Massachusetts Institute of Technology researchers have developed a groundbreaking AI tool that uses satellite imagery to predict the impact of future flooding. By combining generative AI with a physics-based flood model, the tool creates realistic aerial images showing how areas may be affected by flooding from upcoming storms.
Massachusetts Institute of Technology tested the technology on Houston, recreating the city after Hurricane Harvey in 2017. The generated images were compared to actual satellite images and to those made without the physics-based model. The physics-enhanced images proved more accurate, as AI-only images showed incorrect flooding in areas where it wasn’t possible.
The method, called the "Earth Intelligence Engine," aims to enhance public preparedness by providing better flood visuals ahead of hurricanes. "The idea is: One day, we could use this before a hurricane, where it provides an additional visualization layer for the public," said Björn Lütjens, a postdoc at MIT's Department of Earth, Atmospheric, and Planetary Sciences, who led the research. Professor Dava Newman, director of the MIT Media Lab, emphasized the value of hyperlocal perspectives in climate communication, as per the MIT News report.
The project builds on MIT's previous work using generative AI to visualize future climate scenarios. It uses a conditional generative adversarial network (GAN), where two neural networks work together to create synthetic images that closely resemble real satellite photos. The technology’s accuracy is vital, as incorrect images, or "hallucinations," can mislead viewers in critical situations. "We were thinking: How can we use these generative AI models in a climate-impact setting, where having trusted data sources is so important?" explained Lütjens, the researcher behind combining AI with physical models in the same report.
The researchers have made their "Earth Intelligence Engine" available online for testing and further development. Their work, detailed in the journal IEEE Transactions on Geoscience and Remote Sensing, includes contributions from MIT’s Lütjens, Brandon Leshchinskiy, Aruna Sankaranarayanan, and other collaborators, as mentioned in the MIT News report.
