505 / 2019-02-28 16:54:08

Spatial-temporal patterns of urban vitality: Analysis from mobile phone data

vitality,big data,uncertainty,spatial structure

全体主题 > Big data in Urban Planning and Management

Mobile phone signaling data provide a new approach to measure urban vitality. It extends a scale of a neighborhood/block/facility to a scale of a whole city. Urban vitality is also dynamic changing in different time of day. However, spatial-temporal patterns in urban vitality are less discussed in current researches. This paper aims to explore the spatial-temporal patterns in urban vitality, represented by distributions of populations based on mobile phone data in Guangzhou, China.

The study area is the Guangzhou city, China. It is the largest city in Southern China, with an area of 7434.4 km2 and a population of 14.5 million. The data were provided by China Telecom. The company took around 30% share of the user market. The locations information of user trips were aggregated to spatial cells (500m*500m) due to the privacy policy. It provided multi-days records, from Sep-08-2017 to Oct-08-2017 in Guangzhou.

The number of mobile phone users at each location in each time period was observed in multiple days. The vitality at a location in a time period was measured by the mean and the variance of population density in multiple days. The temporal mean and variance of all cells are shown in Fig. 1 and 2. Both mean and variance of spatial vitality are heterogeneous at locations and fluctuated with time.

The spatial distribution of the mean and the variance of the population are mapped in Fig. 3 and 4. From the thirty days’ mean of the population at 8 am, people are mainly distributed in the inner city. The variance also distributes in the inner city but more concentrated. At 8 pm (Fig.4 ) the average population density in thirty days distributes similarly to 8 am. However, the spatial distribution of variance at 8 pm is different from 8 am. It implies that it is necessary to further analyze the similarity and difference of spatial distributions of the mean and the variance at different time slides.

Tensor decomposition is one approach to find the spatial-temporal pattern of both the mean and the variance. Tensor decomposition is a method to decompose multiple dimensions matrix into a set of simpler matrix. Canonical polyadic decomposition (CPD) is one of algorithms in tensor decomposition (Rabanser & Günnemann, 2006). It finds three tensors: all-day mean tensor, day-time variance tensor, and night variance tensor. The spatial distributions of three tensors are shown in Fig. 5. For the first factorial tensor, all-day mean tensor, the high scores are mainly distributed in the inner city. The inner city has a high vitality at all time in a day. There exists the difference in the spatial distributions of day-time variance and night-time variance. Day-time variance is high in the center of the city, Zhujiang New Town, which is the CBD area. Whilst residential areas have higher variance at night, and the city center is low in variance. The tensor decomposition analysis uncovers the spatial-temporal pattern of the aggregate uncertainty. The uncertainty is different at day-time and night-time. At day-time, the spatial vitality is uncertain at the city center, particularly the CBD area; at night-time, it is uncertain at residential areas. The result refers to individuals’ daily activity, that the activity is mainly work-based at day time and home-based at night time.

This research explored the spatial-temporal patterns of urban vitality from mobile phone signaling data. The urban vitality is dynamic changing at different time of day. A tensor decomposition analysis finds the spatial temporal pattern. The city center’s vitality is uncertain at day-time and the residential areas’ vitality is uncertain at night.