Hour-of-day population variation across different functional zones of Beijing

Abstract

Conventional high-resolution population maps typically represent long-term averages and overlook the temporal rhythms of urban life. Using Tencent User Location Data and Amap points of interest (POI), we visualized the intensity and timing of hour-of-day population variation across functional zones in Beijing at 0.01° resolution. For each grid cell, we quantified diurnal fluctuation using the standard deviation of hourly population and identified the hour of peak population presence. These metrics were integrated with POI-derived functional zones to produce a three-dimensional visual representation. The results revealed that commercial areas exhibit the highest fluctuation intensity and tend to peak around midday, whereas residential zones showed the lowest variability and peak at mid-night. This featured graphic highlights the value of time-aware and function-sensitive population representations for urban analytics and planning.

Traditional fine-spatial-resolution population maps typically disaggregate census data using remote sensing and geospatial covariates, yielding long-term (often annual) average population distributions (Dobson et al., 2000; Gaughan et al., 2016). These static representations overlook the diurnal fluctuations in urban population density that are shaped by land use and urban functions. To reveal this temporal dynamic, we employed points of interest (POI) to delineate functional areas and used location-based social media big data as a high-spatiotemporal-resolution proxy for human presence. Taking Beijing as an example, we visualized the intensity of hour-of-day population variation and the hour at which population peaked across different functional areas, revealing fine-grained spatiotemporal rhythms that reflect the city’s real daily life.

Tencent User Location Data (TULD) provide high-frequency geolocated records of user presence from Tencent applications such as WeChat and QQ. With over 1.3 billion monthly active users on WeChat alone (as of March 2024), TULD has been widely used as a reliable proxy for population in China (Xu et al., 2021; Zhang et al., 2020). We aggregated TULD collected from April 15 to May 14, 2019 to hourly intervals and allocated Beijing’s total census population to each cell using the number of Tencent users as the weight. For each grid cell, we computed the standard deviation of hourly population to represent the intensity of diurnal variation and identified the hour at which population reached its daily maximum.

Urban functional zones were derived from 2019 POI data from Amap (also known as Gaode Map, a widely used Chinese mapping and location services platform comparable to Google Maps) at the same 0.01° grid resolution as the TULD-based population surface. Following previous research (Chin et al., 2024), we standardized POI counts by category within developed areas, extracted dominant functional dimensions using principal component analysis, and classified cells into eight functional types using k-means clustering. The inferred functional zones were overlaid with the hour-of day population metrics to visualize how the intensity and timing of daily population dynamics vary across urban functions (Figure 1).

Figure 1.

Diurnal population variation intensity and peak hours of population across functional zones in Beijing. Each 0.01° grid cell is represented by a vertical bar. Bar height indicates the log-transformed standard deviation (SD) of hourly population, representing the intensity of diurnal variation. Bar base color denotes the POI-derived functional zone type. Embedded color indicates the time period during which population reaches its daily peak. Undeveloped areas were defined by land use/land cover image product (Xu et al., 2018) and were not assigned a functional category.

Among the eight urban functional zones delineated by POI data, commercial areas exhibit the highest average intensity of diurnal population variation, followed by leisure and government zones (mean standard deviation of population: 11637 persons/km², 4992 persons/km², and 4539 persons/km², respectively). In contrast, residential areas show the lowest daily fluctuation intensity (1443 persons/km²). Notably, even in undeveloped areas, non-negligible population variations across hours are observed, which may reflect transient human activities. Across Beijing’s administrative boundary, the location with the greatest population variation intensity is found near the Olympic Center commercial district, where the difference between peak and trough hourly population densities reaches 529,198 persons/km². Moreover, the peak hour of population presence is strongly function-dependent: 43.2% of leisure cells reach their population maxima during 15:01–19:00, whereas 36.9% of residential zones peak during mid-night. Dining zones show dual peaks at noon (23.2% of cells during 11:01–13:00) and in the evening (19.6% of cells during 19:01–23:00).

While TULD provides valuable high-resolution insights, it remains a proxy for human presence and may not fully capture all population groups. Similarly, POI-based functional zones approximate dominant urban functions and may not reflect mixed-use complexity. Despite these limitations, the combined framework offers a scalable approach to visualizing spatiotemporal population dynamics and supports more time-sensitive urban planning and infrastructure design.

Software: Tencent User Location Data were collected by Python and processed with the “raster” package in the R statistical computing environment. The 3D population variation map was produced using ArcScene 10.8.