Ai2 releases HiRO-ACEv2 for 3 km extreme weather at about 1/300th the cost of X-SHiELD

Ai2 has introduced HiRO-ACEv2, the newest update to its High Resolution Output Ai2 Climate Emulator (HiRO-ACE), which was introduced earlier in 2026. The system has two parts. ACE is an AI model that operates at a coarser 100 km resolution. HiRO is a model that sharpens those coarse outputs into fine 3 km detail. Ai2 says HiROv2 can reproduce even extremely rare rain, wind and pressure events with high accuracy, while 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 that explicitly simulates storms and other small-scale weather.

The biggest change in v2 is the new downscaling model, HiROv2. It can now predict several high-resolution fields at once instead of just rainfall, so it captures a much wider range of extreme weather, and it has improved skill on extreme events. Ai2 says its main focus was simulating extreme events well, and that changes to how HiROv2 is trained improved its handling of 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. It 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. Winds and pressure around the eye remain somewhat more lopsided than in X-SHiELD. HiROv2 also realistically generates strong winds flowing down mountain slopes.

Ai2 also trained an updated version of ACE for this system. Over a 10-year simulation, its hurricanes, average climate patterns and overall distributions of rain, wind and pressure all line up well with X-SHiELD, which Ai2 reads as a solid foundation for producing long-term high-resolution climate data.

Putting HiROv2 and ACE together, Ai2 generated a full year of global 3 km data. Its statistics closely match X-SHiELD's, even though HiRO was working from a simulation ACE ran on its own rather than from real X-SHiELD data. Across a 10-year run, the energy of HiRO-ACEv2's hurricanes is comparable to X-SHiELD's over the same period, and snapshots show a variety of other extreme events that resemble those in X-SHiELD.

On cost, Ai2 says producing global high-resolution data with HiRO-ACEv2 costs roughly 300 times less than running X-SHiELD. If only one region is needed, the savings can be more than a hundred times greater, depending on the size of the area. Ai2 is also working on a streamlined version of HiROv2 that could make it more than ten times faster still.

There are limits. Because HiRO learned from X-SHiELD, it only works with the version of ACE tuned to match X-SHiELD, so it cannot yet produce high-resolution data for other climates, such as a warmer future or the historical record in datasets like ERA5. Next, Ai2 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, which would allow uses such as snowpack prediction.

To let people see the output without running inference, Ai2 built a web demo showing pre-generated precipitation, wind and sea-level pressure from HiROv1 and HiROv2 side by side with X-SHiELD, with the option to zoom in on a region of interest. For deeper analysis the models are available to download, and tutorial notebooks walk through how to run them.

Key facts

HiRO-ACEv2 combines ACE, an AI model at 100 km resolution, with the new HiROv2 downscaler that sharpens its output to 3 km and predicts several high-resolution fields at once instead of just rainfall.
Given coarsened X-SHiELD data, HiROv2 matches the target simulation's high-resolution statistics to within 3%, even for events that occur only once in a million data points.
Ai2 says the full system produces global high-resolution data at roughly 300 times lower cost than X-SHiELD, with savings more than a hundred times greater if only one region is needed.
Ai2 generated a full year of global 3 km data and a 10-year run in which hurricane energy is comparable to X-SHiELD's.
It cannot yet produce data for other climates such as a warmer future or ERA5-style historical records; training on a 4°C warmer X-SHiELD dataset is planned.

Why it matters

Explicitly simulating storms and small-scale weather at kilometre scale is what X-SHiELD does, and it is expensive. HiRO-ACEv2 is Ai2's attempt to get similar high-resolution rain, wind and pressure statistics from an AI emulator at about 1/300th of the cost. That is what makes it practical to generate many versions of extreme-event data over 10 years or more, which Ai2 pitches as a flexible and inexpensive tool for researchers. The v2 update focuses on the hard part, the rare and intense events, rather than average conditions.

Who it affects

Ai2 frames the system for researchers who need many versions of high-resolution rain, wind and pressure data for extreme events over 10 years or more. Regional users stand to gain most from the cost savings, since Ai2 says the savings can be more than a hundred times greater when only one region is needed. Planned outputs such as near-surface temperature and snow and ice would extend it to uses like snowpack prediction.

How to use it

The models are available to download, and Ai2's tutorial notebooks walk through how to run them. For a quick look without running inference, a web demo shows pre-generated precipitation, wind and sea-level pressure from HiROv1 and HiROv2 beside X-SHiELD, and lets users zoom in on their own region of interest. Ai2 also points to a tech report with quantitative comparisons.

How solid is it

The results are Ai2's own, reported in its blog post, which cites a tech report for the quantitative comparisons. The yardstick throughout is X-SHiELD, the simulation HiRO was trained on, not real-world observations. The within-3% match is for HiROv2 given coarsened X-SHiELD data; the full-year global run, which used a simulation ACE ran on its own, is described as closely matching X-SHiELD statistics. The cost claim is stated only as a ratio to X-SHiELD; no absolute figure is given. The streamlined version that could be more than ten times faster is work in progress.

Risks and caveats

Because HiRO learned from X-SHiELD, it only works with the version of ACE tuned to match X-SHiELD. It therefore cannot yet produce high-resolution data for other climates, such as a warmer future or the historical record in datasets like ERA5. Generated hurricanes have realistic eyes and rain bands, but winds and pressure around the eye remain somewhat more lopsided than in X-SHiELD. Training on new X-SHiELD datasets, including a world 4°C warmer, and adding new outputs are plans, not delivered features.

“Producing global high-resolution data with it costs roughly 300 times less than running X-SHiELD.”

— Ai2, HiRO-ACEv2 blog post