Identifying Spatial and Temporal Patterns of Urban Activities Using Mobile Phone Data

Recently, the use of big data from mobile devices has received considerable attention in transportation studies. The need to do activities is the main inducement for urban trip generation. Furthermore, urban activities and their patterns vary both over space and time. Mobile phone data, as a kind of continuous spatiotemporal data, records the location of people at different times. Therefore such data is suitable for the estimation of urban activity levels and the detection of patterns. In this study, we selected Shiraz as the study area due to its cultural, religious, and tourist significance, as well as the presence of major healthcare centres in the city. The analysis of spatial and temporal patterns of urban trips using continuous spatiotemporal data, such as mobile phone records, can significantly contribute to the improvement of transportation system management, planning, and policy-making for Shiraz.

The variable under investigation in this study is the activity density within a specific time interval and a defined spatial unit. Activity is defined as the number of individuals who either enter or leave a specific area for a specific purpose. Furthermore, activity density indicates the level of activity within the area’s unit of measurement. To investigate activity density across 321 traffic analysis zones (TAZ) in Shiraz, mobile phone data was collected over a one-week period (from 2021-06-24 to 2021-06-30). Following the implementation of data cleaning and preprocessing techniques,  individuals’ stay point and home locations were identified. The population of each TAZ was estimated by utilising the location of individuals within their respective homes. The estimated population and the real population in each spatial unit were employed to calculate the expansion factor. The activity levels within one-hour time intervals on workdays, semi-workdays, and weekends were estimated using an appropriate expansion factor. To examine the spatial dependency of the variable of interest (density of activities), global and local Moran’s I indices were applied to the aggregated density of activities. The study employed exploratory analysis of urban activities time series to identify the trend of activity level, peak periods, intensity change by time, as well as other relevant temporal characteristics. Additionally, the Standardized Normal Homogeneity Test (SNHT) was employed to identify the change point of activity in time series, which indicates the commencement of the activities.

The results not only demonstrated a significant positive spatial autocorrelation of the density of activities within traffic zones (P-Value < 0.001), but also identified the hotspots in the central parts of the study areas. It is notable that the central zones of the city exhibited high activity density, which was influenced by the spatial relationships within the study area. An exploratory analysis of time series revealed variations in activity patterns. These patterns exhibited higher activity levels on workdays compared to semi-workdays, and weekends. The time series observed in the latter half of the semi-workdays exhibited a striking resemblance to that of workdays, yet subsequently exhibited a trend between workdays and non-workdays as the activity level decreased. By examining the time series of activities, it can be observed that the mid-day peak period occurs at 12:00 to 14:00, while the evening peak period occurs at 20:00 to 22:00. Additionally, the lowest level of daily activity was identified between 3 and 6 a.m. The time series uniformity test was employed to ascertain the starting times of activities on workdays and semi-workdays, which were identified as 8:00 am, and on weekends, which were identified as 9:00 am. To validate the detected population and expansion factors and thus the estimated activity level, a spatial correlation between the estimated mobile phone population and the actual population within traffic analysis zones was calculated, which yielded an approximately 82% correlation coefficient. This correlation is statistically significant and therefore acceptable.

The results of these analyses could prove beneficial for the formulation of appropriate transportation planning and policy, as well as for the management of population density at hotspots at any time of the day. Furthermore, they could inform the analysis of urban transportation environmental impacts. With the availability of accurate mobile phone data for a range of spatial units, including traffic zones and even entire countries, it is possible to extract a diverse range of urban activity patterns, including those highlighted in this research.