Journal of Civil Engineering and Urbanism  
Volume 10, Issue 3: 24-31; May 25, 2020  
ISSN-2252-0430  
Comparison of Estimators of Probability Distributions for  
Selection of Best Fit for Estimation of Extreme Rainfall  
Vivekanandan N  
Central Water and Power Research Station, Pune, Maharashtra, India  
Corresponding author’s Email: anandaan@rediffmail.com  
ABSTRACT  
Extreme Value Analysis (EVA) of rainfall is considered as one of the important aspects to arrive at a design value  
for planning, design and management of civil and hydraulic structures. This can be achieved by fitting Probability  
Distribution (PDs) to the series of observed annual 1-day maximum rainfall data wherein the parameters of PDs are  
determined by method of moments and L-Moments (LMO). In this paper, a study on comparison of Extreme Value  
Type-1 (EV1), Extreme Value Type-2, Generalized Extreme Value (GEV) and Generalized Pareto distributions  
adopted in EVA of rainfall for Anakapalli, Atchutapuram, Kasimkota and Parvada sites is carried out. The selection  
of best fit PD for EVA of rainfall is made through quantitative assessment by using Goodness-of-Fit (viz., Chi-  
square and Kolmogorov-Smirnov) and diagnostic (viz., root mean squared error) tests; and qualitative assessment  
by using the fitted curves of the estimated rainfall. On the basis of evaluation of EVA results through quantitative  
and qualitative assessments, the study indicates the extreme rainfall given by EV1 (LMO) distribution could be used  
for the purpose of economical design. The study also indicates the extreme rainfall obtained from GEV (LMO)  
distribution may be considered for the design of civil and hydraulic structure with little risk involvement.  
Keywords: Chi-square, Extreme value analysis, Extreme Value Type-1, Kolmogorov-Smirnov, L-Moments,  
Method of moments, Rainfall, Root mean squared error  
INTRODUCTION  
from which the sample was drawn. It is also reported that  
For planning, design and management of civil and  
hydraulic structures, Extreme Value Analysis (EVA) of  
rainfall is generally considered as one of the important  
aspects to arrive at a design value. This can be achieved by  
fitting Probability Distributions (PDs) to the series of  
observed rainfall data. Depending on the size and the  
design-life of the structure, the estimated extreme rainfall  
corresponding to a desired return period is used (Mujere,  
2011).  
A number of PDs related to the families of normal,  
gamma and Extreme Value Distributions (EVD) are  
generally adopted in EVA of rainfall. Out of which, the  
Generalized Extreme Value (GEV), Extreme Value  
Type-1 (EV1), Extreme Value Type-2 (EV2) and  
Generalized Pareto (GPA) distributions are the members  
of EVD (Rao and Hamed, 2000). Generally, Method of  
Moments (MoM) is used in determining the parameters of  
PDs. Sometimes, it is difficult to assess the exact  
information about the shape of a distribution that is  
conveyed by its third and higher order moments. Also,  
when the sample size is small, the numerical values of  
sample moments can be very different from those of PD  
the estimated parameters of PDs fitted by MoM are often  
less accurate than those obtained by other parameter  
estimation procedures viz., maximum likelihood method,  
method of least squares and probability weighted moments  
(Acar et al., 2008). To address these shortcomings, the  
application of alternative approach, namely L-Moments  
(LMO) is used for EVA (Hosking, 1990). Number of  
studies has been carried out by different researchers  
showed that there is no unique distribution is available for  
EVA of rainfall for a region or country (Bhuyan et al.,  
out a study on developing empirical formula to estimate  
rainfall intensity in Riyadh region using EV1 (commonly  
known as Gumbel), LN2 and LP3. He concluded that the  
LP3 distribution gives better accuracy amongst three  
distributions studied in estimation of rainfall intensity.  
Baratti et al. (2012) carried out flood frequency analysis  
on seasonal and annual time scales for the Blue Nile River  
adopting Gumbel distribution. Esteves (2013) applied  
Gumbel distribution to estimate the extreme rainfall depths  
at different rain-gauge stations in southeast United  
To cite this paper: Vivekanandan N (2020). Comparison of Estimators of Probability Distributions for Selection of Best Fit for Estimation of Extreme Rainfall. J. Civil Eng.  
24  
J. Civil Eng. Urban.,10 (3): 24-31, 2020  
Kingdom. Rasel and Hossain (2015) applied Gumbel  
assessment by using Goodness-of-Fit (GoF) (viz., Chi-  
square (2) and Kolmogorov-Smirnov (KS)) and  
diagnostic (viz., Root Mean Squared Error (RMSE)) tests  
and qualitative assessment through the fitted curves of the  
estimated rainfall. This paper details the procedures  
adopted in EVD for EVA of rainfall with illustrative  
example and the results obtained thereof.  
distribution for development of intensity-duration-  
frequency curves for seven divisions in Bangladesh.  
Afungang and Bateira (2016) applied Gumbel distribution  
to estimate the maximum amount of rainfall for different  
periods in the Bamenda mountain region, Cameroon.  
Studies carried out by Sasireka et al. (2019) indicated that  
the extreme rainfall for various return periods obtained  
from Gumbel distribution could be used for design  
purposes by considering the risk involved in the operation  
and management of hydraulic structures in Tiruchirappalli  
region. However, when number of PDs adopted in EVA of  
rainfall, a common problem that arises is how to determine  
which distribution model fits best for a given set of data.  
This possibly could be answered by quantitative and  
qualitative assessments; and the results are also reliable. In  
this paper, a study on comparison of MoM and MLM of  
estimators of probability distributions for selection of best  
fit for estimation of extreme rainfall is carried out. The  
selection of best fit PD is made through quantitative  
MATERIAL AND METHODS  
The aim of the study is to select the best fit PD for EVA of  
rainfall. Thus, it is required to process and validate the  
data for application such as (i) select the PDs (viz., GEV,  
EV1, EV2 and GPA); (ii) select parameter estimation  
methods (viz., MoM and MLM); and (iii) conduct EVA of  
rainfall and analyse the results obtained thereof. The  
Cumulative Distribution Function (CDF), quantile  
estimator and parameters of GEV, EV1, EV2 and GPA  
distributions adopted in EVA of rainfall is presented in  
Table 1.  
Table 1. CDF, Quantile estimator and parameters of PDs  
Parameters of PDs  
Distri-  
bution  
Quantile estimator  
(RT)  
CDF  
MoM  
LMO  
(1(1k))  
z (2/(3  3 ) (ln(2)/ln(3))  
3 (2(13k )/(12k ))3  
k 7.817740z 2.930462z2 13.641492z3 17.206675z4  
  2k /(12k )(1k)  
R     
k
1/ k  
[1(ln(F))k ]  
1/ 2  
k
k(r)  
(12k) (1k)2  
1  
SR  
RT     
GEV  
F(r) e  
k
(13k) 3(1k)(12k) 23(1k)  
  (sign k)  
1/ 2  
(1k)  2 (1k)2  
  1 (((1k) 1) / k)  
  1 (0.5772157)  
  R (0.5772157)  
r  
F(r) ee  
RT    [ln(ln(F))]  
RT  e[ln(ln(F)] / k  
EV1  
EV2  
2  
ln(2)  
6
   
   
SR  
By using the logarithmic transformation of the observed data, parameters of  
EV1 are initially obtained by MoM and LMO; and are used to determine the  
parameters of EV2 from =exp() and k=1/(scale parameter of EV1).  
r
k  
F(r) e  
R    (/(1k))  
2R  2 /(12k)(1k)2  
CS 2(1k)(12k)1/2 /(13k)  
ξ λ1 (α /(1k))  
k (13τ3 ) /(τ3 1)  
α (1k)(2 k)λ2  
(1(1F)k )  
RT     
1/ k  
k(r  )  
F(r) 11  
GPA  
k
In Table 1,  
,
,
k
are the location, scale and shape  
), (or S ) and C  
distribution and given by 332; RT is the estimated  
extreme rainfall for a return period (T). A relation F, P and  
T is defined by F(r) = 1-P(RT ≥r) =1-P = 1-1/T.  
parameters, respectively; µ (or  
R
R
S
(or  
) are the average, standard deviation and Coefficient  
of Skewness respectively; F(r) (or F) is the CDF of r (i.e.,  
AMR); -1 is the inverse of the standard normal  
distribution function and -1=(P0.135-(1-P)0.135)/0.1975  
wherein P is the probability of exceedance; sign(k) is plus  
or minus 1 depending on the sign of k ; λ1, λ2 and λ3 are  
the first, second and third L-moments respectively; L-  
Skewness is a measure of the lack of symmetry in a  
Goodness-of-Fit (GoF) tests  
GoF tests are applied for checking the adequacy on  
fitting PDs to the observed rainfall data. Out of a number  
GoF tests available, the widely accepted GoF tests are 2  
and KS, which are used in the study. Theoretical  
descriptions of GoF tests statistic are given as below:  
2test statistic is defined by:  
25  
Vivekanandan, 2020  
2
NC  
are missing. However, the data for the missing years were  
(1)  
O j (r) E j (r)  
2  
E j (r)  
not considered in EVA of rainfall. For estimation of  
extreme (i.e., 1-day maximum) rainfall, the Annual 1-day  
Maximum Rainfall (AMR) series of each site was  
extracted from the corresponding daily rainfall data series  
and also used.  
j1  
where, Oj(r) is the observed frequency value of r for  
jthclass, Ej(r) is the expected frequency value of r for  
jthclass and NC is the number of frequency classes.The  
rejection region of 2 statistic at the desired significance  
2
2
level () is given by  
(Zhang, 2002). Here,  
C  1,NCm1  
Table 2. Descriptive statistics of AMR  
m denotes the number of parameters of the distribution  
and  
is the computed value of 2 statistic by PDs.  
Average  
(mm)  
SD  
(mm)  
Max.  
(mm) (mm)  
Min.  
Site  
CS  
CK  
C2  
Anakapalli  
Atchutapuram  
Kasimkota  
Parvada  
107.8  
115.1  
101.2  
98.8  
53.0  
66.9  
41.2  
41.7  
1.539 2.707  
2.588 8.485  
1.270 1.556  
36.8  
34.4  
35.7  
280.0  
378.2  
211.8  
179.0  
KS test statistic is defined by:  
N
(2)  
KS Max  
F
e ri  
FD  
ri   
i1  
where, Fe(ri)=M/(N+1) is the empirical CDF of ri and  
0.260 -0.870 31.2  
SD: Standard Deviation; CS: Coefficient of Skewness; CK: Coefficient of  
Kurtosis; Max: Maximum; Min: Minimum  
FD(ri) is the computed CDF of ri.Here, M denotes the rank  
assigned to the observed values arranged in ascending  
order and N is the number of sample values.  
Test criteria: If the computed values of GoF tests  
statistic given by PD are less than that of the theoretical  
values at the desired significance level, then the PD is  
found to be acceptable for EVA.  
RESULTS AND DISCUSSION  
By applying the procedures of EVA, as described above,  
the parameters of GEV, EV1, EV2 and GPA distributions  
were determined by MoM and LMO; and are used for  
estimation of extreme rainfall. The EVA results of  
Anakapalli, Atchutapuram, Kasimkota and Parvada sites  
are presented in Tables 3 to 6 while the plots are shown in  
Figure 1. For EVA results, it is noted that the estimated  
extreme rainfall by EV2 (LMO) was higher than the  
corresponding values of EV1, GEV and GPA distributions  
for the return periods from 50-year and above.  
Diagnostic test  
A selection of suitable PD for EVA of rainfall is  
carried out through RMSE, which is defined as:  
1/ 2  
N
2
1
(3)  
RMSE   
r r*  
i
i
N i1  
Here, ri and ri* are the observed and corresponding  
estimated extreme values by EVD. The distribution with  
minimum RMSE is considered as better suited distribution  
in comparison with the other PDs adopted in EVA (US  
Analysis based on GoF tests  
In the present study, GoF tests statistic values of GEV,  
EV1, EV2 and GPA distributions were computed and are  
presented in Table 7. Based on GoF tests results, it is  
noted that:  
Application  
In this paper, a study on evaluation of GEV, EV1, EV2  
and GPA distributions through quantitative and qualitative  
assessments for EVA of rainfall is carried out. The daily  
rainfall data (with some gaps) observed at Anakapalli for  
the period 1970 to 2017, Atchutapuram for the period  
1989 to 2017, Kasimkota for the period 1989 to 2017 and  
Parvada for the period 1992 to 2017 was used. Table 2  
gives the descriptive statistics of AMR for the sites  
considered in the study. From the scrutiny of the daily  
rainfall data, it was observed that the data for the  
intermittent period for Anakapalli (2004), Kasimkota  
i) 2test supported the use of GEV, EV1, EV2, and  
GPA distributions for EVA of rainfall for  
Anakapalli, Atchutapuram, Kasimkota and  
Parvada.  
ii) KS test confirmed the applicability of GEV, EV1  
and EV2 distributions for EVA of rainfall for  
Anakapalli, Atchutapuram and Kasimkota.  
iii) For Parvada, KS test results indicated that the PDs  
considered in the study are acceptable for EVA of  
rainfall while determining the parameters by MoM  
and LMO.  
26  
J. Civil Eng. Urban.,10 (3): 24-31, 2020  
Table 3. Estimated 1-day maximum rainfall (mm) by MoM and MLM of EVD for Anakapalli  
GEV  
EV1  
EV2  
GPA  
Return period  
(year)  
MoM  
LMO  
MoM  
LMO  
MoM  
LMO  
MoM  
LMO  
2
97.6  
94.6  
99.1  
99.4  
94.6  
90.3  
92.5  
92.9  
5
10  
143.1  
174.9  
206.8  
217.2  
250.1  
284.1  
319.5  
368.3  
407.0  
138.7  
172.9  
209.9  
222.7  
265.2  
312.7  
365.9  
446.3  
515.9  
145.9  
176.9  
206.6  
216.1  
245.1  
274.0  
302.7  
340.6  
369.3  
144.5  
174.4  
203.0  
212.1  
240.1  
267.9  
295.6  
332.1  
359.7  
129.3  
159.0  
193.9  
206.5  
250.6  
303.8  
368.0  
474.0  
573.8  
131.1  
167.8  
212.6  
229.2  
288.8  
363.3  
456.6  
617.3  
775.3  
144.4  
180.8  
215.0  
225.5  
256.8  
286.1  
313.6  
347.2  
370.8  
145.1  
181.1  
214.2  
224.3  
254.0  
281.4  
306.7  
337.1  
357.9  
20  
25  
50  
100  
200  
500  
1000  
Table 4. Estimated 1-day maximum rainfall (mm) by MoM and MLM of EVD for Atchutapuram  
GEV  
EV1  
EV2  
GPA  
Return period  
(year)  
2
MoM  
99.8  
LMO  
96.8  
MoM  
104.1  
LMO  
105.6  
MoM  
98.5  
LMO  
MoM  
LMO  
94.9  
94.0  
94.8  
5
10  
152.9  
193.5  
237.0  
251.9  
301.1  
355.5  
415.9  
506.1  
583.2  
144.4  
185.3  
233.5  
251.0  
312.3  
386.4  
476.1  
624.6  
765.0  
163.3  
202.4  
240.0  
251.9  
288.6  
325.1  
361.4  
409.2  
445.4  
156.7  
190.5  
222.9  
233.2  
264.9  
296.3  
327.7  
369.0  
400.3  
139.3  
175.3  
218.4  
234.2  
290.4  
359.5  
444.8  
588.9  
728.0  
140.5  
182.1  
233.7  
252.9  
322.6  
410.8  
522.7  
718.1  
913.0  
150.8  
196.6  
244.9  
261.0  
312.9  
367.7  
425.5  
507.0  
572.6  
150.6  
195.0  
241.5  
256.9  
306.2  
357.8  
411.7  
486.8  
546.6  
20  
25  
50  
100  
200  
500  
1000  
Table 5. Estimated 1-day maximum rainfall (mm) by MoM and MLM of EVD for Kasimkota  
GEV  
EV1  
EV2  
GPA  
Return period  
(year)  
MoM  
LMO  
MoM  
LMO  
MoM  
LMO  
MoM  
LMO  
2
94.1  
91.4  
94.5  
94.6  
91.0  
88.4  
90.1  
90.0  
5
10  
130.1  
154.5  
178.2  
185.8  
209.4  
233.2  
257.3  
289.6  
314.4  
126.6  
153.2  
181.5  
191.1  
222.6  
257.2  
295.2  
351.4  
398.9  
130.9  
155.0  
178.1  
185.4  
208.0  
230.4  
252.7  
282.2  
304.5  
130.3  
154.0  
176.7  
183.9  
206.1  
228.1  
250.1  
279.0  
300.9  
119.8  
143.7  
171.1  
180.9  
214.5  
254.1  
300.8  
375.7  
444.5  
121.7  
150.3  
184.2  
196.4  
239.5  
291.6  
354.9  
459.7  
559.0  
132.1  
159.8  
184.5  
191.8  
212.8  
231.5  
248.1  
267.3  
279.9  
131.9  
159.5  
184.2  
191.6  
212.8  
231.7  
248.6  
268.1  
281.0  
20  
25  
50  
100  
200  
500  
1000  
Table 6. Estimated 1-day maximum rainfall (mm) by MoM and MLM of EVD for Parvada  
GEV  
EV1  
EV2  
GPA  
Return period  
(year)  
MoM  
LMO  
MoM  
LMO  
MoM  
LMO  
MoM  
LMO  
2
96.6  
95.4  
92.0  
91.5  
88.5  
83.7  
95.1  
94.8  
5
10  
134.0  
154.4  
171.4  
176.3  
190.1  
202.0  
212.3  
223.9  
231.3  
134.4  
156.7  
175.8  
181.4  
197.6  
212.0  
224.9  
239.9  
249.9  
128.8  
153.2  
176.6  
184.0  
206.9  
229.6  
252.2  
282.0  
304.6  
131.2  
157.5  
182.7  
190.7  
215.3  
239.8  
264.2  
296.4  
320.7  
117.3  
141.4  
169.1  
178.9  
213.1  
253.6  
301.4  
378.7  
450.0  
119.9  
152.1  
191.1  
205.5  
256.8  
320.5  
399.5  
534.4  
665.9  
140.6  
159.5  
170.8  
173.4  
179.2  
182.7  
184.8  
186.3  
186.9  
141.1  
160.5  
172.2  
174.9  
181.0  
184.6  
186.8  
188.5  
189.2  
20  
25  
50  
100  
200  
500  
1000  
27  
Vivekanandan, 2020  
Table 7. Theoretical and computed values of GoF tests statistic by MoM and MLM of EVD  
Computed value  
Theoretical value  
Rain-gauge  
station  
2  
MoM  
LMO  
KS  
GEV  
GEV  
EV1  
EV2  
GPA  
GEV  
EV1  
EV2  
GPA  
EV1  
EV2  
GPA  
2 test statistic  
Anakapalli  
Atchutapuram  
Kasimkota  
Parvada  
3.936  
1.172  
4.000  
2.000  
3.936  
3.759  
2.846  
1.130  
6.489  
3.241  
1.692  
5.843  
5.212  
2.552  
2.846  
2.870  
2.660  
1.862  
2.846  
2.000  
3.936  
3.241  
3.615  
1.130  
0.872  
2.552  
1.308  
3.739  
5.212  
1.517  
2.846  
2.870  
7.82  
5.99  
5.99  
5.99  
7.82  
5.99  
5.99  
5.99  
7.82  
5.99  
5.99  
5.99  
5.99  
3.84  
3.84  
3.84  
-
-
-
-
KS test statistic  
Anakapalli  
Atchutapuram  
Kasimkota  
Parvada  
0.083  
0.085  
0.100  
0.100  
0.098  
0.125  
0.106  
0.116  
0.105  
0.102  
0.088  
0.167  
0.252  
0.446  
0.403  
0.069  
0.072  
0.078  
0.069  
0.095  
0.099  
0.104  
0.105  
0.103  
0.099  
0.098  
0.082  
0.145  
0.231  
0.471  
0.416  
0.065  
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
0.184  
0.228  
0.240  
0.253  
Table 8. RMSE values given by MoM and MLM of EVD  
GEV  
EV1  
EV2  
GPA  
Rain-gauge  
station  
MoM  
11.861  
24.310  
LMO  
10.647  
23.612  
MoM  
LMO  
MoM  
LMO  
MoM  
10.806  
22.792  
LMO  
Anakapalli  
12.736  
27.407  
13.280  
29.044  
16.368  
27.679  
11.793  
23.225  
11.026  
23.506  
Achutapuram  
Kasimkota  
Parvada  
10.290  
6.123  
9.731  
5.634  
10.388  
8.714  
10.643  
7.507  
13.614  
16.698  
10.476  
15.053  
9.689  
4.785  
9.712  
4.873  
ii) But, the rainfall estimates given by MoM are less  
accurate when compared to LMO because of the  
characteristics of moment estimators.  
Analysis Based on Diagnostic Test  
For the selection of suitable PD for EVA of rainfall,  
RMSE values were computed by EVD and the results are  
presented in Table 8. From the diagnostic test results, it is  
observed that:  
iii) Alternatively, GEV (LMO) for Atchutapuram  
while GPA (LMO) for Kasimkota and Parvada is  
found as second best choice for rainfall estimation.  
iv) However, for Kasimkota and Parvada sites, it is  
noted that the most of the observed data are lying  
below the fitted lines of the estimated extreme  
rainfall by GPA (LMO); and hence GPA (LMO) is  
not adjudged as better suited for EVA. In light of  
the above, it is found that GEV (LMO) is the best  
choice for EVA for Kasimkota and Parvada.  
v) By considering the uncertainty involved in rainfall  
estimation for higher order return periods, the study  
suggested that:  
i) RMSE of GPA (MoM) for Atchutapuram,  
Kasimkota and Parvada while GEV (LMO) for  
Anakapalli was found as minimum.  
ii) For Atchutapuram site, it is noted that the RMSE of  
GEV (LMO) is the second minimum next to RMSE  
of GPA (MoM).  
iii) For Kasimkota and Parvada, it is noted that RMSE  
of GPA (LMO) and GEV (LMO) are the second  
and third minimum next to RMSE of GPA (MoM).  
Selection of Probability Distribution  
Based on EVA results obtained from quantitative  
assessment by using GoF and diagnostic tests, it was  
observed that the analysis offered diverging inferences and  
thus called for qualitative assessment using plots of the  
estimated extreme rainfall (Figure 1). Hence, the best fit  
for rainfall estimation was re-assessed through fitted  
curves of the estimated extreme rainfall together with  
RMSE values; and accordingly final selection was made.  
i) Diagnostic test results indicated that GPA (MoM)  
for Atchutapuram, Kasimkota and Parvada while  
GEV (LMO) for Anakapalli could be used for  
EVA.  
a) For the case of economical design of civil and  
hydraulic structures, extreme rainfall obtained  
from EV1 (LMO) distribution may be  
considered even though the RMSE of EV1  
(LMO) was higher than the corresponding  
values of other PDs for Anakapalli,  
Atchutapuram, Kasimkota and Parvada sites.  
b) For the case of little risk involved in the  
operation and management of civil and  
hydraulic structures, extreme rainfall obtained  
from GEV (LMO) distribution may be used for  
design purposes.  
28  
J. Civil Eng. Urban.,10 (3): 24-31, 2020  
Anakapalli  
GEV (LMO)  
Kasimkota  
GEV (LMO)  
800  
600  
400  
200  
600  
Observed  
GEV (MoM)  
EV1 (LMO)  
GPA (MoM)  
Observed  
GEV (MoM)  
EV1 (LMO)  
GPA (MoM)  
EV1 (MoM)  
EV2 (LMO)  
EV2 (MoM)  
GPA (LMO)  
EV1 (MoM)  
EV2 (MoM)  
GPA (LMO)  
500  
400  
300  
200  
100  
0
EV2 (LMO)  
0
1000  
800  
600  
400  
200  
0
1
10  
100  
1000  
1
10  
100  
1000  
Return period(year)  
Atchutapuram  
Return period(year)  
Parvada  
800  
600  
400  
200  
0
Observed  
GEV (MoM)  
EV1 (LMO)  
GPA (MoM)  
GEV (LMO)  
EV2 (MoM)  
GPA (LMO)  
Observed  
GEV (MoM)  
EV1 (LMO)  
GPA (MoM)  
GEV (LMO)  
EV2 (MoM)  
GPA (LMO)  
EV1 (MoM)  
EV2 (LMO)  
EV1 (MoM)  
EV2 (LMO)  
1
10  
100  
1000  
1
10  
100  
1000  
Return period(year)  
Return period(year)  
Figure 1. Plots of estimated extreme rainfall by GEV, EV1, EV2 and GPA distributions with observed data  
Kasimkota  
Anakapalli  
400  
300  
200  
100  
0
500  
400  
300  
200  
100  
0
UCLat 95% level  
EV1 (LMO)  
Observed  
LCLat 95% level  
UCLat 95% level  
EV1 (LMO)  
Observed  
LCLat 95% level  
1
10  
100  
1000  
1
10  
100  
1000  
Return period(year)  
Return period(year)  
Parvada  
Atchutapuram  
500  
400  
300  
200  
100  
0
600  
500  
400  
300  
200  
100  
0
UCLat 95% level  
EV1 (LMO)  
Observed  
LCLat 95% level  
UCLat 95% level  
EV1 (LMO)  
Observed  
LCLat 95% level  
1
10  
100  
1000  
1
10  
100  
1000  
Return period(year)  
Return period(year)  
Figure 2. Plots of estimated extreme rainfall by EV1 (LMO) distribution with confidence limits and observed data  
29  
Vivekanandan, 2020  
Anakapalli  
Kasimkota  
800  
600  
400  
200  
0
800  
600  
400  
200  
0
UCL at 95% level  
GEV (LMO)  
Observed  
LCL at 95% level  
UCL at 95% level  
GEV (LMO)  
Observed  
LCL at 95% level  
1
10  
100  
1000  
1
10  
100  
1000  
Return period(year)  
Return period(year)  
Atchutapuram  
Parvada  
1500  
1200  
900  
600  
300  
0
400  
300  
200  
100  
0
UCL at 95% level  
GEV (LMO)  
Observed  
LCL at 95% level  
UCL at 95% level  
GEV (LMO)  
Observed  
LCL at 95% level  
1
10  
100  
1000  
1
10  
100  
1000  
Return period(year)  
Return period(year)  
Figure 3. Plots of estimated extreme rainfall by GEV (LMO) distribution with confidence limits and observed data  
Figures 2 and 3 present the plots of estimated extreme  
rainfall by EV1 (LMO) and GEV (LMO) distributions with  
95% confidence limits and observed data for Anakapalli,  
Atchutapuram, Kasimkota and Parvada sites. From  
Figures 2 and 3, it is noted that about 80% of the observed  
AMR data iscovered by the confidence limits of the  
estimated rainfall by EV1 (LMO) and GEV (LMO)  
distributions for Anakapalli, Atchutapuramand Kasimkota.  
Likewise, for Parvada, it can be seen that the observed  
data covered by the confidence limits of the estimated  
rainfall using EV1 (LMO) and GEV (LMO) are 100% and  
85%, respectively.  
a) The estimated extreme rainfall by EV2 (LMO) was  
consistently higher than the corresponding values  
of GEV, EV1 and GPA distributions for the return  
periods from 50-year and above.  
b) 2 test results confirmed the applicability of GEV,  
EV1, EV2 and GPA distributions for EVA of  
rainfall for Anakapalli, Atchutapuram, Kasimkota  
and Parvada.  
c) KS test results indicated that GEV, EV1, and EV2  
distributions are acceptable for EVA of rainfall for  
Anakapalli, Atchutapuram and Kasimkota sites.  
d) From KS test results, it was found that GEV, EV1,  
EV2 and GPA are acceptable for EVA of rainfall  
for Parvada.  
CONCLUSIONS  
e) Qualitative assessment of the outcomes was  
weighed together with RMSE values and fitted  
curves of the estimated extreme rainfall.  
Accordingly, GEV (LMO) is considered as the best  
choice for rainfall estimation for all four sites  
considered in the study.  
f) For the case of economical design of civil and  
hydraulic structures, the extreme rainfall obtained  
from EV1 (LMO) distribution could be used for  
design purposes.  
The paper describes the study carried out on comparison  
of MoM and LMO estimators of probability distributions  
adopted in EVA for selection of best fit for estimation of  
extreme rainfall for Anakapalli, Atchutapuram, Kasimkota  
and Parvada sites through qualitative (viz., GoF and  
diagnostic tests) and qualitative (viz., plots of the  
estimated rainfall) assessments. On the basis of evaluation  
of EVA results, the following conclusions were drawn  
from the study:  
30  
J. Civil Eng. Urban.,10 (3): 24-31, 2020  
annual time scales.Hydrology & Earth System Sciences. Dec issue,  
16(12); 4651-4660. Google Scholar  
g) For the case of little risk involved in the operation  
and management of civil and hydraulic structures,  
extreme rainfall obtained from GEV (LMO)  
distribution could be used for design purposes.  
Bhuyan A, Borah M, Kumar R. (2010). Regional flood frequency  
analysis of north-bank of the river Brahmaputra by using LH-  
moments.Water resources management. Jul issue, 24(9): 1779-1790.  
However, by considering the data length of rainfall  
(i.e., 47 years for Anakapalli, 29 years for Atchutapuram,  
26 years for Kasimkota and 23 years for Parvada) used in  
EVA, the study suggested that the extreme rainfall for  
return period beyond 100-year may be cautiously used due  
to uncertainty in the higher order return periods.  
Esteves LS. (2013). Consequences to flood management of using  
different probability distributions to estimate extreme rainfall.Journal  
of Environmental Management. Jan issue, 115: 98-105.  
Haberlandt U, Radtke I. (2014). Hydrological model calibration for  
derived flood frequency analysis using stochastic rainfall and  
probability distributions of peak flows. Hydrology and Earth System  
Sciences, 18(1): 353-365. https://doi.org/10.15488/602.  
Hosking JR. (1990). Lmoments: Analysis and estimation of distributions  
using linear combinations of order statistics. Journal of the Royal  
Statistical Society: Series B (Statistical Methodology). Sep issue,  
DECLARATIONS  
Acknowledgements  
The author is grateful to the Director, CWPRS, for  
providing the research facilities to carry out the study. The  
author is thankful to BARC, Visakhapatnam and India  
Meteorological Department for the supply of rainfall data  
used in the study.  
Malekinezhad H, Nachtnebel HP, Klik A. (2011). Regionalization  
approach for extreme flood analysis using L-moments. Agricultural  
Science and Technology (Iran), 13 (Suppl. Issue): 11831196.  
Mujere N. (2011). Flood frequency analysis using the Gumbel  
distribution. International Journal on Computer Science and  
Engineering. Jul issue, 3(7): 2774-2778. Google Scholar  
Olumide BA, Saidu M, Oluwasesan A. (2013). Evaluation of best fit  
probability distribution models for the prediction of rainfall - runoff  
volume (Case Study Tagwai Dam, Minna-Nigeria). International  
Journal of Engineering and Technology. 3(2):94-98. Google Scholar  
Competing interests  
The author declares that he has no competing  
interests.  
Rao AR, Hamed KH. (2000). Flood Frequency Analysis. CRC Press.  
Boca Raton, Florida, USA. Google Scholar  
Rasel MM, Hossain SM. (2015). Development of rainfall intensity  
duration frequency (R-IDF) equations and curves for seven divisions  
in Bangladesh. International Journal of Scientific and Engineering  
Research. 6(5): 96-101. Google Scholar  
REFERENCES  
Acar R, Celik S, Senocak S. (2008). Rainfall intensity-duration-  
frequency (IDF) model using an artificial neural network approach.  
Journal of Scientific and Industrial Research, 67(3): 198202.  
Sasireka K, Suribabu CR and Neelakantan TR. 2019. Extreme Rainfall  
Return Periods using Gumbel and Gamma Distribution. International  
Journal of Recent Technology and Engineering. 8(4): 27-29.  
Afungang R, Bateira C. (2016). Statistical modelling of extreme rainfall,  
return periods and associated hazards in the Bamenda Mountain, NW  
US Water Resources Council. (1975). Hydrology Committee.Guidelines  
for determining flood flow frequency. US Water Resources Council,  
Hydrology Committee. Google Scholar  
AlHassoun SA. (2011). Developing an empirical formulae to estimate  
rainfall intensity in Riyadh region. Journal of King Saud University-  
Zhang J. (2002). Powerful goodnessoffit tests based on the likelihood  
ratio. Journal of the Royal Statistical Society: Series B (Statistical  
Engineering  
Sciences.  
Jun  
Issue,  
23(2):  
81-88.  
Baratti E, Montanari A, Castellarin A, Salinas JL, Viglione A, Bezzi A.  
(2012). Estimating the flood frequency distribution at seasonal and  
31