306 Chapter 4 Probability and Statistics in Water Resources Engineering
FIGURE 4.10: Comparison
‘between Weibull and
Gringorten probability,
distributions
10 ; °
“Foe eS
og} |__| Sina 02
o7 | —|— 03 §
06 - 04 8
05 - —los 3
oat — tt 06 8
o3{|— | | | | tor §
02 {|| 1 oea
+ TT to
I 10
0 5000 40000 ea 000 —aa.n0
Flow (5)
‘The chi-square test. Based on sampling theory, its known that if the
are divided into M classes, with X,, being the number of outcomes in
‘Pm being the theoretical probability of an outcome being in class m, then
variable,
has a chi-square distribution. The number of degrees of freedom is
expected frequencies can be computed without having to estimate
parameters from the sample statistics, while the number of degrees
M — 1 ~ nif the expected frequencies are computed by estimating
parameters from sample statistics. In applying the chi-square goodness
nll hypothesis is taken as Hy: The samples are drawn from the propo
distribution, The null hypothesis is accepted at the a significance level if
and rejected otherwise
EXAMPLE 4.17
Analysis of a 47-year record of annual rainfall indicates the followi
distribution:
Range | Namberof | Range | Number of
(mm) | outcomes | (mm) | outcomes
1000 2 | is0-1300 | 7
rooo-1050 | 3 | 1300-1350] 5
rosoti0 | 4 | 350-1400 | 3
tioo-110 | 5 | 140-1450 | 2
riso-120 | 6 | 1450-1500] 2
Lrz0-1250 | 7 esistuhts eta
The measured data also indicate a mean of 1225 mm and a standard
151 mm. Using a 5% significance level, assess the hypothesis that the:
is drawn from a normal distribution.Section 4.3 Analysis of Hydrologic Data 307
Solution ‘The first step in the analysis is to derive the theoretical frequency dist
bution, Appendix C.1 gives the cumulative probability distribution of the standard
normal deviate, z, which is defined as
1225
oe 151
z
where x isthe annual rainfall. Converting the annual rainfall amounts into standard
normal deviates, z, yields:
Rafal
(nm)
00
150
1100
iso.
1200
1280
130
1350,
vo | 116
uso | 149
1500 | 192
‘and therefore the theoretical frequencies are given by
Range Theoretical | — Theoretical
(mm) | probability.pm | outcomes, Npm
1000 O07 329
1000-1050 0.05 235
1050-1100 0.08 376
1100-1150, ou 5a7
1150-1200 042 S64
1200-1250 04 658
1250-1300 012 Sos
1300-1350 oan 57
1350-1400 0.08, 376
1400-1450 0.05 238
1450-1500 04 138
1500 0.03 La
where the total number of observations, N, is equal to 47. Based on the observed
and theoretical frequency distributions, the chi-square statistic is given by Eque-
tion 4.105 a5
C= 307 5.62235
329
+f
14l308 Chapter 4
.
Probability and Statistics in Water-Resources Engineering
Since both the mean and standard deviation were estimated from the
data, the 7 statistic has M — 1 — m degrees of freedom, where M
number of intervals), and n = 2(= number of population parameters esi
‘measured data), hence 12 — 1 ~ 2 = 9 degrees of freedom. Using a 5%
level, the hypothesis that the observations are drawn from a normal d
accepted if
05 142 = xbas
Appendix C.3 gives that for 9 degrees of freedom, x3, = 16.919. Since 0
16.919, the hypothesis that the annual rainfall is drawn from a normal ds
accepted at the 5% level.
‘The effectiveness of the chi-square test is diminished if both the num
intervals, called cells is less than 5 and the expected frequency in any cells
5 (Haldar and Mahadevan, 2000; McCuen, 2002a).
Kolmogorov-Smirnoy test. AS an alternative to the chi-square goodies
(to assess the hypothesis that observations are drawn from a population
theoretical probability distribution), the Kolmogorov-Smirnov test «
‘This test differs from the chi-square test in that no parameters from the
probability distribution need to be estimated from the observed data. In
the Kolmogoroy-Smirnov test is called a nonparametric test. The pi
implementing the Kolmogorov-Smirnov testis as follows (Haan, 1977)
1. Let Px(x) be the specified theoretical cumulative distribution fu
the null hypothesis,
2. Let Sy(x) be the sample cumulative distribution function based on:
tions. For any observed x, Sv(x) = k/N, where k is the number of
less than or equal to x.
3. Determine the maximum deviation, D, defined by
D = max|Px(x) ~ Sn(@)l
If, for the chosen significance level, the observed value of D is
cor equal to the critical value of the Kolmogorov—Smirnov statistic
Appendix C.4, the hypothesis is rejected.
An advantage of the Kolmogorov-Smirnov test over the chi-square
it is not necessary to divide the data into intervals; thus any error
associated with the number or size of the intervals is avoided (Haldar and
2000),
EXAMPLE 4.18,
Use the Kolmogorov-Smirnov test at the 10% significance level to assess
sis that the data in Example 4.17 are drawn from a normal distribution