Mapping small-scale irrigation for climate adaptation (Proposals Track)

Anna Boser (UCSB); Jackson Coldiron (UCSB); Karena Lai (UCSB); Madhav Rao (UCSB); Jasper Luo (UCSB); Kathy Baylis (UCSB); Tamma Carleton (UC Berkeley); Kelly Caylor (UCSB)

Paper PDF Slides PDF Poster File Cite
Agriculture & Food Computer Vision & Remote Sensing Time-series Analysis

Abstract

Irrigation is vital for climate resilience and food security in Sub-Saharan Africa (SSA), yet small-scale, farmer-led systems remain poorly measured, especially during the dry season when they are most common. This gap limits research and policy on water management. We created a training dataset of over 2,000 hand-labeled images across Zambia from 2016-2024, which shows that 95% of dry season irrigated fields are small-scale, accounting for one-third of irrigated land, with prevalence rising significantly over the past decade. Building on this, we will produce the first country-scale, multi-year maps of dry season irrigation in SSA by training models on Sentinel-2 time series, evaluating approaches from tree-based baselines to geospatial foundation models. These maps will fill a critical data gap for climate adaptation and water governance while providing a benchmark for geospatial AI in agricultural monitoring.