Hugging Face and Ai2 have released HiRO-ACEv2, an update to their AI climate emulator that produces global 3 km resolution weather data at a fraction of the cost of traditional high-resolution simulations. The system pairs ACE, a coarse 100 km model, with HiROv2, a downscaling model that sharpens those outputs to 3 km detail. The full pipeline costs roughly 300 times less than running X-SHiELD, a conventional high-resolution atmospheric model, and regional runs can be even cheaper.
The main improvement in HiROv2 is its ability to predict several high-resolution fields at once—not just rainfall—and to better capture extreme events. 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 once in a million data points. It also reproduces realistic hurricane structures and accurately estimates tropical cyclone energy, though winds and pressure near the eye remain somewhat more lopsided than in X-SHiELD.
A 10-year simulation combining ACE and HiROv2 produced hurricane energy and extreme-event patterns comparable to X-SHiELD, even though the coarse model ran on its own rather than using real X-SHiELD data. The authors note a key limitation: because HiRO learned from X-SHiELD, it only works with the version of ACE tuned to match that dataset, so it cannot yet generate high-resolution data for other climates. Future work includes training on a 4°C warmer world and adding outputs like near-surface temperature and snow and ice.