Janet Wang

Xiao Wang, Zezhong Zhang, Isaac Lyngaas et al.

Accurate weather and climate prediction relies on data assimilation (DA), which estimates the Earth system state by integrating observations with models. While exascale computing has significantly advanced earth simulation, scalable and accurate inference of the Earth system state remains a fundamental bottleneck, limiting uncertainty quantification and prediction of extreme events. We introduce a unified one-stage generative DA framework that reformulates assimilation as Bayesian posterior sampling, replacing the conventional forecast-update cycle with compute-dense, GPU-efficient inference. At the core is STORM, a novel spatiotemporal transformer with a global attention linear-complexity scaling algorithm that breaks the quadratic attention barrier. On 32,768 GPUs of the Frontier supercomputer, our method achieves 63% strong scaling efficiency and 1.6 ExaFLOP sustained performance. We further scale to 20 billion spatiotemporal tokens, enabling km-scale global modeling over 177k temporal frames, regimes previously unreachable, establishing a new paradigm for Earth system prediction.

Janet Wang, Yunbei Zhang, Lin Zhao et al.

Existing ViT-based weather forecasting models apply uniform computation across all spatial tokens, even though nearby atmospheric grid points often contain similar values and large regions evolve smoothly over time. This makes much of the intermediate per-token computation redundant. Standard token-efficiency methods, such as pruning or merging, reduce cost by removing or fusing tokens. However, weather forecasting is a spatiotemporal dense prediction problem in which a history of atmospheric states must be mapped to future values on the original latitude-longitude grid. Thus, every grid cell must retain a physically meaningful representation, especially under autoregressive rollout. We introduce Sparse-Reslim, a parameter-free plug-in routing module that makes sparse token processing compatible with this fixed-grid requirement. Sparse-Reslim routes only 25% of spatial tokens through the expensive middle transformer blocks and treats those blocks as residual updates: it computes the change produced for the routed tokens and scatters only this delta back to the full sequence. Unselected tokens keep their pre-routing representations exactly, so no grid cell is dropped or replaced by a mask token, and no fusion layer or additional parameters are introduced. Across ERA5 resolutions up to the operational 0.25\textdegree{} standard and two model families, a deterministic Transformer and a diffusion model, Sparse-Reslim improves forecast accuracy on every evaluated variable while substantially reducing cost: training is about 2.5x faster in the main settings and reaches 3.18x speedup at 0.25\textdegree{}, with over 2.2x lower peak memory. A controlled decomposition shows that the accuracy gain comes primarily from sparse routing itself, while random token selection provides an additional regularization benefit without selector overhead.