Detecting Droughts from Space
A computer vision benchmark for satellite-based drought detection in Northern Kenya
A benchmark dataset and competition for predicting forage conditions for livestock in Northern Kenya from satellite imagery. The setup pairs Landsat tiles with expert forage-quality labels collected on the ground; the goal is a model good enough to feed a more accurate index for drought insurance. Better indices mean payouts that actually track the losses pastoralists experience, instead of being driven by noise in coarse vegetation indices.
I built this in collaboration with Weights & Biases, who hosted the public benchmark. The competition let CV researchers without development-economics context contribute to a real-world insurance product, and the W&B writeup walks through the data, baselines, and what the leaderboard taught us.
Presented at the Computer Vision for Agriculture workshop at ICLR 2020. Paper: Satellite-based Prediction of Forage Conditions for Livestock in Northern Kenya (Hobbs & Svetlichnaya).