HiRO-ACEv2: Ai2's 3 km Climate Emulator Cuts Extreme Weather Simulation Cost by ~300x

TL;DR

HiRO-ACEv2 is the newest update to Ai2's High Resolution Output Ai2 Climate Emulator (HiRO-ACE) system, and its new downscaling model HiROv2 can reproduce extremely rare rain, wind, and pressure events while generating 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 update matters because it makes it practical to generate many high-resolution versions of extreme-weather data over 10 years or more, rather than paying the full cost of running an explicit high-resolution simulation.

What HiRO-ACEv2 is and what changed

HiRO-ACEv2 is the newest update to Ai2's High Resolution Output Ai2 Climate Emulator (HiRO-ACE) system. HiRO-ACE was introduced earlier in 2026 and 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 biggest change in HiRO-ACEv2 is the new downscaling model, HiROv2. HiROv2 differs from its predecessor in two ways: it can predict several high-resolution fields at once instead of just rainfall, which lets it capture a much wider range of extreme weather, and it has improved skill in predicting extreme events.

Ai2 states that its main focus with HiRO-ACEv2 was to simulate extreme events well, and that changes to how HiROv2 is trained improved its ability to reproduce the most intense rain, wind, and pressure values.

Accuracy on extreme events

HiROv2 matches the target simulation's high-resolution statistics to within 3% when given coarsened X-SHiELD data, even for events so rare they occur only once in a million data points. That is the headline accuracy claim for the new downscaling model.

Additional reported capabilities of HiROv2:

It accurately estimates the total energy of tropical cyclones, including storms it never saw during training.
It generates hurricanes with realistic eyes and rain bands, though winds and pressure around the eye remain somewhat more lopsided than in X-SHiELD.
It realistically generates strong winds flowing down mountain slopes.

The updated ACE model

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. Ai2 describes this as evidence that ACE's coarse simulations are a solid foundation for producing long-term high-resolution climate data.

End-to-end results over a full year and a 10-year run

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 from that run show a variety of other extreme events that resemble those in X-SHiELD.

Cost and speed

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 also states it is working on a streamlined version of HiROv2 that could make it more than ten times faster still.

Current limitations

HiRO learned from X-SHiELD, so it only works with the version of ACE tuned to match X-SHiELD. As a result, it cannot yet produce high-resolution data for other climates, such as a warmer future or the historical record in datasets like ERA5.

Planned future work

Ai2 plans to train on new X-SHiELD datasets, including one simulating a world 4°C warmer, so the system can handle different climates. It also plans to add outputs such as near-surface temperature and snow and ice, which would allow for additional uses including snowpack prediction.

How to access and inspect the outputs

Ai2 built 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 requires no inference to run and complements the quantitative comparisons in the technical report: it offers a quick visual check of how the outputs look and lets users zoom in on their own region of interest to judge performance there.

For more in-depth analysis, the models are available to download, and Ai2's tutorial notebooks walk through how to run them. The release links to the models on Hugging Face, a paper on arXiv, tutorial notebooks, and a demo site.

Why this matters

HiRO-ACEv2 offers researchers a flexible and inexpensive way to generate many versions of high-resolution rain, wind, and pressure data for extreme events over 10 years or more. The combination of a 3 km output resolution, a roughly 300x cost reduction relative to X-SHiELD for global data, and demonstrated skill on once-in-a-million events is what makes the system useful for long-horizon extreme-weather analysis rather than only for short illustrative runs.