physics.ao-phNov 7, 2026

Guiding future global temperature distributions with simulations and observations

Authors: Francesco Immorlano, Elijah Tavares, Felix Draxler, Juan Nathaniel, Violette Launeau, Duncan Watson-Parris, Padhraic Smyth, Pierre Gentine, 

Organizations: Department of Computer Science, University of California Irvine, Irvine, CA 92697 · Learning the Earth with AI and Physics, New York, NY 10027 · Department of Earth and Environmental Engineering, Columbia University, New York, NY 10027 · Celest Science, Paris, France

Abstract

Assessing climate risks requires projections of daily temperature distributions, including rare and spatially widespread extremes. Climate simulations are costly and biased, while correcting them toward observations risks suppressing inter-model differences. We train a single machine-learning emulator, a conditional diffusion model, on seventeen CMIP6 models and ERA5 reanalysis to generate spatially coherent daily temperature fields conditioned on model identity, date, and CO2-equivalent concentration. At sampling time, diffusion guidance reduces model biases by steering each emulated climate-model distribution toward observations, with its strength selected through perfect-model experiments using withheld simulated futures. Observational influence decreases with projection horizon and forcing. Across seventeen synthetic targets, guidance more than halves median biases in the 1st and 99th percentiles of daily temperature, improves probabilistic scores and reliability, and reduces median annual global-mean temperature RMSE by 32-57% across three emission scenarios. In a detailed benchmark, guidance also outperforms established bias corrections on daily-field probabilistic scores. Efficient sampling generates large ensembles without further climate simulations. For example, our framework corrects CMIP6's roughly threefold historical overestimate of extreme-heat coverage, yet projects that concurrent extremes will affect nearly half of global land on a typical day in the 2080s under SSP3-7.0.

Figures & tables

Explore similar work

CardsList
  1. Regional Climate Model Emulation with Diffusion Approaches: What is the Added Value of Generative Machine Learning?

    Jun 12, 2026Mikel N. Legasa, Antoine Doury, Achille Gellens +4Climate ModelingDiffusion Models

  2. Towards accurate extreme event likelihoods from diffusion model climate emulators

    May 5, 2026Peter Manshausen, Noah Brenowitz, Julius Berner +2Diffusion Model GuidanceClimate Science

  3. Steering Diffusion Priors with Sparse Observations for High-Resolution Temperature Downscaling

    Sep 8, 2026Anirudh Avireddy, Manmeet Singh, Shivanshi Singh +4Diffusion Model GuidanceClimate Modeling