Toward an Improved Global Longwave Downward Radiation Product by Fusing Satellite and Reanalysis Data

Abstract

Surface longwave downward radiation (LWDR) plays an important role in modulating greenhouse effect and climate change. Constructing a global longtime series LWDR dataset is greatly necessary to systematically and in-depth study the LWDR effect on the climate. However, the current multi-source LWDR products (satellite and reanalysis) show large differences in terms of both spatio-temporal resolutions and accuracy in various regions. Therefore, it is necessary to fuse multi-source datasets to generate more accurate LWDR with high spatio-temporal resolution on a global scale. To this end, a downscaling strategy is firstly proposed to generate LWDR dataset with 0.25° resolution from CERES-SYN data with 1° scale, by incorporating the Land Surface Temperature (LST), Total Column Water Vapor (TCWV) and Elevation. Then a machine learning-based fusion method is provided to generate a global hourly LWDR dataset with spatial resolution of 0.25° by combing three products (CERES-SYN, ERA5 and GLDAS). Compared with ground measurements, the performance of generated LWDR product reveals that the correlation coefficient (R), mean bias error (BIAS), and root mean square error (RMSE) were 0.97, -0.95 W/m2 and 22.38 W/m2 respectively over the land, and 0.99, -0.88 W/m2 and 10.96 W/m2 over the ocean. Specially, it shows improved accuracy in the low and middle latitude regions compared with other LWDR products. Considering its better accuracy and higher spatio-temporal resolution, the new LWDR product can provide essential data for deeply understanding the global energy balance and even the global warming. Moreover, the proposed fusion strategy can be enlightening for readers in the fields of multi-source data combination and big data analysis.