Treatment Geometry and Causal Identification with Earth Observation Data

2026-07-22

AI Summary

Geospatial impact evaluations often grapple with ambiguity in defining the exposure units, timing, and intensity of interventions, particularly when multiple plausible exposure definitions exist. This study introduces the concept of “treatment geometry” as a foundational framework to systematically characterize the spatiotemporal footprint of interventions derived from Earth observation data. Centered on key trade-offs—including spatial resolution, temporal alignment, spillover effects, and boundary uncertainty—the framework provides diagnostic tools that enable researchers to identify which geometric definition choices are most critical for causal identification, rather than defaulting to a single methodological approach. Empirical applications to air pollution, wildfires, and forest policy demonstrate that this approach substantially enhances the credibility of causal inference and the rigor of empirical design.

Abstract

A central task in conducting impact evaluations is determining who or what was exposed to a treatment, when, and to what degree. These questions can be especially complex in geospatial settings, where many reasonable definitions of exposure may exist. This chapter introduces treatment geometry as a core concept in geospatial impact evaluation (GIE): the spatial and temporal footprint of a treatment as represented in data. How this footprint is defined shapes identification strategies and the credibility of causal inference. Drawing on cases spanning the air pollution, wildfire, forest policy, infrastructure, pest, and food security literature, the chapter provides practical guidance on navigating key tradeoffs (including spatial resolution, temporal alignment, spillovers, and boundary uncertainty) that arise when translating real-world interventions into analyzable data. Rather than prescribing a single best approach, the chapter equips researchers with a framework for diagnosing which geometry decisions may matter most in their context, closing with synthesis questions to help readers navigate these decisions.

Problem

• treatment geometry
• causal identification
• geospatial impact evaluation
• exposure definition
• Earth observation data

Innovation

• treatment geometry
• geospatial impact evaluation
• causal identification
• Earth observation data
• spatiotemporal footprint