Making AI Work for Environmental Health in Low-Resource Regions: A Narrative Review

Abstract

Artificial intelligence (AI) and machine learning (ML) are increasingly used to monitor environmental conditions, estimate exposures, and support health-related decision making. Yet the settings with substantial environmental-health burdens often have the least complete, representative, and accessible data. This narrative review examines three interconnected constraints (data quality, data bias, and data scarcity) and considers how they affect the reliability, generalisability, fairness, and practical usefulness of AI for environmental health in low-resource regions. The review synthesises 38 sources published between 2018 and 2025 covering air-quality sensing, water-quality prediction, satellite remote sensing, environmental exposure estimation, public-health surveillance, and algorithmic fairness. Across the literature, low-cost sensors can expand spatial coverage but require calibration and maintenance; monitoring and health records may under-represent exposed populations; and models developed elsewhere may transfer poorly without local validation. The review develops an integrated challenge–solution taxonomy linking each data limitation to its consequences for AI, technical mitigation strategies, and policy or community interventions. Case studies from India, Ethiopia, Uganda, and Ghana illustrate how these constraints interact in practice. The review identifies priorities for low-resource settings: shared benchmarks, local and subgroup validation, explicit uncertainty reporting, interoperable data systems, stronger environmental and health surveillance, and governance arrangements that enable responsible use of local data. The central implication is that AI can extend environmental-health monitoring, but technical innovation cannot fully substitute for representative local measurements and health data.