A1123
Title: Generative adversarial models for extreme geospatial downscaling
Authors: Guofeng Cao - University of Colorado Boulder (United States) [presenting]
Abstract: Addressing the challenges of climate change requires accurate and high-resolution mapping of climate variables. However, many existing datasets, such as the outputs of the state-of-the-art numerical climate models, are only available at very coarse resolutions due to model complexity and high computational demand. Deep learning methods, particularly generative adversarial networks (GANs), have proved effective in improving geospatial datasets. A conditional GAN-based extreme geospatial downscaling method is described. Compared to most existing approaches, the method can generate highly accurate datasets from very low-resolution inputs. More importantly, the method considers the uncertainty inherent to the downscaling process that tends to be ignored in existing methods. Given an input, the method can produce a multitude of plausible high-resolution samples. These samples allow for an exploration and inferences of model uncertainty and robustness. With a case study of gridded climate datasets, we showcase the performances of the framework in downscaling tasks with large scaling factors (up to 64) and highlight the advantages of the framework with a comparison with commonly used downscaling methods, including area-to-point kriging, deep image prior, enhanced super-resolution generative adversarial networks (ESRGAN), physics-informed resolution-enhancing GAN (PhIRE GAN), and an efficient diffusion model for remote sensing image super-resolution (EDiffSR).
