Prithvi-EO-2.0 : IBM and NASA launch new geospatial AI foundation model for better satellite data analysis

IBM and NASA have launched Prithvi-EO-2.0, their groundbreaking geospatial AI foundation model, marking a major milestone in environmental mapping and satellite data analysis. Developed in collaboration with Germany’s Jülich Supercomputing Centre, the open-source model boasts improved precision in tracking critical Earth processes, from deforestation to urban heat islands.

A new era for geospatial AI

Prithvi-EO-2.0, with its 600 million parameters, is six times larger than its predecessor, released in August 2023. This advanced model leverages a more geographically comprehensive dataset and sophisticated embeddings to capture intricate temporal and spatial relationships in data. IBM researchers noted its enhanced ability to map long-term environmental changes and assimilate high-resolution information, surpassing previous models by 8% in accuracy on the GEO-bench framework.

The model’s training utilized over 4.2 million data points from NASA’s Harmonized Landsat and Sentinel (HLS) dataset, offering detailed analysis capabilities. Prithvi-EO-2.0 is accessible via Hugging Face and IBM’s TerraTorch toolkit, forming part of NASA’s initiative to democratize satellite data through open technologies.

Addressing global challenges with Prithvi-EO-2.0

The potential applications of Prithvi-EO-2.0 extend across diverse global challenges, from tracking deforestation to responding to extreme weather events and monitoring urban heat. IBM and NASA have already demonstrated its capabilities in three key scenarios:

Amazon deforestation monitoring: By fine-tuning Prithvi-EO-2.0 for canopy height estimation, IBM provided precise visuals of deforestation trends in Bolivia’s Amazon. This analysis supports efforts to verify carbon-credit systems aimed at protecting rainforests, vital for carbon storage and biodiversity.

Valencia flood mapping: The model was used to analyze the catastrophic October 2024 floods in Spain’s Valencia region. Integrating radar and optical satellite data, Prithvi-EO-2.0 produced a detailed map of the 120-square-kilometer flood extent with 30% greater accuracy than traditional methods.

Urban heat island detection in Baltimore: Prithvi-EO-2.0 tracked extreme heat waves in Baltimore, Maryland, using hourly land surface temperature maps. The data highlighted urban heat islands, supporting targeted mitigation measures like tree planting to combat rising temperatures.

Unlocking potential through collaboration

IBM attributes the success of Prithvi-EO-2.0 to its partnership with NASA and Jülich Supercomputing Centre. These collaborations were instrumental in model design, data collection, and performance assessment. Experts also introduced new applications, such as estimating carbon flux and detecting landslides, to expand the model’s capabilities.

Juan Bernabé-Moreno, IBM Research Europe director, emphasized that this spirit of open collaboration serves as a blueprint for developing future tools to tackle large datasets and complex environmental challenges.

Driving climate resilience and sustainability

As climate change accelerates, tools like Prithvi-EO-2.0 play an increasingly critical role in monitoring environmental changes and mitigating disasters. With its ability to process large-scale satellite imagery into actionable insights, this geospatial AI model represents a significant leap forward in understanding and addressing the world’s most pressing environmental issues.