Mid and long-term hydrological classification forecasting model based on KDE-BDA and its application research

Abstract

In view of the uncertainties in the current mid and long-term hydrological forecast results, a mid and long-term hydrological classification forecasting model based on KDE-BDA is built and its application research is performed. The kernel density estimation method (KDE) is used to improve the distribution density function estimation method in the conventional Bayesian discriminant method (BDA). Based on this, a mid and long-term hydrological classification prediction model based on KDE-BDA was built. An example of its application is demonstrated in the runoff forecast of the Danjiangkou Reservoir in the autumn of September. The results show that different forecasting factors have different forecasting ability for runoff classification (flood, normal, dry). Among them, the 500hPa height field factor has higher forecasting ability for the Danjiangkou runoff category (flood, normal, dry), and the 100hPa height field factor, SST and circulation characteristics have a good indication of the runoff’s flood state. The results of Half-Brier score show that the combined forecasting model combines the forecasting advantages of each factor, so the forecasting effect is the best. The pass rate in the simulation period was 89.8%, and the pass rate in the inspection period was 87.5%. The simulation and forecast results were relatively stable.