IBM and NASA have just launched a more advanced version of their geospatial model, Prithvi-EO-2.0, aimed at transforming satellite data into clearer images of Earth’s ever-changing landscape. This new version, with a staggering 600 million parameters, is six times the size of its predecessor, which was the largest model of its kind when it was released just last year. It’s been pre-trained with a more extensive dataset, allowing it to capture not only spatial but also temporal relationships, enhancing its ability to analyze long-term processes like seasonal changes.
Prithvi-EO-2.0 offers significant improvements over its predecessor, particularly in zooming in on detailed landscape features. This model can now identify specific elements, such as crops, trees, livestock, and even solar panels in satellite and drone data. It’s designed to help us better understand and monitor environmental changes, from deforestation in the Amazon to heat waves in urban cities like Baltimore. The model’s performance has been tested and proven on NASA’s extensive Harmonized Landsat and Sentinel dataset, scoring an impressive 75.6% on the GEO-bench framework, marking an 8% improvement over the previous version.
The collaborative effort between IBM and NASA has also led to the development of new applications, including carbon flux estimation and landslide detection. Experts from both organizations worked together to design the model and its training data, ensuring that it could handle complex climate-related tasks. Their combined efforts are helping to make satellite data more accessible and effective in addressing global challenges like climate change and environmental monitoring.
For instance, IBM used the model to map deforestation in Bolivia’s Amazon, showing how vast stretches of land have been transformed from rich, green canopies into urbanized areas. In Spain, during the catastrophic flash floods of 2024, Prithvi-EO-2.0 was able to accurately assess the flood’s extent, combining data from multiple satellite sources to create a more detailed map of the disaster. Similarly, in Baltimore, the model helped track dangerous heat islands exacerbated by urbanization, providing crucial information for future climate resilience efforts.
This innovation is not just about improving technology; it’s about how that technology can be applied to real-world problems. With its enhanced ability to analyze both spatial and temporal data, Prithvi-EO-2.0 is setting a new standard in geospatial AI, offering powerful tools to better understand and respond to our planet’s evolving challenges. The hope is that this technology will drive future collaborations and innovations aimed at unlocking large datasets for environmental and climate action.