不同分辨率再分析资料对浙江省气温刻画能力的对比评估

利用浙江省66个基本气象站1979—2010年的日平均气温数据，系统评估了三套再分析资料R1、R2和CFSR对浙江省气温的刻画能力。结果表明:三套再分析资料的气候平均态与观测均存在一定差异，其中R1、R2的空间分布型与观测较为接近，CFSR与观测差异较大；三套再分析资料均存在系统性冷偏差且这一偏差在32年中稳定存在，其中CFSR的冷偏差更显著，浙南地区是其冷偏差的重要来源。三套资料的均方根误差均存在季节变化:冬季（特别是1月）误差较小而夏季（特别是7-8月）误差较大，R1和R2的季节差异强于CFSR。CFSR对浙江省气温变率的把握能力优于R1和R2，其距平场EOF分解前三模态的空间型态和时间系数与观测更为接近。系统误差订正后，三套再分析资料的可信度得到显著改善，CFSR的改善效果最明显，说明系统性误差是三套再分析资料偏差的重要来源。改善后三套再分析资料的均方根误差和空间相关系数大体相当。CFSR网格点气温插值到观测站点时因海拔差异导致的误差以及CFSR在浙江省的模式地形偏高可能是其有较大冷偏差的重要原因。 

COMPARATIVE EVALUATION OF THE ABILITY OF REANALYSES WITH DIFFERENT RESOLUTION TO PORTRAY TEMPERATURE IN ZHEJIANG PROVINCE

The present study comprehensively evaluates the ability of three reanalyses, namely R1, R2 and Climate Forecast System Reanalysis (CFSR), to reproduce the temperature characteristics of Zhejiang Province using observed daily temperature data from 66 basic stations over the period 1979—2010. The result shows that the climatological normals calculated from all three reanalyses differ from the observed values, with the spatial characteristics presented by R1 and R2 being more consistent with the observation and the CFSR data deviating significantly from the observation. A systematic cold bias appears persistently throughout 1979—2010 in all the three reanalyses, with the CFSR having the largest cold bias, which mainly originates from the southern area of Zhejiang. The root-mean-square error (RMSE) of all the three reanalyses exhibits seasonal variability. The RMSE in winter (especially in January) is the smallest and the RMSE in summer (especially in July and August) is the largest for all three reanalyses. Besides, the seasonal variability of R1 and R2 is stronger than that of the CFSR. The Empirical Orthogonal Function (EOF) analysis reveals that the spatial pattern and temporal coefficient of the first three modes of the temperature anomaly of the CFSR agree with observation better than those of R1 and R2. Therefore, the CSFR can better capture the variability of temperature in Zhejiang. Correction of the systematic cold bias greatly enhances the credibility of the three reanalyses, especially for the CFSR, indicating that systematic error is a major contributor to the discrepancy between the three reanalyses and the observation. After bias correction, the RMSE and spatial correlation coefficient of the three reanalyses have comparable magnitude to each other respectively. The error resulted from elevation difference between the CFSR grid points and observational stations during the interpolation, as well as the possible high bias of the model terrain of the CFSR in Zhejiang may be important factors in producing relatively large cold bias.