IBM and NASA advance climate research with AI

IBM and NASA advance climate research with AI

IBM’s enduring legacy in space exploration and cutting-edge technological innovations position the company as a leader in the federal space community. From supporting early rocket programs to developing advanced AI models for climate research, IBM has consistently advanced space science and technology.

IBM’s involvement with the U.S. space program began in 1949, nearly a decade before NASA was established. The company’s Federal Systems Division collaborated closely with the National Advisory Committee for Aeronautics (NACA), the precursor to NASA. This partnership laid the groundwork for subsequent space missions, including the historic Apollo program. IBM provided essential programming capabilities and technical know-how throughout the 1950s, supporting initiatives like the Redstone Arsenal program developing early rocket technology.

IBM’s contributions to space exploration are not just historical footnotes. On July 20, 1969, NASA’s Apollo program marked a milestone by landing astronauts on the moon, thanks in part to the efforts of 4,000 IBM employees who built computers and developed the software for launching and guiding Apollo missions safely back to Earth.

Over half a century later, IBM continues to be at the forefront of space exploration by leveraging its expertise in AI, data, cybersecurity and quantum computing. In 2019, NASA’s Science Mission Directorate identified AI and machine learning as critical to advancing space science. IBM joined forces with NASA to explore these technologies’ potential and formed a working group to integrate AI and machine learning across the federal agency’s various science divisions. Since then, they have worked together to generate new innovations in Earth science that help combat global challenges like a changing climate.

Addressing global climate challenges with AI

NASA’s most recent collaboration with IBM was driven by the numerous risks posed by a changing climate and the need to better understand how Earth’s landscape is quickly evolving, said Steven Wysmuller, federal AI and sustainability lead at IBM Research.

In late 2022, IBM and NASA signed the Space Act Agreement, outlining 22 initiatives aimed at harnessing IBM’s technological capabilities to support NASA’s missions. One of the first projects under this agreement was developing an AI model to speed up the analysis of satellite images and boost discoveries in climate and weather. The aim was to create an adaptable, reusable foundation model that provides a more efficient way for researchers to analyze and draw insights from nearly 250,000 terabytes of NASA datasets related to Earth’s climate processes and drivers.

Leveraging NASA’s Harmonized Landsat and Sentinel-2 (HLS) dataset, IBM utilized its Vela supercomputer, equipped with 5,000 GPUs, to train an AI model capable of analyzing global weather patterns and climate data, including tracking land use changes, monitoring natural disasters and predicting crop yields. 

The HLS Geospatial Foundation Model (HLS Geospatial FM) can analyze geospatial data up to four times faster than state-of-the-art deep learning models, with half as much labeled data, according to Wysmuller. By leveraging AI, NASA and IBM can model complex climate processes and quickly develop actionable insights to address.

For example, a 2021 NASA study found that about a quarter of the global population resides in flood-prone areas, a figure expected to increase due to rising sea levels and intensified storms. Accurately mapping floods is crucial for both immediate protection and guiding future development away from high-risk zones. The HLS Geospatial FM uses meteorological data and satellite imagery to predict flood likelihood based on weather patterns and river levels. Post-flood, the model assesses damage by analyzing satellite images, aiding disaster response teams and communities in recovery efforts.

“We’re talking to various agencies that are looking at how floods impact agriculture in a particular state or region and how that is impacting overall price necessities like corn or wheat,” Wysmuller said. “But it’s not just in the U.S. What happens 7,000 miles away impacts us here, so globally, we can gain a clearer picture.”

The AI model showed a 15% improvement in accuracy over state-of-the-art deep learning models, highlighting AI’s potential in processing large datasets and paving the way for future advancements in space and climate research.

IBM’s commitment to open-source AI enabled collaboration

IBM is committed to making AI accessible, which is why the HLS Geospatial FM is available on open-source AI platform Hugging Face, according to Wysmuller. Foundation models are highly versatile, and by sharing the model and its code, IBM encourages researchers, developers and organizations worldwide to leverage its capabilities to help address a changing climate. 

For example, large consumer goods companies could use the HLS Geospatial FM to understand macro trends like severe weather to inform raw material sourcing decisions. Agribusinesses could employ this technology to measure, track and mitigate the environmental impact of their farming practices, such as monitoring soil degradation, conserving water and reducing pollution from runoff into local waterways. 

IBM is collaborating with local governments in Africa to monitor deforestation and promote reforestation efforts. By analyzing satellite imagery, the model can help measure the effectiveness of interventions, such as forest fencing, and provide actionable insights to policymakers.

The HLS Geospatial FM can also identify and track wildfires in real time by analyzing satellite images and thermal data, aiding early detection and continuous monitoring of fire spread. By studying historical fire patterns and environmental conditions, the model can help predict potential wildfire outbreaks, enabling proactive measures to prevent or mitigate damage.

In the U.S., utility providers are exploring the HLS Geospatial FM to improve grid resilience. By coupling weather predictions with power grid data, these models help utility providers optimize energy distribution and mitigate adverse weather impacts. This application is particularly relevant for renewable energy sources, like solar and wind, where accurate weather forecasts are crucial for grid stability.

IBM and NASA’s decision to share their geospatial AI foundational model on Hugging Face promotes transparency, accessibility and collaboration in addressing global environmental challenges. Sharing the model encourages community-driven improvements and ensures the latest AI advancements are accessible to all, ultimately contributing to a more resilient and sustainable future.

“Being able to mitigate and balance a changing climate is really important to us,” said Wysmuller. “And making this information available to many so they can protect property and save lives is key.”

Learn more about how IBM is supporting missions in space. 



Read the full article here