Km-scale dynamical downscaling through conformalized latent diffusion models

TL;DR

This work addresses the challenge of reliable uncertainty in km-scale meteorological downscaling by combining residual-corrective latent diffusion with conformal prediction. It formulates the downscaling task as learning the conditional distribution $\,\mathbb{P}(X|Y)$ and uses a residual diffusion in latent space to generate diverse high-resolution realizations, which are then calibrated via asymmetric, grid-wise conformalized quantile regression. The method yields grid-point-level prediction intervals with improved coverage and stable probabilistic scores compared to a diffusion baseline, demonstrating more trustworthy probabilistic downscaling of ERA5-driven fields to approximately 2 km. The approach has practical implications for weather forecasting and energy meteorology by providing calibrated, location-specific uncertainty essential for decision-making under uncertainty, while also highlighting avenues for extending conformal methods to multivariate and temporally aware settings.

Abstract

Dynamical downscaling is crucial for deriving high-resolution meteorological fields from coarse-scale simulations, enabling detailed analysis for critical applications such as weather forecasting and renewable energy modeling. Generative Diffusion models (DMs) have recently emerged as powerful data-driven tools for this task, offering reconstruction fidelity and more scalable sampling supporting uncertainty quantification. However, DMs lack finite-sample guarantees against overconfident predictions, resulting in miscalibrated grid-point-level uncertainty estimates hindering their reliability in operational contexts. In this work, we tackle this issue by augmenting the downscaling pipeline with a conformal prediction framework. Specifically, the DM's samples are post-processed to derive conditional quantile estimates, incorporated into a conformalized quantile regression procedure targeting locally adaptive prediction intervals with finite-sample marginal validity. The proposed approach is evaluated on ERA5 reanalysis data over Italy, downscaled to a 2-km grid. Results demonstrate grid-point-level uncertainty estimates with markedly improved coverage and stable probabilistic scores relative to the DM baseline, highlighting the potential of conformalized generative models for more trustworthy probabilistic downscaling to high-resolution meteorological fields.

Figures (5)

• Figure 1: Grid-wise PICP and PI width for coverage level 1-$\alpha$=0.9. (l)- 2mT, (r)-WS
• Figure 2: 2mT average PICP and % Deviation
• Figure 3: WS average PICP and % Deviation
• Figure 4: Samples of predictions from 2mT test set
• Figure 5: Samples of predictions from WS test set