Statistical Downscaling

Statistical downscaling aims to increase the spatial resolution of climate and weather data from coarser models, improving accuracy for regional and local applications. Recent research heavily utilizes deep learning, employing architectures like U-Nets, GANs, diffusion models, and normalizing flows to achieve this upscaling, often focusing on improving accuracy, uncertainty quantification, and handling diverse data types (e.g., wind, precipitation, sea surface height). This work is crucial for enhancing climate change impact assessments, improving weather forecasting at finer scales, and supporting applications in renewable energy resource management and hazard mitigation. The field is also exploring methods to improve the transferability and efficiency of these models across different regions and variables.