.png)
Team: From Above — November 30th 2021
This is the story of how our team of journalists and data enthusiasts - a collaboration that emerged from the 2021 JournalismAI Collab Challenges - went about using artificial intelligence to identify visual indicators within satellite imagery to chase a story. We thought our tool could be used to identify illegal airstrips built in a remote area or the expansion of deforestation in a jungle or even being able to prove if a public road is indeed being built.
A visual indication of a specific change in the terrain or spotting peculiar infrastructure from above could turn into an investigative story. If you are doing an investigation and are trying to figure out if satellite images could be fundamental to verify findings or gather new evidence and you have no idea where to start, this guide is made for you.
Our team (with members from Bloomberg News, CLIP, DataCritica, and La Nación) was keen on exploring the usage of satellite imagery along with applied computational techniques for writing stories. We knew that satellite imagery possessed information that could potentially enhance our ability to write compelling narratives about the state of our planet cutting across multiple beats. Such a tool tends to be complex and out of the reach for many journalists. So we wanted to create a replicable workflow that could be used for story ideas/projects.
Infographic: 1-Page guide with steps at a high-level
Our common interest was to look at climate crisis through this lens, so we started with the first and vital step of any story: ideas. We gathered research and pitched multiple story ideas within our group that could be extrapolated using imagery of forests, coastal lines, etc and narrowed down on trying to detect illegal cattle ranching within the protected forests.
<aside> 💡 In Mexico: Data Crítica obtained information about the presence of cows in protected areas in the southeast of Mexico. These areas are of high relevance, as some of them represent the largest humid tropical forests in Mexico and Central America. This information is relevant to obtaining images with high likelihood of bringing quality data for training an algorithm, and then detecting other similar areas by applying this algorithm to new areas, in some cases disregarded before this exercise.
</aside>
<aside> 💡 In Colombia: CLIP together with local partner 360, supported by the Pulitzer Center on Crisis Reporting, identified a high density of cows at the border of four National Parks that are key transitional ecosystems to protect the Amazon jungle, through cow vaccination data. To better place the cows on the map, we also collaborated with the Fundación para la Conservación y el Desarrollo Sostenible (Fcds) with ample on-the-ground knowledge of these forests, as sources for geographical shapes in these regions differ. The most updated data available was 2020. Identifying the cows in sat-images would allow us to evaluate the current situation and establish whether cattle and the deforestation that comes with it had continued to expand.
</aside>
Once we set our sites on chasing any indication of illegal cattle ranching within the protected forests of the Amazon, our next step was to acquire imagery that could support us. This would be the main ingredient to the computer vision algorithms that we wanted to try and train. Initially, we set out with an ambitious goal to target the four South American countries (Mexico, Colombia, Brazil & Argentina) that our respective news organizations had covered) but soon we realized we had to narrow down our areas of interest as this step was much more complicated than we had anticipated.
Source: Radiant Earth Insights on Medium
Access to satellite imagery is more widely available than ever, but getting higher-quality imagery (resolutions where object detection was possible), comes at a high cost. Despite trying to get in touch with the biggest providers (like Maxar, Planet, Sentinel, Google Earth), we found it to be hard to get a response to queries when you're interested in a collaboration or a small project. But soon enough, we found there are a growing number of programs that allow free access and/or download of satellite images to organizations and journalists in pursuit of specific goals and pivoted to that as a source.