Climate Downscaling

Climate downscaling estimates local weather and climate patterns from coarse global models or reanalysis, helping assess regional impacts such as extreme heat, heavy precipitation, and infrastructure risk. Approaches range from statistical correction and U-Net-style convolutional networks to continuous-field operators and probabilistic diffusion or flow-matching models, sometimes guided by terrain or physical constraints. Evaluation commonly uses ERA5, regional climate simulations, and station or satellite observations, measuring not only pointwise errors but also extremes, spatial structure, uncertainty, and consistency across variables. A central challenge is that methods that improve average accuracy or fine-scale texture may still misrepresent extremes, dynamics, or future climates, and results can vary across regions and evaluation metrics.