Hugging Face introduces HiRO-ACEv2, an AI climate emulator that generates global 3 km data at about 1/300th the cost of X-SHiELD, with improved extreme event skill.
HiRO-ACEv2: Simulating extreme weather at 3 km resolution for a fraction of the cost
the full system can generate global 3 km climate data at about 1/300th the cost of X-SHiELD
it matches the target simulation's high-resolution statistics to within 3%, even for events so rare they occur only once in a million data points
the energy of HiRO-ACEv2's hurricanes is comparable to X-SHiELD's over the same period
Hugging Face introduced HiRO-ACEv2, the newest update to the High Resolution Output Ai2 Climate Emulator (HiRO-ACE) system. The biggest change is the new downscaling model, HiROv2, which can now predict several high-resolution fields at once instead of just rainfall, capturing a wider range of extreme weather. Changes to training improved its ability to reproduce the most intense rain, wind, and pressure values. When given coarsened X-SHiELD data, it matches the target simulation's high-resolution statistics to within 3%, even for events so rare they occur only once in a million data points. The system also accurately estimates the total energy of tropical cyclones, including storms it never saw during training, and generates hurricanes with realistic eyes and rain bands, though winds and pressure around the eye remain somewhat more lopsided than in X-SHiELD.
Editorial interpretation: The source states the full system can generate global 3 km climate data at about 1/300th the cost of X-SHiELD, a high-resolution global atmospheric model. This cost reduction could make high-resolution climate simulation more accessible for researchers who need many versions of extreme event data. The source also notes that if only one region is needed, the savings can be more than a hundred times greater, depending on the size of the area. The system can generate many versions of high-resolution rain, wind, and pressure data for extreme events over 10 years or more, which may support long-term climate studies. The source reports that across a 10-year run, the energy of HiRO-ACEv2's hurricanes is comparable to X-SHiELD's over the same period.
The source describes a web demo that shows pre-generated precipitation, wind, and sea-level pressure output from HiROv1 and HiROv2 side by side with X-SHiELD, the simulation they were trained on. The demo allows users to zoom in on their own region of interest to judge performance there. The models are available to download, and tutorial notebooks walk through how to run them. The source also mentions a tech report with quantitative comparisons. These resources let readers check the visual and statistical claims directly.
The supplied feed entry is a concise summary; consult the canonical article for full context. The source notes future work: because HiRO learned from X-SHiELD, it only works with the version of ACE tuned to match X-SHiELD. It can't yet produce high-resolution data for other climates, such as a warmer future or the historical record in datasets like ERA5. The source also states plans to train on new X-SHiELD datasets, including one simulating a world 4°C warmer, and to add outputs like near-surface temperature and snow and ice. A streamlined version of HiROv2 is being worked on that could make it more than ten times faster still, but no timeline is given.
HiRO-ACEv2 is the newest update to the High Resolution Output Ai2 Climate Emulator (HiRO-ACE) system, introduced by Hugging Face. It has two parts: ACE, an AI model that operates at a coarser 100 km resolution, and HiRO, a model that sharpens those coarse outputs into fine 3 km detail. The update focuses on simulating extreme events well, with improved skill in predicting extreme rain, wind, and pressure values.
According to the source, the full system can generate global 3 km climate data at about 1/300th the cost of X-SHiELD, a high-resolution global atmospheric model. If only one region is needed, the savings can be more than a hundred times greater, depending on the size of the area. The source also notes that producing global high-resolution data with it costs roughly 300 times less than running X-SHiELD.