Spatial Downscaling of Downward Surface Shortwave Radiation Based on Image Super-Resolution
Fei Cheng, Shuhua Zhang, Qianqian Tian, Weili Duan · IEEE Transactions on Geoscience and Remote Sensing · 2024
Downward surface shortwave radiation (DSSR) is an important component in maintaining the Earth’s energy balance and plays a key role in the stability of the Earth’s climate and ecosystems. DSSR products generated based on remote sensing or reanalysis systems have been widely used in surface-atmosphere processes modeling or analyzing. However, existing DSSR products still have problems in terms of accuracy and spatial resolution. In this study, we proposed a DSSR dataset downscaling method that is based on deep convolutional neural network (DCNN) by learning the mapping relationship between low-resolution DSSR data and high-resolution DSSR data. The method was performed on ERA5-Land DSSR by combining the near-surface temperature data, dew-point temperature data of ERA5-Land, topographic data, and high spatial-resolution H08 DSSR, to directly reconstruct the 0.1° spatial-resolution ERA5-Land DSSR data to 0.05°. We produced DSSR data for long time series at 0.05° spatial resolution. The spatially downscaled DSSR dataset was validated based on ground measurements and other high-resolution DSSR datasets by comparing the spatial distribution. The validation based on 19 ground-based stations in Xinjiang from 1990 to 2017 shows that the mean bias error (MBE), root-mean-square error (RMSE), and relative RMSE (rRMSE) of the downscaled data are improved by 18.66 W/m2, 6.72 W/m2, and 3.76% compared with the original ERA5-Land DSSR product, respectively. Especially, in the validation of the mountainous sites in the Tizinafu River basin, the average RMSE decreased significantly from 88.91 to 61.51 W/m2.