Big Data Analytics and Machine Learning in Urban Environmental Governance: A Systematic Review of Applications, Effectiveness, and Enabling Conditions

Urban environmental governance faces mounting pressure from data complexity and the chronic inadequacy of conventional monitoring to support evidence-based decision-making. While big data analytics and machine learning (ML) are widely heralded as transformative, the conditions under which they translate into improved governance outcomes remain underspecified. This systematic review, conducted per Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) guidelines and registered on the Open Science Framework, examines 26 empirically rigorous studies (2015–2025) using a three-dimensional analytic framework, the Application Typology Dimension (ATD), Effectiveness Evaluation Dimension (EED), and Enabling Conditions Dimension (ECD), to map applications across the urban environmental monitoring lifecycle, evaluate demonstrated effectiveness, and diagnose structural conditions shaping implementation. Analytical capacity is concentrated in air quality and water resources, while biodiversity monitoring is absent and land-use and integrated multi-domain applications remain emergent; this shortfall reflects three distinct, non-exclusive mechanisms: technical, institutional, and disciplinary, each requiring a different remedy. Disaggregating effectiveness into five criteria shows that predictive performance and confirmed governance uptake are empirically distinct outcomes: several high-accuracy studies report no documented governance uptake, and risk anticipation, the least consistently demonstrated criterion, is limited by both training-data gaps and the absence of policy-linkage mechanisms. Enabling conditions are underdeveloped throughout, with no cluster rated above Moderate; coupling analysis indicates these are threshold-bound rather than independent, with technical achievement in one cluster repeatedly paired with an unaddressed shortfall in an adjacent one within the same studies. The evidence base’s concentration in East Asia reflects two distinct patterns: a severe, infrastructure-linked absence of studies from Africa, Latin America, and Central Asia bearing directly on model transferability, and a comparatively modest, coverage-linked underrepresentation of Europe and North America that does not. Layered, actor-specific recommendations for environmental authorities, research institutions, and technology operators show that technological sophistication must be matched by institutional readiness, ethical oversight, and contextual validation to achieve governance impact. This moves the field beyond descriptive accounts of technology adoption towards an exploratory, hypothesis-generating interpretation of governance-relevant patterns. These patterns should be tested against larger evidence bases rather than treated as a validated governance framework.