Extremes & hazards★ Highlight

Surface temperature extremes produced by huge machine learning hindcasts of summer 2023

Mark D. Risser, Ankur Mahesh, William D. Collins, Joshua North, Boris Bonev, Karthik Kashinath, Thorsten Kurth, Michael S. Pritchard, Shashank Subramanian

arXiv preprint · 10 April 2026 · Lawrence Berkeley National Laboratory / NERSC, NVIDIA, University of California, Berkeley

Paper PDF

Why it matters. 7,424-member SFNO ensemble shows ML can reveal worst-case heat extremes beyond NWP extrapolation, demonstrating the value of huge ML ensembles.

In one sentence. Spherical Fourier Neural Operator produces 7,424 storyline forecasts of summer 2023, yielding heat extremes beyond NWP ensemble envelopes for 30% of land.

Abstract

The summer of 2023 brought record-breaking heat extremes across the globe. In this work, we investigate how much worse these extremes might have become, given identical large-scale conditions. Numerical weather prediction (NWP) ensembles simulate alternate heatwave storylines, but their limited sample size diminishes their statistical coverage of observed heatwaves and ability to represent worst-case events. Using the Spherical Fourier Neural Operator machine learning (ML) weather model, we generated a huge ensemble of 7,424 storyline forecasts of summer temperature extremes. Despite being constrained by the same initial and boundary conditions and using training data from the same modeling system, the ML ensemble produced extreme heatwave scenarios exceeding temperatures from NWP ensembles. For 30% of global land areas, ML-generated extreme events were outside the prediction envelope from extrapolating smaller NWP ensembles with extreme value theory. This research demonstrates that very large ensembles substantially improve upon NWP-based characterization of worst-case heat extremes.

Labels

Labels and summary are generated by Claude and reviewed weekly. Wrong? Edit this entry.