Module 5 : Continuous probability distributions

Lecture 26 : Extreme Value distributions

 

Maximum flow.png

Figure 5.24: Gumbel probability plot for the annual maximum flows in the Tiber River, Italy

Fitting the Gumbel distribution to the annual maximum daily runoff data of a river

Daily runoff data is available for a river for about eighteen years (Table 5.1). If only the maximum runoff data observed in a year is considered for modeling eighteen data points are available. Considering only the maximum runoff also obscures the other relatively larger daily runoffs occurred in any particular year. To avoid this, daily runoffs exceeding 500 Cumecs are considered in this example. This results in a total sample size of 120. This data are used in explaining the Gumbel plotting  methodology discussed in the next section.

Table 5.1: Sample calculations involved in Gumbel plotting

S. No.

Run off
(m3/Sec)

Plotting position(p),
m/n+0.3

Empirical reduced
variate, -ln(-n(p)

Gumbel reduced
variate, (x-b)/a

100

867.4

0.826

1.657

0.582

101

871.3

0.835

1.711

0.607

102

881.1

0.843

1.767

0.669

103

882.3

0.851

1.826

0.677

104

902.3

0.860

1.888

0.804

105

912.7

0.868

1.953

0.871

106

939.8

0.876

2.022

1.043

107

947.1

0.884

2.096

1.090

108

955.1

0.893

2.175

1.141

109

973.4

0.901

2.259

1.257

110

992.6

0.909

2.351

1.380

111

999.6

0.917

2.450

1.424

112

1006.2

0.926

2.560

1.466

113

1018.2

0.934

2.682

1.543

114

1034.2

0.942

2.820

1.645

115

1073.3

0.950

2.979

1.894

116

1128.6

0.959

3.165

2.246

117

1138

0.967

3.393

2.306

118

1307.2

0.975

3.685

3.383

119

1506.2

0.983

4.094

4.651

120

1528.6

0.992

4.792

4.794

Gumbel Probability plotting

Two important steps in the Gumbel probability plotting are finding the plotting positions and the reduced variates. Plotting positions are distributed in between zero and one and each plotting position is equally likely to appear. To obtain the plotting positions the data is arranged in the ascending order and the ranks are given from the smallest to the largest variate. In finding the plotting position corresponding to any ordered statistic the rank value of that variate and the rank of the maxima is used. For example, the plotting position corresponding to the 100th ordered statistic is 100/(120+0.3). Sample calculations are shown in the above Table. The parameters used in getting the theoretical reduced variates are: location parameter is 776, scale parameter is 157. Figure 5.24 shows the probability plot of the maximum runoff data of Tiber river, Italy.

Summary

Because the world around is not perfectly reproducible, an understanding of the observation and measurement issues, working knowledge of the theories of probability is necessary to understand the variability. The work of the analyst can be simplified when the problem at hand fits the specifications of one of these common distributions. In this Chapter various observed data have been assumed to follow different discrete and continuous distributions. But, it is necessary to know whether these assumptions are correct or not. For verifying this it is necessary to understand statistics and their distributions that are discussed in forthcoming chapters. In the chapter ‘Hypotheses testing’ the truthfulness of the assumptions is tested.