The study used 178 tropical cyclones from two evaluation settings, comparing outputs from three AI models—GenCast, FourCastNet 3 (FCN3), and FGN—to ECMWF's IFS-ENS ensemble.
AI models excel at cyclone track prediction but struggle with intensity and uncertainty
A benchmark of three probabilistic models reveals up to 90% improvement in storm location over a leading numerical weather prediction system, while intensity biases and calibration errors persist.
Research Lab
Nathan Chalumeau · Xavier Renard · Remi Meynadier · Hugo Rakotoarimanga · Marcin Detyniecki · Emmanuel de Bezenac
INRIA · AXA · Sorbonne Université · CNRS · LIP6
Research Digest··2 min read
Chalumeau et al.
Why this paper
From INRIA and 6 others
In one line
Leading AI weather models outperform numerical forecasts on cyclone tracking but retain intensity biases and calibration errors.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors
- ✓Reports numbers on named benchmarks
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§