Trustworthy AutoML indicators for climate-smart agriculture: A sustainability-governance framework for decision support

Automated machine learning (AutoML) can accelerate data-driven climate-smart agriculture, but environmental and sustainability value cannot be inferred from benchmark accuracy alone. Decisions about irrigation, nutrient management, crop stress, yield risk, and climate adaptation require evidence that automated model search is valid, calibrated, robust across places and seasons, resource-aware, explainable, and governable. This review article reframes agricultural AutoML as an indicator problem: what measurable and auditable evidence is needed before an AutoML-enabled advisory system can be treated as climate-smart decision infrastructure? Methodologically, the article is a structured critical synthesis with scoping elements rather than a completed systematic review; it integrates literature on AutoML, agricultural machine learning, remote sensing, trustworthy AI, sustainable AI, and digital-agriculture governance through an indicator-development lens. The result is TASC-AutoML, an indicator framework that links climate-smart objectives, agricultural decision contexts, data ecosystems, AutoML pipelines, trust gates, deployment constraints, and societal outcomes. We propose a minimum indicator suite covering decision linkage, data provenance, leakage-aware validation, uncertainty and calibration, spatial–temporal transferability, explainability, resource and lifecycle cost, equity, privacy, human oversight, monitoring, and accountability, together with a deployment maturity ladder and reporting checklist. We benchmark the framework against the NIST AI Risk Management Framework, the EU Trustworthy AI guidelines, and the OECD AI Principles, and illustrate it through a retrospective indicator scorecard applied to two published agricultural AutoML studies The framework does not claim operational performance estimates; rather, it defines the evidence structure needed to move from model-search demonstrations towards responsible, resource-efficient, and farmer-centered climate-smart agriculture. Because the framework is conceptual and literature-derived, its indicator suite still requires empirical validation; we outline the participatory, benchmarking, and expert-weighting studies needed to test it.