IBM, NASA team up on open-sourced AI model to track the climate crisis

IBM and NASA are using the same technology that powers ChatGPT to potentially track the ravages of the climate crisis.

The two organizations announced last week that they’ve released a version of IBM’s AI foundation model trained on NASA satellite data through an open-source license on AI model hub Hugging Face.

The newly available AI package is a foundation model, a generalized AI architecture based on massive amounts of data that can be specialized through a process called fine-tuning. In this case, IBM has fine-tuned the model on labeled images of burn scars and floods to help predict where these disasters might strike next.

With more fine-tuning, the model can also be used to track deforestation, predict crop yields, or detect greenhouse gases, according to IBM’s announcement.

The geospatial AI model is designed to cut down on the time it takes for researchers to annotate physical features like trees and crops in satellite imagery, IBM said. With the AI model, users can zoom in on an image and immediately see labeled areas that might have been affected by floods or fires, for instance.

The developers claim it can map floods and fires 15% more accurately than other leading deep learning models. The team also contends that it requires much less labeled training data to perform the tasks accurately, thanks to the foundation model structure, which gives the system a basic understanding of the shape of the data on which to build.

“We believe that foundation models have the potential to change the way observational data are analyzed and help us to better understand our planet,” NASA Chief Science Data Officer Kevin Murphy said in a blog post. “And by open-sourcing such models and making them available to the world, we hope to multiply their impact.”

The new model, which had been in the works for about six months, comes on the heels of IBM’s rollout of its WatsonX AI-building enterprise tools in May. IBM said it would make a commercial version of the geospatial model available through its Environmental Intelligence Suite platform later this year.

IBM and NASA are far from the only ones using AI and data in an attempt to help organizations track floods, wildfires, and other weather events: Google released a tool last fall that helps governments and companies predict flooding, and PwC is helping businesses plan for environmental risk using actuarial science.