Effects of Cloud Microphysics on the Universal Performance of Neural Network Radiation Scheme

Understanding the effect of cloud microphysical changes on the stability of radiation emulator is essential in an operational weather forecasting model with frequent updates. This study examined the effect of 15 microphysics schemes on a radiation emulator for real and ideal cases. In the real case, although the forecast errors against the control run increased with varying microphysics schemes to the trained scheme, the forecast error of 2-m temperature was rather improved by 0.9%–5.4% when compared with observations. The radiation emulator for the real case was applied to an ideal squall-line simulation to test its universal application, resulting in increase of 8.6%–41.3% more forecast errors in heating rates and fluxes for 14 microphysics schemes more than the trained scheme. These errors could be reduced by 26.5%–50.4% with the use of a compound parameterization. Therefore, the stability and accuracy of radiation emulator on microphysics changes was confirmed.