Earthrise Media: Deploying Data and Design to Solve Environmental Issues

Earthrise Media’s Accelerator grant partnership with PJMF is focused on indexing satellite imagery for search. By creating the ability to search, monitor, and characterize locations through imagery, the team aims to allow users to address local, urgent environmental problems.
By the Earthrise Media Team

Our team at Earthrise Media combines machine learning, AI, and human-centered design to guide and measure effective actions on the most pressing climate and conservation issues of our times. We recently built a novel platform to monitor plastic waste from satellite imagery, the Minderoo Foundation’s Global Plastic Watch (GPW). The results are compelling because they can be visually verified through regular Google Earth or Street View — imagery that users are already familiar with. For example, the site will allow governments and local communities to identify illegal dumpsites and those that are at risk of leaking plastic to the environment. Still, we recognize that there is a gap between awareness and action.

With the support of the PJMF Accelerator program, we are working to index the satellite imagery of these sites for search. The work will ultimately allow users to ask and answer decision-relevant questions like:

• Which sites have increased in size by more than three times in the past two years?
• Which sites are expanding towards water bodies?
• Which sites are at risk of leaking plastic to the environment?

The questions are small and idiosyncratic. We have found, however, that it is critical to be able to ask and answer those questions in order to address urgent, local environmental problems. This technology will ultimately support users of all types — including government, communities, researchers, and others — by making it easy to access satellite imagery in a way that can inform decisions.

The methodology that we applied for GPW is one that we have used for a number of years on environmental questions, for example, in Amazon Mining Watch. With the support of the Accelerator program, we are taking steps to turn that methodology into a technology that can be more easily and quickly applied for broader environmental detection and monitoring.

Our methodology reduces a stack of imagery — repeated satellite measurements for the same location over time — to an “embedding” that can be compared across time and space. The “embedding” is essentially a small set of numbers that captures the features of an image. A simplified example could use a scale of 1–10 to rate each feature of the image like:

• How much vegetation (how green)?
• How much brown dirt?
• How many straight lines?
• How uniform is the coloration?

Combining the ratings for all the features, a user will see the results at sequential stages and filter them based on local knowledge or context — much like the process of choosing a restaurant on Google, filtering by location, travel time, cost, and an array of other dimensions.

The key outputs for the Accelerator program will be a new user interface to filter the intermediate results and a database to serve the embeddings to web applications efficiently. This is the first step in transforming our methodology from something that can be applied in a bespoke way to a technology that can be applied more broadly. The user interface will primarily help our data scientists efficiently evaluate model results, dramatically increasing the speed and ease with which a new case study can be addressed. Speed matters for many environmental questions. For example, with Amazon Mining Watch, we have seen that an illegal mine can be built quickly, so quick answers about the locations of new illegal mines are needed to allow action to stop the destruction.

Our organization is working to take these methods and apply them more broadly. The technology we are building will allow our time-tested methods to be effectively applied to address many other environmental questions. Our goal is to allow users to answer questions about the planet relevant to decision-making, leading to better outcomes for organizations, society, and the planet.

The Accelerator program has helped us to refine our vision and map our early ideas toward a longer-term product. In addition to the regular support from the wonderful Accelerator program staff, we are excited to learn from the rest of the Accelerator cohort. The work that these teams are doing is very inspiring. Who knows, perhaps another team’s domain knowledge could inform the next use case for our technology to measure and monitor the planet?