A Comparison of Methods Used

In Flood-Frequency Studies for

Coastal Basins in California

GEOLOGICAL SURVEY WATER-SUPPLY PAPER 1580-E

Prepared in cooperation with the

California Department of ff^ater Resources
A Comparison of Methods Used
In Flood-Frequency Studies for
Coastal Basins in California
By R. W. CRUFF and S. E. RANTZ

FLOOD HYDROLOGY

GEOLOGICAL SURVEY WATER-SUPPLY PAPER 1580-E

Prepared in cooperation with the

California Department of Heater Resources

JNITED STATES GOVERNMENT PRINTING OFFICE, WASHINGTON : 1965

 UNITED STATES DEPARTMENT OF THE INTERIOR

STEWART L. UDALL, Secretary

GEOLOGICAL SURVEY

Thomas B. Nolan, Director

For sale by the Superintendent of Documents, U.S. Government Printing Office

Washington, D.C. 20402 - Price 70 cents (paper cover)
 CONTENTS

Page
Abstract______________________.__-_--_____---___---__ El
Introduction_.____________________________________________________ 1
Purpose and scope-----_----_-___-__-_-_---___--____----__----_ 3
Acknowledgments ____-__-__-_-_-_____----________-___-___-____ 3
Statistical equations used in this study___________________________ 5
Description of the methods of analysis used in this study_______________ 5
Index-flood method.___________________________________________ 6
Multiple correlation_________________________________________ 9
Logarithmic normal distribution____-____-_____-_-_-___-____-_-_- 10
Extreme-value probability distribution or Gumbel method________ 12
Pearson type III distribution_________________________-_________ 13
Gamma distribution.__________________________________________ 14
Analysis of flood frequency in the San Diego area._____________________ 17
Index-flood method.___________________________________________ 17
Multiple correlation_________________________________________ 21
Logarithmic normal distribution_________----_---_____-_-_-_____- 22
Extreme-value probability distribution or Gumbel method. __ _______ 25
Pearson type III distribution.__________--____-____-_--_-_-_-_-- 28
Gamma distribution.____-________________._-_-__-__-_-_____-_- 28
Analysis of flood frequency in north coastal California ______________ 29
Index-flood method.__________________________---_-__-----_--__ 33
Multiple correlation__________________________________________ 41
Logarithmic normal distribution__________-____---_____-________- 41
Extreme-value probability distribution or Gumbel method__________ 43
Pearson type III distribution.__________________________________ 45
Gamma distribution.__________________________________________ 45
Discussion of results of the analyses_______________________________ 47
Index-flood and multiple-correlation methods.____________________ 51
Statistical distributions.___-_-___-----__-----_--_-_-__-----__--_ 51
Summary and conclusions__________________________----_-_-__--__ 54
References cited_____-______-_-_____-_____________---_-----_----- 56
m
IV CONTENTS

ILLUSTRATIONS

[All plates are in pocket]

PLATE 1. Map of San Diego area showing location of stream-gaging

stations used in this study.
2. Map of north coastal California showing stream-gaging
stations and boundaries of hydrologic regions used in
this study.
Page
FIGURE 1. Index map of California_-______--__----_---__----_-_ El
2. Graph showing relation of C in the gamma distribution
to log (M/M0)___-___-______--___---__-_----_-___ 16
3. Bar graph showing period of peak-discharge record for
gaging stations in the San Diego area.______________ 19
4. Flood-frequency curve for Santa Ysabel Creek at Suther-
land Dam__________________________ 20
5. Dimensionless regional flood-frequency curve for the San
Diego area (index-flood method)__________________ 21
6-8. Graph showing relation of
6. Mean annual flood to drainage area in the San Diego
area (index-flood method)__-___--_-----_---___- 22
7. Mean and standard deviation of logarithms of
annual peak discharges to drainage area in the San
Diego area._ _______-_---__-_--_----_----__- 26
8. Mean and standard deviation of annual peak dis-
charges to drainage area in the San Diego area___ 27
9. Bar graph showing period of peak-discharge record for
gaging-stations in north coastal California ___________ 31
10. Flood-frequency curve for Klamath River at Klamath__ 33
11. Flood-frequency curve for Eel River at Scotia__________ 34
12. Dimensionless regional flood-frequency curves for north
coastal California (index-flood method)______________ 35
13-18. Graph showing relation of
13. Mean annual flood to drainage area in the Coast
Ranges (index-flood method).________________ 37
14. C to mean annual basinwide precipitation in the
Klamath Mountains (index-flood method) _____ 38
15. The mean of the logarithms of annual peak dis-
charges to drainage area in the Coast Ranges___ 42
16. The standard deviation of the logarithms of annual
peak discharges to mean annual basinwide pre-
cipitation in the Klamath Mountains.._______ 44
17. The mean and the standard deviation of annual peak
discharges to drainage area in the Coast Ranges. _ 44
18. C to mean annual basinwide precipitation in the
Klamath Mountains (gamma distribution) _____ 46
 CONTENTS

TABLES

Page
TABLE 1. Pearson type III recurrence curve data, Kgt values, in standard
deviations from the mean_______________________________ E15
2. Summary of hydrologic parameters for basins in the San Diego
area--__---_---___-_-_-____-__-___-_-_-_-________-____ 18
3. Qsa for streams in the San Diego area as determined from
graphically derived flood-frequency curves and by various
methods of regional flood-frequency analysis._____________ 23
4. Qioo for streams in the San Diego area as determined from
graphically derived flood-frequency curves and by various
methods of regional flood-frequency analysis_-_-__-_______ 24
5. Summary of pertinent hydrologic parameters for basins in
north coastal California______-____________-____________ 32
6. Qso for streams in north coastal California as determined from
graphically derived flood-frequency curves and by various
methods of regional flood-frequency analysis.------------- 39
7. Qioo for streams in north coastal California as determined from
graphically derived flood-frequency curves and by various
methods of regional flood-frequency analysis.-..---------- 40
8. Summary of differences between discharges determined from
graphically derived flood-frequency curves and discharges
computed by various methods of regional flood-frequency
analysis. ______________________________________________ 48
9. Summary of differences between Qso determined from graphically-
derived flood-frequency curves and Q50 computed from various
distributions using statistical parameters from individual
station arrays.________________________________________ 54
 FLOOD HYDROLOGY

A COMPARISON OF METHODS USED IN FLOOD-

FREQUENCY STUDIES FOR COASTAL BASINS
IN CALIFORNIA

By K. W. GRUFF and S. E. KANTZ

ABSTRACT

This study compares the results of regional flood-frequency studies made by

several methods and appraises the relative reliability of these methods. The
areas selected for study were the subhumid San Diego area in southwestern
California and the humid coastal area in northwestern California. The follow-
ing six methods of analysis were applied to each region: Index-flood method,
multiple correlation, logarithmic normal distribution, extreme-value probability
distribution (Gumbel method), Pearson type III distribution, and gamma distri-
bution. The last four methods named involved not only the computation of the
statistics appropriate to the distributions, but also the relating of these statistics
to basin and climatologic characteristics. On the basis1 of an empirical, non-
statistical test, the following conclusions were reached :
1. All methods of analysis give better results in a humid region than in a sub-
humid region because streamflow is less variable in a humid region.
2. If historical data, either qualitative or quantitative, are available concerning
the magnitude of floods that occurred in the years prior to the collection
of streamflow records, the multiple-correlation method of analysis is
preferred. Only this method and the index-flood method benefit from the
historical data, which, in effect, extend the time base of the analysis.
The multiple-correlation method is superior to the index-flood method be-
cause it has a far more rational basis and in addition gives better results.
3. Where the peak-discharge data are limited entirely to the period during which
streamflow records were collected (no historical data available), a method
based on the distribution of the array of peak flows is preferred because of
its greater objectivity. Of the four distributions tested, the Pearson type III
is the most desirable. It is more flexible than the other three and will
generally fit the peak-discharge data best.
Although this comparison study of flood-frequency methods was based on
small samples from only one part of the United States, the results and con-
clusions appear to be meaningful because they can be explained rationally.

INTRODUCTION

The principle of analyzing flood magnitudes on a probability basis

is almost universally accepted because its use permits economic con-
siderations, as well as hydrologic factors, to govern the planning and
El
E2 FLOOD HYDROLOGY

design of projects that are susceptible to flood damage. There is no

universal acceptance, however, of any single method of making the
flood-frequency analysis. Usually, a basic objective of the analysis
is to derive flood magnitude-frequency relations which may be used
at a site for estimating the magnitude of rare or unusual flood events,
such as the peak discharge that has an average probability of being
exceeded only once in 100 years (Qwo ) or of being exceeded only once
in 50 years (Q50 ). Because our records of flood discharges are gen-
erally short less than 30 years, on the average extrapolation by some
means is required to estimate the magnitudes of unusual floods. How-
ever, the magnitudes so determined depend, to a large degree, on
the method of frequency analysis that governed the extrapolation.
Therefore, it is not unusual for independent workers, using the same
short streamflow records but different methods of analysis, to obtain
widely differing values of discharge corresponding to Q50 or $100-
Furthermore, because of the element of uncertainty that characterizes
any extrapolation, it is seldom possible to decide which of the derived
discharges are the most accurate or which method of analysis is the
most reliable.
For a detailed discussion of the many methods of deriving flood
magnitude-frequency relations, the reader is referred to reports by
Jarvis and others (1936) and Benson (1962). Several of the meth-
ods described in those reports are no longer in favor, and others have
had varying degrees of popular acceptance over the years. At present
(1964), the methods most commonly used in this country are the four
listed below. The agency or agencies shown in parentheses are the
chief proponents of the methods.
1. Index-flood method (U.S. Geol. Survey).
2. Multiple-correlation (U.S. Bur. of Public Roads and U.S. Geol.
Survey).
3. Logarithmic normal distribution (U.S. Army, Corps of Engineers).
4. Extreme-value probability distribution or Gumbel method (U.S.
Weather Bur.).
These four methods and two additional ones the Pearson type III
distribution and the gamma distribution were used in this study.
The Pearson type III distribution was included because the authors
feel that this method is likely to regain the popularity it once had in
probability studies of peak discharge. The gamma distribution was
included because it is increasingly being used, both in the United
States and abroad, for studying the probability of occurrence of hy-
drologic events, including peak discharges.
METHODS USED IN FLOOD-FREQUENCY STUDIES, CALIFORNIA E3

PURPOSE AND SCOPE

The purpose of this study was to compare the results obtained by

applying each of the six methods of flood-frequency analysis to identi-
cal sets of peak-discharge data. Specifically, it was the values of Q50
and (?100, computed by each method, that were compared. One com-
parison was made for streams in the San Diego area in south coastal
California, and another was made for streams in north coastal Cali-
fornia. (See fig. 1.) There were several reasons for selecting these
two areas. First, the comparison is given broader scope by using two
areas that differ greatly in the amount of precipitation they receive;
the San Diego area is subhumid, whereas the north coastal area is very
humid. Second, flood-frequency studies have recently been published
for each area, and the hydrologic factors needed in the analyses were
therefore readily available. The San Diego area study, published by
the California Department of Water Resources (1963), analyzed the
data for 18 stream-gaging stations by two methods the index-flood
method and multiple correlation. Of the 18 peak-discharge records,
2 were longer than 35 years, 11 were between 20 and 35 years, and 5
were shorter than 20 years. The study for north coastal California
published by the U.S. Geological Survey (Rantz, 1964) analyzed the
data for 27 stream-gaging stations by the index-flood method. Of the
27 peak-discharge records, 4 were longer than 30 years, 9 were between
10 and 30 years, and 14 were shorter than 10 years.
The principal reason for selecting the two areas lay in the fact
that, although the discharge records for the streams were relatively
short, the magnitudes of the greatest and second greatest flood peaks
in the past 100 years were generally known within reasonable limits
of accuracy. With this knowledge it was possible to appraise the
results obtained by each of the six methods of analysis and to draw
conclusions concerning the relative reliability of the methods.
Many competent statisticians, using long-term discharge records and
rigorous statistical treatment, have made comparison studies of the
various methods of analyzing flood magnitude-frequency relations
without arriving at definitive conclusions concerning the superiority of
any one method. Hence the plethora of methods of analysis from
which the practicing engineer must choose before making a flood-
frequency study. In view of the many uncertainties involved in a
comparison study of methods of analysis, the authors have eschewed
any rigorous statistical approach, such as an analysis of variance or
a determination of confidence limits. They have adopted, instead, a
simple pragmatic approach in which the basis of comparison is the
relative ability of the various methods to reproduce Q50 and Qi0o at
each of the study sites in the two California areas. This approach
770-401 O J©5 2
E4 FLOOD HYDROLOGY
124° 123" 122° 121" 120"
EXPLANATION

117° H6°

FIGURE 1. Location of areas studied.

has its statistical shortcomings but is not unreasonable. After all,

the objective of all methods is to fit some relation to the available
peak-discharge data, and because it is the magnitudes of the infrequent
floodflows that are usually sought, the method that best fits the data
for the infrequent floods is presumably the most reliable method for
use. Admittedly, the conclusions reached in this report through this
pragmatic approach primarily represent impressions gained during
the course of the study, but the findings appear to be meaningful.

ACKNOWLEDGMENTS

This study was made under the terms of a cooperative agreement

between the U.S. Geological Survey and the California Department of
Water Resources. The report was prepared by the Surface Water
Branch of the Geological Survey under the general supervision of
Walter Hofmann, district engineer. E. D. Cobb, hydraulic engineer,
assisted in the computation and preparation of the data.
The authors are grateful for the many helpful suggestions of D. R.
Dawdy, J. R. Crippen, M. A. Benson, and H. C. Riggs, hydraulic en-
METHODS USED IN FLOOD-FREQUENCY STUDIES, CALIFORNIA E5

gineers of the Geological Survey, who reviewed the manuscript. How-

ever, the subject treated here is not one on which unanimity of opinion
can be expected. The opinions and assertions in this report are those
of the authors and do not, in all particulars, reflect the views of their
coworkers in the Geological Survey.
STATISTICAL EQUATIONS USED IN THIS STUDY

We assume that the reader has some knowledge of elementary

statistics and is familiar with common statistical nomenclature and
with the equations for computing such elemental items as mean, stand-
ard deviation, correlation coefficient, and linear regression equation.
These terms and equations can be found in any standard statistics text
(for example, Ezekiel and Fox, 1959). All other statistical equations
that are used in this report are explained where they first appear and
are thereafter referred to by number as equation 1, equation 2, and
so on.
DESCRIPTION OF THE METHODS OF ANALYSIS USED IN
THIS STUDY
In this section of the report, the six methods of analysis used in this
study are briefly described. In all six methods the streamflow data
analyzed consist of the momentary maximum discharge of each year of
record. Gaging stations whose peak discharges are seriously affected
by manmade storage or diversion are not used. All six methods
attempt to make the most efficient use of the data available. The
index-flood and multiple-correlation methods are to some extent
empirical because the graphical curve-fitting procedures that are basic
to both methods require a certain amount of subjectivity on the part
of the analyst. These methods have the advantage of permitting the
analyst to use whatever qualitative or historical information is avail-
able for extrapolating the flood-frequency relation. The other four
methods are empirical in the sense that a type of distribution loga-
rithmic normal, extreme-value probability, Pearson type III, or
gamma is arbitrarily selected for use. One may theorize concerning
the probability distribution that describes the occurrence of flood
events, but the lack of agreement among hydrologists indicates that
the "true" distribution is not known. Once a distribution is selected
for use, however, the analysis becomes strictly objective, and the ex-
trapolation is automatically made from a mathematical determination
of the statistics mean, standard deviation, and coefficient of skew of
the station data. Qualitative or historical data cannot be assessed,
however, when a rigorous statistical solution is made. On occasion,
this type of information is used to define or modify the high-water end
E6 FLOOD HYDROLOGY

of the computed flood-frequency relation; but when this information

is so used, the statistical distribution no longer controls the all-impor-
tant high-water end of the flood-frequency relation. As in the empiri-
cal index-flood and multiple-correlation methods, the results obtained
will depend to a large degree on the judgment of the analyst in inter-
preting the historical information and assigning probabilities to the
peak discharges. In this report, an objective mathematical treatment
was used for the four methods involving standard statistical distribu-
tions, with no consideration given to qualitative or historical
information.
Regardless of the method of analysis used, it is customary to develop
flood magnitude-frequency relations that are applicable to an entire
region rather than to a single gaging station. Because the flood series
for a single station is a short random sample, it may not be representa-
tive of the long-term distribution of flood events at the gaging station.
Combining records for all stations in a hydrologically homogenous
area tends to reduce the sampling error associated with a nonrepresent-
ative sample. Another advantage of the regional flood-frequency
relation is that it can be applied to ungaged sites in the region. The
boundaries of a homogeneous region must be rationally delineated from
a knowledge of the hydrology of the region; commonly, these bound-
aries will coincide with the boundaries of physiographic sections de-
lineated by Fenneman (1931, 1938) or with the boundaries of regions
delineated on soil classification maps published by the U.S. Soil Con-
servation Service. In making a regional study, a common base period
of years is usually used for each gaging station. This base period is
generally the period of record for the older gaging stations in the re-
gion, and the shorter records are extended by correlation procedures
to cover the base period.
Generally, the term "regionalization," as used by engineers engaged
in flood-frequency analysis, refers not only to the delineation of the
boundaries of hydrologically homogeneous regions, but also to the
establishment of relations between pertinent characteristics of the
flood-frequency curve and basin or climatologic parameters within the
homogeneous region. For example, values of the mean annual flood
are said to be regionalized when a relation is found between mean
annual flood and size of drainage area in the region being studied.
The terms "regionalization," or "regionalized" are used in this broad
sense in this report.
INDEX-FLOOD METHOD
The index-flood method (Dalrymple, 1960) has been the standard
U.S. Geological Survey method of flood-frequency analysis for the
past 15 years. A step-by-step outline of the procedure follows:
METHODS USED IN FLOOD-FREQUENCY STUDIES, CALIFORNIA E7

1. Peak-discharge data within the base period are tabulated for each
station with 10 or more years of record.
2. Historical or qualitative information is noted for each station. As
an example, suppose the base period is 1930-60. Information of
the following types may be available :
"Flood of 1955 is of approximately the same magnitude as that
of 1862, the greatest previously known." Or, "Prior to 1955,
the flood of 1932 was the greatest known since the flood of 1893."
3. Peak discharges needed for each short-term station to complete the
record for all years of the base period are computed from a re-
gression line or equation. The regression is obtained by graph-
ically correlating concurrent peak discharges for the short-term
station and a nearby long-term station.
4. The peak discharges at each station are ranked in order of magni-
tude, starting with 1 for the greatest discharge, 2 for the second
largest discharge, and so on.
5. The recurrence interval for each observed peak discharge is com-
puted. This is n6t done for the peak discharges computed by
regression equation in step 3. The only purpose of the computed
peak discharges is to provide a basis on which to estimate the
recurrence interval for the observed peak discharges. The
formula used to compute recurrence interval is :

where
RI is the recurrence interval in years,
n is the years of record, and
m is the order of magnitude of an annual peak discharge.
6. Recurrence intervals are adjusted, where appropriate, on the basis
of historical or qualitative information. In the example in
step 2, n for the base period is 31 years. The flood of 1955,
being the greatest of record during the base period, would
normally have its recurrence interval computed as 32 years.
However, in the past 99 years, a flood of this magnitude occurred
twice in 1862 and in 1955 but has not been exceeded. There-
fore, a peak discharge equivalent to that of 1955 has orders of
magnitude 1 and 2 in 99 years, giving it recurrence intervals of
50 and 100 years, instead of the single recurrence interval of 32
years that was originally computed. From the statement re-
garding the flood of 1893, we know that the peak discharge of
1932 was not merely the second largest in 31 years, but was the
third largest in at least 68 years, and it therefore has a recurrence
interval of 23 years rather than the 16 years originally computed.
E8 FLOOD HYDROLOGY

(We assume in this example that the flood of 1893 is greater

than that of 1932 but that its peak discharge is unknown.)
7. For each station, the recurrence interval is plotted in relation to
peak discharge on extreme-value probability graph paper. A
straight line or gentle curve is fitted by eye to the plotted points.
8. The mean annual flood ( Q2,33 ), defined as the discharge correspond-
ing to a recurrence interval of 2.33 years on the graph described
in the preceding step, is selected for each station.
9. The peak discharge corresponding to a 10-year recurrence interval
($10) on the graphs described in step 7 is selected for each
station.
10. The comparative slope of the individual curves between Q w and
$2.33, or the ratio of Q10 to $2.33, is computed for each station.
11. From a knowledge of the hydrology of the region being studied,
areas are delineated that are expected to have similar ratios of
Qw to $2.33. Commonly, the boundaries of the areas will be
governed by: (a) physiography (similar ratios would be
expected where drainage basins have similar shape; where pre-
cipitation may occur as either rain or snow, high-altitude basins
would tend to have one ratio and low-altitude basins another)
and (b) mean annual precipitation (generally, the more humid
the area, the smaller the ratio).
12. A homogeneity test (Dalrymple, 1960, p. 38-39) is made of the
ratios of Qlo to $2.33. On the basis of this test, the boundaries of
the areas selected in step 11 may be adjusted, but a rational ex-
planation for any changes that are made is desirable.
13. At each station all recorded discharges are divided by $2.33; so
these discharges are expressed as dimensionless ratios.
14. The median dimensionless discharge ratio for each recurrence in-
terval is determined for each group of stations in the homoge-
nous areas selected in step 11 or 12.
15. For each homogeneous area, median dimensionless discharge ratios
are plotted in relation to recurrence interval on extreme-value
probability paper, and a straight line or curve is fitted by eye
to the plotted points.
16. The procedures described in steps 1-15 apply only to those stations
with 10 or more years of peak-discharge record. Peak dis-
charge data are next tabulated for those stations with 5-9 years
of record within the base period.
17. Concurrent peak discharges for each of these short-term stations
and for a nearby longer term station are correlated graphically.
$2.33 for the short-term station is then determined from the
regression line, it being the discharge corresponding to $2.33
for the longer term station.
METHODS USED IN FLOOD-FREQUENCY STUDIES, CALIFORNIA E9

18. From a knowledge of the hydrology of the region being studied,

areas of probable homogeneity with regard to the mean annual
flood are delineated. Generally, this homogeneity refers to a
similarity in infiltration characteristics. As mentioned earlier,
the boundaries of these homogeneous areas will often coincide
with the boundaries of physiographic sections delineated by
Fenneman (1931, 1938) or with the boundaries shown on Soil
Conservation Service soil classification maps. These bound-
aries, however, may or may not coincide with those delineated in
step 11 or 12.
19. Within each of the areas from step 18, $2.33 for 'both long- and
short-term stations (values obtained from steps 8 and 17) is
correlated graphically with drainage area and with other signif-
icant parameters such as mean annual rainfall, areas of lakes
and ponds, main-channel slope, and mean basin altitude.
The correlation graph from step 19 and the dimensionless flood-
frequency curve from step 15 are the end products of this analysis.
Used for the appropriate region, the first graph provides a means of
estimating Q 2.33 from basin and climatologic parameters; the second
graph is a regional flood-frequency curve in which discharges are ex-
pressed as a ratio to Q->.^.
MULTIPLE CORRELATION
The multiple-correlation method of flood-frequency analysis is be-
coming increasingly popular in the U.S. Geological Survey. In per-
forming a flood-frequency analysis by this method, a region of prob-
able hydrologic homogeneity is first selected as previously described.
For each gaging station in the region with 10 or more years of peak-
discharge record within the base period of years an individual flood-
frequency curve is drawn by following steps 1-7 that are described in
the preceding section entitled "Index-flood method." After the station
flood-frequency curves have been prepared, discharges are read at
selected recurrence intervals, such as 2.33 years ($2.33), 5 years (d)»
10 years (do), 20 years (do), 50 years (do), and 100 years (doo)-
Each set of discharges is then correlated with various basin and clima-
tologic parameters, using a regression equation of the form:
QT =aB*C°D* ....... (2)
where
QT is the discharge corresponding to a recurrence interval of T
years,
«, &, c, d . . . are the constants, and
B, C, D . . . are the basin and climatologic parameters.
E10 FLOOD HYDROLOGY

The basin and climatologic parameters that are considered include

drainage area, mean annual precipitation, area of lakes and ponds, land
slope, main-channel slope, mean basin altitude, a shape factor, and
others. The constants in the equation are computed by least squares,
and statistical tests are made to eliminate from the equation those
parameters that have little or no significance. From the final equa-
tions for discharges corresponding to selected recurrence intervals, a
flood-frequency curve can be constructed for any site in the region,
whether gaged or ungaged, once the values of the significant param-
eters are determined.
The U.S. Bureau of Public Roads uses a variation of this method
for analyzing floods from small drainage areas. The Bureau method
(Potter, 1961) involves a graphical, correlation between discharges
corresponding to a selected recurrence interval and three hydrologic
parameters drainage area, a precipitation index, and a topographic
index. Standard curves have been prepared for several regions in the
country.
LOGARITHMIC NORMAL DISTRIBUTION

In making flood-frequency studies, the U.S. Army, Corps of Engi-

neers, bases its analysis on the premise that the logarithms of annual
peak discharges are normally distributed. Beard (1962) prepared a
detailed description of the method used by the Corps of Engineers; a
brief step-by-step resume of the method is given below. Only those
stations with 10 or more years of peak-discharge record within the
base period of years are used in the analysis.
1. Logarithms of the peak-discharge data within the base period are
tabulated for each station in a region of probable hydrologic
homogeneity.
2. The mean and the standard deviation of the array of data for each
station are computed.
3. The mean and the standard deviation for each short-term station
are adjusted to cover the complete base period. This is done by
first computing a linear correlation for concurrent peak dis-
charges at the short-term station and at a nearby long-term sta-
tion. The adjustment of the mean and the standard deviation
is then made by means of the following equations :

(3)
and __
Mlft-Mla= (Af»-Af*) ( JR) 2OSWS») , (4)
where
M is the mean of the logarithms of the peak discharges,
S is the standard deviation of the logarithms of the peak
discharges,
METHODS USED IN FLOOD-FREQUENCY STUDIES, CALIFORNIA Ell

R is the coefficient of correlation adjusted for lost degrees of

freedom,
1 is the short-term station,
2 is the long-term station,
a is the short-term period, and
6 is the base period.
These formulas have some statistical shortcomings, but they tend
to give an unbiased estimate of the all-important long-term
standard deviation.
4. To regionalize the statistics of the logarithmic normal distribution,
the base-period mean (M) and base-period standard deviation
(S) of the logarithms of peak discharge are each correlated with
basin and climatologic parameters in the homogeneous region.
(Eemember that in a logarithmic normal distribution the antilog
of the mean of the logarithms of the peak discharges corresponds
to the geometric mean or to the median of the natural values of
these discharges and not to their arithmetic mean.)
5. From the relations obtained in step 4, a flood-frequency curve can
be constructed for any site in the region, whether gaged or un-
gaged, by use of the equation:
QT=M+KTS, (5)
where
QT is the logarithm of the discharge corresponding to a re-
currence interval of T years,
M is the mean of the logarithms of annual peak discharges,
S is the standard deviation of the logarithms of annual peak
discharges, and
KT is a characteristic of the normal distribution; for the pur-
pose of this report it may be defined as a coefficient corre-
sponding to a recurrence interval of T years. (The fol-
lowing table gives values of K corresponding to selected
values of recurrence interval, RI.}
RI RI
(years) K (years) K
2____________.________ 0.00 20____________________ 1.64
5___________.________ . 84 50____________________ 2. 05
10____________________ 1.28 100___________________ 2.33
6. If plotted on logarithmic normal probability graph paper, the com-
puted flood-frequency relation will be a straight line.

770-401 O 65 3
E12 FLOOD HYDROLOGY

EXTREME-VALUE PROBABILITY DISTRIBUTION OR GUMBEL

METHOD

About 20 years ago E. J. Gumbel began advocating the use of an

extreme-value probability distribution for analyzing the magnitude-
frequency relation of annual peak discharges, and his method still
enjoys great favor among hydrologist-statisticians. Although a more
recent publication by Gumbel (1958, p. 236, 272) describes three basic
distributions of extreme values which may be used for flood studies,
the distribution he originally advocated is still the most widely used.
The U.S. Weather Bureau is one of the chief proponents of the Gumbel
method and uses it for both precipitation-frequency and flood-fre-
quency studies at individual sites.
To apply the Gumbel method, an area of probable hydrologic homo-
geneity is first selected. The base-period mean and the standard
deviation of annual peak discharges are then computed for each gag-
ing station in the region with 10 or more years of peak-discharge
record within the base period by following steps 1-3 that are described
in the preceding section entitled "Logarithmic normal distribution."
The only difference is that in the Gumbel method natural values of
peak discharge are used, and not their logarithms. The base period
mean (M) and standard deviation ($) are then regionalized by corre-
lation with basin and climatologic parameters in the homogeneous
region. From these regional relations, a flood-frequency curve can
be constructed for any site in the region, whether gaged or ungaged,
by use of the formula:
QT =M+K'T8, (6)
where
QT is the discharge corresponding to a recurrence interval of T
years,
M is the mean of the peak discharges,
8 is the standard deviation of the peak discharges, and
K'T is a characteristic of the extreme-value probability distribu-
tion ; for the purpose of this report it may be defined as a coeffi-
cient corresponding to a recurrence interval of T years. (The
following table gives values of K' corresponding to selected
values of recurrence interval, RI.)
RI
(years) K' IK'
2.33 _______ ..__ __ _ _ __ 0 20 ___ _ __ ___ _ 1.87
5_ ________ . ______ .72 Kt\ _. _ _____ 2.59
10 ____ __ __.__ _ _ _ _ 1.30 100__ ______ - __ _ _ _ _ 3.14
If plotted on extreme-value probability graph paper with arith-
metic ordinate, the computed flood-frequency relation will be a
straight line.
METHODS USED IN FLOOD-FREQUENCY STUDIES, CALIFORNIA E13

PEARSON TYPE III DISTRIBUTION

The Pearson type III distribution, or such variations of it as are

expressed in the Hazen and Foster methods (Jarvis, 1936), was at
one time widely used in probability studies of peak discharge. Its
popularity for this purpose has declined in the past 20 years, but
there is now renewed interest in its use. The Corps of Engineers
uses the Pearson type III distribution in probability studies of flood
volume, in which the annual flood volumes for various durations are
expressed as logarithms. In this study natural values of annual
peak discharges are used.
The Pearson type III distribution is more flexible than either
the logarithmic normal or Gumbel distributions and can be more
closely fitted to the data because it is defined not only by the mean
and the standard deviation of the array of flood peaks but also by
the coefficient of skew of the array. A discussion of the Pearson III
distribution is given by Elderton (1953). A step-by-step descrip-
tion of its application in a flood-frequency study follows. Only
those stations with 10 or more years of peak-discharge record are
used in the analysis.
1. Peak-discharge data within the base period are tabulated for each
station in a region of probable hydrologic homogeneity.
2. The mean and the standard deviation of the array of data for each
station are computed.
3. The mean and the standard deviation for each short-term station
are adjusted to cover the base period. This is done by first com-
puting a linear correlation for concurrent peak discharges at a
short-term station and at a nearby long-term station. The ad-
justments are then made by applying equations 3 and 4 used in
step 3 of the section entitled "Logarithmic normal distribution."
(Note that natural values of the discharges, and not their loga-
rithms, are now used in the two equations.)
4. Before computing coefficients of skew, the individual values of peak
discharge at the short-term stations that are needed to complete
the record for all years of the base period are next computed by
means of the equation :
(7)
where
JTi and X2 are the peak discharges for any given year at sta-
tions 1 and 2, the short- and long-term stations, respectively,
M-L and Mz are the mean values of annual peak discharge for
concurrent periods at stations 1 and 2, respectively, and
$1 and $2 are the standard deviations of the annual peak dis-
charges for concurrent periods at stations 1 and 2,
respectively.
E14 FLOOD HYDROLOGY

Equation 7 is used rather than a regression equation to minimize

bias in the standard deviation of the final array of observed and
computed peak discharges.
5. The coefficient of skew is computed for each station from the
equation :

9 N(N-l)(N-2)(S)*
where
g is the coefficient of skew,
N is the number of years of record (base period) ,
X is the magnitude of a peak discharge, in cubic feet per
second, and
S is the standard deviation, in cubic feet per second.
All peak discharges, both observed and computed by equation 7,
are used in determining coefficients of skew.
6. The base-period mean (M) , standard deviation (S) , and coefficient
of skew (g) are regionalized by correlation with basin and clima-
tologic parameters in the homogeneous region.
7. From the relations obtained in step 6, a flood-frequency curve can
be constructed for any site in the region, whether gaged or un-
gaged, by use of the equation :
QT =M+KgTS, (9)
where
QT is the discharge corresponding to a recurrence interval
of T years,
M is the mean of the peak discharges,
S is the standard deviation of the peak discharges, and
Kg? is a characteristic of the Pearson type III distribution ;
for the purpose of this report it may be defined as a
coefficient corresponding to coefficient of skew equal to g
and a recurrence interval of T years. (Table 1 gives values
of K corresponding to selected values of g and recurrence
interval.)
GAMMA DISTRIBUTION
The gamma distribution is a special case of the Pearson type III
distribution where the locus parameter is zero (Thorn, 1958, p. 117).
It is also similar to the chi-square distribution that is commonly used
in statistical tests. The gamma distribution has two parameters
the arithmetic mean, M, and a shape parameter, C. The parameter,
(7, is a function of log (M/Mg), where Mg is the geometric mean of
the array (fig. 2). The use of the gamma .distribution in hydrologic
METHODS USED IN FLOOD-FREQUENCY STUDIES, CALIFORNIA E15

TABLE 1. Pearson type III recurrence curve data, KgT values, in standard
deviations from the mean

Recurrence Interval, in years

Skew coefficients (g)
1.1 2 5 10 20 50 100

3. 82 i__._ _ <.__ ................ 23.25 24.28

3. 0 .............. -0.65 -0.40 0.45 1.18 2.02 3.13 4.02
2.8............................. -.69 -.38 .48 1.20 2.02 3.09 3.95
2.6.... . . - -.74 -.37 .51 1.23 2.01 3.05 3.87
2.4... .... . .... . . ... -.80 -.35 .55 1.25 2.01 3.00 3.78
2.2 -.86 -.33 .58 1.28 2.01 2.96 3.70
2. 0.... ......................... -.91 -.31 .61 1.30 2.00 2.91 3.60
1.9. .... ... ... ........ . ... . -.93 -.30 .62 1.31 1.99 2.88 3.55
1.8 -.96 -.28 .64 1.32 1.98 2.85 3.50
1.7 -.99 -.27 .66 1.32 1.97 2.81 3.45
1.6............................. -1.02 -.25 .68 1.33 1.96 2.77 3.40
1.5..... ................. -1.04 -.23 .70 1.33 1.94 2.73 3.34
1.4... ..... . -1.07 -.22 .72 1.34 1.93 2.70 3.28
1.3 . -1.09 -.21 .73 1.35 1.92 2.67 3.22
1.2. -1.11 -.19 .74 1.35 1.90 2.63 3.15
1.1 -1.13 -.18 .75 1.34 1.89 2.58 3.09
1.0............................. -1.16 -.16 .76 1.34 1.88 2.54 3.03
0.9 _ __.__..._ -1.18 -.15 .77 1.34 1.86 2.50 2.96
0.8... . .. ... ....... ..... -1.20 -.13 .78 1.34 1.84 2.45 2.90
0.7 -1.22 -.12 .79 1.33 1.82 2.41 2.84
0.6--....--.-...... .... .... -1.24 -.10 .80 1.33 1.80 2.36 2.77
0.5 -1.26 -.08 .81 1.32 1.77 2.31 2.70
O.4.. ... -1.27 -.07 .82 1.32 1.75 2.26 2.62
0.3 -1.29 -.05 .82 1.31 1.73 2.21 2.55
0.2 ... . . ... ... .. -1.31 -.03 .83 1.30 1.70 2.16 2.48
0.1 -1.32 -.02 .84 1.29 1.67 2.11 2.40
0 -1.33 0 .84 1.28 1.64 2.05 2.33

1 Regional value computed for San Diego area.

2 Extrapolated for the purpose of this report.

probability studies has been discussed by Alexander (1962); its ap-

plication in a flood-frequency study is described below. Only those
stations with 10 or more years of peak-discharge record are used in
the analysis.
1. Peak-discharge data within the base period are tabulated for each
station in a region of probable hydrologic homogeneity.
2. The base-period arithmetic mean (M) for each station in the array
is computed, as explained in the earlier discussion of the Gumbel
method.
3. The base-period geometric mean (Mg) for each station in the array
is computed, as explained in the earlier discussion of the logarith-
mic normal distribution. (Mg is the antilog of the mean of the
logarithms of the peak discharges.)
4. Log (M/Mg) is first computed for each station, and is then used
with the curve given in figure 2 to obtain values of the shape
parameter, C.
5. The base-period arithmetic mean (M) and the shape parameter
(C) are regionalized by correlation with basin and climatologic
parameters in the homogeneous region.
6. The regionalized values of M and C are then used with a table of
chi-square to obtain the flood-frequency curve for any site in the
region whether gaged or ungaged. A table of chi-square can be
E16 FLOOD HYDROLOGY
~l I I FT

0.01 0.1
Log IO (WMg)

FIGURE 2. Relation of C in the gamma distribution to log (M/Mg).

found in any handbook of statistical tables (for example, Arkin

and Colton, 1950, p. 121). The number of degrees of freedom
(n) to be used in the table is equal to 2(7. Each value of chi-
square corresponding to 2(7 degrees of freedom is multiplied by
M/2C to give the ordinates of the flood-frequency curve.
7. The use of both logarithms and natural values of discharge in this
method is confusing, at first glance. Logarithms are used only
in the computation of C; natural values of the peak discharges
are used in all other computations.
METHODS USED IN FLOOD-FREQUENCY STUDIES, CALIFORNIA E17

ANALYSIS OF FLOOD FREQUENCY IN THE SAN DIEGO

AREA

A study of flood-frequency relations for coastal streams in the San

Diego area in southwestern California was recently made by the Cali-
fornia Department of Water Resources (1963). According to Fen-
neman (1931) the entire region drained by these streams lies in a
single physiographic province, the Lower California province. Ex-
cept for a narrow coastal plain that is 10-15 miles wide, the region
is mostly mountainous having a maximum altitude of about 6,500
feet at the eastern divide. The precipitation pattern is distinctly
seasonal, and about 75 percent of the rainfall occurs during the 4
months December through March. Average annual precipitation
ranges from 8 inches along the coast to 45 inches at the highest alti-
tudes. The region is generally subhumid, however, and of the 18
basins investigated, 13 had an average annual precipitation of between
15 and 21 inches. Drainage areas at the study sites range from 24
to 740 square miles.-
The locations of the 18 gaging stations used in the study are shown
in plate 1. The numbers identfying these stations on the map and in
the tables and figures that follow are those used in the State flood-
frequency report. Figure 3 shows the period of annual peak-discharge
record at each of the stations, and table 2 summarizes the basin and
climatologic parameters that were considered in the State study. In
the published study the flood magnitude-frequency relations for the
streams were analyzed by both the index-flood method and the multi-
ple-correlation method. These two analyses are briefly summarized
in the pages that follow, and in addition, analyses are made using
the four statistical distributions discussed earlier logarithmic nor-
mal, extreme-value probability, Pearson type III, and gamma. In
all analyses the entire study area was considered a hydrologically
homogeneous unit.
INDEX-FLOOD METHOD
In analyzing the annual flood data by the index-flood method, the
State followed standard U.S. Geological Survey procedures. The
base period selected for use was the 55-year period 1906-60. Annual
peak discharges needed to complete the peak-discharge array for the
base period for each of the 18 gaging stations were obtained by graphi-
cal correlation between stations. The discharges at each station were
then ranked in order of magnitude, and the recurrence interval for
each observed peak was computed by applying equation 1. Through-
out most of the San Diego area the flood of 1916 was the greatest dur-
ing the base period. A study of historic records indicated the peak
 00

TABLE 2. Summary of hydrologic parameters for basins in the San Diego area

Mean
Drainage Drainage Channel Rainfall annual Mean Channel
area basin slope intensity basin wide annual storage
Index Gaging station (sq mi) shape (ft per ft) (inches precipi- basin loss factor
No. factor per day) tation (inches)
(inches)
(A) (Sft) (S) (/) (P) (£) (SO

1 Santa Margarita River at Ysidora _________ ____ _ _ 740 0.63 0. 0097 7.50 16. 57 15.63 7.020
2 Santa Ysabel Creek at Sutherland Dam_ ____ _________ 57 .63 .0301 8. 05 28. 84 24. 23 1.415
3 Temecu la Creek at Vail Dam. _ _ _ _ ___ ____ _ ___ 319 1. 18 . 0202 9. 23 16. 24 15.55 2. 850
4 Murrieta Creek at Temecula. _ ___ _______ 220 1. 21 .0066 6.68 15. 16 14.42 13. 380
5 110 .67 .0243 6.38 19. 08 17.60 3. 945
7 Santa Maria River near Ramona_ ________ ___ ___ _ _ 57 . 77 . 0090 5.45 19. 38 17.67 7.630
8 Sweetwater River near Descanso_ _ _ ___ ___________ 44 .69 .0160 10. 22 28. 39 23. 80 2. 350
9 Arroyo Trabu co near San Juan Capistrano __ _________ 36 . 37 . 0226 6. 30 19. 74 18.02 3. 550
12 592 1. 03 . 0142 7.93 16. 37 15.63 5.650
13 Santa Margarita River near Fallbrook__ _______ ____ 645 .81 . 0132 7. 58 16.65 15. 84 3. 870
14 Guejito Creek near San Pasqual. _ _ _____________ 24 .59 . 0229 6. 35 20.54 18. 20 2. 500
15 San Onofre Creek near San Onofre___ _ _ _ ___________ 35 . 83 .0526 6. 00 14. 76 14. 20 1. 560
16 110 . 64 . 0313 6.65 26.68 22.67 1. 410
17 Jamul Creek near Jamul__ _ ___________________ _ 72 1. 13 . 0291 4. 75 17.43 16. 35 1. 290
18 San Luis Rey River at Oceanside ___ - ___ _______ 348 . 52 .0100 5. 93 20. 63 18. 89 12. 720
19 306 .65 .0130 6. 00 21. 75 19. 98 8. 290
20 San Luis Rey River at Monserate Narrows.. __________ _ 174 .60 . 0171 6. 63 25. 38 22. 00 7.510
21 San Diego River near Santee__ _ ______ _ _ _________ 380 .64 .0093 7. 35 20. 39 17. 81 8. 980
METHODS USED IN FLOOD-FREQUENCY STUDIES, CALIFORNIA E19

Period of record
Drainage
3000000 Gaging station area Station
) f« CM CO *) \C1 <0 (sq mi)

« ^H Santa Margarita River at Ysidora 740 1

^ i ^H Santa Ysabel Creek at Sutherland Dam 57 2

i
Temecula Creek at Vail Dam 319 3

_HH
. Murrieta Creek at Temecula 220 4

San Juan Creek near San Juan Capistrano 110 5

« ^^H Santa Maria River near Ramona 57 7

^^ Sweetwater River near Descanso 44 8

Arroyo Trabuco near San Juan Capistrano 36 9

_ Santa Margarita River near Temecula 592 12

_t Santa Margarita River near Fallbrook 645 13

Guejito Creek near San Pasqual 24 14

San Onofre Creek near San Onofre 35 15

t^m ^H Santa Ysabel Creek near Ramona 110 16

Jamul Creek near Jamul 72 17

i ^^H ^^m San Luis Rey River at Oceanside 348 18

San Luis Rey River near Bonsall 306 19

1 1 mmm San Luis Rey River at Monserate Narrows 174 20

ii ^ San Diego River near Santee 380 21

FIGURE 3. Period of peak-discharge record for gaging stations In the San Diego area.

discharges of this flood to be of approximately the same magnitude as

those of the flood of 1862, the greatest previously known. Therefore, a
peak discharge equivalent to that of 1916 was assumed to have an order
of magnitude of 1 and 2 in about 100 years, giving it recurrence inter-
vals of both 100 years and 50 years. A flood-frequency curve was then
drawn for each station by first plotting annual peak discharge in re-
lation to recurrence interval and then fitting a smooth curve by eye
to the plotted points. Logarithmic extreme-value probability graph
paper was used for this purpose. The upper end of the curve generally
fell between the two plotted first-order points. Figure 4 shows an
example of a station flood-frequency curve developed by the procedure
just outlined.
The discharges corresponding to recurrence intervals of 2.33 years
($2.33) and 10 years (Qw ) were selected from each of the station flood-
frequency curves for use in a homogeneity test. This test indicated that
all 18 stations could be used in constructing a dimensionless flood-
frequency curve for the entire region. To construct this curve it was
first necessary to divide the observed peak discharges at each station
TtfO-401i
E20 FLOOD HYDROLOGY
30,000
i i n
20,000

10,000
Q 8000
8 6000
LJ
w
g 4000
o.
3000

2000

1000
800

600

400

300

100 i i i
1.1 1.2 3 45678910 20 30 40 50
RECURRENCE INTERVAL, IN YEARS

FIGUEB 4. Flood-frequency curve for Santa Ysabel Creek at Sutherland Dam (sta. 2).

by $2.33 s° that each discharge was expressed as a dimensionless ratio.

The median of the 18 peak-discharge ratios for each recurrence in-
terval was then plotted on extreme-value probability graph paper
and a smooth curve was fitted by eye to the plotted points (fig. 5).
In a final step, the values of $2.33, picked from the individual station
flood-frequency curves, were correlated with the parameters listed in
table 2. Only drainage area was found to be significant; the correla-
tion is shown in figure 6.
To obtain the flood-frequency curve for an ungaged site in the study
region, figure 6 is first entered with the drainage area for the ungaged
site. The corresponding mean annual flood ($2.33)? or index flood, is
then read from the curve. The value of $2.33 is used with the frequency
curve in figure 5 to compute the ordinates of the flood-frequency curve
for the ungaged area.
Values of the 50-year flood (Q50 ) and the 100-year flood ($100)
were computed from curves given in figures 5 and 6 for drainage areas
corresponding in size to those of the 18 stations in the region. These
values have been entered in column 4 of tables 3 and 4, respectively,
for comparison with the results obtained by other methods. In
METHODS USED IN FLOOD-FREQUENCY STUDIES, CALIFORNIA E21

column 3 of these tables are listed the values ofQ50 and ^100? picked
from the individual station curves, such as the curve shown in figure 4.
MULTIPLE CORRELATION
The same 18 stations in the San Diego area were also analyzed by
the California Department of Water Resources (1963), using the
multiple-correlation method. The 55-year base period, 1906-60, was
again used, and the entire region was considered hydrologically
homogeneous. From the individual station flood-frequency curves
(such as the curve given in fig. 4) constructed for the index-flood
method of analysis, discharges corresponding to various recurrence
intervals were read. For this study we are interested only in the
discharges corresponding to recurrence intervals of 50 years (Q50 ) and
100 years (Qi00 ). All values of Q50 were correlated with the basin
and climatologic parameters shown in table 2; similar correlations
were made for values of $100- It was found that the only statistically

Q
O 10

1.0

0.1
1.2 1.3 1.5 3 4 5678910 20 30 40 50 100
RECURRENCE INTERVAL, IN YEARS

FIGURE 5. Dimensionless regional flood-frequency curve for the San Diego area (index-
flood method).
E22 FLOOD HYDROLOGY

O 2000

O
CD 800

a 400

10 20 50 100 200 500 1000

DRAINAGE AREA, IN SQUARE MILES

FIGURE 6. Relation of mean annual flood to drainage area in the San Diego area (index-
flood method).

significant parameters were drainage area (A) and a drainage basin

shape factor (Sh). This shape factor is defined as the ratio of the
diameter of a circle of area equal to basin area to the length of the
basin measured parallel to the principal channel. The following
regression equations were obtained:
^ 5o = 1016^L 0- 59^-0- 44 (10)
and
^ 100 =1288^ 0-60^-0- 57. (11)
For each equation the coefficient of correlation was 0.954.
For later comparison in the section entitled, "Discussion of results
of the analysis," the values of Q50 and $100, computed for the 18 sta-
tions by application of these equations, have been entered in column
6 of tables 3 and 4, respectively.
LOGARITHMIC NORMAL DISTRIBUTION
A regional flood-frequency curve for the San Diego region was com-
puted by fitting a logarithmic normal distribution to the original base
data. This was done to compare the results with those obtained by
other methods. The first step in the computation was to convert the
natural values of peak discharge to logarithms. None of the 18 gag-
ing stations had peak-discharge records that were complete for the
TABLE 3. Qso for streams in the San Diego area as determined from graphically derived flood-frequency curves and by various methods of
regional flood-frequency analysis

Index-flood Multiple Logarithmic Extreme-value Pearson type Gamma

Qsofrom method correlation normal probability III distribution distribution
graph- distribution distribution
ically-
Index Station derived
No. station Percent Percent Percent Percent Percent Percent
curves Qso differ- Qso differ- Qso differ- Qso differ- Qso differ- Qso differ-
(cfs) (cfs) ence (cfs) ence (cfs) ence (cfs) ence (cfs) ence (cfs) ence
from from from from from from
col. 3 col. 3 col. 3 col. 3 col. 3 col. 3

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

1 54,000 50,800 -6 60,000 +11 99,800 +85 28,700 -47 34,600 -36 28,900 -46
2 17,000 13,300 -22 13,500 -21 15,500 -9 8,530 -50 10,400 -39 7,590 -55
3 24,000 32,900 +37 28,000 +17 53,200 +122 19,300 -20 23,400 -2 18,600 -22 O
4 24,500 27,300 +11 22,000 -10 42,300 +73 16,200 -34 19, 700 -20 15,200 -38
5 26,000 18,900 -27 19,000 -27 25,400 -2 11,600 -55 14,100 -46 10,600 -59
7 9,300 13,300 +43 12,500 +34 15,500 +67 8,530 -8 10,400 +12 7,590 -18
8 10,500 11,600 +10 11,000 13,000 +24 7,530 -28 9,160 -13 6,600 -37
9 13,400 10,500 -22 13,000 ±1 11,400 -15 6,830 -49 8,300 -38 5,960 -56
12 36,500 45,400 +24 42,500 +16 83,600 +129 25,700 -30 31,100 -15 25,700 -30
13 50,000 47,700 -5 50,000 0 89,000 +78 26,900 -46 32,600 -35 26,900 -46
14 5,500 8,520 +55 8,400 +53 8,590 +56 5,670 +3 6,900 +25 4,790 -13
15 9,900 10,400 +5 8,900 -10 11,100 +12 6,730 -32 8,180 -17 5,840 -41
16 23,600 18,900 -20 19,500 -17 25,400 +8 11,600 -51 14,100 -40 10,600 -55 O
17 13,000 15,200 +17 12,000 -8 18,600 +43 9,510 -27 11,600 -11 8,580 -34
18 32,000 34,400 +8 42,000 +31 56,200 +76 19,900 -38 24,100 -25 19,400 -39
19 36,000 32,100 -11 36,000 0 52,400 +46 18,800 -48 22,800 -37 18,200 -49
20 27,000 24,000 -11 27,000 0 35,400 +31 14,500 -46 17,600 -35 13,600 -50
21 54,000 36,400 -33 40,000 -26 60,400 +12 21,000 -61 25,400 -53 20,400 -62

to
GO
 fe)

TABLE 4. Qioo for streams in the San Diego area as determined from graphically derived flood-frequency curves and by various
methods of regional flood-frequency analysis

Index-flood Multiple Logarithmic Extreme-value Pearson type Gamma

Qioo from method correlation normal probability III distribution distribution
granh- distribution distribution
Index Station derived
No. station Percent Percent Percent Percent Percent Percent
Qioo differ- Qioo differ- Qioo differ- Qioo differ- Qioo differ- Qioo differ-
curves (ob) ence (cfs) ence (cfs) ence (cfs) ence (cfs) ence (cfs) ence
(cfs) from from from from from from
col. 3 col. 3 col. 3 col. 3 col. 3 col. 3

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

1 72,000 70,500 -2 86,000 +19 190,000

27,000" +164 33,600 -53 44,200 -39 35,300 -51
2 26,000 18,500 -29 18,500 -29 +4 10,000 -62 13,200 -49 9,260 -64
3 Temecula Creek at Vail Dam.. _________ ___ 34,000 45,700 +34 37,000 +9 98,200 +189 22,600 -34 29,700 -13 22,700 -33
4 27,500 37,800 +37 28,500 +4 77,400 +181 19,100 -31 25,100 -9 18,600 -32
5 36,500 26,200 -28 26,500 -27 45,400 +24 13,700 -62 18,000 51 12,900 -65
7 12,800 18,500 +45 16,500 +29 27,000 +111 10,000 -22 13,200 +3 9,260 -28
8 15,500 16,200 +5 15,500 0 22,400 +45 8,880 -43 11,700 -25 8,050 -48
9 18,500 14,600 -21 19,000 +3 19,600 +6 8,050 -56 10,600 -43 7,270 -61
12 49,000 63,000 +29 57,000 +16 158,000 +222 30,100 -39 39,500 -19 31,300 -36
13 67,000 66,200 -1 68,500 +2 168,000 +151 31,600 -53 41,400 -38 32,800 -51
14 7,200 11,800 +64 11,500 +60 14,600 +103 6,690 -7 8,820 +22 5,840 -19
15 13,300 14,400 +8 12,000 -10 19,100 +44 7,930 -40 10,500 -21 7,120 -46
16 30,500 26,200 -14 27,000 -11 45,400 +49 13,700 -55 18,000 -41 12,900 58
17 17,000 21,100 +24 15,500 -9 32,700 +92 11,200 -34 14,800 -13 10,500 -38
18 52,000 47,700 -8 60,000 +15 104,000 +100 23,400 -55 30,700 -41 23.700 -54
19 50,000 44,500 -11 50,000 0 96,600 +93 22,100 -56 29,000 -42 22,200 -56
20 39,000 33,300 -15 37,000 -5 64,200 +65 17,000 -56 22,400 -43 16,500 -58
21 75,000 50,400 -33 56,000 -25 112,000 +49 24,700 -67 32,400 -57 24,900 -67
METHODS USED IN FLOOD-FREQUENCY STUDIES, CALIFORNIA E25

entire 55-year base period, 1906-60. The next step was to select a
few stations that were strategically located for use as base stations,
and then to estimate the logarithms of the annual peak discharges
needed to complete the 55-year array for these stations. Three of
the stations having comparatively long records, stations 2, 3, and 8,
were chosen as base stations, and the required individual logarithms
of peak discharge were computed by use of equation 7. (Note that
logarithms are used at this time in equation 7 and not natural values.)
The use of equation 7, instead of a regression equation, results in a
less biased estimate of the standard deviation of the 55-year array of
annual peak discharges, not all of which are observed.
The mean and the standard deviation for the 55-year array at each
of the three base stations were then computed. After this was done,
the mean and the standard deviation of each of the remaining 15
short-term records were first computed and then adjusted to the 55-
year base period by correlation with a base station. Equations 3
and 4 were then applied to the results. At this stage of the computa-
tions the base-period mean (M) and standard deviation ($) were
available for all 18 stations. The values of M and $, still in loga-
rithmic units, were next regionalized by correlation with the basin
and climatologic parameters in table 2. Of these parameters, only
drainage area was statistically significant. The two graphs in figure
7 show the relations between (1) drainage area and M, the mean of
the logarithms of peak discharge; and (2) drainage area and $, the
standard deviation of the logarithms of discharge. By use of these
two graphs and equation 5, a flood-frequency curve can be computed
for any site in the region. Specific equations for computing Q50 and
$10o, in logarithmic units, are:
<?60 =M+2.05£ (12)
and
<?100 =Jl/+2.33£. (13)
Values of Q50 and $100, in logarithms, were computed from curves
given in figure 7 and by application of equations 12 and 13, respec-
tively, for drainage areas corresponding in size to those of the 18
gaging stations. The antilogarithms of Q50 and Quo have been en-
tered in column 8 of tables 3 and 4, respectively, for later comparison.
EXTREME-VALUE PROBABILITY DISTRIBUTION OB GUMBEL
METHOD
A regional flood-frequency study of the San Diego region was also
made by fitting an extreme-value probability distribution to the
original base data. In this analysis natural values of the discharges
were used, and not their logarithms. Stations 2, 3, and 8 were again
E26 FLOOD HYDROLOGY

50 100 200 500 1000

DRAINAGE AREA, IN SQUARE MILES

FIGURE 7. Relation of mean and standard deviation of logarithms of annual peak

discharges to drainage area in the San Diego area.

made the base stations for the study and their records were completed
for the 55-year base period by use of equation 7. The mean and the
standard deviations for the 3 base stations and the 15 short-term sta-
tions were then computed. These statistics for the short-term records
were adjusted to the 55-year base period by correlations with a base
station. Equations 3 and 4 were then applied to the results. (Note
that natural values are used in this method in equations 3 and 4.)
The resulting base-period values of the mean (M) and standard devi-
ation ($) were regionalized by correlation with the basin and clima-
tologic parameters listed in table 2. Again, drainage area was the
only statistically significant parameter. The two graphs in figure 8
show the relation between (1) drainage area and J/, and (2) drainage
area and S. By use of these two graphs and equation 6, a flood-f requen-
METHODS USED IN FLOOD-FREQUENCY STUDIES, CALIFORNIA E27

cy curve can be computed for any site in the region. Specific equations
for computing Q5Q and Qwo are:

and

10,000

8000

21
6000

5000

4000

3000

2000

1000 20

800

600

20,000

10,000

8000

6000 ,16

5000

4000

3000

2000 8«

1000
20 50 100 200 500 1000
DRAINAGE AREA, IN SQUARE MILES
FIGURE 8. Relation of mean and standard deviation of annual peak discharges to
drainage area in the San Diego area.
770-401 O - 65 - 5
E28 FLOOD HYDROLOGY

Values of Q50 and QWo were computed from curves given in figure 8
and by application of equations 14 and 15, respectively, for drainage
areas corresponding in size to those of the 18 gaging stations. These
values are listed in column 10 of tables 3 and 4, respectively, for later
comparison.
PEABSON TYPE III DISTBIBUTION

A fifth flood-frequency analysis was made for the San Diego area
streams by fitting a Pearson type III distribution to the peak-
discharge data. The computation and regionalization of the base-
period mean (M) and standard deviation ($) for this study are
identical with those for the extreme-value probability analysis. Con-
sequently, the graphs in figure 8 were applicable to this analysis. The
next step was to determine the coefficient of skew for each station.
This was done by first using equation 7 to compute the annual
peak discharges needed to complete the 55-year array at each short-
term station and then using equation 8 to compute the coefficient of
skew (g) for each station. The individual values of g ranged from
2.39 to 5.35 but showed no correlation with basin or climatologic
parameters. Consequently, the average value of g, 3.82, was assumed
to be its regional value. A flood-frequency curve can be computed for
any site in the region, gaged or ungaged, by first entering figure 8
with drainage area to obtain M and /S and then using equation 9 and
table 1 with the regional skew of 3.82.
Values of Q50 and Q100 were computed by use of the relations and
table just mentioned for drainage areas corresponding in size to those
of the 18 gaging stations. These values have been entered in column
12 of tables 3 and 4, respectively, for later comparison with the results
obtained by the other methods of analysis.
GAMMA DISTBIBUTION

The gamma distribution was also used to analyze the flood magni-
tude-frequency relation for streams in the San Diego area. A pre-
requisite to the use of this method was a determination of the
base-period arithmetic mean (M) and geometric mean (Mg) for each
of the 18 stations. Values of M were available from the previous use
of the Gumbel-method analysis, and the logarithms of Mg were
available from the previous use of the logarithmic normal distribution.
The logarithms of Mg were then converted to natural values. The
logarithm of the ratio M/Mg was next computed for each station, and
the curve given in figure 2 was then used to obtain corresponding
values of C. The next step was to regionalize the values of M and C.
This had been done for M in the Gumbel method of analysis and the
results are found in figure 8. The individual values of C ranged from
METHODS USED IN FLOOD-FREQUENCY STUDIES, CALIFORNIA E29

0.31 to 0.89, but showed no correlation with basin or climatologic

parameters. Consequently, the average value of £7, 0.43, was assumed
to be its regional value. The regional values of M and O can be used
with a table of chi-square to compute a flood-frequency curve for any
site in the region, whether gaged or ungaged.
Values of Q$Q and QWQ were computed by use of curves given in
figure 8, a regional value of O of 0.42, and a chi-square table, for
drainage areas corresponding in size to those of the 18 gaging stations.
These values are listed in column 14 of tables 3 and 4, respectively, for
later comparison with the results obtained by the other five methods of
analysis.
ANALYSIS OF FLOOD FREQUENCY IN NORTH COASTAL
CALIFORNIA

A study of flood-frequency relations for coastal streams in northern

California was recently made by the U.S. Geological Survey (Eantz,
1964, p. 60-73). According to Fenneman (1931), the region drained
by these streams lies in two physiographic sections the northern
California Coast Ranges and the Klamath Mountains. Except for
a narrow coastal plain, the region is very mountainous with many
peaks above an altitude of 6,000 feet. The precipitation is distinctly
seasonal and about 75 percent of it occurs during the 5 months, Novem-
ber through March. Snow falls in moderate amounts at altitudes
above 2,000 feet, but only at altitudes above 4,000 feet does snow remain
on the ground for appreciably long periods. Average annual precipi-
tation increases from east to west, ranging from a low of 10 inches in
the Shasta Eiver valley to a high of 120 inches in the upper Smith
River basin. The region as a whole is very humid; and of the 27
basins investigated, only 5 had an average annual precipitation of
less than 50 inches. Drainage areas at the study sites ranged from
6 to 12,000 square miles.
The locations of the 27 gaging stations used in the study are shown
in plate 2. The numbers identifying these stations on the map and in
the tables and figures that follow are those listed in the Geological
Survey study. Figure 9 shows the period of annual peak-discharge
record at each of the stations, and table 5 lists the two hydrologic
parameters drainage area and mean annual precipitation that were
found to have a significant effect on the flood magnitude-frequency
relation. In the Geological Survey study this relation was analyzed
by the index-flood method. That analysis is summarized in the pages
that follow, and in addition, analyses were made by five other meth-
ods multiple correlation, logarithmic normal distribution, extreme-
value probability distribution, Pearson type III distribution, and the
 CO
Period of record o
Drainage
-> o o o o o Gaging station area Station
-H CM co ^- m io (sq mi)
n CTI o> CTI CTI ff>

Coast Ranges

I _ IMHH East Fork Russian River near Calpella 93 4615

Middle Fork Eel River below Black Butte River near Covelo 367 4730

Eel River below Dos Rios 1481 4740

North Fork Eel River near Mina 251 4745

m IM South Fork Eel River near Branscomb 43.9 4755

IHBH South Fork Eel River near Miranda 537 4765

Eel River at Scotia 3113 4770

__ Van Duzen River near Dinsmores 80.2 4775

1^ B Van Duzen River near Bridgeville 214 4785

Yager Creek near Carlotta 127 4790

Jacoby Creek near Freshwater 6.07 4800

Mad River near Forest Glen 144 4805

Mad River near Arcata 485 4810

Redwood Creek near Blue Lake 67.5 4815

Redwood Creek at Orick 278 4825

METHODS USED IN FLOOD-FREQUENCY STUDIES, CALIFORNIA E31

c
o if) If) if) u nooif)oof)oif)
tv o> CM u niorvOoooocMCM
(0 i-H i-l CM C MCMCMCMCMrorororo
co If) IT) If) u DiDiDLniDiniDiDin

DO tv CM IO Iorvooooioof)^
(0 ra g If) Id <* C Nl
I-H CO f^ O^ ^ O O^ I""1
C 10 10 rv r vo^rooooOrtCMio j"
(0 o !n - CM CM t
^
Q 4
^

t
i
i
c
\
^ t
b. 4
1
<
C (
k. 0)
(
in 0) O *
C > in
c (u I
o 5 (
1
<n | £ S E ^ « g 5
Gagist:ng O c o «j £ O i
^
£ _
4)
C
(5
ja
»
-WC(OC £ p £ « 1C ?Q i
i
ra S o « i
2S?E>Sio ff'«w
I^ 1*" ro Eo : ^oa^'-o^.Stt) i
>- U- C/5 1J ° ?
» Q)
>
^i
^
tl
« "
fl)-^
«
^ «
*" k. +J - c
5. TO W H -\)cc^-cc>-Ec
8 8 c c 8 £ | g *
<" k. k.k.<U|-k.5W^ 1
SJ >. >
^55g^^|^5
1 .1 = ceircc^oa:.cO£
^
to
(5
^
t;
t>
0
p
i
^cc'x'ScE^-- i
0961
£
C/5
0
C/5
(5
C/5
z 'z 'c w
h -Hl-IWI-yWW
o 'ki J2 o c
i
1 1 1I 1
ji
c
0961 <b
c
o
o
1 1
i
«
0*6 1 p
of
irec
od

q)
OE6I
1 1 0
E
B
C
1

|
O.
M t
e
0261

.
DTfiT
E32 FLOOD HYDROLOGY

gamma distribution. In differentiating regional characteristics of

[Link] in the index-flood method and in all subsequent analyses of the
data for north coastal California, each of the two physiographic
sections the Coast Ranges and the Klamath Mountains was con-
sidered separately as a hydrologic entity.
TABLE 5. Summary of pertinent hydrologic parameters for basins in north
coastal California
[Asterisk indicates long-term station]

Mean
annual
Index Gaging station Drainage area basin wide
No. (sq mi) precip-
itation
(inches)

Coast Ranges

*4615 East Fork Russian River near Calpella__________ 93. 0 40

4730 Middle Fork Eel River below Black Butte River,
near Covelo_______.______________________ 367 60
4740 Eel River below Dos Rios____________________ 1,481 53
4745 North Fork Eel River near Mina_____________ 251 58
*4755 South Fork Eel River near Branscomb_________ 43. 9 79
*4765 South Fork Eel River near Miranda___________ 537 70
*4770 Eel River at Scotia_____________.______ 3,113 59
4775 Van Duzen River near Dinsmores.______________ 80. 2 74
*4785 Van Duzen River near Bridgeville_____________ 214 72
4790 Yager Creek near Carlotta__________--._-____ 127 60
4800 Jacoby Creek near Freshwater._________________ 6. 07 54
4805 Mad River near Forest Glen_________________ 144 60
4810 Mad River near Arcata_______________--------_ 485 64
4815 Redwood Creek near Blue Lake_______________ 67. 5 80
4825 Redwood Creek at Orick_______________ 278 80

Elamath Mountains

*5175 Shasta River near Yreka_________________ » 657 19

*5195 Scott River near Fort Jones________________ 662 33
*5225 Salmon River at Somesbar______________-_---_- 746 57
*5255 Trinity River at Lewiston________________ 726 59
5260 Trinity River near Douglas City_____________ 1,017 56
*5270 Trinity River near Burnt Ranch._______________ 1, 438 57
5285 Hayfork Creek near Hyampom_________________ 379 43
5290 South Fork Trinity River near Salyer_______ _ 899 50
*5300 Trinity River near Hoopa__________________ 2,846 55
*5305 Klamath River near Klamath._______________ 12, 100 42
5320 South Fork Smith River near Crescent City__ __ 295 116
*5325 Smith River near Crescent City_______________ 613 111

1 Actual drainage area above gage is 796 sq mi, but 139 sq mi above Dwinnell Reservoir is noncontributing.
METHODS USED IN FLOOD-FREQUENCY STUDIES, CALIFORNIA E33

INDEX-FLOOD METHOD

In analyzing the annual flood data by the index-flood method, Rantz

(1964) used a 28-year base period, 1932-59. Of the 27 gaging stations
studied, only 13 had records with 10 or more years within the base
period. Annual peak discharges needed to complete the discharge
array for the base period for these 13 stations were obtained by graphi-
cal correlation. The discharges at each of the stations were then ranked
in order of magnitude, and the recurrence interval for each observed
peak was computed by applying equation 1. A flood-frequency curve
was next drawn on extreme-value probability graph paper for each
of the 13 stations by first plotting annual peak discharge against re-
currence interval and then fitting a smooth curve by eye to the plotted
points.
It was possible to extrapolate the flood-frequency curves beyond the
base period with considerable confidence because of the availability of
historical records, both qualitative and quantitative, of major floods
that occurred in years prior to 1932. For Klamath Eiver at Klamath
(sta. 5305), the magnitudes of all major flood peaks in the past 106
years are known and were used in the construction of the flood-
frequency curve (fig. 10). For other stations, where it was known
only that the flood peaks of 1956 were roughly equivalent to those of
1862 and greater than any other since at least 1854, the magnitude of
500,000

450,000 EXPLANATION

Peak discharges that oc-

400,000 curred within the base
period, 1932-59
O
Peak discharges that oc-
curred prior to the base
period

y 250,000

? 200,000

3 4 5 6 8 10 20
RECURRENCE INTERVAL, IN YEARS

FIGURE 10. Flood-frequency curve for Klamath River at Klamath (sta. 5305).
E34 FLOOD HYDROLOGY

the 1956 peak was plotted at both 107 years and 53.5 years to indicate
that this magnitude represented both the highest and second highest
discharges in 106 years. The well-defined part of the flood-frequency
curve for each station, generally a straight line or gentle curve, was
extrapolated to the 100-year recurrence interval, with the provision
that the extrapolation pass through one of these two plotted points
or pass between them. An example of this extrapolation is the flood-
frequency curve for Eel Eiver at Scotia (sta. 4770), shown in fig. 11.
The slopes of the station flood-frequency curves were next tested
for homogeneity. The results of this test are better understood in the
'light of the explanatory remarks that follow. The slope of the flood-
frequency curve for northern California streams is influenced pri-
marily by the difference in severity between the storms that cause the
milder floods, such as the mean annual, flood, and the storms that cause
the infrequent major floods. The greater the disparity between these
two types of storm, the greater the ratio of major flood peak to the
mean annual flood peak and, therefore, the steeper the slope of the
flood-frequency curve. Furthermore, it is almost axiomatic that the
more humid the area, the less variability there is in the precipitation.
Consequently, the areas closest to the coast, since they in general have
the greatest precipitation, would be expected to have flood-frequency
curves that show the flattest slope. Infiltration capacity has a rela-
tively small effect on the peak discharge during major floods because
these floods are generally associated with rains that last for many days,
and consequently, the ground becomes well saturated and the infiltrat-
ing rain amounts to only a small percentage of the storm precipitation.
Altitude may also be a factor because during these prolonged major
storms there is generally some snowmelt which augments the runoff
directly attributable to rainfall. Thus, the flood-frequency curves
600,000
I I nil i i I I I I
EXPLANATION

Peak discharges that oc-

curred within the base
period, 1932-59
O
Peak discharges that oc-
curred prior to the base
O 300,000 period

5 200,000

3 4 5 6 8 10 20

RECURRENCE INTERVAL, IN YEARS

FIGURE 11. Flood-frequency curve for Bel River at Scotia (sta. 4770).
METHODS USED IN FLOOD-FREQUENCY STUDIES, CALIFORNIA. E35

for the basins of higher altitude in northwestern California tend to

have steeper slopes.
The statistical tests for homogeneity of slope of the flood-frequency
curves bore out these premises. These tests resulted in the establish-
ment of the three areas of homogeneity shown in plate 2. The dimen-
sions flood-frequency curves for the three subregions are plotted in
figure 12. Subregion 1 has the flattest station flood-frequency curves;
lying closest to the ocean, it is the most humid area and has the lowest
altitude. Its dimensionless flood-frequency curve is based on the eight
gaging stations in subregion 1 that had been in operation for at least
10 years. The slope of the dimensionless flood-frequency curve for
subregion 2 is steeper than that of the flood-frequency curve for sub-
region 1 owing «& the generally more variable storm precipitation
and higher altitudes found in subregion 2. Only three stations have
10 or more years of record in subregion 2, and consequently the flood-
frequency curve representative of this subregion lacks the high degree
of confirmation obtainable from a large number of gaging stations.
The fact that the dimensionless flood-frequency curve for subregion
3 has the steepest slope of the three regional curves reflects that sub-
region 3 is the least humid of the three subregions. Only two stations
in the subregion have the requisite 10 or more years of record.
The magnitude of the mean annual flood ($2.33) was next investi-
gated. For the 13 stations with 10 or more years of peak-discharge
record, $2.33 was picked from the individual station flood-frequency
I I I TIT

6 7 8 9 10 20

RECURRENCE INTERVAL, IN YEARS

FIGURE 12. Dimensionless regional flood-frequency curves for north coastal California
(index-flood method).
E36 FLOOD HYDROLOGY

curves that had been previously plotted. For the 14 stations with less
than 10 years of record, correlation procedures were used to determine
$2.33- Concurrent peak discharges for each of these short-term sta-
tions and for a nearby longer term station were correlated graphi-
cally. $2.33 for the short-term station was then determined from the
regression line, it being the discharge corresponding to $2.33 for the
longer-term station. The values of Q2.33 for both long-term and short-
term stations were then analyzed.
The magnitude of the mean annual flood in basins in northwestern
California is related primarily to the size of drainage area and to the
magnitude of the mean annual storm. Mean annual precipitation is
an excellent index of the relative magnitude of the mean annual storm
because the bulk of the annual precipitation in the region occurs during
several general storms each year, and all stations experience the same
number of general storms in any given year. Subsurface storage also
exerts a significant influence on the magnitude of the mean annual
flood. Surface storage, on the other hand, is a negligible factor in this
study because there are no sizable lakes or reservoirs that are uncon-
trolled, and streams that are seriously affected by artificially regulated
storage were excluded from the analysis. Because subsurface storage
is related to the infiltration capacity, or the permeability of the soil
and mantle rock, and because much greater permeability is associated
with the Klamath Mountains than with the Coast Ranges, it is logical
to expect the mean annual flood to differ in these two physiographic
sections, when all other factors are equal.
In figure 13 the mean annual floods for basins in the Coast Eanges
have been plotted in relation to the drainage area. Each point is la-
beled with (1) the number of the gaging station for identification, and
(2) the mean annual precipitation for the basin upstream from the
station. (Drainage area and mean annual basin wide precipitation
are given in table 5.) Precipitation in the Coast Ranges basins ranges
from 40 to 80 inches, and within this range no significant correlation
is apparent between mean annual flood ancTmean annual precipitation.
A straight line averaging the plotted points has the equation:
Mean annual flood (£2. 33 ) = 130J.0-91, (16)
where A is the drainage area in square miles.
The wide range in mean annual precipitation in the Klamath Moun-
tains, 19-116 inches, has a very pronounced effect on the magnitude
of the mean annual flood. The relation of $2.33 to drainage area and
mean annual precipitation in this physiographic section can be ex-
pressed by a means of a family of curves, each of which has an equa-
tion similar to equation 16. The equation for this family of curves is
METHODS USED IN FLOOD-FREQUENCY STUDIES, CALIFORNIA E37

T I I I IT
200,000
EXPLANATION

80
100,000 Upper number is gaging
80,000 station. Lower number
70,000 is mean annual basin-
60,000 wide precipitation, in
50,000 inches.
40,000
30,000

Equation of curve:
20,000 Q 2.33 =130A°»

10,000
8000
7000
6000
5000
4000

3000

2000

1000
800
700
600
500
10 10 Is* oo o o oooooo
oo «* in 10 rv oo o

DRAINAGE AREA, IN SQUARE MILES

FIGURE 13. Relation of mean annual flood to drainage area in the Coast Ranges (index-
flood method).

where C is a variable that is related to mean annual precipitation.

The relation of C to mean annual basinwide precipitation is shown
graphically in figure 14. From a comparison of the graphs in figures
13 and 14, it is concluded that for the same size of drainage area and
the same mean annual precipitation, mean annual floods are greater in
the Coast Ranges than in the Klamath Mountains. The basis for this
conclusion is the fact that the coefficient of 130 in the Coast Ranges
formula is equivalent to C corresponding to about 90 inches of mean
annual precipitation in the formula for the Klamath Mountains, yet
the precipitation in the Coast Ranges ranged from only 40 to 80 inches.
This result is not surprising in view of the fact that the Klamath
Mountains have the more permeable soil and mantle rock.
Values of the 50-year flood (Q50 ) and the 100-year flood ($100) were
computed from curves given in figures 12-14 and equation 17 for sites
whose drainage areas correspond in size and in mean annual basinwide
E38 FLOOD HYDROLOGY
300

200 -

100 -

10 -

20 30 40 50 70 100 150

MEAN ANNUAL PRECIPITATION, IN INCHES

FIGURE 14. Relation of G to mean annual basinwide precipitation in the Klamath

Mountains (index-flood method).

precipitation to those of the 13 long-term stations in the region. These

computed values have been entered in column 4 of tables 6 and 7, re-
spectively, for comparison with the results obtained by other methods.
Qso and QWo were not computed for sites similar to the 14 short-term
stations because the station records were too short for use in the other
methods of analysis that follow in this report. In column 3 of tables
6 and 7 are listed the values of Q50 and $i00, picked from the individual
station curves, such as the curves shown in figures 10 and 11.
METHODS USED IN FLOOD-FREQUENCY STUDIES, CALIFORNIA E39

+1 I '+ I I I I I I I

CO CO * Ofc; «0 10

+1 +777i+ +
"53
S
e
fs'sgl^
'3-fe'§ + 1 +1 I I I
2«|

<*
<e, 5 < a> d w ., _,
"53 fisls +1

Ti-J'ia'crf'^iH'ffl'TiJ'oo'
sssss (NtcSS^gS

-Sft + 1 ++

I++

!
03,Q

3!
O]_|
SnS
q s ia
sis

S3 ' S s a s s| 2

2SSSS
TABLE 7. Qioo for streams in north coastal California as determined from graphically derived flood-frequency curves and l>y various
methods of regional flood-frequency analysis

Index-flood Multiple Logarithmic Extreme- value Pearson Gamma

Qioo from method correlation normal probability type III distribution
graph- distribution distribution distribution
Index ically-
No. Station derived
station Percent Percent Percent Percent Percent Percent
curves Qioo differ- Qioo differ- Qioo differ- Qioo differ- Qioo differ- Qioo differ-
(cfs) (cfs) ence (cfs) ence (cfs) ence (cfs) ence (cfs) ence (cfs) ence
from from from from from from
col. 3 col. 3 col. 3 col. 3 col. 3 col. 3

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

California Coast Ranges

o
4615 22,100 23,000 +4 29,500 +33 29,400 +33 25,600 +16 26,500 +20 25,600 +16 <5
_ 44 -25 IE QAA -24 -35 qq tO 7AA -34 °
4755 20,900 11,600 15,700 13,500 13,900
47fi5 170 000 -32 -26 131,000 -23 115,000 -32 no nnn -30 111,000 q. t"1
EQO nnn 1 0
4770 530,000 +10 545,000 560,000 +6 535,000 +1 554,000 +5 507,000 -4 o
1 QA +22 +18 £j
4785 43,600 49,500 +14 59,000 +35 58,600 53,200 55,000 +26 51,500

Klamath Mountains

_ OQ -1.4A i nn -1
5175 6,000 4,280 8,420 9,700 +62 6,650 +11 7,300 5,950
5195 37,500 36,000 -4 24,300 -35 26,200 -30 19,400 -48 21,200 -43 14,500 -61
QO Kflfl net nnn -7 5 -25 is -51
5225 +16 76,400 78,600 61,800 67,400 40 500
1 00 -16 _Q -45
5255 76,700 101 000 79,800 +4 81,800 +7 64,400 70,200 41,900
1 0 -22 -12 -35 -28 -56
5270 170,000 174,000 132,000 149,000 111,000 122,000 75,000
5300 192,000 207,000 1 ,q 220,000 +15 272,000 +42 193,000 +1 210,000 -L.O 134,000 -30
-3 -5 1 A -25
5305 445,000 431,000 445,000 0 695,000 +56 423,000 461,000 335,000
10 -1.9.4 10 -41
5325 186,000 200,000 231,000 202,000 188,000 +1 204,000 +10 110,000
METHODS USED IN FLOOD-FREQUENCY STUDIES, CALIFORNIA E41

MULTIPLE CORRELATION
In analyzing the flood magnitude-frequency relation for north
coastal California by the multiple-correlation method it was neces-
sary to eliminate 14 of the 27 gaging stations from consideration
because the records for these stations were shorter than 10 years. Of
the remaining 13 stations, 5 were in the Coast Ranges and 8 were in
the Klamath Mountains. (See table 5.) The fact that so few records
were available for study severely handicapped the analysis. Because
there were only five stations in the Coast Ranges, no more than one
independent variable could be used in a linear correlation, if the
correlation were to have any significance at all. Because there were
only eight stations in the Klamath Mountains, no more than two
independent variables could be used in the correlation.
The first step in the analysis was to obtain the discharges corre-
sponding to recurrence intervals of 50 years (Q5o) and 100 years
($100) from the 13 individual station flood-frequency curves. These
curves (see, for example, figs. 12 and 13) had been constructed earlier
for the index-flood method of analysis. The values of Q5o and $100
were then correlated with the hydrologic parameters listed in table 5.
For the Coast Ranges, drainage area ( A) was by far the most signifi-
cant parameter, and it was used as the sole independent variable in
the correlation for the five stations in that physiographic region. Two
independent variables drainage area (A) and mean annual basin-
wide precipitation (P) were used in the multiple correlation for the
eight stations in the Klamath Mountains. Regression equations for
$50 and $100 were obtained as follows:
Coast Ranges:
(18)
(19)
Klamath Mountains:
$ 50 = 0.075 [Link].965 (20)
^100 =0.128J.°- 842P1-912. (21)
For later comparison, the values of Q50 and $100? computed for the
13 stations by application of these equations, have been entered in
column 6 of tables 6 and 7, respectively.
LOGARITHMIC NORMAL DISTRIBUTION
Regional flood-frequency curves for north coastal California were
computed by fitting logarithmic normal distributions to the original
base data for the 13 stations used in the previous multiple-correlation
analysis. The first step in the computation was to convert the natural
values of peak discharge to logarithms. The mean and the standard
deviation for each station array were then computed. It was neces-
E42 FLOOD HYDROLOGY

sary to adjust the computed mean and the standard deviation for those
stations whose peak-discharge records were incomplete for the 28-year
base period, 1932-59. The adjustment was accomplished by first
correlating concurrent records for each short-term station and a nearby
long-term station and then applying equations 3 and 4. At this stage
of the computations, the base-period mean (M) and standard deviation
($) were available for all 13 stations.
The values of M and /#, still in logarithmic units, were next region-
alized by correlation with the hydrologic parameters listed in table 5.
In the Coast Ranges region, there was a definite relation between M
and drainage area (fig. 15), but S was apparently independent of any
measured hydrologic parameters. It was expected that S might vary
inversely with mean annual basinwide precipitation, but no correlation
was evident. An average value of $=0.235 was therefore used for
the Coast Ranges. In the Klamath Mountains, M was found to vary
with both drainage area and mean annual basinwide precipitation. No
graph has been provided to illustrate the relation, which is expressed
by the equation:
M= -2.837 + 0.978,4 + 2.421P, (22)
where
M is the base-period mean of the logarithms of annual peak dis-
charges,
A is the logarithm of the drainage area, and
P is the logarithm of the mean annual basinwide precipitation.
CO 5-4
UJ
C3
{£.
I 5.2
O
CO
Q
^ 5.0
20.
_l 4.8
D

1 4.6
u.
O
CO 4.4
I
h-
% 4.2

3
u. 4.0
O

i 3.8

I___I
200 300 500 1000

DRAINAGE AREA, IN SQUARE MILES

FIGURE 15. .Relation of the mean of the logarithms of annual peak discharges to drainage
area in the Coast Ranges.
METHODS USED IN FLOOD-FREQUENCY STUDIES, CALIFORNIA E43

Values of S in the Klamath Mountains were closely related to mean

annual basinwide precipitation as shown graphically in figure 16. By
use of these relations, an isohyetal map of mean annual precipitation,
and equation 5, a flood-frequency curve can be computed for any site
in north coastal California.
By use of the relations just mentioned and equations 12 and 13,
values of Q50 and $100, in logarithms, were computed for drainage
areas corresponding in size and mean annual basinwide precipitation
to those of the 13 long-term gaging stations in the region. The anti-
logarithms of Q50 and Q wo have been entered in column 8 of tables
6 and 7, respectively, for later comparison.

EXTREME-VALUE PROBABILITY DISTRIBUTION1 OR GUMBEL

METHOD

A regional flood-frequency study of the north coastal region of

California was also made by fitting an extreme-value probability dis-
tribution to the base data for the 13 long-term gaging stations in the
region. In this analysis natural values of the discharges were used,
not their logarithms. The mean and the standard deviation for each
station array were computed, and these two statistics were adjusted
for those stations whose peak-discharge records were incomplete for
the 28-year base period, 1932-59. To make this adjustment, concurrent
records for each short-term station and a nearby long-term station
were first correlated, and then equations 3 and 4 were applied. (Note
that natural values, and not logarithms, are now used in equations
3 and 4.) After making the adjustment, the base-period mean (M)
and standard deviation (S) were available for all 13 stations.
The values of M and S were next regionalized by correlation with
the hydrologic parameters listed in table 5. In the Coast Eanges
region, both M and /S were related to drainage area (fig. IT). In the
Klamath Mountains both M and S were related to drainage area and
mean annual basinwide precipitation. No graphs have been provided
to illustrate the two multiple relations, which are expressed by the
equations:
J/=0.0044d°-944P2-164 (23)
S= 0.022J.°.89ipi.86i? (24)
where
M is the base-period mean of annual peak discharges,
S is the base-period standard deviation of annual peak dis-
charges,
A is the drainage area, and
P is the mean annual basinwide precipitation.
E44 FLOOD HYDROLOGY
T" I

|5255

Q 2

z o

20 30 40 60 80

MEAN ANNUAL BASINWIDE PRECIPITATION, IN INCHES

FIGURE 16. Relation of the standard deviation of the logarithms of annual peak discharges
to mean annual basinwide precipitation in the Klamath Mountains.

300,000

200,000

100,000 4770 -\

80,000

60,000
50,000
40,000
1°
* O 30,000

!2 o- 20,000

0 g 10,000
Q _
o: z 8000
Q
< 6000
W 5000

30 50 100 200 300 500 1000 2000 3000 500C

DRAINAGE AREA, IN SQUARE MILES

FIGURE 17. Relation of the mean and the standard deviation of annual peak discharges to
drainage area in the Coast Ranges.
METHODS USED IN FLOOD-FREQUENCY STUDIES, CALIFORNIA E45

By use of the relations expressed in figure 17 and in equations 23, 24,

and 6, a flood-frequency curve can be computed for any site in north
coastal California.
By use of the foregoing relations and equations 14 and 15, values
of Qso and Q1W were computed for drainage areas corresponding in
size and mean annual basin wide precipitation to those of the 13 long-
term gaging stations in the region. These values are listed in column
10 of tables 6 and 7, respectively, for later comparison.
PEARSON TYPE III DISTRIBUTION
A fifth flood-frequency analysis was made for the north coastal
region of California by fitting a Pearson type III distribution to the
peak-discharge data for the 13 long-term gaging stations in the region.
The computation and regionalization of the base period (M) and
standard deviation (8) for this study are identical with those for the
Gumbel method of analysis. Consequently, the graphs for the Coast
Ranges (fig. 17) and equations 23 and 24 for the Klamath Mountains
were applicable for this analysis. The next step was to determine the
coefficient of skew for each station. This was done by first using
equation 7 to compute the annual peak discharges needed to complete
the 28-year array at each short-term station, and then applying equa-
tion 8 to compute the coefficient of skew (g) for each station. For the
streams in the Coast Ranges, the individual values of g ranged from
0.84 to 2.04; for the streams in the Klamath Mountains they ranged
from 1.07 to 3.00. In neither region did g correlate with the selected
hydrologic parameters. Consequently, the average value of g in each
region was assumed to be the regional value of that statistical param-
eter. Under this assumption the regional value of g for the Coast
Ranges is 1.46, and for the Klamath Mountains, 1.83.
By use of the regional relations for M and S as indicated by curves
given in figure 17 and equations 23 and 24, the appropriate regional
value of <7, table 1, and equation 9, a flood-frequency curve can be
computed for any site in the region, whether gaged or ungaged. Val-
ues of Qso and Q100 were computed by use of the relations and table
just mentioned for drainage areas corresponding in size and mean
annual basinwide precipitation to those of the 13 long-term gaging
stations in the region. These values have been entered in column 12
of tables 6 and 7, respectively, for later comparison with the results
obtained by the other methods of analysis.
GAMMA DISTRIBUTION
The gamma distribution was also used to analyze the flood magni-
tude-frequency relation for streams in north coastal California. A
E46 FLOOD HYDROLOGY

prerequisite to the use of this method was a determination of the

base-period arithmetic mean (M) and geometric mean (Mg) for each
of the 13 long-term stations. Values of M were available from the
previous use of the Gumbel-method analysis, and the logarithms of
Mg were available from the previous use of the logarithmic normal
distribution. The logarithms of Mg were then converted to natural
values. The logarithm of the ratio M/Mg was next computed for
each station, and the curve given in figure 2 was then used to obtain
corresponding values of C. The next step was to regionalize the
values of M and C. This had been done for M in the Gumbel method
of analysis and the results are found in figure 17 for the Coast Ranges
and in equation 23 for the Klamath Mountains. Individual values
of C for the five stations in the Coast Eanges ranged from 2.61 to
7.35, but showed no correlation with hydrologic parameters. The
median value of £7, 3.40, was therefore assumed to be the regional
value of C for this area. For streams in the Klamath Mountains,
the individual values of C correlated with mean annual basinwide
precipitation (fig. 18).
The regional values of M and C can be used with a table of chi-
square to compute a flood-frequency curve for any site in the region,
whether gaged or ungaged. Values of Q50 and Q100 were computed by
use of the curves given in figures 17 and 18, equation 23, a regional C
value of 3.40 for the Coast Eanges, and a chi-square table for drain-
age areas corresponding in size and mean annual basinwide precipita-

5225
-5255
5270

0 10 20 30 40 50 60 70 80 90 100 110 12

MEAN ANNUAL BASINWIDE PRECIPITATION, IN INCHES

FIGURE 18. .Relation of C to mean annual basinwide precipitation in the Klamath

Mountains (gamma distribution).
METHODS USED IN FLOOD-FREQUENCY STUDIES, CALIFORNIA E47

tion to those of the 13 long-term gaging stations in the region. These

values are listed in column 14 of tables 6 and 7, respectively, for later
comparison with the results obtained by the other five methods of
analysis.
DISCUSSION OF RESULTS OF THE ANALYSES

Because of the short period covered by most records of peak dis-

charge in California, the true values of the 50-year flood (Q50 ) and the
100-year flood ( Qwo ) are not known. Although the peak discharges of
the greatest and second greatest floods in the past 100 years can often
be deduced within fairly close limits by the use of qualitative historical
information, utilization of this information provides, at best, only a
single 100-year sample of flood events. This sample may or may not
be representative of long-term flood activity. Furthermore, in each of
the two California regions studied, the greatest flood in the past 100
years at each station was caused by a single meteorological event.
This was generally true, also, of the second greatest flood in the past
100 years. Thus, all stations experienced the same major flood events
and, therefore, regionalization of the data has had little effect in re-
ducing the sampling error that results from having a nonrepresentative
sample of floods at a gaging station. We can do nothing about the
sampling error, however, and must proceed on the assumption that our
sampling of flood events adequately represents the norm of flood ac-
tivity and that our values of Q50 and Qwo for each station, as deduced
with the help of historic information, have only random scatter from
their true values.
Tables 3 and 4 present a summary of values of Q50 and $100, respec-
tively, for streams in the San [Link], obtained by various methods
of flood-frequency analysis. Similar tabulations for north coastal
California are given in tables 6 and 7. Because all methods attempt to
fit some relation to the single flood array that is available for each
station in a region, it is presumed that the method that fits the observed
data best is the most reliable one for use. In this study the values of
Q50 and Qwo obtained from the graphically-derived individual station
flood-frequency curves (col. 3 in tables 3,4,6, and 7) are considered the
standard for judging the reliability of the various methods of analysis.
Note that these values of Q50 and Q100 in column 3 are more than mere
extrapolations; they have been determined from a careful assessment
of qualitative historical information. Tables 3, 4, 6, and 7 also list
the differences, in percent, between the values of Q50 and Qwo as deter-
mined by each of the six methods of regional flood-frequency analysis,
and the values of Q50 and Qwo in column 3. These differences are sum-
marized in table 8. Ideally, split-sample testing should have been
TABLE 8. Summary of differences between discharges determined from graphically derived flood-frequency curves and discharges
computed by various methods of regional flood-frequency analysis

Number of sites with differences in indicated range

Quo Qioo
Range of differences
(percent)
Index- Logarith- Extreme- Pearson Index- Logarith- Extreme- Pearson
flood Multiple mic normal value type III Gamma flood Multiple mic normal value type III Gamma
method correlation distribution probability distribution distribution method correlation distribution probability distribution distribution
distribution distribution

San Diego area

+ ,151-222). _ _ 5
+ ,126-150) .-- 1 0
+ (101-125) ____ -_ 1 r 2
+ 81-100)- .-- 1 3
+ 61-80)..- _ .- 4 1 1
+ 2 1 3 1 1 4
+ 21-40).. __ .- 2 2 2 1 4 1 1 1
+ (1-20).. .. 5 4 3 1 1 2 7 2 1
0 0 3 0 0 0 0 2 0
(1-20). ________ 5 5 3 2 6 2 6 4 1 4 1
(21-40). ------- 4 3 5 8 6 4 3 6 4 5
9 2 9 8 8 8
(61-80). . 1 1 3 4
Total ______ 18 18 18 18 18 18 18 18 18 18 18 18
METHODS USED IN FLOOD-FREQUENCY STUDIES, CALIFORNIA E49

COOIN I CO

<NCO I COOCO I (N

COi-4<N I CO

FT
+++
E50 FLOOD HYDROLOGY

used in comparing computed and recorded values of Q50 and Q100. In

split-sample testing, half the stations would be used to derive the rela-
tions necessary to compute Q50 and Qwo, and the remaining stations
would be used to compare computed and recorded values. Unfor-
tunately, there were too few stations to permit split-sample testing,
and consequently all stations were used both in the derivation of the
relations and in the comparison of Q50 and Qw
It should be pointed out that the percentage differences listed in
table 8 do not, in themselves, give a complete picture of the relative
reliability of the various methods; another factor is the number of
degrees of freedom lost in each analysis. The details of this loss will
not be discussed. Suffice it to say that the index-flood method, as used
in the analysis for north coastal California, has more lost degrees of
freedom than the other methods because of the numerous areal sub-
divisions that were established for the region. Consequently, the
results obtained in north coastal California by this method have some-
what less statistical significance than those obtained by the other five
methods.
Table 8 shows that all methods of analysis gave better results for
north coastal California than for the San Diego area. This is to be
expected because the San Diego area, being generally subhumid, has
highly variable streamflow, and it is almost axiomatic that the more
variable the streamflow, the more difficult it is to make a reliable deter-
mination of the flood magnitude-frequency relation. The difficulty is
compounded when the period of record is short. A comparative meas-
ure of the variability of an array of data is the coefficient of variation,
which is defined as the ratio of the standard deviation of the array
to its mean, or S/M. The greater variability of the San Diego area
streams is shown by the fact that the coefficient of variation for these
streams ranged from 1.5 to 2.8, whereas in the north coastal area the
range was from OA to 1.1.
Table 8 shows that for the San Diego area the two empirical meth-
ods the index-flood method and multiple correlation gave results
that are in much closer agreement with the "standard" values of Q50
and $100 than are the values obtained by any of the four methods using
statistical distributions. This agreement is not surprising because the
two empirical methods are the only ones that made use of the qualita-
tive historical data that were available, and it is on these data that the
standard values of Q50 and Qwo are based. In effect, the period of
observed data for the two empirical methods was 100 years or more, as
opposed to the much shorter period of observed data used with the
other four methods. For the north coastal area, the difference in the
quality of results between the two empirical methods and the methods
using statistical distributions is less pronounced. The gamma distri-
METHODS USED IN FLOOD-FREQUENCY STUDIES, CALIFORNIA E51

bution gave the poorest results, but of the other five methods, no single
method was outstandingly superior. The implication is that in a
humid area where all streams are subject to the same general storms,
most methods of analysis will give satisfactory results.
INDEX-FLOOD AND MULTIPLE-CORRELATION METHODS

The multiple-correlation method is superior to the index-flood

method, particularly in subhumid areas. Not only did the multiple-
correlation method give generally better results in this study, as in-
dicated in table 8, but it has a more rational basis than the index-flood
method. The weakness of the index-flood method is that it assumes
that the slopes of the flood-frequency curves for the streams in a given
region vary randomly from some median value. The Geological Sur-
very test for homogeneity of slope in a region has some serious short-
comings and often indicates homogeneity of slope where none exists
(Benson, 1962, p. 21). Actually, there are many physiographic and
climatologic factors that affect the slope of the flood-frequency curve,
and it is often a matter of fortuity that the curves for all streams in
a region have similar slopes. In several previous studies the delinea-
tion of the boundaries of regions of homogeneous slope, based on the
homogeneity test, has resulted in maps that resemble political gerry-
manders. The index-flood method can be used to advantage, however,
for a region where there are insufficient stations with 10 or more years
of record to permit a significant multiple correlation to be made. In
addition, those records that are shorter than 10 years are helpful in
defining the regional relation for the mean annual flood ($2.33)-
STATISTICAL DISTRIBUTIONS

We turn our attention now to the four methods of analysis that are
based on statistical distributions logarithmic normal, extreme-value
probability, Pearson type III, and gamma. As already mentioned,
these methods are empirical in the sense that one of the four distribu-
tions must be arbitrarily selected for use. However, once a distribu-
tion has been selected, the analysis becomes strictly objective. Because
these methods did not make use of the available historical information,
the results obtained from their application were not as satisfactory as
those obtained by use of either the index-flood or multiple-correlation
methods. This was particularly true in the subhumid San Diego area.
If the high-water ends of these distributions had been later modified
on the basis of the historical information, as is often done, the results
would have been more satisfactory, but there would then have been
little point in using one of the theoretical distributions. The all-im-
portant high-water end of the flood magnitude-frequency relation
E52 FLOOD HYDROLOGY

would no longer be controlled by the statistics of the selected distribu-

tion, but would depend on the judgment of the analyst in interpreting
the historical information and assigning probabilities to the peak dis-
charges. Table 8 shows that the logarithmic normal distribution
gave results that were generally high; the other three distributions
gave results that tended to be low. The logarithmic normal distribu-
tion also had a greater spread of differences from the standard values
of Q5o and Q10o than did the other three distributions. It is not to be
inferred, however, that this is a general rule. It is of interest that the
flood-frequency handbook used by the U.S. Army, Corps of Engineers,
which recommends the logarithmic normal distribution method, cau-
tions against its use in subhumid areas (Beard, 1962, p. 22).
Tables 3, 4, 6, and 7 show that the four statistical distributions,
when applied to the same peak-discharge data, gave widely differing
results for individual stations. These distributions have all been ex-
tensively used elsewhere and each has the support of reputable stat-
istician-hydrologists as being the distribution that best describes the
occurrence of flood events. Obviously all methods cannot be best, and
the inescapable conclusion is either that the true distribution that de-
scribes the occurrence of flood events is not known or that no single
distribution is best for all of the many widely varying hydrologic con-
ditions found in a country as large as the United States. In the United
States most of the long-term discharge records are for large streams
draining humid areas, and it is these records that have been used in
the past to test the adequacy of the various distributions. In this
study we have seen that in a humid area all methods give results that
are generally satisfactory. Therefore, it is not surprising that each
distribution has strong proponents among hydrologists. Examina-
tion of the various distributions shows that within the range of prob-
abilities usually used in flood-frequency studies, the extreme-value
probability distribution may be considered a special case of the Pear-
son type III distribution, one in which the coefficient of skew is 1.14.
The gamma distribution is likewise a special case of the Pearson type
III distribution, one in which the locus parameter is zero.
From the preceding paragraph, it would appear that the most de-
sirable distribution for use would be the one that is most flexible and
can therefore fit the peak-flow data closest. Of the four distributions
studied, the Pearson type III distribution is the least rigid because
the Pearson type III distribution uses three statistical parameters,
whereas the other three distributions use two parameters, with the third
METHODS USED IN FLOOD-FREQUENCY STUDIES, CALIFORNIA E53

parameter either constant or determined by one or both of the others.

We know too little of the laws governing the occurrence of flood events
to hazard a guess concerning the value of the third parameter, and it
therefore seems reasonable to use a general or flexible distribution,
such as the Pearson type III, rather than some rigid distribution,
such as one of the other three, which has a built-in restriction on the
value of the third parameter.
If we exclude the two empirical methods and confine our comparison
of results to the four statistical distributions, we should concern our-
selves only with the values of Q50 and not $100- In this study, the values
obtained for Q50 result primarily from the distribution of peak dis-
charges during the base periods; they are not seriously affected by the
consideration of qualitative historical data, as are the values of $100-
Examination of the percent differences for Q50 in table 8 shows that
the Pearson type III distribution fitted the array of observed data
better than any of the other three statistical distributions.
The principal objection to the use of Pearson type III distribution
has been the fact that the skew statistic, which is used in the distri-
bution, has a large standard error when there are relatively few items
in the array of peak discharges. It has been claimed that under these
circumstances the coefficient of skew has little or no significance.
Whether or not the coefficient of skew computed for an isolated station
is statistically significant may be debatable, but it seems reasonable to
accept as significant a regionalized value of the coefficient based on
numerous station records, even though the records may not be entirely
independent.
Because regionalization of the statistics for the individual station
arrays might have obscured the relative abilities of the four distribu-
tions to fit the station data, the individual statistics for each station
array were used to compute individual stition flood-frequency curves.
The computed value of Q50 from each of these curves was compared
with the "standard" values of Q50 listed in column 3 of tables 3, 4, 6,
and 7. The comparison is summarized in table 9. The Pearson type
III distribution gave results that were superior to those obtained by
use of the other three distributions.
Although this comparison study of flood-frequency methods was
based on small samples 18 stations in southwest California and 13
stations in northwest California the results and conclusions appear
to be meaningful because they can be explained rationally.
E54 FLOOD HYDROLOGY

TABLE 9. Summary of differences between Qso determined from graphically derived

flood-frequency curves and Q50 computed from various distributions using statistical
parameters from individual station arrays

Number of sites with differences in indicated range

Range of differences (percent)
Logarithmic Extreme-value Pearson type Gamma
normal dis- probability III distribu- distribution
tribution distribution tion

San Diego area

+ (121-373). __ __. __ _______ 3

+ (101-120)___-_____ _ _______ 2 1
+ (81-100)________ ____________ 0 0
+ (61-80)_____-----___________ 2 1 0
+ (41-60)-____----____________ 4 0 0 1
+ (21-40). _.__-._________.____ 1 0 0 0
+ (1-20)_____ _________________ 0 0 1 0
0____-__- _ _._. 0 0 0 0
-(1-20)______ ________________ 4 1 6 2
-(21-40)_____________________ 0 8 7 6
-(41-60).... _ ______ _ _____ 1 6 1 7
-(61-84). ____________________ 1 2 2 2
Total... ___ _ ____ __ 18 18 18 18

North coastal area

+ (31-34)________. ____________ 1
+ (21-30)_____._______________ 0
+ (11-20). ____________________ 3
+ (1-10)__. ___________________ 5 1 8
0__ _________________________ 0 0 0 0
-(1-10)...... __....__..._..._ 3 11 4 5
-(11-20)... __________________ 1 1 0 7
-(21-26)_--_------.__________ 1 1
Total._________ ______ _ 13 13 13 13

SUMMARY AND CONCLUSIONS

Regional flood-frequency studies were made by six different meth-

ods for the subhumid San Diego area in southwestern California and
for the humid coastal area in northwestern California. A 55-year
base period was used for 18 peak-discharge records in the San Diego
area; a 28-year base period was used for 13 peak discharge records
in north coastal California. All flood-frequency curves were extrap-
olated to the discharges corresponding to the 50-year recurrence in-
terval ($50) and to the 100-year recurrence interval ($100) Two
of the six methods of analysis the index-flood and multiple-corre-
lation methods are to some extent empirical and permit the analyst
METHODS USED IN FLOOD-FREQUENCY STUDIES, CALIFORNIA E55

to use whatever qualitative or historical information is available for

extrapolating the flood-frequency relation. The other four methods
are based on statistical distributions the logarithmic normal, ex-
treme-value probability, Pearson type III, and gamma distributions.
These four methods are empirical in the sense that one of the distri-
butions is arbitrarily selected for use; but once selected, the analysis
becomes strictly objective and the extrapolation is automatically made
from a determination of the statistics of the array of peak discharges.
There is a lack of agreement among hydrologists as to which method
or which distribution is the most reliable for defining the flood magni-
tude-frequency relation at a site. The purpose of this study was to
compare the results obtained by applying each of the six methods of
analysis to the basic data and to appraise the relative reliability of the
methods. The magnitude of the greatest and second greatest flood
peaks in the past 100 years in the two study areas were generally
known within reasonable limits of accuracy, and with this knowledge
it was possible to derive values of Q5n and Qwo for each gaging station
that are representative of flood activity during the past century.
These values of Q50 and $10o were used as standards for the com-
parison and appraisal of the methods of analysis.
It was concluded that all methods of analysis give better results
in a humid region, such as north coastal California, than in a sub-
humid region, such as the San Diego area, because streamflow is
usually less variable in a humid region.
A decision as to the preferred method of flood-frequency analysis
depends on whether or not historical data, either quantitative or
qualitative, are available concerning the magnitude of floods that oc-
curred in the years prior to the collection of streamflow records. If
such information is available, the empirical methods the index-flood
and multiple-correlation methods are superior to any of the four
methods using statistical distributions owing to the fact that only the
empirical methods use historical data, and use of these data gives a
longer time base for the analysis. In this study, our knowledge of
the magnitude of the greatest and second greatest flood peaks in the
past 100 years extended the period of observed data, in effect, to 100
years or more for the empirical methods. This is a much longer period
than the base periods that were used with the other four methods of
analysis. Of the two empirical methods, the multiple-regression
method is superior because it has a much more rational basis than the
index-flood method and in addition gives better results.
Where the peak-discharge data are limited entirely to the period
during which streamflow records were collected (no historical data
available), a method based on the distribution of the array of peak
flows is preferred because of its greater objectivity. Of the four dis-
E56 FLOOD HYDROLOGY

tributions tested, the Pearson type III is the most desirable. It is

more flexible than the other three and will generally fit the peak-dis-
charge data best because the Pearson type III distribution is a three-
parameter distribution with no built-in restriction on the value of the
third, or skew, parameter. The other distributions are two-parameter
distributions with an implied constant value for the third parameter.
Objection to the Pearson type III distribution has been based on the
large standard error of the coefficient of skew, but this objection can be
overcome by using a regional value of the coefficient based on numerous
station records.
Although this comparison study of flood-frequency methods was
based on small samples from only one part of the United States, the
results and conclusions appear to be meaningful because they can be
explained rationally.
REFERENCES CITED
Alexander, G. N., 1962, The use of the gamma distribution in estimating regu-
lated output from storages: Australian Inst. of Engineers, Civil Engineers
Trans., v. CE 4, no. 1, p. 29-34.
Arkin, Herbert and Colton, R. R., 1950, Tables for statisticians: New York,
Barnes & Noble, 152 p.
Beard, L. R., 1962, Statistical methods in hydrology: U.S. Army Corps of Engi-
neers, Civil Works Investigations Project CW-151, 62 p.
Benson, M. A., 1962, Evolution of methods for evaluating the occurrence of floods:
U.S. Geol. Survey Water-Supply Paper 1580-A, 30 p.
California Department of Water Resources, 1963, Flood hazard investigation,
San Diego County: Bull. 112, app. A, 78 p.
Dalrymple, Tate, 1960, Flood-frequency analyses: U.S. Geol. Survey Water-
Supply Paper 1543-A, 80 p.
Elderton, W. P., 1953, Frequency curves and correlation; Washington, Harren
Press, 272 p.
Ezekiel, Mordecai and Fox, K. A., 1959, Methods of correlation and regression
analysis: New York, John Wiley & Sons, 548 p.
Fenneman, N. M.. 1931, Physiography of western United States: New York,
McGraw-Hill Book Co., 534 p.
1938, Physiography of eastern United States: New York, McGraw-Hill
Book Co., 714 p.
Gumbel, E. J., 1958, Statistics of extremes: New York, Columbia Univ. Press,
375 p.
Jarvis, C. S., and others, 1936, Floods in the United States, magnitude and fre-
quency: U.S. Geol. Survey Water-Supply Paper 771, 497 p.
Potter, W. D., 1961, Peak rates of runoff from small watersheds: U.S. Bur.
Public Roads, Hydraulic Design Series No. 2, 35 p.
Rantz, S. E., 1964, Surface-water hydrology of coastal basins of ^northern Cali-
fornia : U.S. Geol. Survey Water-Supply Paper 1758, 77 p.
Thorn, H. C. S., 1958, A note on the gamma distribution: Monthly Weather Rev.,
v. 86, no. 4, p. 117-122.

o
Flood Hydrology
GEOLOGICAL SURVEY WATER-SUPPLY PAPER 1580

This volume was published as

separate chapters A-E
UNITED STATES DEPARTMENT OF THE INTERIOR

STEWART L. UDALL, Secretary

GEOLOGICAL SURVEY

Thomas B. Nolan, Director

 CONTENTS

[Letters designate the separately published chapters]

(A) Evolution of methods for evaluating the occurrence of floods, by Manuel A.
Benson.
(B) Factors influencing the occurrence of floods in a humid region of diverse
terrain, by Manuel A. Benson.
(C) Effect of reservoir storage on peak flow, by William D. Mitchell.
(D) Factors affecting the occurrence of floods in the Southwest, by M. A. Benson.
(E) A comparison of methods used in flood-frequency studies for coastal basins
in California, by R. W. Gruff and S. E. Rantz.