Runoff forecasting for an asphalt plane by Artificial Neural Networks and comparisons with kinematic wave and autoregressive moving average models

Abstract

Summary Event-based runoff forecasting for 1, 2, 4 and 8 time steps ahead, based on rainfall and flow data of ten storm events for an asphalt plane, have been investigated by the Artificial Neural Network (ANN) technique. The investigation includes ANN models with three different types of inputs: (i) rainfall only, (ii) discharge only and (iii) a combination of rainfall and discharge. The results show that inclusion of discharge as an input in general, improved the performance of the ANN. However, model improvements were less significant for longer forecast lead times. Significant time shift errors in the predicted hydrographs were observed for ANN models that used discharge only as input. Although ANN models with the smallest time shift errors were models that included rainfall as inputs, these models produced hydrographs that were noisier. ANN model results were also evaluated by comparisons with results from the kinematic wave (KW) and autoregressive moving average (ARMA) models. It was found that ANN model forecasts compared favorably with runoff predictions by the KW and ARMA models. Specifically, ANN models that included discharge as input were superior to the KW model for all forecast ranges. However, the inclusion of discharge as an input to the ANN models implies that discharge measurements must be available during the model simulation stage; the KW model does not have this requirement. ANN models that did not include discharge as an input were better at long-term forecasts but poorer at short-term forecasts, when compared to the KW model. The poorer performance of the KW model at longer lead times is probably due to errors in the forecast rainfall used.