Habitat Quality Assessment and Driving Factors Analysis of Guangdong Province, China

Abstract

Habitat quality is a key factor in regional ecological restoration and green development. However, limited information is available to broadly understand the role of natural and human factors in influencing habitat quality and the extent of their impact. Based on remote sensing monitoring data of land use over five time points (2000, 2005, 2010, 2015, and 2020), natural factors, and socioeconomic data, we applied the InVEST model to assess habitat quality in Guangdong Province. Using a multiscale geographically weighted regression (MGWR) model, we explored the spatial scale differences in the role of natural and human factors affecting habitat quality and the degree of their influence. The highlights of the results are as follows: ① From 2000 to 2020, land-use changes in the Pearl River Delta (PRD) region were particularly obvious, with the dynamic degree of construction land being higher than that of other land-use types. Construction land has gradually occupied agricultural and ecological land, causing damage to habitats. ② The overall habitat quality in Guangdong Province is decreasing; the areas with low habitat quality values are concentrated in the PRD region and the coastal areas of Chaoshan, Maoming, and Zhanjiang, while the areas with higher habitat quality values are mainly located in the non-coastal areas in the east and west of Guangdong and the north of Guangdong. ③ The MGWR regression results showed that the normalized vegetation index had the strongest effect on habitat quality, followed by road density, gross domestic product (GDP) per unit area, slope, and average elevation, and the weakest effect on average annual precipitation. ④ The effects of average elevation, GDP per unit area, and normalized vegetation index on habitat quality were significantly positively correlated, while road density was significantly negatively correlated. These results provide a scientific basis for adjusting spatial land-use planning and maintaining regional ecological security.

1. Introduction

Biodiversity is an important foundation for life on Earth, and the current widespread decline in biodiversity due to rapid global urbanization is considered a major threat [1]. Habitat quality refers to the ability of a species’ living space to provide suitable living conditions for individuals or populations, that is, the ability of an ecosystem to provide sustainable development within a certain spatial and temporal range [2]. Habitat quality is the basis of functional ecosystems and an important factor that influences biodiversity [3]. Ecosystem fragmentation is increasing with the intensification of industrialization and accelerated urbanization. Habitat quality is significantly declining, and biodiversity is facing a significant threat [4].

In general, scholars in China and abroad study habitat quality mainly through field surveys to obtain habitat quality parameters and construct a comprehensive evaluation system. The evaluation results obtained using this method are highly accurate. However, they are more difficult to implement, consume more human and material resources, and are less efficient than modern sensor-based methods. The use of landscape indices or models to assess habitat quality has become a popular research topic in academia owing to time constraints and data availability. Habitat quality assessment models include the Integrated Valuation of Ecosystem Services and Tradeoffs (InVEST) [5] and the maximum entropy (MaxENT) models [6]; the InVEST model is the most widely used in habitat quality assessment owing to its low amount of imported data, large output data volume, and simple operation.

InVEST is an advanced tool for integrating natural capital as the basis for decision making. This can help spatialize natural capital and ensure that environment-related decisions are conducive to expanding human interests and protecting the environment. Therefore, it has promising applications in governments, non-governmental organizations, and research institutions [7]. Since its release, the InVEST model has been widely used in Europe, North and South America, Asia, Africa, and Europe. For example, in 2011, Fisher et al. applied the model to a forest ecosystem in Tanzania, making the InVEST model more extensive in its assessment content and area [8]. Nelson et al. from Stanford University used the model and its scenario prediction function to study the impacts of land-use change on ecosystem services in the Willamette River Basin in southwestern Oregon, United States [9]. In recent years, scholars have begun comparative assessments based on the scenario prediction function of the InVEST model combined with the establishment of scenarios to explore the conservation effects of different ecosystem services on animal and water service values during drought years [10,11]. Chinese researchers have modified the model parameters to apply InVEST to assess ecosystem service functions in China. Relevant domestic studies have focused mainly on evaluating soil conservation, water supply, and carbon stock functions [7].

To study the spatial and temporal variation characteristics of habitat quality, scholars have used the InVEST model to evaluate habitat quality but have also combined the image dichotomy [12], gray correlation model [13], geographic probe [14], and geographically weighted regression [15] methods to refine and analyze the obtained results and discuss the influence of different land-type factors and their inter-relationships on habitat quality. This provides a theoretical basis for the global construction of ecological security patterns, identification of important habitat patches, and optimization of ecosystem ecological service functions. For applied research on habitat quality evaluation, some scholars currently combine regional habitat quality with the change in land-type share using multi-criteria evaluation (MCE), cellular automata (CA), Markov chain models [16], future land-use simulation (FLUS) [17], or other models to predict the land use, ecological degradation degree, and habitat quality change trends under different scenarios to provide a scientific basis for the government. There are few studies on the attribution analysis of regional habitat quality, and most existing analyses of habitat quality factors are based on traditional statistical methods, such as correlation, multiple regression, and geographical weighting analyses.

The multiscale geographically weighted regression (MGWR) method proposed in 2017 by Fotheringham, a member of the American Academy of Sciences, provides a new way of thinking about the attribution analysis of habitat quality. MGWR is an extension of geographically weighted regression (GWR) that considers geospatial differences and calculates the weight values of different variables using a kernel function to determine the influence scale of the variables, which can effectively avoid capturing too many confounding factors and noise [18].

MGWR has been applied by scholars to studies on urban pollution, social change, population movement, urban construction, and transportation construction to explore the roles of relevant influencing variables. This provides powerful support for social decision making in cities. Fotheringham et al. used MGWR to assess the influencing factors of air pollution in China and compared and explained it with ordinary least squares (OLS) and OLS containing a spatial lag variable (OLSL) with GWR [19]. Yang et al. used a high-resolution map generation method, multi-scale geographically weighted regression kriging (MGWRK) based on MGWR, and area-to-point kriging (ATPK) using random forest (RF) as a scale factor and selected the normalized difference vegetation index (NDVI), normalized difference built-up index (NDBI), and reflectance as influencing factors for surface cooling scales in a city analysis [20]. Zhang et al. used MGWR with neighborhood, traffic, and construction factors as influencing factors to explore the factors influencing housing rentals in Nanjing [21]. Niu et al. used MGWR to investigate the spatial heterogeneity and geographical scale of urban heat island driving mechanisms, improving the understanding of the complex drivers of urban surface heat island effects from a multi-scale perspective and providing a basis for mitigating urban heat islands [22]. This method has been successfully used in empirical studies to identify the action scale of different influences on explanatory variables and in habitat quality studies [23].

At present, scholars at home and abroad have mainly focused on assessing and predicting regional habitats [24,25,26] and exploring spatial and temporal variation characteristics [25], spatial and temporal evolution characteristics [27], relationships with ecological responses [26], and the construction of evaluation systems [28]. However, few studies have been conducted on the mechanisms that influence spatial variation in habitat quality. In terms of research methods, the use of the InVEST model to estimate habitat quality can improve the problems of human and material consumption and the low efficiency associated with field measurements. The assessment results obtained are spatially continuous and suitable for large study areas. Most methods to explore the factors influencing habitat quality use correlation analysis [29], GWR [30], and GeoDetector [14]; however, these methods cannot explore the different spatial scales of action of each influencing factor on habitat quality, and the MGWR model used in this study can fill this gap. In terms of research objects, most studies have focused on small and medium scales, such as rivers [31,32] and nature reserves [13,33], while there are few studies on habitat quality at large scales, such as cities, urban agglomerations, and provincial areas. Moreover, Guangdong Province is strategically located and is undergoing rapid development, and while it is facing many ecological, environmental, and resource challenges, most quantitative studies have focused on the Pearl River Delta and the Guangdong–Hong Kong–Macao Greater Bay Area [34,35], with less attention paid to Guangdong as a whole. In addition, research has mostly focused on single elements [4,31,36,37], and few studies have considered natural conditions, landscape patterns, social development, economic development, or other factors when studying habitat quality.

Therefore, based on the research of scholars at home and abroad, we first analyzed land-use changes based on five periods of land-use data from Guangdong Province in 2000, 2005, 2010, 2015, and 2020, using methods such as the land-use transfer matrix and land-use dynamic degree. The InVEST model was then used to map the spatial distribution and dynamics of habitat quality. Then, a system of indicators of factors influencing habitat quality was constructed based on climate, topography, vegetation, and socioeconomic data. Finally, the MGWR model was used to explore the spatial-scale differences in the roles of natural and human factors affecting habitat quality and the extent of their influence. This study aims to achieve the following: (1) analyze the spatial and temporal evolution characteristics of land-use changes in Guangdong Province over the past 20 years; (2) assess the habitat quality in Guangdong Province over the past 20 years and identify the characteristics of temporal and spatial changes in habitat quality in the study area; (3) quantitatively analyze the response of habitat quality to various influencing factors and reveal the patterns of habitat quality changes caused by natural factors and human activities; (4) highlight the contribution of geography in building an ecologically civilized and beautiful modern city and supplement and improve the relevant data and findings in the field of habitat quality study and impact factor research in Guangdong Province for the period 2000–2020; (5) provide a scientific reference and basis for urban planning, adjusting land-use types and spatial layout, protecting the regional ecological environment, maintaining regional ecological security, and achieving sustainable development in Guangdong Province.

2. Study Area

Guangdong Province, also known as “Yue”, is a province on the southern coast of mainland China with Guangzhou as its capital. It is located south of the Southern Ridge on the shore of the South China Sea (Figure 1).

In terms of economy, Guangdong Province has become the number one economic province in China, accounting for 1/8 of the country’s total economic output. In addition, it has surpassed Hong Kong and Taiwan to become the province with the largest economy in China and has the strongest overall economic competitiveness and financial strength of all provinces, reaching the level of upper-middle-income and medium-developed countries [38]. In terms of ecology, Guangdong Province is located in a tropical and subtropical region with a complex and diverse topography, with the Nanling Mountains in the north and the South China Sea in the south. It is a relatively independent and complete ecosystem. The region has complex climatic conditions, including central subtropical, southern subtropical, and tropical monsoon climate zones, with long summers and warm winters, abundant rainfall, large streams, and long flood periods. On the other hand, meteorological disasters such as heavy rains and floods, tropical cyclones, strong convective weather, lightning strikes, and high temperatures occur frequently. These disasters have long cycles and high frequencies, and can cause severe damage [39]. The Pearl River Delta urban agglomeration (21°25′ N–24°30′ N, 111°12′ E–115°35′ E) is located within Guangdong Province, hereafter referred to as the “PRD”. It consists of Guangzhou, Shenzhen, Zhuhai, Foshan, Dongguan, Zhongshan, Huizhou, Jiangmen, and Zhaoqing, which are adjacent to the two Special Administrative Regions of Hong Kong and Macao. As the most densely populated urban agglomeration in the country, the ecosystem of the PRD has been fragmented by urbanization. Differences in population density and socioeconomic development have exacerbated the uneven development of the region, presenting significant spatial differences. Coordinating the demands of urbanization and maintaining ecosystem services are becoming increasingly important [40,41].

Owing to the complexity of landform types, the fragility of ecosystems, and the impact of human development and construction, the ecosystem of Guangdong Province faces some outstanding problems. Driven by urbanization, industrialization, and population agglomeration, the shares of construction, industrial, mining, and infrastructure land are increasing, which not only destroys the integrity of natural land [42,43,44,45] but also further aggravates habitat degradation and loss. These conditions seriously affect habitat quality and hinder sustainable economic development and the construction of an ecologically sustainable civilization [46]. For example, the hills and mountains in northern Guangdong Province are prone to soil erosion during typhoons and heavy rains, and the simple forest structure of Guangdong Province leads to ecological fragility [47]. However, other natural and social factors leading to increased ecosystem vulnerability remain unclear. Therefore, in Guangdong Province, it is necessary to promptly explore the factors influencing the ecological environmental quality, rationalize land planning, improve the ecological environmental quality, and promote sustainable development.

3. Data Sources and Analysis Methods

3.1. Research Data Sources

In this study, five periods of land-use data from Guangdong Province in 2000, 2005, 2010, 2015, and 2020 were sourced from the Resource and Environment Science Data Centre of the Chinese Academy of Sciences (http://www.resdc.cn, accessed on 1 March 2022), with a spatial resolution of 30 m. Based on the purpose of the study, land-use types were reclassified into arable land, forest land, grassland, water, wetland, construction land, and unused land. Road data were obtained from OpenStreetMap (https://download.geofabrik.de/, accessed on 1 March 2022). The average annual precipitation, average annual temperature, gross domestic product (GDP), population distribution data, and NDVI were obtained from the Resource and Environmental Sciences Data Centre of the Chinese Academy of Sciences at a spatial resolution of 1 km. Topographic data were obtained from the SRTM 90 M digital elevation model (DEM) product from geospatial data (http://www.gscloud.cn/, accessed on 1 March 2022), and the 1 km slope data of Guangdong Province were calculated from the 1 km DEM data of Guangdong Province.

Table 1. Dataset and data sources.

Dataset Names	Data Description	Data Sources
Land-use dataset	30 m land-use data for Guangdong Province	Resource and Environmental Sciences Data Centre of the Chinese Academy of Sciences (http://www.resdc.cn, accessed on 1 March 2022)
Road dataset	Guangdong Province Road Data	OpenStreetMap (https://download.geofabrik.de/, accessed on 1 March 2022)
Natural environment dataset	1 km annual average precipitation data for Guangdong Province	Resource and Environmental Sciences Data Centre of the Chinese Academy of Sciences (http://www.resdc.cn, accessed on 1 March 2022)
	1 km annual average temperature data for Guangdong Province
	Guangdong Province 90 m digital elevation model (DEM) data	Geospatial Data Cloud (http://www.gscloud.cn/, accessed on 1 March 2022)
	Guangdong Province 1 km slope data	Calculated from Guangdong DEM data
	1 km normalized difference vegetation index (NDVI) for Guangdong Province	Resource and Environmental Sciences Data Centre of the Chinese Academy of Sciences (http://www.resdc.cn, accessed on 1 March 2022)
Socioeconomic dataset	Guangdong 1 km population distribution data	Resource and Environmental Sciences Data Centre of the Chinese Academy of Sciences (http://www.resdc.cn, accessed on 1 March 2022), Guangdong Statistical Yearbook
	Guangdong 1 km GDP data

provides the specific descriptions of the dataset.

3.2. Analysis Methods

This study is based on five time points of land-use data from Guangdong Province in 2000, 2005, 2010, 2015, and 2020, and calculates the land-use transfer matrix and land-use dynamics to analyze land-use changes. The habitat quality model implemented in InVEST model version 3.8.7 was used to assess habitat quality, explore the spatial and temporal variation characteristics of habitat quality, and analyze the impact of land-use changes on habitat quality. To further analyze the natural environment and socioeconomic factors influencing habitat quality, the 2015 habitat quality, climate, topography, vegetation, and socioeconomic data were used, with counties as the units of analysis. The raster data were tabulated by county administrative divisions, and the influencing factors were screened through correlation analysis and multiple linear regression analysis using SPSS Statistics 25 software to construct a habitat quality influencing factor index system. To better serve the construction of regional ecological protection and sustainable development, MGWR-2.2 software was used to investigate the spatial scales of influencing factors, the extent to which they affect habitat quality, and the coupling mechanisms of habitat quality influencing factors.

3.2.1. Land-Use Transfer Matrix

The land-use data used in this study were raster data and the size of each pixel was 30 m × 30 m. Each land-use type has a unique code. The land-use transfer matrix clearly expresses the spatial and temporal evolution of the land-use status by analyzing the composition and transformation direction of each category at the beginning and end of the study period [48].

The land-use transfer matrix is derived from the quantitative description of system states and state transfers in system analysis and is in the general form shown in

Table 2. General form of the land-use transfer matrix.

${\mathit{T}}_2$						${\mathit{P}}_{\mathit{i}+}$	Reduction
		${\mathit{A}}_1$	${\mathit{A}}_2$	…	${\mathit{A}}_{\mathit{n}}$
$T_1$	$A_1$	$P_{11}$	$P_{12}$	…	$P_{1n}$	$P_{1+}$	$P_{1+}-P_{11}$
	$A_2$	$P_{21}$	$P_{22}$	…	$P_{2n}$	$P_{2+}$	$P_{2+}-P_{22}$
	…	…	…	…	…	…	…
	$A_n$	$P_{n1}$	$P_{n2}$	…	$P_{nn}$	$P_{n+}$	$P_{n+}-P_{nn}$
$P_{+j}$		$P_{+1}$	$P_{+2}$	…	$P_{+n}$	1
New		$P_{+1}-P_{11}$	$P_{+2}-P_{22}$	…	$P_{+n}-P_{nn}$

, where the rows indicate the land-use type at time $T_1$ and the columns indicate the land-use type at time $T_2$.

$P_{ij}$ indicates the percentage of the total land area converted from land type i to land type j during the period $T_1$ − $T_2$. $P_{ii}$ denotes the percentage of areas where land-use type i remained unchanged during the period $T_1$ − $T_2$. $P_{i+}$ denotes the percentage of the total area of land type i at time $T_1$. $P_{+j}$ denotes the percentage of the total area of land-use type j at point in time $T_2$. $P_{i+}-P_{ii}$ is the percentage reduction in the area of land type i during the period $T_1$ − $T_2$. $P_{+j}-P_{jj}$ is the percentage increase in the area of land class j during period $T_1$ − $T_2$.

3.2.2. Land-Use Dynamic Degrees

Land-use dynamics refer to quantitative changes between land-use types over a certain period of time. This indicator assesses the extent of land-use change caused by regional land-use transfer components. A positive (negative) value indicates a net increase (decrease) in the land area. The absolute value of the land-use dynamics reflects the degree of change in a given land-use type. Therefore, the higher the absolute value, the more dramatic the change in a particular land-use type [49].

$${LC}_1=\frac{U_b-U_a}{U_a}\times \frac{1}{T}\times 100\mathrm{\%}$$

(1)

where ${LC}_1$ is the dynamic degree of a particular land-use type, indicating the dynamic degree (magnitude of change) of a land type in the study area during the study period, T. $U_a$ is the base period quantity (initial quantity). $U_b$ is the final volume.

$$K_s=\left({\sum }_{j=1}^n{∆L}_{j-k}/2{\sum }_{j=1}^n{L}_j\right)\times \frac{1}{T}\times 100\%$$

(2)

where $K_s$ refers to the comprehensive land-use dynamic degree, ${∆L}_{j-k}$ denotes the absolute value of the total area converted from the $j$th land-use type to the $k$th land-use type in the research period, and $n$ is the number of land-use types [50].

The land-use reclassification criteria are shown in

Table 3. Criteria for reclassification of land-use/cover types.

Reclassification Codes	Place Name	Original Place Name and Code
1	Arable land	Paddy fields (11) and dry land (12)
2	Forest land	Forested land (21), shrubland (22), open forest land (23), and other forested land (24)
3	Grassland	High-, medium-, and low-cover grassland (31, 32, and 33)
4	Wetlands	Reservoir ponds (43), mudflats (45), mudflats (46), and marshes (64)
5	Water area	Rivers and canals (41), lakes (42), and seas and oceans (99)
6	Construction land	Urban sites (51), rural settlements (52), and other building sites (53)
7	Unused land	Sandy (61), Gobi (62), saline (63), bare ground (65), bare rocky ground (66), and other (67)

.

3.2.3. InVEST Model

This study used the habitat quality module of the InVEST model version 3.8.7 to assess habitat quality in Guangdong Province. The resultant habitat quality values calculated by this model range from 0 to 1; the better the habitat quality in the region, the higher the value obtained from the assessment.

Five necessary data points were entered to run the habitat quality module: land-use data for the study year, threat source layer data, threat factor table, habitat type, sensitivity table of habitat type to threat factors, and half-saturation constants. Three data points were used: habitat degradation, scarcity, and quality [51].

The calculation formula is as follows:

$$Q_{xj}=H_j\left[1-\frac{D_{xj}^z}{D_{xj}^z+k^z}\right]$$

(3)

where $Q_{xj}$ is the habitat quality in raster $x$ in land-use type $j$, k is the half-saturation parameter, usually taken as 1/2 of the maximum value, $H_j$ is the habitat suitability of land-use type j, $z$ is the normalized constant, usually taken as 2.5, and $D_{xj}$ is the stress level of raster x in land-use type j. The formula for calculating $D_{xj}$ is as follows:

$$D_{xj}={\sum }_{r=1}^R{\sum }_{y=1}^{Y_r}\left(\frac{w_r}{{\sum }_{r=1}^R{w}_r}\right)r_y{i}_{ryx}{\beta }_x{S}_{jr}$$

(4)

where $R$ represents the number of threat factors;$y$ is the grid layer of threat factors; $Y_r$ indicates the set of grid cells on r’s raster map; $w_r$ represents the weight of each threat r that originates in the grid cell; $r_y$ is the value of the stress factor for the raster y; ${\beta }_x$ is the accessibility level of the habitat grid x; $S_{jr}$ represents the sensitivity of land-use type j to threat factor r; and $i_{ryx}$ indicates stress factor value $r_y$ for the raster y stress level for habitat raster x.

Threat factors include biological factors (such as invasive alien species), abiotic factors (such as temperature and precipitation), and anthropogenic disturbances (such as land-use change and roads) [52]. Over the past two decades, urbanization and rapid population expansion in Guangdong Province have accelerated, with anthropogenic factors having a far greater impact on the environment than biological and abiotic factors. Therefore, this study focused on investigating the extent to which anthropogenic disturbance threats affect the environment. Owing to rapid industrialization and urbanization in Guangdong Province, artificial land-use types (cropland and urban construction land) have expanded extensively, causing a loss of ecosystem services and severely affecting biodiversity. Changes in artificial land-use types are a direct manifestation of anthropogenic disturbances to the ecosystem. Thus, artificial land use should be considered a threat factor. In addition, the distribution of road transportation networks also has irreversible impacts on habitat fragmentation.

Therefore, this study considers three land-use types—urban construction land, arable land, and unused land—and five road network categories—railroads, highways, trunk roads, primary roads, and secondary roads. The parameters of these threat factors include the weight, maximum impact distance, and decay function, which should be determined based on empirical values or professional knowledge. We determined the weight, maximum impact distance, and decay function for each threat factor based on relevant studies conducted in similar situations and the user guide of the InVEST model [4,52,53].

The sensitivity parameters included habitat suitability parameters for each land-use type and sensitivity parameters of each land-use type to stressors, all ranging from 0 to 1. These values should be determined based on empirical values and professional knowledge. Therefore, in this study, we set these parameters by referencing the literature from regions similar to the study area [39].

The specific parameters were set based on the habitat stress factors used in this study, as shown in

Table 4. Threat factors and their max distance of influence, weight, and type of decay over space.

Threat Factors	Weight	Max Distance of Influence/(km)	Type of Decay over Space
Unused land	0.2	3	Linear
Construction land	1	10	Index
Arable land	0.68	8	Linear
Railroads	0.9	9	Index
Highways	0.9	9	Index
Trunk roads	1	10	Index
Primary roads	1	8	Linear
Secondary roads	0.75	5	Linear

and

Table 5. The sensitivity of land-use types to threat factors for the habitat.

Land-Use Types	Habitat Suitability	Threat Factors
		Arable Land	Construction Land	Unused Land	Railroads	Highways	Trunk Roads	Primary Roads	Secondary Roads
Arable land	0.35	0	0.4	0.1	0.35	0.35	0.35	0.3	0.2
Forest land	1	0.3	0.8	0.2	0.75	0.75	0.75	0.7	0.6
Grassland	0.4	0.35	0.6	0.1	0.7	0.7	0.7	0.5	0.35
Wetland	1	0.3	0.85	0.3	0.8	0.8	0.8	0.75	0.65
Water area	0.9	0.3	0.9	0.5	0.5	0.5	0.5	0.45	0.3
Construction land	0	0.4	0	0.3	0.6	0.6	0.6	0.5	0.5
Unused land	0	0.4	0.5	0	0.1	0.1	0.1	0.1	0.1

.

3.2.4. MGWR Model

Traditional GWR models compensate for the local limitations of global regression models that do not capture spatially nonstationary relationships among variables by embedding information on the spatial locations of multiple variables [18]. For some variables, the regression coefficients may be smooth (“constant in space”), while for other explanatory variables, the regression coefficients may be non-smooth (“varying with space”) [54,55,56]. Introducing all explanatory variables with nonstationary regression coefficients into the GWR model may reduce the uncertainty in the final prediction due to model fitting [27,54,57]. To address these limitations of traditional GWR, Oshan et al. proposed the MGWR model in 2017 [58], which was further refined and improved using the local parameter statistical inference of Yu et al. in 2019 [59].

The MGWR model can specify exclusive bandwidths for different regional variables to measure multiscale effects during the model calibration process [16]. In contrast to GWR, the MGWR model considers geographic scales when exploring spatial heterogeneity, which improves the model in various ways by allowing it to distinguish between the scales of different variables.

The MGWR model is calculated as follows:

$$y_i={\sum }_{j=1}^k{\beta }_{bwj}(u_i,v_i)x_{ij}+{\epsilon }_i$$

(5)

where $y_i$ is the habitat quality of the i-th sample point, $(u_i,v_i)$ is the spatial coordinate of the i-th sample point, j is the number of independent variables, $x_{ij}$ is the observed value of variable j at i, ${\epsilon }_i$ is the random error term, and $bwj$ represents the bandwidth used for the regression coefficient of the j-th variable [37]. Each regression coefficient of the MGWR ${\beta }_{bwj}$ is based on local regression, and each bandwidth is specific, which is the largest difference from classical GWR.

In this study, we used MGWR-2.2 software, which was developed by the Spatial Analysis Research Center (SPARC) of Arizona State University, to estimate the GWR and MGWR, while the maps were created using ArcGIS 10.2. The impact factors affecting habitat quality were considered in terms of natural and social environments, and with reference to other studies on factors influencing habitat quality, we discovered that economic development, population density, built-up land area, vegetation cover, altitude [60], and traffic conditions [61] were important factors affecting habitat quality. To investigate the influence of the natural environment and social economy on habitat quality, climatic, topographic, vegetation, and economic factors were selected, including the average annual precipitation, average annual temperature, average elevation, slope, NDVI, GDP per unit area, population density, and road density [60,61] in 2015, which were considered as influencing factors for the 2015 Guangdong habitat quality model. Next, habitat quality data and data on various influencing factors were extracted by administrative division (county level) and organized into tables as inputs for the model. This facilitated an exploration of the spatial scale and degree of influence of each factor on habitat quality, as well as an analysis of the coupled mechanisms of habitat quality influencing factors in Guangdong Province. The specific indicator systems are listed in

Table 6. Habitat quality impact factor indicator system.

Tier 1 Indicators	Secondary Indicators	Tertiary Indicators
Natural environment	Climate factors	Average annual precipitation
		Average annual temperature
	Terrain factor	Average altitude
		Slope
	Vegetation factor	Normalized difference vegetation index (NDVI)
Socioeconomic	Economic factors	GDP per unit area
		Population density
		Road density

.

4. Results

4.1. Analysis of Land-Use Changes

According to Figure 2 and

Table 7. Land-use transfer matrix.

Time	Type of Land Use	$\mathbf{Arable}\mathbf{Land}/{\mathbf{k}\mathbf{m}}^2$	$\mathbf{Forest}\mathbf{Land}/{\mathbf{k}\mathbf{m}}^2$	$\mathbf{Grassland}/{\mathbf{k}\mathbf{m}}^2$	Wetland / ${\mathbf{k}\mathbf{m}}^2$	$\mathbf{Water}\mathbf{Area}/{\mathbf{k}\mathbf{m}}^2$	$\mathbf{Construction}\mathbf{Land}/{\mathbf{k}\mathbf{m}}^2$	$\mathbf{Unused}\mathbf{Land}/{\mathbf{k}\mathbf{m}}^2$	Total / ${\mathbf{k}\mathbf{m}}^2$	Percentage
2000–2020	Arable land	22,355.57	13,810.81	1440.03	1537.50	719.76	5372.38	34.80	45,270.85	25.71%
	Forest land	13,487.85	86,140.52	3570.73	1174.72	421.13	2653.25	14.72	107,462.91	61.03%
	Grassland	1509.15	3443.37	2262.08	128.65	49.61	304.29	8.93	7706.08	4.38%
	Wetland	1413.00	1011.10	104.91	1619.83	188.17	898.92	5.30	5241.23	2.98%
	Water area	589.50	384.71	53.76	177.19	574.47	342.45	2.19	2124.26	1.21%
	Construction land	2712.81	1205.11	142.60	320.05	232.95	3540.62	8.90	8163.05	4.64%
	Unused land	36.24	21.61	7.63	13.59	3.66	17.19	24.19	124.12	0.07%
	Total	42,104.12	106,017.24	7581.74	4971.53	2189.76	13,129.09	99.02	176,092.50	100%
	Percentage	23.91%	60.21%	4.31%	2.82%	1.24%	7.46%	0.06%	100%

, the land-use types in Guangdong Province from 2000 to 2020 were mainly arable and forest land, accounting for 25.71% and 61.03% of the total area in 2000 and 23.91% and 60.21% in 2020, respectively, with different degrees of decline in proportion over the 20-year period. Arable land was mainly distributed near water systems, with a high distribution in all areas except for Dongguan and Shenzhen. Over the past 20 years, the area of arable land has decreased by 3166.73 km², with a large amount of arable land being converted into land for construction. Forest land is mainly distributed in the peripheral areas of Guangdong Province, with greater distribution in Zhaoqing, Jiangmen, Huizhou, and the northern part of Guangzhou. The area of forest land was reduced by approximately 1445.69 km², with construction land accounting for a larger proportion of converted land. Grassland was scattered throughout the municipalities with a decrease in proportion and a total decrease in area of 114.34 km², and the largest proportion of land transferred to grassland was forest land. Guangdong Province is located at the mouth of the Pearl River system, with many rivers and widely distributed water bodies, most of which have been gradually replaced by arable land, forest land, and construction land, with a total reduction of approximately 996.37 km². Wetlands are mainly located near water bodies, and in 2000 they were widely distributed in Foshan and gradually replaced by arable land and forest land, with a total reduction of approximately 269.7 km². The total reduction in the area of unused land was approximately 25.1 km², and the largest proportion of land transferred to unused land was arable land. In 2000, the construction land was concentrated in Zhuhai, Zhongshan, Foshan, Guangzhou, Dongguan, and Shenzhen. During the past two decades, the rapid economic development of Guangdong Province has led to a rapid increase in the construction land area. Foshan, Guangzhou, Dongguan, and Shenzhen showed more significant growth in construction land and a trend of encirclement along the PRD, with a total increase of approximately 4966.04 km².

The single land-use dynamic degree of Guangdong Province from 2000 to 2020 was calculated based on the land-use area for each year. As shown in

Table 8. Single land-use dynamic degrees.

Type of Land	Single Land-Use Dynamic Degree
	2000–2005	2005–2010	2010–2015	2015–2020
Arable land	−0.75%	−0.31%	−0.20%	−0.06%
Forest land	−0.04%	−0.01%	−0.14%	−0.03%
Grassland	−0.58%	−0.65%	0.93%	0.00%
Wetlands	−0.11%	0.07%	−0.43%	−0.84%
Water area	−0.47%	1.07%	0.20%	0.00%
Construction land	5.44%	1.71%	1.68%	0.89%
Unused land	−1.34%	−2.62%	−0.36%	−1.40%

, between 2000 and 2005, construction, unused, and arable land changed more drastically, whereas other land-use types such as forest land and wetland changed more steadily. Simultaneously, the overall dynamic degree of each land-use type in this time period was high compared to other time periods, indicating that the rate of land-use changes in Guangdong Province accelerated significantly and the scale of conscious human changes in land-use types expanded. The absolute values of the dynamic degree of unused land, construction land, and watersheds were greater than those of other land-use types in 2005–2010, and there was almost no change in forest land and wetlands over the five-year period. The absolute values of the dynamic degree of construction land and grassland were greater than those of the other land-use types during the study period of 2010–2015 to 2015–2020. The absolute values of the dynamic degrees of cropland, forest land, grassland, and watersheds decreased and then changed steadily during the study period. The absolute values of the dynamic degrees of the unused land, construction land, and wetlands were higher. In general, over the 20-year period, construction land continued to increase, whereas arable, forest, and unused land continued to decrease, and all other land-use types changed to varying degrees.

According to

Table 9. Integrated land-use dynamic degrees.

Year	Integrated Land-Use Dynamic Degree
2000–2005	0.25%
2005–2010	0.12%
2010–2015	0.15%
2015–2020	0.06%
2000–2020	0.13%

, the most dramatic land-use change can be seen in the period of 2000–2005 with a 0.25% dynamic degree, followed by the periods 2005–2010 and 2010–2015, with 0.12% and 0.15% dynamics, respectively; the relatively weaker land-use change is over 2015–2020, with 0.06%.

4.2. Analysis of the Spatial and Temporal Patterns of Habitat Quality Evolution

Using the InVEST model, a raster-based spatial distribution of habitat quality in Guangdong Province was obtained. To facilitate a comparison of the spatial and temporal characteristics of habitat quality, the habitat quality index was divided into five classes according to the equal interval method, considering the actual situation in the study area. As shown in Figure 3 and

Table 10. Rate of change in habitat quality.

Year	Rate of Change in Habitat Quality
2000–2005	−0.63%
2005–2010	−1.14%
2010–2015	0.96%
2015–2020	−1.92%
2000–2020	−2.73%

, the overall habitat quality in Guangdong Province showed a decreasing trend from 2000 to 2020 of 0.7051 to 0.6859, with an annual rate of change of −2.73%. In 2010–2015 a small increase was found, and habitat quality increased slightly with a rate of change of 0.96%, whereas the other years showed different degrees of decrease, with the largest decrease from 2015 to 2020 with a rate of change of −1.92%. It has also been shown that land-use changes can significantly affect habitat quality in Guangdong Province [39]. Land is a carrier of human activities that can directly record patterns of land-use change driven by the social economy. Given the premise that geographical conditions are not easily changed in the short term, land-use change alters the pattern and function of the regional habitat and affects material circulation and energy flow within the habitat, ultimately causing changes in habitat quality. When land-use types and structures change at an accelerated rate, habitat fragmentation, degradation, and even loss can occur, leading to a continuous decline in habitat quality [62,63,64,65,66]. Combining the results of land-use dynamic degrees in Table 6 and Table 7, the land-type stressors for habitat quality included construction land, which had a higher rate of change than cropland and unused land during 2000–2020. Therefore, we assumed that habitat quality is mainly influenced by the type of land used, particularly construction land.

In terms of spatial distribution, areas with low habitat quality values were concentrated in the urban areas of the PRD, the Chaoshan region, and the coastal areas of Zhanjiang in Maoming, whereas areas with high habitat quality values were mainly located in the non-coastal areas of eastern and western Guangdong and the northern regions of Guangdong. The habitat quality trend in Guangdong Province generally increased from the coastal to the inland regions, with an abnormal decrease in the Sui-Fo–Dongguan–Shenzhen region in the PRD economic zone. In addition, the habitat quality in areas near roads mostly decreased in a linear or patchy manner. Combined with the spatial distribution map of land use, it can be seen that areas with low habitat quality values are mainly located in areas with a high distribution of construction land, such as the PRD, where the economy and transportation are well developed, the population is dense, there are fewer forested areas and wetlands, and the altitude is lower. Economic development, population density, land area under construction, vegetation cover, altitude, and road network density are important factors affecting habitat quality.

4.3. Study of Factors Influencing Habitat Quality Based on the MGWR Model

4.3.1. Model Variable Screening and Elimination

Correlations between mean annual precipitation, mean annual temperature, mean elevation, slope, NDVI, GDP per unit area, population density, road density, and habitat quality were analyzed separately using the SPSS Statistics 25 software. As shown in

Table 11. Correlation analysis of factors influencing habitat quality.

Influencing Factors	Average Annual Precipitation	Average Annual Temperature	Average Altitude	Slope	Normalized Difference Vegetation Index	GDP per Unit Area	Population Density	Road Density
Correlation coefficient	0.382 **	0.140	0.743 **	0.758 **	0.903 **	−0.452 **	−0.669 **	−0.760 **

Note: **. The correlation was significant at 0.01 level (two-tailed).

, the mean annual precipitation, mean elevation, slope, and NDVI were significantly and positively correlated with habitat quality (p < 0.01) and negatively correlated with GDP per unit area, population density, and road density (p < 0.01). There was no significant correlation between mean annual temperature and habitat quality. Therefore, mean annual precipitation, mean elevation, slope, NDVI, GDP per unit area, population density, and road density were selected as explanatory variables for habitat quality, and variables with no significant correlation were excluded.

To avoid bias in model prediction caused by multicollinearity in the explanatory variables and to obtain better model-fitting results, this study used habitat quality as the dependent variable and mean annual precipitation, mean annual temperature, mean elevation, slope, normalized vegetation index, GDP per unit area, population density, and road density as independent variables for multiple linear regression analysis using SPSS Statistics 25 software. The variance inflation factor (VIF) was chosen as an indicator to determine whether collinearity existed among the factors. A VIFmax < 10 [67] indicated that there was no significant linear overlap, and explanatory variables with collinearity were excluded. F-values were observed separately to test whether the overall regression model was valid and whether the null hypothesis that all coefficients were zero was rejected.

From

Table 12. Multiple linear regression analysis of influencing factors and habitat quality.

Dependent Variable	Independent Variable	Coefficient	VIF Statistics	F-Test		Adjusted R2
				F-Value	p-Value
Habitat quality	Constant term	−0.152		179.946	$<$0.01	0.923
	Average annual precipitation	2.056 $\times {10}^{-6}$	2.727
	Average annual temperature	0.002	2.032
	Average altitude	2.760 $\times {10}^{-5}$	6.435
	Slope	0.036	7.062
	Normalized difference vegetation index	0.533	4.175
	GDP per unit area	2.735 $\times {10}^{-7}$	3.377
	Population density	1.273 $\times {10}^{-5}$	14.771
	Road density	−108.352	14.411

, the adjusted R² = 0.923 indicates that in this equation, the parameters can explain 92.3% of the habitat quality, and the regression effect is excellent. The p-value of the regression equation was <0.01, F = 179.946; the coefficient of variance expansion of population density and road density was >10, while the tolerance was <0.2, indicating multicollinearity. Therefore, some of the variables were excluded and the regression analysis was conducted again.

A comparison of the linear regression results for different combinations of influencing factors revealed that the best regression results were obtained after excluding two influencing factors, average annual temperature and population density. From

Table 13. Multiple linear regression analysis of the screened influencing factors and habitat quality.

Dependent Variable	Independent Variable	Coefficient	VIF Statistics	F-Test		Adjusted R2
				F-Value	p-Value
Habitat quality	Constant term	0.083		211.261	$<$0.01	0.914
	Average annual precipitation	4.946 $\times {10}^{-6}$	1.403
	Average altitude	2.414 $\times {10}^{-5}$	6.048
	Slope	0.028	6.558
	Normalized difference vegetation index	0.639	3.517
	GDP per unit area	4.145 $\times {10}^{-7}$	2.723
	Road density	−72.723	3.846

VIF, variance inflation factor.

, the adjusted R² = 0.914 indicates that, in this equation, the parameters can explain 91.4% of the habitat quality, and the regression effect is excellent. The p-value of the regression equation was <0.01, F = 211.261. The variance expansion coefficients of the influencing factors were all < 7, indicating that there is no obvious multicollinearity in such factors, the average annual precipitation, mean elevation, slope, normalized vegetation index, GDP per unit area, and road density were selected as the final explanatory variables for the habitat quality simulation, and the habitat quality regression model was

$$y=4.946\times {10}^{-6}{x}_1+2.414\times {10}^{-5}{x}_2+0.028x_3+0.639x_4+4.145\times {10}^{-7}{x}_5-72.723x_6+0.083$$

(6)

where the average annual precipitation is $x_1$, average altitude is $x_2$, slope is $x_3$, normalized vegetation index is $x_4$, GDP per unit area is $x_5$, road density is $x_6$, and the constant term is $0.083$.

4.3.2. Comparison of GWR and MGWR Models

Guangdong habitat quality was used as the explained variable and average annual precipitation, mean elevation, slope, NDVI, GDP per unit area, and road density were used as explanatory variables. The traditional GWR and MGWR models were constructed for each variable, and the model diagnostic information is presented in

Table 14. Comparison of model regression results.

Models	GWR	MGWR
RSS	5.473	5.459
AIC	12.412	8.244
AICc	22.027	16.067
R2	0.954	0.955
Adjusted R2	0.945	0.946

. Compared with the traditional GWR model, the MGWR model, the residual sum of squares (RSS), Akaike information criterion (AIC), and AICc (for small samples, AIC is transformed to AICc) values of the results were smaller, and the R² of the MGWR model and the adjusted R² were higher than those of the GWR model, indicating that the fitting effect of MGWR was better than that of the traditional GWR model; therefore, MGWR was selected as the model for habitat quality impact factors. The explanatory power of MGWR for habitat quality in Guangdong Province was as high as 95.5%, indicating that habitat quality in Guangdong Province can be modeled using the selected influencing factors.

4.3.3. Exploring the Spatial Scale of Habitat Quality Impact Factors

The six influencing factors, namely GDP unit area, NDVI, average annual precipitation, slope, mean elevation, and road density, were entered into MGWR 2.0 to produce the optimal bandwidth and standard deviation of the parameter estimates generated by MGWR, as shown in Figure 4. This determines the spatial scale of the influencing factors and the extent to which they affect the habitat quality.

The black histogram in Figure 4 represents the optimal bandwidth for each variable generated by the MGWR model, the gray histogram represents the standard deviation of the parameter estimates for each variable, and the solid black line represents the individual optimal bandwidths derived from MGWR (119). MGWR can be run at different spatial scales. Larger bandwidths indicate that variables affect habitat quality similarly in space: the larger the bandwidth, the less the spatial heterogeneity, and therefore, the smaller the standard deviation of the parameter estimates. In contrast, variables with smaller bandwidths affect habitat quality at local scales and thus have larger standard deviations in parameter estimates. The mean elevation, GDP per unit area, and NDVI bandwidths of 53, 58, and 47, respectively, only affected habitat quality at local scales and exhibited greater spatial nonstationarity at spatial scales. In contrast, the optimal bandwidths for mean annual precipitation, slope, and road density were 116, 119, and 119, respectively, which are close to the total sample size (120) eventually used by the MGWR model; therefore, these three variables have an impact on habitat quality in Guangdong Province on a global scale.

4.3.4. Exploration of the Intensity of Influence of Habitat Quality Impact Factors

Table 15. Regression coefficients and significance evaluation for each impact factor (MGWR).

Variables	Regression Coefficients for MGWR			Proportion of Study Units (Based on t-Test)/%
	Average	Minimum Value	Maximum Value	$\mathit{p}\le 0.1$	$+$	$-$
GDP per unit area	0.197	−0.049	0.457	95	99.17	0.83
Normalized difference vegetation index	0.522	0.342	0.626	100	100	0
Average annual precipitation	0.039	0.033	0.046	0	100	0
Slope	0.132	0.121	0.147	0	100	0
Average altitude	0.088	−0.049	0.242	22.5	73.33	26.67
Road density	−0.465	−0.468	−0.463	100	0	100

summarizes the statistical information on the parameter estimates generated by the MGWR, including the minimum (min), maximum (max), and mean values of the coefficient estimates for each variable. The proportions of study units with regression coefficients > 0 and <0 are shown. Among the natural elements affecting habitat quality in Guangdong Province, the GDP per unit area and road density were significant for all study units. Among the natural elements, the NDVI had the most significant effect, whereas slope, average annual precipitation, and mean elevation did not significantly affect habitat quality.

4.3.5. Spatial Heterogeneity Analysis of Impact Factors

When analyzing the spatial heterogeneity of the factors influencing habitat quality based on the spatial visualization method, only significant independent variables (p < 0.1) were selected to portray the spatial distribution pattern of their influence on habitat quality. As shown in Figure 5, the effects of mean elevation, GDP per unit area, normalized vegetation, and road density on habitat quality showed significant spatial heterogeneity.

The regression coefficient of the constant term ranged from −0.133170 to 0.175606, with positive values mainly in Zhaoqing City and negative values in Zhanjiang and Maoming City in southwestern Guangdong Province. The regression coefficients changed from negative to positive values from southwest to northeast, indicating that the influence of factors other than those selected on habitat quality was positive in Zhaoqing, Heyuan, and Meizhou and negative in Zhanjiang and Maoming.

The regression coefficients for the mean elevation ranged from 0.162610 to 0.241542 and showed a gradual decrease from south to north in the areas where data were available, with all positive values. This indicates that higher elevations contribute to habitat quality to a certain extent, and the more southward the elevation, the more significant the contribution.

The regression coefficients for the GDP per unit area ranged from 0.125618 to 0.457423, and the overall spatial pattern showed a decreasing trend from coastal to inland and an increasing trend from southwest to northeast. High values were concentrated in Jieyang and Chaoshan on the eastern coast of Guangdong Province, indicating that the increase in GDP per unit area in this region contributed the most to the increase in habitat quality and suggesting that these regions have invested more in ecological protection and achieved certain results.

The NDVI regression coefficients ranged from 0.342145 to 0.625891, with high overall values. High values are concentrated in the central and southwestern regions, whereas low values are concentrated in the northeastern region. This indicates that an increase in NDVI has a greater positive effect on habitat quality than the other influencing factors. Vegetation cover is an important parameter for assessing habitat quality; in general, areas with high vegetation cover have relatively good habitat quality. In recent years, the government has consciously built national forest parks, focusing on creating and protecting the local ecological environment, so that the remaining forest resources in Guangdong Province can be restored and developed.

The regression coefficient of road density ranges from −0.468070 to −0.462777, with a small range of variation and negative values. In terms of spatial pattern distribution, the absolute values of the regression coefficients of road density were centered in the PRD region, with a tendency to increase radially from the coastal areas of the PRD to the inland areas, with the highest absolute values located in southwest Guangdong Province. This indicates that the increase in road density had a more significant negative effect on habitat quality, with the most negative effect observed in the southwest region. Simultaneously, the bandwidth of road density was larger; therefore, this factor affected habitat quality on a larger scale, with less spatial heterogeneity. Road density reflects, to a certain extent, the intensity of human activity. The coastal areas of Guangdong Province and sea entrance areas of the PRD have a higher density of road networks, which also have a greater extent of hill modification and vegetation destruction. In recent years, a number of “cross-river links” have been established on the Pearl River, which have fully stimulated population circulation and economic development on both sides of the river but also brought about certain ecological and environmental problems.

5. Discussion

Combined with changes in policy and economic development over the past 20 years, the economic development model in Guangdong Province has gradually shifted from a crude model to a quality–efficiency one, with an increasing emphasis on protecting and restoring the ecological environment along with economic development. This study showed that areas with low habitat quality values were mainly located in areas with a high distribution of construction land, which are economically and logistically developed, densely populated, and have fewer forested areas and wetlands, which is consistent with the findings of Li et al. [68] and Yang et al. [30]. Economic development, road network density, population density, building land area, vegetation cover, and elevation were assumed to be the important factors affecting habitat quality. Lopes et al. showed that anthropogenic behavior influences habitat quality changes [33,69], which is consistent with the socioeconomic factors selected in this study. Owing to the lack of a full understanding of the importance of ecological protection in earlier years, the economic development of Guangdong Province was not balanced with ecological protection; thus, it incurred huge environmental costs, which led to a fairly large decrease in habitat quality and the regression of ecological functions. A moderate adjustment in population density, an increase in urban greenspace construction, the protection of existing forested areas and wetlands, scientific planning, the adjustment of land-use types, and spatial layout can improve habitat quality and the general ecological environment of Guangdong Province to a certain extent.

One study found that land-use changes had the strongest effect on habitat quality, followed by precipitation and vegetation cover, whereas elevation, slope, GDP, and population density had the weakest effects [60]. In contrast, this study found that NDVI had the strongest effect on habitat quality in 2015, followed by road density, GDP per unit area, slope, and average elevation, and had the weakest effect on average annual precipitation. Moreover, the normalized vegetation index and road density were the main influencing factors in areas with higher values of habitat quality; however, in areas with low values of habitat quality, GDP per unit area was the main influencing factor, indicating that the influencing factors and the degree of influence of habitat quality vary from year to year and from region to region. Unlike the results of Cui et al. on the habitat of the agricultural–pastoral mosaic zone of northern Shaanxi, China [60], our results suggest that precipitation and vegetation cover in Guangdong Province are stronger drivers than elevation or slope. Because the study area is in the southern coastal region, where precipitation is abundant, the effect of precipitation on the local ecology is less significant. In addition, slope and elevation influence the location of human living and production activities, which in turn drive the spatial differentiation of habitat quality [70]. High-altitude areas have less human activity, which improves habitat quality to a certain extent. Areas with denser road networks in the mountains have a greater degree of modification and vegetation damage, indicating a contradiction between human development, construction, and environmental protection. Population growth, intensive agricultural production, and industrial development have led to extensive household, soil, and industrial pollution, and the increasingly dense transportation networks in coastal areas have led to excessive fragmentation of natural ecosystems. Natural and socioeconomic factors drive changes in habitat quality, and the correlations between influencing factors are complex and interactive.

To maintain biodiversity and build cities in line with the principles of ecological civilization, we should increase economic investment in ecological protection, build ecological protection zones, and make efforts to create national forest parks to protect local environments so that forest resources can be restored and developed. While developing the economy, we should consider the scope of transformation of the natural environment and the extent of vegetation destruction, reasonably plan the land for construction, and adhere to the development strategy of “adapting to local conditions and time”. Simultaneously, we should combine the characteristics of regional economic and social development and gradually build a phased and deeply innovative development plan, thus promoting the transition from a sloppy economic growth mode in Guangdong Province and across the entire country to realize regional green and high-quality development.

However, this study has certain limitations. (1) The InVEST model requires a variety of parameters, such as maximum impact distance, original threat weight, and habitat sensitivity, which can affect the evaluation results. The relevant parameters of the InVEST model used in this study were set according to the literature, which lacks a uniform standard reference and may lead to differences in the results. (2) To obtain a better model fit and a higher model run rate, the county was selected as the unit of analysis in this study. A more refined study of the local area of Guangdong Province could be conducted if a smaller-scale analysis was conducted. (3) Changes in habitat quality are the result of many factors such as the normalized water index (NDWI), natural area protection range, land-use changes [2,32], and landscape fragmentation degree [26,37]. However, we failed to obtain these data, which should be improved in future studies.

Therefore, we should further explore the optimal parameters of the InVEST model, select multiple time periods, obtain more extensive and comprehensive data on the influencing factors, construct a multidimensional index system, explore spatial and temporal differences in the factors influencing habitat quality, and clarify the main factors and mechanisms affecting habitat quality across different time periods. In this way, we can better facilitate the spatial planning of land use and regional ecological protection, thereby providing a more comprehensive scientific reference and basis for urban planning, adjusting land-use types and spatial layouts, protecting the regional ecological environment, and maintaining regional ecological security.

6. Conclusions

Land-use changes between 2000 and 2015 were not obvious, while changes between 2015 and 2020 were more drastic. Owing to the rapid population growth and urbanization in Guangdong Province in recent years, the conversion of arable land and forest land into construction land over the last five years has caused greater damage to habitats. Construction land is mainly distributed in economically developed areas such as Guangzhou, Dongguan, and Shenzhen and shows a trend of encircling the coastal areas of the PRD.

From 2000 to 2020, the overall habitat quality in Guangdong Province decreased, and the areas with low habitat quality values were concentrated in the PRD region and the coastal areas of Chaoshan, Maoming, and Zhanjiang, whereas the areas with higher habitat quality values were mainly located in the non-coastal areas east and west of Guangdong and north of Guangdong.

Average annual precipitation, elevation, slope, normalized difference vegetation index, GDP per unit area, and road density were the main factors influencing habitat quality in Guangdong Province. According to the spatial distribution of the regression coefficients of each influencing factor, high elevations in Guangdong Province promoted habitat quality to a certain extent, and the more southward, the more significant the effect. The increase in GDP per unit area promoted an increase in habitat quality; the degree of influence showed a decreasing trend from coastal to inland areas and from southwest to northeast. The increase in the normalized vegetation index had a positive effect on habitat quality, with a greater effect in the central and southwestern regions and a smaller effect in the northeastern region. The overall effect was greater than that of other factors. The increase in road density had a more significant negative effect on habitat quality, showing a gradual radial increase from the coastal area of the PRD to the inland area. The negative effect of road density on habitat quality was greatest in the southwestern part of Guangdong Province, while it was lowest in the PRD. At the same time, this factor influenced the variation in habitat quality on a larger scale with less spatial heterogeneity.