Session summary
This GeoField 2026 methods workshop provides a hands-on introduction to using geospatial impact evaluation to study rainwater harvesting and land restoration in arid landscapes. The session focuses on demi-lunes in Niger, which are small semicircular earthworks designed to capture runoff, improve soil moisture, and support vegetation recovery.
Rainwater Harvesting (Demi-Lune) Impacts in Arid Landscapes
Yuntian Bi leads a workshop on evaluating the landscape effects of demi-lune construction using satellite-derived outcomes and quasi-experimental methods. Because no complete dataset of demi-lune locations existed, the research team manually digitized more than 180,000 demi-lunes from high-resolution imagery and aggregated them to treated grid cells.
Participants work through the impact analysis workflow in R, including treatment and comparison data preparation, nearest-neighbor matching, pre-treatment trajectory checks, and staggered difference-in-differences estimation using the Sun and Abraham estimator. The workshop uses Landsat-derived vegetation and moisture indicators to assess whether treated areas experienced changes in greenness and soil moisture after demi-lune construction.
The session also covers interpretation and robustness checks. Participants examine dynamic treatment effects over time, test for possible spatial spillovers, compare impacts across baseline vegetation conditions, and assess whether results are driven by particular clusters. The example findings suggest sustained improvements in vegetation greenness and soil moisture, with effects becoming more visible several years after implementation.
Together, the workshop demonstrates how manual feature detection, Earth observation, matching, staggered difference-in-differences, and sensitivity analysis can be combined to evaluate small-scale restoration interventions in data-scarce environments. The broader lesson is that satellite data can extend learning beyond localized field measurements, but credible geospatial impact evaluation still depends on careful treatment definition, plausible comparison groups, timing, and transparent robustness testing.