Comparative Analysis And Performance Evaluation Of Two Methods For Integrated Bias Correction Of Precipitation Ensemble Forecasts

Numerical precipitation ensemble forecasts describe the future precipitation situation in an ensemble form,which can effectively improve the forecast accuracy and solve the forecast uncertainty problem of single-value forecasts.However,because of the uncertainty of precipitation elements and numerical models,there are still large errors in precipitation ensemble forecasts,so bias correction is needed for precipitation ensemble forecasts.In this study,to investigate the effects and differences of the two bias correction methods in the domestic basin,a precipitation ensemble forecast correction experiment will be conducted in the Yalong River Basin.First,two modelling methods of the ensemble model output statistics are used to analyse the effects of different methods on the correction results and determine the best modelling method.Then,a Bayesian joint probability model is developed for the precipitation forecasts at the basin stations,and the daily precipitation forecasts are calibrated and compared with the calibration results of the ensemble model output statistics method.Finally,the basin is divided into 5 sub-basins according to the precipitation distribution characteristics of the Yalong River Basin,and the average surface precipitation data are calculated for each of the 5 sub-basins,and the ensemble forecast correction models are established for the day-by-day precipitation forecasts of 2017,2018 and 2019,respectively,to verify the application effectiveness of the 2 methods in the surface precipitation forecast correction.The main results of this study are as follows.(1)In the process of precipitation ensemble forecast calibration tests,it was found that the original forecast tended to overestimate the actual precipitation,and its forecast accuracy,forecast reliability,and forecast skill could not meet the needs of disaster prevention and mitigation as well as actual hydrological forecasting.(2)When using an ensemble model output statistical model based on the left-censored generalized extreme value distribution for the correction of the ensemble precipitation forecast,it is recommended to construct the model with the ensemble forecast mean as the distribution parameter predictor.This approach can effectively avoid the risk of overfitting the model and obtain a more stable ensemble forecast correction than using ensemble members as the distribution parameter predictors.(3)In the calibration process of station precipitation ensemble forecasts,the Bayesian joint probability model has advantages over the ensemble model output statistical method in terms of forecast accuracy,reliability,and skill,and can obtain more comprehensive and high-quality ensemble forecasts;however,in relatively dry regions,the ensemble model output statistical method forecasts outperform the Bayesian joint probability model forecasts in terms of ensemble forecast interval hit observations and deterministic forecast accuracy.(4)In precipitation ensemble forecast correction,if the ensemble forecast system has a long historical data archive and the precipitation observation data are complete,the Bayesian joint probability model is recommended for ensemble forecast statistical post-processing work;when the basin historical precipitation forecast data and observation data are insufficient then the ensemble model output statistical model is more recommended,especially in the case of more precipitation 0 values in the basin,this method has a better correction effect.