The application of a neural network-based scale-adaptive cloud-fraction scheme in the WRF model

Cloud fraction significantly affects the short- and long-wave radiation. However, its realistic representation in models has been difficult due to inadequate understanding of the sub-grid scale cloud processes. Recently, we have developed a neural network-based scale-adaptive (NSA) cloud-fraction scheme using the CloudSat data and found that the new scheme could greatly improve the simulation of cloud spatial distribution and vertical structure. In this study, we present two applications of the NSA scheme in the WRF model. The first is the simulation of the regional winter climate of the Tibet Plateau, where the NSA scheme was shown to significantly reduce the longstanding bias of too-cold surface temperature. The second is a tropical cyclone simulation, showing that the NSA scheme better simulated the track of In-Fa (2021). The underlying mechanisms will be presented.