A combined hydrodynamic model and deep learning method to predict water level in ungauged rivers

Abstract

Forecasting the water level (WL) of rivers is vital for water resource management. Current research of WL prediction mainly focuses on gauged sites and further investigation into effective approaches to predict WL in ungauged rivers, is urgently required. We developed a method for forecasting WL in ungauged sites of rivers by integrating hydrodynamic model and deep learning model. Specifically, two integration modes were included. One mode comprised deep learning model to forecast boundary conditions and hydrodynamic model to forecast WL in ungauged sites (DH mode). The other mode comprised hydrodynamic model to simulate WL in ungauged sites and deep learning model to forecast WL (HD mode). These approaches were tested in a section of the Yangtze River. The results showed that the hydrodynamic model can achieve high simulation accuracy of WL for the entire river, providing the foundation for WL prediction in ungauged sites by the two modes. The HD mode performed better than the DH mode with a satisfactory validation accuracy of the hydrodynamic model. Both modes, however, provided reliable WL prediction results, the forecasting accuracy of the two modes showed further improvement in the dry season, which is important in terms of ship navigation safety. Our proposed method, therefore, has great potential as a scientific reference for forecasting of WL in ungauged sites of rivers and can also be applied for forecasting of other inland water hydrological variables.