Spatiotemporal Decomposition of Urban Traffic Congestion into Recurrent and Incidental Patterns Based on Web Map Data

The acquisition of high-resolution city-scale traffic data is a major challenge for urban studies, as traditional sensors offer sparse coverage and commercial data sources are largely inaccessible. To address this gap, in this paper we present an automated framework to generate the high-resolution spatiotemporal data necessary for decomposing urban congestion into its recurrent and incidental components. The framework operates by systematically extracting quantitative traffic state information from the visual imagery of public web maps using computer vision. Implemented in the megacities of Shenzhen and Hong Kong, this “virtual sensor network” generated a continuous, minute-level spatiotemporal dataset of traffic state across the entire road network. This dataset was then used to perform an in-depth diagnosis of urban congestion using robust principal component analysis. In the analysis, complex traffic dynamics were decomposed into recurrent and incidental patterns, enabling the identification and diagnosis of congestion hotspots based on their intensity and duration. This work provides both a scalable methodology for generating traffic data and an analytical approach to support data-driven decision-making for urban mobility management.