Utilization of the Bayesian Method to Improve Hydrological Drought Prediction Accuracy

This study established a hydrological drought forecasting system based on the Bayesian method and evaluated its utilization for South Korea. The regression result between Historical Runoff (HR) and Ensemble Streamflow Prediction Runoff (ESP_R) was used as prior information in the Bayesian method. Additionally Global seasonal forecast System 5 Runoff (GS5_R) produced using a dynamic prediction method was used in a likelihood function. Bayesian Runoff (BAY_R), as posterior information, was generated and compared with the ESP_R and GS5_R results for predictive ability evaluation. The Standardized Runoff Index (SRI) was selected for the drought prediction, and the BAY_SRI, GS5_SRI and ESP_SRI were computed using BAY_R, GS5_R and ESP_R, respectively. The Correlation Coefficient (CC), Nash-Sutcliffe Efficiency (NSE) and Receiver Operating Characteristic (ROC) score of BAY_SRI were the highest, and the Root Mean Square Error (RMSE) of BAY_SRI was the lowest among the methods. The Bayesian method improved the behavioral and quantitative error of drought prediction and the predictive ability of the occurrence of drought. In particular, the simulation accuracy was significantly improved during the flood season. Additionally, BAY_SRI represented past drought scenarios better than did the other two methods. Overall, we found that the Bayesian method could be applied for hydrological drought predictions for based on 1- and 2-month lead times.