This is the authors' abstract. We don't add key points for this paper.
Urban climate researchers increasingly use machine learning (ML) to examine microclimate conditions and their effects on comfort, health, and energy. Findings from these studies increasingly inform architecture, urban design, planning, and policy. However, how these approaches translate into usable knowledge for spatial decision-making in the built environment remains uncertain. This scoping review, conducted following PRISMA-ScR, examines 211 applied studies that use ML to relate urban form and environmental conditions to microclimatic outcomes across multiple spatial scales, from façade-level performance and street canyons to neighborhood patterns and city-wide assessments. Based on this literature, this review is organized into four principal application domains: (1) association and influence analysis, (2) classification and clustering, (3) data augmentation, and (4) forecasting and simulation. Across these domains, ML is used to identify relationships between built form and environmental conditions, group similar urban situations, extend incomplete datasets, and support predictive analysis relevant to planning and intervention. Several recurring limitations emerge when these studies are considered from the standpoint of practice. Many models are developed and tested within a single empirical setting, and studies rarely show whether the same relationships hold in other urban contexts. Data coverage and representativeness are also often unclear. In addition, urban form is frequently described using simplified indicators that miss geometric detail and typological differences relevant to climate-sensitive design. As a result, ML applications in urban microclimate research currently work best as tools for pattern recognition and exploratory analysis, while their use as a basis for robust design guidance or broadly applicable planning strategies remains more limited. To overcome the limitations, the review outlines a set of cross-context validation principles for researchers and practitioners.
Answers are grounded in this paper and cite their sources.
No takes yet. Share an insight, caveat, or question.
Manapragada et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: