AI accurately visualises future floods

The Earth Intelligence Engine

The visuals could increase hurricane-readiness by showing realistic flood impacts

Spotted: Extreme weather events are intensifying around the world, with hurricanes and floods causing widespread devastation. Flooding alone affects 1.81 billion people, or 23 per cent of the world population, and costs billions in damages. Yet, communicating the urgency of evacuation to communities at risk remains a significant challenge. To address this, MIT scientists have developed a groundbreaking AI tool that generates realistic satellite images showing what a region could look like after a major flood event, helping residents and policymakers better visualise risks and prepare for disasters.

The innovation, dubbed the Earth Intelligence Engine, uses a hybrid approach: it combines generative artificial intelligence with a physics-based flood model to produce satellite-like images of post-storm flooding. Traditional flood predictions rely on colour-coded maps – valuable but often abstract representations of risk. In contrast, this tool creates intuitive, bird’s-eye visuals that resemble real satellite images, offering a tangible, emotionally resonant perspective of flood impacts.

Here’s how it works: the AI system, powered by a type of machine learning called a generative adversarial network (GAN), is first trained on satellite images captured before and after hurricanes. However, AI alone can sometimes produce hallucinations – images that appear realistic but show physically impossible flooding in higher-elevation areas. To counter this, MIT researchers integrated a physics-based flood model that accounts for real-world data like hurricane trajectories, wind patterns, and local flood infrastructure. This combination significantly improves accuracy, producing trustworthy visuals of where flooding is likely to occur.

“We show a tangible way to combine machine learning with physics for a use case that’s risk-sensitive,” explains Dava Newman, MIT professor and co-author of the study. “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.”

The Earth Intelligence Engine is currently a proof-of-concept, tested on Houston with Hurricane Harvey as a case study. Researchers found the physics-reinforced images aligned pixel-for-pixel with real-world flood data, outperforming AI-only models. While promising, scaling the tool to other regions will require additional training with diverse satellite imagery.

Looking ahead, the team envisions the tool being used as a pre-hurricane preparedness resource, offering intuitive flood visualisations to both decision-makers and the public. As lead researcher Björn Lütjens explains, in this case “generative AI is tasked with creating satellite images of future flooding that could be trustworthy enough to inform decisions of how to prepare and potentially evacuate people out of harm’s way.”

Written By: Oscar Williams