Abstract:
Selection of a probability distribution for discharge series is very important in estimating
floods accurately. Since the introduction of L-moments (Hosking 1990), numerous
investigations have been carried out to assess the goodness of fit of various probability
distributions to regional samples. L-moment diagrams were constructed for annual
maximum flood flows, annual average daily stream flows, and annual minimum low
flows at 46 catchments in Sri Lanka. The data were screened to eliminate gross errors and
inconsistencies. For this purpose the Discordancy measure Di introduced by Hosking and
Wallis in 1993 was used. The Discordancy measure allows for an objective determination
of which sites, i, in a region, are grossly discordant with the group as a whole. For each
flow series, sites were dropped with discordancy measures Di,> 3. Suitable distributions
for each site were selected by comparing the observed and theoretical relations between
L-cv and L-skewness of the flood flows. The goodness-of-fit test was then applied in
order to examine the descriptive performance of screened distributions.