Artificial intelligence and numerical weather prediction models: A technical survey

Abstract

Can artificial intelligence (AI) models beat traditional numerical weather prediction (NWP) models based on physical principles? The rapid advancement of AI, inherent computational limitations of NWP models, and the lack of access to big data drive this question in terms of resolution and complexity. This survey offers a systematic review of studies that integrate AI with NWP models at various stages of weather and climate modeling. It aims to address key research questions, including the types of forecasting models, the integration of AI into NWP systems, and the comparative efficacy of AI-based approaches versus conventional NWP models. It covered peer-reviewed literature from 2000 to 2024. This technical survey highlights key advancements in the application of AI within NWP modeling in data assimilation, augmentation, pre-processing, adaptive parameter tuning, optimization, uncertainty quantification, extreme event prediction, post-processing, and the interpretation of NWP outputs. While AI demonstrates significant potential in post-processing NWP outputs, pre-processing remains challenging. This survey also presents state-of-the-art AI-based hybrid models and assesses their applicability to weather data. It highlights the promise of AI in potentially replacing traditional NWP models but emphasizes the need for further advancements in model development and application. The study also offers a detailed classification of forecasting models and outlines promising directions for future research.