HiRO-ACEv2
Ai2's climate simulation framework combines a coarse atmospheric emulator with diffusion downscaling for 3 km precipitation, wind and pressure fields.
What happened lately
HiRO-ACEv2 timeline
HiRO-ACEv2 pairs ACE2S-X-SHiELD, an atmospheric emulator operating at about 100 km resolution, with HiROv2, a diffusion model that produces 3 km precipitation, near-surface wind and sea-level pressure fields. The public repository was created October 12, before Ai2’s October 14 announcement. It includes both checkpoints, forcing data, initial conditions and inference configurations. The English quickstart uses the fme Python package.
The model card lists Apache 2.0 and describes research and educational use. Its training targets are outputs from the X-SHiELD physical climate model, rather than observations. The authors explicitly say it is unsuitable for weather forecasting and is not expected to generalize beyond its training climate. Regional coverage and patch-blending artifacts are additional limitations documented in the card.