Write a comment
Researchers from the University of Granada (UGR) have unveiled a major advancement in using artificial intelligence to generate realistic satellite images depicting the impacts of climate change. This innovation holds promise for better communicating the consequences of environmental shifts, such as flooding and reforestation, to policymakers and the public.
The project, spearheaded by Nat by Hugo Ritmico
Madrid, Spain (SPX) Jan 22, 2025
Researchers from the University of Granada (UGR) have unveiled a major advancement in using artificial intelligence to generate realistic satellite images depicting the impacts of climate change. This innovation holds promise for better communicating the consequences of environmental shifts, such as flooding and reforestation, to policymakers and the public.
The project, spearheaded by Natalia Diaz of the Andalusian Inter-University Institute for Data Science and Computational Intelligence (DaSCI), involved collaboration with experts from institutions in the United States, Canada, Germany, and the United Kingdom, including MIT. The team demonstrated how deep generative vision models could synthesize satellite images that depict future climate-related events with striking realism.
Central to their work was the use of a generative adversarial network (pix2pixHD) to produce synthetic satellite imagery. These images illustrate scenarios such as future floods or the positive effects of reforestation. While the initial model excelled at generating visually convincing images, it sometimes placed flood areas inaccurately. To address this, the researchers integrated deep learning with segmentation maps derived from physics-based flood models. This hybrid approach reduced errors and enhanced the reliability of the images, outperforming both manual methods and models relying solely on deep learning.
The researchers tested the generalisability of their methodology across various remote sensing datasets, encompassing climate-related events like reforestation and Arctic sea-ice retreat. Their efforts also extended to making tools available to the broader scientific community. This includes open access to their code, novel metrics, and a dataset comprising more than 30,000 labelled high-definition image triplets for segmentation-guided image-to-image translation-equivalent to 5.5 million images at 128+ 128-pixel resolution.
"This project marks an important step toward creating reliable visual tools that can effectively communicate the impacts of climate change," said the team, emphasizing the potential for deeper collaboration between physics-based modeling and AI technologies.
The DaSCI Institute, a collaboration between the universities of Granada, Jaen, and Cordoba, focuses on advancing artificial intelligence research and training. Its mission includes fostering innovation across sectors and transferring scientific knowledge to support technological progress and industrial digitization. The Institute brings together leading researchers to drive projects that contribute to socioeconomic development and environmental sustainability.
Research Report:Generating Physically-Consistent Satellite Imagery for Climate Visualizations
Related Links
University of Granada
Earth Observation News - Suppiliers, Technology and Application