Trends in urban flows: A transfer entropy approach
Roberto Murcio, Balamurugan Soundararaj · 2025
Read Trends in urban flows: A transfer entropy approach by Roberto Murcio, Balamurugan Soundararaj (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2501-06316.txt.
Introduction
Roberto Murcio and Balamurugan Soundararaj wrote this research paper, published in 2025. It estimates how crowds flow between places in a city, using a method that detects which location's activity helps predict another's, without tracking individuals.
Counting people as they move is costly, and tracking them closely threatens privacy. The study draws on a sensor network that tallies street-level activity at fine time and place resolution. A tool called transfer entropy then measures how much one spot's past improves forecasts of another spot's future. That reveals the direction and strength of flow links across the city. The authors frame the approach as a privacy-safer way to read urban rhythms, useful for transport planning, retail choices, emergency response, and day-to-day city management.
Catalog record
| CREATOR | Roberto Murcio, Balamurugan Soundararaj |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 25 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2501-06316
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Trends in urban flows: A transfer entropy approach Roberto Murcio, Balamurugan Soundararaj 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. K eywords Human mobility \cdot Information Theory 1 Introduction Accurately estimating the volume and flow of pedestrians in cities is one of the first steps towards understanding the structure of the urban environment [ 1 , 2 , 3 , 4 ] . Human activities such as footfall are highly granular and dynamic in space and time [ 5 , 6 ] , and estimating them with confidence is crucial for decision-making in numerous applications such as urban management, retail, transport planning, and emergency management. Detecting general trends in the inflow of people between spatial locations is neither obvious nor trivial due to the high cost of capturing these movements without compromising the privacy of those involved [ 7 , 8 , 9 , 10 ] . For the last ten years, new data sources on human behaviour that were not accessible before (WiFi, GPS, Bluetooth, for example) allowed for explanation and further understanding of human drives and decisions [ 11 , 12 ] . They also support individual users in decision-making as more shared knowledge becomes available at people’s fingertips daily. Insights into data from mobile crowdsourced sensors demonstrate the power of interconnectivity between people, space, and time [ 13 , 14 , 15 , 16 , 17 ] . The characteristic finding in all these studies is that people do not move randomly, i.e., following a Levy flight [ 18 ] . Instead, there is strong evidence for a conserved quantity in human mobility. [ 7 , 19 , 20 , 21 , 22 , 23 ] . All these models depend on the high granularity of their base data, allowing them to track devices through small areas (cellular towers) or at high detail (GPS). In this work, we also studied the movement of people, but without having the explicit route followed by each device. Using aggregated counts derived from passive Wi-Fi signal probing captured across Great Britain [ 24 ] , we analysed the reciprocal influence—or information transfer—between different pairs of locations. We proposed a complexity path measure that depends not only on urban morphology but also on the type of activity in each area and, consequently, on the time of day this measure is taken. A key aspect of this study is that the temporal resolution of the data employed is sufficiently granular to detect general patterns throughout the day without compromising people’s privacy. Our three main findings are that these data types are not random and capture local activity during the day. We proposed an urban classification based on FF and a complexity index to characterise local routes. 2 Data Description The data for this research form part of the Smart Street Sensor project [ 25 ] - a large-scale study commissioned by Consumer Data Research Center, UCL, to understand footfall (FF) in retail centres across the UK. The project employs a proprietary sensor network that records mobile device signals to discover available Wi-Fi access points. Once captured, the data is processed into five minutes aggregated counts. The precise algorithm followed to map Wi-Fi probe requests into FF counts and the associated bias are described in section Data construction and bias analysis in the Supplementary Materials. For the work presented here, we used data from the 1st of January 2017 to the 31st of December of the same year across 95 cities in Great Britain. Figure 1 shows an example of the sensor’s locations in Central London and Central Edinburgh, UK. (a) Central London (b) Central Edinburgh 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. The volume of FF reported at these locations varies widely. In 2017, the busiest location was inside Waterloo train station, London, with an average of 500 people per hour, while the quietest one was at Rockingham Road, Market Harborough, with only five people on average passing by this location. 2.1 Footfall signals Each location generates a daily time series – FF counts as a function time – or FF signal, that captures the local movements of devices around each sensor (Fig. 2 ). The FF signal is the basic unit of information in this study. 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 To introduce the general characteristics of the FF signals, an average FF was computed for all the five-minute periods in 2017 across all the locations in Great Britain, and the resulting aggregated time series was decomposed into standard trend, seasonal and residual components (Figure 3 ). Since the magnitude of the seasonal variation appears to be constant, the additive model was used to describe the time series. Figure 3: National average footfall in 2017 - decomposed time series The FF signal’s trend component shows volume growth throughout autumn, leading to peak activity during the festive pre-Christmas period. A sharp drop in FF near the end of the year is due to substantially lower counts on Christmas Day compared to the previous days and the next days. FF also exhibits a clear weekly seasonal pattern. Workdays are typically busier than weekends, and Sundays tend to record weekly lows. The residual component is expectantly present, as the probe requests used to construct the FF counts are rather noisy, initially coming with many sources of measurement error that had to be accounted for through data cleaning (Supplementary Information Data construction and bias analysis ). However, the change in the residual component does not exhibit any particular pattern, except for displaying the sharp drop near the year-end, which was already attributed to the low footfall on the 25th of December. 2.2 Assessment The FF signal’s trend and seasonal components indicate this data has a non-random nature and that it detects the ups and downs in FF activity around different urban locations. To further assess these features, we first look if these data are capturing typical human activity patterns, and second, we use as a proxy three different sources: 1) extraordinary events, 2) public transport figures, and Google popular times [ 26 ] and 3)