Estimation Of Downward Surface Shortwave Radiation From The New-generation Geostationary Satellite Of Himawari-8

Downward surface shortwave radiation(DSSR)is an important input parameter for land surface model and climate model,thus to get the high accuracy of DSSR is crucial in land surface radiation balance research and solar energy estimation application.Remote sensing technique with the advantages of large-scale and high-frequency measurements,is playing an more important role for the estimation of DSSR.Currently,although significant progress has been made in this field,some problems remain unclear.Firstly,the existing DSSR algorithm for the cloudy sky often assumes that the cloud phase is liquid,which will bring some errors in the estimation of DSSR.Secondly,even if cloud phase of water and ice are considered,the ice crystals are often assumed to be sphere,which will further introduce some errors in the estimation of DSSR.Thirdly,the computational efficiency of traditional look-up table method will greatly decrease with the increase of input parameters.Fourth,it is difficult to use the current DSSR products for land surface and climate model's simulation,due to its coarse spatial and temporal resolutions(spatial resolution<1°,temporal resolution<3 hours).In view of the above problems,this study focuses on the following aspects of research:(1).Aiming at the problem of the DSSR estimation for cloudy sky,water cloud and ice cloud are both considered in our DSSR estimation algorithm,and we also used the non-spherical ice crystal named Voronoi in the ice cloud model to improve the accuracy of DSSR estimation.(2).Aiming at the problem of low efficiency of traditional look-up-table method with multiple input parameters for estimating DSSR,this study developed a deep learning algorithm for estimating DSSR by training the data of look-up-table(LUT),which is generated from atmospheric radiative transfer model(RSTAR).Compared to the LUT method,the deep learning algorithm for estimating DSSR developed by this study has higher accuracy and computation efficiency.(3).Aiming at the problem of low spatial and temporal resolution of existing DSSR products,this study applied the deep learning algorithm developed by this study to the new-generation geostationary satellite of Himawari-8(with level 2 products of aerosol and cloud),and got the DSSR results with high temporal(10 minutes)and spatial(5km)resolutions for 2016.In order to validate the accuracy of calculated DSSR,a total of123 ground sites from 7 networks(AERONET,BOM,BSRN,CMA,ESRL,GTMBA,Xianghe)were used.The overall mean-bias-error(MBE),root mean square error(RMSE)and the determination of coefficient(R²)for daily mean scale are 8.16 Wm^-2,29.30 Wm^-2 and 0.89,respectively,showing a high accuracy.Compared to LUT method,the deep learning algorithm developed by this study for estimating DSSR can provide the powerful technical support for near real-time monitoring of DSSR.