While artificial intelligence (AI) models like GraphCast and FourCastNet have dramatically improved the speed and accuracy of weather forecasting, their internal workings remain largely opaque, offering little scientific insight into atmospheric processes. However, new pioneering research has demonstrated an AI algorithm that could bridge this gap, translating vast datasets into understandable mathematical equations.
Traditional physics-based models are limited to short-term forecasts (typically less than two weeks) due to the “inherently chaotic” nature of weather and the exponential growth of tiny inaccuracies. These models rely on estimated parameters that represent physical processes, such as wind or temperature factors. In contrast, modern AI models achieve superior performance but contain tens of millions of uninterpretable parameters that do not directly correspond to the underlying physics.
A study by Minor et al. addresses this by introducing the capabilities of a Weak form Scientific Machine Learning {WSciML} algorithm known as Weak form Sparse Identification of Nonlinear Dynamics (WSINDy). Unlike highly-parameterized AI approaches {WSINDy} learns from data to discover and formulate mathematical equations that represent complex, real-world physical processes, such as the relationship between air pressure, density, and vorticity that governs wind speed and direction.
The researchers successfully applied WSINDy to both simulated and real-world turbulent atmospheric fluid data, which includes measurements of temperature, pressure, and wind speed. Most significantly, the algorithm was able to accurately identify the known governing equations of atmospheric physics from a global-scale set of assimilated data incorporating actual weather observations.
These findings suggest that WSINDy holds the potential to not only aid in weather forecasting being especially well-suited for noisy observational data but also to help uncover new physical insights about weather systems.
The researchers note that while the algorithm needs further refinement to accurately identify certain complex atmospheric equations, it is already being explored for application in diverse scientific fields, including fusion research, modeling epidemics, and understanding cell communication during wound healing.