There are many simulations: linked to nuclear wars, tsunamis and even our lives… The objective of all of them is to try to create reaction measures against these disasters. With this in mind, scientists at MIT are developing an artificial intelligence (AI) tool that creates realistic satellite images of possible flood scenarios.
The tool combines a generative AI model with a physics-based flood model to predict areas at risk of flooding and then generate detailed bird’s-eye view images of what the region could look like after the flooddepending on the strength of an approaching storm.
“The idea is that one day we can use this before a hurricane and that it provides an additional layer of visualization for the public – explains Björn Lütjens, from the Department of Earth, Atmospheric and Planetary Sciences at the Massachusetts Institute of Technology (MIT), in a statement -. one of The biggest challenges are encouraging people to evacuate when they are at risk. Maybe this could be another visualization to help increase that readiness.”
To create this tool, Lütjens’ team trained a machine learning model called a conditional generative adversarial network, or GAN, which creates realistic images using two neural networks working against each other.
The first network, called the “generator,” learns by studying real examples, such as satellite images of areas before and after a hurricane. The second network, called the “discriminator”, acts as a critic, trying to distinguish real images from fake ones created by the generator. Together, the two networks improve the results until the generated images look convincingly realistic.
Each network automatically learns and improves based on feedback from the other. This back-and-forth process aims to create images that are almost identical to the real thing. However, GANs sometimes produce “hallucinations”: features in images that appear real, but are factually incorrect or should not be there.
“Hallucinations can deceive viewers – adds Lütjens -. “We were thinking: how can we use these generative AI models in a climate impact context, where reliable data sources are so important?”
To demonstrate the credibility of your model, The researchers applied it to a scenario for Houston, generating satellite images of flooding in the city after a storm comparable in strength to Hurricane Harvey.which had hit the city in 2017. They then compared their AI-generated images with real satellite images, as well as images created without the help of the physics flood model.
The results, published in IEEE Transactions on Geoscience and Remote Sensing, showed that images created without the help of the physics model were highly inaccurate, with numerous “hallucinations”; specifically, some images showed flooding in regions where it would not be physically possible. However, the images from the physics-enhanced method were comparable to the real-world scenario.
Lütjens’ team believes that this technology should be more applicable to predict the outcomes of future flood scenarios, by producing reliable images that help policymakers better prepare and make informed decisions about flood planning, evacuation and mitigation.
In general, the study points out, Policy makers often estimate where flooding may occur based on visualizations in the form of color-coded maps.
“The question is: can satellite image visualizations add another level to this, one that is a little more tangible and emotionally engaging than a color-coded map of reds, yellows and blues, while still being reliable?” Lütjens.
This is an important example of how space-based technology can help manage the developing climate crisis, which is making extreme events, such as floods and hurricanes, more likely.
The team’s method is still in the proof-of-concept stage and It needs more time to study other regions to predict the outcomes of different storms. This will require more training in many more real-world scenarios. Even so, its use has been opened to scientists around the world.
“We are eager to put our generative artificial intelligence tools in the hands of decision makers at the local community level – concludes Dava Newman, director of the MIT Media Lab -, which “It could make a significant difference and perhaps save lives.”