Conditional diffusion models for downscaling and bias  correction of Earth system model precipitation

Aich, Michael; Hess, Philipp; Pan, Baoxiang; Bathiany, Sebastian; Huang, Yu; Boers, Niklas

doi:10.5194/gmd-19-1791-2026

Articles | Volume 19, issue 4

https://doi.org/10.5194/gmd-19-1791-2026

© Author(s) 2026. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/gmd-19-1791-2026

© Author(s) 2026. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 19, issue 4

Development and technical paper

|

03 Mar 2026

Development and technical paper |

| 03 Mar 2026

Conditional diffusion models for downscaling and bias correction of Earth system model precipitation

Michael Aich, Philipp Hess, Baoxiang Pan, Sebastian Bathiany, Yu Huang, and Niklas Boers

Supplement

https://doi.org/10.5194/gmd-19-1791-2026-supplement

Data sets

ERA5 hourly data on single levels from 1940 to present H. Hersbach et al. https://doi.org/10.24381/cds.adbb2d47

NOAA-GFDL GFDL-ESM4 model output prepared for CMIP6 CMIP historical J. P. Krasting et al. https://doi.org/10.22033/ESGF/CMIP6.8597

Model code and software

Model weights for Conditional diffusion models for downscaling \& bias correction of ESM precipitation Michael Aich https://doi.org/10.5281/zenodo.18069119

aim56009/ESM_cdifffusion_downscaling_bc: GMD (Version v0) Michael Aich https://doi.org/10.5281/zenodo.18368891

Short summary

Accurately simulating rainfall is essential to understand the impacts of climate change, especially extreme events such as floods and droughts. Climate models simulate the atmosphere at a coarse resolution and often misrepresent precipitation, leading to biased and overly smooth fields. We improve the precipitation using a machine learning model that is data-efficient, preserves key climate signals such as trends and variability, and significantly improves the representation of extreme events.