the differences are remarkably large and cannot be disregarded.

If certain assumptions can be made about D concerning the number of events

 - 5 -

during the year, theoretical factors can be derived for converting values from the
partial duration series to the annual series and vice versa (Chow, 1964). If these
assumptions do not hold, then empirical conversion factors can be derived. They differ
slightly from the th~oretical ones (WMO, 1974; Beran, 1977; Frederick et. al., 1977)
and for the two-, five- and-ten-year return periods are 1.13, 1.04 and 1.01
respectively (WMO, 1974).

For purposes of comparison both series may be analysed. If the period of

observation is relatively short, or when there are "sample outliers" in the series
(Table 11), as well as in the cases where the design is affected by comparatively
minor events (storm-sewer systems), the partial duration series method has some
advantages (Dickinson, 1977). (For disadvantages see Hershfield and Wilson, 1960;
WMO, 1974). However, the partial duration series methods require 1.65 N items or
more for the same return period in order to estimate the magnitudes more efficiently
than the annual maxima method (N is as above) (Cunnane, 1973; Taesombut and
Yevievich, 1978).

In recent years, increasing attention has also been given to the monthly EP
(Stol, 1971; Snyder, 1975; Dickinson, 1977; BUishand, 1978), and to the five-year
EP (Jenkinson, 1969; Uppala, 1978; Obasi and Nimira, 1977). (Mc Illwrainth (1953)
considers the three-year E~)

As an input of the frequency analysis both the "true" or "clock-hour" or

observational-day data may be used. The difference between the results depends on
the duration of rainfall, and the IItrue ll EP are, on the average, 14 per cent greater
on a daily basis~ (for conversion factors for various duration ~ntervals see:
Jennings (1952), Linsley et al. (1958), Hershfield and Wilson (1958), Weiss (1964),
Guillotand Duband (1968), BUchner (1970), Dr<lschoff (1972),Cao (1974), NERC (1975),
Geiger (1975).)

4.1 Annual maxima series

4.1.1

As may be seen in Table I, different names are frequently used for the same
D. In Table IV an attempt has been made to clarify any possible confusion. There
are different D (Alexander et al., 1969; Thiess, 1975) but there is no firm theore-
tical support for the exclusive USe of one or another D. There is some theoretical
justification for the application of the extreme-value theory (e.g. Gnedenko, 1943;
Gumbe1, 1958; Jenkinson, 1969), the Pearson D (Kryukov, 1973) and the log-normal D
(Chow, 1954) to EP analysis.

As stated by Gumbel (1958), the good fit (to floods) cannot be foreseen from
the theory. Gupta (1970) admitted that the choice of D is often dictated by con-
venience or procedural policy, primarily owing to the lack of a basis for selecting
the distribution of best fit. Frequently, various D are fitted to the data and the
one with the best fit is selected. Thus the decision is made on an empirical
basis by using subjective visual techniques or more objective statistical tests
(Table V). Both techniques meet with criticism (Moran, 1957 and Table V). The
results of tests depend in part on the index of fit and/or on the plotting positions
used (see [Link]); both are controversial. In addition, a test may show that all
 - 6 -

o or none are "correct" (Koopman, 1936). Another procedure for the selection of 0
is based on the best fit between the estimated and the theoretical coefficients of
skewness and kurtosis (Wu and Goodridge, 1974). New techniques based on Bayesian
decision theory (see Bode and Unny, 1976, for further references) and on non-
parametric statistics (Siegerstetter, 1973) discriminate between alternative D
but have only just begun to be used.

As may be seen from Table VI, prabably the mast frequently used 0 are the I
extremal D and the log-normal 0, which may justify them empirically to some extent
for EP analysis. The argument in favour of the use of the I extremal D seems to be
more convincing (Table VII). However, the differences between the estimates by these
two D are small - up to ten per cent, which is almost the same as compared with the
estimates by the PearsonIIID and the log PearsonIIID (Table VIII). In contrast, the
estimates by the 11 extremal D differ by up to 80 per cent (Table VIII). Since we
know that deviations from the I extremal and log-normal straight line exist, the
question is whether these are due to outliers (Hershfield and Wilson, 1960;
Dickinson, 1977; Table 11) and/or to sample errors (Dickinson, 1977) or, rather, due
to climatic and geographical factors. Meanwhile, experience from various countries
shows that there is some climatic and geographical explanation for the greater
suitability of one type of extremal D over another (see [Link].). This would mean
that the general use of one D may not be satisfactory in all cases. However! a
combination of the I extremal D and the 11 extremal D, or eventually the III extremal
o as an alternative, as used in some studies -(Table VI), seems, to some extent, to give
satisfactory results. In this connexion the Generol Extreme Value D proposed by
Jenkinson (1969) indicates a possible but not definitive solution to the problem.
The application of "combined D" (Ludwig, 1979) in EP analysis presents another
possibility, but its theoretical basis is rather poor.

The 11 extremal D has been found to be the most satisfactory of five

compared D for summer data in Illinois (U.S.A.) whereas the log-normal. D was superior
for use with winter, spring and autumn data (Huff and Neill, 1959b). Since many of
the annual maxima were from summer storms, the former D appeared to be as good as,
if not better than, the latter D.

According to Hershfield and Wilson (1960), however, there are no significant

differences in D (I extremal) associated with tropical and other storms. Greater
suitability of the 11 extremal D over I in western Finland was partly ascribed by
Uppala (1978) to the higher intensity and greater can tent of smal~ and large-scale
disturbances in the region. Wurtele and Roe (1978) suggest the 11 extremal 0 for
deserts, which have an effective lower limit but meteorologically no upper limit, and
the I extremal D for moderate-to-high rainfall regimes where no effective lower limit
exists.

According to Kuznetsova (1964) the binomial D may give better results than
the log-normal D in regions or localities where insufficient saturation or atmospheric
condensation as well as low temperature are limiting factors with regard to maximum
daily precipitation and, consequently, where EP are small. On the other hand, in
regions with particularly advantageous precipitation conditions where exceptionally
 - 7 -

heavy storms cause characteristically bent curves on log-normal paper, the log-
normal D is not acceptable due to the great asymmetry of D.

Hershfield (1962) stated: "Visual inspection of hundreds of plots of

extremes during the past several years on both log-normal and extreme-value probabi-
lity paper has revealed that the data exhibit no systematic geographical pattern with
respect ~o a straight line; that is, no region shows a consistent concave-upward
or concave-downward pattern for a particular duration. There appears to be as much
variation among stations for the same duration, even in regions where physiographic
influences on the rainfall regime appear to be absent, as there is within several
durations at the same station".

Krishnan and Kushwaha (1975) found that the deviations from the I extremal D
straight line have a geographical explanation over India. Thus the annual maxima
series with the ratio ~1I~2 =1 (Jenkinson, 1955, Table IX) appears to be more
common in India than the series wi th 0"1/e1'2 = 1 which appears in regions with compara-
tively less intensive precipitation.

In western Finland, the maximum annual series have been better fitted by
the 11 extremal D, whereas the III extremal D was correct in eastern Finland and the
I extremal D along the line from south to north (Uppala, 1978).

In pre-Alpine north-eastern Switzerland the I extremal D has generally been

used. However, in some regions the 11 extremal D was found to be more appropriate
(Zeller et al., 1977).

Most studies analyse one-day or 24-hour rainfall. For shorter rainfall dura-
tion intervals, from five minutes to 24 hours, the same D~achr of cours~ with different
parameters) may be taken (Hershfield, 1962; Rodda, 1970; Miller, 1972; Dickinson,
1977; Frederick et al., 1977). In Belgium,Sneyers (1978) uses the I extremal D for
rainfall durations of one to 30 minutes, the 11 extremal D for durations of one to
24 hours and a mixture of both D for durations in between. In Californi~Wu and
Goodridge (1974) found the PearsonllID to be best for durations of four minutes to
30 hours and the Weibull D to be best for durations of longer than 30 days. Zeller
et al. (1978) use the same D for durations of up to five days. Far longer durations,
the D tends to become normal or can be normalized (Rodda, 1970). For analysing EP of
shorter duratio~Majumdar (1975) presented a method which does not depend upon the
type of D. Thus, for practical reasons, the same D happens to be taken over a range
of various precipitation durations.

4.1.2 Parameter estimation of the distribution

Estimation of the parameters from observational data may be done either nume-
rically (Table IX) or graphically (Dalrymple, 1960; Farmer and Fletcher, 1972). The
latter is based on the assignment of plotting positions and on the use of probability
paper (Table I). This method is simpler, more illustrative and expeditious than
 - 8 -

numerical methods, but not as accurate and objective and thus must be used with
cautionc

From the stotisticol point of view, the method of moximum likelihood is

believed to be the most efficient for lorge somples (Lowery ond Nash, 1970;
Samuelssan, 1972; Matalas and Wallis, 1973), but in the past it was numerically
difficult (Smirnow et al., 1969). For small samples it leods to generally poor
agreement (Bobee and Robitaille, 1972). This method as well as the method of
moments do not need the assignment of plotting positions. Both methods give equi-
valent results for lorge samples which follow the PearsonlIID (Motalas and Wallis,
1973; Eggers 1970 and for classified data (Grant, 1973). However, maximum likeli-
hood estimates yield solutions which are less biassed and less variable than the
comparable moment estimates (Motolos ond Wollis, 1973). For the I extremol D, the
moments are sufficiently accurate and virutally unbiassed and simplest to apply
(Lowery ond Nash, 1970). For highly skewed D, however, they give rather poor
estimates of porameters (Fisher, 1941; Vatnsdol, 1946; Thorn, 1958; Koberg ond
Eggers 1973) ond a correction of the coefficient of skewness has been recommended by
Gloss (1966), Bobee and Robitoille (1975, 1977), Bobee (1975) and Landwehr et al.
(1978) for log PeorsonlIlond PeorsonIIID. Rao (1980b) shows that by using the method
of mixed moments the coefficient of skewness cannot be estimated. The fit by
moments to samples with Qutliers is also unsatisfactory (Jenkinson, 1969; Snyder,
1972; Siegerstetter, 1973). The method of leost squares is not very efficient
(Lowery ond Nosh, 1970; Thiess, 1975) and must not be odopted to particulor D
(Snyder, 1972). With coution, it may be used for extropolotions in the criticol
ronge (Smirnow et 01., 1969). The assignment of plotting positions is needed.
Dolrymple (1960) rejected this method, but hydrologists use it frequently (Droschoff,
1972).

A survey of fitting methods used for the extremal D has been given by Mann
(1968) and Lowery and Nash (1970). The lotter olso compored the aforementioned
numerical methods. The comporison of both Gumbel's and Jenkinson's methods opplied
to EP have been mode by Krishnon and Kushwaho (1975) ond Somuelsson (1972)
(Figure 3). The differences are generally small, up to 18 per cent (Table VIII).
Krishnan und Kushwaha (1975) found that the Jenkinson's (1955) method has 0 tendency
ta underestimate observed values. Thiess (1975) gives a detailed review of 011
fitting methods and suggests 0 new one, which allows D to be fitted correctly using
variable parameters and a graphicol technique.

[Link] Confidence limits

Confidence intervals and the variance of estimates have been studied by

Kaczmorek (1957), Kaller (1974) and Widmoser (1977) for Gumbel's method (Toble IX)
by Nash and Amorocho (1966) for moments, by Kimbal (in Gumbel, 1958) for the moximum
likelihood method, by Matalas and Wallis (1973), Bobee (1973), Bobee and Vileneuv~
(1973) and Eggers (1979) for PearsonIIID, by Kite (1975), Kaller ond Riebe (1979) for
several D and by Lawless (1972, 1974, 1978) for the III extremal D. The results can
be used to construct confidence curves (Draschoff, 1972).
 - 9 -

The selection of the most appropriate formula of plotting positions has been
discussed by many authors, beginning with Hazen (1914b) and surely not ending with
Cunnane (1978), who gives an excellent review of the problem. Many formulae have
been developed, reviewed by Chow (1964), HEC (1975), Thiess (1975), and compared
by Kimball (1960), Thiess (1975) and Cunnane (1978), but those with theoretical
interpretations appear mainly since 1960 (Cunnane, 1978). Controversial results may
be illustrated by the widely accepted Weibull formula which was supparted by Gumbel
(1958) and found by Benson (1962) to be better than the Beard and Hazen formuloe.
However, as stated by Cunnane (1978), the two latter formulae are much more suitable
for normal D and I extremal D than Weibull's formula, which is efficient when D is
uniform. NERC (1975) and Cunnane (1978) recommend the Blom (1958) formula:
T = (N + 1/4) / (m - 3/8) for the normal ~ and the Gringorten (1963) formula:
T = (N + 0.12) / (m·- 0.44) for the I extremal D, where T = return period, N = total
number of items and m = order number of the items arranged in descending magnitude.
According to Thiess (1975), the most efficient formula for the normal D is that by
Alexeyev (1963) and for skewed samples that by Weibull. However, differences between
estimates due to different plotting positions ore generally small (Reich, 1970).
Benson (1967) showed that the limiting value of the average probability of all floods
above the k-year flood is equal to 1/2 k when the California, Weibull or Hozen
formulae are used and Appel (1968) showed that the same is valid in the case where
the plotting pasition far T has the general form of (N + b) I (m ~ a).

4.2 Partial duration series

The vorious methods of analysis af partial duratian serles differ basically

in two ways: firstly, in the assumption about D of the mognitudes (Table X) and
secondly in the assumption about D of the occurrences of corresponding events during
a year (Table XI). For the former, the mast commonly used are the exponential,
gamma, Pearsonllland log PearsonIIID and, for the latter, the Paissan D (Table X, XI).
According to NERC (1975), four models far the analysis of partial duration series
may be distinguished (Figure 2).

Several D are mentioned for analysing the frequency of mognitudes (Table X)

but not all of them seem to represent the behaviour of the magnitudes properly. As
the density of the values is highest just above the base level, the D should be
reverse J-shaped similar to the exponential D; The extremal D and log-normal D
have a peak and therefore cannot represent the behaviour of the data in the neigh-
bourhood of the base level. If the magnitudes in a partial duration series follow
the exponential D, the annual maxima of the same sample follow the I extremal D. In
the case where the latter follow 11 and III extremal D, Draschaff (1972) derived the
corresponding D for the partial series. Musy (1973) compared the PearsonIllD
(gamma) with the exponential D and stated that the former was superior far the
partial series drawn from the stations Lausanne and Geneva (33 and 64 years,
respectively) •

Most methodological problems in analysing partial duration series are of the

same type as with annual series and the pertinent information can be found in the
preceding chapters. An important factor is the selection of the base level, which
influences the selection of D and the estimation of parametera.
 - 10 -

5. CONCLUSIONS

The frequency analysis of extremes of precipitation is generally based on

the annual maxima series. The accuracy of analysis depends on the validity of some
assumptians. The assumptions may be tested. Thp error due to insufficient length
of precipitation record amount to 30-50 per cent in the range of extrapo-
lation. The error due to incorrect selection of the distribution type may be even
greater. In addition, the differences between estimates by the some distribution
type but fitted by different methods amounts to up to 20 per cent.

The most frequently used are the I extremol and log-normal distributions.
On the average the former gives greater estimates (up to ten per cenVand, in
addition, has some advantages compared to the latter. Some distributions, however,
may be superior to others under given seasonal and/or geographical conditions. As
there are no criteria of applicability of different distributions and different
methods of parameter estimation to different precipitation data, their relative
suitability is based on statistical tests and on a comparison of observed values
with theoretical distributions. In this respect the General Extreme Value distribu-
tion proposed by Jenkinson indicates a possible solution.

From all methods of parameter estimation the method of maximum likelihood is

believed to be the mast efficient one but numerically difficult. In contrast, the
method of moments is not as accurate but numerically simpler. In some cases it may
be used with roughly the some effect. Both methods do not need the assignment of
plotting positians which is still controversial: the earlier widely accepted
Weibull formula is generally not efficient.

Some methodological problems in analysing the partial duration series are the
same as in the case of annual maxima series. Conversion factors have been derived
between the estimates by both series.

6. RECOMMENDATIONS

Research based on physical grounds should be supported so that more can be

learned about the distribution af extremes of precipitation and to establish a
sounder basis far the correct selection of the particular distribution. This may be
reached by close co-operation between mathematicians, physicists, and practical hydrolo-
gists. In this context, a workshop on the subject as well as international comparison
tests could be recommended to the WHO authorities os the further step towards 0
generally acceptable solution. Such a salution is urgently needed in many countries
where, as yet, only empirical methods are used and no decision has been made con-
cerning the choice of distribution types for the analysis of extremes of precipitation
on a national basis.
 - 11 -

REFERENCES

Aitchison, J. and J. A. C. BROWN: The lognormal distribution. University of

Cambridge, Monograph 5, 1969.

Alexander, G. N.: Using the probability of storm tronsposition for estimoting the
frequency of rare floods. J. Hydrol.,l (1), 46-57, 1963.

Alexander, G. N., A. Koroly and A. B. Sust: Equivalent distributions with applica-

tion to rain foIl as an upper bound to flood distributions. J. Hydrol., 9 (3),
322-344, 1969 and 9 (4), 345-373, 1969.

Alexeyev, G. A.: To the question of estimation of empirical quantiles and corre-

lotion coefficients (in Russia). Meteorol. Hydrol. (Moscow), 4, 16-23, 1963 0 •

Alexeyev, G. A.: Methods of calculotion of maximum rainfall discharges using pre-

cipitation variation curves (in Russian). [Link] [Link]., No. 107,
3-62, 1963b.

Alexeyev, G. A.: Objective statistical methods for determination of storm pre-

cipitation characteristics. Meteorol. Hydrol. (Moscow) 7, 22-27, 1966.

Alexeyev, G. A.: Objective statistical methods of computation and generalization

of the porameters of maximum roinfoll runoff. lASH publ. 84, 114-127, 1969.

ALLERUP, P. and H. Modsen: Accuracy of point precipitation measurements. Danish

[Link]., No. 5, 84 p" 1979.

Anderson, H. W.: Detecting hydrologic effects of changes in watershed conditions by

double-mass analysis. Trans. AGU, 36 (1), 119-125, 1955.

Anderson, R. L.: Distribution of the serial correlation coefficient. Ann. Math.

Statist., 8 (1), 1-13, 1941.

Anonym: Flow in California streams. [Link] [Link]. Works Bull. 5,

chap. 5, 1923.

Anscombe, F. J.: Rejection of outliers. Technometrics, 2 (2), 123-147, 1960.

Appel, Ch. A.: A note on the average probability of extreme events. Water Resour.
Res., 4 (6), 1359, 1968.

Artino, S. and A. Lamberti: Extreme values law parameters estimation. Rendiconti

Accademia delle Scienze di Bologna, Serie XIII, Tom V, 1978.

Ayoade, J. 0.: A preliminary study of the magnitude, frequency and distribution of

intense rainfall in Nigeria. Proc. Symp. on Flood Hydrology, Nairobi, 1975. TRRL
Supplementary Report 259, 95-112, Crowthorne, Berkshire, 1977.
 - 12 -

Baghirathan, V. R. and E. M. Shaw: Rainfall depth-duration-frequency studies for

Sri Lanka, Journal of Hydrology, 37 (3), 223-239, 1978.

Bardsley, K. E.: A test for distinguishing between extreme value distributions.

J. Hydrol., 34 (3/4), 377-381, 1977.

Belke, D.: Die Anpassung von Verteilungsfunktionen. 6. Fortbildungslehrgang fUr

Hydrologie DVWW, Bad Herrenalp, 1974 (Germony).

BELL, F. C.: Generalized rainfall-duration-frequency relationships. J. Hydroul.

Div. ASCE, 1, 311-327, 1969.

Bensan, M. A.: Plotting positions and economics of engineering planning J. Hydraul.

Div. ASCE, 88 (6), 57-72, 1962.

Bensan, M. A.: Average probability of extreme events. Water [Link]., 3 (1),

225, 1967.

Benson, M. A.: Uniform flood-frequency estimating methods for federal agencies.

Water Resaur. Res., 4 (5), 891-908, 1968.

Beran, M. A. and M. J. Nozdryn-Plotnicki: Estimation of low return floods, Hydrol.

Sci. Bull., 22 (2), 275-282, 1977.

Bernier, M.: Comparaison des lois de Gumbel et de Frechet sur l'estimation des debits
maxima de crues. La Hauille Blanche, 14 (1), 47-54, 1959.

Bernier, J.: Sur la theorie du renouvellement et son application en hydrologie,

Electricite de France, HYD, 67 (10), 1967.

Bernier, J.: Etude de la stationnarite des series hydrometeorologiqueso La Houille

Blanche, 32 (4), 313-319, 1977.

Blom, G.: Statistical estimates and transformed beta-variables. John Wiley & Sons
Inc. New York, Stockholm, 1958.

Bobee, B. and G. Morin: Determination des intervalles de con fiance de la loi

Pearson 3 par les statistiques d'ordre. J. Hydrol., 20, 137-153, 1973.

Babee, B. and R. Robitaille: Correction of bias in the estimation of the coefficient

of skewness. Water Resour. Res., 11 (6), 851-854, 1975.

Bobee, B.: The log Pearson type 3 distribution and its application in hydrology.
Water Resour. Res., 11 (5), 681-689, 1975.

Bobee, B. B. and R. Robitaille: The use of the Pears on Type 3 and log Pearson type
3 distributions revisited. Water Resour. Res., 13 (2), 427-443, 1977.

Bodo, B. and T. E. Unny: Model uncertainty in flood frequency analysis and frequency-
based design. Water Resour. Res., 12 (6), 1109-1117, 1976.
 - 13 -

Borgman, L. E.: Risk criteria. Journal Waterways Harbours Division, ASCE, 89,
1-35, 1963.

Brakensiek, D. L.: Fitting generalized lognormal distribution to hydrologic data.

Trans. AGU, 39, 469-473, 1958.

Braslavsky, A. P., L. R. Struzer, K. B. Sherghina and S. P. Christyayeva:

Practical recommendations for estimation of corrected values of atmospheric precipi-
tation depth by 24-hour observations (in Russian). Trans. [Link]. Res.
[Link]., 52, 34-39, 1975.

Bridges, T. C. and C. T. Haan: Procedure for determining the reliability of preci-

pitation probabilities estimoted from the gammo distribution. Agric. [Link].
Kentucky, [Link]. 3, 41 p.. Lexington, 1971.

Bridges, T. C. and C. T. Haan: Reliability of precipitation probobilities estimated

from the gamma distribution. Mont. Weather Rev., 100 (8), 607-611, 1972.

Bruce, J. P.: Atlas of roinfall intensity-duration frequency data for Canada. Dept.
Trans. Meteorol. Branch, Ottawa, Canada, Climatological Studies No. 8, 1968.

Brunet-Moret, Y.: Etude de l'homogeneite de series chronologiques de precipitation

annuelles par la methode des doubles masses. Cahiers ORSTOM, Serie Hydrologie 8
(4), 1971.

Brunet-Moret, Y.: Cartographie des precipitations et determination de l'averse de

project (design storm) en Afrique Occidentale. Proc. Symp. on Flood Hydrology,
Noirobi 1975. TRRL Supplementary Report 259, 32-40, Crowthorne, Berkshire 1977.

Buchner, H.: Ermittlung des maximal moglichen Regens ouf statistischer Grundloge.
WWT 20 (10), 348-351, 1970.

Buishand, T. A.: Some remarks on the use of doily rainfall models. J. Hydrol., 36
(4), 295-308, 1978.

Burges, S. J., D. P. Lettenmoier ond C. L. Bates: Properties of the three-parameter

log normal ·probability distribution. Water Resour. Res., 11 (2), 229-235, 1975.

Calder, I. R. and C. H. R. Kidd: A note on the dynamic colibration of tipping-bucket

gauges. J. Hydrol., 39 (3), 383-386, 1978.

Calenda, G., A. Petaccia and A. Togna: Theoretical probability distribution of

critical hydrologic events by the partial-duration series method. J. Hydrol., 33
(3/4), 233-245, 1977.

CAD, C.: A contribution to statistical depth-duration-frequency analysis. J. Hydrol.,

22 (1/2), 109-129, 1974.

Cehak, K.: On flood probabilities of east Alpine rivers. J. Hydrol., 20 (1),

65-82, 1973.
 - 14 -

Chang, M. and R. Lee: Objective double-mass analysis. Water Resour. Res., 10(6),
1123-1126, 1974.

Chin, E. H. and J. F. Miller: The distribution of precipitation associated with the

Kansas City flood, September 12-13, 1977. (Abstract) Trans. EOS AGU, 59 (12),
1063, 1978.

Chow, V. T.: A general formula for hydrologic frequency analysis. Trans. AGU, 32,
231-237, 1951.

Chow, V. T.: Frequency analysis of hydrologic data with special application to

rainfall intensities. Univ. Illinois Eng. Expt. Stn. Bull., 414, July 1953.

Chow, V. T.: The log-probability law and its engineering applications. Proc. ASCE,
80 (536), 1-25, 1954.

Chow, V. T.: Determination of hydrologic frequency factor. J. Hydraul. Div. ASCE,

85 (7), 93-98, 1959.

Chow, V. T.: Handbook of applied hydrology. McGraw-Hill Book Comp., New York,
1964.

Clarke-Haftstad, K.g A statistical method for estimating the reliability of a

station-year rainfall record. Trans. AGU, 19, pt. 11., 526-529, 1938.

Clarke-Haftstad, K.: Reliability of station-year rainfoll frequency determination.

Trans. ASCE, 107, 633-683, 1942.

CMWDP, Commissioner Ministry of Water Development and Power, Tanzania: Depth-

duration-frequency relationship for Ubungo rainfall station. Proc. Symp. on Flood
Hydrology, Nairobi, 1975. TRRL Supplementary Report 259, 51-56, Crowthorne,
Berkshire 1977.

Cunnane, C.: A particular comparison of annual maxima and partial duration series
methods of flood frequency prediction. J. Hydrol., 18 (3/4), 257-272, 1973.

Cunnane, C.: Unbiased plotting positions - a review. J Hydrol. 37 (3/4), 205-222,

1978.

Dalinsky, S.: The use of non-dimensional one-parameter functions for the determina-
tion of regional rainfall depth-duration-frequency relations. Hydrol. Sci. Bull.,
16 (2), 67-77, 1971.

Dalrymple, T.: Flood frequency analysis, U. S. Geol. Survey - Water Supply Pap.
1543, A. Part 3. Manual of Hydrology, 1-80, 1960.

Davis, D., L. Duckstein and M. M. Fogel: Uncertainty in the return period of

maximum hydrologic events: A Bayesian approach. J. Hydrol., 31 (1/2), 81-95,
1976.
 - 15 -

Deiniger, R. A. and J. D. Westfield: Estimation of the parameters of Gumbel's third

asymptotic distribution by different method. Water Resour. Res., 5 (6), 1238-1243,
1969.

Della Lucia, D. and S. Fattorelli: Regolirizzazione delle piogge intense nel

Trentino. Monti e Boschi, 25 (1), 1/1 - 1/14, 1974.

Dhar, O. N. and S. M. T. Farooqui: A study of rain falls recorded at the Cherrapunji

observatory. Hydrol. Sci. Bull., 18 (4), 441-450, 1973.

Dickinson, T.: Rainfall intensity-frequency relationships from monthly extremes.

J. Hydro!., 35 (1/2), 137-145, 1977.

Dingens, P., W. Moerdijk and H. Steyaert: De distributie van de dagelijkse neerslag

te Gent, Natuur wet. Tijdschr., 48, 103-140, 1966.

Dixon, W. J.: Analysis of extreme values. [Link]. Statist., 21 (4), 488-506, 1950.

Dixon, W. J.: Processing data for outliers. Biometries, 9 (I), 74-89, 1953.

Dixon, W. J.: Simplified estimation from censored normal samples. [Link].,

31 (3), 385-391, 1960.

Dixon, W. J. and J. W. Tukey: Approximate behaviour of the distribution of win-

sorized t (trimming/winsorization 2), Technometrics, 10(1), 83-98, 1968.

Dodd, £. L.: The greatest and the least variate under general laws of error. Trans.
Amer. Math. Soc., 25, 525, 1923.

Downton, F.: Linear estimates of p-arameters in the extreme value distribution.

Technometrics, 8, 3-17, 1966.

Draschoff, R.: Haufigkeitsanalyse langfristiger Niedersch1agsbeobachtungen a1s

Grund1age der Abflusstatistik k1einer Vorfluten. Inst. Wasserwirtsch., Hydra!. und
Landwirtschaftlicher Wasserbau. Techn. Univ. Hannover, Mitt. 243-277, Hannover, 1972.

Duckstein, L., M. M. Foge1 and Ch. C. Kisie1: A stochastic model of runoff-producing

rainfall for summer type storms. Water Resour. Res., 8 (2), 410-421, 1972.

Eggers, H.: Parameterfreie statistische Methoden zur Analyse van Datenreihen. Mitt.
Th.-Rehbock-F1ussbau1aboratorium, Heft 158, 271-197, Kar1sruhe, 1970.

Eggers, H.: Haufigkeitsanalysen und Vertei1ungsfunktionen. VI. Fortbi1dungs1ehrgang

fUr Hydrologie, DVWW, Bad Herrena1b, 1-16, 1974.

Eggers, H.: Der Einfluss seltener Ereignisse bei der Bestimmung der Hochwasser-
wahrscheinlichkeit. Versuchsansta1t fUr Wasserbau und Ku1turtechnik, Universitat
Fridericaine, Kar1sruhe, Mitt.167, 180 p., 1979.
 - 16 -

Eidgenassisches Amt fUr Strassen- und Flussbau: Die grassten bis zum Jahre 1969
beobachteten Abflu5smengen van schweizerischen Gewassern. Bern, 1974.

Essenwanger, O. M.: Applied statistics in atmospheric science. Port A: Frequencies

and curve fitting. Elsevier [Link]., 412 p., 1976.

Former, E. E. ond J. E. Fletcher: Roinfoll intensity-durotion-frequency relotions

for the Kosatch Mountoins of Northern Utah. Water Resou. Res., 8 (1), 266-271,
1972.

Ferguson, T. S.: On the rejection of outliers. Univ. of California Press; Proc.

Berkeley Symp. Math. Statist. and Prob. 1, 253-285, 1961.

Fiddes, D., J. A. Forsgate and A. O. Grigg: The prediction of storm rainfall in

East Africa. Transport and Road Research Laboratory, Report 623, 1-26, Crowthorne,
Berkshire, 1974.

Fisher, R. A. and L. H. Tippett: Limiting forms of the frequency distribution of the

10rgest ond smallest member of a sample. Proc. Cambridge Phil. Spc., 24, 180-190,
1928.

Fisher, R. A.: Tests of significance in harmonic analysis. Ray. Soc. London Proc.
$er. A, 125 (796), 54-59, 1929.

Fisher, R. A.: Statisticol methods for research workers. 8th ed., 344 pp" Oliver
and Boyd, London, 1941.

Fisz, M.: Wahrscheinlichkeitsrechnung und mathemotische Statistik. VEB, Deutscher

Verlag der Wissenschaften, 1962 (Germon Democratic Republic).

Fogel, M. M. and L. Duckstein: Point rainfall frequencies in convective storms.

Water Resour. Res., 5 (6), 1229-1237, 1969.

Foster, H. A.: Theoretical frequency curves and their application- to engineering

problems. Trans. ASCE, 87, 142-173, 1924.

[Link], H. A.: Duration curves. Trans. ASCE, 99, 1213-1235, 1934.

Frechet, M.: Sur la loi de probabilite de llecart maximum. Ann. Soc. Polonaise
Math. (Krakow),Vol. 6, 93-116, 1927.

Frederick, R. H., V. A. Meyers and E. P. Auciello: Five- to 60-minute precipitotion

frequency for the eastern and central United States. NOAA Tech. Memo, NWS Hydro-35,
1977.

Friedman, D. G. and B. F. Janes: Estimating of rainfall probabilities. Agric.

Expt. Stn., Connecticut Univ., Bull. 332, 22 p., Storrs, 1957.

Fuller, W. Eo: Flood flows. Trans. ASCE, 77, 564-594, 1914.

 - 17 -

Galton, F.: Statistics by intercomparison with remarks on the law of frequency of

error. The London, Edinburg~ Dublin Phil. Mag. [Link]., 4th ser. vol. XLIX, p. 33,
1875.

Gansloser, G.: Hydrologische Untersuchungen in Aethiopien zur Ermittlung der

SpitzenabflUsse aus Kleinen Einzugsgebieten, [Link]., Hydrol. und
Landwirtschaftlicher Wasserbau, Tech. Univ. Hannover, No. 14, 97-106, 1969.

Geiger, H.: Extremwertstatistik. Eidg. Anstalt fUr dos Forstliche Versuchswesen,

Birmensdorf. Interner Bericht H 1, 108 P., Miirz 1975 (Mimeograph).

Ginsburg, Th.: Extremwert-Statistik und kalkuliertes Risiko. Ann. Meteorol. (Neue

Folge), No. 5, 155-164, 1971.

GIos, E.: Eine verbesserte Methode der Parameterschatzung fUr die Pearsonverteilung
Typ Ill, WWT 16 (7/8), 238-241, 1966.

Glos, E. and R. Krause: Estimating the accuracy of statistical flood values by means
of long-term discharge records and historical data. IASH/AIHS publ. 84, 144-151,
1967.

Gnedenko, B_. V.: Sur la distribution limite du terme maximum d·une serie aleatoire.
Ann. Moth. I 44, 423-453, 1943.

Golubev, V. S., N. V. Zotimov and N. A. Zykov: Some results of investigation of

liquid precipitation in the Valday Hills (in Russian). [Link] Hydrol. Inst.
123, 5-14, 1965.

Golubev; V. S." Correction 0 f point precipi tation measurement·s {U. S,. S. R. 'experience j.
WMO [Link], 45p., 1979 (at press).

.Goodridge, J. D.: Regional [Link] frequency for drainageilesign in California.

(Abstract) Trans. EOS AGU, 59 (12), 1061-1062, 1975.

Goodrich, R. D.: Straight line plotting of skew frequency data .. Trans. ASCE, 91,
101-118, 1927.

Grant, J. L.: Statistical frequency analysis by optimization. [Link] ty functions to

histograms. Doctoral Dissertation, Georgia Inst. Tech., Atlanto, U.S.• A., October,.
1973.

Greenwood, J. A• .and D. Durand: Aids for fitting the gamma distribution by maximum
likelihood. Technometrics, 2 (1), 55-65, 1960.

Gringorten, I. I.: Extreme value statistics in meteorology - A method of application.

Air Force Surv. Geophys. 125, 1960.

Gringorten, I. I.: A plotting rule for extreme probability paper. J. Geophy. Res.,
68 (3), 813~814, 19630.
 - 18 -

Gringorten, I. I.: Envelopes for ordered observotions applied to meteorological

extremes. J. Geophys. Res., 68 (3), 815-826, 1963b.

Grobe, B.: Niederschlagsgebiete in Abhangigkeit von Niederschlagshohe und Ein-

trittswahrsheinlichkeit. Wasserhaushalt und Bodennutzung, Heft 3, 1-179,
Braunschweig, 1974.

Grobe, B.: Ermittlung von Bemessungsniederschlagen fUr RUckhaltebechen. DVWT,

Fortbilkdungslehrgang fUr Hydrologie: Hochwasserschutz 15-19 Okt. 1979,
Braunschweig, 38-67, 1979.

Grubbs, F. E.: Sample criteria for testing outlying observations. [Link].,

21 (1), 27-58, 1950.

Grubbs, F. E.: Procedures for detecting outlying observations in samples. Tech-

nometrics, 11, 1-21, 1969.

Grubbs, F. E. and G. Beck: Extension of sample Slzes ond percentage points for
significance tests of outlying observations. Technometrics, 14 (4), 847-854, 1972.

Guillot, Mm. P. and D. Duband: La methade du gradex pour le calcul de la probabilite

des crues a partir des pluies. Proc. Internat. Hydrol. Symp., Fort Collins,
Colorado, U.S.A., p. 506-510, 1967. Soc. Hydrotechnique de Fronce, xmes Journees de
l'Hydraulique, Paris, 1968.

Gumbel, E. J.: The return period of flood flows. Ann. Moth. Staitst., 12 (2),
163-190, 1941.

Gumbel, E~ J.: The statistical forecast of floods. The Ohio Water Resour. Board
Bull., 15, Columbus, Ohio, 1949.

Gumbel, E. J.: Statistical theory of droughts. Proc. ASCE 80 (439), 1-19, 19540.

Gumbel, E. J.: Statistical theory of extreme values and some practical applications.
US Not. Bur. Stds. Appl. Moth. Ser.33, 1954b.

Gumbel, E. J.: Statistics of extremes. Columbia University Press, New York, 375 p"
1958.

Gupta, V.: Selection of frequency distribution models. Water Resour. Res., 6 (4),
1193-1198, 1970.

Guttman, 1. and D. E. Smith: Investigatioh of rules far dealing with outliers in

small samples from normal distribution. 1. Estimation of the mean. Technometrics,
11 (3), 527-550, 1969.

Guttman, I. and D. E. Smith: Investigation on rules for dealing with outliers in

small samples from the normal distribution. 11. Estimation of the variance.
Technometrics, 13 (1), 101-111, 1971.
 - 19 -

Guttman, I.: Premium and protection of severol procedures for dealing with outliers
when sample sizes are moderate to large. Technometrics, 15 (2), 385-404, 1973a.

Guttman, I.: Care and handling of univariate ar multiveriate outliers in detecting

spuriosity - a Bayesian approach. Technametrics, 15 (4), 723-738, 1973b.

Haeuser, J.: Kurze starke Regenfalle in Bayern, ihre Ergiebigkeit, Dauer, Inten-
sitat, Haufigkeit und Ausdehnung. Abhdl. Bayer. Landesstelle fUr Gewasserkunde,
MUnchen, 1919.

Hall, W. A. and D. T. Howell: Estimating flood probabilities with specific time

intervals. J. Hydrol., 1 (3), 265-271, 1963.

Handel, P., Th. Leipold and R. -Po Spiegel: Vergleichende Anwendung verschiedener
Verfahren der Hochwasserstatistik an Pegeln im vorolpinen Raum. DGM, 21 (4), 77-82,
1977.

Hansel, N. und U. Schaefer: Die Doppelsummenanalyse. WWT 20 (5), 145-149, 1970a.

Hansel, N. und U. Schaefer: Beschreibung (Typung) und Deutung von Inhomogenitaten

und Inkonsistenzen von hydrologischen Zeitreihen. Wiss. Z. Tech. Univ. Dresden 19
(4), 1039-1046, 1970b.

Hashemi, F.: A statistical analysis of the a"nual, monthly and daily precipitation
of Teheran. Res. Branch, Iranian Meteorolog. Dept., February 1969. Teheran.

Hazen, A.: Discussion on flood flows by W. E. Fuller. Trans. ASCE, 77, p. 628, 1914a.

Hazen, A.: Storage to be provided in impounding reservoirs for municipal water

supply. Trans. ASCE 77, 1539-1559, 1914b.

Hazen, A.: Discussion of probable variations in yearly runoff by L. S. Hall. Trans.

ASCE, 84, 214-224, 1921.

H.E.C.: Hydrologic frequency analysis. Hydrol. Eng. Center, Davis, California, Vol.
3, 1/01 -9102, 1975.

Hershfield, D. M. and W. T. Wilson: Generalizing of rainfall-intensity~frequency

data. IUGG/IASH publ. 43, 499-506, 1958.

Hershfield, D. M. and W. T. Wilson: A comparison of extreme rainfall depths from

tropical and nontropical storms. J. Geophys. Res., 65 (3), 959-982, 1960.

Hershfield, D. M. and M. A. Kohler: An empirical appraisal of the Gumbel extreme-

value procedure. J. Geophy. Res., 65 (6), 1737-1746, 1960.

Hershfield, D. M.: Rainfall frequency atlas of the United States. US Weather Bur.,
Tech. Pap., 40, 1961.
 - 20 -

Hershfield, D. M. and M. A. Kohler: An empirical comparison of the predictive value

of three extreme-value procedures. J. Geophys. Res., 67 (4), 1535-1542, 1962.

Hershfield, D. M.: On the probability of extreme rainfall events. Bull. Amer.

Meteor. Soc., 54 (10), 1013 - 1018, 1973.

Harler, A. und H. R. Rhein: Die Intensitaten der Starkregen in der Schweiz.

Schweiz. Bauzeitung, 5 (32), 559-563, 1961.

Harler, A. und H. R. Rhein: Die Intensitaten der Starkregen in der Schweiz.

Schweiz. Z. Hydrol., 24 (2), 291-352, 1962.

Harler, A.: Die Intensitaten Van 5tarkregen langerer Douer fUr verschiedene
Ortschaften der Schweiz. Gaz-Eaux-Eaux usees 57 (12), 853-860, 1977.

Horton, R. E.: Frequency of recurrence of Hudson River floods. US Weather Bur.

Bull., 2, 109-11, 1913.

Houghton, J. C.: The incomplete means estimation procedure applied to flood fre-
quency analysis, Water Resour. Res., 14 (6), 1111-1115, 1978.

Huff, F. A. and J. C. Neill: Comparison of several methods for rainfall frequency

analysis. J. Geophys. Res., 64 (5), 541-547, 19590.

Huff, F. A. and J. C. Neill: Frequency relations for storm rainfall in Illinois.

Ill. State Water Surv. Bull. 46, 1959b.

Huff, F. A.: Urban effects on storm rainfall in Midwestern United States. IAHS-
AISH publ., 123, 12-19, 1977.

lEA: Australian rainfall and runoff. Inst. Engrs. Austr., Melbourne, 1958.

Jackson, D. R. and G. Aron: Parameter estimation in hydrology: The state of the art.
Water Resour. Bull., 7 (3), 457-472, 1971.

Jenkinson, A. F.: The frequency distribution of the annual maximum (or minimum)
values of meteorological elements. Quart. J. Roy. Meteor. Soc., 81, 158-171, 1955.

Jenkinson, A. F.: Statistics of extemes. In: Estimation of maximum floods. WMO

Tech. Note 98, 183-228, 1969.

Jenkinson, A. F.: Some quasi-periodic changes in rainfall in Africa and Europe. In:
Proc. WMO/IAMAP, WMO publ. 421, 503 p., 1975.

Jenkinson, A. F.: The analysis of meteorological and other geophysical extremes.

Meteorol. Off. 13, Bracknell, Mem. No. 58, 1977.

Jennings, A. H.: Maximum station precipitation for 1, 2, 3, 6, 12 and 24 hours,

Part Ill. US. Dep. Commerce, Weather Bur. Tech. Pap., 15, Washington, 1952.
 - 21 -

Jevons, W. S.: On the deficiency of rain in an elevated rain-gauge as caused by

wind. London, Edinburgh& Oublin Phil Mag., Ser. 4, (22), 421-433, 1861.

Jowitt, P.K.: The extreme-value type I distribution and the principle of maximum
entropy. J. Hydrol., 42 (1), 23-38, 1979.

Kaczmarek, Z.: Efficiency of the estimation of floods with a given return period.
lASH publ. 45, 145-159, 1957.

Kaller, J.: Die Sicherheit bei der Extrapolation von hydrologischen Ereignissen mit
geringer Auftretenswahrscheinlichkeit bei Anwendung der Gumbel-Verteilung. DGM,
18 (1), 10-13, 1974.

Kaller, J. and W. Riebe: Vertrauensbereiche in hydrologischen Haufigkeitsonalysen.

Deutsche Gewasserkund. Mitt. (DGM) , 23(4), 85-90, 1979.

Kappus, U., J. M. Bleek and S. H. Blair: Rainfall frequencies for the Persian Gulf
coast of Iran. Hydrol. Sci. Bull., 23 (1), 119-129, 1978.

Keeping, E. S.: Distribution-free methods in statistics. Statistical Methods in

Hydrology, Proc. Hydrol. Symp. No. 5, McGill Univ., Canada, 1966.

Kelway, P.S.: The rainfall recorder problem. J. Hydrol., 26 (1), 55-77, 1975.

Kimball, B. F.: Sufficient statistical estimation functions for the parameters of

the distribution of maximum values. Ann. Math. Statist., 17 (3), 299-309, 1946.

Kimball, B. F.: An approximation to the sampling variation of an estimated maximum

value of a given frequency based on fit of double exponential distribution of
maximum values. Ann. Math. Statist., 20 (1), 110-113, 1949.

Kimball, B. F.: The bias in certain estimates of the parameters of the extreme-value
distribution. Ann. Math. Statist., 27, 758-767, 1956.

Kimball, B. F.: On the choice of plotting positions on probability paper. J. Amer.

Statist. Assoc., 55 (291), 546-560, 1960.

Kirsten, M.: Ein neues Verfahren fUr die Bestimmung der wirtschaftlichen Ausbou-
grosse und der WUrdigkeit von Hochwasserschutzmassnahmen. Mitt. Inst. f.
Wasserwirtschaft 17, 1964.

Kite, G. W.: Confidence limits for design events. Water Resour. Res., 11 (1),
48-53, 1975.

Koberg, D. and H. Eggers: Some aspects for the selection of an adequate probability
distribution function for flood analysis. Intermat. Symp. River. Mech. Proc. IAHR,
Vol. 2, 229-239, Bangkok, 1973.

Kohler, M. A.: On the use of double-mass analysis for testing the consistency of
meteorological records and for making the required adjustments. Bull. Amer'. Meteorol.
Soc., VoL 30, 188-189, 1949.
 - 22 -

Kohler, M. A. ond R. K. Linsley: Recent developments in woter-supply fore costing

from precipitation. Trons. AGU, 30, 427-436, 1949.

Koopman, B.O.: On distribution admitting a sufficient statistic. Trans. Ann. Math.

Soc., Vol. 39, 399-409, 1936.

Kotz, S. and J. Neumann: On the distribution of precipitation amounts for periods

of increasing length. J. Geophys. Res., 68 (12), 3635-3640, 1963. .

Krishnan, A. and R. S. Kushwaha: Comparison of Gumbel and Jenkinson's methods of

estimating probable maximum rainfall intensities over India for different return
periods. Centr. Board Irrig. Power J., 30 (2), 223-234, 1975.

Kryukov, V. F.: Theoretical justification on Pears on distribution curves. Soviet

Hydrol., Selected Papers., (1), 71-76, 1973.

Kuksin, 1. E.: Comparison of computation methods of daily maximum of precipitation

(in Russian). Meteorol. Hydrol. (Moscow), 3, 28-31, 1965.

Kumaraswamy, R.: Sinepower probability density function. J. Hydrol., 31 (1/2),

181-183, 1976.

KuznetsQVC, L.P.: The use of cumulative probability curves in processing daily

maximum precipitation (in Russian). Trans. Voyeykov Main Geophys. Observ., 162,
7-21, 1964. .

Lambor, J.: Frequency of intense rainfall for the territory of Poland. J. Hydrol.,
5 (2), 158-162, 1967.

Landwehr, J. M., N. C. Matalas and J. R. Wallis: Some comparisons of flood statistics

in real space and log space. (at press).

Langbein, W. B.: Annual floods and the partial-durdtion flood series. Trans. AGU,
30, 879-881, 1949.

Lauruschkat, B.: Extreme Punktniederschlage und ihre zeitliche Verteilung. 5.

Fortbildungslehrgang fUr Hydrologie. DVWW, Barsinghausen, 1973 (Germany).

Lauterbach, D.: Betrachtungen zur Wahl von Verteilungsfunktionen fUr die Berechnung
von HochwasserdurchflUssen, WWT 18 (1), 10-15, 1968.

Lawless, J. F.: Confidence interval estimation for the parameters of the Weibull
distribution. Utilitas Math., 2, 71-87, 1972.

Lawless, J. F.: Approximations to confidence intervals for parameters in the extreme

value and Weibull distributions. Bionometrika, 61, 123-129, 1974.

Lawless, J. F.: Confidence interval estimation for the Weibull and extreme value
distributions (with discussion). Technometrics, 20, 355-369, 1978.
 - 23 -

Lecher, K.~ Bestimmung der Regenintensitaten vcn Kurzregen verschiedener Haufigkeit

in Chile. Mitt. Inst. Wasserwirtsch., Hydrol. und Landwirtschaftlicher Wasserbau,
Tech. Univ. Hannover, No. 12, 355-366, 1968.

Le Due, S. K. and D. G. Stevens: Maximum likelihood estimates for parameters of the

mth extreme value distribution. J. Appl. Meteor., 16 (3), 251-254, 1977.

Leese, M. N.: Use of censored data in the estimation of Gumbel distribution para-
meters for annual maximum flood series. Water Resour, Res., 9 (6), 1534-1542, 1973.

Lewis, W. A. and L. H. Watkins: Investigation into accuracy of four types of rain-

fall recorders. Quart. J. Roy. Meteor. Soc., 81 (349), 449-458, 1955.

Lieblein, J.: A new method of analyzing extreme-value data. Natl. Advs. Comm. Aero.,
Tech. Note 3053, 88 pp., Washington, 1954.

Liebscher, H.: Hochwasserwahrscheinlichkeit. DGM, 15 (4), 103-110, 1970.

Light, P. and M. A. Kohler: Forecasting seasonal runoff by statistical methods.

Truns. AGU, 24 (2), 719-736, 1943.

Linsley, R. K.,M. A. Kohler and J. L. H. Paulhus: Hydrology for engineers.

Mc Grow-Hill Book Comp. New York, 1958.

Lowery, M. D. and J. E. Nosh: A comparison of methods of fitting the double exponen-

tial distribution. J. Hydrol., 10 (3), 259-275, 1970.

Ludwig, K.: Zusammengesetzte Haufigkeisverteilungen. DGM, 23 (3), 57-61, 1979.

Lyutik, P. M.: Calculation of precipitation on the territory of Eastern Carpathian

Mountains in connection with computation of maximum flood discharges (in Russian).
State Hydrol. Inst., Collection of Papers on Hydrology No. 6, 12-31, 1966.

Majumdar, K. C. and R. P. Sawhney: Estimates of extreme values by different distri-

bution functions. Water Resour. Res., 1 (3), 429-434, 1965.

Maniak, V. und B. Grobe: Verteilungsfunktionen fUr Hochwasserhaufigkeiten und

ihre £ignung. Leichtweiss. Inst. Wasserbau, Tech. Univ. Braunschweig, Mitt. 35,
1-19, 1972.

Mann, H. B. und D. R. Whitney: On the test of weather one of two random variables
is stochastically larger than the other. Ann. Moth. Statist., 18, 50-60, 1947.

Mann, N. R.: Point and interval estimation procedures for the two-parameter Weibull
and extreme-value distributions. Technometrics, 10, 231-256, 1968.

Markovic, R. D.: Probability functions of best fit to distributions of annual pre-

cipitation and runoff. Colorado State Univ., Fort Collins, Hydrol. Pap. 8, 33 p.,
1965.
 - 24 -

Markavic, R. D.: Discriminating the change in means of hydrologic variables. Proc.

Internat. Hydrol. Symp., Fort Collins, Colorado, U.S.A., 581-588, 1967.

Martinec, J.: Expected snow loads on structures from incamplete hydrological data.
J. Glaciolagy, 19 (81), 185-195, 1977.

Massey jr., F. J.: The Kolmogorav-Smirnov test for goadness of fit. J. Amer.
Statist. Assoc., 46, 1951.

Matalas, N. C. and J. R. Wallis: Eureka~ It fits a Pearson type 3 distribution.

Water Resour. Res., 9 (2), 281-289, 1973.

Matthews, L. S~: A cause of error in self~recording rain-gauges. Meteorol. Mag.,

72 (864), 283-287, 1938.

Mc Andrew, D. W.: Snow load analysis and recreational uses of snow data. In:
Santeford. H. S. and J. L. Smith, comp. Advanced concepts and techniques in the
study of snow· and ice resources. Nat. Acad. Sci., Washington D. C., 11-21, 1973.

Mc Illwrainth, J. F.: Rainfall intensity-frequency data for New South Wales station.
J. Inst. Engrs. (Australia), 25 (7/8), 133-139, 1953.

Merriam, C. F.: A comprehensive study of the rainfall on the Susquehanna Valley.

Trans. AGU, Vol. 18, 471-476, 1937.

Miller, J. F.: Physiographically adjusted precipitation-frequency maps: Distribution

of precipitation in mountainous areas. WMO-No. 326 (11), 264-277, 1972.

Miller, J. F., R. H. Frederick and R. J. Tracey: Precipitation-frequency atlas of

the western United States. Vol. I-Vol. XI., NOAA Atlas 2, National Weather Service,
US Dept. of Commerce, Silver Spring. Md., 1973.

Miller, J. F.: Annotated bibliography of NOAA publications of hydrometeorological

interest. NOAA Techn. Memo NWS Hydro-34, Silver Spring, Md., 65 p.. 1977.

Mirotvorskoya, N. K.: Parameters of heavy rains in the south of the Ukraine (in
Russian). Meteorologiya, Klimatologiya i Gidrologiya, No. 5, 35-41, 1969.

Modley, D. A., and H. L. Crutcher: An application of the gamma distribution function

to Indian rainfall. Environmental Data Service, US Dept. of Commerce, Tech. Rep.
EDS, 5, 47 p.. Silver Spring, 1968.

Moron, P. A. P.: The statistical treatment of flood flows. Trans. AGU, 38 (4),
519-523, 1957.

Mosonyi, E.~: Empfehlung fUr die Berechnung der Hochwasserwahrscheinlichkeit.

Wasserwirtschaft, 64 (1), 20-23, 1974.

Mosonyi, E. F., H. Eggers and D. Koberg: Influence of probability analysis on

selection of design flood. 11. World Congress of IWRA, New Delhi, 1975.
 - 25 -

Munster, Ch.: Untersuchungen zur Extemwertstatistik van Niederschltlgen. Dipl.

Arbeit. Univ. Bann, Metearol. Inst., 72 p., 1970 Bonn (West Germany), (mimeograph).

Musy, A.: Repartition spatiale et prevision en temps de retour de facteurs climato-

logiques en vue dlun dimensionnement des ouvrages d'amenagement des eaux. Bull.
Techn. Suisse, Rom. 99 (15), 319-339, 1973.

Nosh, J. E. and J. Amorocho: The accuracy of the prediction of floods of high return
period. Water Resaur. Res., 2 (2), 191-198, 1966.

Nemec, J.: A contribution to the design of recording rain gouge network. IASH/AISH
Bull., 10 (2), 70-73, 1965.

Nerc: Flood studies report. Not. Env. Res. Council, Meteorological Office, London,
1975.

Obasi, G. O. P. and J. K. Nimira: Statistical estimation of the maximum probable

doily rainfall in Kenya using the method of maximum likelihood. Proc. Symp. on
Flood Hydrology, Nairobi 1975. TRRL Supplementary Report 259, 71-94, Crowthorne,
Berkshire, 1977.

O'Mahony, G.: Rainfall and moon phose. Quart. J. Roy. Meteorol. Soc., 91, 196-208,
1965.

Otten, A. and M. A. J. Van Montfort: The power of two tests on the type of distribu-
tions of extremes. J. Hydrol., 37 (4), 195-199, 1978.

Otten, A. and M. A. J. Van Montfort: Maximum likelihood estimation of the general

extreme-value distribution parameters. J. Hydrol., 47 (1/2), 187-192, 1980.

Panchang, G. M. and V. P. Aggarwal: Peak flow estimation by method of maximum

likelihood. Tech. Memo. HL02, India, Central Water and Power Res., Poona, India,
1962.

Panchang, G. M.: Realistic projections from extreme values data. Proe. Internet.
Hydrol. Symp., Fort Collins,Colorado, U.S.A., 546-555, 1967.

Parson, D. A.: Calibration of 0 weather bureau tipping-bucket rain gouge. Month

Weather Rev., 69 (7), 205, 1941.

Pearson, K.: Asymmetrical frequency curves. Phil. Trans. Roy. Soc. London, 186,
1895.

Pearson, K.: Tables for statisticians and biometricians. Part I. The Biometric
Laboratory, Univ. ColI., London, Cambridge Univ. Press, London, 3rd ed., 1930.

Pearson, E. S.: The probability integral transformation for testing goodness of fit
and combining independent tests of significance. Biometrica, 33, 134-148, 1938.

Pierce, B.: Criterion for the rejection of doubtful observations. Astronomical J.,
2, 1852.
 - 26 -

Plate, E. J.: Extreme value models. Inputs for risk analysis in water resources.
Water Resources Publications, Fort Collins, Colorado, 1979, 24 p.

Potter, W. D.: Simplifications of the Gumbel method for computing probability

curves. US Dept. Agr. Soil Conserv. Serv., SCS-TP-78, 1949.

Powell, R. W.: A simple method of estimating flood frequencies. Civil Eng. 13,
105-106, 1943.

Rao, D. V.: Log Pearson type 3 distribution: A generalized evaluation.

J. Hydraulics Div., ASCE, 106 (HY 5), 198000

Roo, D. V.: Log Pearson type 3 distribution: Method of mixed moments.

J. Hydraulics Div., ASCE, 106 (HY 6), 999-1019, 19801:>.

Reich, B. M.: Short-duration rainfall-intensity estimates and other design aids for
regions of sparce data. J. Hydrol. 1 (1 }.3-28, 1963.

Reich, B. M.: Flood series compared to rainfall extremes. Water Resour. Res. 6
(6), 1655-1667, 1970.

Remenieras, G.: Hydrologie de l'ingenieur. Eyrolles, 1965.

Renard, K. G. and H. B. Osborn: Rainfall intensity comparisons from adjacent 6-hour

and 24-hour recording rain gouges. Water Resour. Res. 2 (1), 145-146, 1966.

Richardson, B. F. C. and D. C. Midgley: Analysis of SWA-Namibia rainfall data.

University of the Witwasersrand, Johannesburg Hydrological Research Unit, Report No.
3/79, 67 p., 1979.

Rima, A.: Legge dell'intensita della pioggia per la regione di Bellinzona. Riv. Tee.
Svizzera ital. 7, 254-256, 1964.

Robertson, N. G.: The frequency of high intensity rainfalls in New Zealand.

New Zealand Meteorol. Ser., Mise. Publ. 118, 53 p., 1963.

Rodda, J.: A country-wide study of intense rainfall for the United Kingdom.
J. Hydrol., 5 (1), 58-69, 1967 ..

Rodda, J. C.: Rainfall excesses in the United Kingdom. Trans. Inst. British
Geograph., 49, 49-60, 1970.

Rodda, J. C.: The precipitation measurement paradox-the instrument accuracy problem.

Reports on WMO/IHD Projects No. 16, WMO-No. 316, 42 p., Geneva, 1971.

Rosbjerg, D.: Return periods of hydrological events. Nord. Hydrol., 8 (1), 57-61,
1977.
 - 27 -

Rozhdestvensky, A. V. and A. I. Chebotarev: Stotistical methods in hydrology

(in Russian). Gidrometeoizdat, 424 pp., Leningrad, 1974.
,J
Sama j, F. and ·S. Valovi~: Long-term precipitation averages for the period 1901-1970
in Slovakia (in Slovak). ColI. Pap. Hydrometeorological Institute in 8ratislava,
No. 14, 413 pp., 1978.

Samuelson, B.: Statistical interpretotion of hydrometeorologicol extreme values.

Nord. Hydrol., 3 (4), 199-213, 1972.

Sangal, B. P. and A. K. Biswas: The 3-parameter lognormal distribution and its

applications in hydrology. Water Resour. Res., 6 (2), 505-515, 1970.

Searcy, J. K. ond Ch. Hardison: Double-mass curves. Woter Supply Paper 1541 B,
p. 66, 1960.

Sevruk, B.: Correction of point precipitation measurement. In: Vaw-Eth Mitt., 41,
267-279, 1979, ZUrich.

Sholigram, W. M. ond V. S. Lele: Analysis of hydrologic data using Pearson Type 11

distribution. Nord. Hydrol., 9 (1), 31-42, 1978.

Shane, R. M. and W. R. Lynn: Mathematical model for flood risk evaluotion.

J. Hydroul. Div. ASCE, 90, 1-20, 1964.

Siegerstetter, L. A.: Verwendung parameterfreier Wahrscheinlichkeitsverteilungen

zur Beschreibung hydrologischer Prozesse. Inst. fUr Hydraulik und Gewasserkunde,
Tech. Univ. MUnchen, Mitt. Nr. 11, 104 pp., MUnchen 1973.

Singh, R.: Double-moss analysis on the computer. J. Hydraul. Div. ASCE, 139-141,
1969.

Slakter I M. J.: A comparison of the Pearson chi--square and Kolmogorov goodness

of fit tests with respect to validity. J. Amer. Statist. Assoc., 60, 1965.

Smirnovo, E. A.: . Design roinfall characteristics of the U.S.S.R, territory. Symp.

of floods and their computation, IASH/UNESCO/WMO, Leningrod, lASH publ., 85,
105-114, 1969.

Smirnow N. W., Dunin and I. W. Barkowski:

t Mathematische Statistik in der
Technik. VEB, Deutscher Verlag der Wissenschaften, 1969 (Germon Democratic
Republic.

Smith, R. E. and H. A. Schreiber: Point processes of seasonal thunderstorm rainfall,

2. Rainfall depth probabilities. Water Resour. Res;, 10 (3), 418-423, 1974.

Sneyers, R.: Du test de validite d'un ajustement base sur les fonctions de l'ordre
des observotions. Inst. Roy. Meteorol. Belg., Publ. B, 39, 1963.

Sneyers, R.: Sur I'analyse statistique des series d'observations. WMO Techn.
Note No. 143, IB9 pp., Genevo, 1975.
 - 28 -

Sneyers, R.: L'intensite maxima le des precipitations en Belgique. Publ. ser. 8,

No. 86, 15 pp., Bruxelles, 1977.

Sneyers, R.: L'influence du relief sur les valeurs extremes de l'intensite et la

duree des precipitations en Belgique. [Link]. Tag [Link]. Meteorologie, Grindelwald
19-23 Sept. 1978, 281-283.

Snyder, W. M.: Summary and evaluation of methods for detecting hydrologic changes.
J. Hydrol., 12 (4), 311-338, 1971.

Snyder, W. M.: Fitting of distribution functions by nonlinear least squares. Water

Resour. Res., 8 (6), 1423-1432, 1972.

Snyder, W. M. ond J. R. Waloce: Fitting a three-parameter lognormol distribution by

least squores. Nord. Hydrol., 5 (3), 129-145, 1974.

Snyder, W. M.: Continuous seasonal rainfall probability of extreme rainfall events.

Hydrol. Sci, Bull., 20(2), 275-283, 1975.

Stacy, E. W.: A generalization of the gamma distribution. Ann. Math. Statist., 3,

1187-1192, 1962.

Stahlmann, V., H. D. Schoss and L. Chilla: Haufigkeit von Starkregen im Emscher-

und Lippegebiet. Eigenverlag der Emschergenossenschoft, Essen, 77 pp., 1974.

Stidd, C. K.: Estimating the precipitation climate. Water Resour. Res., 9 (5),
1235-1241, 1973.

Stol, Ph. Th.: On the decomposition of the extreme value distribution of daily
rainfall depths and the derivation of probabilities of compount events. J. Hydrol.,
14 (3/4), 181-196, 1971.

Strommen, N. D. and J. E. Horsfield: Monthly precipitation probabilities by climatic

divisions - 23 Eastern States. US Dept. Agr. Commerce, Mise. Publ. No. 1160, 141 pp.,
Washington D. C., 1969.

Struzer, L. R. and V. G. Golubev: Methods for the correction of the measured sums
of precipitation for water balance computations. Report for the Unesco/WMO
International Workshop on Water Balance of Europe. Bulgaria, lladni Piassadsi
27 Sept. - 2 Oct., 1976.

Szigy6rt6, Z.: A three parameters' gamma distribution function for hydrological

investigations. Proc. Internat. Hydrol. Symp., Fort Collin~ Colorado, U.S.A.,
490-495, 1967.

Taesombut, V. and V. Vevievich: Use of partial flood series for estimating distribu-
tion of maximum annual flood peak. Colorado State Univ., Fort Collins, Colorado,
Hydro!. Pap. 97. 71 p" 1978.

Terry, M. E.: Some rank order tests which are most powerful against specific para-
metric alternatives. Ann. Math. Statist., 23, 346-366, 1952.
 - 29 -

Thiess, N. G., G. Ganz und H. H. Johannsen: Verteilungskurven der Niederschltlge

in Baden-WUrttemberg. Landesanstalt fUr Umweltschutz, Baden-WUrttemberg, Deutscher
Wetterdienst, Karlsruhe und Offenbach/M., 1976.

Thiess, N.: Die Ermittlung von Hochwosserwahrscheinlichkeiten, insbesonders nach

dem Tangentenverfahren. Landesanstalt fUr Wasser-und Abfallwirtschaft, Baden-
WUrttemberg, Karlsruhe, 218 p., Mai 1980 (Vervielftlltigung).

Thiess, N.: Ermittlung aptimaler Verteilungen und Parameterschatzverfahren der

mathematischen Statistik und ihre Anwendung in der Hydrologie. Doctoral Dissertation,
Tech. Univ. Karlsruhe, 127 p., Karlsruhe, 1975.

Thom, H. C. S.: A note on the gamma distribution. Mon. Weather Rev., 86 (4),
117, 121, 1958.

Thom, H. C. S.: Some methods of climatological analysis. WMO Tech. Note, No. 81,
53 pp., Geneva, 1966a.

Thorn, H. C. S.: Distribution of maximum annual water equivalent of snow on the

ground. Mon. Weather Rev., 94, 265-271, 1966b.

Tietien , G. L. et al.: Testing for a single Qutlier in simple linear regression.

Technometrics, ~), 717-721, 1973.

Tippett, L. H. C.: On the extreme individuals and the range of samples taken from
a normal population. Biometrica, 17, 364, 1925.

Todorovic 1 P. and E. Zelenhasic: The extreme values of precipitation phenomena.

Bull. lASH, 13 (4), 7-24, 1968.

Todorovic, P. and E. Helenhasic: A stochastic model for flood analysis. Water

Resour. Res., 6 (6), 1641-1648, 1970.

Todorovic, P. and J. Rouselle: Some problems of flood analysis •. Water Resour.

Res., 7 (5), 1144-1150, 1971.

Todorovic, P. and D. [Link]: A stochastic model of [Link].

J. Appl. Meteor., 14 (1), 17-24, 1975.

Tomlinson, A. I.: Storm rainfall, J. Hydrol. (N. Z.), 17 (2), 57-77, 1978. (Outliers).

Tripathi, N.: Area-depth relations for frequency values of roinfall. Indian J.

Meteor. Hydrol. Geophys., 27 (2), 173-176, 1976.

Tukey, J. W.: The future of data analysis. Ann. Moth. Statisti., 33 (1), 1-67, 1962.

Turner, A. K.: Storm intensities versus seasonal rainfalls. J. Hydro1., 12 (4),

377-386, 1971.
 - 30 -
Underhill, H. W.: Rainfall recorders - a comparison of different types. Bull.
into Ass. scient. Hydrol., 11 (3), 50-55, 1966.

Uppalo, S.: Extreme distribution functions for daily and monthly precipitation
in Finland •. Geophysica, 15 (1), 1978.

US Weather Bureau: Frequency of maximum water equivalent of March snow cover in

narth central United States. US Weather Bureau, Tech. Pap. No. 50, 1-24, 1964.

Vatnsdal, J. R.: Minimal variance and its relation to efficient moment-test. Ann.
Math. Statist., vol. 17, 198-207, 1946.

Van Mantfort, M. A. J.: On testing that the distribution of extremes is of Type I

when Type 11 is the alternative. J. Hydrol., 11 (4), 421-427, 1970.

Van Montfart, M. A•. J.: An asymmetric test on the type of the distribution of
extremes. Mede. Landbouwhogesch., Wageningen, 73, 1-16, 1973.

Vinogradov, U. B.: Method of computing rainfall floods characteristic. AIHS publ.

84, 312-320, 1969.

Van Mises, R.: La distribution de la plus grande de n valeurs. Revue Moth. l'Union
Interbalcanique (Athens), voL 1, 1-20, 1936.

Wald, A. and J. Wolfowitz: An exact test for randomness in the nonparametri~ case
based on serial correlation. Ann. Moth. Statist., 14, 278-388, 1943.

Wallis, W. A. and G. H. Maore: A significance test for time-series analysis.

J. Amer. Statist. Assoc., 36, 401, 1941.

Wallis, J.R. and N. C. Matalas: Correlogram analysis revisited. Water Resour.

Res., 7 (6), 1448-1459, 1971.

Wan, P.: Paint rainfall characteristics of Saudi Arabia. Proc. Inst. Div. Engrs.,
Part 2 (62), 179-187, 1976.

Weibull, W.: A statistical theory of the strength of materials. Ing. Vetenskops

Akad .. HandL (Stockholm). VoL 151, p. 15, 1939.

Weingartner, S.: Berechnung der Hochwasserwahrscheinlichkeit. 6. Fortbildungs-

lehrgang fUr Hydrologie, DVWW, Bad Herrenalb, 1974 (Germany).

Weingtirtner, W.: Extremverteilung Starkregen-Hochwosser. Hydrologie Fortbildungskreis

Wiener Mitt., Wasser-Abwasser-Gewtisser, 1975. Inst. fUr Wasserwirtschaft, Hochschule
fUr Bodenkultur, Band 18, Wien, 1975.

Weisner, C. T.: Hydrometeorology. Chappman and Hall, Ltd., London, 232 p., 1970.

Weiss, L. L. J. and W. T. Wilson: Evaluation of significance of slope changes in

double-mass curves. Trans. AGU, vol. 34, 393-896, 1953.
 - 31 -

Weiss, L. L. J.: Ratia of true to fixed-interval moximum rainfall. J. Hydraul.

Div. ASCE, 90 (1), 77-82, 1964.

Widmoser, P.: Das stochastische und statistische Risiko bei der Wahl des Bemessuns-
wertes aus Extremwertbeobachtungen. DGM, 21 (2), 32-38, 1977.

WMO: Annotated bibliography on precipitation measurement [Link]/IHD

Project Report No. 17. WMO-No. 343, 278 p., Geneva, 1973.

WMO: Guide to hydrological practices; Ch.5, Hydrological Analysis. WMO-No. 168,

3rd Edition, 5.17-5.36, Geneva, 1974.

World water balance and water resources of the Earth (in Russian). Gidrometeoizdot,
637 p., Leningrad, 1974.

Wu, B. and J. D. Goodridge: On the selection of probability distributions for

hydrologic frequency analysis. Paper presented at AGU, 1974, Fall Annual Meeting.

Wurtele, M. G. and J. M. Roe: Fisher-Tippett representations of extreme rainfall

and temperature data. (Abstract) Trans. EOS, AGU, 59 (12), 1063, 1978.

YarneII, D. L.: Rainfall intensity-frequency data. US Dept. Agr. Mise. Publ. 204,
9-23, 1935.

Yevievich, V. and R. J. Jeng: Effects of inconsistency and non-homogeneity on

hydrologic time series. Proc. Internat. Hydrol. Symp., Colorado State Univ., Fort
Collins, Vol. 1, 451-458, 1967.

ZeIler, J.: Die Bedeutung der FrUhjahrschneeschmelze fUr den extremen Hoch-
wasserabfluss in Wildbachgebieten, Eidg. Anst. Farstl. Versuchswesen, Birmensdorft,
Switzerland, Mitt. 51 (1), 171-188, 1975.

Zeller, J., H. Geiger and G. Roethlisberger: Starkniederschlage des schweizerischen

Alpen- und Alpenrandgebietes. Teil 1-3, Eidg. Anstalt fUr das Forstliche
Versuchswesen, Birmensdor~ Schweiz, 1976_1978.
 - 32 -

R ETU R N PER I o0 I N YEA R S

10 1 10 2
l~
10- 3 10- 2 10- 1 1
•
---------------
Vl
Time series modelling
c
z
0
::l:
---
-----------------
-- - ---
0
~

f--
«
...
f--

::I:
Frequency analysis
of partial duration
u
~

--'
0-
0-
series «
--- - ------ "-
-- - -- ...
0

Frequency analysis '"

z
of annual maxima «
series ""
- - --
EXTREMES OF PRECIPITATION
f-- Data on a clock-hour Data composed of all Oata composed of the Vl
z
=> or observational-day values that exceed a maximum value for '"
~

0-
basis
/
given base stage each year of record f--
~

:z: z
~

~ "-
LU

'"

Vl
U
~

f-- f--
=> (;> >,
0- ., .,
0=
U
.,
0=
Vl

-'"
~
:>
:> :>

'-
f--
.,er .,er
~
=> >
.... .... ...
<Y

IIII
I- I-
0 f--
11

"
u
time value value «
""0«

/ increasing number and interdependence of data

increasing accuracy and costs of analysis
::l:
u

Figure 1 - Range of application. definitions and characteristics of three

possible methods of analysis for extremes of precipitation
 - 33 -

Oistribution Distribution
of of
MAGNlTUDES EVENTS
IV

one for each season one for each season

III

one for the year 1I one for the year

I constant number of events

Figure 2 Classification of models I - IV using partial duraTion series with some

additianal references to NERC (1975) (madel I - widely used;
madel 11 - Borgman (1963); Shone and Lynn (1964); Bernier (1967);
Todorovic and Zelenhasic (1976); model III - Borgman (1963);
Todorovic and Zelenhasic (1970); model IV - Todorovic and Rouselle
(1971)
 - 34 -

By Jenkinson and Gumbel Methods

[Link] belw8fn e~t,..m. Ycluu [Link] yaart ,..tu~n)
·30

·25

0
•,
I Type H
c + 5 Type Jl1
j 0
0·82 0·98 1-02 1·06 1'10 1-14 H8 1-22

•• 5 •• • Vo'' ' of f.
::-
~
-'0 • : •• !
•• •
.. :
• ft
c - 15 • • • •
:! • •
.!; -20
.!;
•
•uc
-25
I• • • •
•
~
~

0
•• -30

- 35 • '.
• .
-40 •
-45 •
-50 - Regression line with e~uotion

-55 Y=172·48 -189·05 X IR=-O'SOI

T)'~ I ,Hand IlJ refal" to inten$ity types
-60

Figure 3 Differences between lOO-year daily precipitation maxima estimates by

Jenkinson (1955) ond Gumbel (1954b) methods accarding to Krishnan and
Kushwaha (1975). Types I, 11 and III refer to distribution types
defined by Jenkinson (1955) which are not identical with I, 11 and III
extremal types (Table I V). Paints = annual maxima series fram different
precipi tetion stations; er 1 = the standard deviation of the annual
maxima; u2 the standard deviation of the greater members in pairs of
annual maxima
200 -
P
{mm/
1.75

150

125
V
1/ ... vv
100
vV V
vty r I

/1/1, ...... VVl-- • W

'"
I
75
£~~
VJ";?'V MI 7 ' 7 ,r
50
,A~ O.l I .. " /'
~&~
25 [Link] :> ' ..... ;'

0.. J 75.......
o I
I I 11, I '" I I
0.0 t i I I i i I i I I I r n
1001 10 I 2033 10 50 100 T 500 2.00 ",Ca to £/) m 40 la U 10 1r -1~

Figure 4 - Frequency diagrdm of annual daily maximum Figure 5 - Critical values of r for various
precipitation (p) for Sever 1901-1970; sample sizes n and confidence levels
T - return period in years of 90 and 95 per cent (Van Montfort,
1970)
 1 11

,I~ I rt=lU I I! J-l

H-EEI
;-t-~-!~~!-
,-~l
i
I
Cl I n::r:tO rnl ,

.,
11
I. i
II
I

I I I' I' ,iI

I

!±I=="_.
1 !
I!
. _.121' , I I ,w

R
I I . i ' '"
!
lj
el _.11

I
I . I rTn I
I
,

l'
I
I
H '

I i ,1
r---t-t-rl
r ,",
1-;-1
1
'

, I I I r1
+--i_Li
/ I / !

I
I I I II 1111 I I I Ill: I: II! 11 I, II I!
! I 1I
i j

1.01 1.1 2.33 10 50 100 T 5001.01 1.1 2.33 10 50 100 T 500

Figure 6 - Probobility poper for I ond 11 extremol distributio") T = return period in yeors
 - 37 -

TABLE I

History of frequency onalysis of floods

and precipitation

Year Items* Re f e r e nee

1852 S Criterion for the rejection of an Peirce (1852)

Qutlier

1875 S Studies on the log-normal distribution Galton (1875)

1880- G Herschel and Freeman applied flow- Foster (1934)

1890 duration curves to flood studies

1895 5 Derivation of a series of distributions Pearson (1895, 1930)

to fit any distribution

1896 F G.W. Rafter suggests the use of Fuller (1914)

probability methods in runoff studies

1913 F Discussion of the application of Horton (1913)

Gaussian law in flood studies

1914 F The agreement with the normal law of Hazen (1914a)

errors is closer if the logarithm of
annual floods is used

1914 Comprehensive study of statistical Fuller (1914)

methods applied to floods. First flood-
frequency formula. Semi-log paper
introduced to plot flood flows

1914 G Suggestion of the graphical Hazen (l914b)

linearizotion of the normal distribu-
tion. The problem of plotting position
is discussed.

1918 EP,G Frequency distribution curves of pre- Thiess (1976)

cipitation probably first constructed
by the Italian Montanari for the
station Marozzo in Ferrare district

1919 EP A study on heavy rainfall intensity Hauser (1919)

and frequency

* EP = extremes of precipitation; G = graphical methods; F = flood;

5 = statistics, general
 - 38 -
TABLE I cont.
History of frequency analysis of floods
and precipitation

Year Items Re f e r e nee

1921 G The use of log-normal probobility paper Hazen (1921)

is proposed

1923 S,F The N/m formula for plotting position Anonymous (1923)
applied to flood studies

1923 S The lorgest value for other than the Dodd (1923)
normal distribution first studied

1924 F Peorson type I ond III curves put in a Fos ter (1924)
form convenient for use in flood
studies

1925 S Calculation of the probabilities of the Tippett (1925)

largest value taken from the normal
initial distribution for different somple
sizes

1927 S Extremal distribution type 11 (second Frechet (1927)

asymptotic) sometimes known as
Fisher-Tippett 11

1928 S All three extremal distributions studied Fisher and Tippett

(Type 11 independently of Frechet, 1927). (1928)
Type I distribution is sometimes known as
Gumbel distribution, or double exponential
distribution since Gumbel (1941) first
applied it to flood-frequency analysis.
Type III is known as Weibull (1939) distri-
bution since he first applied it for
practical purposes.

1927 G Introduction of probability paper, which Goodrich (1927)

was also used later, in modified form,
for extreme-value distribution by Fisher
and Tippett, Weibull, and Gumbel

1935 EP Extensive rainfall frequency analysis in Yarnell (1935)

U.S.A. (updated later by Hershfield, 1961)

1936 S Further contributions by classifying the Von Mises (1936)

initial distributions (Types I, 11, Ill)
 - 39 -
TABLE I cont.
History of frequency analysis af floods
and precipitation

Yeor Items R e f e r e nee

1939 S Type III extremal distribution opplied Weibull (1939)

for proctical purposes

1941 F,S Application of the Frechet (Fisher- Gumbel (1941)

Tippett) theory of extreme values to
flood data. Method of moments pro-
posed to fit Type I extremal
distribution

1943 S Systematic ~tudy of extremal distri- Gnedenko (1943)

butions. lIOnly extreme-value distri-
butions obey the stability postulate,
which an extreme-value distribution
has to doll

1943 G Extremal probability paper for graphical Powell (1943)

application of Gumbel's method (Type I)
(1941)

1946 S Maximum-likelihood method suggested for Kimball (1946)

fitting extremal distribution

1949 G Method for establishing the confidence Gumbel (1949)

limits in the plotting of annual maximum
values

1949 F The empirical relationship between the Langbein (1949)

probability of annual-maxima flood series
and the expenctancy of partial flood series

1953 EP Type I extremal distribution (fisher- Chow (1953)

Tippett - Gumbel) applied to EP

1953 EP Log-normal distribution applied to three- McIllwrainth (1953)

day maximum rainfall

1954 S Extreme-value law is practically a Chow (1954)

special case of the log-normal
probability law. Suggests the use of
third parameter to log-normal distribu-
tion, the coefficient of variation,
which is shown ta be related to the
coefficient of skew (log-skew distribution)
 - 40 -
TABLE I cont.
History of frequency analysis af floods
and precipitation

Year Items R e f e r e nee

1954 S Extremal distributian type III Gumbel (1954a)

(Weibull) applied ta draught
frequency analysis. This distribution
is still the mast accepted ane for this
purpase (Deininger and Westfield, 1969)

1955 S One farmula for three types of distri- Jenkinson (1955)

butions distinguished by u /o test
l 2

1957 EP Gamma distribution used in the Friedman and Janes

estimation of rainfall probabilities (1957)
for Connecticut

1958 S A least squares procedure developed for Brakensiek (1958)

fitting a normal, log-normal and
extremel distribution

1958 S Basic textbook on the statistics of Gumbel (1958)

extremes and applications

1959 G A graphical method for the determination Chow (1959)

of frequency factors of a log-normal
distribution proposed

1960 EP Evidence for the acceptability of the Hershfield (1960)

type I extremal distribution (Gumbel
procedure) far EP analysis

1962 S Practical procedure for fitting extremel Panchang and Aggarwal

distributions by the method of maximum (1962)
likelihood
I
1963 G Set of graphs for use in constructing Gringorten (1963b) I
the confidence limits

1964 S Probability model for analysis of partial Shane and Lann (1964) I
duration series

1964 EP The log-normal distribution is acceptable Thom (1966b) I!

in characterizing the winter season U.S. Weather Bureau i
maximum accumulated water equivalent (1964)
series J
 - 41 -
TABLE I cont.
History of frequency analysis of floods
and precipitation

Year Items Re f e r e nee

1968 F The log Pears on type lIT distribution has Benson (1968)
been selected as the base method of
flood-frequency estimating in the U.S.A.

1969 5 One formula for I, II and III extremal Jenkinson (1969)

distribution types (General Extreme Value
Distribution) •

1970 EP I extremal distribution applied to maximum Zeller (1975)

depth of snow cover and to rate of snow
melt

, 1973 EP Log PearsonlIIdistribution applied to McAndrew (1973)

: extremes of water equivalent of snow
I1975 F The general extreme-value distribution NERC (1975)
recommended to achieve standardization
of flood-frequency onalysis used in
the United Kingdom

1976 F PearsonIlI and Log Pearson liT distribu- Mosonyi (1976)

tion types generally recommended for
flood estimation in West Germany

1977 F In a general way the Pearson type 111 Bobee and Robitaille
distribution fits the annual flood data (1977)
more ~dequately than the log Pearson
typel1i

1977 EP I extremal distribution applied to Martinec(l9Z7)

maximal snow loads

1978 EP General Extreme Value Distribution lJppalo (1978)

applied on nation-wide basis
 TABLE II
Basic assumptions for a frequency analysis of extreme precipitation and methods for testing them

~ s sum p t ion Causes of error Tests [Link] of correction R e f e r e nee s

ICORRECTED or accurate Wind-field defor- Gouges set up in sites shielde~

fileosured values mation above the from the wind or ground-level
The measured and orifice of gouge gauges instead of eleYdtea
true values of elevoted above gauses
precipitation do the ground and
not deviate exposed to the
wind Jevons (1861); Rodda (1971);
WMO (1973); Struzer and Go1ubev
Wetting and eva- Wetting and ~mptying tests
(1976); Sevruk (1979); Alletup
poration loss of Evaporation tests
and Madsen (1979)
can-type gouge
Instrument error Calibrdtion Motthews (1938); Porsons (19414
of recording Spldsh-out Lewis and Watkins (1955);
gauges Underhill (1966); Renard and
Osborn (1966); WMO (1973); ....
N
Kelwoy (1975); Colder ond I
Kidd (1978)
Human errot Control of raw data

CONSISTENT data Alteration of Double-moss analysis:

series gauge or gauge Merriam (1937); Light and Kohler
No change occur- high; . (1943); Kahler (1949); Kohler anc
red during the introduction of Lindsley (1949); Weiss and
period of obser- shield; Wilson (1953); Anderson (1955);
vation in the chonses in bbset~ Searcy and Hordison (1960); Sing
internal conditions vdtion procedure; Double moss analysis: (1969); Honsel ond Sch~fer
of the measurement etc. Plotting of cumulative" precipi- (1970o,b); Brunet-Moret (1971);
system tation sums against a control Farmer and Fletscher (1972);
which is highly correlated Chong and Lee (1974); WMO (1974)
HOMOGENEOUS data Growing woods or
with the tested varioble,
series new constructions, General:
consistent and homo~eneous
No chonge occur,;. elevated in the 1 Thorn (19660); YeYievich and ~eh9
red during the ViCinity. of gauge, (1967); Markovic (1967); Snyder
period 'of obser- gouge moved to 0 (1971); Bernier (1977)
vation in the diff~rent site, "I
external conditions etc. ~
of measurement
 TABLE II cont.
Bosic assumptions for a frequency analysis of extreme precipitation and methods for testing them
i i
~ s sum p t ion Causes of error Tests and methods of correction R e f e r e nee s

'ISTA~IONARY* data Alterations in Statistical tests of central

ser~es weather-pattern tendency (averaoes):
The mean, standard T - test (Student) Frederick et al. (1977
deviation and the U - test (by Wilcoxson, Monn Mann and Whi tney (1947); Hershfield'
moments of higher Effect of urbani- and Whitney) (1962); Eggers (1974); Babee and
order do not chan- zotion (Huff, 1977 Rabitai11e (1977)
ge with time. Terry test Terry (1952fi Babee and Robitaille
(This assumption Analysis of variance (1977)
cannot hold true
**Stati~tical tests of variability
in partial dura-
(st. deviation, variance):
tion series; Fisher (1929)
F - test
Models III and IV Ronk dispersion test Eggers (1975)
in Figure 2)
Run - test Thorn (19660); Eggers (1970)
trend analysis
INDEPENDENCE of data
no correlation
Fluctuation of
extreme precipi-
Tests for significance of
auto correlation coefficient
Wol1is and Motolos (1971) ...w
exists among the tation (extreme of
members of data precipitation can Rank correlation - test Eggers (1970; Bobee and Robitoille
series (especially have the same phy_ (1977); etc.
important for sical origin)
partial duration
series) (high precipi ta- Counting ranks Eggers (1970; Turner (1971)
tion follows high
precipitation Goutereou - test Turner (1971)
Phose - length - test by wallis Wallis ondMoore (1941); Eggers
and Moore I etc.• (1970)
Test by Anderson; Wald and Anderson (1941); Wold ond
Wolfowitz; Shuster; Walker Wolfowitz (1943); Chow (1964);
Bobe e and Robitoil1e (1977)

* HEC (1975)suggests that no examination is needed because the extreme-values series are practically
stationary
** Sneyers (1975) suggests that the same methods as for the central tendency may, also be used for the
t8st of variability if the absolute values of deviations of indiVidual measurements from the mean
are kno~n
 TAllL~ II Gont,
Basic assumptions for a fr~quenGY analYsis of extreme pr~cipitation and methods for testing them
i
Ass u m p t ion Causes of error Tests and methods of correction R e fer e nee 5

SUFFICIENT length Due to the shortness Hershfield and Wilson

of time interval from of record the sampl~ (1958); Huff and Neill
~hich the data are does not represent (1959); KuznetsoYa (1964),
dra....n ih papl,llation Lambor (1967); Liebscher
The length of record (1970); Weisner (1970);
of 25 years* is to Miller
short for good def~~ The errors in estimates are Hershfield and Kohler
nition of the distri... 1 a function of record length (1960); KuznetsoYa (1964),
bution type and (Table Ill) Nosh and Amorocha (1966);
rainfall estimates Glass and Krause (1967);
Liebscher (1970); Maaley
and Crutcher (1968);
Bridges and Haan (1971,
1972); Kaller (1974);
DaYi' et al, (1976)
The individual station data Clarke-Hafstad (1938,
may b~ compQunded in ,regional 1942); Lindsley et al.
studies (stat~on~years concept) (1958); Kuznetsava (1964); t
Bell (1969); Caa (1974);
NERC (1975), Dickinson
(1977); Baghirathan and
Shaw (1978)
An outli~r can be identified Chow (1953) i Ferguson
Outliers: the rainfall
arises from weather by using thG ratio of 100 to (1961); Hershfield (1962~
situations quite unlike 50-year return period values Kuksin (1965); Rodda
those of other events in from nearby stations and, if (1967); Grubbs (1969);
needed, replaced by the next Snyder (1972, 75), See
the sample (Tomlinson!
.978) largest value in the same yeqr 3.4.1
"XTREMES OF PR"C,PITA- The distribution which I See Table ~ 4.1.1 $ee 4.1.1
TION follow 0 speci~ seems to fit the $ample
fied distribution does not apply to its
function population

THE ESTIMATES OF THE Di ff~renc;es betw~en See Table xm, JX,4.1. 2 see 4.1.2
PARAMETER values of the parameter values
the distribution ore estimated by different
unbiassed procedures

• Jenkinson (1955~ 69)'·indlcates 0 procedure which is applicable to annual series with 0 shorter length
of record (eight to ten years)
 - 45 -

TABLE III

Comparison of daily maximum precipitation estimates using various

lengths of observation period (occording to Kuznetsova (1964»*

Station n N = 25 years 50 S N < 60 years Max.

dai 1y
Difference betl~een- the maximum and the minimum prec i pi
estimates in per cent of the maximum tation
year T=10 20 50 T=100 .T=lO 20 50 T=100 (mm)
n > 100 years
Leningrad 124 6 9 12 13 76
Sverd10vsk 124 48 51 53 52 28 27 31 26 94
Barnau1 121 73 69 58 60 40 45 45 43 61
LU9ansk 121 48 49 52 52 13 12 10 10 80
Orenburg 108 22 33 49 58 8 16 26 33 60
Nerchinsky Zavod 106 36 36 67 85 17 29 44 46 154
Karpi nsk 104 44 39 34 29 25· 24 24 25 77
Astrakhan 102 38 40 45 48 'i 7 18 22 23 73
Gorky 102 34 41 55 65 23 26 32 33 79
Kiev 101 16 21 25 27 4 7 13 18 104
MEDIAN 108 38 40 52 52 23 26 31 33 80
50 < n ~ 80 years
Kostroma 80 11 12 7 1; 62
Liepaya 74 36 59 90 103 119
V1adivostok 74 29 28 24 21 178
Penza 72 25 38 57 71 100
Surgut 71 22 25 28 27 87
Sol e<:hard 69 2 2 10 15 53
Turukhaus 68 26 31 41 51 61
Kzyl- Drda 57 41 42 45 43 41
Yakutsk 52 5 0 2 2 4~

MEDIAN n 25 28 28 27 62

*N'" length of record used for estimation; n = t<>tal length of record; ..

T = return period .in years
 - 46 -

TABLE IV

Alternative names for some distributions and some of their characteristics

Distribution family Distribution type Alternative name and/or

relationship (see Table I)

(a) Continuous distributions

General Extreme Value (GEV)

Jenkinson (1969) defined - I extremel IFisher-TiPpett type I,
Gumbel, Double exponentiol
GEV distribution using
- II extremal Frechet, Fisher-Tippett type 11,
one formula for all
log Gumbel
three types
- III extremal Fisher-Tippett type Ill, Weibull

Pearson III
This has three parameters; - Exponential
II shape parameter = 1 (negative .
exponentiol)
for certain values of these - Gamma ,I scale parameter d
parameters the distribution - Erlangian .1 sbap~ paro:eter equals an integerl
~ay change name as follows: - Normal I coefficient of skewness = ~

! Log: distributions
I
I Derived by taking logorithm - log-normal I Gibrat '-5 law
of the value or shifting the - log I extremal I1 extremel l log Gumbel
values and taking their - log Pearson III
!
logarithm
1

j (b) Discrete distributions

I (Wu and Goodridge 1974)

I Binomial distribution
- Poisson number of Bernoulli trials very
i large
I Pascal distribution
I~ - Poisson number of successful trials very
large
- Gauss probability of success of each
trial approaches zero
- Geometric number of successful trials =1
- Negative binomial number of successful trials not
an integer
 - 47 -

TABLE V

Tests for a selection of distribution type

and their criticism+-

T est Re f e r e nee

*+Chi-square-test Markovic (1965); Keeping (1966); Draschoff (1972);

(Goodness-of-fit test) Maniak and Grobe (1972); Koberg and Eggers (1973);
Musy (1973); Belke (1974); Dello Lucia and
Fattorelli (1974); Weingartner (1975); Samuelsson
(1972)
+ Kolmogorov-Smirnov·test Keeping (1966); Droschoff (1972); Koberg and
(Goodness-of-fit test) Eggers (1973); Belke (1974); Weingartner (1974);
Thiess (1975)
Likelihood ratio test Koberg and Eggers (1973); Weingartner (1974);
Bodo and Unny (1976)
Sign test Keeping (1966); Del10 Lucia ond Fottorel1i (1974)
++Tests for distinguishing Jenkinson (1955); Van Montfort (1970, 1973);
between I ond 11 extremal Bardsley (1977); Otten and Van Montfort (1978)
distributions
er 1/••-2 ratio test for Jenkinson (1955); Krishnan and Kushwaha (1975);
extremel distributions Dickinson (1977)
Test for distinguishing Pearson (1938); Maniak and Grobe (1972); Bobee
between Peorson types and Robitaille (1977)
Coefficient of determinotion Gringorten (1960); Gupta (1970)
Friedman's two-way Gupta (1970)
onolysis by ronks
Comparison between estimates Benson (1968); Bobee and Robitaille (1977)
and observed data
Control curves or bands Gumbel (1958), Sneyers (1963)

+ Inadequacy of test: Chi-square test (Keeping, 1966; NERC, 1975), Kolmogorov-

Smirnov-test (Belke, 1974; NERC, 1975; Bobee and Robitaille, 1977). Massey (1951)
compared Chi-square test and Kolmogorov-Smirnov test
• Bodo and Unny (1976) and Plate (1979) concluded that no statistical test is accurate
enough to discriminate between distributions
* The results may depend on the arbitrary assignment of the class interval (Gumbel,
1958; Brunet-Moret, 1971; Belke, 1974; Weingartner, 1974)
++Otten and Van Montfort (1978) compared Bardsley (1977) and Van Montfort (1970) tests
 _ 48 -

TABLE VI

Distribution types used for the frequency onolysis of the annual

maxima of precipitaTion

R e f e r e nee s
pistribution
General Application to extremes of precipitation

EXTREMAL Fisher and Tippett (1928); Chow (1953); Hershfield and Wilson
.distribution von Mises (1936), Gumbel (1958,6~; Huff and Neill (1959a); Hersh-
type (Fisher- (1941, 54b, 58); Potter field and Kohler (1960); Hershfield
Tippet I; (1949); Jenkinson (195~ (1961, 62); Reich (1963,70) Robertson
Gumbel - see 69); Chow (1964), Lowery (1963); Kuksin (1965); Majumdar and
Table IV) and Nash (1970); Sawhney (1965); Rodda (1967,70); Guillot
Samuelsson (1972) and Duband (1967); Bruce (1968); Lecher
(1968); Fogel and Duckstein (1969);
Ganzloser (1969); Hoshemi (1969),
Richardson ond Midgley (1979); Jenkinson
(1969); Reich (1970); Ginsburg (1971);
Stol (1971); Draschoff (1972); Miller
(1972); Hershfie1d (1973); Della Lucia
and Fattorelli (1974); Fides et al. (1974
Kaller (1974); Krishnan ond Kushwaha
(1975); Tripathi (1976); Wan (1976);
Ayoade (1977); CMWDP (1977); Dickinson
(1977); Sneyer (1977); Baghiratan and
Show (1978); 8uishand (1978); Kappus
et al. (1978); Wurtele and Roe (1978);
leller et al. (1978); Samaj and Volovic
(1978);~ further ref. see Miller (1977)
ttI EXTREMAL Frechet (1927; Fisher and Huff and Neill (1959a,b); Jenkinson (1959/
, distribution Tippett (1928); von Mises 69) Reich (1970); Draschoff (1972); Farmer
,type (Fr<khet, (1936); Gumbel (1941,58); and Fletscher (1972); Dhar and Farooqui
: Fisher-Tippett Jenkinson (1955,69); Grin- (1973); Della Lucia and Fattorelli (1974);
, II; Log-Gumbel gorten (1960); Chow Krishnan and Kushwaha (1975); Sneyers
'-see Table IV) (1964) (1977); Kappus et al. (1978); Wurtele and
Roe (1978); Uppala (1978), leller et al.
(1978)
and II EXTRE- Jenkinson (1955,69; Huff and Neill (1959a,b); Reich (1970);
MAL distribu- Gumbel (1941, 58) Krishnan and Kushwaha (1975); Sneyers
tions as al- (1978); Uppala (1978); leller et 01.
ternative .or (1978)
GENERAL EX-
TREME VALUE
distribution
(Table IV)
 - 49 -
TABLE VI cont.
Distribution types used for the frequency onolysis of the onnuo1
maxima of precipitation

Distribution
R e f e r e nee 5

General App1icotion to extremes of precipitotion

I II EXTREMAL Jenkinson (1955, 69); Wu and Goodridge (1974)

(Weibull ) Gumbe1 (1958)
LOG-NORMAL Ga1ton (1875; Hazen Mc I11wrainth (1953); Hershfie1d and
distribution (1914a,b, 1921); Chow Wi1son (1958); lEA (1958); Huff and Nei11
with 2 or 3 (1954,59,64); Aitchisan (1959a); Hor1er and Rhein (1961);
parameters and Brown (1969); Burges Hershfie1d and Koh1er (1962);Kuznetsova
et al. (1975) (1964); Kuksin (1965); Majumdar and
Sawhney (1965); A1exeyev (1966); Lyutik
(1966); Mirotvorskaya (1969); Da1insky
(1971); Turner (1971); De11a Lucia and
Fattore1li (1974); Cao (1974); Burges
et al. (1975); Chin and Miller (1978);
Kappus et al. (1978)

PEARSON III Foster (1924); Pearson Draschoff (1972) Lauruschkat (1973);

(1895,1930); Chow (1964) Goodridge (1978)
Bobee and Morin (1973); Wu and Goodridge (1974)
Bobee and Robitai11e
(1977)

LOG PEARSON III Chow (1964); Benson Reich (1970); Draschoff(1972); Kappus
(1968); Bobee (1976) et al. (1978)
Bobee and Robitai11e
{l977)j Rao (19800)

BINOMIAL Chow (1964) Kuznetsova (1964); Lyutik (1966)

Distribution
'---------'------------'---~---~.~----_ .. -_._--
 - 50 -
TABLE VII
Argument for the use of I extremol ond log-normal distributions for
the analysis of annual maximo of precipitation (also see Table VIII)

I extremal distribution Reference Log-normal distribution Reference

There is some theoretical Jenkinson The applicability of 011 Thiess (1975)

basis for the choice (1955/69) logorithmic forms of
distributions is, in
It is consistent in goodness- NERC (1975) general, limited and not I
af-fi t tests justi fied I
Comparing the expected and Hershfield The deviations from the Kuznetsova(1964) I
computed values for various ond Kohler log-normal straight line I
return periods on 128 widely (1960) averaged ten per cent fo
separated stations in U.S.A., daily extremes of preci-
the I extremal distribution pitation in the U.S.S.R.
(Gumbel procedure) was found ond only seldom omounted
to be acceptable for the to twenty per cent
estimation of extreme whereas for lOO-year
roinfall period of observation
there were no deviations
at all

I extremal distribution (Gumbel procedure) is better than the log- Hershfield

normal distribution (1962) and Kuksin
(1965)

On a theoretical basis the I extremal distribution will give good Majumdar and
estimates for return periods of up to 1000 yeors when the Sawhney (1965)
coefficient of variation is less than 0.5, even when the population
distribution is log-normal

An empirical appraisal shows Reich Log-normal distribution Lyutik (1966)

the general applicability (1970) applied only when Cs'" 3
of the I extremal distri- and binomial when Cs"'" 3
bution (Gumbel procedure) (C = coefficient of
to extremes of rainfall sk~wness)

By applying the general Samuelsson The log-normal D and Della Lucia

extreme value distribution (1972) the 11 extremal D have and Fattorelli
to extreme values which been mare frequently (1974)
sig ni ficantly belong accepted than the I
to several different kinds extremal D (Gumbel
of distributions, it was procedure)
shown that this method could
be applied as a standard
one.
The results show that the Dickinson
Gumbel distribution should (1977)
rearesent the data well
 _ 51 -

TABLE VIII
Comparison of lOO-year daily rainfall estimates by various distribution
types and fitting procedures (methods of parameter estimation)

Compared Number Length Di ffe- Note, Reference!

distribution of of rence Argument Country
types* stations record A- B
(year) (% A)
I EXTREMAL D vs. LOG-NORMAL C
I extremal D (G) A 1 38 10 The difference is almost Hershfield and
lag-normal D B trivial compared with the Wilson (1958)
possible sampling error North Carolina
of a single record U.S.A.
I extremal D (G) A 9 40 5 Both D give smaller va- Huff and Neill
log-narmal D B lues than the ranked (1959)
data Illinois,
U.S.A.
I extremal D (G) A 339 50 10 The I extremal D (G) was Hershfield
lag-normal D (M) B the best (1962) U.S.A.
1 extremal D !1O Conversion factors when Majumdar and
log-narmal D the coefficient of Sawhney (1965)
variation <: 0.5 India
I extremal D (J) A 3 14-30 +
- 1 Samuel son
log-normal D B (1972) Crete,
Greece
- -
I extremal D (G)
I extremal D The I extremal D (G) was Kuksin (1965)
(g by Alekseyev) the best of the five D. White Russia
Only in cases with U.S.S.R.
log-normal D oufliers did the graphic
(g by Alekseyev) 20 40_70 5-10
procedures show values
log-normal D up to 30 -per cent greater
than the analytical pro-
log-normal D
procedures.
(Blakinov proc.) ,
*D = distribution type; (G) = Gumbel's fitting procedure; (g) = graphical fitting
procedure; (J) = Jenkinson's fitting procedure; (L) = Lieblein's fitting procedure;
(M) = moment's fitting procedure; (SQ) = least squares
Note: The comparison of estimates by different D have been made for floods more
frequently and extensively than for extremes of precipitatian. See Rozhdestvensky
and Chebotarev (1974) for references from the U.S.S.R. and for further references:
Bernier (1959), Benson (1968), Lauterbach (1968), Liebscher (1970), Reich (1970),
San gal and Biswas (1970), Maniak and Grobe (1972), Cehak (1973), Eidg. Amt. (1974),
Mosonyi et al. (1975), NERC (1975), Bobee and Robitaille (1977), Handel et al.
(1977), Shaligram and Lele (1978), Kaller and Riebe (1978), Platte (1979~
 52 -
TABLE VIII cont.
Comparlson of lOO-year daily raiQfall estimates by various distribution
types and fitting orocedures (methods of oarometer estimation)
Compared Number Length Di ffe- Note, Reference/
distribution of of rence Argument Country
types" stations record A- B
(year) (% A)
I EXTREMAL D vs. PEARSON m D or LOG PEARSON mD:
1 extremal D (G) A 4 17-29 -8 to Reich (1970)
log Pearson 3 D B +14 Pennsylvania
U.S.A.
- -
I extremel D (G) several** No significant diffe- Miller et al.
thousands rences between the (1973) cited
I extremel D (L) four D
of statioo by Frederick
Pears on III D years et al.(1977)
U.5.A.
log Pearsonill D

I extremal D
= =
Kappus et 01.
II extremal D 1 39 log PearsonIIID was (1978)
log-normal D the best Ir-an

log Pearson m D
-
log Pearsan HID The Pears on III D is Majumdar and
quite good for all Sawhney (1965
practical purposes and India
perhaps better when
the nature of D cannot
be decided

"D = distribution type; (G) = Gumbel's fitting procedure; (g) = graphical fitting
procedure; (J) = Jenkinson's fitting procedure; (L) = Lieblein's fitting procedure;
CM) = moment's fitting procedure; (SO) = least squares
*"Duration of rainfall = 1, 6, 24 hours
 _ 53 -

TABLE VIII cont.

Comparison of lOO-year doily rainfall estimates by various distribution
types and fitting procedures (methods of parameter estimation)

Compared Number Length Di ffe- Note.. . Reference!

distribution of of rence Argument Country
types* stations record A- B
(year) (% A)
I EXTREMAL vs. 11 EXTREMAL D:
I extremal D (G) A 9 40 -30 11 extremal D shows a Huff and Neil
II extremal D (G) B . more even D of posi- (1959)
tive and negative Illinois
deviations from the U.S.A.
ranked data

I extremal D (G) A 9 40 -26 11 extremol D over- Huff and Neil

11 extremal D(J) B estimates are larger (1959)
for the longer return Illinois
periods U.S.A.

I extremal D (G) A 4 17-29 -50 to II extremol D estimates Reich (1970)

II extremal D B -80 were unduly larger than Pennsylvania
the biggest observa- U.S.A.
tions

11 extremal D was best

of 011 D for return
periods of less than
10 years

I extremal D (G) A 70 70 -20 to II [Link] o is Zeller et 01.

11 extremal D(G) B -30 better than I extremal (1977)
o in the north-east Switzerland
preAlpine region

I extremel D (G) A 339** 50 10 Both D give equivalent Hershfield

I extremel D (L) B results (1962) U.S.A.

I extremel D (G) A 1 30 18 Samuelsson

I extremel D (J) B*** (1972)
Crete, Greece

* D = distribution type; (G) = Gumbel's fitting procedure; (g) = graphical fitting

procedure; (J) = Jenkinson's fitting procedure; (L) = Lieblein's fitting procedure;
(M) = moment's fitting procedure; (SO) = least squares
** 2 - 10 OOO-y~ar eyents have been estimated
***Jenkinson's ~1969) procedure
 - 54 -
TABLE VIII cont.
Comparison of lOO-year daily rainfall estimates by various distribution
tunes and fittinn nrocedures (methods of aorometer estimation)
Compared Number Length Di ffe- Note, Reference/
distribution of of rence Argument Country
types* stations record A- B
(year) (% A)
I EXTREMAL D FITTED BY DIFFERENT PRDCEDURES:

I extremel D (G) A 9 2D-5D 6 Hardly any difference ** Krishnan and

extremel D (J) B ( from between the two methods Kushwoho
(If/62
= 1) 150) os compared to observed (1975)
values from 150 stations India
(Figure 3)
I extremel D (G) A 23 20-50 -3 Both methods under-
extremel D (J) B ( from estimate the observed
k/"2 '" 0.9) 150) values but (G) fit
them better (150 [Link]'-
tions)

I extreme! D (G) A 12 20-50 15 Observed values are

extreme.! D (J) B ( from underestimated by
("/"2 ~ 1.1) 150) (J) and overestimated
by (G) but (J) fit
them better (150
stations)

I extremel D (G) A 1 70 Geiger (1975)

I extremel D (M) B (from 4 Switzerland
I extremol D (SO)B 350) 0

* D = distribution type; (G) = Gumbel's fitting procedure; (g) = graphical fitting

procedure; (j) = Jenkinson's fitting procedure; CL) = Liebleinls fitting procedure;
(M) = moment's fi tting proc~dure; (SO) = least squares
** Compiled by the authors from Krishnan and Kushwaho (1975)
Note: Mdrkovic (1965) found that the yearly precipitation sums from the 1614 U.S.A.
stations with a minimum of 30 years of observation ore best fitted by the log-normal
D with two parameters. However, the differences between the results of all tested D
(normal, log-normal, Gamma both with two and three parameters) have not been sig-
nificant and thus practically all five D are applicable for annual precipitation.
According ta Muster (1970) the differences between the estimations of EP by gamma D
and I extremal D are negligible
 _ 55 -

TABLE IX

Methods of parameter estimation for annual maxima series

in the order of their spreading and numerical feasibility
(decreasing with the arrow) and general efficiency (increasing)
with the arrow)

Method Re f e r e nee

Moments* Pearson (1939), Gumbel (1941), Huff and Neill (1959a),

Fisz (1962h Hershfield (1962), Dawntown (1966), Thom(1966),
Lowery and Nash (1970), Bobee and Robitaille (1977), Dickinson
(1977), (Roo 1980b)

Least squares** Chow (1951, 1953), Gumbel (1954b),

Brakensiek (1958), Huff and Neill (1959a), Draschoff (1972),
Snyder (1972), Snyder and Wallace (1974), Richardson and
Midgley (1979)

Maximum likelihood* Ifl Fisher (1922), Kimball (1946, 49), Greenwood and Durand (1960)
Panchang and Aggarwall (1962), Panchang (1967), Jenkinson (1969
1977), Lowery and Nash (1970), Samuelsson (1972), Leese (1973),
Matalas and Wallis (1973), Le Due and Stevens (1977), Otten
and Van Montfort (1980)

Others Deininger and Westfield (1969), Thiess (1975), Bobee and

Robitail1e (1977),Houghton (1978), Jowitt (1979), Artina and
Lamberti (1978)

* Jenkinson's (1955) procedure is based on moments. 61/62 ratio is used to select

the distribution (01 = standard deviation of the annual maxima;
62 = standard deviation of the greater member in pairs of annual
maxima). Later Jenkinson (1969) uses a ratio based on the mean
values of sextiles of the ordered series to select between the
I, 11 and III extremal distributions (Table IV) and the methad
of maximum likelihood for parameter estimation

** Gumbel's (1954b) procedure 15 a special case of least squares and uses the
geometric mean of the two regression coefficients

Lieblein's (1954) procedure; data fitted to the I extremal distribution using

the original order of the series and the weights according
to the rank in subgroups
 _ 56 -

TABLE X

Distributions used for the frequency onalysis of the magnitudes of partiol~

duration series (some references deal with floods)~

Distribution Reference Distribution Reference

type type

Exponentiol Dingens et 01. (1966)/ Polyo Dingens et 01. (1966)

distribution Vinogradov (1969), (Negative binomial)
Stol (1971), Musy (1973),
Stidd (1973), Smith et al. Gamma* Kotz and Neumann
(1974, Dickinson (1977), (1963), Musy (1973),
Buishand (1978) Muster (1970) Handel
et 01. (1977),
Thiess et al. (1976),
Buishandl"I978)
Log-normal Horler (1962, 1977), Sinepower Kumaraswamy (1976)
Rima (1964)

Pearson III Draschoff (1972)/ Normol Stidd (1973)

Musy (1973), (for transformed
Handel et 01. (1977), values)
Brunet-Moret (1977) Geometric Duckstein et al.
(1972)

Log Grobe (1974, 1979), Goodrich N~mec (1965)

Pearson III Stahlmann et 01. (1974),
Handel et al. (1977) I extremal Bell (1969)

Peorson IV Dingens et al. (1966)

* Further references (general): Friedman ond Janes (1957), Chow (1964),

Szigyart6 (1967~ Mooley and Crutcher (1968), Bridges ond Haan (1971), Thiess (1975).
Applications to the extremes of precipitation: Thom (1958, 660), Stacy (1969),
Strommen and Horsfield (1969) and in Miller (1977)
 · - 57 -

TABLE XI

Distributions for the number of events above the base

level of partial-duration series

Distribution R e f e r e nee
Poisson Thorn (1959), Borgman (1963), Hall and Howell (1963),
Shone and Lynn (1964), Fagel and Duckstein (1969),
Todorovic and Yevievich (1969), Zelenhasic (1970),
Duckstein et al. (1972), Cunnone (1973), NERC (1975),
Davis et al. (1976), Calenda et al. (1977), Rosbjerg (1977).

Polya NERC (1975), Beran (1977), Calenda et al. (1977)

(Negative binomial)
 _ 58-

ANNEX I

Part A

Camputation of the first and second extremal distributions with the choice
between them by the test according to Van Montfort - an example from Switzerland

The following analysis concerns the annual maxima series (1901-1970) from the
station Sever in south-eastern Switzerland, in the basin of the river Inn. The
altitude is 1712 m and the mean annual precipitation 879 mm (1901-1940). The analysis
is in three parts: Parameter estimation of the distribution using Gumbel's method
(Gumbel, 1958); testing whether the I or 11 extremal D is apprapriate using the
Van Montfort test (Van Mantfort, 1970); computation af confidence limits, according
to WMO (1974). Each part is described in detail and presented separately (Parts B-F).
In Part E a summary of results is made. Here, only the first three and last four
values (i = 1, 2, 3, ••• 67, 68, 69, 70) are given but, for the sake of recomputation,
all annual daily precipitation maxima of the sample can be found in Part F. The
probability papers for the I and 11 extremal D are presented in Figure 6. The
following symbols have been used:

c. J. confidence limits sx' sn - standard deviation

e, se scale and location parameters T - return peri od. in (years)
D distribution function t - fractil of the standardized
normal distribution
i rank in descending order
x - annual maximum of daily
K - frequency factor precipitation. in (mm)
1 leaps x. y - mean ~

1" logarithm naturalis llX. lly difference between successive

n sample size (70 years) . values
n' - n - 1 - confidence level
r - correlation coefficient a - abreviation
of 1 and Yi+V2 E - sum
- standard error of estimate
 - 59 -

ANNEX I cont.

Part B

Parameter estimation of the I extremal distribution for the annual

maxima series from Bever (Switzerland) by Gumbel's method
Step Step For m u 1 a e and res u 1 t s Note
No. description

1 Ranking of xi in see Part F

descending order
2 tXi 3324 mm
3324
~=~ =10= 47.3 mm
n

3 x?1
t x?
1
169 516 mm 2
11 048 976 mm2
(t Xil)2-----~­
t x? _ (tx;) I The numerator
= 1 n = 12.9 mm is nand
Sx n
not n - 1.
n + 1
4 Calculation of Ti =-r Part E
the return period
by using Weibull 's
formulae
5 Yi = -In (-In(l - iil)
tYj 38.g
- _ t y; _ 38.8 __
Y - -n-'" - 1 0 0.554

6 y?1
ty?1 119.9

= 1.19
x-x
Calculation of the F(x) = e
-e---
c
parameters of the
I extrema 1 D . c=~=12.9= 10.9 Steps3,6
sn T:l9
x =x - cy = 47.3 - 10.9 x O.55 = 41.4
 - 60 -

ANNEX I cont.

Port B cont.
Parameter estimation of the I extremel distribution for the annual
maxima series from Sever (Switzerland) by Gumbells method
Step Step For m u 1 a e and res u 1 t 8 Note
No. description

Plot of X against The distribution is fit~ed by a straight Figure 4

T in extremal line with equation for x: Step 4
probability paper
x(T) = XtC . y (T) wgere This function
y(T) = -In (-In (1 - T )) can be used
for extrapo-
lation
Example:
Calculation of 4.6 Figure 4
the lOO-year daily 91. 5 mm
maximum precipitation

Note: Visual inspection of Figure 4 shows that the distribution fits very well.
Anyway it may be desirable to know whether the 11 extremel distribution could
be used as well
 61 -

ANNEX I cont.

Port C

Choice between the I and 11 extrema1 distributio~

according to Van MontFort (1970)

Step Step r Cl r m u 1 a e a nd res u 1 t s Note

No. description

7 Parts B F f

8 6Yi = Yi+l - Yi Part B f

Step 5
1,· Mi
9 Calculation of = "':::':;"--
the leaps 1i 6Yi
767.9
10 21322.5
1
11 Yi+V2 = -In(-ln(l Ti + )) Part E
V2
n + 1 Port F
Ti+V2 = i +V2

rYi+V2 37.3

12 2 108.7
rYi+V2
13 The test statistic rl i • Yi+V2 371 .1
is the correlation
coefficient r of E1Y - ~rl~jL....;:,r,It'j,-±+JLlIz.?
i i+1I2 nI n - 1
1 and Yi+V2 r= --:...:..::..:.....---..::..----~2,.-- =-0.042 nl ::

(rlf- (rli)2 ) (rY~+1I2 _ (rYi~Y2))

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

Reading of the 0.22 Because

critical value of r « rcrit
r from Fi gure 5 the I extrem31
for the confidence D can be
level of 95% accepted
 - 62 -

ANNEX I cont.

Port D

Computation of the confidence limits, according to WMO (1974)

Step Step Formulae and results Note

No. description

14 Calculation of the K; = Yi -Y Port B

frequency factor K sn

15 s; = .Vl + 1.14 K + 1.1 KZ

16 Standard error of B' s Port B
se;=
estimate 1fnA Step 3

Assuming normal D. x :!: t (a} • sei See Figure 4.

the confidence limits are: For the range
of extraoolatico
the same' formulael
can be used '

Note: Assuming that the lower limit is zero, the 11 extremal D is the log transform
of I. If the Van [Link] Test- indicates that the IT extremel D is appropriate, one
takes the logarithm of the measured values and calculates the same steps as mentioned
above. Taking the antilogarithm leads to the final re-suIts. For plotting purposes
the 11 extremel D paper in Figure 6 can be used
 ANNEX I cont.

Port E
Porameter estimation of the I extremal distribution by Gumbel's method and computotion for the choice
of distributian by the Montfort test and of confident limits occording to the WHO
(Station Sever-Switzerland)

Parameter est1mat1oh Choice of the d1str1but1ol1 (Van Montfort test) Conf1dence limits
5 Y nJ b 0 1 5
i x x2 T y i oy
" 1 ,2 y;tV2 y2;tV2 l'YitV2 K s· c.1.
Numbering of columl,5; identical to numbering of steps in Parts 8, C and [)
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

Res u 1 t S
r
1 90 8100 71 4.26 18.11 9 .70 12.85 165.16 3.85 14.'80 49.43 3.12 3.91 11.91
2 81 6561 35.5 3.55 12.64 • 8 .41 19.38 375.72 3.33 [Link] 64.52 2.53 3.31 10.07
3 73 5329 23.6 3.14 9.87 1 .~o 3.39 11.49 2.96 . 6.91 10.11 2.18 2.95 9.00 lJ'o
W
, I

67 30 900 1.06 -1.06 1.12 3 .09 31.49 990.46 -1.10 1.21 -34.68 -1.36 1.22 3.71
68 27 729 1.04 -1.15 1.33 2 .12 16.59 275.13 -1.20 1.46 -20.03 -1.44 1.26 3.90
69 25 625 1. 03 -1. 27 1.62 1 .i8 5.63 31.75 -1.35 1.82 " 7.60' -1. 54 1.36 4.15
70 24 5766 1.01 -1.45 2.10 - - - - - - - -1.69 1.49 4.54

Sum'
- 3324 169516 - 38.83 119.90 - - 767.92 21322.49 37.34 108.69 371.1; - - -
 - 64 -

ANNEX I co"t.

Part f

Annual maxima of daily precipitation for Bever

(Switzerland) 1901-1970

rank da [Link] year rank [Link] year

maximum ma,[Link]
i (11Ul1)
i (mm)
,

I 90.0' , 1926 36 "7.U' I ge'l

2 e 1.\1 I'ISQ 37 47.0 192J
3 73.0 ' l'IbV 'fd 47.0 1'146
4 7l.0, 19"ll .19 ':"';).(1 14'56
'5
6
11.0
;'7.0 I
I,
I 1%2
1'122 .
4U
41
qLl.O
43.11
1'124
1'15'1
7 66.0 I "
1'101 42 "5. o· 1'1(,7
il &5..0 1'140 43 42.11 1'17 U
'I 60.0 190H Q2.0 1'130
10 So.v 1950 ""
"·5 "2.V 1,'14'1
11 S.~. 0 1928 46 41.0 1923
12 S8.v 1903 47 "LV 1'152
13 58.0 19'.2 '3 "11 40.0 1957
I" 58.0 1963 49 LlO.,O l'lI.1
15 57,.0 19.B 50 "0.0 1'14'1 "'.
11> 51.0 IqLf2 SI 3'1.0 1905
17 Sh.1J. 1965 52 39.0 1,9" 7 .
18 55.0 1917 53 38.0 1951
19 55.0 1906 S" 38.11 1934
20 53.0 1956 55 36.0 1938
21 5-~. 0- 1'/10 5& 36.0 1939
2Z 52'.0 1911 51 30..0 1912
23 52.0 1914 5ll 35.0 1904
,,,I 5e.V 1'110 59 35.0 1961
25 50.0 1'136 60 3~.0 19"1
2b 4'1.0 1929 bl 34.0 19(,4
i'.7 Q'I.O [Link] 62 34.11 190'1
26 49.~ 1'131 , 63 3".0 1932
29 lIY.O 19\6 64 ,<'.0 1901
30 0
4'~. 19:>3 bS 31.0 1 'I SS
31 48.0 196'1 66 31.0 J 945
32 /.18. iJ 1915 67 30.0 1943,'
33 48.0 19&6 68 27.0 191'1
34 48.0 19O1l 69 l5.0 1'lcZ "
35 43.0 1931 70 24.0 1921