Runoff Analysis for a Small Watershed of Tono Area Japan by Back Propagation Artificial Neural Network with Seasonal Data

Abstract

The rainfall runoff (R-R) process was studied for two small sub-basins having different sizes in a mountainous catchment of Tono area Japan. The runoff and other meterological data have been collected in this catchment for the last 14 years. The major objective of this study was to construct numerical models for these sub-basins to predict runoff after 1/2 and 1 h. The effects of season and the size of the catchment on R-R process were also investigated. The hydrogeological conditions of the catchment were studied prior to the analyses. The data obtained for summer (rainy) and winter (dry) seasons were treated separately in order to study the seasonal effects on the model development. The back propagation artificial neural network technique (BPANN) and the multivariate autoregressive and moving average models (ARMA) were adopted for the analysis. It was found that for very small catchments the seasonal effects are dominant and therefore separate models should be developed for each season to obtain better forecasting estimates. It was also found that the predictions by BPANN models were better than multivariate ARMA models for intense rains having complex R-R relationships in summer. On the other hand, both the modelling techniques yielded almost similar results for smaller rains in winter. It was also found clearly that the accuracy of prediction decreased with the increase of the time period for prediction.