A Generalizable and Accessible Approach
to Machine Learning with Satellite Imagery

Professor Tamma Carleton

The combination of satellite imagery and machine learning is transforming our ability to map, monitor, and influence many global challenges, ranging from deforestation to poverty eradication to illicit activity. However, this emerging research area is data-intensive and computationally demanding, limiting its accessibility and use. Here, we show that a single encoding of satellite imagery can generalize across diverse prediction tasks (e.g., forest cover, house price, road length). Our method achieves accuracy competitive with deep neural networks at orders of magnitude lower computational cost, scales globally, delivers label super-resolution predictions, and facilitates characterizations of uncertainty. To democratize access to the rich information contained within satellite imagery, we share task-agnostic imagery features and a wide range of programming tutorials in a public web interface.

Biography: Tamma is an Assistant Professor at the Bren School of Environmental Science and Management at the University of California, Santa Barbara. Her research combines economics with datasets and methodologies from remote sensing, data science, and climate science to quantify how environmental change and economic development shape one another. Her work focuses on climate change, water scarcity, equity, and physical and mental health. Tamma is a member of the Climate Impact Lab, a research associate at the Environmental Markets Lab, and a faculty research fellow at the National Bureau of Economic Research. She holds a PhD in Agricultural and Resource Economics from the University of California, Berkeley, and MSc.'s in Environmental Change and Management as well as Economics for Development from the University of Oxford.