Machine learning-based evaluation of dynamic thermal-tempering performance and thermal diversity for 107 Cambridge courtyards

The dynamic thermal conditions profoundly impact on the quality of physical, cultural, and social experiences in courtyard spaces. This research aims to identify microclimatic dissimilarities between courtyards terms tempering seasonal–diurnal extremes enriching ground-level textures. methodology included field measurements summer-2021 winter-2022 Cambridge, UK; simulations 107 ENVI-met model validations; machine learning-driven clustering using Super Organising Maps (SuperSOM). results indicate that diurnal range spatial-UTCI mean summer (DTR(M)<24?C) is double winter (DTR(M)<12?C); meanwhile maximum deviation three times as significant (?>3?Cat 7:00 BST versus ?>1?Cat 12:00 GMT). SuperSOM analysis was performed K-means hierarchical agglomerative partition all into seven subclusters its graph-lattice structure. Clusters Km_I, Hac_I, Hac_IV feature a positive synergy thermal-tempering thermal-enriching potentials. In contrast, other four clusters exhibit conflicting scenarios during day night across two seasons analysed. These data-driven outcomes enabled us optimise spatial landscape strategies for designing retrofitting microclimates, contributing current discussions climate-responsive sensation-inclusive design historical urban contexts.