A hybrid artificial intelligence and semi-distributed model for runoff prediction

Abstract

Abstract Hydrological simulations perform a vital role in river discharge forecasts, which is very essential in water resources engineering. The present study has been carried out using a semi-distributed model developed using HEC-HMS, an artificial neural network (ANN), and a hybrid model (HEC-HMS-ANN) for simulation of daily discharge in the Kallada River basin, Kerala, India. The HEC-HMS model did not perform well with the available dataset. So for simulating daily runoff, a hybrid model is developed by coupling HEC–HMS output with ANN. The model prediction accuracy is assessed using statistical metrics. Precipitation, lagged precipitation, and lagged discharge were used as input variables for the ANN model. The optimal number of lags was determined using partial autocorrelation. The hybrid model integrating the output from HEC-HMS into ANN shows better performance than the other models in simulating daily discharge and estimating the accuracy of yearly peak discharge. The accuracy evaluation of yearly peak discharge values demonstrates that simulation error is reduced by 66% and 26.5% in the hybrid model compared to the HEC-HMS and ANN models, respectively.