> ## Content Index
> Fetch the complete content index at: https://globalfeed.ai/llms.txt
> Use this file to discover other available public pages before exploring further.

# Ai2's open HiRO-ACEv2 simulates extreme weather at 3 km for a fraction of the cost
- URL: https://globalfeed.ai/en/ai2s-open-hiro-acev2-simulates-extreme-weather-at-3-km-for-a-fraction-of-the-cost/
- Published: 2026-10-20T02:32:09.000Z
- Updated: 2026-10-20T02:32:09.000Z
- Description: Ai2 has released HiRO-ACEv2 under Apache 2.0. It takes coarse climate simulations down to 3 km detail and reproduces heavy rain, wind and hurricanes at roughly 1/300th the cost of the physics model it was compared with.
- Author: GlobalFeed Editor
- Tags: Ai2, HiRO-ACEv2, climate, weather forecasting, open model, x-allen_ai, dil-en, elle

The Allen Institute for AI (Ai2) introduced [HiRO-ACEv2](https://huggingface.co/blog/allenai/hiro-acev2?ref=globalfeed.ai) on October 14, 2026\. The system lets climate researchers generate global rain, wind and pressure data at 3 km resolution. The models can be downloaded from [Hugging Face](https://huggingface.co/allenai/HiRO-ACEv2?ref=globalfeed.ai) under Apache 2.0, with tutorial notebooks for running them on your own server. The technical report is on arXiv.

## How the system works

HiRO-ACE has two parts. ACE is an AI model that simulates the climate at a coarse 100 km resolution. HiRO then sharpens that coarse output into fine 3 km detail. The model was trained on data from X-SHiELD, a high-resolution global atmospheric model that explicitly simulates storms and other small-scale weather. The biggest change in the new version is HiROv2: the earlier version predicted only rainfall, while the new one produces several high-resolution fields at once and captures extreme events better.

## Performance on extreme weather

According to Ai2, when given coarsened X-SHiELD data, HiROv2 matches the target simulation's high-resolution statistics to within 3 percent, even for events that occur once in a million data points. It accurately estimates the total energy of tropical cyclones, including storms it never saw in training, and generates hurricanes with realistic eyes and rain bands. The team notes that winds and pressure around the eye remain somewhat more lopsided than in X-SHiELD. The model also realistically produces strong winds flowing down mountain slopes.

With an updated version of ACE, the team generated a full year of global 3 km data. In a 10-year run, the energy of its hurricanes was comparable to X-SHiELD's over the same period. Ai2 says producing global high-resolution data this way costs roughly 300 times less than running X-SHiELD, and when only one region is needed, the savings can be more than a hundred times greater depending on the area. The team is also working on a version of HiROv2 that could be more than ten times faster.

## Limits and next steps

Because HiRO learned from X-SHiELD, it currently works only with the version of ACE tuned to match X-SHiELD. That means it cannot yet produce high-resolution data for other climates, such as a warmer future or historical records in datasets like ERA5\. The team plans to train on new X-SHiELD data, including a world 4°C warmer, and to add outputs such as near-surface temperature, snow and ice. A web demo shows pre-generated output so users can zoom in on their own region.

The results rely on Ai2's own comparisons, and the model emulates a single physics simulation it was trained on. It is worth following independent studies that compare its output with real observations, especially before using it for regional planning.