Can AI be enabled to dynamical downscaling? Training Deep Generative Models to downscale ERA5 to high-resolution COSMO-CLM dynamical reanalyses over Italy

Downscaling techniques are one of the most prominent applications of Deep Learning (DL) in Earth System Modeling. A robust DL downscaling model can generate high-resolution fields from coarse-scale numerical model simulations, saving the timely and resourceful applications of regional/local models. Moreover, specific DL models can generate uncertainty information and provide ensemble-like pool scenarios, hardly achievable using traditional numerical simulations due to their high computational requirements. In this work, we present the application of deep generative models, namely a Generative Adversarial Network (GAN) and a Latent Diffusion model (LDCast, Leinonen et al., 2023), to perform the downscaling of ERA5 (Hersbach et al., 2018) data over Italy up to a resolution of 2km. The target high-resolution data used for training consists in the Italian high-resolution dynamical reanalyses obtained with COSMO-CLM (Raffa et al., 2021). The goal of the study is to show that recent advancements in generative modeling can learn to provide comparable results with numerical dynamical downscaling models, such as the COSMO-CLM model, given the same input data (i.e., ERA5 data), preserving the realism of fine-scale features and flow characteristics. The training and testing database is composed of hourly data from 2000 to 2020 (~184000 timestamps), and the target variables of the study are 2-m temperature and horizontal wind components. A selection of predictand variables from ERA5 is used as input to the DL models (e.g., 850hPa temperature, specific humidity, and wind). The generative models are compared with reference baselines, both DL-based (e.g., UNET) and statistical methods. Preliminary results are presented, highlighting the improvements introduced with this architecture with respect to the baselines. The results are evaluated by different quantitative verification scores: RMSE, predicted spectra, frequency distributions, and spatial distribution of errors.