NWP is the core methodology of modern forecasting, utilizing massive supercomputers to solve complex fluid dynamics and thermodynamic equations. By treating the atmosphere as a grid of three-dimensional cells, these models simulate the movement of air, moisture, and energy over time to predict future states from minutes to weeks in advance.
The industry is currently moving toward “convective-permitting” resolutions (typically <4km) globally, but the computational cost of such fine grids is astronomical. Traditionally, running a global 3km model required tens of thousands of CPUs; however, the shift toward GPU-accelerated computing is allowing agencies to maintain high resolution while reducing power consumption and processing time.
Even the best NWP models cannot explicitly simulate features smaller than their grid boxes, such as individual clouds or turbulent eddies in the boundary layer. Scientists must use “parameterizations”—simplified mathematical shortcuts—to estimate these effects. In 2026, a major search area for pros is “AI-based parameterization,” where neural networks replace traditional physics-based shortcuts to improve the accuracy of rainfall and solar radiation forecasts.
The UFS is the push toward a single, community-based modeling framework that allows researchers and operational forecasters to work on the same code. This “open-science” approach is designed to accelerate the transition of research innovations (R2O) into daily weather operations, reducing the multi-year lag that previously plagued the industry.