Causes of Changing Woodland Landscape Patterns in Southern China

Abstract

Forests are composed of landscape spatial units (patches) of different sizes, shapes, and characteristics. The forest landscape pattern and its trends are closely related to resistance to disturbance, restoration, stability, and the biodiversity of the forest landscape and directly influence the benefits and sustainable exploitation of forest landscape resources. Therefore, forest landscape patterns and the driving forces have increasingly attracted the attention of researchers. The present study analyzed the spatial and temporal dynamics of woodland landscape patterns in typical hilly mountainous areas in southern China using ArcGIS, landscape pattern index, and morphological spatial pattern analysis. Meanwhile, a logistic regression model was used to analyze the drivers of woodland change in Anyuan County from three aspects: natural, geographic location, and socio-economic conditions. The total area of woodland decreased during the 10-year study period, with a net decrease of 4959.27 ha, mainly due to conversion into cultivated land, garden land, and construction land. Patch density, edge density, and aggregation index of woodlands increased over time, indicating enhanced fragmentation, stable and complex patch edges, and increased patch connectivity. Conversely, the highest patch index values exhibited decreasing trends, indicating decreases in the dominant patch type. Morphological spatial pattern analysis results showed that the core area was dominant and the islet area increased over time, which also indicates enhanced fragmentation. Forest landscape change is the result of environmental change, ecological processes, and human disturbance, with geographical location and social economy having greater influences on forest landscape change. Human activities such as navel orange cultivation, returning cultivated land to forest, and land occupation for construction were the major factors driving woodland change. The results provide reference that could facilitate forest management and sustainable forest resource utilization.

1. Introduction

Forests play fundamental roles in the maintenance of terrestrial ecosystems, safeguarding of basic human living conditions, and mitigation of climate change; in addition, they are essential for sustainable socio-economic development [1,2]. As the carrier of forest resources, forest land is not only the basis for sustainable forestry development but also a prerequisite for the maintenance of the ecological services of forests [3,4,5]. Different forest landscape systems are observed under various forest ecological types, and geographical, climatic, and socio-economic conditions, leading to distinct and diverse forest landscape mosaics [6,7,8].

The study of forest landscape structure, function, the dynamics and mechanisms of mutual influence and control, and the rational spatial configuration according to the characteristics of different ecosystems helps to reveal the relationship between the natural environment and human society. It can also facilitate the maximization the production potential of natural resources [9,10,11,12]. Consequently, forest landscape patterns have increasingly attracted the attention of various stakeholders, particularly changes in forest landscape patterns and the driving forces over the long term [13].

Current studies on forest landscape patterns have focused on heterogeneity over time and space. For example, Wu et al. quantified the degree of human interference by constructing the Hemeroby index (HI) and explored the spatio-temporal characteristics of landscape patterns and human interference in the course of urbanization [14]. Coops et al. combined remote sensing techniques with landscape pattern indices to analyze changes in forest fragmentation following beetle infestation over time [15]. However, landscape pattern research mainly analyzes the type, quantity, and spatial distribution and configuration of landscape units, which are largely limited to two-dimensional planes [16]. In contrast, the introduction of morphological spatial pattern analysis (MSPA) into forest landscape analysis can facilitate the rapid and intuitive identification of spatial morphology [17], which is conducive for the optimization of the ecological security barrier system, establishment of ecological corridors and biodiversity conservation networks, forest ecological network construction, and decision-making and data support in forest resource management [18,19,20]. Therefore, the application of MSPA to landscape ecology is increasing.

The forest landscape pattern refers to the composition and spatial distribution of forest in the landscape, reflecting landscape heterogeneity, and is the result of interactions between natural and human factors across space and time [21]. It is of great scientific significance to clarify the characteristics and driving forces of spatial and temporal changes in forest landscape, which can facilitate the understanding of shifts in regional ecological environment, rational use of resources, and harmonization of the relationship between people and land [22]. Therefore, researchers have applied different variables to explain the forces that drive woodland change; however, previous analyses often ignore the driving influence of socio-economic factors or the spatial heterogeneity of institutions (policies) on land use change, making it difficult to objectively and accurately reveal the causes of woodland change [23]. As for the research methodology, the influence of each driving factor in different geographical locations is mainly obtained through Moran’s I index analysis [24], geographically weighted regression and principal component analysis [25]; however, at present, there are fewer studies that involve exploring the influence of each factor on the degree of land use from the overall to the local level [26]. Logistic regression models can not only take account of the spatial heterogeneity of land use change drivers but can also quantify the relationship between regional land use dynamics and drivers, such as social, economic, technological, policy, and natural environment factors [24]. In recent years, some scholars have used logistic regression models to analyze the drivers of urban spatial expansion, arable land change, and land use change; however, they are relatively less frequently applied to woodland landscape analysis [25].

The Gannan area of Jiangxi Province is an ecological geographical barrier in southern China. The forest ecosystem in the area is influenced by geographical and human factors simultaneously so that a single landscape index may not adequately reflect changes in forest landscape [27]. Consequently, there is an urgent need for a variety of methods to analyze forest landscape trends. Therefore, the present study attempts to collect landscape pattern data and morphological data and systematically analyze the spatio-temporal changes in forest landscapes in Anyuan County, a typical county in the hilly mountains of southern Jiangxi Province, using the logistic regression model to analyze the factors driving forest landscape change in the region. The objectives of the present study are: (1) to determine the change of woodland area using a land-use conversion matrix; (2) to explore the change of woodland landscape using morphological spatial pattern analysis (MSPA) and landscape pattern indices; (3) to identify the drivers of forest landscape change using a logistic model; and (4) to propose an optimal management strategy for the forest landscape in the study area.

2. Materials and Methods

2.1. Study Area and Data Collection

Anyuan County is located in the south of Jiangxi Province, China (115°9′112″–115°37′13″ E, 24°52′18″–25°36′52″ N). It belongs to the humid monsoon hanging area at the southern edge of the Central Subtropics. The topography of the study area is undulating and mainly consists of low and medium mountains and hills extending from the Nanling Mountain Range (Figure 1). The total land area of Anyuan County is 235,924.1 ha, out of which 83.4% is mountainous. Anyuan County is a typical inland mountainous area dominated by woodlands and is the origin of the upper Ganjiang River (Yangtze River system) and the Dongjiang River (Pearl River system), which provide drinking water for Hong Kong.

2.2. Data Sources and Processing

The 2009 Second National Land Survey Data, 2014 Land Use Change Data and 2019 National Land Survey Data were obtained from the Natural Resources Bureau of Anyuan County. Digital elevation model (DEM) data for Anyuan County with a 30 m × 30 m resolution were downloaded from the Geospatial Data Cloud (http://www.gscloud.cn, 17 August 2022) to extract slope and topographic relief information. Soil texture and soil type maps of Anyuan County were also extracted from the corresponding maps of Jiangxi Province. Socio-economic statistical data, such as navel orange cultivation, returning farmland to forest, and land occupancy for construction were derived from the Anyuan County Statistical Yearbook (

Table 1. Table of data sources.

Data	Source
Land survey data	The Natural Resources Bureau of Anyuan County
Digital elevation model (DEM) data	The Geospatial Data Cloud (http://www.gscloud.cn, 17 August 2022)
Soil texture and soil type maps	The corresponding maps of Jiangxi Province
Socio-economic data	The Anyuan County Statistical Yearbook

). Using ArcGIS v10.6 (ESRI Inc., Redlands, CA, USA), land use data for 2009, 2014, and 2019 were used to reclassify the study area into seven categories: farmland, garden land, woodland, grassland, construction land, water area, and other land. Woodland was further divided into forestland, shrubland, and other woodland (Figure 2).

2.3. Transfer Matrix-Based Analysis of Woodland Change

The land use transfer matrix was derived from a Markov model to analyze woodland change in the study area [28,29]. The Markov model can not only quantify the transferred area between different land use types but also reveal the rate of land use transfer. The equation is as follows:

$$D_{ij}={\displaystyle \sum _{ij}^n\left[\frac{S_{i-j}}{S_i}\right]}\times 100\%$$

where D_ij is the proportion of land use type i converted to type j in the study period; S_i is the initial area of land use type i in the study area; S_i−j is the area of land use type i converted to type j in a study period; and n is the number of land use types converted in the study area. Vector data were converted into 30 m × 30 m raster data and subjected to overlay analysis using ArcGIS v10.6. The results were imported into Excel v2021 (Microsoft Corp., Redmond, WA, USA), and a pivot table was inserted to derive the woodlands transfer matrix.

2.4. Selection of Landscape Pattern Indices

Landscape pattern analysis is the core process in landscape ecology research. Understanding the relationships between landscape elements and the development of landscape patterns is essential for accurately revealing trends in ecological environment change. At present, this analysis is the most widely used method to quantitatively study the landscape pattern [30,31,32]. Therefore, 11 landscape pattern indices were selected based on their ecological connotations combined with the characteristics of landscape types in the study area. Five indices were selected at the level of patch type: largest patch index (LPI), aggregation index (AI), landscape shape index (LSI), patch density (PD), and edge density (ED). Six indices were selected at the landscape level: patch density (PD), number of patches (NP), edge density (ED), landscape shape index (LSI), aggregation index (AI), and Shannon diversity (SHDI). The ecological connotations of each index are summarized in

Table 2. Landscape pattern indices used in this study and their ecological connotations.

Index	Calculation Formula	Parameter Description and Ecological Connotations
Number of patches (NP)	NP is obtained using ArcGIS v10.6.	It is positively correlated with landscape fragmentation.
Aggregation index (AI)	$AI=\left[g_{ii}/mas\to g_{ii}\right]\left(100\right)$	The gii is the number of similar neighboring patches of the corresponding landscape type. The smaller the AI value, the more dispersed the land landscape is.
Landscape shape index (LSI)	$Sc=L/2{\left(\Pi A\right)}^{1/2}$	The A and L are the mean area and perimeter of patches, respectively. The smaller the Sc value, the more regular and simpler the shape of the patches and landscape.
Patch density (PD)	$P_d=N/A$	N is the number of patches and A is the total area of the landscape or patches. The larger the Pd value, the greater the fragmentation of the landscape.
Largest patch index (LPI)	$LPI=a_{\mathrm{m}\mathrm{a}\mathrm{x}}/A\times 100$	amax is the area of the largest patch in the landscape or patch type, while A is the total area of the landscape. The LPI determines the dominant patch type in the landscape.
Edge density (ED)	$ED=E/A$	E is the total length of the edge of all patches, and A is the total area of the landscape. The ED value is highest when the components account for equal proportions.
Shannon diversity (SHDI)	$SDHI=-{\displaystyle \sum _{i=1}^m}\left(P_{i}In{P}_i\right)$	Pi is the proportion of landscape patch type i. Higher SHDI values indicate a more balanced distribution of patch types in the landscape.

[33,34,35]. The landscape indices of woodlands in each study period were calculated using Fragstats v4.2 (Department of Forest Science, Oregon State University, Corvallis, OR, USA) [36].

2.5. Morphological Spatial Pattern Analysis

MSPA identifies important habitat patches and corridors that play a vital role in the landscape connectivity of woodlands at the pixel level. This morphological approach emphasizes structural connectivity. First, it divides the foreground into seven non-overlapping classes (i.e., core, islet, perforation, edge, bridge, loop, and branch) [17,37]. Next, the important classes for maintaining landscape connectivity are identified, and the current status of morphological landscape distribution and corridor structure is obtained via the spatial analysis function, which is superior to traditional landscape analysis methods. The MSPA classes and their ecological connotations are summarized in

Table 3. Classes of morphological spatial patterns and their ecological connotations.

Morphological Spatial Pattern	Definition	Ecological Connotations
Core	A set of pixels with foreground pixels that are far away from background pixels, at a distance larger than the specified value for a certain parameter.	Larger habitat patches in the foreground pixel provide larger habitats for species, which are essential for biodiversity conservation and are source sites in the ecological network.
Islet	Patches that are not connected to any foreground areas and that are smaller than the minimum threshold of the core area.	Small, isolated, and fragmented patches that are not connected to each other, have low connectivity between patches, and low potential for internal exchange and transfer of material and energy.
Perforation	A hole inside the central area, with the edge outside the foreground made up of the background.	It is a transition area between the core and non-green landscape patches, namely internal patch edges (edge effects).
Edge	The edge outside the foreground.	It is a transition area between the core area and the major non-green landscape areas.
Bridge	At least two points connected to different cores.	The narrow area linked to the core area represents a corridor connecting patches in the ecological network, which is crucial for biological migration and landscape connectivity.
Loop	At least two points connected to the same core area.	Corridors connecting the same core area are shortcuts for species migration within the same core area.
Branch	Only one side is connected to the edge, bridge, or loop.	Areas that are only connected at one end to an edge, bridge, loop, or perforation.

[18,38,39]. MSPA was implemented using GuidosToolbox v3.0 (Peter Vogt.,European Commission Joint Research Centre (JRC), Brussels, Belgium) [40], with woodland as the foreground and other land use types as the background. Woodland was assigned a value of 2, and other land use types were assigned a value of 1, while the attribute value of blank areas was set to 0. Data were transformed into binary raster data in TIFF format, and the morphological spatial patterns of woodlands were identified using eight-neighborhood analysis with an edge width of 1 and a raster size of 30 m × 30 m. Seven groups of morphological indices were obtained for the MSPA classes [41,42]. MSPA analysis of the data was performed to obtain seven landscape types that do not overlap. Afterward, the images in TIFF format were imported into ArcGIS v10.6 (ESRI Inc., Redlands, CA, USA), and unique values were obtained using the symbol system in the layer properties, and then multiple layers could be observed. The layers were merged with the same category to obtain the above seven landscape types [43,44].

2.6. Logistic Regression Model

The dichotomous logistic regression model allows regression modeling of dependent and independent (or continuous or mixed) variables [45,46]. The independent variable is x_i = (x₁, x₂…x_n), and the dichotomous dependent variable Yi takes the value 0 or 1 (Yi = 0 means that the subject event does not occur and Yi = 1 means that the subject event does occur). The forest land changes when Y = 1, and 0 otherwise. The model illustrates the relationship between the dependent variable and the independent variable of the model. Pi is the probability of occurrence of an event, and the probability can be calculated using a logistic function [47]. The formula is as follows:

$$P=\left(Y=1| x_1,x_2\dots x_n\right)=\frac{exp\left({\beta }_0+{\sum }_{}{\beta }_i{x}_i\right)}{\left[1+exp\left({\beta }_0+{\sum }_{}{\beta }_{i}x\right)\right]}$$

A logit transformation is usually performed on the formula P to transform it into a linear formula. The formula is as follows:

$$In\left(\frac{P}{1-P}\right)={\beta }_0+{\beta }_1{x}_1+\dots +{\beta }_n{x}_n$$

where: x₁, x₂......x_n are independent variables; parameters are regression coefficients to be determined; P/(1 − P) is the occurrence ratio of 1 set of events; and the occurrence ratio expβ is an important indicator of the independent variable on the dependent variable. expβ indicates the change in the occurrence ratio of events for each one-unit increase in the independent variable. Wald χ2 is used to evaluate the degree of explanation of the dependent variable by each independent variable. A p value less than 0.05 indicates a good model fit. The Homsmer–Lemeshow indicator (HL) was used to test the fit of the logistic regression equation, and HL values greater than 0.05 showed that the model fit was good [48].

Using the ArcGIS v10.6 platform, this study spatially analyzed the land use data in 2009 and 2019, assigned the raster for increasing or decreasing forest land from 2009 to 2019 as 1 and the raster for unchanging forest land as 0. According to the characteristics of woodland change in Anyuan County, the representativeness and accessibility of data, and with reference to relevant research results [49,50], the woodland change indicators were finally constructed based on natural environment, socio-economic factors, and geographical location. Natural factors include single factors such as elevation, slope, soil type, soil erosion, temperature and precipitation, as well as comprehensive indicators such as biodiversity evaluation, wind and sand control evaluation, soil and water conservation evaluation, and surface roughness factor. Geographic location indicators include distance to rivers, roads, towns, rural settlements, urban settlements, orchards, industrial and mining sites, etc. Socio-economic indicators consist of the seven economic indicators of total rural population in townships, total rural labor force, total rural electricity consumption, rural educated population, garden fruit production, navel orange and citrus production, and population density in townships. To facilitate analysis, relevant indicators were analyzed, quantified, and reclassified. According to the “Technical Regulations for Current Land Use Survey” promulgated by China Agricultural Zoning Commission in 1984, the slope was divided into five classes, which were ≤2°, 2°~6°, 6°~15°, 15°~25° and >25°. Soil type was divided into six classes according to the results of the Second National Soil Census: I (yellow soil), II (red soil), III (yellow-red soil), IV (waterloggogenic paddy soil), V (submergenic paddy soil), and VI (intermediate purple soil). Temperature and precipitation were normalized; geographical location was based on the current land use map of Anyuan County; and roads, rural settlements, etc. were extracted. The indicators were first analyzed by Euclidean distance analysis in ArcGIS v10.6 (ESRI Inc.), and then the raster data were classified into six levels using the natural breakpoint method using reclassification tools for each factor. Population density and navel orange and citrus production data were from the 2019 Anyuan County Statistical Yearbook. Logistic regression model analysis was performed in SPSS25.0 for the above multiple indicators, and the following less significant indicators were excluded: soil erosion degree and comprehensive indicators in natural factors; distance to rivers and urban settlements in geographical location indicators; and five economic indicators—total population of villages and towns, total rural labor force, total rural electricity consumption, rural educated population, and garden fruit production. Finally, the index system shown in

Table 4. Driving indicators of forest land change in Anyuan County, 2009–2019.

Driving Force Index	Variables	Types of Variables	Units
Dependent variable	Woodland change (Y)	Secondary classification	0, 1
Natural drivers	Elevation (X1)	Continuous type	m
	Slope (X2)	Multi-category	I~V
	Soil type (X3)	Multi-category	I~V
	Temperature (X4)	Continuous type	°C
	Precipitation (X5)	Continuous type	mm
Geographical location drivers	Distance to road (X6)	Continuous type	m
	Distance to town (X7)	Continuous type	m
	Distance to rural settlements (X8)	Continuous type	m
	Distance to orchards (X9)	Continuous type	m
	Distance to industrial and mining sites (X10)	Continuous type	m
Socio-economic drivers	Population density (X11)	Continuous type	Person/km2
	Navel orange and citrus production (X12)	Continuous type	t

was formed.

3. Results

3.1. Woodland Transfer in Anyuan County

Land use data for woodlands in the study area (2009–2019) are summarized in

Table 5. Area of different woodland types in Anyuan County and the change over time (Unit: ha).

Woodland Type	Woodland Area			Change Over Time
	2009	2014	2019	2009–2014	2014–2019	2009–2019
Forestland	158,503.17	157,976.33	155,439.01	−526.83	−2537.32	−3064.15
Shrubland	495.31	486.07	645.18	−9.24	159.11	149.88
Other woodland	19,865.19	19,727.51	17,820.19	−137.68	−1907.32	−2044.99
Total	178,863.66	178,189.92	173,904.39	−673.74	−4285.52	−4959.27

after reclassification. The total woodland area decreased over the 10-year period, from 178,863.66 ha in 2009 to 173,904.39 ha in 2019, a net decrease of 4959.27 ha. Among the different woodland types, the largest decrease occurred in forestland area, with a net decrease of 3064.15 ha. ‘Other woodland’ area also decreased markedly, with a net decrease of 2044.99 ha, while the net decrease in shrubland area was only 149.88 ha. Overall, the decrease in woodland area over the latter 5 years (2014–2019) was greater than that in the previous 5 years (2009–2014).

To investigate the transfer-in and transfer-out of different woodland types, a transfer matrix of woodlands in the study area was obtained (

Table 6. Transfer matrix of woodlands in Anyuan County, 2009–2019 (Unit: ha).

Woodland Type	Transfer Direction	Farmland	Garden Land	Construction Land	Water Area	Grassland	Other Land	Total
Forestland	Transfer-out	1154.46	8484.33	2447.94	731.19	60.43	6.01	12,884.36
	Transfer-in	1986.44	4235.68	465.83	120.25	433.76	113.81	7355.77
Shrubland	Transfer-out	9.94	74.02	21.53	8.88	1.38	0.00	115.75
	Transfer-in	20.00	24.35	3.84	0.97	3.66	1.08	53.90
Other woodland	Transfer-out	307.23	2442.28	613.64	147.74	18.57	4.68	3534.14
	Transfer-in	1190.31	2187.52	343.32	78.13	297.73	68.29	4165.30
Total	Transfer-out	1471.63	11,000.62	3083.11	887.81	80.37	10.69	16,534.25
	Transfer-in	3196.75	6447.55	812.99	199.36	735.15	183.18	11,574.97

). During the 10-year study period, the total area of woodland transferred to non-woodland was 16,534.25 ha, and this transfer-out area was mainly attributed to garden land (11,000.62 ha/66.53%), construction land (3083.11 ha/18.65%), and farmland (1471.63 ha/8.90%). Meanwhile, 11,574.97 ha of some land types were transferred to woodland, and this transfer-in area was mainly derived from garden land (6447.55 ha/55.70%), farmland (3196.75 ha/27.62%), and construction land (812.99 ha/7.02%). The spatial transfer map of woodland in the study area from 2009 to 2019 is shown in Figure 3.

3.2. Landscape Pattern Change of Woodlands

3.2.1. Landscape Pattern Change at the Patch Level

The landscape pattern indices of woodlands at the patch level are summarized in

Table 7. Landscape pattern change in woodlands based on patch-level indices in Anyuan County.

Year	Woodland Type	PD	LPI	ED	LSI	AI
2009	Forestland	1.76	81.58	12.58	80.62	93.99
	Shrubland	0.13	0.04	0.53	20.57	73.16
	Other woodland	2.03	1.06	12.19	86.65	81.71
2014	Forestland	1.84	78.30	12.56	81.30	93.93
	Shrubland	0.13	0.03	0.53	20.81	72.50
	Other woodland	2.09	1.05	12.18	86.77	81.62
2019	Forestland	3.44	59.44	14.59	95.06	92.83
	Shrubland	0.12	0.20	0.50	14.98	83.25
	Other woodland	5.30	0.32	14.23	116.42	73.99

PD, patch density; LPI, largest patch index; ED, edge density; LSI, landscape shape index; AI, aggregation index.

. In both forestland and other woodland, PD showed an overall upward trend over time. PD values of forestland increased from 1.76 in 2009 to 3.44 in 2019, and those of other woodland correspondingly increased from 2.03 to 5.30, indicating an enhancement in patch fragmentation for both woodland types. By contrast, PD values of shrubland decreased over time, which signified a decline in patch fragmentation for this woodland type. In addition, the highest LPI value was observed for forestland, indicating that forestland was the dominant type of woodlands in the study area. However, there was a striking decrease in the LPI values of both forestland and other woodland over time, indicating that their dominant patch types tended to decrease.

Generally, ED values of forestland and other woodland increased over the study period, while the opposite trend was observed for shrubland, especially from 2014 to 2019. Accordingly, the edge complexity of forestland and other woodland was greater than that of shrubland. LSI values of forestland and other woodland increased over time, suggesting an increase in the complexity of their patch shapes. LSI values of shrubland decreased markedly in 2019 compared with those in the earlier periods, indicating that the patch shapes tended to be regular. The AI values for both forestland and other woodland decreased from 2009 to 2019, indicating increased landscape fragmentation level and a strikingly dispersed spatial distribution, with less inter-patch connectivity. On the contrary, the AI values for shrubland first decreased slightly then increased markedly, indicating an overall enhancement of patch connectivity and a tendency for landscape aggregation.

3.2.2. Landscape Pattern Change at the Landscape Level

At the landscape level, both NP and PD values of woodlands showed an overall upward trend from 2009 to 2019, and their change over the last 5 years was distinctly faster than that in the first 5 years (

Table 8. Landscape pattern change of woodlands based on landscape-level indices in Anyuan County.

Year	NP	PD	ED	LSI	SHDI	AI
2009	7011	3.92	12.65	92.51	0.37	92.57
2014	7247	4.07	12.63	93.23	0.37	92.51
2019	15,411	8.87	14.66	112.77	0.35	90.87

NP, number of patches; PD, patch density; ED, edge density; LSI, landscape shape index; SHDI, Shannon diversity; AI, aggregation index.

). NP increased from 7011 in 2009 to 15,411 in 2019, and PD correspondingly increased from 3.92 to 8.87, indicating an overall increase in the landscape fragmentation of woodlands. Similarly, ED and LSI values increased over the 10-year study period, while AI and SHDI values tended to decrease. ED increased from 12.65 to 14.66, and LSI increased from 92.51 to 112.77, indicating that patch edge complexity and landscape fragmentation in woodlands increased over time. AI values reached 90.87–92.57, indicating that woodland patches dominated land use types. The AI exhibited a downward trend over time, indicating an increase in landscape fragmentation and a decrease in landscape connectivity of woodlands. SHDI values decreased from 0.37 in 2009 to 0.35 in 2019, indicating an improvement in the ecological environment of woodlands with a balanced distribution of ecological space.

3.3. Morphological Spatial Patterns of Woodlands

The total area of different MSPA classes in woodlands exhibited a downward trend over time, decreasing from 178,843.3 ha in 2009 to 174,730.95 ha in 2019, a net decrease of 4959.27 ha (

Table 9. The area and proportion of MSPA classes in woodlands of Anyuan County (Unit: ha).

Morphological Spatial Pattern	2009		2014		2019
	Area	Proportion	Area	Proportion	Area	Proportion
Core	144,342.9	80.71%	143,057.34	80.32%	135,145.53	77.34%
Islet	1421.82	0.80%	1513.35	0.85%	2529.9	1.45%
Perforation	7097.49	3.97%	6883.02	3.86%	5947.47	3.40%
Edge	13,762.17	7.70%	13,852.44	7.78%	15,885.72	9.09%
Bridge	3546.27	1.98%	3827.52	2.15%	4856.22	2.78%
Loop	4255.74	2.38%	4320.09	2.43%	4742.19	2.71%
Branch	4416.93	2.47%	4648.14	2.61%	5623.92	3.22%
Total	178,843.3	100.00%	178,101.9	100.00%	174,730.95	100.00%

). The core area was widely distributed and well connected in a faceted manner (Figure 4). As the primary MSPA class in the study area, the core area accounted for 80.71%, 80.32%, and 77.34% of the total woodland area in 2009, 2014, and 2019, respectively. By comparison, the edge area accounted for 7.70%–9.09% of the total woodland area, while the proportions of other classes were relatively low.

Based on the morphological indices, the core area displayed a decreasing trend over time, from 144,342.9 ha in 2009 to 135,145.53 ha in 2019, a net decrease of 9197.37 ha. The proportion of the core area decreased from 80.71% to 77.34%, and the area of larger habitat patches in pixels decreased. Conversely, the islet area increased over time, from 1421.82 ha in 2009 to 2529.9 ha in 2019, a net increase of 1108.08 ha. This trend indicates a dramatic increase in small patches in disconnected, isolated, and fragmented woodlands. Overall, the perforation area showed a downward trend, decreasing from 7079.47 ha in 2009 to 5947.47 ha in 2019, a net decrease of 1150.52 ha. Accordingly, there was a decrease in perforations within the central area and further enhancement of patch connectivity in woodlands. Furthermore, an increase in the area of edges and bridges coincided with an increase in the area of branches and loops, which optimized structural corridors in the landscape and conveyed benefits to biological migration and landscape connectivity.

3.4. Logistic Regression Analysis of Woodland Change

In the HL test, the Sig value is 2.96 (

Table 10. Analysis of drivers of forest land change in Anyuan County, 2009–2019.

Driving Force Index	Variables	β Factor	Standard Error	Wald χ2	Sig.	Exp (β)
Natural drivers	X1	0.006	0.000	2963.778	0.000 **	1.006
	X2	-	-	521.312	0.000 **	-
	X2 (I)	−18.264	10,815.227	0.000	0.999	0.000
	X2 (II)	1.733	0.119	213.336	0.000 **	5.659
	X2 (III)	1.689	0.128	174.276	0.000 **	5.412
	X2 (IV)	1.327	0.115	133.821	0.000 **	3.768
	X2 (V)	0.688	0.119	33.517	0.000 **	1.990
	X3	-	-	17.142	0.004 **	-
	X3 (I~II)	2.258	0.715	9.978	0.002 **	9.560
	X3 (II~III)	2.161	0.758	8.135	0.004 **	8.683
	X3 (III~IV)	2.313	0.716	10.449	0.001 **	10.108
	X3 (IV~V)	2.388	0.716	11.116	0.001 **	10.896
	X3 (V~VI)	25.246	40,192.969	0.000	0.999	9.205
	X4	0.084	0.004	362.030	0.000 **	1.088
	X5	−0.281	0.104	7.312	0.007 **	0.755
Geographical location drivers	X6	−0.131	0.039	11.414	0.001 **	0.878
	X7	0.056	0.023	5.984	0.014 *	1.058
	X8	−0.648	0.091	51.197	0.000 **	0.523
	X9	−0.537	0.117	21.206	0.000 **	0.585
	X10	−0.0001	0.000	104.382	0.000 **	1.000
Socio-economic drivers	X11	0.0004	0.000	7.312	0.007 **	1.000
	X12	0.000	0.000	81.771	0.000 **	1.000
	Constants	−21.804	1.236	311.377	0.000	0.000

** and * indicate significance at 0.01 and 0.05 levels, respectively.

), which is greater than 0.05, and the model fits well. The area under the receiver operating characteristic (ROC) curve is 0.869, greater than 0.7, and the model passes all tests and can be used for the driving force analysis of forest land area reduction. In terms of coefficients, precipitation (X₅), distance to road (X₆), distance to rural settlement (X₈), distance to orchard (X₉), and distance to industrial and mining land (X₁₀) were negatively correlated with forest land area change; that is, the lower the values of X₅, X₆, X₈, X₉, and X₁₀, the greater the forest land change, and the remaining indicators were positively correlated. According to the significance level and Wald χ² statistic, among the natural environmental drivers, X₁~X₅ were significantly correlated, among which, X₂ (Ⅰ) and X₃ (V) were significantly greater than 0.05, indicating that slope (0°~2°) and soil type V (purplish soil) did not have significant effects on woodland area. Specifically, the four slope ranges of slope II (2°~6°), slope III (6°~15°), slope IV (15°~25°), and slope V (>25°) all had significant effects on woodland change. In addition, according to the Wald χ² statistic in Table 10, the Wald χ² statistic of slope II (2°~6 °) was the largest, which indicates that its positive influence on woodland area change is greater than other slope ranges; that is, the probability of woodland area change was 566% with slope I (≤2°) as the reference object. Similarly, among the six soils, X₃ (IV~V) had a greater effect on woodland change, i.e., with soil I as the reference object, the probability of woodland area change was 1090%.

Among the geographical location drivers, X₆~X₁₀ indicators were significantly correlated; only X₇ was positively correlated, and the others had a negative correlation. Among them, woodland change had a significant rural settlement effect, i.e., the closer the distance to the rural settlement, the faster the decrease in woodland is, and the conversion area of woodland to another land use decreases gradually as the distance from the road increases.

Among the socio-economic factors, population density (X₁₁) and navel orange and citrus production (X₁₂) had significant effects on woodland landscape variation.

4. Discussion

Forest landscape change is most significantly influenced by socio-economic factors [51]. Firstly, Anyuan County is one of the most important navel orange production areas in southern Ganzhou, with an area of 20,667 ha under navel orange cultivation in 2022, and forest land tends to be in a “reciprocal” relationship with navel orange plantations. From 2009–2014, the core area decreased from 144,342.9 to 143,057.34 ha, a net decrease of 1285.65 ha. In addition, since 2012, there have been large-scale outbreaks of citrus yellow shoot in Ganzhou. To prevent the spread of the disease, many farmers cut down diseased trees and stopped planting and managing navel oranges, and some abandoned orchards, which were naturally restored to woodland (Figure 5a). From 2014 to 2019, the patch density (PD) of shrublands decreased from 0.13 to 0.12, with a trend toward patches. Second, ionic rare-earth resources are abundant in Anyuan County, and rare earths are mostly distributed in hilly mountain areas, and the mining of rare-earth mines causes direct damage to woodlands. According to remote sensing monitoring data, from 2009 to 2014, the area of forest land damaged by rare earth mining in Anyuan was 180.5 ha, and the area of isolated islets increased, from 1421.82 ha in 2009 to 1513.35 ha in 2014. Concurrently, since 2014–2019, the local government has continued to promote mining environment management and ecological restoration, restoring 103 mining plots with an area of 230.8 ha, with many exposed rare earth mining sites restored to woodland and the landscape status improved, with the shrub landscape shape index (LSI) value decreasing from 20.81 to 14.98 from 2014 to 2019 (Figure 5b). Furthermore, the expansion of urban and rural constructed land has had major impacts on forest landscape change. Anyuan County is a hilly and mountainous county, the periphery of the town is surrounded by forest land and cultivated land, and most of the expansion of the town is achieved by leveling the surrounding hills, and the edge density (ED) of woodlands around towns increased from 2009–2019, with an increase in ED from 12.65 to 14.66. Additionally, to promote rural revitalization, the state has supported Anyuan County through the Ministry of Transport (Figure 5c). Over the past 10 years, RMB 3.8 billion of has been disbursed to complete the upgrading of 131 km of national and provincial roads in addition to the construction of 1875 km of rural roads. Whereas transport infrastructure has been improved, many woodlands have been separated by roads, with considerable impacts on woodland landscape patterns (Figure 5d).

Moreover, physical and geographic conditions, such as elevation, slope, and soil type are influencing woodland landscape patterns, which is consistent with the results of previous studies [52,53]. Elevation and topographic slope influence the temperature and rainfall distribution in a woodlot, thus influencing plant growth [54]. The topography of Anyuan County rises in the middle and slopes to the north and south, and woodlands are mainly distributed in areas with higher elevation and slope between 2° and 25°, and in the three soil types yellow loam, red loam, and yellow–red loam, which are more widespread. Geographical location is also a key factor influencing variation in woodland in Anyuan County. Accessibility through roads favors the development and occupation of woodlands. The closer a woodland is to a rural settlement, the more vulnerable it is to disturbance from human activity.

The changes in woodland area and landscape pattern not only mark trends in forestry development but also characterize trends in land evolution [55]. The rationality of development trends is directly related to the scale and layout of forestry production, which plays a pivotal role in people’s welfare and livelihood. Through the analysis of the evolution of the landscape pattern of woodland, it is of great significance to identify the existing problems and strengthen the management of forest resources in order to promote the sustainable development of forestry. Based on the observation of the evolution of woodland changes and landscape patterns in Anyuan County, the following strategies for woodland development and protection are proposed. Firstly, woodlands must be protected, especially within the core ecological source areas, to minimize disturbance, and prevent adverse impacts such as landscape fragmentation. Secondly, forest land should be exploited rationally, balancing the relationships between navel orange cultivation and construction projects and forestry to ensure the sustainable development of forest land. Thirdly, the monitoring of forest land should be enhanced, and the damage that rare-earth mining and navel orange cultivation cause to forest land should be minimized. Furthermore, restoration activities should be undertaken in areas that have already been damaged to optimize forest landscape resources and morphological structure.

It is better to analyze the changes of various types of landscapes during the study period by combining the landscape pattern index and morphological spatial pattern analysis (MSPA). The introduction of logistic regression models into the study of the drivers of woodland change can identify which factors have influenced woodland change. Compared with qualitative research methods, this method is more convincing. Inevitably, there are limitations of this study. In the process of selecting the driving factors, the authors are limited to the information of the study area, and when selecting economic data, few indicators were selected, which cannot fully reflect the real situation in the study area. In the follow-up study, the screening of influencing factors can be expanded to enrich the research content. Second, extending the research period and combining research with remote sensing information technology may make the study more convincing and representative. In the follow-up study, the author will try to make up for the above deficiencies and make the study more representative.

5. Conclusions

Overall, from 2009 to 2019, the total area of woodland in Anyuan County shows a decreasing tendency, with a net decrease of 4995.27 ha, which is mainly due to the conversion of woodland to agricultural land, garden land, and construction land. Woodland change is the result of a combination of internal and external factors such as nature, geography, and socio-economy. In this study, the influence of socio-economic factors is the most significant. Firstly, the area of navel orange plantation and woodland are in a “wane and wax” relationship, and navel orange plantation has encroached on a large amount of woodland. Secondly, mining, especially rare-earth mines, has produced a large amount of bare land which exerts serious effects on the surrounding ecological environment. Moreover, there is a large amount of woodland around the cities in the hilly mountainous areas of southern China, and the process of urban expansion inevitably encroaches on some woodland for construction. The construction of roads and other infrastructure also has an important impact on the landscape pattern of woodland and also facilitates the development and utilization of woodlands. Meanwhile, among the natural environmental drivers, the influence of elevation on woodland change cannot be ignored. The natural, geographical, and socio-economic influences have caused important changes in woodland patches and shape indices, such as enhanced patch fragmentation, decreased core area, enhanced isolation and fragmentation, etc. The results of this study can provide a basis for regulating the impact indicators of woodland change and optimizing woodland landscape patterns, which can promote sustainable management of woodland.