Journal of Water and Land Development

Abstract

The aim of the study was to compare two grouping methods for regionalisation of watersheds, which are similar in respect of low flow and chosen catchments parameters (physiographic and meteorological). In the study, a residual pattern approach and cluster analysis, i.e. Ward's method, were used. The analysis was conducted for specific low flow discharge q 95 (dm 3 •s -1 •km -2 ). In the analysis, 50 catchments, located in the area of the upper and central Vistula River basin, were taken. Daily flows used in the study were monitored from 1976 to 2016. Based on the residual pattern approach (RPA) method, the analysed catchments were classified into two groups, while using the cluster analysis method (Ward's method) -into five. The predictive performance of the complete regional regression model checked by cross-validation R 2 cv was 47% and RMSE cv = 0.69 dm 3 •s -1 •km -2 . The cross validation procedure for the cluster analysis gives a predictive performance equal to 33% and RMSE cv = 0.81 dm 3 •s -1 •km -2 . Comparing both methods, based on the cross-validated coefficient of determination (R 2 cv ), it was found that the residual pattern approach had a better fit between predicted and observed values. The analysis also showed, that in case of both methods, an overestimation of specific low flow discharge q 95 was observed. For the cross-validation method and the RPA method, the PBIAS was -10%. A slightly higher value was obtained for the cross-validation method and models obtained using cluster analysis for which the PBIAS was -13.8%.