Probabilistic AI models outperform numerical weather prediction models on record breaking heat extremes
Abstract
AI weather prediction (AIWP) models have emerged as a key complement to physics-based numerical weather prediction (NWP) models, and now often outperform them on aggregate skill. Yet, several studies report that physics-based models still forecast extreme events better, in particular record-breaking extremes, based on the evaluation of deterministic forecasts. Here we evaluate heat extreme forecasts from probabilistic AIWP models over global land and show that they clearly outperform the ECMWF IFS ensemble, both raw and post-processed, including for record-breaking events, in all regions and at all lead times considered. AIWP models better forecast the occurrence and intensity of extremes, are better calibrated in the tail and miss fewer records. Station observations confirm these results. We argue that many reported AIWP limitations stem from unsuitable verification choices, specifically comparing deterministic forecasts that target different quantities. Instead, a fair evaluation of extremes necessitates probabilistic verification.
Related
- Can AI Forecast the Extremes? Evaluating Probabilistic Weather Models on Tropical Cyclones
- Dynamical-Model-Informed Stochastic Flow Matching for Global Probabilistic Subseasonal Forecasting of Means and Extremes
- EastAsiaClimateExtremes: An AI-Ready Dataset of Weekly Atmospheric and Oceanic Extremes over East Asia for Subseasonal Prediction Research
- Butterfly Effect and the Kinetic Energy Cascade in Probabilistic Machine Learning Weather Prediction Models
- Comparison of data-driven and numerical weather prediction models for rainfall nowcasting in major cities of Bangladesh
BibTeX
@misc{wessel2026probabilistic,
title = {{Probabilistic AI models outperform numerical weather prediction models on record breaking heat extremes}},
author = {Jakob Benjamin Wessel and Fiona Spuler and Christopher A. T. Ferro and Marlene Kretschmer and Sebastian Sippel},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.20519},
eprint = {2610.20519},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.20519},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Probabilistic AI models outperform numerical weather prediction models on record breaking heat extremes
AU - Jakob Benjamin Wessel
AU - Fiona Spuler
AU - Christopher A. T. Ferro
AU - Marlene Kretschmer
AU - Sebastian Sippel
PY - 2026
DA - 2026/10/14
PB - arXiv
DO - 10.48550/arxiv.2610.20519
UR - https://arxiv.org/abs/2610.20519
AB - AI weather prediction (AIWP) models have emerged as a key complement to physics-based numerical weather prediction (NWP) models, and now often outperform them on aggregate skill. Yet, several studies report that physics-based models still forecast extreme events better, in particular record-breaking extremes, based on the evaluation of deterministic forecasts. Here we evaluate heat extreme forecasts from probabilistic AIWP models over global land and show that they clearly outperform the ECMWF IFS ensemble, both raw and post-processed, including for record-breaking events, in all regions and at all lead times considered. AIWP models better forecast the occurrence and intensity of extremes, are better calibrated in the tail and miss fewer records. Station observations confirm these results. We argue that many reported AIWP limitations stem from unsuitable verification choices, specifically comparing deterministic forecasts that target different quantities. Instead, a fair evaluation of extremes necessitates probabilistic verification.
ER -