IPSL-AID: Generative Diffusion Models for Climate Downscaling from Global to Regional Scales

Kishanthan Kingston; Olivier Boucher; Freddy Bouchet; Pierre Chapel; Rosemary Eade; Jean-Francois Lamarque; Redouane Lguensat; Kazem Ardaneh

physics.ao-phMar 23, 2026

IPSL-AID: Generative Diffusion Models for Climate Downscaling from Global to Regional Scales

Authors: Kishanthan Kingston, Olivier Boucher, Freddy Bouchet, Pierre Chapel, Rosemary Eade, Jean-Francois Lamarque, Redouane Lguensat, Kazem Ardaneh

Organizations: Climate Modeling Center, Sorbonne University, CNRS, IPSL, Paris, France · LMD, CNRS, ENS/PSL, Sorbonne University, IPSL, Paris, France · Climate Modeling and Analysis, LLC, Superior, CO, USA · Climate Modeling Center, Sorbonne University, IRD, IPSL, Paris, France

Abstract

Effective adaptation and mitigation strategies for climate change require high-resolution projections to inform strategic decision-making. Conventional global climate models, which typically operate at resolutions of 150 to 200 kilometers, lack the capacity to represent essential regional processes. IPSL-AID is a global to regional downscaling tool based on a denoising diffusion probabilistic model designed to address this limitation. Trained on ERA5 reanalysis data, it generates 0.25 degree resolution fields for temperature, wind, and precipitation using coarse inputs and their spatiotemporal context. It also models probability distributions of fine-scale features to produce plausible scenarios for uncertainty quantification. The model accurately reconstructs statistical distributions, including extreme events, power spectra, and spatial structures. This work highlights the potential of generative diffusion models for efficient climate downscaling with uncertainty

Figures & tables

Figure 1: An example of randomly sampled spatial blocks used during training of the global model

Figure 2: Surface plots are shown for t2m, u10, v10, and tp (columns 1 to 4). The rows, from top to bottom, represent the CU ERA5 input, the HR ERA5 reference, the model prediction, and the difference between the prediction and the HR reference. All panels correspond to 2021-01-01 06:00 UTC

Figure 3: Model evaluation for the complete 2021 evaluation dataset. Columns represent t2m, u10, v10, and tp, respectively. Rows display density scatter plots (top), PDFs (middle), and PSDs (bottom)

	\TCH t2m (K)				\TCH u10 (m s -1 )				\TCH v10 (m s -1 )				\TCH tp (mm hr -1 )
\TCH Quantile ( $q$ )	\TCH Truth	\TCH Diff	\TCH U-Net	\TCH CU	\TCH Truth	\TCH Diff	\TCH U-Net	\TCH CU	\TCH Truth	\TCH Diff	\TCH U-Net	\TCH CU	\TCH Truth	\TCH Diff	\TCH U-Net	\TCH CU
0.90	300.09	300.07	300.08	300.01	7.61	7.60	7.56	7.09	6.10	6.09	6.06	5.64	0.20	0.21	0.21	0.27
0.95	301.20	301.17	301.14	300.97	10.24	10.24	10.19	9.71	7.93	7.91	7.89	7.31	0.47	0.49	0.49	0.49
0.975	302.57	302.57	302.52	302.13	12.33	12.33	12.27	11.77	9.60	9.57	9.55	8.82	0.92	0.94	0.89	0.75
0.99	306.40	306.41	306.35	305.55	14.48	14.47	14.40	13.85	11.62	11.58	11.56	10.66	1.71	1.73	1.56	1.13
0.995	309.21	309.21	309.16	308.30	15.81	15.80	15.72	15.11	13.04	13.00	12.97	11.94	2.43	2.45	2.15	1.45

Table 1: Extreme true, diffusion model, U-Net, and CU input quantiles for the downscaled variables.

Figure 4: Spatial distribution of the MAE, averaged over the 2021 dataset, for the downscaled variables: t2m, u10, v10, and tp

Figure 5: Rank histograms for the 10-member diffusion ensemble. Columns show t2m, u10, v10, and tp. Red dashed lines mark the expected uniform frequency

Figure 6: Density plots of ensemble spread versus RMSE for t2m, u10, v10, and tp. The red dashed line indicates the 1:1 relationship corresponding to perfect spread–skill (SSR = 1)

Appendix figures & tables7 assets

Supplementary material from the paper’s appendix.

Appendix

Figure A1: Data flow and model architecture for the conditional EDM model

	\TCH $N$			\TCH $\sigma_{\min}$			\TCH $\sigma_{\max}$			\TCH $\rho$
\TCH Metric	\TCH 10	\TCH 20	\TCH 40	\TCH 0.002	\TCH 0.02	\TCH 0.2	\TCH 8	\TCH 80	\TCH 100	\TCH -10	\TCH 7	\TCH 10
MAE (K)	0.63	0.55	0.53	0.63	0.58	0.46	0.55	0.63	0.64	0.64	0.63	0.61
RMSE (K)	0.99	0.89	0.85	0.99	0.93	0.76	0.87	0.99	1.00	1.04	0.99	0.96
$R^{2}$	0.998	0.998	0.998	0.998	0.998	0.999	0.998	0.998	0.998	0.998	0.998	0.998
Time	2h38	3h43	6h05	2h38	2h33	2h35	2h33	2h38	2h32	2h32	2h38	2h38

Appendix

Table A1: Ablation study of EDM sampler parameters.

Figure B1: Surface plots for t2m, u10, v10, and tp (columns 1 to 4). The first three rows show three representative members from the 10-member ensemble. The last row shows the ensemble standard deviation. All panels correspond to 2021-05-02 19:00 UTC

Figure B2: Surface plots are shown for t2m, u10, v10, and tp (columns 1 to 4). The first row shows the full-domain ground truth. The second to fifth rows present a zoomed-in view over South America, showing respectively the truth, the U-Net prediction, the diffusion model prediction, and the difference between the diffusion model prediction and the truth. All panels correspond to 2021-01-01 06:00 UTC

Figure B3: Evaluation of residual fields for the complete 2021 evaluation dataset. Columns represent t2m, u10, v10, and tp, respectively. Rows display density scatter plots (top), PDFs (middle), and PSDs (bottom)

Figure B4: Spatial distribution of relative error for t2m, u10, and v10 (first three panels), and absolute error for precipitation (rightmost panel)

Figure B5: Spatial map of the temporal Pearson correlation coefficient between pair variables

Climate Downscaling

Climate downscaling translates coarse climate simulations and observations into finer regional detail for local impact and risk assessment.

See all 27 papers

High-Resolution Climate Projections Using Diffusion-Based Downscaling of a Lightweight Climate Emulator
Feb 13, 2026Haiwen Guan, Dibyajyoti Chakraborty, Moein Darman +3Climate Science Climate Modeling
1Information Sciences and Technology, The Pennsylvania State University, University Park, Pennsylvania · Department of Applied Mathematics, University of California, Santa Cruz, California · 3Allen Institute for Artificial Intelligence (AI2), Seattle, Washington +1
Climate downscaling with likelihood functional flow matching
Oct 24, 2026Zhi-Song Liu, Helmi Toropainen, Ammar Kheder +1Climate Downscaling Flow Matching
Department of Computational Engineering, LUT University, Lappeenranta, Finland. · Institute for Atmospheric and Earth System Research (INAR), University of Helsinki, Helsinki, Finland. · Atmospheric Modelling Center, Lahti (AMC-Lahti), Lahti, Finland.
Physics-Constrained Adaptive Flow Matching for Climate Downscaling
Apr 3, 2026Kevin Debeire, Aytaç Paçal, Pierre Gentine +3Climate Downscaling Physics-Informed Generative Modeling
Deutsches Zentrum für Luft- und Raumfahrt (DLR), Institut für Physik der Atmosphäre, Oberpfaffenhofen, Germany · School of Computation, Information and Technology, Technical University of Munich, Germany · LEAP Center, Columbia University, New York, NY, 10027, USA +1

Feb 13, 2026J/K move · Enter open · S save

High-Resolution Climate Projections Using Diffusion-Based Downscaling of a Lightweight Climate Emulator

Haiwen Guan, Dibyajyoti Chakraborty, Moein Darman, Troy Arcomano+2

1Information Sciences and Technology, The Pennsylvania State University, University Park, Pennsylvania · Department of Applied Mathematics, University of California, Santa Cruz, California · 3Allen Institute for Artificial Intelligence (AI2), Seattle, Washington+1

Climate Science Climate Modeling Conditional Diffusion Models Climate Downscaling Diffusion Models

Open paper AskPDF arXiv

The proliferation of data-driven models in weather and climate sciences has marked a significant paradigm shift, with advanced models demonstrating exceptional skill in medium-range forecasting. However, these models are often limited by long-term instabilities, climatological drift, and substantial computational costs during training and inference, restricting their broader application for climate studies. Addressing these limitations, Guan et al. (2024) introduced LUCIE, a lightweight, physically consistent climate emulator utilizing a Spherical Fourier Neural Operator (SFNO) architecture. This model is able to reproduce accurate long-term statistics including climatological mean and seasonal variability. However, LUCIE's native resolution (~300 km) is inadequate for detailed regional impact assessments. To overcome this limitation, we introduce a deep learning-based downscaling framework, leveraging probabilistic diffusion-based generative models with conditional and posterior sampling frameworks. These models downscale coarse LUCIE outputs to 25 km resolution. They are trained on approximately 14,000 ERA5 timesteps spanning 2000-2009 and evaluated on LUCIE predictions from 2010 to 2020. Model performance is assessed through diverse metrics, including latitude-averaged RMSE, power spectrum, probability density functions and First Empirical Orthogonal Function of the zonal wind. We observe that the proposed approach is able to preserve the coarse-grained dynamics from LUCIE while generating fine-scaled climatological statistics at ~28km resolution.

Abstract

Climate Downscaling

Explore similar work

High-Resolution Climate Projections Using Diffusion-Based Downscaling of a Lightweight Climate Emulator

Climate downscaling with likelihood functional flow matching

Physics-Constrained Adaptive Flow Matching for Climate Downscaling