Options for framing NOAA weather and water modeling as a foundational component of the broader US weather and water enterprise

Numerical weather prediction (NWP) is at a crossroads. Until recently, most of the advances at operational modeling centers have been via improvements to physics-based models: increasing resolution, implementing advances in parameterizations, improving coupling among model components, assimilating new types of data, adding ensembles to variational data assimilation systems, and reconfiguring large code bases to run more efficiently on ever-expanding High Performance Computing (HPC) systems. However, there has been a significant shift within the past year. Major weather forecasting centers around the world have rapidly implemented AI techniques, and are implementing data-driven surrogate models alongside their physics-based models. While these AI-based models have shown skill superior to the best NWP models in some respects, their most salient characteristic is a dramatic increase in speed, enabling the generation of very large ensembles and a corresponding change in how observations are used in data assimilation systems. While industry in the US has been at the forefront of the transition to AI-based NWP, federal agencies have lagged in their adoption of AI-based modeling. Meanwhile, recent cuts to the US federal budget have caused a depletion of modeling expertise at NOAA and throughout the US research and operations community.