1. Introduction
The land covers and landscapes in many of China’s cities have been modified significantly in recent years due to the rapid economic development and population growth. Nowadays, more than 50% of China’s population lives in cities, and this rate will continue to rise to 75% by 2030 [
1]. For this reason, urbanization has become one of the most important factors affecting ecosystem services and environmental quality, especially in some emerging cities or city clusters [
2,
3]. To address the adverse impacts and promote sustainable urban development, the Chinese government has issued the “New Urbanization Plan” with the intent to emphasize ecological progress, urbanization quality, domestic demand expansion, and rural–urban coordination simultaneously [
4,
5]. To achieve these goals, it is necessary to monitor and understand urban expansion continuously [
6]. Mapping urban growth accurately is a prerequisite for studying and managing the urbanization process, revealing the drivers, and evaluating the consequences [
7,
8,
9,
10].
Co-evolution, the process of reciprocal evolutionary change that occurs between cities or within a city as they interact with each other [
11,
12], has been regarded as the key step for a city to become a megalopolis [
13,
14]. The international megalopolises in developed countries, such as the Northeast megalopolis in USA [
15], the Great Tokyo Area in Japan, and other megalopolis [
16], all have experienced long co-evolution processes to integrate multiple cities, gradually differentiating mutually beneficial and coordinated functions, and became the regional, national, and ultimately international economics centers [
13]. The development of satellite remote sensing and geographic information system technologies has promoted the fascinating possibility of the analysis of the urbanization and co-evolution of multi-cities [
17,
18,
19], especially in emerging cities in developing countries such as India, South Africa, and China, where there are relatively fewer references for sustainable urban development [
10]. Previous studies on the spatiotemporal changes of urban land use and land cover (LULC) in China have largely focused on the Jing-Jin-Ji Megalopolis, Pearl River Delta Megalopolis, and Yangtze River Delta Megalopolis [
10,
20,
21,
22,
23]; less attention has been paid to urban expansion in emerging cities in the land-locked regions, calling for more research and understanding of the processes and consequences of urban development in inland [
24,
25,
26]. Besides, landscape ecology approaches and metrics have been widely used for analyzing the urbanization process, which could support a better understanding of the evolution of a city [
20,
27,
28,
29].
The temporal resolution of remote sensing data plays a crucial role in studying the co-evolution of multi-cities [
30,
31]. However, time intervals of 5 to 10 years have often been used to map urban expansion in most studies [
27], which is apparently not suitable for mapping and monitoring the expansion processes in fast-growing cities where the annual urban expansion can exceed 5% [
20,
32]. Some small but important changes may be ignored in remote sensing monitoring over a large time interval, resulting in barriers analyzing fast transformations between the different LULC types, urbanization processes, and driving forces in emerging cities [
20,
33,
34]. All of these were mainly due to the limitations from available remotely sensed data [
7,
35,
36]. Specifically, images with inferior quality were unusable due to their low frequency and quality (i.e., seasonal variation and excessive cloudiness) [
37], leading to most urban monitoring studies being based on a comparison of a classical supervised classification between two dates, instead of a proper time series algorithm [
37,
38]. The rapid developing cities, with great inter-annual changes, may require continuous time series models to reveal their series of urban expansion processes [
22,
23,
39,
40].
An efficient and stable remote sensing data processing method could integrate a volume of images, which would be meaningful in monitoring the urbanization process and analyzing the co-evolution of multi-cities [
41,
42,
43]. Several methods have been developed in recent years to take advantages of time series of remotely sensed images for LULC detection, including the Landsat-based detection of Trends in Disturbance and Recover (LandTrendr) [
44,
45], the vegetation change tracker (VCT) [
46], and the Continuous Change Detection and Classification (CCDC) algorithm [
17,
33,
47,
48,
49,
50]. These methods are designed to analyze the changes of ground objects through different mechanisms. LandTrendr detects abrupt change by the segmentation method and, between the abrupt changes, a slope is fitted for each segment to capture the gradual changes, and this approach has been used for insect infestation detection and forest change detection [
44,
47,
51,
52]. The VCT normalizes each Landsat image into a forest probability index and uses a thresholding method to detect forest disturbance. These two methods are usually used to monitor forest dynamics and rarely used to monitor urban expansion because limited indices could not support the classification of many land cover types [
53]. The CCDC algorithm, on the other hand, uses all spectral bands in Landsat images to detect many kinds of surface change by fitting multiple surface reflectance models with sines and cosines simultaneously [
54], and has been applied to study the changes in vegetation, impervious surface expansion, and surface temperature dynamic after land use change [
55,
56,
57].
The Chang–Zhu–Tan urban agglomeration (CZTUA), located at the middle reach of the Yangtze river basin, is the fastest growing region in Hunan Province, showing the typical developing pattern that can be currently found elsewhere in China [
26,
58]. The development of CZTUA has benefited from a series of urban development policies such as “The Rise of Central China Plan” in 2004, “Resource saving and Environment friendly” dual-type society experimental areas in 2008, and “Middle Yangtze River Urban Agglomeration Planning” in 2015 [
26,
59,
60]. These urban development policies, which promoted the prosperity of provincial cities and poorer neighbor regions, have been among the key drivers of the co-evolution of multi-cities [
28]. Therefore, a comprehensive understanding of urban expansion in CZTUA is not only a prerequisite for comprehending the urbanization process and economic growth, but also the basis for better urban planning. We focused on the urban expansion dynamics in CZTUA during 2001 to 2017 when the majority of the development urban policies were formulated and carried out. The objectives of this study were to (1) map the annual LULC and landscape dynamics using the Landsat data and CCDC; (2) quantify the spatiotemporal patterns of urban expansion on the regional and city scales; and (3) analyze the co-evolution pattern, driving forces, and implemented policies of multi-cities.
4. Result
4.1. Accuracy of the LULC Classification
For the entire study area, we selected five years of the LULC maps (i.e., 2001, 2005, 2010, 2015, and 2017) for our spatiotemporal accuracy assessment. Validation samples were the remaining 10% of the samples. The classification results achieved an overall accuracy of 90.44%–92.31% across these five years (
Table 2). Our results show that the highest overall accuracy (92.31%) was in 2001, followed by 92.20% (2005), 90.89% (2010), 90.63% (2017), with the related lowest one being 90.44% in 2015.
The classification results of the cropland, forest, and water classes were more accurate, while those of grassland and wetland were relatively poor. The difference in classification accuracy of different land covers may be due to the fact that the number of training samples for the grasslands and wetlands were less than those of the other land-cover types, typical for rare land cover types [
82,
83]. Besides, seasonal variations may also lead to the related low classification accuracy. For example, the probability of a misclassification between a cropland and forest was still high, and the grasslands and wetlands were confused easily. Nevertheless, the small differences in overall accuracy among the different years suggest that the CCDC model performed well in mapping the time series of LULC. So, the results could meet the requirement for monitoring and analyzing the urban expansion process.
Figure 3 shows the Landsat images and resultant land cover maps in 2001 and 2017, respectively, in a typical area.
4.2. Temporal LULC Change from 2001 to 2017
Based on the yearly classification maps, the land cover in CZTUA has significantly and rapidly changed in the past 17 years (
Figure 4). In general, the impervious surface area increased while the cropland and forest decreased continuously, and the water area, wetland, and grassland remained relatively stable. The land cover was composed of cropland of 5297 km
2, forest of 4355 km
2, grassland of 12 km
2, wetland of 152 km
2, water of 344 km
2, and impervious surface of 808 km
2 in CZTUA in the year of 2001 (
Table A1). Overall, an increase of 371 km
2 in impervious surface, and a decrease of 169 km
2 and 206 km
2 in cropland and forest, respectively, were observed from 2001 to 2017 in the study area (
Figure 4). Most of the impervious surface was gained from both forest and cropland and the transformation from wetland, water, and grassland was not obvious (
Table A2).
The impervious surface area increased from 808 km
2 to 1179 km
2 over the past 17 years with the net increase rate of 45.90% (
Table A1). There was large spatial variation in the area of impervious surface increase among subregions: from 211 to 246 km
2 (16.59%) for Changsha City; 145 to 227 km
2 (56.55%) for Changsha County; 126 to 217 km
2 (72.22%) for Wangcheng District; 82 to 106 km
2 (29.27%) for Zhuzhou City; 31 to 64 km
2 (106.45%) for Zhuzhou County; 63 to 94 km
2 (49.21%) for Liling City; 83 to 103 km
2 (24.10%) for Xiangtan City; 57 to 110 km
2 (92.98%) for Xiangtan County; and 9 to 13 km
2 (44.44%) for Shaoshan City (
Table A4).
Figure 5 demonstrates the annual increment (AI) and annual growth rate (AGR) of the impervious surface for nine regions over the past 17 years. The AIs of impervious surface for Changsha County in the year of 2006, Wangcheng District in the year of 2011 and 2012, and Xiangtan County in the year of 2012 were the highest, which were greater than 10 km
2 in the observation. A very small decrease in impervious surface was also presented in Changsha City during the years from 2013 to 2015. The relatively low AIs of impervious surface were appeared in Shaoshan City, which was lower than 0.1 km
2 for most of the time. The AIs in the rest of the regions of CZTUA were increasing steadily. The highest AGR was presented in the Xiangtan county in 2012. AGRs in Changsha City, Xiangtan City, and Zhuzhou City became negative during certain periods. Zhuzhou County, Xiangtan County, and Wangcheng District had a relatively high AGR among the nine regions (
Table A3).
4.3. Spatiotemporal Dynamics of Urban Expansion
CZTUA demonstrated great spatial variability in urban development from 2001 to 2017 (
Figure 6). Changsha City, Xiangtan City, and Zhuzhou City are the belt-shaped cities that were initially developed along the river. In recent years, urbanization in the southern part of Changsha City, the northeast of Xiangtan City and the northwest of Zhuzhou City has been active, showing a trend of triangular convergence (
Figure 6c). For Changsha County and Wangcheng District, the primary core has been constantly expanding outward, especially in the eastern part of Changsha County and western part of Wangcheng District (
Figure 6a,b). Meanwhile, the urban area of Zhuzhou County and Xiangtan County expanded rapidly after the year 2010, resulting in a doubling growth. Shaoshan City and Liling City were developmentally restrained.
Figure 7 shows the dynamics for the proportion of three urban growth types (i.e., leapfrogging, edge-expansion, and infilling) of the newly developed impervious surface patches in CZTUA during the study period. It could be seen that there was large spatial and temporal variability during the urban expansion processes across regions. Over the entire study period, infilling was the dominant growth type with a steady trend to decrease, possibly as the gaps became filled. On the other hand, leapfrogging had a marginal occurrence at the beginning of the time series, to later become the dominant urban growth type in 2016. Edge-expansion remained more stable throughout the time series. The proportion of infilling reduced steadily from about 80% in 2001 to 40% in 2016 in Changsha City, while the proportion of edge-expansion increased smoothly from about 20% to 40% and the share of leapfrogging increased from about 0% to 10% (
Figure 7). In 2016, leapfrogging increased sharply in Changsha County and Wangcheng District. Most of the newly developed urban land in Changsha County concentrated around the existing urban area and is distributed with the newly developed highways and airport in the region (
Figure 3b).
4.4. Landscape Changes during Urban Expansion
Figure 8 shows the annual changes in landscape metrics in CZTUA. The PLAND and LPI increased slowly for all nine regions during whole period. Changes in LPI suggested the proportion of the largest urban patch (the existed impervious surface) in each region. Specifically, Changsha City presented the highest proportion in PLAND and LPI with a steady increased rate among the nine regions, while the rest of the regions showed a similar pattern in PLAND and LPI though the values were lower than those for Changsha City. The LPI in Xiangtan City, Shaoshan City, and Wangcheng District has greatly increased from 2007 to 2010.
The landscape metrics PD and LSI described the complexity and fragmentation for the impervious surface. The PDs in Wangcheng District, Changsha County, and Zhuzhou City were relatively higher than those in other regions, with a slow decline in the early stage and rapid growth in the latest period. The remaining regions had a relatively lower PD and could form a different group. Changsha City was different from all other regions by presenting a steady decrease in PD over the past 17 years. Contrarily, Liling City, Xiangtan County, and Zhuzhou County increased in the latest period and had no significant changes in the earlier stage. In the total area of CZTUA, PD changed a little during the observation and kept a relatively low value. For the LSI, CZTUA had the highest LSI with the U-shape function changed. In this study, the LSI was lower than 150 in the past 17 years. According to the value of the LSI, the cities in CZTUA could be roughly divided into three groups, the first group of which were greater than 90, including Changsha County, Shaoshan City, Liling City, and Xiangtan County; for the second group, which includes Zhuzhou County, the LSI was between 60 to 90; the last group, with an LSI lower than 60, includes Zhuzhou City, Changsha City, Xiangtan City, and Shaoshan City. Most of regions in CZTUA were under a steady and similar pattern in LSI. The IJI and COHESION metrics represent aggregation and connectivity. IJI in Changsha City was the highest throughout the period, rising from 50 in 2001 to just under 70 until 2013, when there was a levelling off for three years. The remaining regions presented a similar pattern, which decreased first and then increased or kept the value without change for IJI. Wangcheng District and Changsha County had a reverse U-shape during this observation. For COHESION, Changsha City, Xiangtan City, Zhuzhou City, Wangcheng District, and Changsha County had a relatively higher value, which was close to 100, and presented a steady value or slight increase from 2001 to 2017. Xiangtan County, Shaoshan City, and Liling City increased steady before 2009 and stabilized in the latest period, while Xiangtan County, with a little bit of hysteresis, persisted by increasing up to 2013. Zhuzhou County presents sustainable growth in COHESION.