Technology giants IBM and NASA have announced a new collaboration to develop an AI foundation model aimed at significantly improving the accuracy and efficiency of weather forecasting and climate applications.
The new model combines IBM's extensive AI capabilities with NASA's vast trove of Earth science data and expertise. It is designed to ingest diverse meteorological data sets and run complex simulations faster than existing forecasting technology, while also enabling new climate insights.
While AI is already being applied in weather prediction models like NVIDIA's Fourcastnet and Google DeepMind's GraphCast, IBM points out that those are narrowly focused AI-based "emulators," not versatile foundation models.
Those AI emulators produce rapid weather forecasts but are limited by their training data and lack the fundamental understanding of weather physics. The IBM-NASA foundation model has several key advantages:
- Analyzes more diverse data types and sources
- Much faster at data ingestion and inference
- More easily adapted to new climate modeling needs
- Boosts accuracy for existing forecast uses
Potential applications include inferring high-resolution insights from low-fidelity data, forecasting extreme weather, predicting disasters like wildfires, and more.
This isn't IBM and NASA's first foray into climate-focused foundation models. Their geospatial AI model debuted earlier this year to help analysts extract insights about climate change impacts from satellite imagery data.
Chris McKay
At the upcoming COP28 climate conference, IBM will highlight how that model has been fine-tuned for various applications. In Kenya, IBM's model assists in visualizing tree planting, addressing water scarcity, and combating deforestation. In the UAE, in partnership with MBZUAI, it's used to study urban heat islands, helping to understand and mitigate their formation.
"AI foundation models using geospatial data can accelerate discovery of climate solutions," Curioni said. "We hope these technologies help address planetary health threats faster than previously possible."
The new weather and climate model will similarly synthesize multi-petabyte data sets using IBM's AI platform and NASA's simulation capabilities. Over time, it could provide invaluable guidance on adaptation strategies as global warming intensifies.
With lives and ecosystems increasingly vulnerable to extreme weather, IBM and NASA's model may prove to be a vital tool for officials, businesses, and communities on the front lines of the climate crisis.