再分析降水资料的适用性评估与偏差校正——以长江中下游地区为例

Applicability evaluation and deviation correction of reanalysis precipitation data:case of middle and lower reaches of Changjiang River

为弥补长江中下游地区台站观测资料的部分缺失,探究再分析数据集的适用性,从时间和空间上评估ERA5和CFSR再分析降水数据集精度,建立了包含相关系数、相对偏差、均方根误差和Kling Gupta系数多指标定量综合评价体系,提出了结合校正系数法与自回归模型的组合校正法(C-AR);并将C-AR校正法与单一的校正系数法、AR模型以及校正系数法与最邻近抽样法组合的校正法(C-KNN)进行对比,探究C-AR组合校正法的校正效果。研究结果表明:①ERA5和CFSR与实测降水数据在年、季尺度上降水趋势变化和年内分配规律上较为一致,且与实测数据的相关性均较好;②ERA5和CFSR在实测降水量多的区域均存在正偏差,而在实测降水量少的区域存在负偏差;③C-AR组合校正模型不仅可在量级上校正数据集,还可提升数据集与实测数据的相关性,可多方面综合改善数据集精度,提高数据集适用性,校正效果要优于校正系数法、AR模型和C-KNN组合校正法。

In order to make up for the partial lack of observation data from stations in the middle and lower reaches of the Changjiang River,we explored the applicability of the reanalysis dataset and evaluated the accuracy of the ERA5 and CFSR reanalysis precipitation datasets in time and space.Then we established a multi-index quantitative comprehensive evaluation system including correlation coefficient,relative deviation,root mean square error and Kling Gupta coefficient,and proposed a correction method(C-AR)that combines correction coefficient method and autoregressive model.Finally,we compared C-AR correction method with single correction coefficient method,AR model,and the combined correction method of correction coefficient method and nearest neighbor sampling method(C-KNN)to explore the correction effect of C-AR combined correction method.The results show that:①ERA5 and CFSR are consistent with the measured precipitation data in terms of the annual and seasonal scales of precipitation trend changes and the precipitation distribution rules during the year.②Both ERA5 and CFSR have positive deviations in strong precipitation areas and negative deviations in weak precipitation areas.③C-AR combined correction model can not only correct the datasets in magnitude,but also improve the correlation between the dataset and the measured data.The model can comprehensively improve the accuracy of the dataset from many aspects and improve the applicability of datasets in the middle and lower reaches of the Changjiang River,which is significantly better than the correction coefficient method,the AR model and the C-KNN combined correction method.