Mapping the urban heat Island at the territory scale: An unsupervised learning approach for urban planning applied to the Canton of Geneva

•Efficient unsupervised learning method for UHI analysis at the city scale.•Urban feature extraction with GMM clustering: 10 microclimatic clusters.•Simulated UHII with UWG tool and validation with urban on-site measurements.•Yearly average UHII analysis of Canton of Geneva ranging from 1.7 °C to 2.2 °C.•Methodology supports urban planning is reproducible and computationally efficient. This study presents a fully reproducible clustering-based methodology for the assessment of the urban heat island intensity (UHII) at the territory scale, using parametric microclimate models and limited computational resources. In large-scale climate modeling, a common preliminary operation is to utilize the well-established Local Climate Zone classification to characterize the thermal response of urban areas based on morphology. With the increasing availability of urban datasets, data-driven approaches can be implemented to quantitatively derive meaningful urban features without relying on a standardized classification. The proposed methodology employs a Gaussian Mixture Model clustering algorithm to partition the urban territory into a suitable number of homogeneous microclimate zones, enabling the calculation and mapping of the UHII for each zone through the Urban Weather Generator (UWG) tool. The developed approach is applied to the Canton of Geneva, Switzerland, identifying ten microclimatic areas and analyzing the spatiotemporal variation of UHII. Results show yearly average values of UHII ranging from 1.7 °C to 2.2 °C, depending on urban morphology. The simulated values are partially validated by comparison with on-site measurements from two urban weather stations, yielding a satisfactory agreement. The methodology can support urban planning with the goal of avoid overheating through a large-scale mapping. [Display omitted]