West Virginia Agricultural and Forestry Davis College of Agriculture, Natural Resources

Experiment Station Bulletins And Design

1-1-1961

Rainfall probability and its applications

Jerry C. Burchinal

W. H. Dickerson

Follow this and additional works at: https://researchrepository.wvu.edu/

wv_agricultural_and_forestry_experiment_station_bulletins

Digital Commons Citation

Burchinal, Jerry C. and Dickerson, W. H., "Rainfall probability and its applications" (1961). West Virginia
Agricultural and Forestry Experiment Station Bulletins. 454.
https://researchrepository.wvu.edu/wv_agricultural_and_forestry_experiment_station_bulletins/429

This Bulletin is brought to you for free and open access by the Davis College of Agriculture, Natural Resources And
Design at The Research Repository @ WVU. It has been accepted for inclusion in West Virginia Agricultural and
Forestry Experiment Station Bulletins by an authorized administrator of The Research Repository @ WVU. For more
information, please contact [email protected].
 W%i^M'

3h-

BULLETIN 4 54T
MARCH 1961

WEST VIRGINIA UNIVERSITY AGRICULTURAL EXPERIMENT STATION

 Digitized by the Internet Archive
in 2010 with funding from
Lyrasis IVIembers and Sloan Foundation

http://www.archive.org/details/rainfallprobabil454burc
CONTENTS

Introduction 5

Probability Methods 6

Applications for Frequent Estimates 7

Case I. Average Annual Precipitation for the State 7

Case II. Annual Rainfall by Water Years 9
'-
Case III. Weekly RainfaU .. - 10
Case IV. Monthly Rainfall 13

Summary 15

Appendix A
Determination of a Frequency Line by the Kimball Equation 16

Appendix B
Frequency of Rainfall by the Log-Probability Method 1

Appendix C
The Incomplete Gamma Distribution 20
 THE AUTHORS
J. C. Burchinal is Associate Professor of
Civil Engineering in the College of Engineering
and W. H. Dickerson is Professor of Agricul-
tural Engineering in the College of Agriculture,
Forestry, and Home Economics, and Agricul-
tural Engineer in the Agricultural Experiment
Station.

This study was supported in part by the North-

east Regional Project NE-35, "Application of Chma-
tology toNortheastern Agriculture," a cooperative
effort involving the agricultural experiment stations
of the Region and the Weather Bureau of the United
States Department of Commerce.

(Cover photograph by David R. Creel)

West Virginia University

Agricultural Experiment Station
College of Agriculture, Forestry, and Home Economics
A. H. VanLandingham, Director
Morgantown
Rainfall Probability
And Its Applications
J. C. BURCHINAL and W. H. DICKERSON

Introduction

PRECIPITATION, or rainfall, as it is often called in popular usage,

has been measured and recorded for a number of West Virginia
60 years. Continuous records of good quality
stations for about
are available for 24 back to 1926. Data from these stations on
stations
precipitation and temperature have been entered on punched cards for
ease of tabulation and analysis under a cooperative arrangement with
the U. S. Weather Bureau, and the Northeast Regional Project, NE-35,
"Application of Climatology to Northeastern Agriculture."
These records have many uses in engineering and agriculture. One
useful purpose is in the preparation of rainfall probability estimates. These
estimates, which can be conveniently represented as frequency curves, can
be used to show the probable intensity or amount of rainfall. Frequency
may be expressed as a percentage, or in terms of the return period in years.
It should be understood that the expression of frequency as a return

period in years implies no periodicity. Annual, seasonal, monthly, weekly,

or shorter periods can be treated by various probability methods.
Estimates of rainfall probability are essential for solving many prob-
lems dealing with drainage, stream flow, and the design of
irrigation,
hydraulic structures.For example, the normalcy of rainfall during the
period of an experiment in which rainfall influences the outcome should
always be examined. Experimental work in irrigation, the measurement
of runoff or erosion, or the determination of crop yields are always af-
fected by the amount, The use
distribution, or intensity of precipitation.
of probability estimates offers a convenient method of comparing the
period of record to the long-time expectancy. Thus it can be determined
if the period in question is above, below, or near normal expectancy, or
if the period contains years that represent a wide range in precipitation
frequency.
Several methods have been advanced to determine rainfall probability.
Some are relatively simple and easy to apply, whereas others involve
mathematical formulas that are time consuming to solve. An ideal method
of estimating probability would: (1) apply to any period of time such
as a year, season, month, or week; (2) provide reliable estimates for
a return period of 50 to 100 years by extrapolation; and (3) be easily
and quickly computed. None of the probability methods now in common
use meets all of these requirements. A consideration of the merits of
the various probability beyond the scope of this bulletin, the
methods is

purpose of which is to illustrate some possible applications of rainfall

frequency data which can be derived from the available climatological
records.

Probability Methods
Three methods are illustrated for determining the probability, or
frequency, of rainfall occurrences. These are: (1) the equation pro-
posed by Kimball [5]; (2) the log-probability method; and (3) the
incomplete gamma distribution. It is believed that each method is gen-
erally applicable to the type of problem for which it is used. In some
cases, similar results can be secured and the same conclusion reached by
any of the three methods.
The Kimball formula (5) and other similar plotting equations de-
termine a plotting position for hydrologic events such as rainfall amounts
in terms of recurrence intervals or per cent frequency. These points can
be plotted (probability paper is often used for convenience) and the line
of best fit drawn by eye. If the data are normally distributed, the points
will approach a straight line on arithmetic probability paper. If the log-

arithms of the data are normally distributed, the frequency will appear
as a straight line on logarithmic probability paper. Such methods are
the simplest to use. However, a high degree of subjectivity is sometimes
involved in drawing the curve of best fit.

The method was devised by Hazen (4) and later

log-probability
improved by Chow (2). It is used later to study annual runoff and rain-
fall. Annual hydrologic events, such as flood peak or volume and annual

rainfall amounts, often approach a normal distribution. The data is said

to be skewed if it does not fit the normal curve. The log-probability
method offers one means of handling such data. By use of the method,
a transformation is obtained that will result in a straight line on logarithmic
probability paper.
The incomplete gamma function has been suggested by Barger and
Thom (1) as a model for representing rainfall distribution. Weekly rain-
fall do not usually exhibit normal distribution characteristics be-
totals
cause of the large number of weeks with zero rainfall. It has been pro-
posed that the incomplete gamma distribution may offer an appropriate

6
model for smoothing rainfall probability curves for weekly, monthly, or
seasonal amounts. Friedman and Janes (3), in applying the method to
Connecticut precipitation data, concluded that most irregularities in the
shape of cumulative probability curves based on a 30-year sample are
not significant, and that the incomplete gamma affords a sound basis for
smoothing out such irregularities.

Applications for Frequency Estimates

In the following section, frequency curves are illustrated and there
is a brief explanation of how they may be used for various purposes and
by different interests. The curvesexamples are based on the Kim-
in the
ball equation, the log-probability method, and the incomplete gamma
model. This order is arrayed somewhat on the ease of computation of
a frequency line, beginning with the simple and proceeding to the more
complex. Examples of the computations required for each case are given
in the Appendixes.

CASE 1. Average Annual Precipitation for the State

The frequency line for the average rainfall over the State as determined
by the Kimball equation is shown in Figure 1. The average annual rain-
fall over the State was determined by the station average method, i.e.,

the arithmetic average of all stations reporting, disregarding differences

in areal coverage.

A plotting position was determined for each annual rainfall amount

by means of the plotting equation. The points were plotted on logarith-
mic probability paper, and the line of best fit drawn by eye. Individual
points that fell far out of line were given small weight in determining the
position of the line.
Figure 1 shows that the points between 5 and 95 per cent frequency
approach a straight line. This can be interpreted as indicating that more
confidence can be given to the line between these limits than can be ascribed
to the extremities. The point representing the 25.43 inches for 1930 is

not in good agreement with the remainder of the data. This may be due
to any of a number of reasons, for example, the data are not normally
distributed, the data are not a representative sample, or thelow figure
of 25.43 inches may represent a drought with a recurrence interval much
longer than the 3 3 -year return period assigned by the plotting equation.
The frequency line of average annual rainfall over the State is of
general interest for studying the droughtiness, normalcy, or excessive
rainfall characteristics of selected years or periods. Appendix Table A
contains the annual average rainfall for the State for the period 1926-1957,
 — r~

1 i
"T^ ;
i!!^ H'\ 1

1
Hu
j
I I
1
TT

:nr _,_ _— —
4:J= + :i+:=j: — -p liil ^i==; ^— ;:;::" == =
\ '
1
,
41 lu""""
Y
1 i
i

i 1
1
1

c) _ !

'
.
':'.
JL -y
> Q>
r-- C ,c ::.-
~ z:r.: rii iir44"^
_i i_
=
•— =
:::;:: ^ZJ^± 4--
— ^— - —
: ITr
a:
f T' u
1-^ ^
/
srS
gJ 0>
|i
/
::::::::^: " — f

_. _ _L^ Ll '
— X~
n—
-

—
../..
J ..
i __^
—L 4- —— '

-"J —k-| -^— -

!

\
—— ;::: ;;:: :: — ~ 11
"^^'^^^T—rt- E =
r.'-z:

—-h- = —
eh; p:--^^
+-r-j
--
h- 1 1 ~; ^ EE „... 1^
1
f
^„ J
1

—
..., . .

————
1

d •»i—
\

—
=
::::

r^
;rE
...;
4— -*— '
'
'
-t-
E
=S

Er -
CM
E
——
1
__.
f
i:ii [::'. \'.:r^
— o—
ZL
1
« « -

"*^' .,.-- „ IZIZI _J «_ E -

EE E^ =3%-. '
! ;
i

::::iz:t:-
—
— - f 1

— ,
:

S: ^
:Hr:i: ::E:
—
iz '.^
r tt+--t;
:0 E
—— !
•
:-—•
jf—
|- - '-:-
ff---J---
-

- il;. !

:i
xl 4- -I- --

—— Lr i^. ill
§ ==
"- -z:r: --- o
= = s rr- = [=
== — z: lit::^~"7 ir:-:.;i::ri::;::
r- p
.
1

!'
[[
'

U'

—— ;: + li-!. l:,:i!i —
n—
—
'tT'
= 3: ::::
1 1 II
=:
' r
t 1 1 1

•""f**/ "/ ii'^^^i^a

jjiiiiL:: ..lii 1
c 6
n i

CV
>°\\ N <0 8 1-
1*)
inclusive. It has been mentioned that the dry year of 1930 is outstanding.
In comparison, the droughty season of 1953, which may have been as
severe as any experienced at some localities (this point has not been in-

vestigated) did not approach the State-wide deficiency that existed in

1930. At the other extreme, a high rainfall such as the 54 inches ex-
perienced in 1950 can be expected to have a return period of about 15
years, as determined from the frequency line.
Data such as the foregoing are needed in connection with work in-

volving drainage, erosion control, irrigation, flood control, water supply

and related problems.

CASE II. Annual Rainfall by Water Years

A hypothetical case which serves to illustrate the use of an annual
rainfall frequency curve in engineering follows: Assume that an industry
wishes to use the runoff supplied by the 63.59-sq. mi. Deckers Creek drain-
age area near Morgantown, West Virginia. Further assume that this in-

dustry has determined that it will need 14 inches of runoff annually in

order to supply its water needs. A tabulaton of actual data is shown in
Table 1.

TABLE 1. Annual Runoff, Deckers Creek Drainage Area, and Rainfall,

Morgantown Lock and Dam, Morgantown, West Virginia*

Precipitation in
Water Inches Runoff from Deckers
Inches Morgantown
Year Creek Drainage Area
Lock and Dam
1947 14.00 35.16

1948 25.65 47.18

1949 21.76 39.12

1950 27.31 47.43

1951 28.64 46.34

1952 27.12 48.69
1953 18.48 35.08

1954 15.98 35.85

1955 24.58 42.12

1956 33.44 54.68

* Runoff from United States Geological Survey data and rainfall from United States Weather
Bureau records.

The record of runoff for Creek drainage area is only

the Deckers
as long as is shown in Table and statistical tests may be applied to
1,

show that these are insufficient data from which to plot a runoff frequency
curve. However, the rainfall records go back a sufficient number of
years so that a frequency curve may be constructed. Computations,

9
based on a 26-water-year period, October 1 to September 30, inclusive,
utilizing the log-probability method provided the frequency line shown
in Figure 2 and the data shown in Table 2.

TABLE 2. Probability of Annual Rainfall, Morgantown Lock and Dam,

Morgantown, West Virginia

Frequency Recurrence Interval

Inches
(Per Cent) (Years)

99 1.01 30.61

50 2 42.08
20 5 47.19
5 20 52.71

1 100 57.81

Table 1 shows that 1 947 gave the lowest percentage of runoff for the
annual rainfall, namely 40 per cent. The amount of precipitation in that
year was 35.16 inches. The probability that this amount would be
equalled or exceeded in any year is, according to Figure 2, approximately
90 per cent. Stated in another way, the chances are that in one of ten
years a rainfall of 35 inches or less will be experienced.
Since 14 inches was the minimum amount the industry had decided
it would need, it could then decide whether it could afford to function
with the probability of having insufficient water one year in ten on the
average, or it could make some provision to take care of this deficiency.
If it can be assumed that the relationship of 40 per cent of the
rainfall will be runoff for the dry years, the frequency curve might pro-
vide an estimate of how much additional water will be required by the
industry in such years. The 95 per cent probability shows about 33.50
inches of rainfall. Forty per cent of this equals 13.4 inches. The deficiency
would be 0.6 inch or about 660,000,000 gallons which would have to
be provided on a once in 20-year basis.

CASE III. Weekly Rainfall

Weekly frequency curves based on the incomplete gamma model
are shown in Figure 3 for Elkins, West Virginia. Week numbers refer
to the weeks of the climatological year beginning March 1. Thus, week
19 is the period of July 5-11, inclusive, and week 32 is October 4-10.
At Elkins, week 19 and week 32 have, respectively, the greatest and
least average precipitation for any period of the climatological year. For
the driest week, the chances are about 15 per cent or one year in seven that
rainfall of one inch or more will occur. For week 19, the wettest, the
probability for this is 58 per cent.

10
 """""
f
1 1 Hi
i!
rT'"
1

h '

„ J
S*
-^^^ ts- ft
"— TtxE: rr — ^^-f —X —+- -A P^
i-^ -—.-\~- __ —
tSl'J 1 iCi
:, i 1

J i<> 1
'
1
1

5*'^" !
1? .« i
I
:i '
''\\ \\

5 d 1

—
'

, '

U 'i
"E
1

E —E
^

iiisc: ^
SHk-^^^ ':.:z
r- : ;;:!:::: ZZ'.
+ =f=- ~^

l^^-=^#^^ ^ t --- -^1 '^1

—
1
i ! 1

-i^-i^
T^i—
-±11*- 35 - - ^ ^ —
o <> [
< >

7^
£
Jt <) "^ ci
:

i£ Ci '
> / '

S
1 1

»>
-*- ^-i — _
— -
_.
J
L ..-•_!_
1
1

_L _1_ \

—
-r^ '

^ / M
1

'

^^-
,
1

'

— — — rr izz z-;rr^ii
1
1

\;=3
1

1 !

1
.

' '
z:;
:E;;r;
~E
^-^z
-^V-^^— —— e-
LL '
'

z:: !
.,
1---
:...
^ /. ,. .
___t:
#fR# ~
1
1

- E :E
i
:;£

i T^E^
h^^

^E
"~~
-"—
f^— :
! 1
^ •

k.

E rrr ur:---:-::: p-|-3

-•--^--— -—
i_ — —
——
E :;;; :;:
E
"-T '

— i i

! — — "
"^~ '
'

'
E
^T^
-h^
"
i

1
E1 1

i 1
,
1
[ !

I Q 1
1

ii Mi CM '
1 i
'

1 1

ij f
III IJ-LiL 1

IT
1 '

II l! !

'-Z 'ZZ
- '

~ = ^z

j
" 1
E:
— rr
1= ::r
-- "-

'
.
'
'
-T- ^4E
!
--

-^
^=
'

' '

'
1

ti,.
'

I'M
\t' - i 1
i 1

1 1

{'i <
'

-U L_L-| — !

^= irr =r
.

till
1 E "1 —=:=
1

— — zzr-'i^
—
E rr: :;t::

J-Ij , ,

1
:
:
- -" -^
1 1 1 1 !
Ef:^y— Ti
^^ /f 1 HI -Wttji in gliiLLlT____Z
1
1
ll-l

^ w _Ei!_e! 1
o
1

O
C
ij ^S S s^

11
12
 For an example of how such data might be put to practical use,
amounts might be considered as one
the probability of weekly rainfall
of the factors in scheduling farming or other operations that are influenced
by the weather. This presupposes that some latitude of choice exists.

In the case cited, week 19 is be a good period for growing forage

likely to

crops in the general vicinity of Elkins, but unfavorable for harvesting

such crops because of the relatively high rainfall. With sufficient informa-
tion, the farm operator should be in a position to know whether to ( 1 :

produce forage crops that can be harvested in a favorable climatic period,

(2) ensile crops that can generally be expected to mature when weather
conditions are unfavorable for making hay, or (3) provide drying equip-
ment and select machinery and methods that minimize the weather hazard
in harvesting forages.

CASE IV. Monthly Rainfall

One use of a monthly rainfall probability curve may be illustrated
by the determination of the normalcy of May rainfall on the Reymann
Memorial Farms, Wardensville, West Virginia, during the period in which
irrigation experimental work was in progress there. May rainfall is prob-
ably a good indication of the need for, or the response to be expected
by, the irrigation of the first cutting of forage crops. If the average rain-
fall for the months of May assumed to
(3.31 inches at Wardensville) is

be adequate, then the months of May during the period

rainfall for the
of the experiment can be compared to the computed frequency line to
determine if they were near normal, or the relative wetness or dryness.
The May frequency curve for WardensviUe (Figure 4) was determined
by the incomplete gamma method. It has been determined that monthly
rainfall is often skewed and not normally distributed. In Figure 4 the
50 per cent probability rainfall does not coincide with the average or
mean of 3.31 inches. The frequency curves indicate that only about 42
per cent of the Mays wiU have precipitation equal to or greater than the
average, or conversely that 58 per cent of the months of May can be
expected to be equal to or less than this figure. Skewness of monthly
rainfall is not confined to the station and period considered here, as 60
per cent of the months have less than normal precipitation and 40 per
cent have more, according to Linsley, et al. (6). This feature of the
distribution may sometimes lead to an over-optimistic view of the adequacy
of the average monthly rainfall for crop production.
During the period of the irrigation experiment (1953-58), May of
1953 had the highest rainfall, although it is a matter of record that the
following months in 1953 were very dry. Table 3 shows the relationship
of the actual precipitation to the computed frequency Une.

13
14
TABLE 3. Comparison of Recorded May Rainfall for the Period
1953-58 to the Expected Frequency of May Rainfall at
Wardensville, West Virginia

Probability from
Recorded Rainfall Ordered Rainfall
Fig. 4 of rain-
May May fall ^ ordered
1953 5.94 ins. 2.29 35%
1954 2.34 2.34 37%
1955 4.81 2.44 40%
1956 2.44 2.68 46%
1957 2.29 4.81 80%
1958 2.68 5.94 90%
Av. (1926-1956) = 3.31 inches.

50 Per Cent Probability = 2.90 inches.

These comparisons would seem to indicate that the 1953-58 period

contained a good sample of wet and dry years, but that the period may not
be indicative of results to be expected by the irrigation of forage crops
when rainfall approximates either the long-time average or the 50 per
cent chance which represents the median value.

Summary
Weather data, including temperature and precipitation records, have
been entered on IBM punch cards for 24 West Virginia Stations beginning
with January 1, 1926. This work was done in cooperation with the
United States Weather Bureau and the Northeast Regional Project NE-35,
"Application of Climatology to Northeastern Agriculture."
One of the possibilities for utilizing the precipitation data is for
the preparation of rainfall probability or frequency Such
estimates.
estimates have been illustrated by computations for weekly, monthly, and
seasonal rainfall amounts. Three methods of determining frequency
curves have been used — the equation proposed by Kimball, the log-
probability method, and the incomplete gamma distribution. Some pos-
sible applications illustrating the use of frequency curves in engineering
and agriculture are suggested.

References
1. Barger, G. L. and Thorn, H.C.S., "Evaluation of Drought Hazard." Agronomy
Journal, Vol. 41 (1949).

2. Chow, Ven Te, "The Log-Probability Law and Its Engineering Applications."
ASCE, Separate No. 536, Vol. 80 (November 1954).
15
3. Friedman, Don G., and Janes, Byron E., Estimation of Rainfall Probabilities',
Conn. Agr. Exp. Sta. (Storrs), Bull. 332, 1957.

4. Hazen, Allen, Flood Flows. A Study of Frequency and Magnitudes. New York,
John Wiley and Sons, Inc., 1930.

5. Kimball, B. F., "Assignment of Frequencies to a Completely Ordered Set of

Sample Data." Tra?7s. A.G.U., Vol. 27, No. VI (1946).
6. Linsley, Ray K. Jr., Kohler, Max A., and Paulhus, Joseph L. H., Applied Hy-
drology. New York, McGraw-Hill Book Company, Inc., 1949.

7. Pearson, K., Tables of the Incomplete Gamma Function. London, 1922.

APPENDIX A
Determination of a Frequency Line by
the Kimball Equation

The equation is:

N + 1 m
F = per cent frequency or probability
T = return period in years, or recurrence interval
m = rank number in an array of decreasing order of magnitude
N = number of years of record

Procedure

1. Tabulate the average annual rainfall for the State.

2. Array the rainfall amounts in order of decreasing magnitude, as
in Table A.

3. Compute and tabulate the plotting position, i.e., per cent fre-
quency or return period, by the formula.
4. Plot the arrayed rainfall amounts against the per cent frequency
on probability paper, see Figure 1

5. Draw the line of best fit by eye, giving less weight to single
points that plot far out of line.

16
 Table A

Av. Annual
Year Precipitation m Arrayed % Frequency
Inches (F)
Inches

1926 49.60 1 54.43 3.0

1927 49.14 2 53.07 6.1

1928 43.28 3 51.55 9.1

1929 46.77 4 51.13 12.1

1930 25.43 5 50.61 15.2

1931 42.36 6 49.86 18.2

1932 43.86 7 49.60 21.2

1933 48.65 8 49.14 24.2

1934 37.61 9 48.68 27.3

1935 51.13 10 48.65 30.3

1936 46.12 11 48.02 33.3

1937 49.86 12 46.81 36.4

1938 42.39 13 46.77 39.4

1939 43.65 14 46.12 42.4

1940 45.12 15 45.12 45.4

1941 36.59 16 45.00 48.5

1942 48.68 17 44.52 51.5

1943 40.72 18 43.86 54.5

1944 44.52 19 43.65 57.6

1945 50.61 20 43.28 60.6

1946 36.69 21 42.39 63.6

1947 37.22 22 42.36 66.7

1948 53.07 23 41.28 69.7

1949 45.00 24 41.19 72.7

1950 54.43 25 40.72 75.8

1951 48.02 26 39.82 78.8

1952 41.28 27 37.61 81.8

1953 36.53 28 37.22 84.8

1954 46.81 29 36.69 87.9

1955 39.82 30 36.59 90.9

1956 51.55 31 36.53 93.9

1957 41.19 32 25.43 97.0

17
 APPENDIX B
Frequency of Rainfall by the Log-Probability Method

Hazen (4) developed a method of computing frequency curves based

on the assumption that the logarithms of the variable are normally dis-
tributed. The procedure used empirically determined log-probability fre-
quency factors to handle skewness. Chow (2) later computed a table of
theoretically correct log-probability frequency factors for a range of skew
coefficients from to 5.0.

Plotting positions are determined by the formuJa:

ye = y (KCv + 1) -

y, = plotting positions, per cent frequency, on the theoretical

curve

y = mean value of the data

K = log-probability frequency factor from table calculated by

Chow
Cv = coefficient of variation

Cv = V N 2 y' — {tyY

N = number of years of record

y = annual value of rainfall in inches

Procedure

1. Tabulate the average annual water year rainfall, see Table B.

2. Determine N, and calculate 2y, 2y', and (2y)'.

3. Calculate Cv.

4. Using Cv, determine K from tables prepared by Chow.

5. Solve the equation for yc

6. Plot the frequency line of yc versus per cent frequency, as in

Figure 2.

18
Table B. Calculations for Plotting Position y< for use in the Theoretical
Log-Probability Equation. Station —Morgantown
Lock and Dam, West
Virginia. Annual Rainfall for Years 1932-1957, inclusive, based on Water
—
Year October 1 to September 30, inclusive

Recurrence Per Cent

Year Rainfall
Interval (Yrs.)
K KCv yc
Frequency
1932 38.92 1.01 -2.04 —0.28 30.61 99
1933 53.34 2 -0.07 —0.01 42.08 50
1934 35.73 5 0.81 0.11 47.19 20
1935 39.77 20 1.75 0.24 52.71 5

1936 40.05 100 2.62 0.36 57.81 1

1937 46.99
1938 42.50
1939 47.00
1940 40.26
1941 37.72
1942 42.50
1943 43.89
1944 37.08
1945 53.15
1946 39.52
1947 35.16
1948 47.18
1949 39.12
1950 47.43
1951 46.34
1952 48.69
1953 35.08
1954 35.85
1955 42.12
1956 54.68
1957 35.13
N 26

2y 1105.20

2y= 47,857.66

(2y)' 1,221,467.04

y 42.51

19
 Cv =\N2y= — (Sy)2 = "V 1,244,299.16 — 1,221,467.04 = "V 22,832.12

Sy 1105.20 1105.20

151.10
— — 0.137
1105.20

Yc = (KCv + 1) Y
Ye = (-.28 + 1) 42.51

Ye = 30.61

APPENDIX C
The Incomplete Gamma Distribution

Barger and Thorn ( 1 ) have suggested that the incomplete gamma

function could be used for studying weekly, monthly, or annual rainfall
distributions. On the basis that this model is appropriate, the expression
for cumulative rainfall probability is, according to Friedman and Janes,

r'[ x^-' e"^'

(x) = ^^('-^) /
dx
I ^ ' r (/)
G(x) is the probability of occurrence of rain amounts less than or
equal to x inches.

p ^ P= —y where y is the number of times no rain occurred and N

N is the total sample size.

"
1
+ V/-I L
In X — TT
n
1

?
,
In Xi -1

- g -
/ 1
^ 1
4 In X - tIt 2 In Xi
/5=b = i

f
The average rainfall x and the sum of the natural logarithms 2 hi Xi are
i

based on periods when rain actually occurred, that is n N-y. =

The procedure used to obtain the desired solution of the equation
is as follows:

20
 1. Tabulate monthly rainfall amounts as has been done in Table
CI for May rainfall at Wardensville, West Virginia. The data were ar-
rayed in Column 1 for ease in handling the logarithms.
2. Record in Column 2 the natural logarithms of the rainfall amounts.

_ _ 1 n
3. Determine N, n, y, x, In x, — :§ In Xi. Calculate g, b, and P
n i

from the relationships given above.

4. Calculate values in Column 3 from u =

b V g
5. Using the calculated values of u and p, read the cumulative
probability from Pearson's Tables of the Incomplete Gamma Function
(7) and enter in Column 4.
6. Column 4 is then plotted against the corresponding rainfall
amount shown in Figure 4.
to obtain the curve as
7. When periods with no rain occur, as in week 32 for Elkins,
Table C2, Column 5 is calculated by multiplying the probability by the
proportion of rain cases, i.e., (1-P) F(x). Adding to Column 5 the pro-

portion of no rain cases, P = —2 = .067 = .07, gives the proba-

30
bility to plot against the corresponding rainfall amounts as is shown in

Figure 3.

Table CI. Wardensville, West Virginia —

May Rainfall Prob-
ability Incomplete Gamma Calculations

1 2 3 4
Rainfall Ln X u=.504 X F(x)*

.16 - 1.8326 .08 .00

.64 - .4463 .32 .02

.69 - .3711 .35 .03

1.47 .3853 .74 .15

1.48 .3920 .75 .16

1.83 .6043 .92 .24

2.29 .8286 1.15 .35

2.34 .8502 1.18 .36

2.44 .8920 1.23 .40

2.55 .9361 1.28 .42

2.58 .9478 1.30 .43

2.65 .9746 1.34 .44

2.68 .9858 1.35 .44

2.71 .9970 1.37 .47

21
 TABLE CI. CONTINUED
1 2 3 4
Rainfall Ln X u=.504 X F(x)*

2.77 1.0188 1.40 .48

2.99 1.0953 1.51 .51

3.23 1.1725 1.63 .57

3.56 1.2698 1.79 .63

3.67 1.3002 1.85 .64

3.67 1.3002 1.85 .64

3.75 1.3218 1.89 .66

3.77 1.3271 1.90 .66

4.15 1.4231 2.09 .72

4.61 1.5282 2.32 .78

4.81 1.5707 2.42 .80

5.06 1.6214 2.55 .83

5.25 1.6582 2.65 .85

5.34 1.6752 2.69 .86

5.94 1.7817 2.99 .90

6.24 1.8310 3.14 .91

6.70 1.9021 3.38 .94

Totals 102.02 + 30.9410

N = 31
* From Pearson's Tables p =^ 1.75; u = .504x

Calculations for May Rainfall

Incomplete Gamma Distribution
N = 31 y =
X = 102.02 = 3.29" In X = 1.1909
31

1 30.9410
— 2 In Xi = = .9981
n 31

Inx 2 In Xi = 1.1909 — .9981 = .1928

n

=,1+^1+4/3 (.1928) 1 + "V 1-2570

4 X .1928 .7712

22
 1 + 1.1212 2.1212
= 2.750
.7712 .7712

b = —X 3.29
= 1.196
g 2.750

X X X
.504 X

bVg 1.196 X 1.658 1.983

P = g 1 = 2.75 — 1.00 = 1.75

Table C2. Elkins Week 32 (Least Av. Precip.)

Incomplete Gamma Calculations
1 2 3 4 5 6
Add No
Arrayed In X U F(x)* (l-P)F(x) Rain (.07)

— .00 .00 .00 .07

— .00 .00 .00 .07

.01 - 4.6052 .02 .02 .02 .09

.02 - 3.9120 .04 .03 .03 .10

.04 — 3.2189 .08 .06 .06 .13

.05 - 2.9957 .09 .07 .06 .13

.06 - 2.8134 .11 .09 .08 .15

.14 - 1.9661 .26 .19 .18 .25

.16 - 1.8326 .30 .23 .21 .28

.19 — 1.6607 .36 .27 .25 .32

.27 - 1.3093 .51 .36 .33 .40

.28 - 1.2730 .53 .38 .35 .42

.34 - 1.0788 .64 .45 .42 .49

.37 - .9942 .70 .47 .44 .51

.42 - .8675 .79 .52 .48 .55

.47 - .7550 .89 .57 .53 .60

.51 - .6733 .96 .60 .56 .63

.54 - .6162 1.02 .62 .58 .65

.57 - .5621 1.08 .64 .60 .67

.66 - .4155 1.25 .69 .64 .71

.66 — .4155 1.25 .69 .64 .71

.67 - .4005 1.27 .70 .65 .72

.86 - .1508 1.62 .79 .73 .80

.95 - .0513 1.80 .82 .76 .83

.97 - .0305 1.83 .82 .76 .83

.99 - .0100 1.87 .84 .78 .85

23
 TABLE C2. CONTINUED

1 2 3 4 5 6
Add No
Arrayed In X U F(x)* a-p)F(x) Rain (.07)

1.08 + .0770 2.04 .86 .80 .87

1.11 + .1044 2.10 .87 .81 .88

1.22 + .1988 2.30 .89 .83 .90

1.77 + .5710 3.34 .96 .89 .96

15.38 —31.6569

F(>.) from Pearson's Tables p .08; u 1.89x; 1-P — 1 .067

2 P = .067
30

15.38
X = = .55" In X = —.5974
28

1 —31.6569
— Sin Xi = -1.1306
n 28

In X S In Xi = —.5974 — (—1.1306)
n

= .5332

1 +V 1 + 1-33 (.5332) 1 + 1.31 2.31

2.1328 2.1328 2.1328

X .55
b = — = = .51
g 1.08

— —
p = g

XX 1 = 1.08 1 = .08

bVg .51 \ 1.08 .51 (1.04) .53

u = 1.89 X

1 — P = 1.00 — .067 = .93

24
 Table C3. Elkins Week 19 (Greatest Av. Precip.)
Incomplete Gamma Calculations

Arrayed In X u F(x)*
.04 — 3.2189 .03 .00

.11 — 2.2073 .09 .02

.28 — 1.2730 .23 .08

.45 — 0.7985 .37 .16

.45 - 0.7985 .37 .16

.55 — 0.5978 .45 .20

.67 — 0.4005 .55 .26
.92 — 0.0834 .75 .37

.98 — 0.0202 .80 .40

.99 — 0.0100 .81 .40
1.01 + 0.0100 .83 .41

1.04 + 0.0392 .85 .42

1.10 + 0.0953 .90 .45

1.11 + 0.1044 .91 .45

1.22 + 0.1988 1.00 .49
1.28 + 0.2469 1.05 .51

1.53 + 0.4253 1.25 .59

1.54 + 0.4318 1.26 .59
1.57 + 0.4511 1.29 .61

1.79 + 0.5822 1.47 .67

1.81 + 0.5933 1.48 .68
1.82 + 0.5988 1.49 .69
1.84 + 0.6098 1.51 .69
2.01 + 0.6981 1.65 .73

2.13 + 0.7561 1.75 .75

2.53 + 0.9282 2.07 .82
2.63 + 0.9670 2.16 .84
3.46 + 1.2413 2.84 .92
3.51 + 1.2556 2.88 .93
5.72 + 1.7440 4.69 .99

46.09 + 2.5691

F(x) from Pearson's Tables p ^ .59; u = .82x

25
_ 46.09 _
X = = 1.54" In X = .4318
30

1 2.5691
— 2 In Xi = = .0856
n 30

In X 2 In Xi = .4318 — .0856 = .3462

n

1 +'Vl — 1.33 (.3462) 1 +-V 1.4606

4 X .3463 1.3852

1 + 1.2086 2.2086

1.3852 1.3852

g = 1.59

P = g — 1 = 1.59 — 1 = .59

T 1.54
b = — = =: .97

g 1.59

X X

bVg .97 V 1-59 .97 x 1.26 1.22

u = .82 X

26