FAQ

What is Satlas?

Satlas is a platform for visualizing and downloading global geospatial data products generated by AI using satellite images. Currently, it includes three types of data: marine infrastructure (offshore wind turbines and platforms), renewable energy infrastructure (onshore wind turbines and solar farms), and tree cover, but we hope to include many more over time. It is our hope that this data will be useful for earth and environmental scientists, as well as for organizations working in a variety of geospatial domains. The marine infrastructure data is already being used by Ai2’s Skylight to detect vessels involved in illegal fishing.

Can I download the data?

All of the data displayed in Satlas, along with training data and model weights, can be downloaded for offline analysis here. Data is available as monthly maps starting from January 2016. Marine infrastructure and renewable energy infrastructure are published as one GeoJSON for each month, and tree cover is published as a set of GeoTIFFs for each month.

How accurate is the data?

The data generated by our AI models has high accuracy. However, AI systems are never perfect and several factors tend to degrade performance. An analysis of the accuracy of each Satlas geospatial data product is made available in our Data Validation Report.

Why are satellite pictures missing for my time and area?

The satellite imagery in Satlas comes from the Sentinel-2 satellites, which do not cover the whole planet on each pass. If satellite imagery is missing for your time and location, try choosing an adjacent month to get a picture from another orbit.

Why is it hard to annotate features in satellite imagery?

The Sentinel 2 satellite imagery is fantastic in that it has frequent (weekly) global coverage (although there are missing components on each pass). However, compared to commercial satellite feeds it is comparatively low resolution (10m/px), meaning even a large off-shore wind turbine only occupies ten or so pixels in the image. In addition, changing climatic conditions through the year and erratic cloud cover cause noise that the model must ignore.

Are there other sources of global geospatial data?

Other websites include:

• Global Forest Watch collects various forest-related datasets including tree cover, tree gain, and tree loss. Satlas also includes tree cover data: our hope is that this data will be complementary to Global Forest Watch, with high frequency (monthly) updates, especially as we work to improve the quality of our models over time.
• OpenStreetMap is an open map dataset that includes offshore infrastructure features. We’ve observed many objects identified by our models that are missing from OpenStreetMap, and believe that these detections may be helpful for improving OpenStreetMap.
• Earthrise Media maintains several websites highlighting data relevant to environmental issues. Global Plastic Watch shows plastic pollution sites, and Amazon Mining Watch tracks mining activities in the Amazon rainforest.
• ArcGIS and Google Earth Engine data catalogs publish numerous datasets developed by various organizations, including Global Power Plant Database and World Port Index. Unfortunately many of these datasets are regional or no longer maintained. We anticipate Satlas Explorer being a more unified hub for global, frequently updated geospatial data.

Is there code I can download?

Yes! Model parameters and code are available on Github.