Pedestrian volume modelling using space syntax and agent-based models

Pedestrian volume models (PVMs) enable evidence-based urban planning and help understand variables that influence pedestrian activity. This study compared agent-based and Space Syntax-based linear regression PVMs in Tel Aviv-Jaffa, Israel, focusing on three objectives. First, it compared the predictive power of Space Syntax-based linear regression analysis and agent-based models in two distinct urban areas: the urban core and a zoned, modernist neighborhood. Second, it explored the optimal variables and parameters for each model in order to explain and predict the pedestrian volume distribution in the street network. Third, it evaluated the feasibility of fitting a well-performing city-wide PVM. Regression models used 70 Space Syntax measures as well as land use distributions; agent-based models utilized agent route length, land use attractiveness and agents’ preferred distance minimization – metric, topological or angular – implicitly including Space Syntax measures. 247 hourly pedestrian observations were used for calibration and validation. Findings revealed that agent-based models predicted pedestrian volumes more accurately in separate urban areas, while Space Syntax-based linear regression performed better city-wide. The variables used by each model aligned with existing PVM literature, showing local Space Syntax centrality measures’ significance in the zoned neighborhood, while higher-radius measures dominated the mixed-use core. The agent-based model demonstrated some unexpected results, such as a preference for metric distance minimization in the urban core. The results demonstrate that novel modelling techniques may improve performance of pedestrian volume modelling in urban areas and could be adopted for future planning, and they reaffirm previously established variables for pedestrian volume modelling.