ESMs Latent Space Exploration for Uncertainty Quantification and Spatiotemporal Downscaling

Summary

Climate impact assessments are hampered by the increasing complexity and computational demands of large Earth System Model (ESM) ensembles and the need for high-resolution regional data. This project developed advanced AI-driven methods, including novel GCM selection workflows and Visual Transformer-based deep learning models (ViSIR and ViFOR), to manage model uncertainty and perform spatiotemporal downscaling. The project, now completed, demonstrated impressive performance in both selecting representative climate model subsets and reconstructing high-resolution climate data.

Key Facts

Source Document

https://example-government.gov/policy-document-link