Semi-distributed neural network models for streamflow prediction in a small catchment pinang

Abstract

This paper applied an artificial intelligence methodology for streamflow prediction in a flash flood in Pinang catchment based on TOPMODEL input and output data sets. TOPMODEL is a semi-distributed rainfall runoff model widely used in numerous water resource applications. However, literature has indicated relative weakness in TOPMODEL performances in streamflow prediction. Thus, radial basis function neural network (RBF-NN) has been employed to improve the accuracy of streamflow prediction and then compared with TOPMODEL and multilayer perceptron neural network (MLP-NN) performances. Four years of daily hydro-meteorological data sets (for the period between 2007 to 2010) were used for calibration and validation analysis. The results have shown an improvement from 0.749 and -19.2 of the calibration period to 0.957 and 0.001, and from 0.774 and -19.84 of the validation period to 0.956 and -3.611 of Nash-Sutcliffe model (NS) and Relative Volume Error (RVE), respectively. RBF-NN performance has been established to improve the daily streamflow prediction; however, the MLP-NN was better in contrast with the involved method in the study. It can be concluded that TOPMODEL performance showed a high ability to simulate the peaks compared with both AI methodologies.