Comprehensive assessment of global atmospheric downward longwave radiation in the state-of-the-art reanalysis using satellite and flux tower observations

The atmospheric downward longwave radiation at the Earth's surface (L-d) is an important parameter for investigating greenhouse effects and global climate changes. Reanalysis data have been widely applied to obtain surface radiation components. Since new generation reanalysis data have been released, a comprehensive evaluation of the L-d predictions from the latest reanalysis data using ground measurements is still necessary. In this study, the L-d estimates of four representative reanalysis data (CFSR, JRA-55, ERA5, and MERRA2) were evaluated using ground observations at 383 stations from the AmeriFlux, AsiaFlux, BSRN, Buoy, FLUXNET, and SURFEAD networks. The evaluation results manifested that the overall root mean square errors (mean bias errors) of daily mean L-d values over the global surface were 21.1 (- 1.8) W m(-2), 22.4 (- 3.9) W m(-2), 19.3 (- 3.6) W m(-2), 25.2 (- 12.6) W m(-2), and 20.5 (3.1) W m(-2) for CFSR, JRA-55, ERA5, MERAA2, and CERES-SYN, respectively. Compared with the CERES-SYN satellite retrievals, the ERA5 (CFSR) daily mean L-d estimates had relatively smaller overall root mean square errors (mean bias errors) over the global land surface. Over the global ocean surface, the JRA-55 daily mean L-d estimates had comparable mean bias errors (MBEs) with CERES-SYN. After removing the MBEs, the best annual mean L-d estimate was 344.0 (+/- 3) W m(-2) over the global surface of 2001 to 2020. The spatial distributions and long-term trends of L-d for the selected four reanalysis data and CERES-SYN were also investigated in this study. The comprehensive assessment of the L-d products from reanalysis data and satellite retrievals in this study would be helpful for climate change studies.

Automated summary

In this study, the estimates of atmospheric downward longwave radiation (L-d) from four reanalysis data sets were evaluated using ground observations. The results showed that ERA5 and CFSR had relatively smaller errors compared to other reanalysis data, indicating their better performance in estimating L-d. This comprehensive assessment of L-d products from reanalysis data is important for climate change studies.