Building Urban Resilience: A Tensor-Based Simulation Framework for Sensitivity Analysis

Abstract

This study presents a simulation-based framework for analysing and predicting changes in urban activity patterns under disruptive conditions using aggregated mobile phone location data and tensor decomposition techniques. The research employs Mobile Spatial Statistics data from Kyoto, Japan, organised as a multi-dimensional tensor incorporating date, time, area, and age-group dimensions. Using Non-negative Tucker Decomposition, the framework identifies latent spatial, temporal, and demographic patterns in urban mobility data. The methodology simulates disruptive scenarios by exchanging the core tensor from a disruption event into the factor matrices of a normal condition, allowing the reconstruction of alternative population distribution patterns. Application of the framework to a simulated typhoon scenario during Kyoto’s peak tourist season reveals significant temporal, spatial, and demographic shifts in urban activity, including reduced mobility during commuting and leisure periods, greater spatial concentration of activity, and varying levels of sensitivity among age groups. The paper argues that tensor-based simulations provide urban planners and decision-makers with a flexible tool for exploring “what-if” scenarios, supporting contingency planning, disaster preparedness, and resilience-building strategies in complex urban environments.