Improving categorical and continuous accuracy of precipitation forecasts by integrating Empirical Quantile Mapping and Bernoulli-Gamma-Gaussian distribution
Lingjie Li,Zhaode Yun,Yong Liu,Yintang Wang,Wenpeng Zhao,Yan Kang,Rui Gao
DOI: https://doi.org/10.1016/j.atmosres.2023.107133
IF: 5.965
2023-12-01
Atmospheric Research
Abstract:Statistical post-processing is a pivotal approach in enhancing the statistical accuracy and applicability of precipitation forecasts from numerical weather prediction models. The existing methods primarily focus on correcting errors in either categorical or quantitative aspects individually, failing to address both simultaneously. Furthermore, the benefits of post-processing in extending the effective lead time (ELT) have not been quantitatively examined. In this study, an integrated statistical post-processing method by combining the Empirical Quantile Mapping (EQM) and the Bernoulli-Gamma-Gaussian model (BGG) (EQM-BGG for short) is constructed, and a comprehensive accuracy metric for evaluating the ELT is also proposed. Taking the Bengbu Basin located in the upper and middle reaches of the Huai River Basin as the study area, the statistical post-processing using the three methods (EQM, BGG, EQM-BGG) is performed on the multi-model weighted integration forecast precipitation (MWIFP), which are derived from three numerical models, namely ECMWF, NCEP and CMA. The results demonstrate that the integration of EQM-BGG capitalizes on the combined strengths of EQM and BGG, facilitating a dual correction of categorical and quantitative errors. The forecast accuracy (FA) and mean absolute error (MAE) of the post-processed basin-average forecasts are enhanced by more than 10%, and the ELTs for both the entire basin and sub-basins are extended by 12–78 h. At the grid scale, EQM-BGG significantly reduces the decay rate of forecast accuracy and the expansion speed of low-accuracy regions as the lead time increases, the improvement of FAs and MAEs surpasses 10% and 20%, respectively, across all grids within the lead time of 96–102 h. Furthermore, certain grids in the northwestern part show an extension of ELTs by 18 to 54 h. These findings highlight that EQM-BGG significantly enhances both categorical and quantitative accuracy simultaneously, resulting in precipitation forecasts that exhibit greater reliability over longer lead times. This advancement enriches the methodologies for hydro-meteorological forecasts.
meteorology & atmospheric sciences