Session summary
This GeoField 2026 session presents three applied use cases exploring how Earth observation can support retrospective evaluation of agricultural resilience interventions. The presentations span Nepal, Bangladesh, and Ethiopia, with examples focused on disaster risk reduction, stress-tolerant crop varieties, and emergency seed aid.
Moderator: Jessica Wells
Reducing Environmental Risks through Sugarcane in Nepal
Dinee Tamang and Kunwar Singh present a use case from Mercy Corps’ Managing Risk through Economic Development program in Nepal. The study examines whether promoting sugarcane cultivation along barren riverbanks in Kailali and Kanchanpur districts helped reduce environmental risks such as erosion and flood-related land degradation. Using satellite imagery, retrospective feature digitization, matching, and difference-in-differences, the evaluation assesses changes in sugarcane cultivation and soil erosion while highlighting the importance of geospatial readiness during program design.
Stress-tolerant Rice Varieties at Scale in Bangladesh
Matt Hallas and Jeff Michler present research on submergence-tolerant rice varieties in Bangladesh. These varieties are designed to withstand moderate flooding, creating a challenging measurement problem for Earth observation. The session discusses both national-scale adoption and impact estimation, as well as a GeoField use case that attempted plot-level classification using Sentinel-1, Sentinel-2, survey data, and random forest methods. The work shows that rice can be identified reliably from space, but distinguishing specific stress-tolerant varieties requires stronger training data and more precise crop labels.
Diffusion of Sweet Potatoes in Ethiopia
Katherine Nolan and Ariel BenYishay present an evaluation of an emergency seed aid program implemented by the International Potato Center in Ethiopia. The program distributed orange-fleshed sweet potato and improved potato planting materials alongside agricultural and nutrition training. The evaluation combines a regression discontinuity design with survey data, PlanetScope imagery, field-level labels, and deep learning methods to assess whether the program increased sweet potato cultivation and whether adoption persisted after the intervention.
Together, the session highlights the value and constraints of geospatial impact evaluation after programs have already been implemented. Earth observation can help recover evidence, improve precision, assess persistence, and extend learning beyond household survey outcomes. At the same time, these cases show that retrospective evaluation depends heavily on the quality of field boundaries, crop labels, training data, ground truth, and early planning.