Flood frequency analysis at ungauged sites in Peninsular Malaysia using least square support vector machine

Abstract

The ability of hierarchical cluster analysis (HCA) and least square support vector machine (LSSVM) in the estimation of flood quantiles at ungauged sites in Peninsular Malaysia were studied. Comparison between the multiple linear regression (MLR) models, LSSVM, HCA with MLR and HCA with LSSVM were performed. To assess the effectiveness of this model, 70 catchments in the province of Peninsular Malaysia with five inputs variables which namely catchment area, longest drainage area, slope of mainstream, altitude of the mainstream and mean annual rainfall were used as case studies. The performance of HCA with LSSVM was compared with the MLR, HCA with MLR and LSSVM models using various statistical measures such as relative bias (RBIAS), mean absolute relative error (MARE) and mean squared relative error (MSRE). The results of the comparison indicate that the proposed model which is HCA with LSSVM has the lowest RBIAS, MARE and MSRE compared to the other three models. This model predicts flood quantile more accurately and provides a promising alternative technique in estimation of flood quantiles in ungauged sites