Seasonal streamflow forecasting by surrogate modeling in the Yarlung Zangbo River Basin, China

Abstract

Study region: The Yarlung Zangbo River (YZR) basin on the Tibetan Plateau, China Study focus: Due to global climate change, the risk of drought disaster is increasing. Seasonal hydrological forecast can be beneficial for drought early warning and help reduce risks in water resources and drought management. However, the computational burden of distributed hydrological models remains a limitation for their wide use in seasonal streamflow forecast. This study designs a seasonal streamflow forecast framework based on the surrogate model (SM) for VIC. Both the impacts of pre-processing and post-processing on the seasonal streamflow forecast, and the accuracy, reliability and efficiency of the forecast framework based on SM, are carefully evaluated in the YZR basin, China. New hydrological insights for the region: The results show that the VIC model forced by CFSv2 has a good predictability and reliability in seasonal streamflow forecast in the YZR basin. Both pre-processing and post-processing can improve streamflow forecast accuracy, while post-processing also improves the forecast reliability more significantly. The SM-simulated streamflow is identical with that of VIC, with NSE larger than 0.95 and Pbias smaller than 5% at Nuxia station, in the YZR basin. The proposed SM-based seasonal streamflow forecast framework has been proven to be a good alternative for the VIC-based framework, with similar forecast accuracy and reliability, and higher computational efficiency, reducing up to 97% computation time.

Keywords

• Surrogate model
• The Variable Infiltration Capacity model
• Seasonal streamflow ensemble forecasting
• Deep-learning