Machine learning and remote sensing for smallholder precision agriculture in Ethiopia

Abstract

While machine learning (ML) and remote sensing (RS) are frequently heralded as the definitive solutions for agricultural resilience in Sub-Saharan Africa, a profound ‘implementation gap’ persists between laboratory-validated computational maturity and field-level utility for smallholder farmers. This systematic review, conducted under preferred reporting items for systematic reviews and meta-analyses (PRISMA) 2020 guidelines, critically analyzes why sophisticated models optimized for large-scale monocultures fail within the fragmented, intercropped landscapes of Ethiopia. By synthesizing empirical evidence across four domains—in-season crop yield forecasting, digital soil mapping, real-time biotic stress detection, and agro-meteorological modeling—the review uncover a fundamental scale mismatch between coarse-resolution satellite observations and sub-hectare micro-plots. The critique identifies localized data scarcity, hardware constraints, and the ‘last-mile’ connectivity divide as the primary friction points obstructing the transition from macro-level pixels to actionable, site-specific agricultural intelligence. Moving beyond simple summary, the study propose a strategic roadmap centered on decentralized edge computing, tinyML optimizations, and a restructuring of extension services to integrate digital intelligence into daily smallholder decision-making. These structural shifts are essential to bridge the digital divide and secure Ethiopia’s national food security against escalating climate variability. This review foregrounds the significance of digital agriculture within the context of the sustainable development goals (SDGs), specifically addressing SDG 2 (zero hunger) and SDG 13 (climate action) by enhancing crop productivity and building resilience in smallholder systems.