Wind speed super-resolution and validation: from ERA5 to CERRA via diffusion models

TL;DR

This paper tackles the delayed availability of high-resolution reanalysis data by proposing a diffusion-model–based wind-speed downscaling from ERA5 to CERRA for the Italian region. By conditioning a DDIM-based diffusion model on low-resolution ERA5 frames and training on existing CERRA data, the authors generate high-resolution wind fields that closely match ground truth and in-situ observations, while avoiding additional data requirements typical of physics-based downscaling. The Ensemble Diffusion approach yields the best quantitative gains, nearing CERRA performance and outperforming bilinear baselines, with validated improvements against IGRA V2 measurements. This data-driven method offers timely, resource-efficient downscaled wind fields that could complement or substitute traditional downscaling during data lags, and highlights a path toward neural reanalyses for broader meteorological variables and regions.

Abstract

The Copernicus Regional Reanalysis for Europe, CERRA, is a high-resolution regional reanalysis dataset for the European domain. In recent years it has shown significant utility across various climate-related tasks, ranging from forecasting and climate change research to renewable energy prediction, resource management, air quality risk assessment, and the forecasting of rare events, among others. Unfortunately, the availability of CERRA is lagging two years behind the current date, due to constraints in acquiring the requisite external data and the intensive computational demands inherent in its generation. As a solution, this paper introduces a novel method using diffusion models to approximate CERRA downscaling in a data-driven manner, without additional informations. By leveraging the lower resolution ERA5 dataset, which provides boundary conditions for CERRA, we approach this as a super-resolution task. Focusing on wind speed around Italy, our model, trained on existing CERRA data, shows promising results, closely mirroring original CERRA data. Validation with in-situ observations further confirms the model's accuracy in approximating ground measurements.

Figures (11)

• Figure 1: Comparison of wind speed over Italy between ERA5 and CERRA
• Figure 2: Three different stages of reanalysis: (left) the Global Reanalysis ERA5 will be used as boundary condition, (middle) CERRA Regional Reanalysis, and (right) a CERRA 2D Reanalysis for the near surface. The arrows indicate that the amount of observational information used for the reanalysis progressively increases from the global to the regional reanalysis. Source: Copernicus-ECMWF CERRA Presentation, 2024 presentationCerra
• Figure 3: Selected area for our study, delimited by the following geographical coordinates: North at 47.75°, South at 35°, East at 18.75°, and West at 6°.
• Figure 4: The downscaling diffusion process utilize as input a set of low resolution images. For predicting the high resolution image at time $t_0$ the conditioning information include the low resolution images at times $t_{-6}$ , $t_{-3}$, $t_0$, $t_{+3}$ where each step is 3 hours long.
• Figure 5: The super-resolution diffusion framework employs a sequence-to-one strategy, where pre-upsampling is achieved via bilinear interpolation. The diffusion process is comprised of five diffusion steps, using a denoising U-net to progressively denoise the input. A noise scheduler orchestrates this process. This procedure is replicated 15 times, generating an ensemble of potential super-resolution outcomes. These outcomes are then amalgamated using a mean operation to produce the final output.