Simons, Gareth D.; (2021) Detection and prediction of urban archetypes at the pedestrian scale: computational toolsets, morphological metrics, and machine learning methods. Doctoral thesis (Ph.D), UCL (University College London).

Preview

Text phd_final.pdf - Accepted Version Download (92MB) | Preview

Abstract

Granular, dense, and mixed-use urban morphologies are hallmarks of walkable and vibrant streets. However, urban systems are notoriously complex and planned urban development, which grapples with varied interdependent and oft conflicting criteria, may — despite best intentions — yield aberrant morphologies fundamentally at odds with the needs of pedestrians and the resiliency of neighbourhoods. This work addresses the measurement, detection, and prediction of pedestrian-friendly urban archetypes by developing techniques for high-resolution urban analytics at the pedestrian scale. A spatial-analytic computational toolset, the cityseer-api Python package, is created to assess localised centrality, land-use, and statistical metrics using contextually sensitive workflows applied directly over the street network. cityseer-api subsequently facilitates a review of mixed-use and street network centrality methods to improve their utility concerning granular urban analysis. Unsupervised machine learning methods are applied to recover ‘signatures’ — urban archetypes — using Principal Component Analysis, Variational Autoencoders, and clustering methods from a high-resolution multi-variable and multi-scalar dataset consisting of centralities, land-uses, and population densities for Greater London. Supervised deep-learning methods applied to a similar dataset developed for 931 towns and cities in Great Britain demonstrate how, with the aid of domain knowledge, machine-learning classifiers can learn to discriminate between ‘artificial’ and ‘historical’ urban archetypes. These methods use complex systems thinking as a departure point and illustrate how high-resolution spatial-analytic quantitative methods can be combined with machine learning to extrapolate benchmarks in keeping with more qualitatively framed urban morphological conceptions. Such tools may aid urban design professionals in better anticipating the outcomes of varied design scenarios as part of iterative and scalable workflows. These techniques may likewise provide robust and demonstrable feedback as part of planning review and approvals processes.

Download activity - last month

Export as CSVJSONXML

Download activity - last 12 months

Export as CSVJSONXML

Downloads by country - last 12 months

Export as JSONXMLCSV

Archive Staff Only

View Item