CLDAS温度产品在湖北区域的适用性评估及订正研究

Applicability Assessment and Correction Research of CLDAS Temperature Products in the Hubei Region

摘要: CLDAS提供了高时空分辨率的陆面融合数据集，为湖北精细化气象服务提供了重要的数据支撑，而数据的适用性评估及其订正是开展数据应用的重要基础。本文以湖北省82个国家级自动气象站逐日温度为基准，采用偏差、平均误差、平均绝对误差、均方根误差和相关系数等指标，分析2018~2022年CLDAS气温产品在湖北区域的适用性，并用一元线性回归模型对CLDAS温度产品开展订正研究，且用2023年资料对模型进行回代检验。结果表明：(1) CLDAS温度产品在1月、3月、5月、10月、11月和12月的空间分布与实况差距较大，在2月的空间分布与实况匹配度较高；汛期(5月、6月、7月、8月、9月)为暖偏差，1月、2月、3月、11月、12月为冷偏差。(2) CLDAS温度产品插值和站点观测值相关性不强，但4月和10月的质量优于其他时间。相关系数以正相关为主，中度相关的站点数比例为21%~40%，高度相关的站点数比例7%~17%。(3) CLDAS温度插值在4月、10月分别降低0.5℃、3℃后，与站点实况基本相同，可作为站点稀疏地区的参考。

Abstract: CLDAS provides a high spatiotemporal resolution land surface merged dataset, offering crucial data support for refined meteorological services in Hubei. The applicability assessment and correction of this data serve as an important foundation for its application. Using the daily temperature data from 82 national-level automatic weather stations in Hubei Province as a benchmark, this study employed indicators such as BIAS, ME, MAE, RMSE, and R to analyze the applicability of the CLDAS temperature product in the Hubei region from 2018 to 2022. A univariate linear regression model was used to conduct correction research on the CLDAS temperature product, and data from 2023 were used for back-substitution testing of the model. The results indicate that: (1) The spatial distribution of the CLDAS temperature product showed significant discrepancies with observations in January, March, May, October, November, and December, while the spatial distribution in February matched observations relatively well. A warm bias was observed during the flood season (May, June, July, August, September), and a cold bias was observed in January, February, March, November, and December. (2) The correlation between the interpolated CLDAS temperature product and station observations was not strong, but the quality in April and October was better than in other periods. The correlation coefficients were predominantly positive, with the proportion of stations showing moderate correlation ranging from 21% to 40%, and the proportion showing high correlation ranging from 7% to 17%. (3) After reducing the interpolated CLDAS temperature by 0.5˚C in April and 3˚C in October, it became essentially the same as the station observations and can be used as a reference for areas with sparse stations.