In partnership with the City of Vancouver, elementsLab developed, tested and evaluated a machine learning-based approach for identification, analysis and comparison of existing and future-state neighbourhood types. The types were applied to a test study site using a rules-based approach to change urban form and density. Select indicators from the City of Vancouver’s Resilient Neighbourhood Design Tool (RNDT) were used to evaluate physical design against resilience objectives.

A machine-learning method identified eight urban form types in Vancouver— a representative set of neighbourhood scale typical patterns, based on physical characteristics. The method utilizes primarily spatial data and a Gaussian Mixture Model (GMM), operated in a Geographic Information System (GIS). GMM is an unsupervised machine learning technique designed to probabilistically cluster data. It was trained on four facets of the City’s urban fabric: urban trees, street network, buildings, and parcels. This method can generalize and expedite the dynamic generation and measuring of urban form for rapid iteration of future what-if scenarios generated from the contemplated policy options. It includes data to enable comparative estimates of select indicators of urban resiliency- climate action, neighbourhood equity, built form, open space/amenities, housing mixes, and sustainable mobility.