FLOOD FLOW REGIONALIZATION BASED ON L-MOMENTS AND ITS USE WITH THE INDEX FLOOD METHOD

摘要

The method introduced in this paper is an alternative approach to flood flow regionalization. A homogeneous region is defined as a group of catchments with equal sample distribution of Q_max~AN except of the location parameter. Reliability of this concept is dependent on how well the sample distribution represents the true distribution of Q_max~AN. Over 260 selected river catchments were grouped into regions by means of hierarchical and splitting (nonhierarchical) cluster analysis methods according to L-Cv, L-Cs and L-Ck, which are in a strictly homogeneous region constant. The K-means splitting method with Euclidean distance metrics gave better results in terms of between and within cluster variances than the hierarchical method. Acquired regions are not geographically compact, but this is not regarded as an indicator of similarity of the frequency distribution. Good correlations between the average annual maximum flood and catchment area were found in all regions. For making this approach applicable also to ungauged catchments, a method for attributing such catchments to a region was suggested. This method is based on the use of contour maps of L-Cv and L-Cs. From these maps, values of L-Cv and L-Cs were inferred for 10 test catchments which were excluded from the analysis. The average relative error of L-Cv was rather small, 8.5% and the error of L-Cs was 18.9%. The index flood was estimated from catchment characteristics by means of a regression model and a supplementary map of relative errors. In this case, the average relative error was 12.3%. The quantiles Q_N were estimated by the index flood method. This method, in combination with the proposed regionalization, showed promising results with the average relative errors of Q_(10) and Q_(10) less than 13% and Q_(100) not exceeding 15%. Future work will be directed towards the issue of improving the estimation of the index flood and the coefficient of linear variation, which can be estimated from catchment characteristics. Regionalization can be improved by including the flood and rainfall seasonality measures.