aceModerate evidence
More
AI weather models tracked Hurricane Helene but underestimated its winds, a case study finds
Study title: Artificial Intelligence Weather Prediction Model Performance for Hurricane Helene (2024)
- Evidence type
- Benchmark or lab evaluation
- Sources
- 1 sourcePeer-reviewed
- Published
- 2 min read
- Truvace record
- TRV-2026-2612cite or check it
Study title: Artificial Intelligence Weather Prediction Model Performance for Hurricane Helene (2024)
A peer-reviewed case study in Weather and Forecasting compared short-range forecasts of Hurricane Helene (2024) from three open-source AI weather models (Pangu-Weather, FourCastNet version 2 and GraphCast) with the NCEP Global Forecast System (GFS). The AI models accurately captured the hurricane's track but underestimated its intensity in low-level winds and surface pressure, unlike the physics-based GFS.
Read the study abstract
Abstract The recent emergence of machine learning in the field of weather prediction has shown remarkable promise. Deep learning models have proven to be a promising alternative to traditional physics and dynamics-based models with significantly lower computational costs given their high efficiency. While the performance of new artificial intelligence weather prediction (AIWP) models has demonstrated continuous improvement for general weather patterns, their efficacy in predicting high-impact, extreme weather ev…
48Gain side
How the two sides scoreClose call
50Problem side
The gain
In a case study of Hurricane Helene (2024), three AI weather models (Pangu-Weather, FourCastNet version 2 and GraphCast) accurately captured the storm's track, as did the physics-based GFS.
The problem
In the same Hurricane Helene case study, the three AI models underestimated the storm's intensity in low-level winds and surface pressure, unlike the physics-based GFS model.
A fixed formula sets both scores from the evidence and its sources. Reader votes don’t move them.
What the scores mean
G is the gain side, out of 100. It goes up when the gain is specific, when it measurably helped people and when the sources behind it are strong.
P is the problem side, out of 100. It goes up when the problem is specific, when it measurably hurt people and when the sources behind it are strong.
In brief
The AI models and the GFS predicted similar environments supporting Helene's intensification, including deep-layer vertical wind shear and high midlevel relative humidity. That suggests the intensity shortfall comes from how the models represent the storm itself, not from misreading the surrounding conditions.
The result matters because AI models run at much lower computational cost than traditional physics-based models, yet their performance on high-impact extreme events has been little studied. Intensity is a key input to hazard forecasts, so underestimating it limits how far these models can be relied on operationally.
Main points
- Three open-source AI models were tested: Pangu-Weather, FourCastNet version 2 and GraphCast.
- The comparison baseline was the NCEP Global Forecast System, a physics-based model.
- The AI models and the GFS predicted similar environments for intensification, including deep-layer vertical wind shear and high midlevel relative humidity.
- The AI models accurately captured Helene's track.
- The AI models underestimated cyclone intensity in low-level winds and surface pressure, unlike the GFS.
- The authors frame the work as a case study to complement research on AI forecasts of extreme events and their operational utility.
The full story
The study evaluates short-range forecasts of Hurricane Helene (2024), comparing three open-source machine learning weather models with the NCEP Global Forecast System. It looks at both the large-scale environment and the storm's own track and intensity.
On environment and track, the AI models performed well: they predicted conditions similar to the GFS that supported intensification and accurately captured the hurricane's path. On intensity, they fell short, underestimating low-level winds and surface pressure where the GFS did not.
The authors say the work aims to complement existing research on how faithfully AI forecasts represent extreme events, particularly for operational forecasting where successive initialization updates provide guidance for refining hazard predictions.
What this doesn’t fix
The analysis covers a single storm, Hurricane Helene, and three open-source AI models, so the findings may not apply to other extreme events or models.
Sources
1 on file- Weather and Forecastingdoi:10.1175/waf-d-25-0175.1
Talk it over
This record found a gain and a problem in the same place. Which matters more to you, and who might answer differently?
AI weather models tracked Hurricane Helene but underestimated its winds, a case study findshttps://truvace.com/item/artificial-intelligence-weather-prediction-model-performance-for-hurricane-helen
This record found a gain and a problem in the same place. Which matters more to you, and who might answer differently?
The gain In a case study of Hurricane Helene (2024), three AI weather models (Pangu-Weather, FourCastNet version 2 and GraphCast) accurately captured the storm's track, as did the physics-based GFS.
The problem In the same Hurricane Helene case study, the three AI models underestimated the storm's intensity in low-level winds and surface pressure, unlike the physics-based GFS model.
Answer in the threadHow do we know? What can this kind of evidence show, and what can’t it?
The evidence This rests on a benchmark or lab test, from Weather and Forecasting.
Answer in the threadWhat does it leave open? What else would you want to know before acting on it?
The record’s own caveat The analysis covers a single storm, Hurricane Helene, and three open-source AI models, so the findings may not apply to other extreme events or models.
Answer in the threadHave you seen AI used this way where you study, work or live? Did it go the same way?
Answer in the thread
The questions are the same on every record. The details under them are taken from this record as written.
What readers say