Weather Forecasting Using Advanced Artificial Intelligence Techniques Based on Swin Transformer Network

It is still difficult for weather prediction (NWP) representations to exactly anticipate rainfall, especially the highest rainfall events. The severely intermittent character rainfall, together with subgrid-scale are critical to the multi-scale dynamics that produce rainfall, might be the reasons for this. Because of this inherent fallibility of weather prediction model projections, it is standard practice to fix these forecasts by post-processing the model's output. To improve weather prediction, the study suggests using a Spectral Swin Transformer Network (SSTN). An operating high-resolution model system, the PRUFS model, which runs for East China at 3 km grid intermissions, was used to test the model. The 311 conventional ground-based weather stations to collect data on 2 m temperature, 2 m relative speed. A year's worth of historical PRUFS perfect outputs and surface comments make up the training dataset for SSTN, while two real-time SSTN outputs form the basis of the act evaluation.