AI2 Releases HiRO-ACEv2 Climate Emulator with 3 km Global Data at 1/300th of X-SHiELD's Cost

The Allen Institute for AI (AI2) has released HiRO-ACEv2, the latest version of its High Resolution Output Ai2 Climate Emulator (HiRO-ACE). The system can generate global 3 km climate data at roughly 1/300th the cost of X-SHiELD, a high-resolution global atmospheric model that explicitly simulates storms and other small-scale weather.

HiRO-ACE combines ACE, an AI model running at a coarser 100 km resolution, with HiRO, a downscaling model that sharpens coarse outputs into 3 km detail. The main update in HiRO-ACEv2 is the new downscaling model HiROv2, which can predict several high-resolution fields—such as precipitation, wind, and sea-level pressure—simultaneously and shows improved skill in reproducing extreme weather events.

In tests, HiROv2 reproduced intense rain, wind, and pressure values with high-resolution statistics matching the target simulation within 3%, even for events as rare as one in a million data points. It also accurately estimated the total energy of tropical cyclones, including storms not seen during training, and generated hurricanes with realistic eyes and rain bands, though winds and pressure around the eye remained somewhat more lopsided than in X-SHiELD.

The updated ACE model produced hurricanes, average climate patterns, and overall distributions of rain, wind, and pressure that aligned well with X-SHiELD over a 10-year simulation. AI2 also released a web demo, model downloads, and tutorial notebooks. Future work includes training on new X-SHiELD datasets simulating a world 4°C warmer, and adding outputs such as near-surface temperature and snow/ice.