In this episode, I talk with David Schurman, co-founder and CTO of Perennial, about their verification platform for climate-smart agriculture. Perennial uses geospatial data and machine learning to unlock agricultural soils as the world’s largest carbon sink.

Highlights:

  • How Perennial gathers and annotates training data from satellites and ground-based observations.
  • Handling variations across satellites and geographic locations.
  • Stratifying training data across the kinds of variables that matter.
  • Collaboration between machine learning engineers, remote sensing scientists, and crop scientists.
  • The importance of gathering more training data than you think you’ll need.
  • Respecting the data.
  • The nuance of communicating performance metrics.

Quotes:

“Perennial’s mission is to unlock agricultural soils as the world’s largest carbon. And it goes back to that theoretical potential to store carbon. Almost nothing compares to the potential of natural ecosystems, soil being one of those. Today, soil actually contains more than three times the amount of carbon that the atmosphere does. So a really small percentage change in the soils carbon pool is a huge impact on atmospheric carbon.”

“What we do is we take a ton of different geospatial data sets. There’s all sorts of stuff in there, like topography, climatology, hydrology, satellite, remote sensing, and we do a bunch of feature engineering and pre-processing on those. And the target variable is soil organic carbon. And so all of those factors are things that are highly correlated with how the soil ecosystem works, the factors that drive carbon formation, and the soil. And so we’re able to set up the supervised machine learning problem that way and detect changes in soil, carbon at scale, back through time in different geographies, focusing right now on the United States and Australia.”

“The kind of secret sauce, I guess is in the ground data collection. So we have teams of soil samplers who go out every year, multiple times a year, and they collect the gold standard of soil samples from these fields.”

“One of the things I really love about this team is it’s just so multidisciplinary. So we have machine learning engineers with classical statisticians working with remote sensing scientists and crop scientists, and it all kind of comes together.”

“Collect more training data than you think you need. I think that’s especially important for startups because when you’re early especially it can be really tempting to save on costs, maximize runway.”

Links:


Resources for Computer Vision Teams:

LinkedIn – Connect with Heather.
Computer Vision Insights Newsletter – A biweekly newsletter to help bring the latest machine learning and computer vision research to applications in people and planetary health.
Computer Vision Strategy Session – Not sure how to advance your computer vision project? Get unstuck with a clear set of next steps. Schedule a 1 hour strategy session now to advance your project.
Foundation Model Assessment – Foundation models are popping up everywhere – do you need one for your proprietary image dataset? Get a clear perspective on whether you can benefit from a domain-specific foundation model.