Study of Rainfall Pattern using Extreme Value Distributions

Prof. Shreenivas Londhe Ph.D

Professor in Civil Engineering, Vishwakarma Institute of Information Technology, Pune 411048
[email protected]
Nikhilesh Gandhre
Post Graduate Student, Water Resources and Environmental Engineering
Vishwakarma Institute of Information Technology, Pune 411048
[email protected]

Abstract

The study aims to analyze the rainfall pattern in various districts of Maharashtra using extreme
value distributions. The data used for analysis is from the year 1901-2000. The data is divided in
two sets with first set from 1901 to 1950 and second set from 1951 to 2000.Rainfall for return
periods 50 & 100 years is calculated using Gumbel & Log Person distributions. There is a
considerable difference between the rainfall values of each return period between both the sets
for all the three distributions. An increase in rainfall is seen in the Raigad, Sindhudurg, Pune,
Kolhapur, Sangli, Nanded, Gadchiroli, Chandrapur districts while a decrease is seen in Ratnagiri,
Nashik, Ahemadnagar, Jalgaon, Aurangabad, Beed, , Amravati, Buldhana, Yavatmaal, Nagpur,.
This may also be attributed to effect of climate change over last 100 years.

Key words-Rainfall Pattern, Extreme Value Distribution, Gumbel, Log Pearson

Introduction

Rainfall is the most important hydrological parameter and is largely responsible for deposition of
fresh water on the globe. It provides a suitable condition for many types of ecosystems as well as
water for hydroelectric projects and crop irrigation. Rainfall has a unique characteristic of
diversifying with respect to space and time thus resulting in variations from place to place and
time to time. Analysis of rainfall distribution over an area helps in planning strategies for
irrigation and cropping pattern. Probability and frequency analysis of rainfall enables us to reach
at conclusions which can be used to prevent floods, droughts and can be applied to the
development of water resources like dams, canals, etc. In recent years floods as well as drought
like conditions are being observed in various areas of Maharashtra. Changes in rainfall pattern in
any phase might affect an agro based economy which largely depends upon the monsoon.
Extreme Value Distributions are used in this analysis to study the changes in rainfall pattern over
a period of 100 years.
Study Area & Materials

The state of Maharashtra is considered as a study area. It is situated at 19.7515° N latitude and
75.7139° E longitude having a total area of 307,713 km². Out of 36 districts, the study is being
carried out on 18 of them. Annual precipitation data of these 18 districts for a period of 100 years
i.e from 1901 to 2000 is collected from Indian Meteorological Department (IMD).

Figure 1-Map of Study Area

Methodology

The 100 years precipitation in mm data of each district is divided into two sets i.e 1900-1950 and
1951-2000.Rainfall for returns periods of 50 and 100 is calculated using extreme value
distributions for each data set. Details about the extreme value distributions used are given
below. The fitness of a particular distribution to a particular data is judged using Chi-squared test
details of which are given in subsequent sections.

Extreme Value Distributions

Extreme Value Distributions arise as limiting distributions for maximums or minimums (extreme
values) of a sample of independent, identically distributed random variables as the sample size
increases. In hydrology these distributions can be used to predict the extreme events such as
maximum rainfall, floods and droughts. In this analysis Gumbel & Log Pearson Distributions are
used for analysis of rainfall pattern.
Gumbel Distribution (1941) (Subramanya, 2008)
The following steps are used to analyze the annual maximum rainfall probability distribution.
i. Means of annual rainfall (X)
Standard deviation (sn)
ii. Reduced variate is calculated using the formula.

Yt = -ln( ln (T/(T-1))) (1)

Where
Yt = Gumbel reduced variate, a function of ‘T’
T = return period in years
iii. Frequency factor K for Gumbel Distribution is calculated using the formula

K = (Yt-Yn)/Sn (2)

Both Yn (reduced mean) & Sn (reduced standard deviation) of the Gumbel Variate and
are functions of sample size N and their values are available in tabular form for several of
N (Subramanya, 2008)
iv. Magnitude of rainfall for the given return period is calculated by

Xt =X+ (K*sn) (3)

Xt= magnitude of rainfall for given return period ‘T’

K=Frequency factor
X=Mean of annual
sn=standard deviation

Log Pearson Distribution (Subramanya, 2008)

In this method, the sample (i.e. Z in this case) is first transformed into logarithmic form before
analyzing. For Log Pearson, Kz is a function of which is calculated by both the return period
and the coefficient of skew (Cs). The values of Kz are given in the book (Subrmanya, 2008)
Following steps were followed for implementing Log Pearson distribution.

i. Log X = Z of all rainfall data

ii. Mean (Z) of the log values
Standard deviation (Sn) of the log values
Coefficient of skew of the log values (Cs)
iii. The value of frequency factor (Kz) was taken from the statistical table in Engineering
Hydrology by K Subramanya,2008 corresponding to Cs to “T” (Return period).
iv. Magnitude of rainfall Xt for a given return period ‘T’ is calculated by formula

Zt = Z + (Kz * Sn) (4)

Where Xt = Antilog (Zt)
Test for good fit

Chi Square Test

The test intends to evaluate the difference between the sample data and the probability
distribution. The formula for Chi Square test is given by

(5)

Where:
χ2 = calculated chi-square
Ei = frequency that is hoped regarding to the class division
Oi= frequency on the same class
N =number of class intervals
The formula of Ei is as follow

(6)

Where:
n = number of data
N= number of class intervals

The criterion for goodness is value of χ2 calculated should be less than value of χ2 from the
table (Reddy, 1997). The χ2 value is read out from the table by knowing two parameters
a) Degree of freedom=k-h-1
Where, k=number of class intervals
h=number of estimated parameters
b) Level of Significance

Results and Discussions

Table 1 shows consolidated results of Gumbel and Log Pearson distributions at all the districts

Table 1-Rainfall for 50 yr & 100 yr Return periods (mm)

RAIGAD GUMBEL LOG PEARSON

XT 1900-1950 1951-2000 DIFF 1900-1950 1951-2000 DIFF
50 4636.34 4995.36 359.02 3992.03 4566.57 574.55
100 4972.11 5339.27 367.16 4065.81 4734.09 668.28
SINDHUDURG GUMBEL LOG PEARSON
XT 1900-1950 1951-2000 DIFF 1900-1950 1951-2000 DIFF
50 4172.98 4540.79 367.81 3592.80 4010.97 418.16
100 4457.11 4828.09 370.98 3648.55 4082.13 433.58
 RATNAGIRI GUMBEL LOG PEARSON
XT 1900-1950 1951-2000 DIFF 1900-1950 1951-2000 DIFF
50 5243.53 5203.57 -39.96 5026.40 4970.20 -56.20
100 5623.24 5569.24 -53.99 5290.51 5217.42 -73.08
PUNE GUMBEL LOG PEARSON
XT 1900-1950 1951-2000 DIFF 1900-1950 1951-2000 DIFF
50 1661.21 1790.72 129.52 1482.79 1565.53 82.73
100 1802.52 1933.99 131.47 1550.06 1614.70 64.63
SANGLI GUMBEL LOG PEARSON
XT 1900-1950 1951-2000 DIFF 1900-1950 1951-2000 DIFF
50 1079.56 1287.72 208.17 1016.94 1195.41 178.48
100 1169.68 1407.30 237.62 1084.06 1289.57 205.51
KOLHAPUR GUMBEL LOG PEARSON
XT 1900-1950 1951-2000 DIFF 1900-1950 1951-2000 DIFF
50 2560.60 2787.23 226.63 2451.26 2561.96 110.70
100 2799.48 2971.24 171.76 2668.16 2655.69 -12.47
NASHIK GUMBEL LOG PEARSON
XT 1900-1950 1951-2000 DIFF 1900-1950 1951-2000 DIFF
50 1717.90 1524.25 -193.65 1536.07 1369.28 -166.79
100 1865.52 1643.82 -221.71 1604.99 1421.13 -183.86
AHMEDNAGAR GUMBEL LOG PEARSON
XT 1900-1950 1951-2000 DIFF 1900-1950 1951-2000 DIFF
50 1049.83 950.22 -99.62 974.89 805.61 -169.29
100 1148.89 1029.47 -119.42 1043.35 826.43 -216.92
JALGAON GUMBEL LOG PEARSON
XT 1900-1950 1951-2000 DIFF 1900-1950 1951-2000 DIFF
50 1244.64 1162.61 -82.03 1153.45 1030.70 -122.74
100 1349.37 1253.32 -96.05 1220.58 1061.79 -158.78
AURANGABAD GUMBEL LOG PEARSON
XT 1900-1950 1951-2000 DIFF 1900-1950 1951-2000 DIFF
50 1331.04 1166.38 -164.66 1135.03 1091.99 -43.03
100 1456.37 1259.75 -196.62 1172.79 1237.29 64.50
BEED GUMBEL LOG PEARSON
XT 1900-1950 1951-2000 DIFF 1900-1950 1951-2000 DIFF
50 1352.68 1253.12 -99.56 1088.92 1253.12 164.20
100 1493.33 1365.66 -127.68 1126.93 1365.66 238.72
 NANDED GUMBEL LOG PEARSON
XT 1900-1950 1951-2000 DIFF 1900-1950 1951-2000 DIFF
50 1636.78 1783.59 146.81 1571.53 1684.40 112.87
100 1792.94 1949.86 156.92 1722.97 1824.45 101.48
AMRAVATI GUMBEL LOG PEARSON
XT 1900-1950 1951-2000 DIFF 1900-1950 1951-2000 DIFF
50 1616.14 1598.98 -17.16 1465.93 1388.21 -77.71
100 1756.56 1739.40 -17.16 1543.41 1461.96 -81.45
BULDHANA GUMBEL LOG PEARSON
XT 1900-1950 1951-2000 DIFF 1900-1950 1951-2000 DIFF
50 1325.30 1220.46 -104.84 1174.54 1102.64 -71.89
100 1435.06 1317.54 -117.52 1219.40 1149.26 -70.14
YAVATMAL GUMBEL LOG PEARSON
XT 1900-1950 1951-2000 DIFF 1900-1950 1951-2000 DIFF
50 1619.92 1618.32 -1.61 1423.55 1514.94 91.40
100 1750.35 1749.50 -0.85 1470.66 1608.22 137.57
NAGPUR GUMBEL LOG PEARSON
XT 1900-1950 1951-2000 DIFF 1900-1950 1951-2000 DIFF
50 1878.94 1695.57 -183.36 1806.23 1645.92 -160.31
100 2026.30 1820.74 -205.56 1916.58 1740.18 -176.40
CHANDRAPUR GUMBEL LOG PEARSON
XT 1900-1950 1951-2000 DIFF 1900-1950 1951-2000 DIFF
50 2069.81 2147.16 77.35 1815.09 1996.65 181.56
100 2227.95 2326.03 98.08 1866.40 2119.83 253.43
GADCHIROLI GUMBEL LOG PEARSON
XT 1900-1950 1951-2000 DIFF 1900-1950 1951-2000 DIFF
50 2234.04 2316.35 82.31 2013.70 2229.86 216.16
100 2402.11 2501.43 99.31 2081.31 2389.28 307.98

Raigad, & Sindhudurg districts show an increase rainfall by more than 300mm.Pune, Kolhapur,
Sangli, Nanded, Buldhana districts show an increase in rainfall by more than 100mm.More than
100mm decrease is seen in the Nashik, Ahemdnagar, Aurangabad, Jalgaon, Beed districts.
Nagpur district shows an increase in rainfall by 150mm.Chandrapur, Gadchiroli districts show an
increase by 75mm.Amravati & Yavatmaal districts also show a decrease less than 20mm.These
changes in rainfall can be attributed to the effect of climate change over rainfall pattern at each
of the location discussed above.
Chi square test

The test is carried at 5% significance level (Suhartanto et al. (2018))

i. Gumbel

Degree of Freedom = k-h-1 = 10-2-1 = 7

k = number of class intervals = 10
h = number of derived parameters = 2 (sn, X)

Value of χ2 from the table = 14.067

ii. Log Pearson

Degree of Freedom = k-h-1 = 10-3-1 = 6

k = number of class intervals = 10
h = number of derived parameters = 3 (Z,Sn,Cs)

Value of χ2 from the table = 12.59

Table 2 shows consolidated results of all the locations while table 3 gives results of χ2 test
which indicate whether the distribution is suitable for a particular data set or not.

Table 2- χ2 calculated

DISTRICT GUMBEL LOG PEARSON

1900-1950 1951-2000 1900-1950 1951-2000
RAIGAD 13.6 7 66.4 47.8
SINDHUDURG 15 10 72 72
RATNAGIRI 6 7.2 58.4 63.6
PUNE 6.8 5.6 65.6 65.6
SANGLI 5.6 10.8 49.6 52
KOLHAPUR 8 6.8 81.2 62
NASHIK 6 4.4 56.4 64.8
AHMEDNAGAR 7.2 6.8 64.4 69.6
JALGAON 8.8 14.8 56.4 93.6
AURANGABAD 21.2 9.6 52.7 68.8
BEED 2.8 5.2 86.6 60.8
NANDED 8 8.8 61.6 71.6
AMRAVATI 2.4 4.4 67.6 84.4
BULDHANA 8.8 11.2 69.2 54.6
YAVATMAL 8.4 5.2 82 76.8
NAGPUR 8.4 12.8 66.4 57.2
CHANDRAPUR 8 6 71.2 62
GADCHIROLI 6 16 35.6 45.6
 Table 3-Chi Square Test Results

GUMBEL DISTRIBUTION LOG PEARSON DISTRIBUTION

2 SUITABLE 2 NOT SUITABLE
χ calculated < 14.067 χ calculated > 12.59

Conclusions

Districts considered for analysis follow Gumbel Distribution. Maximum increase is seen in the
district of Sangli (17%). Minimum increase is seen in the districts of Chandrapur & Gadchiroli
(3% to 4%). Maximum decrease is seen in the district of Aurangabaad (15%) while minimum
decrease is seen in Yavatmaal (0.01%). Raigad, Sindhudurg, Pune, Kolhapur districts show an
increase in rainfall by 7% to 8%. In Nashik, Nagpur & Ahmendnagar districts the decrease is by
11% to 12% while Jalgaon & Beed districts show a decrease of 7% to 9%. Nanded & Buldhana
districts show an increase by 8%. Amravati & Ratnagiri district shows a decrease by 1%. The
rainfall trend has increased in the exterior parts of Maharashtra while it has decreased in the
interior parts.

References

Reddy J (1997), Stochastic Hydrology, Laxmi Publlications (New Delhi) Pvt.Ltd,

Subramanya K (2008), Engineering Hydrology, Tata McGraw Hill Education (India) Pvt.Ltd,

Suhartanto. E. et al. (2018) “Frequency Analysis for Estimating Design Rainfall - Case Study on
the Upstream Lesti Sub-watershed” Asian Journal of Applied Science and Technology (AJAST),
2,(1), 87-100.