Estimation of ANN prediction bounds for the suspended sediment load modeling

Abstract

Abstract In this paper, the point prediction of the Artificial Neural Network (ANN) for the suspended sediment load modeling was evaluated for the Lighvanchai River located in Iran, in monthly and daily scales. Since point prediction of ANN convey no information about the accuracy of prediction, so prediction intervals (PIs) were constructed by the Bootstrap method as a most frequently used technique for assessing the uncertainty of ANN. In this way, the accuracy of PIs was quantified by coverage and width criteria. The results showed that the ANN-based modeling in daily scale had better performance compared to that in monthly scale and Nash Sutcliff efficiency was 32% higher in daily scale compared to monthly. Moreover, the width and coverage of the constructed PIs in daily scale were 14% and 24%, lower and higher compared to that in monthly scale and the Bootstrap method could appropriately capture the target values.