Lumora: Seeing the Farm from Above, A Locally-Built Intelligence Pipeline for Smallholder Crop-Stress Monitoring in Data-Scarce African Landscapes

Across much of Africa's smallholder agricultural landscape, crop stress goes unnoticed until it is visible to the naked eye, by which point yield loss is often already underway. Formal monitoring infrastructure, whether satellite derived or field surveyed, has historically been built for large scale commercial agriculture in data rich regions, leaving smallholder farmland largely invisible to systematic oversight. This paper presents Lumora, a locally built, open source intelligence pipeline that brings crop stress monitoring to exactly this kind of landscape, using only free tier satellite infrastructure and no dedicated field budget. Grounded in a case study of a 6$\times$6 km smallholder agricultural region in Zanzibar, Lumora combines automated Sentinel-2 ingestion, a transparent proxy labelling framework built from literature derived spectral thresholds in place of unavailable field surveys, and a benchmarked classification pipeline (Random Forest and 1D CNN, both reaching around 98.7\% test accuracy against a naive baseline of around 82\%) delivered through a dashboard designed for non specialist extension workers rather than remote sensing experts. More than a single deployment, Lumora is offered as a template. It shows that rigorous, field survey free agricultural intelligence can be built by and for the communities it serves, using infrastructure already available to researchers across similarly under surveyed African farmland, without waiting for the data infrastructure that better resourced regions take for granted.