Exploring the Spatially Heterogeneous Relationships Between Biodiversity Maintenance Function and Socio-Ecological Drivers in Liaoning Province, China

Abstract

Biodiversity maintenance function (BMF) denotes the capacity of ecosystems to sustain genetic, species, ecosystem, and landscape diversity. Assessing the spatial distribution and underlying drivers of BMF at the regional scale is essential for biodiversity management. However, research on the socio-ecological drivers of BMF from a geographical perspective remains scarce. Therefore, this study developed an integrated assessment framework encompassing climatic factors, species richness, vegetation status, ecosystem protection, and anthropogenic disturbance. We analyzed the BMF spatial patterns across Liaoning Province, China, and identified the dominant drivers and their spatial heterogeneity using multi-scale geographically weighted regression and geographical detector. The results show that (1) the eastern/western mountainous regions and Liaohe River estuary are critical BMF zones for prioritized conservation; (2) BMF spatial variation is mainly shaped by precipitation, temperature, slope, and forestland/farmland proportion, with factor interactions amplifying their impacts; (3) drivers show distinct spatial heterogeneity. Specifically, precipitation, slope, and NDVI exert homogeneous effects, whereas elevation, temperature, farmland/wetland proportion, and GDP exhibit pronounced heterogeneity. Natural factors generally exert positive effects, while the farmland/urban proportion tends to exert negative impacts—for example, farmland’s negative influence is stronger in the west, whereas the forestland and temperature exert more positive effects in the east. The results enhance the methodological framework for elucidating the spatial relationships between BMF and drivers, providing a scientific basis for biodiversity conservation and ecosystem management in Liaoning Province and similar regions.

1. Introduction

Biodiversity is the essential foundation for human well-being and sustainable development [1]. Biodiversity maintenance function (BMF)—which denotes the capacity of ecosystems to sustain genetic, species, ecosystem, and landscape diversity is among the most important ecological functions of ecosystems, representing a fundamental ecological process [2,3]. Nevertheless, rapid population growth, expanding industrialization and urbanization, and accelerating climate change have jointly induced a global biodiversity crisis, which has attracted close attention from governments and the international community [4,5,6]. In an era of constrained conservation budgets, limited personnel, and tight timelines—as well as pronounced spatial variation in both biodiversity and threat intensity—regional scale evaluation of BMF is essential for identifying priority conservation areas and developing targeted protection strategies [7,8,9,10].

Traditional biodiversity assessments—which mainly rely on plot- or site-level species inventories, community surveys, and subsequent diversity indices—have limited applicability in regional-scale management [11]. Therefore, contemporary studies have shifted toward multifactor, integrated assessment frameworks that support decision-making at broader scales [12,13]. The emergent paradigm exploits remote sensing data to develop grid-level assessment models of biodiversity patterns, while integrating multisource datasets (including climate variables, land cover, species occurrence, and ecosystem integrity metrics) with spatially explicit modeling to account for landscape heterogeneity across jurisdictional boundaries [14]. For example, Liu et al. [3] constructed a BMF model in the Beijing–Tianjin–Hebei region, incorporating biomass, topography, climate, and land cover. Wu and Meng [15] developed a biodiversity assessment model based on temperature, elevation, and NDVI, refined using national-level nature reserves. Li et al. [16] quantified BMF in the Yellow River Basin through net primary productivity (NPP), precipitation, temperature, and elevation. The InVEST habitat quality module has also been widely applied as a proxy for biodiversity [2]. These pioneering efforts provide valuable frameworks and variable selections for BMF assessment. However, a systematic integration of these diverse approaches within a unified framework is still lacking.

Understanding the spatial distribution of biodiversity and its underlying drivers has, therefore, become a focal topic in ecological research [17,18]. Specifically, elucidating the spatial heterogeneity and driving mechanisms of biodiversity provides vital insights into the complex interactions between natural and socioeconomic factors while informing conservation practice [19,20]. Although existing evidence shows that biodiversity patterns reflect the combined influence of climatic, topographic, and anthropogenic factors [21,22], the relative importance of each factor often varies substantially across geographic contexts [23,24], making it difficult to quantify their intertwined effects [5]. To address this challenge, the geographical detector provides a robust approach for assessing both individual and interactive contributions of drivers, without relying on strict statistical assumptions or complex parametrization [25]. It has been successfully applied to the driving analysis of various ecosystem services such as soil conservation [26], carbon storage [27], and water conservation [28]. Moreover, geographically weighted regression (GWR) incorporates spatial locality into regression modeling, thus capturing spatial non-stationarity in driver effects [29]. However, classic GWR’s key limitation is its assumption that all processes operate at a single spatial scale—an assumption often violated in real-world ecological systems with multiscale processes [30]. Fotheringham et al. [31], therefore, advanced multi-scale geographically weighted regression (MGWR), which assigns an optimal bandwidth to each variable, thereby enhancing analysis in highly heterogeneous regions [32] and leading to its increasing application across ecological studies [33]. Despite the two methods’ complementary strengths, their combination application to identify the BMF drivers remains scarce.

Liaoning Province, located in the southern part of Northeast China, serves as a transitional zone between multiple biogeographic regions. It hosts diverse and interconnected ecosystems—including temperate deciduous forests in the eastern Changbai Mountain extension, grasslands in the western Mongolian Plateau transition zone, and coastal wetlands along the Bohai and Yellow Seas [34,35]. This ecosystem diversity supports irreplaceable biodiversity: It is the key habitat and migratory corridor for numerous terrestrial and marine species, including Oplopanax elatus, Abies holophylla, Otis tarda, Grus japonensis, Larus relictus, Ciconia boyciana, and Phoca largha [36,37]. However, the attention it receives is disproportionate to its high biodiversity values. Moreover, only a small fraction of Liaoning falls within the nationally designated priority areas for biodiversity conservation set out in the National Biodiversity Conservation Strategy and Action Plan of China (2011–2030) [38]. Furthermore, existing regional BMF studies in China have primarily focused on ecologically renowned basins (e.g., Qinghai–Tibet Plateau [19], Yellow River Basin [16]) and economically developed urban agglomerations (e.g., Beijing–Tianjin–Hebei [3], Yangtze River Delta [39]). In contrast, research on biodiversity in Liaoning remains fragmented. Relevant studies have been mostly limited to single-ecosystem surveys (e.g., temperate forest biodiversity [40]), small-scale assessments (e.g., wetland biodiversity pattern in Liaohe Delta [41]), or oversimplified pattern recognition (e.g., importance evaluation based on NPP [42]), and they have lacked systematic, large-scale BMF evaluations integrating natural and anthropogenic indicators. Thus, the spatial heterogeneity of the key BMF drivers remain unknown. These knowledge gaps have prevented a comprehensive understanding of ecosystem services in Liaoning Province and hindered evidence-based provincial conservation planning.

This study integrated climate, species richness, vegetation status, ecosystem protection, and anthropogenic disturbance to construct a regional-scale BMF assessment for Liaoning Province, Northeast China. Specifically, we aimed to (1) characterize the spatial patterns of BMF importance in Liaoning Province and identify priority zones for biodiversity conservation; (2) quantify and map the influence and spatial heterogeneity of socio-ecological drivers using geographical detector and MGWR; and (3) derive recommendations for biodiversity conservation and management in Liaoning Province based on the research findings.

2. Materials and Methods

2.1. Study Area

Liaoning Province occupies the southern portion of Northeast China between 38°30′–43°24′ N and 118°53′–125°46′ E and includes 14 prefecture-level cities spanning over 148,000 km². Local geomorphologists summarize its terrain as “mountainous regions constitute about 60%, water bodies cover approximately 10%, and plain regions for the remaining 30%” [37]. The province features mountainous and hilly terrain in its eastern and western regions, with mean elevations of 500–800 m. In contrast, the central region comprises the Liao River Plain, which has an average elevation of 200 m (Figure 1a). The province experiences a temperate continental–monsoon climate characterized by coincident thermal and precipitation maxima during the growing season [43]. The mean temperature increases from east to west, while precipitation displays a pronounced gradient from humid in the east to arid in the west [34]. Forest and cropland constitute the principal land-use types (Figure 1b). Intensifying anthropogenic pressures superimposed on inherent environmental vulnerability—particularly in the west—have caused severe ecological problems, including land desertification and soil erosion [44].

Figure 1. Study area: (a) location and elevation; and (b) land cover in 2020.

2.2. Data Sources and Preprocessing

The analysis integrated climate, vegetation, land use and land cover (LULC), topographic, and socioeconomic data layers. The slope was extracted from the digital elevation model (DEM). The forestland, grassland, wetland, farmland, and urban proportion were obtained from LULC data. More details about the multisource data are shown in Table 1. All rasters, except the 30 m LULC, were resampled to 250 m and re-projected to WGS_1984_Albers in ArcGIS 10.3 for spatial congruence.

Table 1. Data sources, types, timeframes, and spatial resolutions used in the study.

Data	Sources	Type	Spatial Resolution	Time
Average annual precipitation	National Earth System Science Data Sharing Platform (http://www.geodata.cn, accessed on 10 November 2024)	Raster	1 km	2000–2020
Average annual temperature	National Earth System Science Data Sharing Platform (http://www.geodata.cn, accessed on 10 November 2024)	Raster	1 km	2000–2020
Digital elevation model (DEM)	Shuttle Radar Topography Mission (SRTM; http://srtm.csi.cgiar.org/, accessed on 2 December 2023)	Raster	90 m	—
Normalized difference vegetation index (NDVI)	SPOT/VEGETATION products (https://www.earthdata.nasa.gov, accessed on 10 November 2024)	Raster	500 m	2020
Net primary productivity (NPP)	SPOT/VEGETATION products (https://www.earthdata.nasa.gov, accessed on 10 November 2024)	Raster	500 m	2020
Fractional vegetation cover (FVC)	National Tibetan Plateau Data Center (https://data.tpdc.ac.cn, accessed on 10 November 2024)	Raster	500 m	2020
Aboveground biomass (AGB)	National Tibetan Plateau Data Center (https://data.tpdc.ac.cn, accessed on 10 November 2024)	Raster	500 m	2020
Amphibian, mammal, and bird species richness	IUCN (https://iucn.org/, accessed on 14 September 2024)	Raster	1 km	2015
GDP Density	Resource and Environment Science and Data Center (http://www.resdc.cn, accessed on 10 August 2024)	Raster	1 km	2020
Population Density	Resource and Environment Science and Data Center (http://www.resdc.cn, accessed on 10 August 2024)	Raster	1 km	2020
Road	Peking University’s Geographic Data Sharing Infrastructure (http://geodata.pku.edu.cn, accessed on 10 August 2024)	Vector	—	2015
Land use and land cover (LULC)	Resource and Environment Science and Data Center (http://www.resdc.cn, accessed on 10 August 2024)	Raster	30 m	2020

2.3. Assessment of BMF at the Regional Scale

2.3.1. Indicator System Construction

Building upon previous BMF assessments [3,7,15,19], we developed a comprehensive framework simultaneously considering climate, vegetation, species, current conservation status, and anthropogenic pressures (Table 2). Five major criteria were defined: climatic factors, vegetation status, species richness, ecosystem protection, and anthropogenic disturbance. Each specific indicator selected represents a widely recognized and commonly used proxy variable within its respective dimension.

Table 2. Indicator system for evaluating the importance of BMF at the regional scale.

Criteria Hierarchy	Indicator Hierarchy
Climatic factors	Long-term average precipitation
	Long-term average temperature
Vegetation status	Net primary productivity (NPP)
	Vegetation fraction cover (FVC)
	Aboveground biomass (AGB)
Species richness	Avian richness
	Mammalian richness Amphibian richness
	Floral richness
Ecosystem protection	National or provincial nature reserve
Anthropogenic disturbance	Anthropogenic disturbance index

2.3.2. Calculations of Criteria Hierarchies

The formula for calculating the climatic factor criteria hierarchy (C_i) is as follows:

$$C_i = {w_1\cdot \mathrm{P}\mathrm{r}\mathrm{e}}_i+{w_2\cdot \mathrm{T}\mathrm{e}\mathrm{m}}_i,$$

(1)

where Tem_i denotes the standardized value of long-term average precipitation; Pre_i is the standardized value of long-term average temperature; and w₁ and w₂ are the weights of temperature and precipitation, respectively. In this study, calculations were performed with equal weights for temperature and precipitation.

The formula for calculating the species richness criteria hierarchy (S_i) is as follows:

$$S_i= {w_3\cdot \mathrm{B}\mathrm{i}\mathrm{r}}_i+{w_4\cdot \mathrm{M}\mathrm{a}\mathrm{m}}_i+{w_5\cdot \mathrm{A}\mathrm{m}}_i+{w_6\cdot \mathrm{V}\mathrm{e}\mathrm{g}}_i,$$

(2)

where Bir_i, Mam_i, Am_i, and Veg_i denote the standardized value of bird, mammal, amphibian, and plant species richness, respectively; and w₃, w₄, w₅, and w₆ are the weights of birds, mammals, amphibians, and plants, respectively. In this study, calculations were performed with equal weights for the four indicators.

The formula for calculating the vegetation status criteria hierarchy (V_i) is as follows:

$$V_i = {w_7\cdot \mathrm{N}\mathrm{P}\mathrm{P}}_i+{w_8\cdot \mathrm{F}\mathrm{V}\mathrm{C}}_i+{w_9\cdot \mathrm{A}\mathrm{G}\mathrm{B}}_i,$$

(3)

where NPP_i, FVC_i, and AGB_i denote net primary productivity, vegetation fraction cover, and aboveground biomass, respectively; and w₇, w₈, and w₉ are the weights of NPP, FVC, and AGB, respectively. In this study, calculations were performed with equal weights for the three indicators.

The value of ecosystem protection criteria hierarchy (E_i) is assigned based on the level of nature reserves. In the study, E_i was designated as 0.7 for provincial-level nature reserves; 1.0, for national-level nature reserves; and 0, for other regions.

The anthropogenic disturbance criteria hierarchy is derived from LULC data, and its formula is as follows:

$$A_i= \left({\mathrm{F}\mathrm{a}\mathrm{r}\mathrm{m}}_i+{\mathrm{T}\mathrm{o}\mathrm{w}\mathrm{n}}_i\right)/{\mathrm{A}\mathrm{r}\mathrm{e}\mathrm{a}}_i,$$

(4)

where A_i is the anthropogenic disturbance index; Farm_i, and Town_i denote the farmland and urban area, respectively; Area_i refers to the area of the grid i. In this study, we extracted farmland and urban area based on LULC (30 m) and created a fishnet (cell size width/height = 250 m) to calculate the anthropogenic disturbance index of each cell.

2.3.3. BMF Importance Index

Each criteria hierarchy derived above should be processed by min–max standardization, after which the BMF importance index (Bio_i) is calculated using the formula specified below:

$${\mathrm{Bio}}_{\mathrm{i}} = {w_c\cdot C}_{i,norm}+{w_s\cdot S}_{i,norm}+{w_v\cdot V}_{i,norm}+{w_e\cdot E}_{i,norm}-{w_h\cdot A}_{i,norm},$$

(5)

where C_i,norm, S_i,norm, V_i,norm, E_i,norm, and A_i,norm denote the standardized value of the climatic factor, species richness, vegetation status, ecosystem protection, and anthropogenic disturbance criteria hierarchies, respectively; and w_c, w_s, w_v, w_e,and w_a are the weights assigned to the respective criteria hierarchies. In this study, calculations were performed with equal weights for the five criteria hierarchies.

The additive linear model with equal weights was adopted based on its theoretical transparency, previous studies in regional ecological assessments [37,45], and its utility as an objective baseline in the absence of definitive empirical evidence to justify differential weighting. This approach ensures interpretability while avoiding unsubstantiated subjective bias, treating all five core criteria as fundamentally equally important for regional biodiversity maintenance.

2.4. Exploration of the Main Factors Influencing BMF

2.4.1. Selection of Influencing Factors

The selection of explanatory variables was guided by a conceptual framework that recognizes biodiversity as shaped by interacting climatic, topographic, habitat, and anthropogenic factors [19,23,46]. Specifically, climate variables (temperature and precipitation) represent the energy–water balance that limits species distributions and ecosystem productivity. Topographic factors (DEM, slope) capture microclimatic variation, soil conditions, and exposure to human disturbance. Habitat types—expressed as forest, grassland, and wetland proportions—reflect the landscape’s capacity to support diverse species assemblages. NDVI was included as a proxy for overall vegetation productivity and habitat quality. Furthermore, anthropogenic factors (urban and farmland proportions, population, GDP, and road density) were used to quantify the intensity of human activity. Detailed information is provided in Table 3.

Table 3. Detailed information on the influencing factors.

Variable Category	Variable	Code	Description	Unit
Topographic factors	DEM	x1	Digital elevation model	m
	Slope	x2	Slope gradient	◦
Climatic factors	Pre	x3	Long-term average precipitation	mm
	Tem	x4	Long-term average temperature	°C
	NDVI	x5	Normalized difference vegetation index	–
Habitat factors	Forest	x6	Forestland proportion	%
	Grass	x7	Grassland proportion	%
	Wetland	x8	Wetland proportion	%
Anthropogenic factors	Urban	x9	Urban proportion	%
	Farmland	x10	Farmland proportion	%
	GDP	x11	GDP density	10,000 yuan/km2
	Population	x12	Population density	People/km2
	Road	x13	Road density	km/km2

In this study, global Moran’s I in ArcGIS 10.3 was implemented to validate the spatial autocorrelation characteristic of the study area. Positive values indicate the existence of a cluster over space, while negative values suggest a geographically dispersed spatial pattern [47].

To analyze the comprehensive response of BMF to its driving factors at the grid scale, we employed GIS grid technology to partition the study area into 6123 grids (5 km × 5 km). Average values of both the dependent variable (BMF) and each independent variable were extracted for every grid using the “Zonal Statistics as Table” tool, located within the “Zonal” analysis toolkit under “Spatial Analyst Tools” in ArcGIS 10.3.

2.4.2. Geographical Detector

This study employed the geographical detector [25] to quantify the explanatory power of the candidate drivers, utilizing a 5 km × 5 km grid as the spatial analysis unit. Specifically, the factor detector and interaction detector modules of the geographical detector were employed. The factor detector identifies the individual contribution of each independent variable, while the interaction detector evaluates whether paired factors exert enhanced or weakened joint influences [48]. The explanatory power is measured by the q-statistic:

$$q=1-{\displaystyle \frac{{\displaystyle \sum _{h=1}^L{N}_h}{\sigma }_h^2}{N{\sigma }^2}},$$

(6)

where q (0 ≤ q ≤ 1) represents the quantitative contribution of drivers on dependent variables, and a higher q value indicates a stronger influence of the explanatory variable on the explained variable. Moreover, h = 1, 2, 3, …, L refers to the classification or stratification of variables; N_h and ${\sigma }_h^2$ represent the number of samples and the variance of layer h, respectively; and N and σ² represent the total sample size and the variance over the study area, respectively.

Both individual and pairwise q-values were computed to identify key drivers. As the geographical detector can only process discrete variables, all 13 continuous variables must be reclassified into discrete ones. This process is implemented using the Natural Breaks method available in ArcGIS 10.3, which clusters data based on their intrinsic properties. This approach minimizes within-class variance while maximizing between-class variance, making it a widely adopted method for discretizing continuous independent variables [27,48].

2.4.3. Multi-Scale Geographically Weighted Regression

To further identify spatially varying relationships between BMF and driving factors, we employed multi-scale geographically weighted regression (MGWR), which assigns an optimal bandwidth to each predictor and thus outperforms traditional GWR when processes operate at disparate spatial scales [49]. GWR/MGWR forms regression relationships between independent and dependent variables at a local scale, effectively mitigating errors arising from spatial heterogeneity [47]. The regression coefficient of a specific driving factor reflects the BMF response intensity to that factor. A higher adjusted R² value indicates greater explanatory power and better model fit. Independent variables with pronounced spatial heterogeneity are better captured with narrower bandwidths, while those exhibiting greater spatial stationarity are more appropriately modeled using larger bandwidths [50]. The MGWR model is expressed as follows:

$$y_i={\beta }_0\left(u_i,v_i\right)+{\displaystyle \sum _{j=1}^k{\beta }_{bwj}\left(u_i,v_i\right)x_{ij}}+{\epsilon }_i, i\in \left\{1,2,\cdots ,n\right\}$$

(7)

where y is the dependent variable; (u_i, v_i) is the center-of-mass coordinate of the i-th sample; β₀ (u_i, v_i) denotes the intercept; k is the number of driving factors; x_ij is the j-th independent variable; β_bwj (u_i, v_i) is the regression coefficient of the i-th sample for the j-th driving factors, and bwj in β_bwj indicates the bandwidth used for calibration of the j-th conditional relationship; and ε_i is the error term.

Models were fitted in MGWR 2.2. The dependent variable field was BMF, while the independent variable fields comprised the 13 driving factors listed in Table 3 (continuous variables). An adaptive bi-square kernel and the corrected Akaike Information Criterion (AICc) guided bandwidth selection, which was optimized via a golden-section search [51]. Detailed algorithmic procedures are described by Oshan et al. [52] and Li and Fotheringham [51].

3. Results

3.1. Spatial Characteristics of BMF in Liaoning Province

The climatic factor criteria hierarchy across Liaoning Province shows a “High in the East and South, Low in the West and North” distribution pattern (Figure 2). The vegetation status criteria hierarchy achieves high-level values in the eastern region, moderate-level values in the western region, and low-level values in the central region. The species richness criteria hierarchy exhibits a spatial pattern analogous to that of the vegetation status criteria hierarchy. The ecosystem protection criteria hierarchy reflects the current spatial pattern of biodiversity conservation levels established based on the nature reserve system, whereby the western, southern, and eastern regions, as well as the Liaohe River, have been effectively protected. Concerning the anthropogenic disturbance criteria hierarchy, the eastern Liaoning has been relatively less impacted by human activities, while the West exhibits a fragmented pattern of human influence. By contrast, the central and coastal areas have experienced more severe human disturbances.

Figure 2. Spatial patterns of the five criteria hierarchies: (a) climatic factors; (b) vegetation status; (c) species richness; (d) ecosystem protection; and (e) anthropogenic disturbance in Liaoning Province.

Using the Natural Breaks (Jenks) method in ArcGIS 10.3, we divided the BMF importance index into four categories: low, moderate, high and critical. Highly important and critically important zones encompass 30,788.19 km² and 30,483.25 km², respectively, accounting for 21.24% and 21.03% of the provincial area. These zones are principally concentrated in the eastern and western mountainous regions, along the main stem of the Liao River, and within the Liao River estuary (Figure 3). Moderate- and low-importance areas span 31,226.01 km² (21.54%) and 52,451.43 km² (36.19%), respectively, and are mainly distributed across the central Liao River Plain and parts of western Liaoning.

Figure 3. Classification of biodiversity maintenance function (BMF) importance in Liaoning Province.

3.2. Effects of Socio-Ecological Factors on BMF

3.2.1. Spatial Autocorrelation Test

Heterogeneity is observed in the geospatial relationships between driving factors and BMF. Therefore, we carried out spatial autocorrelation analysis to determine whether there was spatial agglomeration in BMF and its drivers. The results show that all Moran’s I values for the driving factors and BMF exceed zero, with corresponding p values below 0.001. This indicates that the distributions of the 13 factors and BMF are not random but instead exhibit positive spatial correlations (Table 4).

Table 4. Moran’s I of BMF and driving factors (x1: DEM, x2: slope, x3: precipitation, x4: temperature, x5: NDVI, x6: forestland proportion, x7: grassland proportion, x8: wetland proportion, x9: urban proportion, x10: farmland proportion, x11: GDP density, x12: population density, x13: road density).

Variable	Moran’s I	Z-Scores	p-Value
BMF	0.859	131.583	0.000
x1	0.916	140.367	0.000
x2	0.869	133.147	0.000
x3	0.994	152.281	0.000
x4	0.931	142.706	0.000
x5	0.653	100.179	0.000
x6	0.840	128.778	0.000
x7	0.510	79.164	0.000
x8	0.525	80.714	0.000
x9	0.665	102.035	0.000
x10	0.785	120.270	0.000
x11	0.609	96.956	0.000
x12	0.706	109.711	0.000
x13	0.619	95.316	0.000

3.2.2. Single-Factor Effect on the Spatial Heterogeneity

Geographical detector was used to identify key factors governing BMF distribution. A higher q value indicates a stronger influence of the factor on BMF. The results indicate that all 13 candidate drivers significantly influence the BMF spatial pattern (p < 0.05), though their explanatory power (q-value) varies markedly (Figure 4). Four distinct gradients emerge: (1) q > 0.5—forest proportion, farmland proportion, slope, and precipitation; (2) 0.3 < q ≤ 0.5—urban proportion and NDVI; (3) 0.1 < q ≤ 0.30—population density, DEM, temperature, and GDP; (4) q ≤ 0.1—road density, wetland proportion, and grassland proportion.

Figure 4. Individual contributions (q-values) of 13 drivers to the BMF pattern (x1: DEM, x2: slope, x3: precipitation, x4: temperature, x5: NDVI, x6: forestland proportion, x7: grassland proportion, x8: wetland proportion, x9: urban proportion, x10: farmland proportion, x11: GDP density, x12: population density, x13: road density).

Hence, forestland proportion, farmland proportion, slope, and precipitation are the foremost determinants of the BMF spatial differentiation, while urban proportion and NDVI exert secondary influences. Among climatic variables, precipitation surpasses temperature, and among topographic variables, slope exceeds DEM in explanatory strength.

3.2.3. Interactive Effects Among Drivers

Interaction detector analysis reveals that paired factors consistently exert stronger influences than single factors, exhibiting a prevailing “bi-factor enhancement” pattern (Figure 5). Moreover, three principal interaction clusters are evident: (1) precipitation with forestland proportion, farmland proportion, and slope; (2) farmland proportion with NDVI and urban proportion; (3) forestland proportion with wetland proportion.

Figure 5. Interactive influence of paired drivers on BMF (x1: DEM, x2: slope, x3: precipitation, x4: temperature, x5: NDVI, x6: forestland proportion, x7: grassland proportion, x8: wetland proportion, x9: urban proportion, x10: farmland proportion, x11: GDP density, x12: population density, x13: road density).

Secondary but still notable synergies include the following: (1) forestland proportion with DEM, farmland proportion, temperature, and population density; (2) farmland proportion with population density, DEM, slope, and GDP; (3) slope with DEM. Significantly, any interaction involving forestland proportion, farmland proportion, slope, or precipitation outperforms interactions between the remaining nine variables, thus underscoring the pivotal role of these four drivers.

3.2.4. Spatial Heterogeneity Captured by MGWR

The MGWR model was used to identify the spatial distribution of the influences of different drivers on BMF. MGWR explains 97.1% of the variance in BMF (adjusted R² = 0.971). Model fit (local R² > 0.81) is strongest in western and southern Liaoning (Figure 6). Moreover, the spatial response of local coefficients reveal the following patterns: forestland proportion, precipitation, temperature, and farmland proportion together dominate provincial BMF heterogeneity. Precipitation, slope, and NDVI contribute positively across nearly all grids. Temperature is generally positive but attenuates in the arid west. DEM yields dual effects, being positive in the mountainous east and negative in western lowlands. Forestland proportion exerts a robust positive influence, which is maximized in the northeast. Grassland proportion exerts low influence province-wide. Wetland proportion exhibits a mixed response, being significantly positive along the Liao River and negative or positive elsewhere. Farmland proportion correlates negatively—most strongly in the semi-arid west—while urban proportion also shows negative relationships, particularly in the northwest. GDP exerts positive effects in the west, likely reflecting restoration investments, but shows negative impacts in the industrialized central and southern regions. Population density has a mildly positive influence, whereas road density remains weakly negative across the province. These patterns confirm pronounced spatial non-stationarity; thus, distinct driver suites predominate in different physiographic and socioeconomic settings.

Figure 6. MGWR results: (a) local R²; and (b–n) coefficients between BMF and 13 drivers (x1: DEM, x2: slope, x3: precipitation, x4: temperature, x5: NDVI, x6: forestland proportion, x7: grassland proportion, x8: wetland proportion, x9: urban proportion, x10: farmland proportion, x11: GDP density, x12: population density, x13: road density).

4. Discussion

4.1. Spatial Heterogeneity of BMF in Liaoning Province

BMF spatial heterogeneity in Liaoning Province reflects the intricate interplay of hydro-thermal regimes, topography, vegetation structure, and socioeconomic pressures. By integrating climatic factors, vegetation status, species richness, ecosystem protection, and anthropogenic disturbance into a multi-criteria framework, this study mapped a clear “High-East-and-West, Low-Center” spatial pattern of BMF. The eastern and western regions exhibit optimal habitat conditions and robust ecosystem services, presenting critical ecological barriers that safeguard regional biodiversity belts [34]. Furthermore, the eastern mountains exhibit the highest BMF levels due to the convergence of favorable climatic conditions, dense vegetation, diverse habitat mosaics, limited human disturbance, and rich species pools [35]. Western Liaoning centered on the Nuru’erhu–Songling–Yiwulu range (300–1000 m a.s.l.) and draining in the Daling and Xiaoling Rivers [53]—this region features geomorphologically complex terrain that supports heterogeneous ecosystems and, consequently, elevated BMF values.

Shaped by multiriver alluvium, the Liaohe River Delta sustains a spectrum of wetland habitats—including reed marshes, riverine waters, and intertidal mudflats [54]. These wetlands serve as critical sanctuaries for vulnerable and endangered waterbirds (e.g., O. tarda, G. japonensis, L. relictus, and C. boyciana) and host the world’s largest breeding colony of Larus saundersi. The delta also functions as a pupping ground for P. largha in the western Pacific [41]. Consequently, it represents an indispensable region for safeguarding BMF at the provincial scale.

4.2. Driving Mechanism of Socio-Ecological Factors on the BMF Spatial Pattern

Climatic suitability is a fundamental regulator of habitat quality and biodiversity patterns [23]. Our geographical detector and MGWR analyses consistently identify climatic factors as the principal determinants of BMF distribution. Specifically, precipitation and temperature jointly modulate vegetation productivity, faunal behavioral rhythms, and resource availability, thus shaping BMF patterns. In line with previous studies by MacFadyen et al. [55] and Sun et al. [19], we find that precipitation exerts a stronger influence on biodiversity than temperature. Notably, the MGWR bandwidths (temperature = 544 vs. precipitation = 5516) reveal substantially greater spatial non-stationarity in temperature effects. This trend is particularly evident in arid western Liaoning, where elevated temperatures worsen drought stress and habitat degradation, resulting in negligible or even negative regression coefficients for temperature.

Topography has a multifaceted influence on BMF by modifying local microclimates and mediating the intensity of human disturbance [16,56]. Although DEM alone has limited explanatory power, its interactions with forestland proportion, farmland proportion, and precipitation demonstrates significant enhancement effects, underscoring its indirect regulatory role. Slope exhibits a strong positive association with BMF, which is consistent with Liaoning’s predominantly hilly terrain (~46% of land area), where moderate relief and gentle gradients promote habitat heterogeneity and support a diversity of niches that collectively enhance biodiversity maintenance capacity.

Forest ecosystems, characterized by their multilayered structure—comprising canopy, sub-canopy, shrub, herb, and litter layers—create heterogeneous habitats that facilitate niche differentiation and support high species richness. In Liaoning, where forests cover 42.56% of the land surface, both the geographical detector and MGWR identify forestland proportion as the foremost determinant of BMF spatial differentiation. Although wetlands occupy a relatively small area, they function as critical habitats for aquatic and semi-aquatic organisms and serve as essential stopover sites for migratory birds. Our findings align with those of Luo et al. [57], revealing a significant positive correlation between wetland proportion and BMF along the Liao River. Although wetland proportion alone exhibits modest explanatory power, its strong interaction with forestland proportion underscores the complementary role of forest–wetland mosaics in sustaining regional biodiversity. Grasslands, comprising only 0.58% of provincial areas, exert a negligible influence on BMF. Notably, NDVI shows a consistent positive relationship with BMF, particularly in western Liaoning, suggesting that restoring vegetation cover can contribute to biodiversity recovery in this ecologically vulnerable region.

Anthropogenic pressures exhibit pronounced spatial heterogeneity in their impacts on BMF, as evidenced by small MGWR bandwidths, which indicate highly localized effect scales. In northwestern Liaoning, located on the southern edge of the Horqin Sandy Land, agricultural expansion and urban development exert intense negative effects on BMF. This region is inherently vulnerable to desertification, drought, and sandstorms, which heightens ecosystem sensitivity to human disturbances [37,44]. In western Liaoning, GDP is positively correlated with BMF in the west, reflecting targeted regional investments in ecological restoration rather than the benefits of economic activity. By contrast, in the more economically developed central and southern regions, GDP shows a negative correlation with BMF, suggesting that intensive economic development exerts adverse impacts on biodiversity, primarily through the conversion of natural habitats to agricultural, industrial, and residential uses [58]. Moreover, road density demonstrates negative regression coefficients across western and southern Liaoning, reinforcing that transportation infrastructure expansion poses an additional significant threat to regional biodiversity [47].

Overall, the formation of Liaoning’s BMF pattern reflects a complex interplay of ecological and anthropogenic drivers, with distinct spatial hierarchies and regional specificities. While ecological factors—particularly forest proportion, climate, and topography—are the core drivers shaping the fundamental pattern of BMF, their dominance does not preclude the critical but localized influence of human activities, thus corroborating previous studies [21,34]. In ecologically fragile western Liaoning, agricultural expansion exerts strongly negative effects that can surpass those of natural drivers, while in the central and southern developed regions, GDP and road density exert clear adverse impacts. This spatial bifurcation in anthropogenic effects aligns with findings from other regions [19,23]. Importantly, nonlinear interactions, such as those between topography and human accessibility or between forest and wetland cover, highlight the need to transcend single-factor analyses. As a transitional zone spanning floristic, climatic, and land-use gradients, Liaoning exemplifies how regional biodiversity patterns emerge from context-dependent synergies and trade-offs. This finding underscores the importance of developing spatially tailored conservation strategies that address both ecological connectivity and anthropogenic pressures.

4.3. Management Implications and Recommendations

Because climatic, topographic and anthropogenic factors interactively determine provincial BMF pattern, the following adaptive and region-specific strategies are essential: (1) Prioritize climate adaptation. Precipitation is identified as the chief driver, so climate considerations must guide biodiversity conservation. Persistent drought monitoring, early-warning systems, and water-saving agronomic practices are vital for the drought-prone northwest [59]. (2) Institutionalize nature-based solutions, including coupled protection and restoration projects (afforestation, grassland rehabilitation, and wetland renewal) to boost vegetation recovery, enhance ecological function, and raise BMF, particularly in fragile western landscapes. (3) Optimize land-use patterns, such as through crop diversification and biodiversity-friendly farming, to enhance agricultural landscape heterogeneity and reduce habitat fragmentation [60,61,62]. Rigorous ecological impact assessments must accompany infrastructure development to limit habitat loss during urban expansion [23,63]. (4) Differentiate conservation priorities. For example, coastal wetlands on the East Asian–Australasian Flyway require strict regulation of reclamation, landfill, and mass tourism. Establish additional sanctuaries, buffer zones, and wetland parks. Similarly, eastern forests demand enhanced reserve networks, bans on primary-forest logging and wildlife hunting, and development of ecological corridors to mitigate habitat fragmentation.

4.4. Limitations and Future Directions

Although this study aimed to quantify the key factors influencing the spatial patterns of BMF in Liaoning Province, several limitations warrant attention in future research. First, the distribution of biodiversity is governed by multifaceted interactions among ecological and anthropogenic factors. Our analysis, constrained by the substantial spatial heterogeneity across Liaoning Province, potentially excluded some critical variables, including the following: (1) Natural environmental factors: these include soil characteristics (e.g., nutrient availability, pH) and geological processes (e.g., erosion, sedimentation) [64]. (2) Anthropogenic activities: reclamation, mining, tourism, photovoltaic projects, and ecological restoration projects may exhibit nonlinear or lagged effects on biodiversity [65]. The partial inclusion of these drivers can give rise to discrepancies between our findings and actual driving mechanisms. Therefore, future studies should adopt a more holistic framework, integrating interdisciplinary datasets to capture these complexities. Second, data availability was limited. The uniform 250 m resolution of our spatial data may have obscured fine-scale ecological patterns, potentially biasing accuracy. Advanced remote sensing platforms or drone-based surveys can help mitigate this limitation in future work. Owing to data scarcity, key biological groups (reptiles, fish, insects, and macrofungi) were omitted from the BMF model, despite their ecological significance. Collaborative efforts with local biodiversity monitoring networks can help fill these gaps. Nevertheless, the methodology of our study remains applicable to broader regional or national scales to meet the needs of different decision-makers. Overall, the aforementioned limitations warrant further consideration and should be addressed in future research.

5. Conclusions

This study developed an integrative assessment model to evaluate the importance of BMF, mapped its spatial distribution across Liaoning Province, and employed both the geographical detector and MGWR to analyze the drivers underlying its spatial heterogeneity. The results reveal a “High-East-and-West, Low-Center” spatial pattern of BMF, identifying the eastern and western mountains and the Liao River estuary as biodiversity priority conservation zones. Natural variables—particularly vegetation, climate, and terrain—play a dominant role in shaping BMF, while human activities exert context-dependent yet non-negligible influences. Forest proportion, precipitation, temperature, farmland proportion, and slope emerge as the principal drivers, each exhibiting marked spatial non-stationarity. Natural factors generally promote BMF, while farmland expansion and urbanization tend to suppress it. Notably, GDP displays positive effects only in western restoration zones. These spatially varying effects underscore the need for region-specific conservation strategies, such as prioritizing ecological restoration in western ecologically fragile zones, developing biodiversity-friendly agriculture (e.g., organic farming, crop diversification) in the central plains, and maintaining estuarine hydrological connectivity by strict regulation of reclamation, landfill, and mass tourism. These insights not only advance the current understanding of the drivers of BMF but also provide evidence-based guidance for optimizing biodiversity conservation efforts and land-use planning.