Earth Observation and Causal Machine Learning for the Social Sciences

Wednesday, September 10, 2025
5:30 PM - 7:30 PM
1028 HN (in-person only)

Presented by
Kazuki Sakamoto
Adjunct Assistant Professor of Urban Planning
Columbia University

This presentation will explore cutting-edge methodologies that bridge earth observation data with causal inference for development research. Drawing from a recent book chapter, the talk will present a comprehensive framework for integrating satellite imagery and machine learning into causal analysis workflows, synthesizing five key approaches from outcome imputation to image-informed causal discovery. The presentation will also share findings from research applying geographic regression discontinuity designs with vision transformer models to address pre-treatment heterogeneity in satellite-based evaluations. Together, these works demonstrate how advanced computer vision techniques can move beyond prediction to enable rigorous causal analysis of development interventions, offering new methodological tools for researchers studying poverty, health outcomes, and sustainable development using earth observation data.

About the Speaker. Kazuki Sakamoto is a PhD Candidate at the Institute of Analytical Sociology at Linköping University, Sweden, and an Adjunct Assistant Professor of Urban Planning at Columbia University's Graduate School of Architecture, Planning, and Preservation. His research focuses on applying causal machine learning methods to satellite imagery for development impact evaluations, with particular emphasis on infrastructure and aid programs. He has previously served as a consultant for the World Bank, Inter-American Development Bank, New York City Economic Development Corporation, and Goldman Sachs.

Open to all students, faculty, and staff.