High-resolution geospatial data are essential for capturing fine-scale climate processes, but generating them through physically based models is computationally intensive, especially at continental and global scale. Standardising climate indicies further increases this burden and limits their practical use in regions with complex terrain and strong microclimatic variability, such as the Alps. To address these challenges, we propose 3DcGanAE, a 3D generative deep learning framework that reconstructs high-resolution (HR) climate fields from low-resolution (LR) inputs by learning residual structures between the two scales. The model incorporates Distributed Gaussian Noise (DGN) to generate realistic residual fields and uses a denoising autoencoder to reduce Gaussian artefacts and systematic biases. A Bayesian Gaussian Mixture Model (BGMM) is used to estimate scale‑adaptive noise parameters and preserve statistical consistency. HR fields are decomposed into LR feature components, enabling the model to reconstruct fine-scale HR fields while maintianing spatial coherence. The framework is evaluated across the Alpine region for drought related indicies, including the Standardised Precipitation–Evapotranspiration Index (SPEI) and Standardised Soil Moisture Index (SSI) across the Alps. Results demonstrates that the model reliably generates 1 km drought datasets from LR inputs (5–25 km), outperforming state-of-the-art spatiotemporal GANs and multitask 3D convolutional models. It achieves adjusted R2 values of 0.98—0.92 and Kolmogorov–Smirnov statistics below 0.21, effectively capturing seasonal extremes, maintaining statistical fidelity, and recovering fine‑scale spatial variability. These findings highlight the potential of the proposed framework for scalable high‑resolution climate applications across large and heterogeneous domains.
Aieb et al. (Wed,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: