Research Areas
Mapping agriculture
What crops are farmers growing? How is crop choice changing over time? Which agricultural practices, technologies, and policies lead to better outcomes for farmers and for the environment? We use satellite imagery and novel geospatial data to map agriculture at landscape scales, with a focus on low- and middle-income regions where data gaps are historically prevalent.
Geospatial machine learning
Geospatial data is a new frontier for machine learning. Unlike natural images, satellite imagery is multi-spectral, multi-temporal, and multi-modal; it contains spatial structure; it varies in resolution, and objects in it tend to be small and cluttered. Ground labels are also scarce for the majority of applications, while there are vast amounts of unlabeled imagery. We develop machine learning motivated by these characteristics.
Remote sensing for causal inference
Ultimately, we want geospatial data to help us make better choices for people and the environment. It’s not enough to make maps — we want to answer questions like: Do protected areas reduce deforestation? Which crop rotations improve soil health? How many stream miles and wetland acres are regulated by the Clean Water Act? We use remote sensing and machine learning to shed light on causal questions while correctly accounting for the errors that arise in such methods.