NASA, IBM team up to build LLMs that can help fight climate change

IBM on Thursday said it has partnered with the US space agency NASA to co-develop a foundation large language model based on geospatial data that it claims will help scientists and their organizations fight climate change.

The open source model, which will be available on Hugging Face, was developed on IBM’s watsonx.ai platform and trained on Harmonized Landsat Sentinel-2 satellite data (HLS) over one year across the continental US before being fine-tuned on labelled data for flood and burn scar mapping — a scientific process to map large environmental fire incidents, the company said.  

While testing the accuracy of the model, researchers at IBM saw a 15% improvement in precision compared to existing learning models for mapping floods and burn scars from fires, using half as much labelled data.

This improvement, according to the company, could speed up geospatial analysis by three to four times, and help reduce the amount of data cleaning and labelling required in training a traditional deep learning model.

“With additional fine-tuning, the base model can be redeployed for tasks like tracking deforestation, predicting crop yields, or detecting and monitoring greenhouse gasses,” IBM said in a statement.

The release of the model, according to both IBM and NASA, assumes significance because access to the latest geospatial data and analyzing them remains a significant challenge in climate science despite large amounts of data being added regularly.

The model can act as the base to analyze datasets for advancing applications of AI in combating climate change. A commercial version of the geospatial model, which is part of IBM watsonx, will be made available through the IBM Environmental Intelligence Suite later this year, the company said.