Why trust Greenbot
Massachusetts Institute of Technology (MIT) researchers have developed a groundbreaking tool called “Earth Intelligence Engine” to predict and visualize future floods by combining generative artificial intelligence (AI) and physics-based modeling. The study aims to help communities prepare for local and upcoming natural disasters. The research was published in IEEE Transactions on Geoscience and Remote Sensing journal this November 2024.
“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 and the research lead. “One of the biggest challenges is encouraging people to evacuate when they are at risk. Maybe this could be another visualization to help increase that readiness.”
Overcoming AI Hallucinations with Physics Integration
Initially, they tested the model using generative AI alone. The tool was able to create realistic satellite images but had pitfalls in hallucinations (made-up facts by AI), sometimes predicting flooding in physically impossible areas, as modern models are prone to do.
“Hallucinations can mislead viewers,” said Lütjens. “We were thinking: How can we use these generative AI models in a climate-impact setting, where having trusted data sources is so important?”
So they combined it with a physics-based flood model that incorporates real, physical parameters like hurricane trajectories, storm surges, and existing flood infrastructure. This significantly enhanced the model’s reliability.
The Earth Intelligence Engine was tested using data from Hurricane Harvey. The generated images of Houston matched the actual flood extent with remarkable accuracy, providing a pixel-by-pixel visualization of the flooding as predicted by the physics model. These images could offer a more tangible way for residents to understand their risk compared to traditional color-coded flood maps. According to Lütjens, realistic satellite imagery might be more “emotionally engaging” and easier for the general public to understand than standard flood predictions.
AI’s Potential in Reducing Climate Change Challenges
Dava Newman, a co-author of the study and director of the MIT Media Lab, highlighted the tool’s potential impact on local communities.
“We show a tangible way to combine machine learning with physics for a use case that’s risk-sensitive, which requires us to analyze the complexity of Earth’s systems and project future actions and possible scenarios to keep people out of harm’s way,” she explained. “We can’t wait to get our generative AI tools into the hands of decision-makers at the local community level, which could make a significant difference and perhaps save lives.”
One can argue that environmental science is the best field for AI technology to mature. Many of its applications demonstrate its potential to address climate challenges.
For instance, Google’s DeepMind has used AI to predict wind patterns to optimize wind energy production. Similarly, Microsoft’s AI for Earth is focused on developing tools for managing water, agricultural practices, and conserving biodiversity. These efforts show AI can take part in empowering understanding, preparation, and mitigation of climate change.
The Earth Intelligence Engine is available online for public use. The researchers noted that additional training may still be required to be applicable to other regions. MIT’s research received support from multiple organizations, including NASA, Google Cloud, and the MIT Portugal Program.
