Luxury Effect, Heritage Effect, and Land Use Hypotheses Revealing Land Cover Distribution in Hainan Island, China

Abstract

Land cover analysis is a key method used to understand past land use patterns and explore the driving forces and processes behind them. This study focuses on land cover in 18 counties and cities of Hainan Island, delving into the driving factors of land cover in specific areas of Hainan Island, including the southern, northern, eastern, western, inland, and coastal regions. The effects of socio-economic factors, landscape pattern, and urban functional units on land cover are considered, and three hypotheses are proposed to explain the observed trends. The results indicate that house prices are positively correlated with construction area, woodlands land, and urban green space, thereby supporting the luxury effect hypothesis on land cover. In addition, construction age was negatively correlated with the woodlands area, confirming the role of the inverse legacy effect hypothesis in land cover. Other relationships between urban functional units and land cover emphasized the importance of the land use hypothesis in land cover planning. These results will help decision-makers and managers to better understand the current drivers of land cover, allowing for a more scientific basis when planning and managing urban land cover.

1. Introduction

Land cover (LC) refers to the coverage of various natural and anthropogenic features across different regions and territories of the Earth’s surface, reflecting the vegetation, water bodies, buildings, and other physical characteristics present in different areas [1]. Changes in LC directly reflect the impact of humans on land resources. The importance of LC was emphasized in the August 1992 Annual Meeting of the American Academy of Ecology, where a panel discussion was devoted to highlighting key research questions on LC change [2]. A few years later, in 1995, the International Humanities Programme (IHDP) and the International Geosphere Biosphere Programme (IGBP) jointly published a scientific study of LC changes, opening up this new area of research [3]. Historically, LC was mainly affected by natural factors, but recent technological advancements and continued increases in human activities have led to greater anthropogenic influence, resulting in dramatic LC changes in some areas [4,5]. Urbanization has provided significant opportunities for economic and social development [6,7,8]. However, irrational and inefficient land development practices have led to adverse consequences, such as the shrinking of ecological green spaces and the urban heat island effect [9,10]. Therefore, to promote sustainable urban development, the key lies in the rational utilization of land resources [9].

The LC of a growing city constantly shifts with changes in the city’s economy, development time, population density, and urban planning decisions [11]. To better understand the heterogeneity of urban landscape changes due to variations in LC and their impacts on urban ecosystems, as well as to predict future LC changes, it is imperative to thoroughly study the factors influencing LC. Several hypotheses have been proposed, with the most widely recognized being the luxury effect hypothesis [12,13], the legacy effect hypothesis [12,14,15,16], and the land use hypothesis [17].The luxury effect hypothesis proposes that urban residents with higher socioeconomic status have a greater ability to allocate resources to vegetation and habitat building/preservation, thus affecting the overall dynamics of an integrated geological survey system [12,13]. In addition, wealthier residents may have greater lobbying power and access to decision-makers (both public and private investors), often leading to greater influence and agency over public and private decisions, including community LC planning and investment, thus influencing LC planning and resource allocation [13,18]. Their political connections, social capital, knowledge, and access to information further enhance their ability to advocate for their own interests, which can ultimately shape urban LC over time [14,19]. Another hypothesis is the legacy effect hypothesis, which suggests that current LC patterns in urban areas are influenced by the legacy of past land policies, including urban green space (UGS) management practices and urban planning strategies [12,14,15,16]. This is most obvious in older neighborhoods (i.e., areas with older housing and urban development) that have higher UGS (including woodlands and grasslands), reflecting the long-term trajectory of management practices [15,19,20]. The final hypothesis considered in this study is the land use hypothesis. It emphasizes that different functional areas in a city have different land use distributions [17]. For example, business districts in cities often choose to prioritize accessibility, visibility, and a high population density during their development, as this is conducive to attracting customers and developing business activities [21,22]. Accordingly, land use in commercial areas tend to be dominated by large commercial buildings and facilities, such as wholesale markets and supermarkets [23]. In contrast, residential areas are more inclined to prioritize comfort and good environmental quality, so the LC model of residential areas often includes residential buildings and public green spaces for residents [24]. However, excessive population density usually negatively affects both business districts and residential areas through an increase in congestion diseconomies [25,26]. Therefore, the LC distribution of a city should take into account regional development conditions comprehensively, aiming to improve the living environment and promote the sustainable development of urban areas. Studying these hypotheses allows for a better understanding and prediction of the driving factors behind land use decisions, providing theoretical support and empirical evidence for formulating land management policies and planning [12,14,17].

Rapidly urbanizing areas produce highly fragmented ecosystems, resulting in landscapes consisting of heterogeneous, complex, and discontinuous patches [27]. Landscape pattern refers to the spatial arrangement of various landscape elements of different sizes and shapes [28]. Changes in landscape pattern usually show the changes in the size, shape, and aggregation degree of various landscape patches in space, which is characterized by spatial heterogeneity [29]. This change reflects the spatial dimension of LC change, which is the evolution of ecological and spatial environment systems in a certain region, arising from the interaction of natural and human factors [30,31]. In addition, the landscape pattern index is widely used to study the characteristics, change analysis, and driving forces behind LC in different cities and in simulations to predict future urban spatial forms [32,33,34,35]. Therefore, the study of urban landscape characteristics is an important component of LC research, and it is also an important aspect of linking urban problems and proposing corresponding countermeasures.

The description above indicates that there are complex interactions between socio-economic factors, landscape patterns, and urban LC. This further suggests that the level of urbanization in different regions may lead to different LC distribution patterns [36]. However, few studies have attempted to explore the relationship between socioeconomic landscape patterns and urban LC through UFUs [28,37]. Urban Functional Units (UFUs) refer to physically defined spaces within urban areas with clear boundaries (such as residential areas, roads, etc.), utilized by humans for specific functions such as residence or transportation [17,38]. UFUs represent unique components within the urban matrix with distinct economic and social functions [39], leading to diverse LC patterns and resulting in varied landscape configurations. Similar types of LC often generate comparable landscape patterns [40,41].

In recent years, with the continuous development of remote sensing technology, research on LC has become increasingly in-depth [42,43,44,45]. LC research relies on obtaining image data of the Earth’s surface from satellites or aircraft, which provides a wealth of information and clarifies land cover distributions, subsequently informing prospective regional management and development planning [38]. In addition, landscape index—as a quantitative index to analyze the characteristics of landscape pattern—has formed a unique research paradigm of the spatial patterns of large-scale ecosystems in recent years by combining with remote sensing satellites and geographic information systems [3,46]. Furthermore, recent work combining remote sensing and socio-economic variables has identified a close relationship between these changes and social-economic factors, such as housing price, construction age, and population density [17,38,47,48,49,50,51]. Therefore, it is particularly important for us to apply remote sensing technology to understand the driving modes of LC in different regions for a more thorough study of urban sustainable development.

Most LC studies have been restricted to single cities or metropolitan areas, even at the national level [52,53,54]. In addition, it is worth noting that metropolitan areas have received the most attention in almost all previous studies [55,56,57], while small- and medium-sized cities—especially those in developing countries—are strongly underrepresented. However, it is expected that almost all of the world population growth and most global economic growth in these cities will occur in the coming decades [58,59]. Thus, a key overlooked task is to consider the LC in these developing cities in the analysis.

Hainan Province is the southernmost provincial-level administrative region in China, and it is also the youngest province in the country. Its urbanization started relatively late, highlighting the characteristics of developing cities [60]. Moreover, as a special economic zone and pilot-free trade zone in China, Hainan Province has a good ecological environment, obvious regional advantages, and more space for sustainable development [60,61]. Therefore, it is important to study LC changes in this region. This study focuses on 18 counties and cities in Hainan Island, Hainan Province, China, aiming to investigate the impacts of the luxury effect, legacy effect, and land use hypotheses on land cover. The main purposes are as follows: (1) To use social and economic factors, landscape patterns, and UFUs. We conducted an in-depth study on the driving factors influencing the LC of 18 counties and cities in the eastern, western, northern, southern, inland, and coastal areas of Hainan Island, as well as within Hainan Province to accomplish this purpose. (2) To characterize the heterogeneous effects of socio-economic indicators, landscape indicators, and UFUs on the LC in cities in different regions of Hainan Island from the perspective of regional heterogeneity. (3) Finally, it raises policy implications for sustainable development in similar cities by considering the drivers affecting urban LC.

2. Materials and Methods

2.1. Study Area

The study area was 18 counties and cities of Hainan Island, located in southern China (north latitude 18°09′–20°10′, east longitude 108°37′–11°03′), which forms part of Hainan Province, China. Sansha City of Hainan Province was not included in the study area due to its small land area, small permanent resident population, and insufficient data. The 18 counties and cities of Hainan Island are divided into northern, southern, eastern, western, inland, and coastal areas according to specific administrative divisions [62] (Figure 1). Hainan Province is one of China’s established economic zones and currently the only free trade port [63]. Since the establishment of the Hainan International Tourism Island in 2009, Hainan has experienced rapid socioeconomic development and swift urban expansion [64]. By the end of 2023, the total population of Hainan Province had grown to 10.43 million, with a GDP of 755.118 billion yuan, achieving a year-on-year growth rate of 9.2%, ranking second among all provinces in China [65,66,67].

2.2. Sampling Design of Urban Functional Units (UFUs)

To designate sampling locations across the 18 counties and cities of Hainan Island, Landsat 7 satellite imagery is employed. Based on the methodologies of Guo et al. [68] and Cui et al. [48], and taking into account the unique characteristics of each city, such as urban density, land use distribution, and urban form, the grid size was adjusted according to the dimensions of each city’s study area [68]. Additionally, to ensure statistical significance of the results, each city was incorporated with a minimum of 80 urban functional units [48]. Haikou City was divided into a 1 km × 1 km grid, with Dongfang City divided into a 0.65 km × 0.65 km grid; Wenchang City into a 0.6 km × 0.6 km grid; Lingao County into a grid of 0.55 km × 0.55 km; Chengmai County, Danzhou City, Ding’an County, Tunchang County, Qionghai City, Lingshui City, and Wanning City into 0.5 km × 0.5 km grids; Sanya Yazhou into a 0.45 km × 0.45 km grid; Baoting and Changjiang counties into 0.4 km × 0.4 km grids, Ledong County into a 0.38 km × 0.38 km grid, Wuzhishan City into a 0.35 km × 0.35 km grid; and Baisha County and Qiongzhong City were divided into 0.3 km × 0.3 km grids. We aim to select at least one UFU in each grid within the main urban area wherever possible. To accurately identify the type of each UFU, we employ a combination of online and offline methods. “Online,” we use Google Maps imagery to determine the type of UFU. For example, if an area is identified as a hospital on the map, we initially classify that UFU as a hospital. “Offline,” we conduct field surveys to verify the accuracy of identified UFUs. For instance, if an UFU is classified online as a residential area but our on-site investigation reveals it to be uninhabited, we will then disregard that UFU.

In this study, based on the Urban Forest Effects model [69] and the 2018 Basic Urban Land Use Classification map of China (EULUC-China), following [39] research methodology, each of the 18 counties and cities is divided into six primary UFUs. These include utilities service districts, government agencies service districts, industry and business districts, recreation and leisure districts, residential districts, and transportation areas (for the specific numbers of UFUs in each city/county region, see Table S1). Utility service areas include water supply and sewage treatment plants, garbage treatment plants, crematoriums, and cemeteries. Government agency service districts include government agencies and institutions, higher education institutions, primary and secondary schools, scientific research institutes, and hospitals. Industry and business districts include factories, commercial office areas, wholesale markets, and supermarkets. Recreation and leisure districts encompass parks, public squares, leisure squares, food squares, sports centers, and museums. Residential districts include hotels, high- and low-story residences, and urban villages. Finally, transportation areas include airports, railway stations, car parking lots, and main roads. This approach allowed us to systematically explore a variety of urban forms and functions in different regions, including differences in urban density and LC space [39,70].

2.3. LC Data and Landscape Indicators

For each grid cell across the 18 counties and cities, we evaluated the LC type using remote sensing technology. We obtained 1 m spatial resolution Landsat 7 imagery of 18 counties and cities in Hainan Island for the year 2020 from Google Maps. A supervised classification method (ENVI 5.3; Exelis Visual Information Solutions Company, Texas City, TX, USA) was used to divide the LC into different types: impervious ground, woodlands, grassland, water area, and bare land. The overall accuracy assessments were all above 92.5% (see Figure 2 and

Table 1. Overall accuracy assessment of 18 counties and cities in Hainan Island, Hainan Province.

18 Counties and Cities	Haikou City	Lingshui County	Danzhou City	Baoting County	Wuzhishan City	Sanya City Yazhou District	Tunchang County	Dingan County	Changjiang County	Ledong County	Wanning City	Wenchang City	Chengmai County	Qiongzhong County	Lingao County	Qionghai City	Dongfang City	Baisha County
Overall Accuracy	92.5%	94.4%	95.5%	96.7%	96.8%	96.8%	97.5%	97.6%	97.8%	97.9%	98.3%	98.5%	98.7%	99.3%	99.4%	99.6%	99.8%	99.9%

), confirming the reliability of this approach [71] (see Table S2 for details). Next, we used ArcGIS to clip the data of each UFU that was classified through supervised classification, ultimately obtaining the distribution of LC for each urban functional unit sample site. The trimmed distribution map of LC (Figure 3) was then used as the input for FRAGSTATS 4.2.1 (available from https//fragstats.org/index.php/downloads, accessed on 1 December 2022), following the method described in [28,70]. From this, the 8 pattern indexes of landscape organization were calculated, as detailed in

Table 2. Descriptions of metrics of patterns in landscape organization used in this study [28,70].

Landscape Index	Definition	Formulas
Patch Number (NP)	The total count of patches within a specific landscape type.	$\mathrm{NP}=n_i$
Patch Density (PD)	The frequency of patches per unit area in a given landscape type.	$PD={\displaystyle \frac{ n_i}{A}} \left(1000\right)\left(100\right)$
Landscape Shape Index (LSI)	A metric indicating variations in the shape of the landscape.	$LSI={\displaystyle \frac{ 0.25E}{2\sqrt{A}}}$
Contagion Index(CONTAG)	A measure of the degree of tight connection among patches in the landscape.	$CONTAG=\left[1+{\displaystyle \frac{{{\displaystyle \sum }}_{i=1}^m{{\displaystyle \sum }}_{i=1}^m\left[\left(p_i\right)\left(\frac{g_{ik}}{{{\displaystyle \sum }}_{k=1}^m{g}_{ik}}\right)\right]\left[ln\left(p_i\right)\left(\frac{g_{ik}}{{{\displaystyle \sum }}_{k=1}^m{g}_{ik}}\right)\right]}{2 ln\left(m\right)}}\right]\left(100\right)$
Connectance Index (CONNECT)	The degree of functional linkages or connectivity among patches.	$CONNECT=\left[{\displaystyle \frac{{{\displaystyle \sum }}_{j>k}^{n_i}{c}_{ijk}}{\frac{n_i\left(n_i-1\right)}{2}}}\right]$
Patch Cohesion (COHESION)	The degree of structural and functional connectedness of patches, reflecting the connectivity of a plant’s habitat.	$COHESION=\left[1-{\displaystyle \frac{{{\displaystyle \sum }}_{j=1}^n{p}_{ij}}{{{\displaystyle \sum }}_{j=1}^n{p}_{ij \sqrt{a_{ij}}}}}\right]{\left[1-{\displaystyle \frac{1}{\sqrt{A}}}\right]}^{-1}\times 100$
Splitting Index (SPLIT)	A measure of the degree of landscape division or fragmentation.	$SPLIT={\displaystyle \frac{A^2}{{{\displaystyle \sum }}_{i=1}^m{{\displaystyle \sum }}_{j=1}^n{a}_{ij}^2}}$
Shannon’s Evenness Index (SHEI)	A measure of landscape richness, assessing the distribution and abundance of different patch types.	$SHEI={\displaystyle \frac{-{{\displaystyle \sum }}_j^n\left(p_i\times \ln p_i\right)}{\ln n}}$

. We subsequently explored the impact of these landscape indices on LC.

2.4. Factors of Social and Economic Variables

Working within the hypotheses of luxury effect, heritage effect, and land use effect, we chose housing price, construction age, and population density as the main representatives of each UFU, mainly for the following reasons: (1) These three metrics can be quantified with relative accuracy. (2) The housing price represents the degree of urbanization of functional units, and the land price is high in areas with high urbanization, such as the urban center [72,73]. (3) The completion period refers to the number of years since its construction; for example, if a UFU was built in 2004, it will have been in existence for 20 years by 2024, which is associated with its urbanization phase and habitat duration [74,75]. (4) The distribution of LC is easily affected in areas with high population density [76,77]. (For the specific data calculation route method, see Figure 4.)

2.5. Statistical Analysis

In this study, socioeconomic variables, landscape index, and urban functional unit type were used as predictors. Each of the 7 different types of LC studied, namely impervious area, forest land, grassland, bare land, water area, total area, and green area, was taken as a separate response variable. Initially, all predictor and response variables were normalized using the z-score method, and outliers with a z-score greater than 3.0 or less than −3.0 were removed [79,80]. For the data analysis, we used a two-step linear model (LM) approach. In the first step, each predictor variable was paired with the response variable. In the second step, we performed multiple LM models, including only those predictor variables with a p-value of less than 0.05 in the simple LM model. To find the optimal subset of models that best explained these variables, we used a model selection procedure based on the Akaike information criterion (AIC) and selected the model with the smallest AIC value. All statistical analyses were performed using R version 4.1.0 (R Foundation for Statistical Computing, Vienna, Austria).

3. Results

3.1. Driving Factors Affecting LC Distribution in the South and North of Hainan Island

In socioeconomic indicators, house prices have the greatest impact on building area (β = 0.028 *) in the southern region compared to the northern region (

Table 3. Regression model results for LC in the south and north of Hainan Island. “-” Represents that the variable was not included in the final model.

Different Factors		Impervious Area β Coefficient		Woodlands Area β Coefficient		Water Area β Coefficient		Bare Land Area β Coefficient		Grassland Area β Coefficient		UGS Area β Coefficient		Total Area β Coefficient
Region		Southern Region N = 742	Northern Region N = 1112	Southern Region N = 742	Northern Region N = 1112	Southern Region N = 742	Northern Region N = 1112	Southern Region N = 742	Northern Region N = 1112	Southern Region N = 742	Northern Region N = 1112	Southern Region N = 742	Northern Region N = 1112	Southern Region N = 742	Northern Region N = 1112
Intercept		0.076 ***	−0.017	−1455.736 *	0.046	−0.018	0.097 *	−0.059	−0.110 ***	−0.041	−0.027	−0.178 **	0.033 ***	0.026	0.009
Socioeconomic variables	Housing price	0.028 *	-	-	0.040 ***	-	0.028 *	-	−0.030 *	−0.036	−0.039 **	-	0.027 **	0.023	0.019 **
	Construction age	-	-	-	−0.033 *	-	−0.043 *	−0.145 *	0.024 *	-	-	-	−0.022	-	-
	Population density	-	-	-	-	-	-	-	-	-	0.038	-	-	-	-
Landscape index	NP	1.110 ***	0.355 ***	0.953 ***	0.419 ***	-	0.266 ***	0.107 ***	0.120 *	−0.360 ***	-	0.877 ***	0.382 ***	1.028 ***	0.409 ***
	PD	−0.257 ***	−0.223 ***	−0.247 ***	−0.089 ***	0.096 ***	−0.091 ***	0.065 ***	−0.028 *	−0.142 ***	−0.170 ***	−0.262 ***	−0.123 ***	−0.257 ***	−0.194 ***
	LSI	-	-	−0.231 ***	-	0.100 **	-	-	0.094 **	0.456 ***	0.332 ***	−0.166 **	0.077 **	-	0.260 ***
	CONTAG	−0.173 ***	-	-	−0.033 ***	0.444 ***	−0.059 ***	0.193 ***	-	−0.129 *	0.216 ***	−0.095 ***	-	−0.089 **	0.037 ***
	CONNECT	−0.517 **	-	-	−0.639 ***	-	−0.888 ***	-	0.300 ***	−0.886 **	−0.249 **	-	−0.600 ***	−0.357 *	−0.533 ***
	COHESION	-	-	-	-	0.068	0.031	0.031 *	-	-	−0.070 ***	-	−0.025	-	−0.042 ***
	SPLIT	−0.105 ***	-	0.108 ***	0.081 ***	0.099 **	-	-	−0.031	0.049	-	0.123 ***	0.068 ***	−0.043 *	−0.052 ***
	SHEI	−0.177 ***	-	0.106 ***	-	0.421 ***	-	0.190 ***	-	−0.137 *	0.166 **	-	0.016	−0.066	-
Primary UFUs	Utilities service districts	-	0.030	0.087	−0.061	−0.230	0.088	−0.085	-	−0.023	−0.003	0.086	-	-	0.013
	Government agencies service districts	-	0.042	0.081	−0.030	−0.031	−0.039	−0.010	-	0.052	0.088	0.087	-	-	0.027
	Industry and business districts	-	-	-	-	-	-	-	-	-	-	-	-	-	-
	Recreation and leisure districts	-	−0.002	0.237 **	0.039	−0.155	−0.014	−0.007 *	-	0.063	−0.003	0.242 **	-	-	0.013
	Residential districts	-	−0.008	0.212 **	0.015	−0.116	−0.082	−0.071	-	−0.058	−0.039	0.188 **	-	-	−0.011
	Transportation	-	−0.062	0.207 **	−0.022	−0.018	−0.009	−0.032	-	−0.200	−0.018	0.198 **	-	-	−0.046
R2		0.800	0.747	0.583	0.576	0.238	0.320	0.258	0.115	0.211	0.376	0.601	0.643	0.831	0.832
Akaike information criterion (AIC)		−1895.48	−3334.53	−1455.73	−3350.93	−1383.92	−2635.79	−2488.14	−2734.4	−1326.73	−2523.51	−1525.17	−3514.91	−2108.04	−4051.4
p-value		<2.2 × 10−16 ***	<2.2 × 10−16 ***	<2.2 × 10−16 ***	<2.2 × 10−16 ***	<2.2 × 10−16 ***	<2.2 × 10−16 ***	<2.2 × 10−16 ***	<2.2 × 10−16 ***	<2.2 × 10−16 ***	<2.2 × 10−16 ***	<2.2 × 10−16 ***	<2.2 × 10−16 ***	<2.2 × 10−16 ***	<2.2 × 10−16 ***

Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05.

). In contrast, in the northern region, house prices have the largest impacts on forest area (β = 0.040 ***), water area (β = 0.028 *), grassland area (β = −0.039 **), UGS area (β = 0.027 **), and total area (β = 0.019 **). Building age negatively impacts forest area (β = −0.033 *) and water area (β = −0.043 *) the most in the northern region (Table 3).

In landscape pattern indicators, except for water area in the southern region and grassland area in the northern region, all other LC types are influenced by NP. PD has significant impacts on all types of LC. LSI has the most negative impact on forest area (β = −0.231 ***) and UGS area (β = −0.166 **) in the southern region. Conversely, it positively impacts water area (β = 0.100 **), grassland area in both regions (β = 0.456 *** for the southern region, β = 0.332 *** for the northern region), UGS area in the northern region (β = 0.077 **), and total area (β = 0.260 ***) in the southern region. Except for forest area in the southern region, CONTAG negatively impacts other LC types. Apart from impervious surface area and bare land area in the northern region, as well as forest area, water area, bare land area, and UGS area in the southern region, CONNECT has significant negative impacts on LC in other regions. SPLIT has significant positive impacts on forest area (β = 0.108 *** for the southern region, β = 0.081 *** for the northern region) and UGS area (β = 0.123 ***) in both regions, while negatively impacting total area (β = −0.043 * for the southern region, β = −0.052 *** for the northern region). SHEI has large negative impacts on impervious surface area (β = −0.177 ***) and grassland area (β = −0.137 *) in the southern region, but significant positive impacts on forest area (β = 0.106 ***), water area (β = 0.421 ***) in the southern region, and grassland area (β = 0.166 **) in the northern region. In UFUs, recreation and leisure areas, residential areas, and transportation areas positively influence forest area and UGS area in the southern region (Table 3).

3.2. Driving Factors Affecting LC Distribution in the East and West of Hainan Island

In terms of socio-economic indicators, housing prices have a positive effect on forest area (β = 0.026 **) in the eastern region and on water area (β = 0.042 **) in the western region (

Table 4. Regression model results for LC in the east and west of Hainan Island. “-” Represents that the variable was not included in the final model.

Different Factors		Impervious Area β Coefficient		Woodlands Area β Coefficient		Water Area β Coefficient		Bare Land Area β Coefficient		Grassland Area β Coefficient		UGS Area β Coefficient		Total Area β Coefficient
Region		Eastern Region N = 1191	Western Region N = 675	Eastern Region N = 1191	Western Region N = 675	Eastern Region N = 1191	Western Region N = 675	Eastern Region N = 1191	Western Region N = 675	Eastern Region N = 1191	Western Region N = 675	Eastern Region N = 1191	Western Region N = 675	Eastern Region N = 1191	Western Region N = 675
Intercept		−0.003	−0.014	0.018	−0.041 **	0.182 ***	−0.006	−0.105 ***	−0.073 ***	−0.057	−0.063	0.048 ***	−0.060 ***	0.019 **	−0.039 **
Socioeconomic variables	Housing price	0.017 *	0.035 *	0.026 **	-	-	0.042 **	−0.024 **	-	−0.069 ***	-	-	-	0.011 *	0.032 *
	Construction age	-	-	-	-	−0.075 *	0.500 **	-	-	-	-	-	-	−0.017	-
	Population density	−0.024	-	-	-	-	-	-	-	-	-	-	-	-	-
Landscape index	NP	0.929 ***	0.282 ***	0.646 ***	0.649 ***	0.294 ***	0.543 ***	-	0.207 **	0.354 ***	−0.276 **	0.650 ***	0.574 ***	0.911 ***	0.462 ***
	PD	−0.188 ***	−0.260 ***	−0.112 ***	−0.103 ***	−0.030 *	−0.026 *	0.086 ***	−0.061 ***	−0.120 ***	−0.142 ***	−0.124 ***	−0.125 ***	−0.164 ***	−0.218 ***
	LSI	-	0.595 ***	−0.166 ***	−0.096	−0.141 **	−0.133 ***	0.028 *	0.076 *	-	0.425 ***	−0.148 ***	-	−0.115 ***	0.354 ***
	CONTAG	-	0.107 ***	−0.129 ***	0.137 *	-	−0.039 **	0.333 ***	−0.133 **	−0.034 *	0.245 **	−0.121 ***	0.221 ***	−0.074 ***	0.097 **
	CONNECT	-	−0.549 ***	−0.530 ***	-	−1.076 ***	-	0.131	-	−0.230	−0.342	−0.513 ***	-	−0.237 ***	−0.347 **
	COHESION	−0.045 ***	−0.038	-	0.055	0.107 ***	-	0.057 ***	-	-	−0.074	-	-	-	-
	SPLIT	−0.110 ***	−0.138 ***	0.089 ***	0.130 ***	0.064 **	-	-	-	0.036 *	0.088	0.086 ***	0.114 ***	-	−0.027
	SHEI	−0.013	-	−0.077 *	0.235 ***	0.099 ***	-	0.353 ***	−0.118 *	-	0.192 *	−0.066 *	0.299 ***	−0.044 *	0.067
Primary UFUs	Utilities service districts	-	-	−0.052	-	−0.021	-	-	-	0.134	0.134	-	-	-	-
	Government agencies service districts	-	-	0.020	-	−0.068	-	-	-	0.121 *	0.043	-	-	-	-
	Industry and business districts	-	-	-	-	-	-	-	-	-	-	-	-	-	-
	Recreation and leisure districts	-	-	0.060	-	−0.039	-	-	-	0.101	−0.048	-	-	-	-
	Residential districts	-	-	0.049	-	−0.151 **	-	-	-	0.015	−0.087	-	-	-	-
	Transportation	-	-	0.050	-	−0.039	-	-	-	0.052	−0.071	-	-	-	-
Adjusted R2		0.810	0.769	0.622	0.525	0.285	0.287	0.191	0.169	0.288	0.282	0.679	0.556	0.880	0.828
Akaike information criterion (AIC)		−3694.09	−1758.57	−3693.11	−1570.46	−2742.28	−1790.52	−3642.22	−1636.82	−2594.98	−1250.75	−3904.11	−1595.89	−4592.03	−2132.94
p-value		<2.2 × 10−16 ***	<2.2 × 10−16 ***	<2.2 × 10−16 ***	<2.2 × 10−16 ***	<2.2 × 10−16 ***	<2.2 × 10−16 ***	<2.2 × 10−16 ***	<2.2 × 10−16 ***	<2.2 × 10−16 ***	<2.2 × 10−16 ***	<2.2 × 10−16 ***	<2.2 × 10−16 ***	<2.2 × 10−16 ***	<2.2 × 10−16 ***

Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05.

). Conversely, they negatively impact grassland area (β = −0.069 ***) in the eastern region. Building age has the most significant negative impact on water area (β = −0.075 *) in the eastern region, whereas it has the largest positive impact on water area (β = 0.500 **) in the western region (Table 4).

Regarding landscape pattern indices, LSI negatively affects forest area (β = −0.166 ***), water area (β = −0.141 **), and UGS area (β = −0.148 ***) in the eastern region, whereas it positively affects impervious surface area (β = 0.595 ***) and grassland area (β = 0.425 ***) in the western region. CONTAG has substantial negative impacts on forest area (β = −0.129 ***), grassland area (β = −0.034 *), and UGS area (β = −0.121 ***) in the eastern region, but its effects are positive on forest area (β = 0.137 *), grassland area (β = 0.245 **), and UGS area (β = 0.221 ***) in the western region. CONNECT negatively affects both regions’ total area. COHESION negatively affects impervious surface area (β = −0.045 ***) in the eastern region, whereas it positively affects water area (β = 0.107 ***) in the same region. SHEI negatively impacts forest area (β = −0.077 *) and UGS area (β = −0.066 *) in the eastern region, whereas it positively affects forest area (β = 0.235 ***) and UGS area (β = 0.299 ***) in the western region. In terms of UFUs, government agencies service districts (β = 0.121 *) have the largest impact on grassland area in the eastern region. Residential areas (β = −0.151 **) have the greatest impact on water area in the eastern region (Table 4).

3.3. Driving Factors Affecting the LC Distribution in the Inland and Coastal Areas of Hainan Island

In terms of socio-economic factors, house prices are positively correlated with inland forests (β = 0.070 ***) but negatively correlated with inland water bodies (β = −0.076 ***) and grasslands (β = −0.089 ***). Building age negatively affects impervious surfaces in inland areas (β = −0.946 *) and water bodies in coastal areas (β = −0.069 **) (

Table 5. Regression model results for LC in in the inland and coastal areas of Hainan Island. “-” Represents that the variable was not included in the final model.

Different Factors		Impervious Area β Coefficient		Woodlands Area β Coefficient		Water Area β Coefficient		Bare Land Area β Coefficient		Grassland Area β Coefficient		UGS Area β Coefficient		Total Area β Coefficient
Region		Inland Region N = 485	Coastal Region N = 1380	Inland Region N = 485	Coastal Region N = 1380	Inland Region N = 485	Coastal Region N = 1380	Inland Region N = 485	Coastal Region N = 1380	Inland Region N = 485	Coastal Region N = 1380	Inland Region N = 485	Coastal Region N = 1380	Inland Region N = 485	Coastal Region N = 1380
Intercept		1.996 ***	−0.040	0.231	−0.122	2.994 ***	0.714	−0.153 ***	−0.164	2.166	0.032	0.248	−0.006	1.866 ***	−0.007
Socioeconomic variables	Housing price	-	0.0140	0.070 ***	0.017	−0.076 ***	0.048 ***	−0.063 **	-	−0.089 ***	−0.026 *	0.029	-	-	0.016 *
	Construction age	−0.946 *	-	-	−0.033	-	−0.069 **	-	-	-	−0.049	-	-	-	−0.029 *
	Population density	-	-	-	-	-	-	-	-	-	-	-	-	-	-
Landscape index	NP	0.819 ***	0.506 ***	0.597 ***	0.523 ***	−0.383 ***	0.428 ***	−0.264 **	-	0.284 ***	-	0.634 ***	0.469 ***	0.790 ***	0.535 ***
	PD	−0.145 ***	−0.218 ***	-	−0.123 ***	-	−0.034 **	-	0.018 *	-	−0.119 ***	-	−0.138 ***	−0.104 ***	−0.189 ***
	LSI	0.110	0.259 ***	−0.150	−0.105 ***	0.421 ***	−0.179 ***	0.346 ***	0.142 ***	-	0.149 ***	−0.164	−0.061 *	-	0.125 ***
	CONTAG	0.321 **	-	0.560 ***	−0.077 ***	0.891 ***	−0.145 ***	0.504 ***	0.072 *	0.357 *	0.021 *	0.558 ***	−0.065 ***	0.410 ***	-
	CONNECT	−15.401 ***	−0.327 ***	-	−0.208 ***	−23.801 ***	−0.245 ***	-	0.096 *	−17.000 *	−0.335 ***	-	−0.253 ***	−12.669 ***	−0.326 ***
	COHESION	−0.060	−0.033 **	-	0.032 *	-	0.090 ***	-	-	0.095 ***	-	0.064	0.030 *	-	-
	SPLIT	−0.149 ***	−0.108 ***	0.234 ***	0.062 ***	−0.116 ***	0.101 ***	−0.065	-	-	0.078 ***	0.221 ***	0.074 ***	-	−0.028 **
	SHEI	0.261 *	−0.057 ***	0.578 ***	-	0.897 ***	−0.093 *	0.456 ***	0.067 *	0.462 ***	-	0.629 ***	-	0.403 ***	-
Primary UFUs	Utilities service districts	-	0.115	−0.357	0.043	-	−0.756 ***	-	0.057	-	0.100	−0.355	-	−0.205	-
	Government agencies service districts	-	0.055	−0.358	0.087	-	−0.687 ***	-	0.103	-	0.002	−0.374	-	−0.205	-
	Industry and business districts	-	0.007	−0.284	0.090	-	−0.683 ***	-	0.069	-	−0.037	−0.322	-	−0.161	-
	Recreation and leisure districts	-	0.043	−0.063	0.113	-	−0.684 ***	-	0.061	-	−0.005	−0.187	-	−0.193	-
	Residential districts	-	0.044	−0.333	0.152	-	−0.747 ***	-	0.036	-	−0.091	−0.378	-	−0.253	-
	Transportation	-	−0.022	−0.207	0.108	-	−0.619 ***	-	0.105	-	−0.067	−0.279	-	−0.250	-
Adjusted R2		0.835	0.753	0.540	0.553	0.2456	0.262	0.103	0.145	0.250	0.292	0.622	0.580	0.884	0.805
Akaike information criterion (AIC)		−1353.4	−3935.07	−1050.81	−4037.35	−980.7	−3341.21	−950.63	−3719.91	−697.86	−3390.85	−1131.16	−4154.62	−1630.04	−4674.2
p-value		<2.2 × 10−16 ***	<2.2 × 10−16 ***	<2.2 × 10−16 ***	<2.2 × 10−16 ***	<2.2 × 10−16 ***	<2.2 × 10−16 ***	9.509 × 10−11 ***	<2.2 × 10−16 ***	<2.2 × 10−16 ***	<2.2 × 10−16 ***	<2.2 × 10−16 ***	<2.2 × 10−16 ***	<2.2 × 10−16 ***	<2.2 × 10−16 ***

Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05.

).

Regarding landscape pattern indices, LSI positively influences water bodies in inland areas (β = 0.421 ***) but negatively influences water bodies in coastal areas (β = −0.179 ***). CONTAG has a positive effect on all LC types in inland areas. CONNECT negatively affects all LC types in coastal areas. COHESION has the strongest impact on impervious surfaces (β = −0.033 **), forest area (β = 0.032 *), water area (β = 0.090 ***), and UGS area (β = 0.030 *) in coastal areas. SPLIT negatively affects water bodies in inland areas (β = −0.116 ***), whereas it positively affects water bodies in coastal areas (β = 0.101 ***). SHEI has a positive effect on all LC types in inland areas. Utilities service districts (β = −0.756 ***), government agencies service districts (β = −0.687 ***), industry and business districts (β = −0.683 ***), recreation and leisure districts (β = −0.684 ***), residential areas (β = −0.747 ***), and transportation areas (β = −0.619 ***) have the strongest negative impact on water bodies in coastal areas (Table 5).

3.4. Driving Factors Affecting the LC Distribution of Hainan Island as a Whole and of 18 Counties and Cities

House prices positively influence forest area (β = 0.033 ***), UGS area (β = 0.015 *), and total area (β = 0.017 **) on Hainan Island overall. Apart from bare land area, NP, SPLIT, and CONNECT have the greatest impact on other LC types on Hainan Island. PD and LSI significantly affect all LC types on Hainan Island. CONTAG positively affects impervious surfaces (β = 0.071 ***) and bare land area (β = 0.127 ***) but negatively affects forest area (β = −0.064 ***). COHESION has the greatest positive impact on water bodies (β = 0.076 ***) and the greatest negative impact on impervious surfaces (β = −0.072 ***) across Hainan Island. SHEI has the largest effects on water bodies (β = 0.071 ***), bare land area (β = 0.130 ***), and UGS area (β = 0.056 ***) overall on Hainan Island, while public service areas (β = −0.691 ***), administrative and public service areas (β = −0.616 ***), industrial and commercial areas (β = −0.623 ***), recreation and leisure areas (β = −0.601 ***), residential areas (β = −0.683 ***), and transportation areas (β = −0.569 ***) all negatively influence water bodies on Hainan Island (See

Table 6. Regression model results for LC in the 18 counties and cities of Hainan Island. “-” Represents that the variable was not included in the final model.

Different Factors		Impervious Area N = 1832 β Coefficient	Woodlands Area N = 1832 β Coefficient	Water Area N = 1832 β Coefficient	Bare Land Area N = 1832 β Coefficient	Grassland Area N = 1832 β Coefficient	UGS Area N = 1832 β Coefficient	Total Area N = 1832 β Coefficient
Intercept		0.059	−0.027	0.656 ***	−0.089 ***	0.072	0.020 *	0.015
Socioeconomic variables	Housing price	0.011	0.033 ***	0.018	−0.019 *	−0.054 ***	0.015 *	0.017 **
	Construction age	-	-	-	-	-	-	-
	Population density	-	-	-	-	-	-	-
Landscape index	NP	0.510 ***	0.561 ***	0.373 ***	-	0.146 **	0.500 ***	0.588 ***
	PD	−0.239 ***	−0.122 ***	−0.042 ***	0.025 **	−0.149 ***	−0.142 ***	−0.208 ***
	LSI	0.337 ***	−0.102 ***	−0.140 ***	0.097 ***	0.113 ***	−0.035	0.148 ***
	CONTAG	0.071 ***	−0.064 ***	-	0.127 ***	-	-	-
	CONNECT	−0.423 ***	−0.365 ***	−0.468 ***	-	−0.548 ***	−0.445 ***	−0.485 ***
	COHESION	−0.072 ***	0.018	0.076 ***	0.018	-	-	−0.018 *
	SPLIT	−0.152 ***	0.085 ***	0.059 ***	-	0.043 *	0.069 ***	−0.044 ***
	SHEI	-	-	0.071 ***	0.130 ***	0.022	0.056 ***	-
Primary UFUs	Utilities service districts	0.011	−0.021	−0.691 ***	-	0.084	-	-
	Government agencies service districts	−0.027	0.022	−0.616 ***	-	0.007	-	-
	Industry and business districts	−0.072	0.023	−0.623 ***	-	−0.061	-	-
	Recreation and leisure districts	−0.065	0.083	−0.601 ***	-	−0.039	-	-
	Residential districts	−0.045	0.065	−0.683 ***	-	−0.099	-	-
	Transportation	−0.104	0.051	−0.569 ***	-	−0.077	-	-
R2		0.751	0.525	0.207	0.119	0.256	0.572	0.807
Akaike information criterion (AIC)		−5097.35	−5151.43	−4215.75	−4921.48	−3811.52	−5352.69	−6134.81
p-value		<2.2 × 10−16 ***	<2.2 × 10−16 ***	<2.2 × 10−16 ***	<2.2 × 10−16 ***	<2.2 × 10−16 ***	<2.2 × 10−16 ***	<2.2 × 10−16 ***

Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05.

).

4. Discussion

4.1. Luxury Effect Test of the Mechanism Driving LC Changes

The luxury effect has been widely used to explain the distribution of LC in cities across the globe [17,81,82]. In our analysis of the main UFUs in the 18 counties and cities of Hainan Island in 2020, we found a significant positive correlation between building area in the southern region, forested areas, UGS areas, and property prices across the island (Figure 5). While these variables are agents of wealth, more wealth helps to invest more in building houses and maintaining woodlands and UGS. Thus, our findings support the luxury effect hypothesis as a driver of the distribution of impervious area, woodlands, and UGS area in these areas. In more affluent areas, urban residents have greater capacity to allocate resources towards building areas, forest land, and UGS, thereby influencing the overall LC of the region [83]. Additionally, wealthier residents possess stronger lobbying abilities and greater opportunities to interact with decision-makers, including public and private investors, enhancing their influence and agency over public and private land use planning and investments [83]. Moreover, wealthier regions may place greater value on the quality of life and environmental comfort of residents, modifying ecosystems to this end [84]. Woodlands and UGS provide a better living environment, including fresh air, beautiful scenery, and places for leisure activities [85]. Therefore, affluent suburbs may invest in the construction and protection of UGS systems to pay higher prices for these quality living environments [86]. Moreover, a (Pan 2020 [87]) study found that house prices generally reflect the supply and demand dynamics in the housing market. High demand has led to more buildings developing in cities, and people are willing to pay higher prices for homes [87]. Thus, the influence of affluent residents can ultimately promote different LC distributions, driven by lifestyle choices, social status, and investment capacity for environmental management [88].

Grassland area decreased with increasing housing prices, mainly because grassland and woodlands areas tend to grow alternately, which is consistent with previous reports of a negative correlation between woodlands and grassland [89]. This trend is believed to arise from ecological competition between trees and grassland. Large trees compete for light, water, and soil nutrients, thus inhibiting grassland expansion [38,90]. As the number of planted trees increases in affluent areas, the shadows formed by the canopy partly changes the microclimate of the surface, reducing the growth space of shadow-sensitive herbs and consequently resulting in a reduction in grassland area [91,92].

Our study also found that with higher house prices, landscapes became more fragmented and had poorer connectivity. This phenomenon is strongly associated with an increased degree of urbanization [93]. With the expansion of urban construction, the density of buildings and infrastructure increases, and the originally continuous natural landscape is interrupted by buildings and artificial facilities, forming a scattered and fragmented landscape [94,95], increasing landscape fragmentation and decreasing connectivity. In addition, high house prices introduce pressure on land developers to maximize LC developmental value while meeting people’s demand for high-quality housing and commercial land. Often this leads to large-scale developments across the original green space, fragmenting it into much smaller—often disconnected—pieces of green space [96,97]. In such situations during urban development, it is recommended that policy formulation actively advocate for sustainable urban policies, give priority to the development and maintenance of green space, establish systems such as ecological corridors and green roofs, and protect the connectivity and integrity of the ecosystem [98]. It is also important to actively call on the general public to participate, cultivate public awareness, encourage reasonable distribution and accessibility of UGS to each resident, promote a comfortable living environment, improve the social awareness of landscape problems, promote social participation from all walks of life, promote the implementation of the solution, and create an inclusive and vibrant urban environment [99]. Only by considering ecological, economic, and social factors and taking scientific and comprehensive measures combined with long-term participation can the challenges of sustainable urban development be effectively met.

4.2. Heritage Effect Test of the Mechanism That Drives LC Change

In this study, a reverse legacy effect was found, which mirrors the results of another recent study [88]. Our results indicated that in the northern part of Hainan Island, the older the construction age of houses, the lower the woodlands area (Figure 5). This may be due to the dynamic impact of regional development on urban ecosystem quality and the historical decisions and urban planning adopted in the past, which have lasting effects on the current woodlands cover pattern [100,101]. Since the 1950s, Hainan Island has suffered considerable deforestation due to the increasing demand for land and resources—both within the island and outside the island. Due to the mining industry and insufficient forestry investment, Hainan Island has changed from a net timber export area in the 1950s and 1970s to a net importer in the 1980s [102]. It is estimated that 2700 hectares of woodlands were destroyed annually in the 1970s and 1980s [103]. After 30 years of destruction, the woodlands area of Hainan Island shrank to its lowest level in the late 1970s: around 15% of the total land area [102]. Research indicates that during the period of 2010–2014, there was excessive and unbalanced land development in the northern region of Hainan Island, particularly in cities such as Haikou, Lingao, and Danzhou [104]. Haikou City is an important growth pole of economic development of Hainan Province, and its GDP accounts for about 30% of the whole Hainan Province. Its economic development is heavily dependent on land development and construction. Construction land reserve resources are inadequate to meet the development needs, so the land development intensity is much higher than the provincial average. Economic radiation from the rapidly developing Haikou City, along with the Danzhou construction of Yangpu Industrial Zone and the Haihua Island Project, have imposed strong environmental and ecological pressures, especially on woodlands.

We also observed a reverse legacy effect on the impervious area in the inland areas of Hainan Island. First of all, the urbanization process in inland areas of Hainan Island is relatively slow, and the population inflow and urbanization process may not be as rapid as those in coastal areas [64]. Therefore, the growth rate of building demand may be slow, and sometimes there are vacant and idle buildings, leading to the decrease of impervious area [105]. Since 2016 [106], Hainan Province has undertaken organized development and revitalization efforts to reclaim and utilize vacant and idle land, yet the phenomenon of idle land still exists [107]. Secondly, because the high-speed railway around Hainan only revolves around coastal cities, the inland areas have an underdeveloped economy, inconvenient transportation routes, and poor infrastructure [108,109]. At the same time, the construction of the Hainan Free Trade Port and the development of tourism are mainly concentrated in the coastal cities, which further leads to the migration of the population in inland areas to the coastal areas, resulting in a smaller population in inland areas [110,111], which in turn reduces the demand for construction. Additionally, buildings will accumulate damage as they age, and some may be abandoned or demolished for other purposes [112]. According to relevant data [113,114], the government of Hainan Province has implemented planning and adjustments in the inland areas, reallocating building areas for other purposes such as farmland and ecological conservation. Consequently, this has led to a further reduction in building areas.

Another interesting phenomenon that was noted is that the water area of the eastern and coastal areas decreased with construction age, while that in the western area increased with construction age. The economy of the eastern coastal areas is more developed than that in the western areas, and the urbanization process is more mature, especially in Sanya and Haikou [115]. Rapid urbanization began in Haikou and Sanya in the 1990s. After this period, the Haikou High-tech Industrial Development Zone, the Yalong Bay National Tourism Resort in Sanya, and the Hainan Yangpu Bonded Port Zone were set up, which further stimulated the rapid development of the eastern coastal areas [115]. Especially since the establishment of the Hainan Free Trade Port, rapid development in coastal areas has attracted a significant influx of external investments for urban development. This surge has led to an increase in imports and exports, resulting in a heightened demand for infrastructure such as roads and ports. The rapid expansion of urban spaces has led to the filling in of water bodies, further diminishing the water area within cities [116].

However, the economy in the western region is relatively poor. From 2000 to mid-2020, the GDP of Hainan Island has increased significantly, with eastern Haikou and Sanya having the highest GDP [64]; however, the western region lags behind. For example, in 2020, the per capita GDP of Sanya (eastern coastal area) was 88,900 CNY, while the per capita GDP of Ledong County (western area) was only 31,000 CNY. The Western Baisha and Baoting also had a low GDP per capita in 2000–2020 [64]. This highlights the serious imbalance in the economic development of Hainan Island, consistent with the conclusions of previous research [64]. In the western regions, the level of economic development is relatively low, and there are many water bodies that remain undeveloped [117]. With the nation’s increasing emphasis on the development and construction of Hainan [111], the western regions have also begun to develop. Simultaneously, as awareness of environmental protection and sustainable development grows among the populace, there is a greater emphasis on environmental conservation in new urban construction. Both the government and society have implemented more measures to protect and increase water resources. This trend is particularly pronounced against the backdrop of Hainan Island’s status as a key ecological conservation area [118].

4.3. The Land Use Hypothesis Tests the Mechanism Driving the LC Change of UFUs

Urban planning plays a key role in resource allocation [119]. Our study found significant positive associations with UGS and woodlands in recreational areas, residential areas, and transportation areas in the southern region, which is consistent with previous studies [47] (Figure 5). The UGS provides important ecosystem services and functions, such as habitat conservation, climate regulation, and pollution removal to improve air quality and reduce runoff [120]. Not only that, UGS can also bring people to relax, exercise, and socialize by providing entertainment space and an attractive outdoor environment [121]. Forests, as a component of UGS, play an indispensable role that cannot be overlooked. Due to their height and large canopy cover area, trees can provide people with abundant shade from sunlight and wind, noise reduction, and other protection [122]. Some studies have noted that trees can reduce the street temperature by up to 19 °C, especially in tropical regions [123,124].

Therefore, urban planners should recognize that UGS distribution is critical to the wellbeing of residents in urban design. Whether in residential, recreational, or transportation areas, the visual, recreational, aesthetic, and other functions of UGS should be taken into account, along with its development and ecological functions [125]. For example, in residential areas, UGS can meet the needs of residents for leisure and sports activities, while UGS in transportation areas can serve the needs of pedestrians and street traffic while ensuring the continuity of the landscape and showcasing the beauty of the city [126]. By planning more UGS in appropriate places, urban planners can promote the sustainable development of urban areas, coordinate the relationships between urban population, economy, resources, and environment, and guide the city towards a more natural and balanced direction [127]. Furthermore, urban planners and decision-makers should also consider the long-term costs and benefits of UGS development and the important role of biodiversity in supporting the stability and resilience of urban ecosystems.

In addition, with the acceleration of urbanization, the LC mode of cities continues to change. For example, land reclamation can convert water areas into other LC types (most commonly, shallow straits into land) in order to expand urban land area [17,128]. For example, in the northern coastal area of Haikou city, the newly reclaimed areas are mainly near the Haidian District of Haikou city, which mainly comprises buildings and similar infrastructures [129]. Therefore, the practice of land reclamation can increase the available construction area. On the one hand, this alleviates the contradiction between supply and demand of land in coastal areas and expands the living space and development space of the society [129,130], but it has also led to some negative effects. In the past 40 years, massive land reclamation in China has resulted in a loss of about 21,900 square kilometers of wetlands, equivalent to about 50% of the country’s total coastal wetland area [17,129].Therefore, in the process of urban development, it is suggested to make rational use of the existing waters, incorporate them into the urban ecosystem and adopt the concept of multi-functional urban design, organically combine the natural water landscape with urban functions, and improve the local optimization within the city [131]. For example, retaining natural wetlands in cities can be used as both a biological habitat and a functional area for urban flood control and water resources regulation [132,133].

4.4. Study Limitations

While our study provides valuable insights into the factors affecting LC in tropical China, there are some limitations that should be acknowledged. First, in this study, we did not observe an effect of population density on LC, probably because Hainan Province is regarded as a migratory area that mainly attracts older migrants from cold areas. These cities are therefore called migratory bird destinations, and these elderly people or “migratory birds” prefer to buy property in their temporary tourist cities to live out their retirement [134]. Although this migration is seasonal and temporary, its effect on the destination area may be quite significant [134]. For example, temporary migration can have effects on public transport, housing, population density, and infrastructure development [135]. During the peak season, population mobility increases rapidly, but public services such as infrastructures may not be enough to meet the growing demand [134]. In the off-season, the population decreases sharply, resulting in underutilization of these houses and infrastructure [134]. Therefore, it is difficult to judge the drivers of LC solely from the large fluctuations in population density. This situation requires more in-depth analysis and research to fully understand the impact of these factors on LC.

Some of the house price data may also be inaccurate, as transport areas or those belonging to government agencies had no relevant data. In these situations, the average house price data from around the area were utilized, and thus there may be some deviation. There is room for future work to identify and appraise more accurate and sensitive indicators to evaluate the urban areas of LC drivers.

The construction time is only a static index, which can fail to fully reflect the actual situation and use of the house. Certain buildings may have been renovated, expanded, or rebuilt in their past, reducing the value of construction age as a metric. However, the distribution of LC characteristics and building dates may differ significantly in different regions; therefore, the results may not be directly generalized to other regions. Additionally, it is important to consider the influence of spatial heterogeneity on the results.

Finally, the geographical scope of our study was limited to a specific region of China. This means that our results may not be directly transferable to all other tropical cities, as different regions with different environmental, socioeconomic, and historical backgrounds can have a significant impact on urban LC. Therefore, future studies should focus on more regions and more cities, with a greater sample size.

5. Conclusions

Since the establishment of the free trade port, Hainan Island has been undergoing rapid urbanization. In order to explore the driving factors affecting LC in Hainan Island, this paper investigated the relationship between social economy, landscape pattern, urban functional unit, and LC and systematically and comprehensively analyzed the driving factors of urban LC in 18 counties and cities across Hainan Island in 2020. Our study found that areas with higher housing prices had more UGS and floor area, with increased landscape fragmentation and decreased connectivity, confirming the impact of the luxury effect hypothesis on LC. Furthermore, we found the effect of a reverse legacy effect on LC, indicating that urban planning decisions adopted in the past may have a lasting impact on the current woodlands cover pattern. Finally, we confirm the role of the land-use hypothesis in urban planning through the relationship between UFUs and LC. The above conclusions provide some important policy implications for the urban planning and land use management in Hainan Island and similar areas. While we also explored the impact of LC drivers within a limited scope, future studies should incorporate greater sampling efforts and other critical factors (e.g., climate, soil) in order to fully understand the driving mechanism of urban LC.