Climate-Resilient Agriculture: Systems, Science and Solutions for a Warming World

Dairy production must expand to meet rising demand while reducing greenhouse gas emissions, using scarce feed, land and water more efficiently, and protecting animal welfare under a warming climate. Artificial intelligence is frequently presented as the technology capable of reconciling these objectives, yet the evidence supporting that expectation has developed unevenly. This critical narrative review examines the state of knowledge on artificial intelligence in sustainable and climate-smart dairy farming, with attention to the strength, consistency and methodological quality of the underlying research rather than to the volume of publication. Literature was identified through structured searching of open scholarly databases and indexes, supplemented by backward and forward citation searching and by examination of authoritative institutional sources, with a final search date of 15 June 2026. Four analytical domains are synthesised: estimation and mitigation of enteric methane; monitoring of health, welfare and productive efficiency as an indirect route to lower emission intensity; adaptation to thermal stress; and the systemic, economic and governance conditions that determine whether algorithmic capability becomes farm-level benefit. The evidence is strongest for narrow, well-instrumented detection tasks in intensive housed systems, where models trained on sensor and image data achieve performance approaching or exceeding trained human observers. It is substantially weaker for the claims that matter most to the climate-smart argument. Very few studies trace a validated causal path from improved prediction to measured reductions in emission intensity, and external validation across farms, breeds, housing systems and climates remains uncommon. Recurrent methodological weaknesses include data leakage through inappropriate partitioning, imprecise reference standards, incomplete reporting, and a pronounced concentration of evidence in high-income intensive systems. Economic evaluation is fragmentary, and governance questions concerning data ownership, transparency and the welfare implications of algorithmic management remain largely unresolved. Progress requires multi-farm external validation, prospective evaluation against emission and welfare outcomes rather than classification metrics alone, and deliberate investment in evidence from smallholder and pasture-based systems.