Earth Index: A step-by-step guide to searching satellite imagery with AI

In November, I sat in a room in Kuala Lumpur and watched a journalist from Mongabay explain how they identified dozens of illegal narco-trafficking airstrips in Peru.

The team used a tool called Earth Index to label a map with an initial set of 10 airstrips they’d found. Earth Index took the samples and scoured surrounding areas for additional airstrips, returning a set of possible matches within seconds. It provided the reporters with a list of leads to verify with additional reporting and analysis.

In my round-up of tools and insights from the conference, I called Earth Index “the coolest new tool I saw.” The only downside was that at the time it was invite-only. Well, good news: Earth Index is now publicly available. Anyone can use it for free (with some limitations).

It’s a really useful tool for investigations that involve identifying specific landscapes or structures on earth — factories, mines, roads, structures, airstrips and other large topographic features — using satellite imagery. Earth Index can take a few labeled samples of the structure or feature you’re trying to find (the larger and the more unique, the better) and search for it using an AI model trained satellite imagery of the entire earth’s surface. 

Earth Index puts the technical and time-consuming process of geospatial machine learning in the hands of people without advanced technical skills, or the need to gather a large corpus of data. That’s pretty exciting.

Along with the Peru airstrip investigation, the tool has also been used to look into illegal logging in Albania, palm oil production in Brazil, and mining activity in the Amazon.

Here’s a look at what you can do with Earth Index, a step by step guide to setting up your first project, and some insights and tips from my interview with Ed Boyda, a data scientist with Earth Genome, the nonprofit behind the tool.