Application of Hybridized ANN-GARCH, ANN-SETAR, MARS-SPSO and CANFIS-SPSO Meta-Models for Improving Accuracy of Monthly Streamflow Prediction

Abstract

Abstract Among the components of the hydrological cycle, stream flow has a major role in integrated water resources management. Establishing an accurate and reliable forecasting method for prediction of stream flow is very useful. Nowadays, data-driven methods are variously applied for river flow prediction. By hybridizing, one can take advantage of the cons of different methods for the proposed purpose. In the present research, we have combined SETAR and GARCH methods with ANN and also coupled MARS and CANFIS with SPSO to predict the monthly flow of the Maroon River in south west of Iran. Thus, four hybridized data-driven models of ANN-GARCH, ANN-SETAR, MARS-SPSO and CANFIS-SPSO are developed and compared to see which method has the best performance. Although all the models yielded good results but it was seen that the ANN-SETAR model found more accurate answers in prediction of the stream flow with an average 5% higher accuracy in the results. The IQR of ANN-SETAR model is similar to observed value that this showed the efficiency of the ANN-SETAR for dependable simulation of extreme values of river flow compared to other models. So, it was concluded that the ANN-SETAR model is better than the other methods for forecasting the monthly streamflow.