Create and explore imaginative 3D worlds with our new product Dreamfield (opens in a new tab)

Determine the cause of the lower IFS-ENS cyclone-track detection rate

Determine whether the lower cyclone-track detection rate of IFS-ENS in the 2025 operational comparison reflects the storm sample, forecast uncertainty, or differences in forecast structure and tracking.

Background

The operational comparison evaluates FGN, operational GenCast, and IFS-ENS on 2025 tropical cyclones using both probabilistic location scores and track-detection rates. Although both AI models achieve higher detection rates and better negative log-likelihood scores than IFS-ENS, the paper notes that the changing numerical baseline prevents attributing differences between the reanalysis and operational settings solely to AI fine-tuning.

The authors identify an unresolved attribution problem concerning IFS-ENS’s lower detection rate. It may result from the composition of the 2025 storm sample, uncertainty in the forecast, or differences in the forecast fields and the tracking procedure. Resolving this issue would clarify whether the observed detection-rate gap reflects model performance, event selection, or methodological effects.

References

Whether its lower detection rate reflects the storm sample, forecast uncertainty, or differences in forecast structure and tracking remains unresolved.

— Can AI Forecast the Extremes? Evaluating Probabilistic Weather Models on Tropical Cyclones  (2610.22077 - Chalumeau et al., 15 Oct 2026) in Section 4, “Operational setting: tracking, intensity, and spatial agreement for models fine-tuned on operational analyses,” paragraph “Cyclone tracking”