Spatiotemporal Characteristics and Determinants of Rural Construction Land in China’s Developed Areas: A Case Study of the Yangtze River Delta

Abstract

Rural construction land (RCL) received less attention but played an important role to control rural land use. Studying the RCL of developed areas may provide valuable references for underdeveloped areas to optimize land use. The Yangtze River Delta (YRD) is the most economically developed region in China. The study is intended to explore the spatiotemporal characteristics and determinants of RCL in the YRD based on a period of data from 1990 to 2017. The results show that the RCL in the YRD increases at an average annual rate of 5.38% but the growth rate tends to decrease. There is a weak spatial linkage of the RCL growth between cities. Clear spatial differences exist in the effects of every determinant of RCL. The correlation between the rural population and the RCL is unstable, which proves the existence of hollow villages. There is no clear correlation between the RCL and the local economy and accessibility, as the rural population normally goes to few big cities for higher salary work but spends the money in their hometowns on building homes. These findings help optimize rural land use in the YRD and provide an important reference for planning land use in underdeveloped regions.

1. Introduction

Land use is the most direct form of human activities affecting the natural environment, and it has always been a hot topic in academic circles [1,2]. Studying land use change can deepen the understanding of the interaction mechanism of human–land relationship system. The rural construction land (RCL) refers to the construction land in rural areas, mainly including land for public utilities in towns and villages (e.g., roads) and residential land for rural residents. The RCL has traditionally been an active area for research on human–land relations and the development of rural areas because it supports both production and the lives of the rural population [3,4,5]. As a global issue, when the urbanization accelerates, the rural population is also decreasing [6,7]. According to the latest global demographic analysis, approximately 45% of the world’s populations are living in rural areas, but this number is projected to decrease to 30% by 2050 [8,9]. Growing urban expansion is decreasing rural land. To make up for the shortage of RCL, more and more arable land or ecological land is being developed for human activities. This has led to irreparable damage of the ecological environment and loss of cultivated land resources [9]. Moreover, idle land and abandoned residential properties are common in rural areas. This not only wastes land resources and damages the environment, but it also greatly increases the pressures on arable land and may even threaten future national food security [10,11,12].

With rapid economic growth as well as industrialization and urbanization [4,13,14,15,16], China’s rural population is migrating to cities, and the permanent population in rural areas has been decreasing continuously. Houses in villages are idle, land use efficiency is low, a large amount of land resources is wasted, and the contradiction between land supply and demand has become increasingly prominent [17]. In fact, reduced rural population with rapid urbanization does not necessarily mean less RCL in China. On the contrary, the RCL is expanding. This forced China to confront the issue of rational planning of RCL. At the same time, the development between urban and rural areas in China is unbalanced. The RCL is basically excluded from the land market. In 2013, China issued a document on further reform and the establishment of a unified construction land market for urban and rural areas. Under the premise of complying with planning and usage regulation, the rural collective management construction land is allowed to be sold, leased, and bought as shares. Since then, the RCL has been undergone major changes.

Compared with urban land use, the RCL received much less attention, and due to the unique institutional mechanism, i.e., urban–rural dual structure, China’s countryside is much different from the city. As a result, researches of rural land use in China and other countries have different focuses. Few researches solely focused on the RCL; instead, many of them focused on the change of cultivated land, Land Use–Land Cover Change, land acquisition conflicts, etc. [2,6,18]. Focusing on RCL in China, some scholars carried out research in the value analysis of RCL [19], land circulation [20], and the relationship between RCL and the rural population [21]. Most of these studies focused on small and medium study areas (e.g., counties, towns, and villages) [22,23,24]. A few researchers carried out provincial, region-specific, or even national macroscale analyses [25,26,27,28], and they found that RCL expanded faster in the developed eastern regions, slower in the central and western regions [29]. Topography, rivers, transportation, and food production all have a significant impact on RCL. However, macroscale quantitative analyses of RCL in economically developed areas remain rare [28]. To study the changes of RCL in developed areas would help optimize local land use and provide valuable references for other underdeveloped regions to formulate scientific land planning.

The Yangtze River Delta (YRD) region is the most economically developed area in China. Despite the fact that resources and population continue to concentrate in this region, the contradictions between land use and social development system have become increasingly acute in rural areas. By taking the YRD region as a case study, this research uses Exploratory Spatial–Temporal Data Analysis (ESTDA) and the Multiscale Geographically Weighted Regression (MGWR) model to explore spatiotemporal characteristics and determinants of RCL in developed areas. The findings will be conductive to the establishment of a unified urban and rural construction land market, promote the reform of rural land system and rural revitalization, and provide a basis for ongoing reform and preparation of the five-level and three-type territorial spatial planning system.

The paper consists of five sections. The second section describes the data and methodology adopted in the study. The third section presents the results. The fourth section discusses the results, contributions, and limitations. The final section concludes with the findings.

2. Study Area and Methodology

2.1. Study Area

The YRD region is an economic center that has been at the frontier of China’s reform and opening-up in recent decades. This economically developed region spans the middle and lower reaches of the Yangtze River plain with a total area of 358,000 km². By the end of 2020, the YRD had a GDP of US$3.49 trillion. This region generates a quarter of the national economy while only accounting for 3.71% of the land area. We selected Jiangsu Province, Zhejiang Province, Anhui Province, and Shanghai as the study area based on the outline of the YRD region in the Regional Integrated Development Plan issued by the State Council of China in 2019. The study area includes 41 prefecture-level cities/municipalities (Figure 1). The eastern coastal area of China is much more developed than the central and western regions; similarly, in the YRD region, the three eastern coastal provinces, Shanghai, Jiangsu, and Zhejiang, are a little more developed than the central inland province (Anhui).

2.2. Data

Considering that China’s development occurred relatively slowly in the early stages, and that social and economic data before 1990 are difficult to obtain, the period from 1990 to 2017 was chosen as the study period. The RCL data of the YRD region from 1990 to 2017 used in this paper were obtained from the Long-Term Impervious Surface (1978–2017) dataset created by Gong et al. [30], which was derived from Landsat sequence imagery with resolutions of 30 m and 16 d, respectively. NPP-VIIRS nighttime lighting data were used to aid in the calculation. The accuracy of the dataset is in the range from 85% to 95% in most regions (a few regions in western China have an accuracy of less than 70%). The long-term dataset includes two sets of data: the total impervious surface area of the entire region, and the urban impervious surface area of the town. The impervious surface area data in this dataset have been adopted by a few studies regarding the construction land [31,32]. We obtained the RCL by excising urban impervious surface from the total impervious surface. By superimposing the vector’s administrative divisions with the RCL data, the RCL of each prefecture-level city/municipality was then determined. The above dataset can be downloaded for free from this website (http://data.starcloud.pcl.ac.cn/zh (accessed on 15 April 2022) [33].

From 1990 to 2017, the RCL in the YRD region increased steadily each year, with a total increase of 14,768 km², an average annual growth area of 527 km², and an average annual growth rate of 5.38% (Figure 2). The growth rate for the entire study period first decreased and then increased with some fluctuations. The annual growth rate was the highest from 1990 to 2000, with an average annual growth rate of 7.22% and an average annual growth area of 402 km². The average annual growth rate in RCL from 2010 to 2017 was 5.91%; however, the average annual growth of the RCL area was 808 km², much higher than that in the preceding period.

Digital elevation model (DEM) data were obtained from the Resource and Environmental Science and Data Center of the Chinese Academy of Sciences [34]. The resolution of the data was 30 m. The slope was calculated from the DEM data using Arcgis10.6. Road network vector data were obtained from Navinfo [35] and Tianditu [36]. The scale of the map was 1:450,000. Social and economic data were obtained from the Statistical Yearbook of Anhui Province, the Statistical Yearbook of Jiangsu Province, the Statistical Yearbook of Zhejiang Province, the Statistical Yearbook of Shanghai, and statistical bulletins of national economic development [37] and social development [38] for various cities in 2000, 2010, and 2017.

Regarding the determinants of RCL, both natural and socioeconomic factors including height, altitude, population, and GDP of primary industry have important effects on construction land [23,25,39]. Natural factors generally determine the basic development conditions of an area, and their influence on construction land remains relatively stable. On the other hand, social and economic factors often dominate local development decisions and often have important effects on construction land. The GDP of primary industries and the area of arable land is obviously closely related to RCL, and they have a trade-off relationship with RCL. The rural population is the body of rural economic activities on RCL and has a great influence on RCL. Transportation accessibility reflects the connectivity of the area to other areas. The accessibility of a region is the average of the minimum traffic time from the region to all other regions. The minimum traffic time between regions is calculated based on road network, in which different type roads are assigned different velocity. GDP is a frequently used indicator to characterize regional economy. The elevation and slope are the main natural factors affecting construction conditions. Accordingly, seven indicators closely related to rural development were selected as explanatory variables in this study (

Table 1. Determinants of RCL.

Category	Variable	Reference
Socioeconomic factors	GDP of primary industry	[31]
	rural population	[28]
	area of arable land	[40,41]
	accessibility	[42]
	GDP	[43]
Physical factors	elevation	[31,44]
	slope	[31,44]

). The observation scale of all indicators is the prefecture-level city.

2.3. Methodology

The methodological flow diagram is shown in Figure 3. We first applied the ESTDA framework to investigate the dynamics of the spatial pattern of RCL based on the RCL datasets of the YRD region containing 41 cities from 1990 to 2017, then revealed the transitions of local spatial structures, and lastly visualized the spatial coevolution between cities and their neighbors. MGWR was employed to estimate the effects of various influencing factors on RCL.

2.3.1. Annual Growth Rate of Construction Land

The annual growth rate of construction land (K), which is used to quantitatively describe the annual growth of a certain land type, is given by

$$K=\frac{U_b-U_a}{U_a}\times 100\%$$

(1)

where U_a and U_b are the areas of a certain type of land at times a and b, respectively.

2.3.2. Gini Coefficient

The Gini coefficient is used to investigate differences in the distribution of geographic elements in space. We use this coefficient to observe the differences in the growth of RCL across the YRD region. The closer the Gini coefficient is to 1, the more balanced the growth of RCL is in the region. It can be calculated by Equation (2) [45].

$$G=\frac{1}{2n^2\overline{x}}{\displaystyle \sum _{i=1}^n{\displaystyle \sum _{j=1}^n\left|x_i-x_j\right|}}$$

(2)

where x_i and x_j are the growth rate of RCL in city i and city j, respectively; $\overline{x}$ is the average growth rate of RCL of all cities; n represents the number of cities.

2.3.3. Exploratory Spatiotemporal Data Analysis (ESTDA)

As a classical temporal GIS analysis method, ESTDA can be used to identify the spatial–temporal changes in research variables [46]. The method is widely used to explore the spatiotemporal patterns in land use [27,31]. Therefore, in this study, ESTDA was used to effectively integrate spatiotemporal information and explore the spatiotemporal dynamics of RCL in the YRD region.

ESTDA involves several components: Moran’s I, a time path, a spatiotemporal transition, and a spatiotemporal network [47,48]. The specific ESTDA method is detailed below.

(1)

Moran’s I

Moran’s I is frequently used to analyze the spatial autocorrelation of research objects. Spatial autocorrelation contains two models, namely global spatial autocorrelation and local spatial autocorrelation. Global spatial autocorrelation describes the distribution of an element attribute over the entire study area. It is used to judge the agglomeration characteristics and intensity of the element [49,50]. The global Moran’s I, which takes a value between −1 and 1, is given by

$$I=\frac{N}{{\displaystyle {\sum }_{ij}{\omega }_{ij}}}\frac{{\displaystyle {\sum }_i{\displaystyle {\sum }_j{\omega }_{ij}(x_i-\overline{x})(x_j-\overline{x})}}}{{\displaystyle \sum {}_i{(x_i-\overline{x})}^2}}$$

(3)

A positive or negative value of I reflects a positive or negative spatial association of a city with its neighbors, respectively. In Equation (3), x_i and x_j represent the rate of change in RCL in city i and city j, respectively; $\overline{x}$ is the average of the change in RCL in each city; ω_ij is the spatial weight matrix, which describes the adjacency of two samples in space; and N is the number of samples.

The local Moran’s I (I_i) is used to investigate the local spatial autocorrelation of neighboring objects (i.e., their difference and relation):

$$I_i=\frac{{\mathrm{z}}_i}{S^2}{\displaystyle \sum _j{\omega }_{ij}{z}_j}$$

(4)

where I_i represents the local Moran’s I, where z_i = y_i − $\overline{\mathrm{y}}$, z_j = y_j − $\overline{\mathrm{y}}$, S² = Σ(y_i − $\overline{\mathrm{y}}$)²/n, y is the annual growth rate of RCL; ω_ij is the row-standardized contiguity matrix. S² is positive, so the result of local Moran’s I can be judged by the positive and negative of z_i and Σω_ijz_j. The LISA coordinate system is constructed with z_i/S² as the horizontal axis and Σω_ijz_j as the vertical axis. There are four kinds of results in pairs of positive and negative combinations, corresponding to the four quadrants in the LISA coordinate system, to describe the local spatial correlation between a sample and the adjacent sample.

(2)

LISA (Local Indicators of Spatial Association) time path

In the LISA coordinate system, the coordinate of each sample element represents the cluster type of the element and its neighbors. Connecting the positions of each city in the LISA scatterplot along the time series will form a dynamic time path with geometric features mainly including the relative length and tortuosity. Relative length reflects the relative magnitude of changes in each city. When the relative length of the time path is greater than 1, city i’s moving distance is longer than average, so the change of city i is relatively more dynamic; otherwise, it is stable [51]. Tortuosity is used to characterize local spatial dependence. The greater the tortuosity, the stronger the spatiotemporal dependence of inter-regional RCL [47]. The above two indicators not only describe the synergistic change in RCL, but they also characterize local spatial differences to reveal the spatiotemporal interactions among geographical elements. The relative length Γ_i and the tortuosity Δ_i are calculated as follows:

$${\Gamma }_i=\frac{n\times {\displaystyle {\sum }_{t=1}^{T-1}d(L_{i,t},L_{i,t+1})}}{{\displaystyle {\sum }_{i=1}^n{\displaystyle {\sum }_{t=1}^{T-1}d(L_{i,t},L_{i,t+1})}}}$$

(5)

$${\Delta }_i=\frac{{\displaystyle {\sum }_{t=1}^{T-1}d(L_{i,t},L_{i,t+1})}}{d(L_{i,t},L_{i,t+1})}$$

(6)

where L_i,t is the position of city i in the Moran scatter plot in year t; d (L_i,t, L_i,t+1) is the distance city i moved from year t and year (t + 1); and n is the number of cities (in this paper, n = 41).

(3)

LISA space–time transition

The LISA space–time transition describes the spatial relationship between a city and its neighbors characterized by changes over time. Sixteen different spatiotemporal transition in four categories are defined by a modified Markov chain measure of sample migration [47]. Conversion types are based on the location of the city in the Moran scatterplot and are categorized as high and high agglomeration (HH, that is, a city and its neighbors all have high values for an indicator), low and high agglomeration (LH), low and low agglomeration (LL), high and low agglomeration (HL). The specific transition types are described in

Table 2. Space–time transition of city agglomerations.

Type of Transition	Description	Transitions
I	Self-transition with neighborhood stabilization	HHt→LHt+1, LHt+1→HHt+1, HLt→LLt+1, LLt→HLt+1
II	Self-stabilization with neighborhood transition	HHt→HLt+1, LHt→LLt+1, HLt→HHt+1, LLt→LHt+1
III	Self-transition with neighborhood transition	HHt→LLt+1, LHt→HLt+1, HLt→LHt+1, LLt→HHt+1
IV	Self-stabilization with neighborhood stabilization	HHt→HHt+1, LHt→LHt+1, HLt→HLt+1, LLt→LLt+1

Note: HH means a city and its neighbors all have high values for an indicator (i.e., RCL growth rate in this study), HL means a city has high value but its neighbors have low value, etc.

[47].

The transition between HL and LH is a heterogeneous transition, whereas the transition between HH and LL is a homogeneous transition. The spatial homogeneity trend indicator (SHTI) is defined as

$$\mathrm{SHTI}=\frac{H_m-H_t}{2}$$

(7)

where H_m and H_t represent the probabilities of a heterogeneous-to-homogeneous transition and a homogeneous-to-heterogeneous transition, respectively. SHTI ranges between −1 and 1. If SHTI is greater than 0, the spatiotemporal transition has a homogeneous trend; if SHTI is less than 0, the spatiotemporal transition has a heterogeneous trend.

(4)

Correlation network

A spatiotemporal network is also known as a spatial coevolution or a spatiotemporal interaction visualization. The covariance coefficients of the RCL in neighboring cities from 1990–2017 were calculated. The correlation intensity of the change in RCL in neighboring cities was shown by connecting lines in different colors. The association strengths corresponding to different ranges of correlation coefficients (

Table 3. Coevolution strength.

ID	Correlation Coefficient	Coevolution Strength
1	−1 to −0.5	Strong decoupling
2	−0.5–0	Weak decoupling
3	0–0.5	Weak coevolution
4	0.5–0.9	Moderate coevolution
5	0.9–1.0	Strong decoupling

) were defined with reference to a previously published work [51].

2.3.4. Multiscale Geographically Weighted Regression (MGWR)

MGWR was used to explore the spatial distributions of the influences of different factors on RCL. MGWR is a nonparametric local spatial regression method. MGWR can be used to observe the spatial heterogeneity of the effects of different factors influencing RCL to facilitate the discovery of spatial patterns of influence [31,52,53]. The MGWR calculation model is expressed as

$$y_i={\displaystyle \sum _1^k{\beta }_{b\omega j}({\mu }_i,{\upsilon }_i)x_{ij}+{\epsilon }_i}$$

(8)

where y_i is the RCL value; x_ij is i determinant j in area i; β_bwj (µ_i, v_i) is the coefficient of area i for location (µ_i, v_i), where bwj is the optimum bandwidth of determinant j; ε_i is a mutually independent random error. The estimate can be described as

$${\beta }_{b\omega j}({\mu }_i,v_i)=\left(X^{\prime }{}_j{W}_{b\omega j}\right({\mu }_i,v_i\left)X_j\right){}^{-1}{}_j{W}_{b\omega j}({\mu }_i,v_i)y$$

(9)

where β_bwj (µ_i, v_i) is the spatial weight, and y is the observation of RCL. In the MGWR model, each determinant has its own spatial scale, and the optimum bandwidth is detected by the corrected Akaike information criterion. The study sample is 41 cities in the YRD region.

3. Results

3.1. Spatiotemporal Dynamics of RCL in the YRD Region

3.1.1. Spatial and Temporal Patterns of RCL in the YRD Region

(1)

Global features

The center of gravity of RCL distribution in the YRD region continued to move westward from west of Yangzhou in 1990 to near the border of Nanjing and Chuzhou in 2017 (Figure 4a). The western part of the YRD region is mostly under the jurisdiction of Anhui Province; therefore, the change in the RCL area in Anhui Province has accelerated. Zhejiang Province is located in the southern part of the YRD region, where the RCL area is much smaller than in Anhui and Jiangsu Provinces.

The global Moran’s I can be used to determine the spatial autocorrelation of the change in RCL. Moran’s I was positive and gradually increased from 1990 to 2017, although there was a slight decrease after 2012; thus, rural growth showed a significant positive correlation among cities (Figure 4b). The Gini coefficient decreased over time, indicating that the difference in the change in RCL between cities decreased slowly during the study period.

(2)

Spatial and temporal patterns of changes in RCL

Based on previously published research, the cities were classified into four categories based on the rate of growth in RCL [54]: decreasing (<0%), low growth (0–5%), medium growth (5–10%), and high growth (>10%).

From 1990 to 2000, RCL in the YRD region showed a pattern of low growth in the central cities and high growth in the peripheral cities (Figure 5). Among cities, Xuzhou, Suzhou, Wuxi, and Suqian showed the highest growth in RCL; the average growth rate in these cities exceeded 19%, adding a total of 748 km² of new RCL and accounting for 19% of the total growth in the study area. Ma’anshan, Nanjing, Xuancheng, and other central cities had a slower rate of RCL growth. Only Ningbo experienced a negative growth rate (decrease in RCL).

The decade from 2000–2010 was a period of low RCL growth in the YRD region, without any area of rapid increase in RCL. Only 14 cities in southern Jiangsu Province, northern Zhejiang Province, and eastern Anhui Province showed medium rates of growth in RCL. Among the cities, Lu’an recorded the fastest RCL growth rate of 8.14%. The area of negative RCL grown extended southward from Ningbo to include Zhoushan, Shaoxing, Taizhou, Jinhua, and Wenzhou, with an average rate of decrease of 2.87%.

The growth rate of RCL in the YRD region recovered between 2010 and 2017. Medium growth rates were observed in the eastern and southeastern cities, whereas the growth rate was high in Huangshan (Anhui Province) and Quzhou (Zhejiang Province). Reductions in RCL were not observed in any part of the study area during this time period. With the improvement in the living standards of rural residents and the national “village access” project, land for rural roads and residential land have increased rapidly. The village access project is a systematic project in China that aims to construct cement roads (or other type of roads with hardened surfaces) in every village and to connect them.

3.1.2. Dynamics of the Local Spatial Dependence of RCL

The numbers of cities for which the relative length of the LISA time path of RCL was greater than 1 during the three study periods were 14, 19, and 16, respectively, accounting for more than one-third of the total cities (Figure 6). This shows that the local spatial dependence of RCL growth was strongly dynamic. The cities with longer relative paths were mainly located in Jiangsu Province and Shanghai; this group also included Anhui Province after 2000. From a global spatial perspective, the spatial dynamics of the change in RCL were significantly stronger in the northern part of the study area compared with the southern part. On a provincial scale, the dynamics were strongest in Jiangsu Province and weakest in Zhejiang Province.

The tortuosity of most cities was low during both 1990–2000 and 2000–2010, indicating that the change in RCL was stable in most cities of the YRD region (Figure 7). The spillover effect and the fluctuation in the growth of RCL were not significant in most cities. There was a large change in tortuosity from 2010 to 2017. The regional distribution of tortuosity showed strong spatial agglomeration with higher volatility in the eastern part of the study area than in the western part.

3.1.3. Local Clustering of Changes in RCL

The position of the sample in the LISA coordinate system can be calculated to obtain the spatial distribution of LISA clusters according to the local Moran’s I (Figure 8). The cluster types of RCL change in the YRD region were dominated by high–high and low–low clusters. High–high cluster indicates a rapid growth in RCL in an area and its neighboring areas, which are referred to as hotspots. In contrast, the low–low cluster means low or even negative growth in RCL in an area and its neighboring areas. During the study period, the high–high cluster shifted from Suqian and Lianyungang in the northern part of the study area to Quzhou and Hangzhou in the southern part. In contrast, the low–low cluster showed the opposite pattern, gradually shifting from the south to the north. Overall, the relatively economically backward cities tended to form clusters with a high rate of RCL growth, whereas economically advanced cities were more likely to form low–low clusters.

The LISA spatiotemporal transition matrix shows the characteristics of local spatial dependent shift. This matrix is a visual representation of the dynamic shift in LISA coordinates in the Moran scatter plot [31]. The change in agglomeration is further described on the basis of the local spatial characteristics. This change is reflected by transitions between quadrants in the LISA coordinate system. As shown in

Table 4. Spatial and temporal transition matrix of RCL.

Time Period	t/t + 1	HH	HL	LH	LL	Type		SHTI
1990–2000	HH	0.238	0	0	0	I	0.023	0.023
	HL	0.023	0.071	0	0	II	0.023
	LH	0.023	0	0.095	0	III	0
	LL	0	0	0	0.523	IV	0.954
2000–2010	HH	0.285	0	0	0	I	0.023	0.012
	HL	0.023	0.047	0	0	II	0.046
	LH	0.023	0	0.071	0	III	0
	LL	0	0	0.023	0.5	IV	0.931
2010–2017	HH	0.357	0	0	0	I	0.07	−0.001
	HL	0	0.023	0	0	II	0.023
	LH	0.023	0	0.047	0.023	III	0
	LL	0	0.047	0	0.452	IV	0.907

, the area located in the third quadrant (LL) accounted for the largest proportion of the total area (approximately 50% on average), representing the low–low cluster with the adjacent cities all having low values. There was a gradual decreasing trend for this kind of city cluster over time. The first quadrant (HH), which represents the high–high cluster with the adjacent cities all having high values, accounted for the next largest proportion of the total area; the proportion of agglomeration gradually increased. A clear spatial polarization of the change in RCL was observed in the YRD region. Type IV (see Table 1) transition was the most abundant, accounting for more than 90% of the regional share. No Type II transition, which indicates a large difference between a city and its neighbors, was observed, indicating that the spatial dependence of rural growth between a city and the surrounding cities tended to remain stable in the short term. SHTI denotes a shift from heterogeneity (HL and LH) to homogeneity (HH and LL) during the transition. The SHTI was greater than 0 in both 1990–2000 and 2000–2010, suggesting a homogeneous trend for the spatiotemporal transition. However, during 2010–2017, the SHTI was less than 0 and the local dependence of rural economic growth tended to be heterogeneous. This also reflects a decrease in the proportion of Type IV transitions and SHTI values less than 0 over the three time periods.

3.1.4. Inter-City Competition for RCL

The spatial synergistic network of RCL characterizes the dynamics of inter-city competition for RCL. The inter-city RCL relationships were dominated by synergistic development and weak competition (Figure 9). However, this collaborative development has weakened over time, and the number of cities with medium and weak synergy have gradually increased. From 1990–2000, 69 groups of cities showed strong synergy, and 19 groups of cities had medium and weak synergy. From 2010–2017, the groups of cities with strong synergy was 57, while 24 groups of cities had medium or weak synergy. The additional groups of cities with medium and weak synergy were distributed in the southern Jiangsu and Zhejiang provinces. There was no significant change in the synergistic relationships in cities with relatively low economic levels, namely Anhui Province. Weakly synergistic or competitive relationships were found between neighbors in this area. The spatiotemporal network of RCL provides a visualization of the degree of synergistic development between each sub-region and neighboring cities in the study area and clarifies the synergistic relationship between each city and its neighboring cities. The maps shown in Figure 8 can be used to help formulate development plans according to the local conditions and promote regional integration.

3.2. Determinants of RCL

3.2.1. Baseline Results

To explore the determinants of RCL, regressions were conducted using a stationary panel regression model with RCL as the dependent variable for each city. According to the results in

Table 5. Results of panel regression.

Adj-R2	X1	X2	X3	X4	X5	X6	X7
0.697	0.219 ** (2.26)	−0.167 (−1.06)	0.728 *** (6.43)	0.011 (0.26)	−0.256 (−1.27)	0.418 ** (3.25)	−0.819 *** (−4.87)

** p < 0.05; *** p < 0.01.

, the GDP of primary industry (X₁), area of arable land (X₃), elevation (X₆), and slope (X₇) were significantly correlated with RCL, whereas rural population (X₂), accessibility (X₄) and GDP (X₅) were not significantly correlated with RCL.

The GDP of primary industry reflects the income of rural farmers, especially in the early stages of industrialization. The significant positive correlation between the GDP of primary industry and RCL indicates that an increase in farmers’ income will promote the RCL. The area of arable land is also closely related to farmers’ income. In areas rich in arable land resources, the restrictions on land development and construction will be relaxed to increase farmers’ income. A significant positive correlation was observed between the area of arable land and RCL. Elevation was significantly and positively correlated with RCL, likely because areas that are relatively high in elevation (e.g., mountainous areas) are not suitable for cultivation and have few restrictions on development and construction. In contrast, slope was significantly and negatively correlated with RCL because large slopes make the land less suitable for building and development.

There was no significant correlation between RCL and the rural population because the rural population and household income did not change regularly during the study period. A large number of people from rural areas moved into the cities for work as a result of the rapid urbanization that occurred after 2000. The rural population therefore decreased significantly after 2000. However, the increased income brought in by migrant workers allowed farmers’ unions to build housing and other developments, resulting in an increase in RCL. One possible explanation is that the decrease in RCL caused by the decrease in the rural population and the increase in RCL caused by the increase in income caused by migrant workers’ going out to work offset each other, so the rural population index is not significant.

The insignificant relationship between GDP and RCL and accessibility may be because the rural population normally migrated to work in minority big cities, and after they earned money, they went back to hometown to build houses which affect local RCL.

3.2.2. Spatial and Temporal Variations in the Effects of the Determinants

To further explore the effects of the determinants on RCL, we used MGWR to analyze the spatial and temporal differences in the strength of the effect of each determinant.

(1)

GDP of primary industry

As shown in Figure 10, the correlation between the GDP of primary industry and RCL presented a significant east–west zonation. The correlation coefficient gradually decreased moving from the coastal cities to the inland cities, indicating that the effect of the GDP of primary on RCL weakened moving from the coast to inland. Temporally, the average proportions of GDP provided by primary industry in 2000, 2010, and 2017 were 18.6%, 10.1%, and 7.5%, respectively, indicating a gradual decrease over time. This indicates that the contribution of primary industry to the GDP of each city along with the effect on RCL gradually decreased as economic development increased.

(2)

Area of arable land

As shown in Figure 11, the area of arable land was positively correlated with RCL. Arable land is an important land resource. The dependence of rural residents’ income on arable land decreased from 2000 to 2017 as a result of rapid economic development; therefore, the effect of arable land on RCL weakened during this time period. The effect of arable land on RCL gradually weakened moving from north to south in 2000. But in 2017, this effect weakened from the coastal to inland areas. It could be attributed to the fact that as the economy developed, rural population were less dependent on the arable land.

(3)

Elevation and slope

Elevation was positively correlated with RCL (Figure 12). The effect of elevation on RCL was greatest in mountainous areas such as Anqing, Huangshan, and Quzhou in the western and southern parts of the YRD region. Temporally, the effect of elevation on RCL gradually became weaker over time. Because increasing the slope makes construction more difficult, slope was negatively correlated with RCL (Figure 13). Over time, the negative correlation coefficient between slope and RCL gradually decreased, indicating that the effect of slope on RCL became weaker over time.

(4)

Rural population

The rural population and RCL showed an unstable correlation. As shown in Figure 14, the rural population was positively correlated with RCL in the YRD region in 2000. The positive correlation showed a significant east–west divergence, with the effect of the rural population on RCL gradually weakening by moving from the coast to the inland area. In 2010, the rural population in the YRD region was negatively correlated with RCL; the strongest negative correlation was found in Anhui Province, where the phenomenon of rural hollowing out was pronounced. In 2017, the rural population and RCL in the YRD region were positively correlated, and the spatial distribution of the effect was basically the same as in 2000, although the correlation strength was greatly reduced.

4. Discussion

4.1. Mechanisms of the Effects of Determinants on RCL

The GDP of primary industry, the area of arable land, the elevation, and the slope all significantly affected the RCL. The influence of each factor on RCL decreased year by year with the progression of economic development and urbanization. Spatially, it shows the gradually weakened role of coastal–inland factors.

(1) In general, the primary industry affects the income of rural residents and the ability of the rural population to engage in construction. The results indicate that the dependence of farmers’ income on primary industry decreased from the coast to inland areas. This is because more people in inland areas go out to work, making them less dependent on primary industry for income. It is also the case for economy. since the rural population normally migrates to other developed regions to work, their income has little relation with the local economy of their hometown, so the GDP is also insignificantly correlated with local RCL.

(2) The area of arable land has a strong influence on RCL in the northern and eastern parts of the study area and a weak influence in the southern and western parts. The northern areas of the YRD region are richer than the southern areas in terms of topographic conditions and land quality. In general, areas rich in arable land tend to be more favorable for construction; therefore, the correlation between RCL and arable land is stronger in the northern areas. Since 2000, the demand for land has greatly increased due to the need for economic development in coastal areas; thus, the RCL was significantly affected by the amount of available arable land. The conversion of arable land to RCL occurred in inland areas to a lesser extent than in the more developed coastal areas.

(3) Elevation is positively correlated with RCL, which shows that the elevation does not limit the expansion of RCL. This can be explained as follows. Mountainous areas are normally classified as rural areas, and much of the construction land in plains with low elevation is classified as urban construction land. The influence of elevation on RCL gradually weakened over time, which may be related to the completion of construction projects (e.g., residential houses, roads).

(4) Slope often directly determines the difficulty of construction, but the effect of slope on RCL weakens significantly over time, which is due to the progress of technology that helped overcome natural constraints [53]. Spatially, the effect of slope on RCL becomes weaker from coastal areas to inland areas. This may be because there are more mountainous areas in inland locations, causing more facilities (e.g., houses, roads) to be built on relatively high slopes.

(5) The relationship between RCL and rural population was not stable during the study period. In 2000 and 2017, the two were positively correlated, whereas they were negatively correlated in 2010. Because of the accelerated urbanization in China since 2000, numerous rural residents moved to cities; however, their household registration and some family members remained in the countryside [55]. When they earned money, they chose to build houses and purchase properties in the countryside, leading to a continuous increase in RCL and decrease in population. This phenomenon was most obvious in the most undeveloped areas. This is consistent with previous studies that found urbanization leads to the phenomenon of “hollow villages” [5,56,57]. After 2010, the rate of urbanization slowed significantly, corresponding to a slowing down in the flow of rural residents to cities; thus, the relationship between the rural population and RCL returned back to a positive correlation.

4.2. Other Factors Influencing RCL

Li’s study shows that the RCL in Zhejiang Province is increasing rapidly [28]. However, this study finds that, in fact, the cities in the southeast of Zhejiang Province showed a decrease in the rural construction land. As a result of economic development, some rural areas are counted as urban areas, which may have caused the construction land in these areas to be counted as urban construction land rather than RCL. For example, Yiwu in southern Zhejiang Province is the first county-level city in China. The change in status of Yiwu from a county to a county-level city may cause its RCL area to be counted as urban construction land. In addition, the consolidation of rural settlements is also an important reason for the decrease in RCL in some areas. However, the decreases in RCL were only limited to specific areas at specific times and did not affect the overall trend of rapid growth in RCL.

4.3. Policy Implications

These findings serve as a guide for Chinese policymakers. The increase or decrease in construction land is generally in line with the population. As the rural population decreases, the RCL should also gradually downsize according to traditional planning theory. However, the RCL in many areas increased while the rural population decreased. In view of the reality of the YRD region, we propose to carry out the consolidation of construction land in rural areas, clear idle and wasted construction land through measures such as confirmation and registration, and strictly implement the policy of linking the increase in urban construction land to the reduction in RCL. Due to the differences in regions and environments, the planning according to the local condition should be introduced to improve the efficiency of RCL utilization [5].

In addition, the results also indicate that the integrative development in the YRD region needs to be strengthened. On the one hand, it is necessary to gradually reduce the development gap between regions and between urban and rural areas and rationally guide the scale and speed of rural population flowing into cities. On the other hand, it is necessary to gradually improve the factor market, especially the unified construction land market in urban and rural areas, allow RCL to enter the market, and promote the optimal allocation of rural land resources.

4.4. Contributions and Limitations

By taking the YRD region as a case study, this research explores the spatiotemporal dynamics and determinants of RCL in China’s developed areas. The results clarify the spatial patterns of changes in RCL over time, the spatial dependence of RCL, and the mechanisms of each RCL determinant based on long-term time-series data. The findings not only help optimize local rural land use, but also provide a scientific reference for rural land use planning in other regions. The findings also enrich the macro-scale research focusing on developed areas.

This study also has some limitations. On the one hand, although the indices selected cover the economy, transportation, and nature, the number of factors in each category is limited and remains to be enriched. Furthermore, policies and historical and cultural factors are different between cities, so it is necessary to consider these factors in in-depth studies in the future. On the other hand, land is publicly owned and there is no private land in China, which is different from Europe and the United States and other countries. There are two forms of public land in China, namely urban state-owned land and rural collective land. There is no unified land market between urban and rural areas. In view of this, we did not carry out a comparative analysis of RCL with other countries in the paper.

5. Conclusions

Focusing on the RCL changes in economically developed areas of China, we analyzed the spatiotemporal characteristics and influential factors of RCL by taking the YRD region as a case study. The study used the rural impervious surface data of 41 prefecture-level cities from 1990 to 2017 and socioeconomic and natural factors. The main conclusions are as follows.

Firstly, the RCL in the YRD region was in a state of growth from 1990 to 2017. During this time, the RCL growth rate first increased rapidly, then decreased but finally increased again. The average annual growth rate is 5.38%, and the total increase in the RCL area is 14,768.3 km². The spatial autocorrelation of the RCL growth in the YRD tends to increase over time, whereas the regional differences tend to decrease.

Secondly, the RCL change in the prosperous eastern coastal area of the YRD is more pronounced and fluctuant than that in the western inland area. However, the spatial linkage of the RCL growth between cities of the YRD is weak, showing that the integrated development needs to be strengthened.

Thirdly, the factors that significantly affected the RCL in the study area include the GDP of primary industry, the amount of arable land, the elevation, and the slope. The first three factors have positive correlations with the RCL, whereas the slope has a negative correlation with the RCL. Clear spatial differences can be observed in the effects of these determinants across the study area. In addition, there is no clear correlation between the RCL and the local economic development, as the rural population normally goes to few big cities for higher salary work but spends the money in their hometowns on building homes. It is also the case for the indicator of local accessibility.