Measuring City-Level Transit Accessibility Based on the Weight of Residential Land Area: A Case of Nanning City, China

Abstract

A large body of research on transit accessibility emphasizes the importance of methods to simulating the real-world travel process. Few efforts have been made to conduct empirical research and comparative analysis of overall city-level transit accessibility. In addition, most of literatures utilize census tracts combined with the buffer method to estimate transit travel demand or available service areas, failing to take into account the reality that different land-uses have their own population. This research aims to develop an overall index of city-level transit accessibility based on the weight of residential land area. We integrated five types of destinations and the coverage of residential area within the transit stop service area to evaluate the overall structural problems of land use and public transportation in the process of urban development. Based on a case study on Nanning City, it was found that the weighted average travel distance is increased by 5.42 km, but the overall weighted travel time of the city is shortened by 7.65 min. In addition, an increase in coverage within the stop threshold and a decrease in the number of residential communities outside the threshold indicate that transit accessibility facilitates urban expansion. The empirical results show that the overall transit accessibility index can provide a reasonable measure criterion for the compact spatial structure and support urban strategic planning and address the problem of land use and public transport in the process of urban development.

1. Introduction

Transit accessibility represents how easy to travel by transit from origin to destination. It is often used to evaluate the relationship between public transport and land use [1,2,3,4,5]. After over one decade of improvements, indicators of transit accessibility have shifted from distance-and-time based schemes to cost-and-utility based ones [5]. Its measurement tools and methods have also transited from the simple Euclidean analysis to the road network and transit network analysis [4]. Nevertheless, due to different research intentions, little consensus has been reached on the measurement of transit accessibility in the process of urban development [6]. A large number of accessibility methods pay much attention to the formulation of models in an attempt to simulate the real travel process, such as the multi-stop or active space [7,8], and detailed parameters such as the frequency of transit services and transfer efficiency [9,10,11], resulting in more sophisticated models and difficulties [12]. Little attention has been paid to empirical research and comparative analysis of overall urban transit accessibility, which reduces the role of accessibility methods on evaluating the urban strategy for land use and transportation planning, and guiding in urban planning strategy analysis and policy improvement [13].

The spatial connection between residential communities (RCs) and relevant destinations connected by transit embodies the special relationship of urban spatial structure. Trips from dwellings to the nearest transit stop is the first mile of entire transit trip [14,15]. The man-land relationship associated with residential areas with access to transit is an important parameter to quantify stop travel demand [16,17,18]. Census tracts or administrative units have been widely used as proxies to quantify travel demand around stops. However, such quantification often results in overestimation errors [3,19]. Additionally, the distance buffer combined with census tracts is often used to estimate population within transit service areas which also ignores the characteristics of uneven population distribution in large-scale census tracts.

In addition, previous studies are concerned with effective transit accessibility, using distance standards as a threshold to estimate population with access to transit, such as 400 m, 600 m range [16,20,21,22,23] or 0.5 miles et al. [24,25]. the population outside the given distance were excluded from the study. The given threshold approach may be not suitable for all cases [14]. Moreover, this is inappropriate for city-level studies on land use and public transport development planning that require entire urban data not only within-threshold but also outside-threshold.

To address previous problems, this study aims to develop an overall city-level index to reflect the accessibility of an entire city from RCs to related destinations. Our measure uses data of residential land area as the weight, which is related to population as a proxy for population distribution, and combines with the threshold method to examine the city-wide accessibility relationship between RCs and transit facilities. The city-level overall transit accessibility can reflect the problems of urban land use and public transit layout in the process of urban development, and help evaluate the overall structural problems and the future planning improvement of land use and public transportation.

The paper has a few sections. Section 2 presents the literature review of transit accessibility. Included in definition of transit accessibility and population with access to transit stops. Section 3 is research method, the sources of case data. Section 4 introduces the data analysis results. The last two sections are about discussion and conclusion.

2. Literature Review

2.1. Definition of Transit Accessibility

Accessibility is often regarded as how easy to reach potential destinations for specific activities opportunity [10,26,27]. Similarly, transit accessibility refers to travelling only by transit or walk [14,28,29], indicating the degree of ease for a person participating in various activities from the origin to the destination. In view of different perspectives of accessibility studies, existing literatures can be classified into four categories. The first group is objective journeys of a person to the specific destinations by transit, which includes trips on walk from the origin and the destination to transit stops, as well as transit trips [10,27]. In other words, these objective journeys reflect the geographical spatio-temporal relationship between urban land use and public transport [14,30,31], which is specifically caused by the spatial separation of land use [31]. They are accessibility measures of these objective journeys in terms of distance or time, including access to transit stops, access to destinations and overall accessibility. These measures belong to the objective expression of the parcel spatial relationship based on land use.

The second group is the quantity and quality of opportunities available to the destination, including the basic essential needs (such as basic medical care, shopping, education, etc.) provided by the facilities, and the gravity formed by the quality and scale of high-grade facilities. The measurement of gravity involves the quality and quantity of supplies that match with the socio-economic level and conform to individual choices. In addition to measure of the spatio-temporal relationship, these measures also include differences in the quantity and quality of facilities, which use a gravitational model to express the magnitude of gravity [5,6,32].

The third group is the impact of different socio-economic factors on transit accessibility. Due to the different economic conditions of individuals, factors of the time and cost affect the transit travel of different social members. These measures require a lot of time-consuming household surveys to obtain a more accurate socio-economic situations of individuals [33,34] or special accessibility for a special group [27].

The fourth group is the provision of transit services, including the transit departure frequency, the waiting and transfer times at transit stops, which are associated with the departure frequency [10,12,29], and influence the ease of travel through the waiting and transfer time of passengers. The specific measures can be conducted using the details such as schedule, frequency and limited time, peak and off-peak hours, weekdays and non-weekdays [9,10,11]. These measures are closer to the real time of transit travel.

Although transit accessibility measures incorporating more contents and parameters maybe more fully reflect the details of public transit travel, such as quantities of opportunities and public transit services, it is relatively difficult to find a measure that can accurately capture all of these components [12], and more detailed parameters will lead to more complexity, thus it is more difficult to achieve the purpose of research on the overall urban accessibility by comparison. Furthermore, accessibility approaches for urban strategic planning do not require particularly detailed data [13]. This paper will focus on the measure of the overall potential transit accessibility at city-level, that is, the objective trips from the origin to the destination by transit, so that variable factors are excluded, such as socio-economic factors and available transit services, etc. In addition, destinations are expressed in terms of higher-grade facilities, ignoring community-level facilities, which means that there will be no shortage of supply within the service area of high-grade facilities. It is essential to simplify appropriately to focus on city-level potential accessibility.

2.2. Population with Access to Transit Stops

It is a key issue for researchers to obtain the number or proportion of the population around transit stops to accurately estimate the effective demand of public transit [19]. Many studies use demographic administrative units or census tracts as proxies, simplified to polygon centroids to estimate the distance from administrative units to transit stops [35,36]. Due to the too large extent of administrative units, the simplified method will result in overestimation of accessibility where the centroid within the threshold and underestimation if outside the threshold [37]. Alternatively, the buffer approach combined with the census polygons is utilized to estimate the proportion of the area served by public transit. Although the method narrows the scope of census tracts, the coverage areas are still based on the assumption of uniform population distribution and ignores the characteristics of uneven distribution, which also leads to estimation errors [12,17].

Some scholars attempt to propose an improved method of network ratio, which uses the ratio of the length of pedestrian network to the total length of streets in the accessible analysis areas to calculate the number or proportion of population with access to transit [38]. It is also assumed that the population is evenly distributed per street and proportional to the length of the street. In fact, the population of residential, retail and industrial land in the analysis areas will obviously not be the same [17]. Some studies advocate the subdivided grid method to estimate transit accessibility, either using the density-weighted of population [13], or the census data and grid unit overlaid to calculate population with access to transit [12]. Although the grid unit may be more precise than statistic administrative units, the allocation of population from large areas to a smaller grid still relies on the assumption of average population distribution, if the grid distributed population is located in the transit service area and just belongs to industrial areas or other areas without fixed population, we know it is an obviously incorrect estimate. In addition, the census tracks corresponding to administrative boundaries include many rural areas in China, therefore, the problem of the population or proportion with access to transit stops is still not well solved.

Many scholars agree that addressing this problem involves using smaller units or the decomposition of residential areas [13,39]. The parcel-based method of transit accessibility will provide a more accurate source of travel demand data due to population associated with parcels [17,19]. There is a literature developing parcel-network method to measure the transit accessibility, which correlates residential units with census tracts, and allocates the number of residential units to the subdivided parcels based on the total number of residential units in census tracts, even considering the residential type in the parcel. Based on the method, the population of each block is allocated to the subdivided parcels to measure the transit accessibility [17]. The parcel-network method improves results of the simple area and line-length allocation in census tracts by correlating the population with the residential units within the parcel. Unfortunately, the residential units and population allocated in the parcels are still based on total number of population and residential units in census tracts, and accurate data on population and residential units on subdivided parcels can only be obtained through expensive and time-consuming household surveys, so the case study provided by the author is limited to two northern communities in Dallas City.

RCs play an important role as the origin for parcel-based trips in the measurement of transit accessibility. There is a strong correlation between family travel time and accessibility conditions of RCs, and the proximity of the dwellings to activity locations will affect the travel time [11,18,40,41]. The land distribution of RCs can define the static urban population distribution attached to the residential lands, and using the proportion of urban residential land area gotten from the vector map and google images as a proxy for the proportion of the population fixed in residential land, has more objective and realistic characteristics than population in the census tracks arranged evenly to the subdivided grid. In addition, it is easier to be obtained compared with the population and specific building data.

This paper uses land area of RCs to measure the city-level potential transit accessibility, aiming to reveal the problems of land use, related facilities and public transportation in urban development by comparison.

3. Methodology

3.1. Research Steps

The measure of transit accessibility based on the land areas of RCs is divided into two levels. The first level takes the RCs as the origin of the entire trips to five types of destinations by transit (expressed in distance and time), and calculates the city-level composite index of transit accessibility with the weight of the residential land area. The second level measures the coverage of transit stops by means of given thresholds of the stop, reflects the distribution of RCs with access to transit facilities on foot, including the coverage of the residential and facilities of the stops, to analyze the relationship between land use and transport facilities from a micro perspective of RCs in the urban area.

In addition, for the destination facilities in the accessibility research, there will be some differences due to different research purposes, and should generally include facilities for the daily life of citizens. There is a literature suggesting that destination facilities should include shopping, education, entertainment, and employment [28]. Another research believes that health, financial and postal services should be added, but does not consider leisure facilities [13]. In addition, existing another literature believes that the social and recreational facilities should be added, but the employment place was excluded [12]. This research intends to develop a composite transit accessibility at city-level to evaluate the changes in land use and transport facilities in the process of urban development. It uses five types of facilities as destinations, including above community-level commercial, health, park and culture facilities, which further include large shopping center, Grade 2A hospital, parks, amusement parks, libraries, science and technology museums, et al. While they don’t consider other service facilities and primary and secondary schools, because the majority of facilities can be caught within walking distance from the dwelling in high-density China. The measure of accessibility for the employment place is represented by the city center [4,42], implying the assumption of the urban center-peripheral structure, the city center is the concentration of employment, and the arrival of the city center represents the employment commuting.

The measurement is divided into five steps (see Figure 1): The first step is to establish network datasets for transit accessibility analysis. In order to highlight the role of the subway on promoting public transit travel, the network dataset in 2020 is designed into two scenarios: with subway and without subway. The integrated subway and bus network datasets are processed by grouping and a connection strategy in the intersection of subway stations. The second step is to find nearest transit stops to RCs and destination facilities. Since the polygon centroids of RCs and points of facilities often deviate from the road network, and network analysis requires that all of the elements analyzed are strictly located on the network, so that the Euclidean distance method is used to find nearest stops. In addition, the nearest walking distance between RCs and facilities is obtained at the same time, and compared with the result of the third step to avoid errors. The third step is to obtain the minimum travel time onboard from the origin stops to the destination stops through the Arcgis10.5 network analysis function. If the walking time from the origin RCs to the destinations in the second step is less than or equal to the sum of the two walking times to the stops and the travel time onboard of the third step, the walking time without transit is taken as the accessibility time. The fourth step is to use the residential land area as the weight and calculate the weighted accessibility for each residential community (RC) to each destination type. Finally, the overall transit accessibility at the city level is calculated by summing the weighted accessibility of the five types of destinations. The fifth step analyzes RCs covered by the stops within the given thresholds and the distribution of outside the thresholds, based on the distance between the RCs and the nearest stops in the second step the corresponding RCs are selected within the thresholds of 400 m, 600 m, 800 m and 1200 m, respectively, and summarizing the land areas of RCs selected as the coverage of the stops within the thresholds and analyzing the distribution outside the thresholds. The results can reflect the efficiency of spatial structure in land use and public transportation.

The transit accessibility measure based on residential land uses the following equation:

$$D_i={{\displaystyle \sum }}_j{D}_{ij}={{\displaystyle \sum }}_j(d_{is}+d_{sp}+d_{pj})$$

(1)

j = 0,1,2,3,4, represent city center, shopping center, hospitals, park and culture facilities, specifically.

$$T_i={{\displaystyle \sum }}_j{T}_{ij}$$

(2)

d_is represents the distance from i RC to the nearest s stop, d_pj represents the distance from j facility to the nearest p stop, d_sp represents the shortest distance through the transit network between the two nearest stop s and p, D_ij represents the total distance from the i RC to the j facility by transit and walking, consisting of the sum of the above three-part distances, D_i represents the objective total trips from i RC to all types of facilities by transit and walking.

Since the total trip D_i of each RC, contains walking trips and transit trips, the time indicator is introduced for unified measurement, and the conversion speed is set as walking 5 km/h, bus 15 km/h, subway 40 km/h [42]. T_ij represents the travel time from RC i to the facility j through transit network, converts D_ij of the three-part trips into travel time, T_i represents the total time from i RC to all types of facilities by transit, indicating the transit accessibility in each RC.

The city-level transit accessibility measure based on the weight of residential land area uses the following equation:

$$HFD={{\displaystyle \sum }}_{j}HF{D}_j={{\displaystyle \sum }}_j{{\displaystyle \sum }}_i\frac{A_i{D}_{ij}}{{{\displaystyle \sum }}_i{A}_i}$$

(3)

$$HFT={{\displaystyle \sum }}_{j}HF{T}_j={{\displaystyle \sum }}_j{{\displaystyle \sum }}_i\frac{A_i{T}_{ij}}{{{\displaystyle \sum }}_i{A}_i}$$

(4)

Here A_i represents the land area of RC i; HFD and HFT represent the overall city-level transit accessibility based on the weight of residential land area expressed by distance and time, representing the sum weighted distance and time of the entire city from RCs to five types of destinations; HFD_j and HFT_j represent transit accessibility based on the weight of residential land area for each the different destinations, such as HFD₀ represents transit accessibility in total travel distance for all of RCs to the city center. The smaller the composite index of the HFT/HFD becomes, the better the overall transit accessibility in the city becomes, and the HFT/HFD has become bigger, indicating that the overall transit accessibility in the city has deteriorated. The same goes for the different destinations HFD_j and HFT_j.

Many studies set a threshold to determine the coverage of transit stops, such as 400 m or 800 m to measure the accessibility of facilities [10,43], and data outside the threshold are excluded, so it is not conducive to the overall accessibility measurement. In this paper, on the basis of measuring the overall accessibility in the city, the thresholds of 400 m, 600 m, 800 m and 1200 m are decided by the RCs distribution within the thresholds of stops (see Figure 2), respectively corresponding to about 8–20 min range of walking distance, are set for the analysis criteria to evaluate the efficiency of public transport facilities. And then the selected residential land areas are summarized within the given threshold, and further divided by the total residential land areas of the city to obtain the coverage rate around the stops. By means of the distribution analysis of RCs with different thresholds, we can identify the RCs with available or unavailable transit service areas precisely, and therefore determine the important areas needed to be improved in transit facilities and land use planning in the future.

3.2. Research Area

Nanning City is the capital of Guangxi Province (Figure 3), having land areas of 1834 square kilometers, had experienced rapid urban development from 2000 to 2020, the population expanded from 1.77 million to 5.29 million, and the built-up areas from 110.2 square kilometers to 326.7 square kilometers during 20 years, the population and built-up area in 2020 is close to 3 times in 2000. Since 1997, Nanning has taken the relocation of the municipal government as an opportunity to start an urban expansion eastward, and the city center has changed from monocentric, Chaoyang Square, to polycentric pattern, and a new city center, Jinhu Square has been formed. In the process of urban expansion, not only numerous new RCs have been built, but also many new hospitals, shopping centers, parks and other living and leisure facilities have been built. In addition, the construction of the subway began in 2012, now four subway lines have been operated by 2020.

3.3. Resource of Data

In Chinese cities, administration is hierarchically organized into different levels: RC, street, district, and city. RC is the basic administrative unit including villages in the built-up areas, we therefore use RC to measure transit accessibility. Data of Nanning in 2000 are derived from Map of Nanning 2000 and Google image (2002.5). Combined with Map of Nanning, the outline polygons of RCs land are extracted from Google image. While the data in 2020 come from the OSM vector map (2020) and Amap. Compared with travel surveys and household surveys, it is relatively easy that using GIS technology plots RCs outline and facility locations based on the images and vector map [41]. We got 1336 RCs in 2000, and 2531 RCs in 2020, the average land area is 16,024.9 m² and 26,767.9 m², respectively. Because real estate companies prefer the large scale, RCs have become bigger. Due to the park being large enough, each park entrance is used as the park point for accessibility measurement. The information of other data, such as transit facilities, and other facilities, sees

Table 1. Data source of origin, destinations and transit facilities.

Domains	Types	Count		Source
		2000	2020	2000	2020
Residential community	Origin	1336	2531	Google image2002 and Map of Nanning 2000	OSM vector map2020 and Amap2020
Bus route	Transit facilities	30	189	Map of Nanning 2000 and Google image2002	OSM vector map2020 and Amap2020
Subway line	Transit facilities		4		OSM vector map2020
City center	Occupation	1	2	Google image2002	OSM vector map2020
Shopping	Department store and shopping center	10	41	Map of Nanning 2000	Amap2020
Health	Hospital(2A)	18	31	Map of Nanning 2000	Amap2020
Park	Park and amusement park at district-level and above	15	49	Map of Nanning 2000	Amap2020
culture	Library, museum of science, technology and cultural	10	13	Map of Nanning 2000	Amap2020

.

4. Results of Data Analysis

4.1. Statistical Analysis of Result

Table 2. Descriptive statistics.

Index	Unit	2000				2020
		Min	MAX	Mean	Std.	Min	MAX	Mean	Std.
dis	M	18.06	2032.32	327.45	271.98	25.59	1426.49	215.64	135.44
Di	M	2364.82	50,965.53	16,236.31	8705.04	5928.22	71,742.41	20,633.19	9917.53
Ti	Min	17.78	262.41	84.72	42.22	27.03	272.04	78.31	32.75
Di0	M	329.32	11,877.73	4506.93	2360.90	259.04	21,229.14	6417.42	4158.20
Ti0	Min	3.57	59.47	21.92	10.81	3.11	74.35	20.43	10.51
Di1	M	175.75	11,350.10	3648.29	2392.55	182.99	13,842.91	2686.59	1909.44
Ti1	Min	1.96	58.72	18.60	10.85	2.20	42.86	11.98	6.37
Di2	M	172.22	8891.72	2073.17	1336.62	192.94	12,941.68	2713.47	1927.56
Ti2	Min	2.07	45.69	12.60	6.83	2.32	41.71	12.06	6.02
Di3	M	192.71	11,562.04	3011.83	2291.05	91.23	14,253.85	3085.10	1882.29
Ti3	Min	2.29	57.23	16.29	10.05	1.09	50.69	14.85	7.51
Di4	M	0.00	10,464.48	2950.54	1989.45	139.50	22,540.60	5730.61	3812.03
Ti4	Min	0.00	51.79	15.23	8.81	1.67	80.71	18.99	10.45

is a descriptive statistic of accessibility parameters of Nanning City in 2000 and 2020 (with subway). From 2000 to 2020, the mean value of the total travel distance (D_i) increases, but the total travel time (T_i) decrease, which denotes the transit accessibility of inhabitants in Nanning city improved in the context of urban expansion. Specifically, in comparison among the five types of destination, the mean value of D_i0, D_i2, D_i4 have a great increase, and D_i3 has a litter increase, while only the mean value of D_i1 has declined, which indicates an increase in the average distance from RCs to different facilities except shopping. In contrast, the mean value of T_i0, T_i2, T_i3 has a decrease in the context of distance increase, only T_i4 is an exception. The statistical results suggest that there exists urban expansion in Nanning, and this is a common phenomenon in developing cities in China [4,42]. On the other hand, the accessibility in terms of time improves clearly.

The result of the hotspot distribution of RCs shows characteristic of central agglomeration clearly according to the total travel time (Figure 4). Further analysis according to the uniform standard of transit travel time, 10 min, 15 min, 20 min and 30 min classification, from RCs to five types of destinations (Figure 5), the travel time from RCs to the city center (Figure 5a,f) and culture facilities (Figure 5e,g) in 2000 and 2020 shows a consistent central agglomeration characteristic. The same phenomenon is observed to the shopping center (Figure 5b) in 2000, while it is different in 2020 (Figure 5g), which may be the strategy of shopping centers to follow population migration under the action of marketization mechanism, so the time to shopping is shortened in some new developed suburban RCs. In addition, the hospital (Figure 5c,h) and park (Figure 5d,i) don’t show the characteristic of central agglomeration too, the times to arriving at these destinations in the new developed suburbs aren’t longer than the inner city. It is owing to the government-oriented development strategy of public facilities for politic fairness and local revenues. Because the better the facilities are, the more expensive the land price is, so the local government has the motive to build public facilities in new developed areas.

4.2. Transit Accessibility Based on the Weight of Residential Land Area

The transit accessibility based on the weight of residential land area gives different travel weights to RCs of different sizes, and the larger the residential land areas are, the greater the weight is, which means the more people traveling in the communities. This method can more objectively and truly reflect trips of the city as a whole than administrative unit arrangement. The overall transit accessibility based on the weighted residential land area (HFD and HFT) is 18.99 km and 99.36 min in 2000, respectively, and the HFD and HFT in 2020 (with the subway) are 24.41 km and 91.71 min (

Table 3. Average area weighted travel distance and time of Nanning City in 2000 and 2020.

	HFD (km)						HFT (min)
	HFD	HFD0	HFD1	HFD2	HFD3	HFD4	HFT	HFT0	HFT1	HFT2	HFT3	HFT4
2000	18.99	5.22	4.46	2.30	3.58	3.39	99.36	25.50	22.43	14.28	19.21	17.69
2020 without subway	21.17	6.82	2.86	2.88	3.06	5.55	103	30.32	14.64	14.79	17.22	26.02
2020 with subway	24.41	7.82	3.11	3.24	3.42	6.83	91.71	24.15	13.88	14.05	16.96	22.67

). The urban weighted average distance of trip is increased by 5.42 km, but the time is shortened by 7.65 min. It indicates that the weighted overall transit accessibility at city-level has increased in the context of weighted distance increase.

Furthermore, the travel distance of the three types of destinations has increased, especially under the condition that two urban centers have been formed in 2020, the distance to the city center (HFD₀) is still getting longer, which further confirms that Nanning has a significant outward expansion in space, and the size of urban planning layout has become larger, so the distance from RCs to various destinations has become longer. Whereas the weighted travel distance to the shopping centers has declined, because among the five types of destinations, the shopping centers takes market-oriented development strategy of following population suburbanization. While other facilities are basically government-oriented investment and construction, which have the consideration of the expected economy of scale, and the development plan is expected to cover more people, so it shows the opposite trend in the weighted travel distance.

The range of variation in the distance and time of the weighted land area from 2000 to 2020 is an obvious distinction for different destinations, compared with those of simple statistical average method (Figure 6), especially for the hospitals and parks. It denotes that the deviation of the land area gravity of RCs from the simple statistical average. If the positive variation of the weighted distance becomes larger, the proportion of RC land with longer distance is relatively large, vice versa. Therefore, the weighted land area method can more accurately express the relationship of the distance and number between the residential land and of transportation facilities.

In order to analyze the role of subway on transit accessibility in large cities, a comparative analysis is used by constructing two network datasets with subway and without subway in 2020. Compared with the index without subway network, HFD and HFD_j (different destinations with subways) are longer, but the HFT is shortened by 11.29 min, the HFT_j have been shortened too, especially the HFT₀ (city center) has been shortened by 6.17 min. The reason is when combining with the subway, the network calculation uses the nearest time algorithm, giving the preference to a path with the shortest time, so the HFD with the subway may be longer, but the HFT is indeed shorter. The results of comparative analysis indicate that the subway has a significant impact on the overall transit accessibility in large cities. It is very advantageous for suburban residents to go to the city center by transit for commercial activities. This is also the main reason for encouraging the construction of rapid transit in the suburbs, as well as the joint development of the residential following public transport [24,42].

4.3. Coverage of the Transit Stops Incorporating Residential Land Area

The distance from RCs to the stop is an important factor in the transit travel choice. In this paper, four thresholds of 400 m, 600 m, 800 m and 1200 m are set to analyze the residential land area covered by transit stops. The results show that the coverage of each given threshold of stops is much higher in 2020 than that in 2000 (

Table 4. Coverage rate of residential land area (RLA) within a threshold from the stop.

Threshold	2000			2020
	RLA (m2)	Coverage Rate of RLA (%)	Number of Residential	RLA (m2)	Coverage Rate of RLA (%)	Number of Residential
Dis ≤ 400 m	14,065,295.37	65.70	1006	59,540,425.1	87.88	2329
Dis ≤ 600 m	17,619,242.4	82.30	1184	65,671,390	96.93	2473
Dis ≤ 800 m	19,051,584.8	88.99	1253	67,212,790	99.21	2513
Dis ≤ 1200 m	20,265,924.9	94.66	1311	67,703,188	99.93	2530
Sum	21,409,262.7		1336	67,749,441		2531

). In 2020, the and 99.21%, specially. While the coverage rates within the same threshold are 67.5%, 82.3% and 88.99% in 2000, specially. It indicates that more of RCs of Nanning are located within the given walking range of the stops in 2020, and the transit stop has better residential coverage. The result further implies that during 20 years of development in Nanning, the density of public transit routes and stops has been greatly improved, the connectivity of RCs to transit stops is well-matched.

In addition, the distribution of RCs outside the stops of 400 m, 600 m, 800 m and 1200 m (Figure 7) shows that most of them are all located in the new development area or the outer suburbs of the city. When the threshold is relaxed from 400 m, 600 m to 800 m and 1200 m, the trend of suburbanization of RCs away from the stops is more obvious. In addition, comparing with the changes in transit accessibility of RCs outside the stop threshold in 2000 (Figure 7a) and 2020 (Figure 7b), it is consistent with the conclusions of Sultana [44], the transit accessibility in early RCs in new developing areas was poor, and with the urban development and commercial facilities emerging, the travel distance of RCs changes. Moreover, although the number of suburban RCs far from the stop decreased in 2020, compared with 2000, there are many new RCs uncovered by transit. The main reason is that with the urban development and transport improvement, new areas are developing outward constantly. It is evidence that transit accessibility facilitates urban expansion, and a phenomenon worthy of vigilance, especially under the current conditions of slowing urban population growth in China.

5. Discussion

5.1. Advantages of the Residential Land Area Weights Index of Transit Accessibility

This paper presents the composite index of transit accessibility at city-level based on the weight of residential land area (HFT/HFD), as well as the indexes for five different destinations (HFT_j/HFD_j) to evaluate transit accessibility at city-level. The method has the following advantages: First, compared with most studies on travel distances using small travel surveys and unofficial household surveys [45,46], it is relatively easy to plot RCs outline and facility locations in a map [41]. In addition, compared with the census tracts by area or by block route to evenly distribute the population to transit service areas, due to RCs with the characteristics of fixed and accurate spatial location and correlation with population, the measurement depended on the travel distance may be more objective, and the information of travel flows caught by the residential land area may be more actual.

Second, the transit accessibility measurement based on each RC to different destinations, can form a more comprehensive framework of transit accessibility in the whole city, and further discover the spatial structure problem of the dwelling, facilities and public transport in the process of urban development. In addition, the residential land data combined with the coverage of the stop thresholds can analyze the data not only within the transit service areas but also the distribution outside the service areas, which can reflect the transit efficiency of the whole city and provide the intuitive and visual results that precisely focus on the problems to be addressed in our future urban planning for land use and transit facility investments.

5.2. Interaction between Accessibility, Quantity of Facilities and Sustainable Urban Structure

Empirical study on transit accessibility of Nanning city shows that the weighted travel distance gets longer, while the weighted travel time gets shorter. It indicates that Nanning city has expanded outward in the past 20 years, but the overall accessibility has been improved. The empirical result shows that the urban structure in terms of five destinations including new developing areas gets better than ever, and transit accessibility of weighted residential land area can reflect urban structure changes.

When further analyzing the different destinations, we find that market-oriented shopping center has the best accessibility and biggest change in 2020 (Figure 6). It denotes that the quantity of shopping centers has increase more than other facilities. In general, there are more facilities and reasonably distributed, the accessibility is the better. In the other hand, If the accessibility improvements of shopping center are achieved by an increase in the quantity, which means the service areas get smaller and the market becomes more competitive. Thus, it would bring about operational risks of the shopping center in the market. The same goes for other facilities. Therefore, it is a large problem worthy of vigilance, especially avoiding economic risk caused by the debts of local governments and private real estate enterprises, which are formed by overbuilding in new developing areas [47].

In the cities or areas with rapid urbanization, it is almost inevitable that numerous people will enter the city, and the city will expand outward. Spatial proximity has become important indicators for measuring whether the urban structure is compact and low-carbon [48,49]. The overall index of transit accessibility proposed by this paper, combining with the stop threshold method, can provide a reasonable measure criterion for the spatial structure by quantifying travel relationship between urban land use and public transportation in the process of urban development. Therefore, we are able to guide the layout of the city to avoid the problem of non-compact spatial structure, especially in the development of new urban areas, to achieve sustainable development.

5.3. Limitation

Although the method proposed in this paper wants to incorporate as many details as possible to improve the model performance, there are still some shortcomings. First, when measuring the transit accessibility, although the weight of residential land area is better than the method of evenly distributing the population of census tracts by the buffer area, but due to the lack of detailed cadastral data, the difference in plot ratio is not taken into account, such as the difference plot ratio in the same residential land area, the associated population will be different, so there will be some errors. The availability of detailed cadastral data on residential land will improve the performance of the methodology.

Secondly, the five types of destinations in the method use the nearest time method to match the surrounding RCs, without considering the scale and supply capacity of the facilities, which implied that the supply capacity of the facilities can match the surrounding residential communities, although the measurement adopts high-grade facilities, such as hospitals above 2A level and supermarkets above designated scale, etc., in principle, it ensures that the nearby residents through the nearest time can get a certain quality facility service, but people may choose the nearest facility unnecessarily. Instead, they are willing to select facilities of high quality that match their income [33].

In addition, in the transit accessibility trips, only walk-transit travel mode is considered, and bicycle-subway, car-subway travel mode is not considered, and bicycle travel will broaden the coverage of stops, such as the 1.5 km or 2 km green travel circle of stops, and the car-subway travel chain has more advantages in suburbs with low-density population. Only considering walk-transit travel mode that may mislead the suggestions for suburban public transport planning, and the construction of parking facilities in suburban combined with rapid rail transit will probably have an advantage over the effect of increasing the transit outlets [42,50]. Therefore, subsequent research should integrate the comparative analysis of the three travel modes to further optimize the comprehensive index of transit accessibility at city-level.

6. Conclusions

This study proposes an overall index based on the weight of residential land area for measuring transit accessibility of RCs. Different from the method with detailed parameters, we simplify the parameter of transit travel by using objective travel distance/time and high-level destinations to measure transit accessibility, aim to establish the structural framework for the comparison and analysis of the overall transit accessibility at city-level.

We take Nanning city as a case, the results show that the urban weighted average distance trip is increased, but the time is shortened. It indicates that the weighted overall transit accessibility at city-level has increased in the context of weighted distance increase. The results mean there is no so-called suburban sprawl from the perspective of transit accessibility during the 20 years of rapid development in Nanning. Although urban area expanded outward, the investment in public facilities and transit facilities is also growing, so the transit accessibility has increased too. The empirical results show that the overall index proposed can provide a reasonable measure criterion for the spatial structure. It can be used to guide the urban planning and construction to achieve urban compact structure, especially in rapid urbanization regions and countries.

When comparing the index of the weighted land area with that of the simple statistical average, there is an obvious deviation in the results of the two methods. The greater the variation between the two methods becomes, the more the method of simple statistical average deviates from the gravity of population. Therefore, the weighted land area method can more accurately reflect the relationship of the distance and number between the residential land and of transportation facilities.

In addition, the comparative analysis of the two network datasets with and without subway shows that the subway has a significant impact on the overall transit accessibility in large cities. The result supports the construction of subways in large cities to improve transit accessibility. Moreover, combined with coverage within the different thresholds around the stops, it can evaluate RCs covered and uncovered by the stops of the whole city, and the results can be used as a basis for evaluating the efficiency of public transport and improving the future transport planning.