Physic Based Numerical Weather Prediction

Physics-based numerical weather prediction (NWP) traditionally relies on complex physical models to forecast weather, but recent research explores the integration of data-driven approaches, particularly deep learning, to enhance accuracy and efficiency. Current research focuses on developing and comparing hybrid models that combine the strengths of physics-based and data-driven methods, employing architectures like deep UNet++, transformers (e.g., SwinV2), and neural operators (e.g., in FourCastNet), often trained on large datasets like ERA5. This interdisciplinary effort aims to improve the skill and speed of weather forecasting across various timescales, from nowcasting to seasonal predictions, ultimately leading to better preparedness for extreme weather events and improved societal resilience.