Weather Forecast: Sharp Focus express
The atmosphere is a chaotic system, and a detailed global forecast requires monstrous computing power. StretchCast sidesteps this by working like the human eye: it creates 'sharp vision' over a selected region with an ultra-fine grid, while seeing the rest of the world blurry, but sharp enough to capture global connections.
The model was trained on 25 years of observational data, where Sun heats the oceans, Water evaporates, and winds are born. A grid of just 8,000 cells (hundreds of times fewer than conventional models) spans the planet and accurately reproduces typhoons, seamlessly stitching forecasts at zone boundaries. The atmosphere's chaos is entropy, a measure of disorder, and variable resolution prevents computational noise from swamping the calculations.
The idea isn't new: back in the 1940s, John von Neumann dreamed of numerical weather prediction. Today, StretchCast proves that detailed regional forecasts are achievable even without supercomputers: it’s a sharp eye for every corner of the Earth.
Original abstract
as written by the authors arXiv:2603.27288
Global AI weather forecasting still relies mainly on uniform-resolution models, making it hard to combine regional refinement, two-way regional-global coupling, and affordable training cost. We introduce StretchCast, a global-regional AI forecasting framework built on a variable-resolution stretched cubed-sphere (SCS) mesh that preserves a closed global domain while concentrating resolution over a target region. Within this framework, we develop a one-step predictor, SCS_Base Model, and a rollout-oriented multistep predictor, SCS_FCST4 Model, to test the feasibility of SCS-based forecasting and the benefit of joint multistep training. Experiments use ERA5 with 69 variables over 1998-2022. Because training compute remains limited, this study uses a coarse-resolution proof-of-concept configuration rather than a final high-resolution system. Even with only about 7,776 effective global grid cells and roughly 0.875 degree resolution over the center-refined face, the 23M-parameter SCS_Base Model yields stable multivariate forecasts. With 83M parameters and training cost on the order of hours, SCS_FCST4 Model delivers competitive medium-range anomaly-correlation evolution over the target region after unified reprojection, especially for geopotential height, specific humidity, and part of the lower-tropospheric winds, while maintaining smooth cross-face continuity and realistic multiscale structure in typhoon and spectral analyses. These results support StretchCast as a practical lightweight foundation for global-regional AI weather forecasting.
To cover the Earth without distortion at the poles, StretchCast uses a grid like a soccer ball's — panels of different sizes.