FY-4A GIIRS数据与ERA5再分析资料融合的中国区域大气加权平均温度模型

Using FY-4A GIIRS data and ERA5 reanalysis data to build a regional atmospheric weighted mean temperature model in China

大气加权平均温度(T m)的精度直接影响全球导航卫星系统(GNSS)水汽反演的结果。针对现有T m模型的参数、建模数据源有待优化及模型构建时仅依赖于单个探空站点或单一格网点数据等问题,本文提出融合FY-4A GIIRS数据与ERA5再分析资料,在此基础上引入滑动窗口算法对融合数据进行处理同时顾及经度、纬度和高程因子构建空间分辨率为0.5°×0.5°的T m经验模型(FY-ET m模型)。采用偏差(Bias)和均方根误差(RMS)作为精度评定指标,联合未参与建模的2020年探空数据、ERA5再分析资料及天顶对流层延迟产品,对FY-ET m模型及其反演的大气可降水量进行精度评定。结果表明:以探空数据为参考值,FY-ET m模型的年均Bias、RMS分别为-0.02、5.79 K,相比较于Bevis和GPT3模型分别提高了3.62(Bias)、0.8(RMS)和2.54(Bias)、0.63 K(RMS);以ERA5再分析资料为参考值,FY-ET m模型的年均Bias、RMS分别为0.01、3.32 K,相比较于Bevis和GPT3模型分别提高了0.97(Bias)、0.13(RMS)和2.94(Bias)、1.71 K(RMS),同精度优异的GPT3模型相比,FY-ET m模型在中国西部和北部地区也表现出了明显的精度改善;以GNSS站点得到的PWV为参考值,FY-ET m模型反演的PWV与GNSS站得到的PWV值精度相当,Bias变化范围为-0.5~0.5 mm。FY-ET m模型准确度高稳定性良好,只需输入位置和时间信息就能获取目标点的T m,能够在GNSS水汽反演中发挥重要的作用。

The accuracy of atmospheric weighted mean temperature(T m)directly affects the results of Global Navigation Satellite System(GNSS)precipitable water vapor inversion.For the existing T m model parameters,modeling data sources to be optimized and the model construction only relies on a single sounding site or single grid point data.It is proposed to fuse FY-4A GIIRS data with ERA5 reanalysis data.A sliding window algorithm is introduced to process the fused data while taking into account the longitude,latitude and elevation factors to construct an empirical T m model(FY-ET m model)with a spatial resolution of 0.5°×0.5°.Bias and root mean square error(RMS)are used as accuracy metrics,combined with the 2020 sounding data,ERA5 reanalysis data,and zenith tropospheric delay(ZTD)products that are not involved in the modeling.Subsequently,the accuracy of the FY-ET m model and its inverse precipitable water vapor(PWV)are evaluated.The results show that the average annual Bias and RMS of FY-ET m model are-0.02、5.79 K,respectively,which are 3.62(Bias)、0.8(RMS)、2.54(Bias)and 0.63 K(RMS)higher than those of Bevis and GPT3 models.With the reanalysis data of ERA5 as the reference value,the average annual Bias and RMS of FY-ET m model are 0.01、3.32 K,respectively,which are 0.97(Bias)、0.13(RMS)、2.94(Bias)and 1.71 K(RMS)higher than those of Bevis and GPT3 models.Compared with GPT3 model with excellent accuracy,FY-ET m model also shows obvious accuracy improvement in western and northern China.With the PWV obtained from GNSS station as the reference value,the accuracy of PWV inversion from FY-ET m model is similar to that from GNSS station,Bias ranges from-0.5 mm to 0.5 mm.The FY-ET m model has high accuracy and good stability.It can obtain the T m of the target point only by inputting the position and time information,and it can play an important role in GNSS precipitable water vapor inversion.