数值集合降雨预测的校正后处理方法研究

Post-processing method of numerical ensemble rainfall prediction

利用2017年4月1日-9月30日全球集合预测系统的降雨预测数据和雅砻江流域气象站点的降雨观测数据,采用基于左删失广义极值分布的集合模式输出统计方法对流域内降雨预测进行校正,对比分析该方法两种建模形式在校正结果上的差异。结果表明:采用集合成员均值校正的方式可以有效改善原始预测对于降雨过分高估的问题,其预测结果明显优于采用集合成员校正方式的预测结果,后者由于模型参数增加而出现过度拟合问题,限制了其在雅砻江流域中的应用。另外采用集合成员均值校正方式的预测结果的准确性在不同流域范围存在明显差异并倾向低估流域内较大降雨量,因此在后续的研究中需要进一步针对该方法无法对极值降雨量进行准确预测的问题进行改进。

Using the rainfall prediction data of the Global Ensemble Forecast System from April 1 to September 30,2017 and the rainfall observation data from the meteorological stations in the Yalong River Basin,the rainfall prediction data was calibrated by the ensemble model output statistics based on the generalized extreme value distribution,and the calibration results obtained from two models were compared and analyzed.The results show that the ensemble member mean calibration model can effectively address the problem of rainfall overestimation,which is always the case with the original prediction model.Furthermore,its prediction result is significantly better than that of the ensemble member calibration model,which is limited by the over-fitting problem due to the increase of model parameters,thus the latter is not applicable to the rainfall prediction of the Yalong River Basin.However,the accuracy of the prediction results of the ensemble member mean calibration model varies significantly in different basins and tends to underestimate the large rainfalls in the basin.Therefore,further research on this method should be conducted targeting at improving the prediction accuracy of extreme rainfalls.