Why Are Physics‐Based Models Taking so Long to Run?

Physics‐based models provide a reliable and interpretable framework based on established physical laws, allowing them to deal with unforeseen future conditions much better than statistical data‐driven methods. Physics‐based models in Earth science (e.g., climate models, hydrological models, and groundwater models) are commonly used for understanding long‐term trends, predicting near‐term variations, and projecting future patterns and changes under different scenarios. However, one major challenge for physics‐based models is that they often require considerable computational resources and long runtime. Here, through a revisit to the CFL (Courant‐Friedrichs‐Lewy) condition with illustrative examples, I demonstrate that the long runtime needed for physics‐based models is internally constrained by the CFL condition. Ignoring the CFL condition may speed up the model run, but it will lead to unstable model performance with unrealistic results, which further dissolves the benefits of using physics‐based models. To maximize the benefits of physics‐based models in real‐world applications, it is important to meet the CFL condition and make strategic trade‐offs among finer model resolution, limited computational resources, and shorter model runtime.