Trends in urban flows: A transfer entropy approach

TL;DR

The paper tackles the challenge of inferring urban pedestrian flows with privacy-preserving, high-resolution FF data from a Wi‑Fi probe network. It leverages Transfer Entropy to quantify directional information transfer between nearby locations and introduces a route-complexity measure to relate movement patterns to urban morphology. The findings show that FF signals are non-random indicators of local activity, enabling detection of ordinary patterns and extraordinary events, and that TE can uncover persistent directional biases beyond simple correlation. This approach offers a scalable framework for urban flow analysis with practical implications for planning, retail, and emergency management while highlighting the role of street structure and activity-type in shaping pedestrian dynamics.

Abstract

The accurate estimation of human activity in cities is one of the first steps towards understanding the structure of the urban environment. Human activities are highly granular and dynamic in spatial and temporal dimensions. Estimating confidence is crucial for decision-making in numerous applications such as urban management, retail, transport planning and emergency management. Detecting general trends in the flow of people between spatial locations is neither obvious nor easy due to the high cost of capturing these movements without compromising the privacy of those involved. This research intends to address this problem by examining the movement of people in a SmartStreetSensors network at a fine spatial and temporal resolution using a Transfer Entropy approach.

Figures (11)

• Figure 1: Distribution of sensors in a) Central London and b) Central Edinburgh. The brightest areas correspond to cities with a higher number of sensors in operation (which does not necessarily imply a higher FF, although it is the case in London.
• Figure 2: FF signals exhibit weekly circadian rhythms. For this particular location -inside Birmingham New Street rail station- we can observe an early peak in FF volume and high volume every five minutes during the middle of the day. This signal also reflects the lack of activity around the station on Christmas and boxing
• Figure 3: National average footfall in 2017 - decomposed time series
• Figure 4: Monthly FF counts. There is an evident drop in activity on the 25th of December and the 1st of January, while the peak in FF was reached on the 16th of December (a Saturday two weeks from Christmas)
• Figure 5: FF signal at some London streets during the day