WRF simulation knowledge sharing based on the MC-GAT model and multi-head attention mechanism: A case study in China

Abstract

Meteorological processes are a critical component of environmental systems. The management and sharing of meteorological simulation knowledge are essential for the scientific modeling of these processes. Numerous meteorological simulation studies have provided detailed descriptions of regional configurations and parameterization schemes. However, effectively managing and sharing WRF simulation knowledge remains a significant challenge. This study proposes a knowledge prediction framework based on a Multi-Condition Graph Attention Network (MC-GAT), which employs a multi-head attention mechanism to aggregate multi-condition feature information for parameter prediction. The framework focuses on provincial regions in China, using 10 provinces as examples, each serving as the innermost nested region for WRF parameter prediction. The results indicate that as the number of input conditions increases, the model's prediction accuracy improves significantly. Furthermore, a preliminary WRF knowledge prediction platform for Chinese regions has been developed to enable users to quickly obtain WRF parameters, thereby enhancing the efficiency of parameter selection.