The Allen Institute for AI, founded by Microsoft co-founder Paul Allen, has launched Satlas, a groundbreaking tool that combines generative AI with satellite imagery from the European Space Agency’s Sentinel-2 satellites.
Satlas focuses on renewable energy projects and tree cover worldwide, offering monthly updates for most parts of the planet, excluding parts of Antarctica and remote ocean areas.
The tool utilizes a feature called “Super-Resolution,” employing deep learning models to enhance the clarity of satellite images, including generating high-resolution images of buildings and landscapes. This technology allows users to identify solar farms, onshore and offshore wind turbines, and track changes in tree canopy coverage over time.
Satlas is a valuable resource for policymakers working to address climate and environmental goals, as it provides comprehensive and publicly accessible data on renewable energy projects and tree cover.
While Satlas is a pioneering use of super-resolution in a global map, it is not without its limitations. The AI model may occasionally exhibit inaccuracies, referred to as “hallucination,” which can lead to misrepresentations of buildings and objects. These discrepancies may arise from regional architectural differences or the model’s attempt to predict the placement of objects based on training data.
Developing Satlas required extensive manual labeling of satellite images to identify wind turbines, offshore platforms, solar farms, and tree cover percentages. Deep learning models were trained on this labeled data to recognize these features automatically. For super-resolution, the models were fed numerous low-resolution images of the same locations taken at different times, allowing them to predict sub-pixel details in the high-resolution images they generate.
The Allen Institute plans to expand Satlas to provide other types of maps, such as identifying global crop types. The overarching goal is to create a foundational model for monitoring the planet’s changes, enabling scientists to study climate change and other Earth phenomena more effectively.