We also introduce and evaluate the Pacific Northwest Windstorm (PNW)-Typhoon weather satellite dataset to verify the effectiveness of DGDM in high-resolution regional forecasting.
Climate change has led to an increased frequency of natural disasters such as floods and cyclones.
Future frame prediction has been approached through two primary methods: autoregressive and non-autoregressive.
Ranked #1 on Weather Forecasting on SEVIR
In this paper, we propose a training strategy to make the weather prediction model robust to spatial-temporal shifts.
Traditional weather forecasting relies on domain expertise and computationally intensive numerical simulation systems.