© IJCIRAS | ISSN (O) - 2581-5334

December 2017 | Vol. 1 Issue. 1

Rainfall frequency analysis using Gumbel

distribution
Sylvester Richard Chikabvumbwa1, Suribabu C.R2, Dessalegn Worku3
1
Department of Hydraulic Engineering, University of Tlemcen, BP 319, Tlemcen, Algeria (sylvechika@[Link])
2
School of Civil Engineering, SASTRA University, Thanjavur-613 401, India

Abstract variations in rainfall have dramatic effects on

water balance of rivers and reservoirs and on
The study of rainfall frequency analysis is crop yields. Mondal et al [1] stated that climate
important as it guides engineers, policy change affects temperature and precipitation
makers and water planners to design water more than other climatic parameters. When
structures, predict extreme events such as rainfall is higher than its mean value in a
floods or droughts and prepare disaster risk particular area, floods are expected and when it is
management plans in advance. This paper below its mean value, dry spells and
discussed how Gumbel distribution can be consequently droughts occur. Different studies [2]
used in analyzing frequency of rainfall in [3] [4] showed that crop planning and watershed
Dowa district. 30 years of rainfall data management can be performed through
(1986–2015) for Dowa district was analyzed probability analysis of rainfall. Suribabu [5] found
using Excel and Hydrognomon software out that the yield of crops particularly in rain-fed
applications by employing different areas depends on the rainfall pattern, which
statistical analysis and distribution types. makes it important to predict the probability of
The more practical result of this paper was occurrence of rainfall from the past records.
that the 2-year, 5-year, 10-year, 20-year,
25-year and 50-year return periods of In order to understand these phenomena, it is
maximum monthly rainfall were calculated important to forecast and predict the probability of
for the district. It was found out that occurrence of rainfall from the past records of
Gumbel’s distribution is suitable for hydrological data using statistical analysis [6].
predicting expected maximum rainfall in the Different statistical distributions such as Gumbel,
district. From the Gumbel’s distribution Log Pearson, and Galton have been used in
using return periods (T) above, the different studies to quantify the likelihood and
estimated maximum monthly rainfall intensity of extreme events but none is standard
obtained are: 272.5 mm, 340.4 mm, 385.3 and specific to any place[6]. Rainfall Frequency
mm, 428.5 mm, 442.1 mm and 484.3 mm Analysis is the estimation of how often a rainfall
respectively. These values are useful for event will occur [7]. The analysis involves fitting a
irrigation systems designing, reservoir probability model for a sample of maximum rainfall
designing, storm management and water over a specified period of time within a specified
balance modelling in the district. The result area. The model parameters established are then
of this study can also be used to develop used to predict the extreme events of large
more accurate and reliable models of flood recurrence interval [8][9][10]. Reliable rainfall
risk management. frequency estimates are necessary for flood zones
management, calculation of crop water
Keyword: probability distributions, requirements, proper designs of irrigation schemes,
maximum monthly rainfall, plotting and for designing water infrastructures.
position, return period
The main objective of the study was to carry out
1. INTRODUCTION
the rainfall frequency analysis for Dowa District
Rainfall is the prime source of water for using statistical approach. The results of the
agriculture (and/or irrigation) and domestic uses analysis will give detailed information of likely
such as drinking in Dowa district in Malawi. Any maximum rainfall to be expected in the district at

PAPER ID [Link]
1
 © IJCIRAS | ISSN (O) - 2581-5334
December 2017 | Vol. 1 Issue. 1

the various return periods based on the observed Figure 1. Map showing study area
data. This information will be very useful for
irrigation engineers and water planners when
determining best crops to cultivate in the district,
determining crop water requirements, designing
irrigation schemes, designing rainwater harvesting 2. METHODOLOGY
structures and designing water infrastructures such
as dams. The method employed in this study was statistical
based on probability distributions. Rainfall
descriptive maximum rainfall data of Dowa
1.1Study area district, from 1986 to 2015 (30 years rainfall data)
was considered for the rainfall frequency analysis
Dowa district lies roughly between 33° 20’ E and
by applying the Gumbel’s distribution tied with
34° 10’ E and between 13° 20’ S and 13° 40’S in Gringorten plotting position.
the Central region of Malawi. It is bordered by Firstly, the data was arranged in increasing order
Lilongwe to the South and South West, Kasungu then ranks were assigned to the arranged data
to the North West, Ntchisi to the North, and with the lowest dataset having the highest order.
Salima to the East. The total land area is roughly
Secondly, estimates of exceedance probabilities
3041 km2 and constitutes 3.2 percent of the total
associated with historic observations were
size of the country. The district ranks 7th out of 9
districts in the region in terms of size of land and calculated using Gringorten plotting position
has population density of 210 people per km 2. formula which is given in equation 1:
The district altitude varies from 1,036 m to 4,000
i−a
m above mean sea level. The people in the q i=
district depend on agriculture as their main N + 1−2 a
source of living. (Equation 1)

Where, q i= Exceedance probability associated

with a specific observation; N = Number of
annual maxima observations (30 in this case); i =
Rank of specific observation with i=1 being the
largest to i=N being the smallest; a = constant
for estimation=0.44. From this value, the non-
exceedance probability of the distribution can be
determined using the equation 2 below:

pi=1−qi (Equation 2)

where pi is the non-exceedance probability of the

distribution.

2.1. Statistical Definition of Return Period

Assuming that X is a random variable which has

a cumulative distribution function, F x (X ). The
probability that X is less than or equal to a given
event X p is given as:

F x ( X )=P ( X ≤ X p ) =p
(Equation 3)

The probability that this event will be exceeded is

then equal to 1− p and the percent exceedance is
PAPER ID [Link]
2
 © IJCIRAS | ISSN (O) - 2581-5334
December 2017 | Vol. 1 Issue. 1

denoted as 1 00(1− p) . For such an event, X p , √ 6 S 2x

the return period corresponding to this α= (Equation
π
exceedance probability is denoted by T . 10)
Therefore, the return period is given by:
Using the Gumbel distribution described above,
1 T} rsub {p} theoretical ¿is then calculated in the
T= (Equation 4)
(1− p) same way as T} rsub {p} estimated ¿ .

Using this definition, the 100-year return period Rainfall analysis curve is then plotted in Excel. If
can be understood as an event with a probability an eye fit to this plot suggest a straight line, then
of exceedance 1− p=0.01 or a non-exceedance it is reasonable to conclude that the Gumbel’s
probability, p=0.99. In other words, there is a distribution is a good fit for the observed rainfall
99 % chance that this event will not be exceeded data. In case, it does not, Hydrognomon software
may be used to test other probability distributions
within a given year. Using this concept of T ,
such as Log Normal or Log Pearson Type III.
estimated distribution of the 30 years of data was
determined as T} rsub {p} estimated ¿ and
evaluated as (T = 1/(1-p)). The assumption at this
level is that the data follows a specific
distribution, namely the 'Gumbel' or Extreme
Value Type 1' distribution. The Gumbel
3. RESULTS
distribution is given as follows:
The annual rainfall plot and maximum versus
mean rainfall data for Dowa district for the 30

[ (
F x ( x ) =exp −exp
(Equation 5)
α )]
−x −u
=p year period collected from Blantyre
Meteorological station are presented in Figure 1
and Figure 2 respectively below.

where x is the observed discharge data, and u

and α are the calculated parameters of the
distribution. This distribution is then used to
calculate the theoretical estimate of p.
specifically p theoretical . This p theoretical , is
calculated as below:

[ { ( )}]
p theoretical=exp −exp −1∗

(Equation 6)
x−u
α
Figure 2: Annual rainfall pattern of Dowa district

Where x, u and α are determined as below:

n
xi
x=∑ (Equation 7)
i=1 n
n
1
2
S=
x ∑
(n−1) i=1
( xi −x )
2

Figure 3: Maximum versus Mean Annual Rainfall

(Equation 8)
Table 1 and Table 2 show descriptive statistics
u=x−0.5772 α (Equation 9) and regression analysis results for the maximum
annual rainfall for the district respectively.

PAPER ID [Link]
3
 © IJCIRAS | ISSN (O) - 2581-5334
December 2017 | Vol. 1 Issue. 1

Rainfall statistical analysis was used to determine

the measures of central tendency (mean, range,
maximum, minimum etc.) and dispersion
(standard deviation, coefficient of variation,
reduced centred index, coefficients of skewness
and kurtosis). COV is coefficient of variation, RCI

rainfall. The equation 𝑌 = m𝑋 + c, was used

is reduced centred index and PE is the effective

where 𝑋 is the independent variable (the year)

and 𝑌 is the dependent variable (rainfall). The
slope line is m and c is the intercept (value of 𝑌
when 𝑋 = 0). The sign of the slope defined the
direction of trend variable: increasing if the sign is
positive and decreasing if the sign is negative.
Mean Max Min Variance [Link] Skew Kurtosis COV RCI PE
JAN 239.86 413.90 113.70 6151.74 78.43 0.58 0.19 0.33 5.44 166.89
FEB 194.36 328.90 58.30 4798.34 69.27 0.07 -0.46 0.36 4.02 130.49
MAR 123.37 315.20 12.50 6807.07 82.50 0.77 -0.22 0.67 1.81 73.70
APR 23.57 94.80 0.00 569.52 23.86 1.71 2.36 1.01 -1.30 4.14
MAY 3.10 45.00 0.00 80.05 8.95 3.91 17.27 2.89 -1.94 0.00
JUN 0.47 4.80 0.00 1.67 1.29 2.84 7.25 2.77 -2.02 0
JUL 0.72 12.10 0.00 5.67 2.38 4.20 19.14 3.31 -2.01 0
AUG 0.28 2.40 0.00 0.50 0.71 2.53 5.16 2.55 -2.03 0
SEP 0.71 11.90 0.00 5.52 2.35 4.17 18.88 3.29 -2.01 0
OCT 6.47 56.80 0.00 158.08 12.57 2.90 9.06 1.94 -1.83 0
NOV 43.27 156.40 0.00 1377.78 37.12 1.29 1.77 0.86 -0.69 15.96
DEC 147.82 346.00 11.80 4292.18 65.51 0.66 2.10 0.44 2.57 93.26
Table 3: Computation table

Table 1: Descriptive statistics for the maximum

rainfall data for Dowa district

Figure 4. Rainfall frequency curve

Table 2: Regression analysis of rainfall for the
maximum rainfall data in Dowa district
The rainfall frequency curve was plotted where
The Gumbel’s distribution analysis was done return period (T) in years was linked with
following the above analysis and the results maximum rainfall. From the rainfall frequency
obtained are shown in Table 5. A plot of return curve above, the following values in Table 4 were
period v/s maximum rainfall was plotted for Dowa determined for different return periods.
district, which is shown in Figure 6.
Return Period, T Maximum rainfall
(years) (mm)
2 272.533
5 340.409
10 385.349
20 428.457
25 442.131
50 484.255
Table 4. Predicted maximum rainfall

PAPER ID [Link]
4
 © IJCIRAS | ISSN (O) - 2581-5334
December 2017 | Vol. 1 Issue. 1

It should be noted that the curve follows the respect to Gumbel, Log Normal and Log Pearson
Gumbel distribution very well for smaller Type III distributions. Using this curve, maximum
maximum values flows but drifts away from the rainfall values corresponding to any return period
theoretical distribution at higher values. This is from 1 to 100 can be predicted. The table below
why it is wise to use multiple distributions such as shows the predicted values for 2, 5, 10, 20, 25, 50
Log-normal and Log Pearson type III and check and 100 years.
which distribution works best for a specific site.
Hydrognomon version 4 software was also used Return Gumbel Log Log
to validate the results. Gumbel , Log-Normal and Period, T Normal Pearson
Log Pearson Type III distributions were compared (years) III
using Gringorten plotting position to see if there 2 272.533 275.344 275.401
are significant differences in the predicted rainfall 5 340.409 344.011 345.380
values. Figures 3, 4 and 5 show the results. 10 385.349 386.472 388.773
20 428.457 425.461 428.689
Gringorten LogNormal LogPearsonIII Gumbel Max 25 442.131 437.541 441.070
Exceedance probability (%) - scale: Normal distribution 50 484.255 474.066 478.538
99.95%

Table 5. Predicted maximum rainfall for different

99.8%
99.5%

.05%
99%
98%

95%

90%

80%
70%
60%
50%
40%
30%
20%

10%

.5%
.2% probability distributions
5%

2%
1%

1,000

4. DISCUSSION
Table 1 shows that the district receives mean and
effective rainfall from November to March.
100
Understanding the behavior of this data would
enable irrigation engineers and water planners to
design effective rainwater harvesting structures
mm

to save the water for irrigation for the months

10 when the effective rainfall is very low ( from April
to October). The RCI parameter indicates the
relative sequences of wet and dry conditions. An
RCI of 1 and above shows wet conditions and
1 below 1 show dry conditions. The RCI indicates
-3 -2 -1 0 1 2 3 that Dowa district has a near normal to
moderately dry sequences. Skew and kurtosis
Figure 5. Probability distributions were analyzed to check whether the annual and
seasonal rainfall data was normally distributed.
Probability Density Functions (PDF) - Histogram
Skew is a measure of evenness, or symmetry. If
LogNormal LogPearsonIII Gumbel Max data shows a mirror image from the centre point,
it is said to be symmetric. Normally distributed
data has a skew factor of zero or values near
zero. Positive values for the skew indicate that
data is skewed to the right and negative values
for skew indicate data is skewed to the left. Dowa
district rainfall is positive indicating skew to the
right. Kurtosis on the other hand is a measure of
data peakedness or flatness relative to a normal
distribution. When the coefficient of kurtosis is
zero, it means the data is normally distributed.
When coefficient of kurtosis is positive, it means
0 50 100 150 200 250 300 350 400 450 500 550 600 650 data has a peaked distribution and when kurtosis
Figure 6 . Probability Density Function Histogram
is negative, it indicates a flat distribution. Dowa
for Probability distributions rainfall is peaked.

Table 5 shows the comparison of the theoretical From the preliminary study and statistical
distribution and maximum rainfall data with analysis as shown in Table 4, variation in rainfall
is found to be statistically insignificant (p>0.05).
PAPER ID [Link]
5
 © IJCIRAS | ISSN (O) - 2581-5334
December 2017 | Vol. 1 Issue. 1

expected maximum monthly rainfall as

Table 5 presents the calculation charts from determined by the return period.
which the rainfall frequency analysis curve (Figs.
2), was generated. It was observed that the data 5. CONCLUSION
followed Gumbel distribution with Gringorten From the rainfall probability analysis on the
plotting position. The distribution gave similar rainfall of Dowa district, it is evident that Gumbel
values near the center of the distribution but Distribution (Extreme Value Type – I) is
varied considerably in the tails as seen in Figure ascertained as the best fit distribution type.
2. It should be noted however that the curve
follows the Gumbel distribution very well for The more practical result of this paper was the
smaller maximum values flows but drifts away return period calculation using Gumbel
from the theoretical distribution at higher values. distribution for Dowa district. Here, 2, 5, 10, 20,
Furthermore, Figure 2 was used to predicting 25 and 50-year return periods of rainfall return
maximum rainfall with a return periods of 2, 5, levels were illustrated. An irrigation engineer or
10, 20, 25 and 50-year periods. From the policy maker making a risk analysis for a 50 year
Gumbel’s distribution using return periods (T) as plan could use the 50 year return period result of
shown in Table 6 above, the estimated maximum projected maximum monthly rainfall to determine
monthly rainfall obtained are: 272.5 mm, 340.4 water rainfall harvesting structures, determine
mm, 385.3 mm, 428.5 mm, 442.1 mm and 484.3 height of flood control dyke height, flooding risks,
mm respectively. These values are useful for damage projections, etc. The 100 year return
irrigation systems designing, reservoir designing, period results are high intensity but low
storm management and water balance modelling probability which has an indirect link with 10 year
in the district. The Gumbel distribution gave the return period results. It depends on the engineer
best fit for maximum rainfall. or policy makers to choose the duration and risk
level. The results of this study can be used to
Figures (3 and 4) were generated from develop better strategies of risk and damage
Hydrognomon software to further compare three from extreme rainfall events and flooding. In
distributions namely Gumbel, Log Normal and Log addition, Knowledge of the pattern of extreme
Pearson Type III tied to Gringorten plotting rainfall will be useful in various disciplines such as
position. The figures definitely show that Gumbel the irrigation, agronomy, environment, and civil
distribution fitted the data well than the other two engineering.
distribuitons. This agrees with Eslamin et al. [10]
who found out that Gumbel distribution best fits ACKNOWLEDGEMENT
for maximum monthly rainfall in Iran. The authors acknowledge the Director of Blantyre
Meteorological Services for providing rainfall data,
Ewemoje [11] noted that best distributions from the District Commissioner of Dowa District for
Log Normal and Log Pearson Type III normally providing district profile and irrigation engineers
emerge with high coefficient of determinations in Dowa district for providing feedback on the
when tied with Weibull plotting position. This usage of the results found by this research.
therefore would however have been improved if
the study changed plotting position from REFERENCES
Gringorten to Weibull. The use of Gumbel
distribution with Gringorten plotting position [1] Mondal A, Khare, Kundu, “Spatial and
hence is suggested for the prediction of 50-year temporal analysis of rainfall and temperature
maximum rainfall return period for Dowa district. trend of India”, Theor Appl Climatol , 122:143–
Observed deviations from the distributions were 158, 2015
small when the data was transformed on a
logarithmic scale. This could be a good standard [2] Singh R.K., “Probability analysis of rainfall for
of transforming data to a log scale to check crop planning and watershed management in
distribution characteristics. Sikkim”, Journal of Hill Research, 11(2): 239-
2411998,1998
The information of return periods can help guide
irrigation engineers, policy makers and planners [3] Vaidya V.B., Karande B.I., Pandey V.,
in water sector. They can more easily make their Lunagaria M.M. and Shekh A.M., “Rainfall
decisions for a given span of time using the probability analysis for crop planning in Gujarat
PAPER ID [Link]
6
 © IJCIRAS | ISSN (O) - 2581-5334
December 2017 | Vol. 1 Issue. 1

state”, Journal of Agrometeorology, 10(Special

issue 1): 183-185, 2008.

[4] Suchit K.R., Sunil K., Arvind K.R., Satyapriya

and Dana R.P., “Climate Change Variability and
Rainfall Probability for Crop Planning in Few
Districts of Central India”, Atmos. Climate Sci. 4
394-403, 2014

[5] C.R. Suribabu, G. Arvind, P.A. Kumar, S.

GirishKarthi, “Statistical Analysis of 30 years
Rainfall Data: A Case Study”, IOP Conf. Series:
Earth and Environmental Science 80 (2017)
012067, 2017. doi :10.1088/1755-
1315/80/1/012067

[6] J. R. Stedinger and R. M. Vogel, “Frequency

analysis of extreme events.” Handbook of
Hydrology, chapter 18, McGraw-Hill, New York,
1993. Vol. 3, No. 10, 2013.

[7] N. Mujere, “Flood frequency analysis using the

Gumbel distribution.” International Journal on
Computer Science and Engineering (IJCSE), vol. 3,
pp. 2774–2778, July 2011.

[8] U. N. Ahmad, A. Shabri, and Z. A. Zakaria,

“Flood frequency analysis of annual maximum
stream flows using L-Moments and TL-Moments.”
Applied Mathematical Sciences, vol. 5,pp. 243–
253, 2011.

[9] N. Bhagat. Flood Frequency Analysis Using

Gumbel's Distribution Method: A Case Study of
Lower Mahi Basin, India. Journal of Water
Resources and Ocean Science. Vol. 6, No. 4,
2017, pp. [Link]:
10.11648/[Link].20170604.11

[10] S.S. Eslamian and H. Feizi,” Maximum

Monthly Rainfall Analysis Using L-Moments for an
Arid Region in Isfahan Province”, Iran. J. Appl.
Meteorol. Climatol, 46, 494–503, 2007

[11] T.A. Ewemoje and O.S. Ewemooje ,” Best

Distribution and Plotting Positions of Daily
Maximum Flood Estimation at Ona River in Ogun-
Oshun River Basin, Nigeria”, Agricultural
Engineering International: CIGR Journal. Vol.13,
No.3, 2011. Manuscript No.1380

PAPER ID [Link]
7