Construction and Simulation of Global Land Surface Microwave Emissivity Atlas Using FY-3D Satellite Data

Microwave land surface emissivity (MLSE) atlas and instantaneous simulation of all-sky/all-surface MLSE are important prerequisites for satellite data assimilation. A ten-day/month synthesized FengYun-3D MLSE atlas (New_FY3D) was constructed by the two global MLSE daily product datasets, clear-sky (FY-3D1) and clear/cloudy (FY-3D2) retrieved from the same FY-3D Micro-Wave Radiation Imager (MWRI) Level-1 brightness temperature (BT) data from 2021 to 2022, respectively. Then, a set of global MLSE label samples based on the New_FY3D including 14 surface geophysical parameters was obtained for instantaneous global MLSE simulation with 0.10o spatial resolution by adopting the extreme gradient boosting (XGBoost) machine learning method. Finally, the FengYun-3F (FY-3F) MWRI-II BT simulations using Advanced Radiative Transfer Modeling System (ARMS) based on above different MLSE products were evaluated. The results show that the New_FY3D atlas performs well, and the BT simulation at the top of atmosphere is better than that of FY-3D1, FY-3D2 and the international mainstream TELSEM2 atlas. Surface roughness, vegetation coverage, land cover type and snow cover are vital parameters for MLSE simulation. The XGBoost model can simulate instantaneously all-sky/all-surface MLSE well at 10.65-89.0 GHz frequencies. The average simulation determination coefficients (R2) under clear-sky and cloud-sky conditions are 0.925 and 0.901, respectively, and the average root-mean -square errors (RMSE) are 0.018 and 0.021, respectively. The large simulation errors occur in permanent wetland, ice and snow, and urban and build-up areas. The BT simulation based on XGBoost simulated MLSE with the standard deviation of 6.6 K is better than that based on New_FY3D and TELSEM2.