Weather prediction is entering a new phase: AI systems are now targeting the rare events that matter most. HiRO-ACEv2 can generate global climate data at 3 km resolution for about 1/300th the cost of X-SHiELD, while Microsoft’s PBC improves forecasts two to six weeks ahead.
Those advances reach beyond better day-to-day predictions. They could give scientists more usable data for studying hurricanes, rainfall, wind, pressure, climate patterns, and the conditions linked to extreme heat. The central question is no longer whether AI can imitate weather, but whether it can reproduce the events that shape decisions and cause the greatest damage.
HiRO-ACEv2 Targets Weather’s Rarest Extremes
HiROv2 can reproduce extremely rare rain, wind, and pressure events with high accuracy, including events that occur only once in a million data points. HiRO-ACEv2 matches the high-resolution statistics of its target simulation to within 3%, giving researchers a way to examine extreme conditions without paying the full cost of the X-SHiELD model.
Over a 10-year run, HiRO-ACEv2’s statistics closely match X-SHiELD’s. The system also captures details inside tropical cyclones: HiROv2 estimates the total energy of tropical cyclones, including storms it never saw during training, and generates hurricanes with realistic eyes and rain bands.
There is still a clear area for improvement. Winds and pressure around the eye remain more lopsided than they are in X-SHiELD, even as the system reproduces the broader structure of hurricanes. HiROv2 also generates strong winds flowing down mountain slopes, adding another demanding test to its simulation abilities.
The cost difference drives the biggest opportunity. Producing global high-resolution data with HiRO-ACEv2 costs roughly 300 times less than running X-SHiELD. When researchers need only one region, the savings can become more than a hundred times greater, depending on the area’s size.
That shift turns high-resolution climate data into a more accessible research tool. Lester Mackey captured the promise: “I have a lot of hope both for the increased resource efficiency of weather forecasting and the extended reliability of forecasts, especially for extreme weather.”
Machine Learning Extends Forecasts Beyond Today
Microsoft’s PBC tool takes a different path. Instead of generating a high-resolution climate simulation, it improves weather forecasts two to six weeks in advance and can run on CPUs, making it resource-efficient.
In the study’s analysis, PBC improved the accuracy of ECMWF’s weather model for 98 percent of precipitation targets and 92 percent of temperature targets. It also doubled the accuracy of ECMWF’s new AI forecasting system.
PBC is lightweight and easy to train, and scientists can configure it without huge resources. That combination matters because a forecasting tool must fit the working conditions of researchers, not demand an entire computing system before it becomes useful.
Genevieve Flaspohler, executive director of Rhiza Research, is an author of the PBC research. Rhiza Research collaborated with the Kenya Meteorological Service Authority, with funding from the Gates Foundation. The work connects machine learning research with forecasting needs that reach beyond the largest research centers.
Karishma V. Patel, a co-author of heat prediction research, described the practical test for AI: “Where AI can actually add value is where prediction is central to the decision-making, and that’s in scale or providing more granularity.” She added, “What is important is not just that we can build AI-enabled technology, but it actually solves a problem or removes a bottleneck.”
Climate Patterns, Cyclones, and the Rising Heat Risk
ACE2 simulations add another layer to the picture by testing how ocean temperature patterns affect tropical cyclones. The simulations reveal that Atlantic and global tropical cyclone numbers decreased in the 15th Century under an El Niño-like SST pattern.
Global tropical cyclone frequency decreased mainly because stronger vertical wind shear and a drier midtroposphere emerged as the Hadley circulation strengthened. Idealized SST experiments with ACE2 show the opposite result under a La Niña-like SST pattern: global tropical cyclone frequency would increase.
Wenchang Yang and Mu-Ting Chien are authors of the ACE2 study, along with Elizabeth A. Barnes. Yang is faculty at Princeton University, while Eric D. Maloney is faculty at Colorado State University and Gabriel A. Vecchi is faculty at Princeton University.
The ACE2 results also connect with the growing need to understand heat. U.S. deaths associated with high temperatures rose 53%, from an annual average of 2,670 in 2000–2009 to more than 4,000 in 2010–2020.
Rebecca Falconer is an author of the AI’s extreme heat challenge article, and Patel is a co-author of heat prediction research. Together, these developments point toward a common goal: better forecasts, more detailed simulations, and stronger evidence about the risks communities face.
HiRO-ACEv2, PBC, and ACE2 approach weather from different directions, yet their value comes from the same breakthrough. AI can reduce the cost of high-resolution modeling, extend useful forecasts, and test climate patterns that shape extreme events. As these tools advance, the next measure of success will be clear: whether they help turn difficult weather data into decisions made early enough to matter.
Based on
- HiRO-ACEv2: Simulating extreme weather at 3 km resolution for a fraction of the cost — huggingface.co
- Microsoft has a new machine learning tool to improve weather forecasts | The Verge — theverge.com
- Deep-learning climate emulator ACE2 reveals a global decrease in 15th-century tropical cyclone frequency under El Niño-like sea surface temperatures | npj Climate and Atmospheric Science — nature.com
- AI’s extreme heat challenge is moving from prediction to protection — axios.com
