.png)
The world of climate AI is fast-moving — and can feel overwhelmingly so. In August, I knew nothing about geospatial foundation models. Fast forward to now, I’ve learned from an incredible team and shipped contributions to geospatial AI. Here are my reflections.
I’ve just completed a 6-month rotation with Earth Genome through the Schmidt Futures Technologists for Global Transformation Fellowship program. Serving as a Product Manager, my focus has been on expanding user access to Earth Index, our visual search tool for the environment. I was embedded within Earth Genome, focusing on the systems and resources needed to open up access to alpha users and provide the team a foundation for responding to rapidly growing demand. This rotation helpfully included in-person work at our fall offsite, where we built the product roadmap for the next year and brainstormed our long-term goals for Earth Index. I also connected with our user base and collaborators at Climate Week NY and AGU 2024 to demonstrate the importance of application-focus on foundation models, particularly building end-user interfaces to simplify access to these complex systems. I learned extensively about the challenges and successes of scaling a philanthropically funded technology product to a variety of users.
Advanced technology is only impactful if made accessible to those who are directly impacted by the problem. All the key guiding principles of Earth Index’s development revolve around accessibility, such as ensuring no specialized expertise is required to use the product and that data about local communities is owned by local communities. It has become even clearer that technologies like AI and machine learning can be really impactful for people’s lives and for environmental monitoring, but using models is a complex task for someone without an ML background. Earth Index helps to make this simple and close the gap between cutting-edge research and putting technology into practice in real life. This has pushed me to think deeper about prioritizing accessibility for tech products in the future. At DevSeed’s Large Earth Convening during AGU, I saw firsthand how usability is a dominant issue for geographic foundation models. Tools like Earth Index play a crucial role in making foundation models accessible and impactful, especially for indigenous peoples and local communities.
Building an initial set of users and scaling a product from 10 to 100 is really challenging, especially when developing a philanthropically funded tool. This fall, we began to transition users from the Earth Index alpha waitlist into active usage of the tool, with approximately 100 users by December. In preparation, I helped build Earth Index’s user management/support infrastructure by creating public-facing documentation, putting together user tracking mechanisms, and monitoring key user metrics to evaluate our success. From this process, I learned directly from our users about what they need urgently, such as global coverage and labeled data upload, which directly informed our roadmap planning. We faced challenges keeping new users engaged and retaining users without all their needed features, but we tried to understand their pain points and plan for improvements accordingly. With so many possible uses of Earth Index, we’re striving to fulfill a whole swath of goals for all of our users, while being limited by how fast we can build. Watching and helping this impressive team scale the tool has been an incredible learning opportunity in bringing a tool into the hands of users and growing quickly while meeting demands.
The Earth Observation and geospatial foundation model space is ripe with opportunity, and these technologies are really powerful if sufficiently funded and ethically utilized. As we’ve seen with the launch of groundbreaking models like Prithvi and Softcon, geospatial foundation models are extremely capable of processing Earth’s many data streams and completing complex downstream tasks without needing labeled training data. When I started at Earth Genome, Sentinel-2 or geo foundation models were foreign to me, but by December, I was able to engage productively in DevSeed’s Large Earth Convening discussing eminent challenges like usability and benchmarking. Additionally, I spent time this fall exploring ways to harness Earth Index to solve agricultural challenges, thinking about problems like crop mapping or yield prediction to build last-mile solutions for farmers. Conducting product research in a new domain was exciting, brainstorming how Earth Index might be effective for a new problem and where it might actually not be useful. As I move on from Earth Genome to a new rotation, I plan to continue learning about new methods and tools using EO data and geospatial foundation models.
It has been a fantastic and rewarding time at Earth Genome. Happy to have contributed to advancing Earth Index, poised to be the primary public option for environmental monitoring, protecting people and land under threat from destruction.
Are you interested in gaining experience and contributing to action on the most pressing climate and conservation issues of our times? Earth Genome welcomes placements of all kinds — internships, externships, and fellowships — to learn and contribute along with us. Get in touch to talk.