Downscaling weather forecasts from Low- to High-Resolution with Diffusion Models

2026-03-30

AI Summary

This study addresses the challenge of effectively downscaling global weather forecasts from 100 km to 30 km resolution while preserving fine-scale structures and characteristics of extreme events. It presents the first global probabilistic atmospheric downscaling system based on diffusion models, integrated within the Anemoi framework. The approach employs a conditional diffusion model that learns the residual distribution between high-resolution fields and interpolated low-resolution inputs, incorporating multivariate coupling and spectral consistency constraints. Experiments demonstrate that the system significantly improves probabilistic forecast skill for surface variables—measured by the fair continuous ranked probability score (FCRPS)—accurately reproduces the target power spectrum, maintains physically consistent relationships such as those between wind and pressure, and generates realistic extreme values for events like tropical cyclones that align well with observational ensembles.

Abstract

We introduce a probabilistic diffusion-based method for global atmospheric downscaling implemented within the Anemoi framework. The approach transforms low-resolution ensemble forecasts into high-resolution ensembles by learning the conditional distribution of finer-scale residuals, defined as the difference between the high-resolution fields and the interpolated low-resolution inputs. The system is trained on reforecast pairs from ECMWF IFS, using coarse fields at 100 km to reconstruct fine-scale variability at 30 km resolution. The bulk of the training focuses on recovering small-scale structures, while fine-tuning in high-noise regimes enables the generation of extremes. Evaluation against the medium-range IFS ensemble target shows that the model increases probabilistic skill (FCRPS) for surface variables, reproduces target power spectra at small scales, captures physically consistent multivariate relationships such as wind-pressure coupling, and generates extreme values consistent with those of the target ensemble in tropical cyclones.

Problem

• downscaling
• weather forecasts
• high-resolution
• extreme events
• probabilistic forecasting

Innovation

• diffusion models
• atmospheric downscaling
• probabilistic forecasting
• ensemble refinement
• extreme weather generation