Journal of Research in Mathematics Trends and Technology (JoRMTT) Vol. 01, No.

02, 2019 | 39-45

Journal of Research in Mathematics Trends and

Technology

Probability Distribution of Rainfall in Medan

Elly Rosmaini1* and Yoni Yolanda Saphira1
1
Department of Mathematics, Universitas Sumatera Utara, Medan, 20155, Indonesia

Abstract. In this paper we chose three stations in Medan City, Indonesia to estimate Monthly
Rainfall Data i.e. Tuntungan, Tanjung Selamat, and Medan Selayang Stations. We took the
data from 2007 to 2016. In this case fitted with Normal, Gamma, and Lognormal
Distributions. To estimate parameters, we used this method. Furthermore, Kolmogorov-
Smirnov and Anderson Darling tests were used the goodness-of-fit test. The Gamma and
Normal Distributions is suitable for Tuntungan and Medan Selayang Stations were stated by
Kolmogorov-Smirnov's test. Anderson Darling's test stated that Gamma Distribution was
suitable for all stations.

Keyword: Probability Distribution, Test of Goodness-Of-Fit, The Method of Moments.

Abstrak. Pada makalah ini kami memilih tiga stasiun pada kota Medan, Indonesia untuk
memperkirakan Data Curah Hujan Bulanan yaitu Stasiun Tuntungan, Tanjung Selamat dan
Medan Selayang. Kami mengambil data dari 2007 sampai 2016. Pada kasus ini dicocokkan
dengan distribusi Normal, Gamma dan Lognormal. Untuk memperkirakan parameter, kami
menggunakan metode ini. Lebih lanjut, uji Kolmogorov-Smirnov dan Anderson Darling
digunakan untuk uji goodness-of-fit. Distribusi Gamma dan Normal cocok untuk stasiun
Tuntungan dan Medan Selayang ditetapkan oleh uji Kolmogorov-Smirnov. Uji Anderson
Darling menyatakan bahwa distribusi Gamma cocok untuk semua stasiun.

Kata Kunci: Distribusi Probabilitas, Uji Goodness-Of-Fit, Metode dari Momen

Received 25 April 2019 | Revised 14 June 2019 | Accepted 24 July 2019

1. Introduction

The distribution function of rainfall and rainfall prediction in Libya has been studied by Sen and
Eljadid (1999). Distribution of monthly rainfall in Libya has met the Gamma probability
distribution functions are tested using Chi-Square test. The series of all rainfall recorded at least
20 years in Libya investigated statistically and Gamma distribution parameters are calculated on
existing stations. Prediction rainfall amounts around 10, 25, 50, and 100 mm is achieved make
use of the probability function.

Suhaila Abdul Aziz Jamaludin and (2007) classified daily rainfall data according to four types of
rain (types 1, 2, 3, and 4). Gamma, Weibull, Kappa, and mixed Exponential are four distributions
which are tested to conform to the number of daily rainfall in Malaysia peninsula. Parameter
estimation for each distribution uses the maximum likelihood method. Based on tests of goodness-

*Corresponding author at: Universitas Sumatera Utara, Medan, 20155, Indonesia

E-mail address: elly1@usu.ac.id

Copyright ©2019 Published by Talenta Publisher, e-ISSN: 2656-1514, DOI: 10.32734/jormtt.v1i2.2835

Journal Homepage: http://talenta.usu.ac.id/jormtt
Journal of Research in Mathematics Trends and Technology (JoRMTT) Vol. 01 , No. 02, 2019 40

of-fit, Mixed Exponential distribution declared as the most appropriate distribution to describe
the amount of daily rainfall in Malaysia peninsula.

Based on the book "Statistical Methods in Hydrology" (Hann CT, 1977), it has summarized that
the Gamma distribution is the most appropriate for monthly and annual precipitation. In the book
"Arid Lands and Water Evaluation and Management" (Maliva and Missimer, 2012), has shown
that one of the distribution Normal, Log-Normal, Gamma, Weibull and Gumbel in accordance
with the rainfall data from the arid and semi-arid.

We take three stations as samples for the area of Medan and use the Normal distribution, Gamma
and Log-Normal. The goodness-of-fit test is used by Kolmogorov-Smirnov dan Anderson
Darling. The rainfall pattern at the three stations in Medan can be used as consideration in
predicting rainfall so that it can anticipate problems such as problems in the development of
infrastructure and flood prevention in Medan.

2. Materials and Methods

a) Normal distribution
A random variable 𝑋 is said to be normally distributed, if its density function has the form
(Montgomery and Runger, 2014):

1 −(𝑥−𝜇)2
[ ]
𝑓(𝑥) = .𝑒 2𝜎 2 ; −∞ < 𝑥 < ∞ (1)
√2𝜋𝜎 2

with parameter 𝜇 where −∞ < 𝜇 < ∞ and 𝜎 > 0.

Mean of normal distribution: 𝐸(𝑋) = 𝜇

Variance of the normal distribution: 𝑉𝑎𝑟(𝑋) = 𝜎 2

b) Gamma distribution
A random variable 𝑋 is said to be Gamma distributed if and only if its density function has
the form (Montgomery and Runger, 2014):

𝜆𝑟 𝑥 𝑟−1 𝑒 −𝜆𝑥
𝑓(𝑥) = ; 𝑥 > 0, 𝜆 > 0, 𝑟 > 0 (2)
𝛤(𝑟)

𝑟
Mean of gamma distribution: 𝐸(𝑋) = 𝜆

𝑟
Variance gamma distribution: 𝑉𝑎𝑟(𝑋) = 𝜆2

c) Lognormal Distribution
Let 𝑊 has a normal distribution with mean 𝜃 and variance 𝜔2 , then = exp(𝑊) is a random
variable which is lognormal distributed with the following probability density function
(Montgomery and Runger, 2014):

1 (𝑙𝑛(𝑥)−𝜃)2
[− ]
𝑓(𝑥) = 𝑒 2𝜔2 ; 0<𝑥<∞ (3)
𝑥𝜔√2𝜋
𝜔2
Mean of Lognormal distribution: 𝐸(𝑋) = 𝑒 𝜃+ 2
2 2
Variance of Lognormal distribution: 𝑉𝑎𝑟(𝑋) = 𝑒 2𝜃+𝜔 (𝑒 𝜔 − 1)
Journal of Research in Mathematics Trends and Technology (JoRMTT) Vol. 01 , No. 02, 2019 41

d) Moment Method
Let 𝑋1 , 𝑋2 , … , 𝑋𝑛 are random samples of probability distribution 𝑓(𝑥) where 𝑓(𝑥) can be
discrete probability mass function or continuous probability density function. The 𝑘-th
pulation moment is (Wackerly, Mendenhall, and Scheaffer, 2008):

𝜇𝑘′ = 𝐸(𝑋 𝑘 ) (4)

The appropriate 𝑘-th sample moment is:

𝑛
1
𝑚𝑘′ = ∑ 𝑋𝑖𝑘 (5)
𝑛
𝑖=1

Moment method is based on the intuitive idea that states that the moment of the sample
should provide a good estimate from appropriate population moment. That is, 𝑚𝑘′ has to be
a good estimation of 𝜇𝑘′ , where 𝑘 = 1, 2, …. Next, since population moment 𝜇1′ , 𝜇2′ , … , 𝜇𝑘′
is a function of population parameter, then moment of population can be equated to the
moment of sample and completing the desired estimator. Therefore, the method of
momentscan be expressed as follows:

Choose as estimation of parameter values which is a solution of the equation, 𝜇𝑘′ = 𝑚𝑘′ ,
where 𝑘 = 1, 2, … , 𝑡, and 𝑡 is the number of parameters to be estimated.

e) Test goodness-of-fit
The procedure steps to execute the Kolmogorov-Smirnov test to give a collection of sample
values 𝑥1 , 𝑥2 , … , 𝑥𝑖 examined from the population 𝑋 by article Soong (2004) is as follows:
a. Arranging the sample values from the smallest to biggest, denoted by 𝑥𝑖 .
𝑖
b. Determining the observed distribution function 𝐹 0 [𝑥𝑖 ] at each 𝑥𝑖 using 𝐹 0 [𝑥𝑖 ] = 𝑁.
c. Determining theoretical distribution function 𝐹𝑥 [𝑥𝑖 ] at each 𝑥𝑖 using distribution
hypothesis which has been obtained and the deviation is determined from the equation:
𝑑2 = max{|𝐹 0 [𝑥𝑖 ] − 𝐹𝑥 [𝑥𝑖 ]|} (6)
d. The maximum absolute value 𝑑2 obtained from the equation 𝑑2 = max{|𝐹 0 [𝑥𝑖 ] −
𝐹𝑥 [𝑥𝑖 ]|} , compared to the critical value shown in the statistics table. If the critical value
is greater than equal 𝑑2 , then d tested distribution is suitable to describe the observed
data, if it is not holds, that is the critical value greater than equal 𝑑2 , we get the tested
distribution is not suitable for Kolmogorov-Smirnov, which is shown as follows
(Lothar, 1984):
Table 1 The critical value of Kolmogorov-Smirnov test

limit to 𝑑2 𝛼
1,073
0,20
√𝑛
1,138
0,15
√𝑛
1,224
0,10
√𝑛
1,358
0,05
√𝑛
1,628
0,01
√𝑛
 Journal of Research in Mathematics Trends and Technology (JoRMTT) Vol. 01 , No. 02, 2019 42

Anderson Darling test is used to determine the distribution of the sample data. This test is
a modification of the Kolmogorov-Smirnov test. Anderson Darling test uses a specific data
distribution in calculating critical value.

The formula used to test Anderson Darling is:

𝑛
1 − 2𝑖
𝐴𝐷 = [∑ {ln(𝐹0 [𝑍𝑖 ]) + ln(1 − 𝐹0 [𝑍𝑛−𝑖+1 ])}] − 𝑛 (7)
𝑛
𝑖=1
where 𝐴𝐷 = Anderson Darling Test
𝑛 = Amount of data
𝐹 = Cumulative distribution function
Anderson Darling test uses the critical value according to the tested probability distribution
as follows (Ang and Tang, 2007):
0,3
0,2 + (8)
∗
𝐴 = 𝐴𝐷 + 𝛼 ; 𝑓𝑜𝑟 𝑔𝑎𝑚𝑚𝑎 𝑑𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛
𝑛
𝛼
𝐴∗ = ; 𝑓𝑜𝑟 𝑛𝑜𝑟𝑚𝑎𝑙 𝑎𝑛𝑑 𝑙𝑜𝑔𝑛𝑜𝑟𝑚𝑎𝑙 𝑑𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛
0,75 2,25 (9)
1+ + 2
𝑛 𝑛
where 𝐴∗ = Critical value of Anderson Darling
𝑛 = Amount of data
𝛼 = The level of significance
Table 2 𝛼 significance level of Anderson Darling test

𝛼 Level of Significance
K 0.25 0.10 0.05 0.025 0.01 0.005
1 0.486 0.657 0.786 0.917 1.092 1.227
2 0.477 0.643 0.768 0.894 1.062 1.190
3 0.475 0.639 0.762 0.886 1.052 1.178
4 0.473 0.637 0.759 0.883 1.048 1.173
5 0.472 0.635 0.758 0.881 1.045 1.170
6 0.472 0.635 0.757 0.880 1.043 1.168
8 0.471 0.634 0.755 0.878 1.041 1.165
10 0.471 0.633 0.754 0.877 1.040 1.164
12 0.471 0.633 0.754 0.876 1.038 1.162
15 0.470 0.632 0.754 0.876 1.038 1.162
20 0.470 0.632 0.753 0.875 1.037 1.161
∞ 0.470 0.631 0.752 0.873 1.035 1.159

3. Results and Discussion

The method of moment is used to estimate the parameters in the Normal, Gamma and Lognormal
distribution. The result of parameter estimation by moment method using equation (4) and (5)
shown in the following table:

Table 3 The result of parameter estimation using Method of Moments

Station Normal Gamma Lognormal

𝜇̂ 𝜎̂ 2
𝜆̂ 𝑟̂ 𝜃̂ ̂2
𝜔
Medan
229,742 14.885,5416 0,015433903 3,545815743 5,312735695 0,24843959
Selayang
Tanjung
182 17.288 0,010527534 1,916011188 4,994033351 0,419946678
Selamat
Tuntungan 241,925 20.636,786 0,011722998 2,836086291 5,33761395 0,30202762
 Journal of Research in Mathematics Trends and Technology (JoRMTT) Vol. 01 , No. 02, 2019 43

Parameter estimation results are used to the test goodness-of-fit, in this case, Kolmogorov-
Smirnov and Anderson Darling test. This goodness-of-fit test is used to examine the appropriate
distribution of the monthly rainfall in Medan Selayang, Tanjung Selamat, and Tuntungan stations.
The results of the goodness of-fit test for Normal Distribution, Gamma and Lognormal are as
follows:

Table 4 The results of Goodness-of-Fit test using the Kolmogorov-Smirnov

Station 𝑛 Critical value Normal Gamma Lognormal

Medan
120 0,123237575 0,102710069 0,102710011 0,424515726
Selayang
Tanjung
120 0,123237575 0,125003484 0,125003474 0,39424123
Selamat
Tuntungan 120 0,123237575 0,086193286 0,086193283 0,407912624
Based on the results of the Kolmogorov-Smirnov test using equation (6) and Table 1, it is clear
that the Normal and Gamma Distribution is apropriate with the monthly rainfall data in Medan
Selayang and Tuntungan stations. While the monthly rainfall data at Tanjung Selamat station is
not appropriate with the three distributions.

Table 5 The results of Goodness-of-Fit test using Anderson Darling

Station Normal Gamma Lognormal

∗ ∗ ∗
𝐴 𝐴𝐷 𝐴 𝐴𝐷 𝐴 𝐴𝐷
Medan
0,747213166 1,71859701 1,723586083 1,718594948 0,747213166 1,529518631
Selayang
Tanjung
0,747213166 3,277427696 3,282419502 3,277428367 0,747213166 1,486938223
Selamat
Tuntungan 0,747213166 1,791307447 1,796298604 1,791307469 0,747213166 1,518192187
Based on the result of Anderson Darling test using equation (7), (8), (9), and Table 2, it is easy to
see that the Gamma distribution is appropriate with all the stations in this study.

Histogram
0,004

0,0035

0,003

0,0025
Densitas

0,002

0,0015

0,001

0,0005

0
0 100 200 300 400 500 600
Curah hujan (mm)

Figure 1 Histogram of Medan Selayang station

Journal of Research in Mathematics Trends and Technology (JoRMTT) Vol. 01 , No. 02, 2019 44

Histogram
0,0045
0,004
0,0035
0,003

Densitas
0,0025
0,002
0,0015
0,001
0,0005
0
0 200 400 600 800
Curah hujan (mm)

Figure 2 Histogram of Tanjung Selamat station

Histogram
0,0035

0,003

0,0025
Densitas

0,002

0,0015

0,001

0,0005

0
0 200 400 600 800
Curah hujan(mm)

Figure 3 Histogram of Tuntungan station

4. Conclusion

The results of Kolmogorov-Smirnov goodness-of-fit test states that the Normal and Gamma
Distribution is appropriate with the monthly rainfall data in Medan Selayang and Tuntungan
stations. The results of Anderson Darling goodness-of-fit stated that the Gamma distribution is
appropriate with all of the stations in this study.

REFERENCES

[1] A. HS. Ang and W. H. Tang, Probability Concepts in Engineering: Emphasis on

Applications to Civil and Environmental Engineering, 2nd Ed., John Willey and Sons,
Hoboken, New Jersey, 2007.
[2] C. T. Haan, Statistical Methods in Hydrology, The Iowa State University Press, Ames,
1977.
[3] R. Maliva and T. Missimer, Arid Lands and Water Evaluation and Management, Springer,
Berlin, 2012.
[4] D. Montgomery and G. C. Runger, Applied Statistics and Probability for Engineers, John
Wiley and Sons, USA, 2014.
[5] L. Sachs, Applied Statistics: A Handbook of Techniques, Springer, Verlag, 1984.
Journal of Research in Mathematics Trends and Technology (JoRMTT) Vol. 01 , No. 02, 2019 45

[6] Z. Sen and A. G. Eljadid, “Distribution Function for Libya and Rainfall Prediction”,
Hydrological Sciences Journal, Istanbul, Turkey, vol. 44, no. 5, pp. 665-680, 1999.
[7] S. Jamaludin and A. A. Jemain, “The Statistical Distributions to The Daily Rainfall Amount
in Peninsular Malaysia”, Technology Journal, vol. 46C, pp. 33-48, 2007.
[8] T. T. Soong, Fundamentals of Probability and Statistics for Engineers, John Wiley and
Sons, New York, USA, 2004.
[9] D. Wackerly, W. Mendenhall, and R. L. Scheaffer, Mathematical Statistics with
Applications, Thomson Brooks / Cole, USA, 2008.

Attribution-NonCommercial-ShareAlike 4.0 International License. Some rights reserved.