Abstract:
This thesis briefly deals with regional low-flow frequency analysis. Low-flow
quanti lcs arc commonly used to describe stream characteristics and are used as
parameters in various studies. Unfortunately, it can be difficult to estimate extreme
low-flow quantiles, especially at sites with short record lengths. A great deal of
research has been directed to the improvement of extreme quantile estimation
procedure and regional frequency analysis has been found to be one of the best ways.
The present study is undertaken with the objective of providing a method to obtain
better estimates of low-flow of the streams in the north-west region of Bangladesh
through regionalization.
Three steps are involved in this study. Selection of low-flow data series that
satisfy such statistical criteria as randomness, stationarity, independence and
homogeneity; choosing a theoretical distribution for modelling the data sample; and
estimation of low-flow. Two different distribution selection methods were applied for
the selection of appropriate probability distributions, namely the method of L-moment
ratio diagram and the method of product moment ratio diagram. Weibull and the log
Pearson type 3 (LP3) distributions were found to fit the low-flow data series of the
selected rivers. However, the procedure based on moment ratio diagrams depends on
subjective judgement. These two distributions are further subjected to a quantitative
statistical test, namely Kolmogorov-Smimov test for further screening. This test also
suggested the Weibull and the LP3 as suitable probability distributions to fit the
selected low-flow data.
The shape parameters of both the distributions have been regionally estimated
over the north-west region yielding a regionally estimated skewness of 0.21 for
Weibull distribution and -0.88 for LP3 distribution. These regional estimates were
used to calculate low-flow quantiles for recurrence intervals of 10, 25, 50 and 100
year. These low-flow quantiles are compared to the low-flow quantiles obtained by atsite
approach. It is found that, in case of Weibull distribution, for recurrence interval
10 year, the maximum relative difference between regional and at-site quantile is
about 40 percent whereas that for LP3 distribution is 13 percent. Generally regional
estimates of low-flows showed better agreement with the corresponding values at
sites with longer records. Moreover regionally estimated statistics have smaller
sampling variations than those estimated only on the basis of at-site data. As such
when low-flow estimates are not significantly affected in magnitude by process of
regionalization, they are likely to exhibit smaller variances.
