Objectives Commuting is arguably the most common behavior for urban residents. It describes the traveling behaviors of urban residents between residential and working places. Theoretically, every people will find a job near to his/her residential place. However, due to housing prices, children's education, and other reasons, most people do not live close to their working place. Therefore, there is usually excess commuting. There is a quite rich literature on excess commuting from the perspective of definition, calculation, social, and spatial dimensions of excess commuting, etc. Most of these studies are carried out based on two types of data. The first is the survey data, which is obtained by census survey, questionnaire and a face-to-face interview. This kind of data is very detailed and with socio-economic information, but it is time-consuming and with low spatiotemporal resolution. Besides, its sample size is usually limited. The second type of data is big data such as the GPS traces of taxis, the smart card data, and mobile phone data, etc. Such kind of data has a very high spatiotemporal resolution, small-scale, and covers a broad geographical range. The limitation is that such kind of data is usually hard to access and update, and with a lack of socio-economic information.
Methods Considering the limitation of the above two types of data, we propose a new approach to estimate excess commuting based on open-source data, which is near real-time and easily accessible. We investigate the spatial heterogeneity of excess commuting based on distance and time. We also explore the possible influencing factors of excess commuting. The main data involved are residential and workplace point of interest(POI), road networks, bus stops, metro stations and, metro lines, which are all downloaded from application programming interface(API). The population of each district is also collected. We define the POI sampling size as 1 500 and use a spatial sampling approach to derive representative residential and workplace POIs. For residential POIs, the complete spatial random(CSR) sampling approach is used. That is, we choose residential randomly from the set of residential POIs. While for workplace POIs, the types of the POI are considered. In other words, the more the POIs with the given type, the more likely POI with such type will be selected.We suppose that every residential POI has only one person, and every workplace POI has only one job. Also, each people should have and only have one job. Then we derive a commuting matrix of 1 500×1 500, indicating 2 250 000 commuting pairs. For each commuting pair, we record its commuting distance and time by Amap under public transportation mode. Three navigation strategies are considered, which are the fastest, the lowest cost, and the minimum transfer. For each commuting pair, its commuting distance is defined as the average of commuting distances with three navigation strategies. Similarly, we obtain the value of commuting time. The histograms of commuting distance and cost are drawn. The spatial heterogeneity of commuting distance and time is illustrated by the kernel density maps. We also compare the excess commuting of Chengdu with other cities.And, the possible influences of the spatial pattern of road networks, the spatial distribution of metro stations and metros, bus lines, and population density are explored by qualitative analysis and regression analysis.
Results The experimental results show that: (1) The maximum, minimum, and average commuting time for the area within the third ring of Chengdu City are 7 418 s, 434 s, and 2 922 s, respectively. The average commuting time for the first, the second and the third ring of Chengdu are 2 126 s, 2 439 s, and 2 922 s, respectively. (2) The maximum, minimum and average commuting cost for Chengdu City are RMB ￥ 8, 0, and 3.215 respectively.The average commuting cost for the first, the second and the third ring of Chengdu are RMB ￥ 2.54, 2.84, and 3.23, respectively. (3)The wasteful commuting rate of Chengdu is 80.69%, which is very close to Aurora and Baidu extracted by big data of GPS traces. (4) There is a significant core-periphery pattern for commuting time and cost. Most residential POIs in the central urban area is of low commuting time and cost. (5)The length density and branch density are negatively related to commuting time, while the shape index of the road network has a position relationship with commuting time. Our results are also closed to other empirical studies carried out by questionnaires. Besides, areas with dense population density tend to have high commuting costs.
Conclusions The proposed approach could be used for near real-time estimation of excess commuting. There are significant excess commuting for Chengdu City. People who live in the urban central area are more likely to waste less in commuting. Both spatial distribution of public transportation facilities and population will impact excess commuting. Development of the public transportation helps alleviate excess commuting.
DownLoad: