An improved meta-Gaussian distribution model for post-processing of precipitation forecasts by censored maximum likelihood estimation

Abstract

Statistical post-processing methods have been applied in hydrometeorological forecasting to correct the bias and spread error in raw forecasts. Among various post-processing methods, the meta-Gaussian distribution model (MGD) is one of the early successful methods for post-processing of precipitation forecasts and has been applied in the National Weather Service’s Hydrologic Ensemble Forecast System (HEFS), together with the mix-type meta-Gaussian distribution model (MMGD). However, recent studies have shown that the original MGD cannot yield reliable forecasts especially for sub-daily precipitation forecasts (e.g., 6-hourly). In this paper, we improved the MGD model by applying the censored maximum likelihood estimation (CMLE) method. We conducted experiments using GEFS reforecasts in Huai river basin in China to evaluate its performance. The results show that the proposed method performs better than the original MGD for sub-daily precipitation forecasts. The proposed method also achieves similar forecast skill with the state-of-the-art censored, shifted Gamma distribution-based ensemble MOS (CSGD-EMOS) if both use ensemble mean as the only predictor. The results indicate that the proposed CMLE-MGD can be useful for further applications such as flood forecasting that needs forecasts of high temporal resolution.