Enhancing a Multi-Step Discharge Prediction with Deep Learning and a Response Time Parameter

Abstract

Flood forecasting is among the most important precaution measures to prevent devastating disasters affecting human life, properties, and the overall environment. It is closely involved with precipitation and streamflow data forecasting tasks. In this work, we introduced a multi-step discharge prediction framework based on deep learning models. A simple feature representation technique using a correlation of backward lags was enhanced with a time of concentration (TC) concept. Recurrent neural networks and their variants, coupled with the TC-related features, provided superior performance with over 0.9 Nash–Sutcliffe model efficiency coefficient and substantially high correlation values for multiple forecasted points. These results were consistent among both the Upper Nan and the Loei river basins in Thailand, which were used as case studies in this work.