Spatiotemporal Variation in Regional Habitat Quality and Its Driving Factors: A Case Study of Ningxia, Northwest China

Abstract

Habitat quality is critical for spatial planning strategies and ecological conservation initiative, evaluating the health of the natural environment that supports human survival. However, current approaches pay insufficient attention to revealing the evolution and spatial heterogeneity of the habitat quality simultaneously. In this study, a comprehensive and practical framework was therefore developed for mechanistic habitat quality analysis, which incorporates an adaptable evolutionary model alongside multiple spatial statistical methods. Ningxia, located in Northwest China, was selected as a case study area due to its fragile ecosystem. The proposed framework was then applied to characterize the evolutionary process and spatial heterogeneity of habitat quality in Ningxia. Key factors driving spatial heterogeneity were also found at the same time. From 2000 to 2024, habitat quality in Ningxia is characterized by good habitat and shows significant improvement, following a progressive trajectory. The proportion of poor habitat has been significantly reduced from 29.26% to 24.63%, while that of excellent habitat has been increased from 1.68% to 2.33% over the past two decades. Variation in habitat quality is more pronounced in northern and southern regions, while remaining relatively stable in the central Yellow River ecological corridor. Both natural and socioeconomic factors have an impact on the habitat change in this region, such as the Normalized Difference Vegetation Index (NDVI), Net Primary Productivity (NPP), and Gross Domestic Product (GDP). Vegetation factors play vital roles in spatial variation in habitat quality, while the influences of socioeconomic factors are relatively small. The spatial heterogeneity is driven by nonlinear synergistic effects among numerous factors. This paper developed a feasible framework to retrieve the evolution and spatial heterogeneity pattern of habitat quality, which provides a robust methodology for further habitat assessment at the ecologically fragile regions worldwide.

1. Introduction

Habitat quality refers to the ability of a specific ecosystem to provide suitable conditions for species survival and reproduction, which depends on its internal ecosystem property and external interference (i.e., human intervention) [1,2]. It is an effective indicator of ecosystem health that reflects the balance between socioeconomic development and ecological functions, especially for ecologically fragile regions [3]. Therefore, habitat quality is critical for regional biodiversity conservation and environmental sustainability. Climate change and rapid socioeconomic development pose challenges to biodiversity and ecosystem health [4,5,6]. In recent years, the intense land use and land cover changes, such as urbanization and expansion of crop land, have resulted in habitat fragmentation and reduced ecological capacities for supporting species survival, which increases risks of ecosystem habitat degradation [7]. Water resource pressures stemming from climate change and plantation degradation further exacerbate risks to ecosystem [8]. Consequently, establishing a framework to assess habitat quality and investigate its driving forces is imperative to developing effective ecological conservation strategies under a changing environment.

To assess regional habitat quality, many scholars have developed models, such as the Integrated Valuation of Ecosystem Services and Tradeoffs (InVEST) [9,10], Artificial Intelligence Ecosystem Service Model (ARIES) [11,12], Social Values for Ecosystem Services (SolVES) [13,14] and the Multi-Scale Integrated Model of The Ecosystem (MIMES) [15,16]. Among these models, the InVEST model is proven to be the most applicable for regional habitat quality assessment, because of its lower data requirements, user-friendly operation, cost efficiency, high accuracy, and advanced spatial analytics [17]. It can also describe the distribution of habitat quality in different regions and the degree of degradation at multiple scales. However, many studies employing the InVEST model have primarily concentrated on delineating spatial patterns of habitat quality, while fewer have undertaken in-depth explorations of the underlying drivers, particularly considering the complex interactions between multi-factors and their relative contributions to habitat quality changes. Remote sensing and geographic information system (GIS) tools provide effective means of capturing long-term variations in regional habitat quality and exploring its driving force. By utilizing multitemporal and multi-source remote sensing data, it is possible to accurately detect the long-term spatiotemporal variation in habitat quality and quantify their evolutionary processes with the support of GIS spatial analysis tools. However, despite these technical advances, current applications still often fall short in mechanistically explaining the driving forces behind observed phenomenon. Therefore, integrating dynamic remote sensing monitoring with InVEST-based habitat assessment, and further incorporating spatial statistical methods (e.g., Geographical Detectors, spatial regression), can provide a more comprehensive framework to detect long-term dynamics of regional habitat quality and explore the coupled effects and relative importance of multiple driving forces.

The northwest of China is located in the arid and semi-arid climate region. The scarcity of water resources and severe soil erosion generates the fragile ecological background of this region. Most parts of this region serve as an ecological security barrier and key implementation area for national ecological restoration projects such as the Three-North Shelterbelt Forest Program and the Grain for Green Program. Driven by socioeconomic development and ecological restoration projects, land use and land cover in this region experience significant changes, which may result in more complex spatial patterns of habitat quality dynamics. Ningxia Hui Autonomous Region, located in the upper and middle reaches of the Yellow River, serves as typical representative of the ecologically fragile area in Northwest China. However, existing studies have rarely integrated long-term dynamic evolution, spatial heterogeneity, and driving mechanisms of habitat quality under the dual influence of human activities and ecological restoration. Investigating its habitat quality dynamics is critical for understanding the complex interactions between socioeconomic development and ecological restoration across the region.

Therefore, the objectives of this study are to: (1) develop a framework to analyze the spatiotemporal variation in habitat quality and its driving forces; (2) assess the long-term spatiotemporal evolution of habitat quality; (3) reveal the driving forces of habitat quality variation and their interactions; (4) determine the spatial heterogeneity of habitat quality and demarcate its ecological management zones. The findings are expected to provide a feasible framework to retrieve the spatiotemporal evolution of habitat quality and scientific basis to ecological conservation and environmental sustainability.

2. Study Area and Data Source

2.1. Study Area

The Ningxia Hui Autonomous Region (Ningxia), an ecologically vulnerable region of the Yellow River Basin, is located in the northwest of China (Figure 1). It contains five cities, the Shizuishan, Yinchuan, Zhongwei, Wuzhong and Guyuan, with a total area of 66,400 square kilometers. This region is characterized by a continental semi-arid climate, with an average annual precipitation of 343 mm. Precipitation in this region shows uneven spatiotemporal distribution, which is temporally concentrated in summer and spatially reaching higher amounts in the southern part. The land cover types of this region are mainly grassland, crop land and forest. The crop land is threatened by salinization, and over 90% of the natural grassland in the region faces ecological degradation risk. According to the Soil and Water Conservation Bulletin of Ningxia Hui Autonomous Region in 2024 (https://slt.nx.gov.cn/xxgk_281/fdzdgknr/gbxx/sltjgb/202509/t20250908_5013736.html (accessed on 10 January 2026)), the soil erosion area of Ningxia is 14,793 square kilometers, accounting for 22.28% of the total area. The Helan and Liupan mountain areas in this region are part of the National Nature Reserve.

2.2. Data Sources and Processing

Multi-source datasets are used in this study, including Digital Elevation Model (DEM), precipitation (PRE), Net Primary Productivity (NPP), Normalized Difference Vegetation Index (NDVI), population density (PD), Gross Domestic Product (GDP), and land use and land cover data. The detailed information of datasets used in this study are listed in

Table 1. Detailed information of the datasets employed in this study.

Item	Name	Resolution	Time Series	Source
DEM	ASTER GDEM	30 m	-	https://www.gscloud.cn
PRE	1 km monthly precipitation dataset for China	1 km	2000, 2005, 2010, 2015, 2020	https://zenodo.org/records/3114194 (accessed on 8 September 2025)
NPP	MOD17A3HGF	500 m	2000, 2005, 2010, 2015, 2020	https://earthengine.google.com/
NDVI	MOD13A2	1 km	2000, 2005, 2010, 2015, 2020	https://earthengine.google.com/
PD	China population density 1 km grid dataset	1 km	2000, 2005, 2010, 2015, 2020	https://www.resdc.cn/DOI/DOI.aspx?DOIID=32 (accessed on 15 September 2025)
GDP	China GDP 1 km grid dataset	1 km	2000, 2005, 2010, 2015, 2020	https://www.resdc.cn/DOI/DOI.aspx?DOIID=33 (accessed on 15 September 2025)
Land use and land cover	China land cover dataset [20]	30 m	2000, 2005, 2010, 2015, 2020, 2024	https://zenodo.org/records/15853565 (accessed on 4 September 2025)

. DEM is derived from ASTER GDEM dataset with a spatial resolution of 30 m. Precipitation is obtained from 1 km monthly precipitation dataset for China and annual precipitation were aggregated by monthly data. NPP is obtained from MOD17A3HGF with a spatial resolution of 500 m. NDVI is obtained from MOD13A2 with a spatial resolution of 1 km. PD and GDP data are obtained from China population density and GDP 1 km grid dataset. Land use and land cover (LULC) data is obtained from China land cover dataset (CLCD) with a spatial resolution of 30 m. LULC was reclassified from nine types to six types, i.e., crop land, forest, grass land, built-up land, water body and barren land. Datasets for the study area were all extracted and reprojected to the WGS_1984_UTM_Zone_47N using ArcGIS 10.2. In order to capture spatial dynamics while minimizing short-term noise, a five-year interval was employed to assess habitat quality [18,19]. As habitat quality assessment is mainly based on LULC inputs, habitat quality evaluation was extended to 2024 to capture the latest spatiotemporal dynamics of regional ecological conditions.

3. Methodology

Spatiotemporal variation and clustering features of habitat quality in Ningxia during 2000–2024 are investigated using long-term multi-source data and InVEST model (Version 3.13). Then, the driving forces of habitat quality spatial pattern are explored to provide implications for ecosystem restoration and habitat quality improvement. Figure 2 shows the framework of this study.

3.1. Assessment of Habitat Quality

The habitat quality module of the InVEST model is used to evaluate the habitat quality of Ningxia. Based on the input of land use and threat sources, the module can quantify negative impacts of threat sources on habitat and calculate habitat quality index of the region [19,21]. The calculated habitat index ranges from 0 to 1, with higher values indicating better habitat quality and greater capacity to provide ecosystem services. The formula of habitat quality index is as follows:

$$Q_{xj}=H_j(1-{\displaystyle \frac{D_{xj}^Z}{D_{xj}^z+k^z}})$$

(1)

where Q_xj represents the habitat quality index; H_j denotes the habitat suitability of the j-th land type; D_xj indicates the habitat degradation index of the x-th grid cell within the j-th land type; k is the semi-saturation constant; and Z is the model conversion coefficient.

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

(2)

where R represents the total number of threat sources; r is the threat source; Y_r is a set of grid cells of the r-th type threat source; y is the grid cell where the r-type threat sources located; w_r is the weight of the r-th type threat source; i_rxy is the maximum coercion distance of the r-type threat source; β_x represents the reachability of the threat source to the x-th grid cell; S_jr represents the sensitivity of the j-th type of land to the r-th type threat source.

$$i_{rxy}=1-{\displaystyle \frac{d_{xy}}{d_{rmax}}} \mathrm{for} \mathrm{linear} \mathrm{decay}$$

(3)

$$i_{rxy}=exp\left[-\left({\displaystyle \frac{2.99}{d_{rmax}}}\right)\times d_{xy}\right] \mathrm{for} \mathrm{exponential} \mathrm{decay}$$

(4)

where d_xy represents the straight-line distance between grid x and grid y; d_rmax is the maximum value of the threat factor r.

Land use and land cover types related to anthropic disturbances, i.e., crop land, built-up land and barren land, are selected as threat factors. Parameters and sensitivity of InVEST model are listed in

Table 2. Parameters of the InVEST model for Ningxia.

Threat Factor	Maximum Distance (km)	Weight	Decay Type
Crop land	3	0.8	Linear
Built-up land	5	1	Exponential
Barren land	2	0.4	Exponential

and

Table 3. Sustainability and sensitivity of the InVEST model in Ningxia.

Land Use and Land Cover Type	Habitat Sustainability	Threat Factor
		Crop Land	Built-Up Land	Barren Land
Crop land	0.4	0.3	0.6	0.2
Forest	0.9	0.7	0.8	0.3
Grassland	0.8	0.6	0.7	0.3
Built-up land	0.1	0.1	0.1	0.1
Water body	0.95	0.8	0.9	0.4
Barren land	0.6	0.4	0.5	0.1

. In addition, habitat quality is divided into five levels using natural breaks method. The five levels are Poor (Class I), Fair (Class II), Medium (Class III), Good (Class IV), and Excellent (Class V).

3.2. Analysis of Habitat Quality Spatial Transformation

The transformation matrix is used to quantitatively describe spatial transformation of habitat quality by determining the specific quantity of a habitat quality level to other levels in a region [22,23]. By constructing a habitat quality transformation matrix, the spatial transformation processes among different levels of habitat quality in different periods can be quantitatively characterized, thereby reflecting the dynamic process of habitat quality. The general formula of transformation matrix is as follows:

$$S_{ij}=\left[\begin{array}{ccc}{S}_{11}& \cdots & S_{1j}\\ ⋮& \cdots & ⋮\\ S_{i1}& \cdots & S_{ij}\end{array}\right]$$

(5)

where i represents the type of habitat quality before the transformation; j represents the type of habitat quality after transformation; S_ij is the area of habitat quality type i transfers to habitat quality type j.

3.3. Identification of Habitat Quality Spatial Autocorrelation

Spatial autocorrelation based on Moran’s I is used to measure the correlation between the habitat quality of a given spatial unit and that of its adjacent units, indicating presence of aggregation, randomness, and dispersion in spatial pattern [18,24]. This approach includes both the global Moran’s I to evaluate spatial autocorrelation pattern and the local Moran’s I index to examine localized spatial associations [25]. In this study, the global Moran’s I is employed to detect the significant presence of aggregation of habitat quality in Ningxia and the local Moran’s I is employed to identify the presence of spatial clustering phenomenon of habitat quality in Ningxia. The global Moran’s I is calculated as follows:

$$I={\displaystyle \frac{{\sum }_{i=1}^n{\sum }_{j=1}^n{w}_{ij}\left(x_i-\overline{x}\right)\left(x_j-\overline{x}\right)}{S^2\sum _{i=1}^n\sum _{j=1}^n{w}_{ij}}}$$

(6)

where I represents the global Moran’s I; n is the total number of grid units; x_i and x_j are the observed values for unit i and j; $\overline{x}$ is the mean value of observed values; w_ij is the spatial weight matrix; and S is the sum of the weight matrices.

The global Moran’s I ranges from −1 to 1. A value ranging from −1 to 0 indicates a negative spatial correlation among the variables, whereas a value between 0 and 1 indicates a positive spatial correlation. A value of approximately 0 implies the absence of spatial autocorrelation.

The local Moran’s I is calculated as follows:

$$I_i={\displaystyle \frac{\left(x_i-\overline{x}\right)}{S^2}}\sum _{j=1}^n{w}_{ij}\left(x_j-\overline{x}\right)$$

(7)

where I_i represent the local Moran’s I; x_i, x_j, n, $\overline{x}$, S, and w_ij are the same as Equation (6).

The local Moran’s I can be visualized by local indicators of spatial association (LISA) cluster map. The clusters are divided into high–high (H–H), low–low (L–L), high–low (H–L), low–high (L–H) and non-significant types. The H–H and L–L clusters reflect positive spatial correlation (synergistic effects), whereas H–L and L–H clusters indicate negative spatial correlation (tradeoff effects). Clusters identified as non-significant indicates no statistically discernible spatial pattern.

3.4. Detection of Habitat Quality Spatial Pattern Dominant Forces

Habitat quality is jointly influenced by human intervention and environmental mechanism. According to the ecological background of Ningxia and previous studies [5,26], human intervention is characterized by population density (PD) and Gross Domestic Product (GDP), while environmental mechanism is characterized by Elevation (ELE), precipitation (PRE), Net Primary Productivity (NPP), and Normalized Difference Vegetation Index (NDVI) in this study. To investigate the driving forces causing the spatial heterogeneity of habitat quality, the single-factor detection and interaction detection in Geographical Detector are used [27,28]. The single-factor detection can be used to effectively identify the presence of spatial heterogeneity in habitat quality and determine the explanatory power of different factors for the phenomena. And the interaction detection can be used to evaluate the corresponding interaction relationship between influencing factors and the dependent variable (i.e., habitat quality). Interaction relationship can be detected weakening, enhancing, or independent between influencing factors and habitat quality (

Table 4. Judgment criteria of interaction relationship based on Geographical Detector Model.

Judgment Criteria	Interaction Relationship
q(X1∩X2) < min(q(X1), q(X2))	Nonlinear attenuation
min(q(X1), q(X2)) < q(X1∩X2) < max(q(X1), q(X2))	Single-factor nonlinear attenuation
q(X1∩X2) < max(q(X1), q(X2))	Dual-factor enhancement
q(X1∩X2) = q(X1) + q(X2)	Independent
q(X1∩X2) > q(X1) + q(X2)	Nonlinear Enhancement

). The calculation formula is as follows:

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

(8)

where N represents the total sample size; ${\sigma }^2$ represents the total variance; N_h indicates the sample size of the h-th sub-region; ${\sigma }_k^2$ represents the variance of the h-th sub-region; L represents the number of sub-regions; q represents the detection capability of each detection factor with a range of [0, 1]. To reduce the complexity associated with data processing, multi-source data were resampled to a 1 km grid. The continuous factors were discretized using quantile, natural breaks and geometric interval method. A detailed description of the classification parameters for Geographical Detector is presented in

Table 5. Discrete of driving factors for Geographical Detector Model.

Factors	Discrete Method	Number of Intervals
PRE	geometric	8
ELE	natural	9
NPP	natural	9
NDVI	natural	9
PD	quantile	9
GDP	quantile	9

.

4. Results

4.1. Habitat Quality of Ningxia

The spatial distribution and area proportion of habitat quality in Ningxia for 2000, 2005, 2010, 2015, 2020, and 2024 is shown in Figure 3 and

Table 6. Area proportion (%) of poor, fair, medium, good, and excellent habitat in Ningxia from 2000 to 2024.

Year	Poor	Fair	Medium	Good	Excellent
2000	29.26	0.74	5.55	62.76	1.68
2005	24.43	1.17	3.97	68.63	1.80
2010	25.99	1.55	3.30	67.20	1.96
2015	24.36	1.88	3.48	68.21	2.07
2020	26.28	2.12	3.46	65.78	2.37
2024	24.63	2.22	2.95	67.86	2.33

. The average value of habitat quality in Ningxia for 2000, 2005, 2010, 2015, 2020 and 2024 were 0.563, 0.601, 0.588, 0.597, 0.581, and 0.594, respectively. Good habitat accounted for the largest proportion (more than 65%) in Ningxia, followed by poor, medium, excellent and fair habitat. Spatially, habitat quality is higher in the central part than the northern and southern.

In 2000, good habitat accounted for 62.76% of the total area, followed by poor habitats (29.26%); medium, fair, and excellent habitats occupied minor proportion of 5.55%, 0.74%, and 1.68%, respectively. Medium, fair, and excellent habitats exhibit a dispersed spatial pattern. In 2005, the proportion of good habitat increased to 68.63%, while poor habitat declined significantly to 24.43%. changes in medium, fair, and excellent habitats remained relatively stable. In 2010, good habitats remained dominant (67.20%), with poor habitats rebounding slightly to 25.99% but still less than 2005. In 2015, habitat quality maintained an overall optimizing trajectory: good habitats rebounded to 68.21%, poor habitats dropped to 24.36%, and excellent habitats climbed steadily to 2.07%; fair and medium habitats also recorded mild increases to 1.88% and 3.48%, respectively. In 2020, the proportion of good habitat in Ningxia slightly decreased compared to the previous years, whereas poor habitat increased. By 2024, good habitat recovered to 67.86%, and poor habitat decreased to 24.63%, reaffirming the downward trend of severely degraded habitats.

In summary, habitat quality in Ningxia was characterized mainly by good and poor habitat. From 2000 to 2024, the proportion of poor habitat decreased from 29.26% to 24.63% while that of good and excellent habitats increased from 64.44% to 70.19%. Despite the presence of fluctuation in some periods, the overall variation in habitat quality indicates positive improvement in habitat quality in Ningxia over the past two decades.

4.2. Spatial Evolution Process of Habitat Quality

4.2.1. Spatiotemporal Variation in Habitat Quality in Ningxia

The change in habitat quality in Ningxia during 2000–2024 is illustrated in Figure 4 and

Table 7. Area proportion (%) of habitat quality change in Ningxia from 2000 to 2024.

Period	Large Decline	Moderate Decline	No Change	Moderate Improvement	Large Improvement
2000–2005	2.09	0.85	87.97	2.29	6.80
2005–2010	4.11	1.02	90.85	15.51	2.45
2010–2015	2.66	1.34	90.69	1.16	4.15
2015–2020	4.25	1.31	90.74	1.37	2.32
2020–2024	2.13	0.94	91.82	1.34	3.77
2000–2024	6.10	2.78	75.93	4.73	10.46

. It can be seen that the predominant trend of habitat quality within the Ningxia region was unchanged, accounting for over 75% of the study area. The area with moderate and large improvement in habitat quality accounted for 4.73% and 10.46% of the study area, respectively, and were mainly concentrated around the Liupan Mountain in the southern part of Ningxia. Land use and land cover in this area are mainly restored grassland and forestland. Notably, habitat quality improved significantly from 2000 to 2010, especially peaking in 2005–2010, which was mainly attributed to the implementation of national ecological restoration projects, as well as intensified protection efforts for nature reserves. The proportions of moderate and large habitat decline areas were relatively limited, accounting for 6.10% and 2.78% of the study area, respectively. Areas with moderate and large decline in habitat quality are clustered in urban expansion regions, industrial sites, agricultural and pasturing interlaced zone. It is worth mentioning that the area of habitat decline expanded considerably in 2015–2020, with decline proportion reaching 4.25%. The habitat degradation in this period is mainly driven by extensive energy base construction (e.g., the Ningdong Energy and Chemical Industry Base). The proportion of area with improved habitat quality significantly exceeded that of degraded proportion across most periods, indicating an overall upward trend in habitat quality across the study area.

4.2.2. Spatial Transformation of Habitat Quality Grade in Ningxia

From 2000 to 2005, Ningxia’s habitat quality exhibited a pattern characterized by “dominance along the Yellow River corridor with stable regional differentiation.” The main corridor along the Yellow River was a densely transformed zone, primarily characterized by transformation from mutual conversions between Class I, Class II and Class III. In the southern Guyuan–Liupan Mountain area, there were transitions between Class V, Class IV and Class III. The central arid zone was dominated by no change in habitat class. This feature highly matches the “maintenance at the same level” and “corridor-type transformation” patterns depicted in Figure 5.

From 2010 to 2015, land use conversion exhibited distinct characteristics of localized leaps and corridor branch expansion. In the northern Yinchuan Plain, high-grade land experienced downgrades from Class I to Class IV and Class I to Class V, while the eastern foothills of the Helan Mountains saw upgrades from Class II to Class I. Conversion density increased along Yellow River tributary corridors, primarily involving interactions between adjacent land use classes. In the southern regions of Xiji and Haiyuan counties, frequent conversions occurred between Class V and Class IV land.

From 2020 to 2024, land conversion exhibited characteristics of “refined adjustments and optimized grade structure.” In the northern Yinchuan Plain, the retention rate of high classes increased, accompanied by minor adjustments to Classes I-II. In the central region, the refined interaction between Class IV and Classes III and V intensified. In southern counties such as Pengyang and Jingyuan, the conversion from Class V to Class IV continued, with significant improvements in lower classes.

From 2000 to 2024, long-term trend indicates that land use conversion exhibits a pattern of “expanding across the entire region with enhanced gradient interactions.” The scope of conversion has expanded from the main corridor of the Yellow River to encompass the entire region. Northern areas primarily feature interactions between high and medium conversion classes, while southern areas concentrated on low-to-medium class conversions.

4.3. Spatial Autocorrelation of Habitat Quality in Ningxia

To investigate spatial clustering characteristics of habitat quality, global and local Moran’s I are calculated. The global Moran’s I for Ningxia from 2000 to 2024 is shown in

Table 8. Global Moran’s I of habitat quality in Ningxia from 2000 to 2024.

Year	Moran’s I	Z	p
2000	−0.259	−275.99	<0.001
2005	−0.226	−229.16	<0.001
2010	−0.242	−230.07	<0.001
2015	−0.186	−183.54	<0.001
2020	−0.205	−188.00	<0.001
2024	−0.170	−151.40	<0.001

. The global Moran’s I from 2000 to 2024 were −0.259, −0.226, −0.242, −0.186, −0.205, and −0.170, respectively. The global Moran’s I for the study period were less than 0, indicating a significant negative spatial autocorrelation and discrete spatial distribution of habitat quality in Ningxia. Furthermore, the global Moran’s I exhibited an overall upward trend (gradually approaching zero) over the study period, changing from −0.259 in 2000 to −0.170 in 2024, indicating a gradual weakening of spatial dispersion and an enhancement in spatial agglomeration of habitat quality. The local Moran’s I is further calculated to reveal spatial cluster patterns (Figure 6). The local Moran’s I of habitat quality in Ningxia exhibited significant spatial clustering, with high-value (H–H) clusters concentrated in Helan Mountain and Liupan Mountain and low-value (L–L) clusters concentrated in central Ningxia plains and southern Loess Hills. Integrating statistics of global and local Moran’s I, spatial heterogeneity of habitat quality in Ningxia from 2000 to 2024 emerged from combined effects of topography, human activities, and policy interventions. From a topographic perspective, the forested mountain areas of Helan Mountain and Liupan Mountain serve as core carriers of high–high (H–H) aggregation. Their topographic isolation and superior vegetation coverage (over 60% in Helan Mountain and exceeding 70% in Liupan Mountain) have sustained contiguous clusters of high-quality habitats. Conversely, the farmlands of the Ningxia Plain and the desert grasslands of the southern Loess Hills constitute the primary distribution zones for low–low clustering (L–L). The plain region experiences intense human disturbance, while the hills suffer severe soil erosion, forming sensitive zones for low-value clustering. Human disturbances continued to shape landscape heterogeneity. Between 2000 and 2005, the expansion of the Shizuishan Industrial Park (coal chemical industry) and urbanization in Yinchuan (20% growth in built-up areas) led to localized shrinkage in the H–H zone of Helan Mountain and a surge in H-L anomalies (high values surrounded by low values) in the plains region. After 2010, ecological restoration in Helan Mountain’s mining areas (closing over 20 mines) and afforestation in southern hilly regions (reclaiming over 100,000 mu of farmland) integrated H-H zones across the entire mountainous area, while L–L zones continued to shrink in hilly areas. From 2020 to 2024, ecological corridor construction (e.g., Helan Mountain Biological Corridor) further enhanced the contiguity of H–H zones across Helan and Liupan mountains. The “ecological blank space” policy in the Yinchuan metropolitan area restricted disorderly expansion of construction land, reducing new H–L anomalies by 60%. The green transformation of the Ningdong Energy and Chemical Industry Base (achieving 95% recycling rate for coal chemical wastewater) effectively curbed low-value industrial disturbance to habitats. Spatial dispersion continued to decrease as agglomeration patterns strengthened. The negative evolution of the global Moran’s I index (gradually improving from −0.259 in 2000 to −0.170 in 2024) corroborates this process at the macro level. Although habitat quality continues to exhibit a discrete distribution characterized by alternating high and low values, the persistent strengthening of H–H and L–L clusters gradually diminishes this dispersion. This ultimately formed a heterogeneous pattern where “mountainous high-value cores remain stable, while low-value zones in plains and hills are controllable.” This outcome stemmed both from the topographical barriers of Helan Mountain and Liupan Mountain shaping spatial segregation of human activities, and directly reflected Ningxia’s ecological policy evolution from “localized restoration” to “holistic management” through targeted interventions.

4.4. Driving Factors of Spatial Evolution in Habitat Quality

To investigate the dominant drivers of spatial variation in habitat quality, the factor detection module of the Geographic Detector is applied. The influence of each driver on habitat quality is quantified by q-statistic (Figure 7). Vegetation factors (NDVI and NPP) were the dominant drivers influencing spatial patterns of habitat quality, with consistently highest explanatory power than topographic (ELE) and socioeconomic factors (GDP and PD) from 2000 to 2020. Nevertheless, the relative importance of each driving factor exhibited diverse variability in this period. Specifically, the explanatory power of the ELE increased remarkably from 0.024 in 2000 to 0.144 in 2020, indicating a pronounced enhancement in the topographic constraint and influence on spatial pattern of habitat quality. This trend may be associated with the differentiated implementation effects of windbreak and sand fixation projects across diverse topographic units in Ningxia. In contrast, the explanatory power of GDP and population density (PD), rose slightly over the study period but remained lower than that of natural ecological factors. Among vegetation factors, NPP exhibited a substantial increase in its q-statistic (from 0.151 to 0.308) and replaced NDVI as the dominant driver, indicating that the controlling factor of habitat quality has shifted from vegetation cover to productivity.

To further elucidate the mechanisms by which multiple factors jointly influence habitat quality, this study employed an interaction detection module to analyze the interaction effects among factors (Figure 8 and Figure 9). The interaction detection results for the year 2000 revealed that the q-values for pairwise interactions among all driving factors in the study area were generally higher than those for any single factor. The interaction types predominantly exhibited nonlinear enhancement, with some showing dual-factor enhancement, while only a very small proportion of factors exhibited antagonistic relationships. This indicated that the combined effects of two driving factors generally exert a stronger influence on habitat quality than their individual effects, suggesting that habitat quality in Ningxia is a complex outcome of multiple factors acting in concert.

From a temporal perspective, in the year 2000, the interaction pairs with the strongest explanatory power for habitat quality were ELE∩NPP and ELE∩NDVI. This indicated that during the initial phase, the habitat quality pattern in Ningxia was primarily determined by the strong correlation between ELE and vegetation factors (NPP, NDVI). As a relatively stable natural baseline, topography profoundly constrains vegetation growth and productivity distribution by influencing hydrological and thermal factors. Together, these two elements formed the core driving mechanism for spatial differentiation of habitat quality at that time. Additionally, the interaction between GDP and NPP also exhibited significant influence, reflecting the direct impact of economic activities on ecosystem productivity during that period.

In summary, the spatial pattern of habitat quality in Ningxia is the result of the combined effects of natural and socioeconomic factors, with natural factors playing a dominant role. Over the past two decades, the primary factor influencing habitat quality has shifted from NDVI to NPP. A widespread nonlinear enhancement effect existed among the various drivers, particularly the strong interactions between natural factors such as NPP, NDVI, and ELE, which constituted the key mechanism shaping the current spatial differentiation of habitat quality.

5. Discussion

5.1. Analysis of Applicability of Results

5.1.1. Applicability and Reliability of Results

This study assessed habitat quality in Ningxia from 2000 to 2020 using the InVEST model. Results demonstrated that the simulated outcomes align closely with regional natural geographic patterns and existing research findings, confirming the model’s strong applicability and reliable results in ecologically fragile areas like Ningxia. The habitat quality module of the InVEST model systematically quantified the stress intensity imposed on ecosystems by threat factors such as cultivated land and urban areas. It effectively identified the spatial patterns of habitat quality differentiation across Ningxia’s three major landform units: the northern Yellow River irrigation area, the central arid belt, and the southern mountainous region. The Liupan Mountains in the south, as a key biodiversity conservation area, maintain consistently high habitat quality. In contrast, the Yinchuan Plain in the north exhibited significant habitat degradation due to urban expansion and intensive agricultural activities. This distribution pattern aligned with the findings of studies on Ningxia’s ecologically priority zones [12,29,30]. Regarding parameter settings, this study referenced the existing literature by assigning a threat weight of 0.8 to cultivated land, 1.0 to urban land, and 0.4 to unutilized land. Such parameter configurations align with regional human activity intensity characteristics, enhancing the model’s simulation rationality. Regarding result reliability, Moran’s I analysis revealed significant spatial autocorrelation in habitat quality, manifesting as a distinct “high–high” and “low––low” clustering pattern. This indicated that habitat quality distribution was strongly influenced by spatial dependence on ecological background conditions. The “high–high” clusters in the southern mountainous region aligned closely with the distribution of the Liupan Mountain–Ziwuling biodiversity conservation area, while the “low–low” clusters in the north correspond to areas of intense human activity, such as the Ningdong Energy and Chemical Industry Base. This further validated the consistency between the model outputs and actual ecological degradation patterns. Long-term research also indicated that although Ningxia’s habitat quality remains at a moderate level overall, its spatial heterogeneity was significant, with degraded areas highly overlapping with human activity hotspots. Furthermore, the “high–low” outliers identified in this study were predominantly distributed in ecological transition zones, revealing habitat fragmentation in urban–rural fringe areas. This finding was mutually supportive with related research on the mismatch between ecosystem service supply and demand in Guyuan City. The model’s ability to identify ecological engineering responsiveness further demonstrated its strengths in dynamic assessment. After 2010, habitat quality in Ningxia showed a degradation trend in the central arid belt, aligning with local desertification and northern urban expansion. Conversely, southern mountainous areas maintained stable habitat quality due to ecological conservation policies, indicating InVEST’s capability to effectively capture responses to ecological effects from land use and land cover changes. Recently, researchers have coupled the InVEST model with multi-objective optimization methods to predict an improvement in Ningxia’s habitat quality under ecological conservation scenarios [31,32]. This outcome further highlighted the model’s utility in regional ecological planning and policy scenario simulation. In summary, the model’s applicability in Ningxia stemmed from its parameter localization capability and strong resolution of spatial heterogeneity. Result reliability was validated through spatial autocorrelation analysis and alignment with actual ecological patterns.

5.1.2. Uncertainty

Although the model demonstrates good applicability, it was important to recognize its inherent uncertainties. The results of the InVEST model were highly dependent on the accuracy of input data, particularly the precision and resolution of land use/land cover data. Additionally, the parameter settings within the model—such as threat source weights and attenuation patterns—involve a degree of subjectivity that may influence the absolute values of the results. Nevertheless, these limitations do not undermine the reliability of the primary conclusions: on one hand, the study focused on the relative spatial variation and temporal trends of habitat quality, which were less dependent on absolute values. On the other hand, spatial autocorrelation tests provided strong support for the robustness of the patterns identified [33,34]. Future work could enhance the accuracy of complex ecosystem assessments by integrating multiple models with high-resolution land use simulations. Therefore, the main conclusions of this study were robust.

5.2. Zone-Based Strategies for Ecological Conservation and Land Planning

Based on spatiotemporal dynamics of habitat quality in Ningxia, this study proposed zone-based strategy for ecological conservation and land management. According to long-term trend in habitat quality and vegetation factors [35,36], Ningxia is divided into three sub-regions, i.e., Ecological Protection Zone (EPZ), Priority Control Zone (PCZ), and Buffer Zone (Figure 10). Land planning strategies for three zones are as follows.

Ecological Protection Zone (EPZ): EPZ mainly distributes in the northwest and southeast regions of Ningxia, which contains nature reserve. Land use and land cover type in EPZ are mainly grassland and wetland. Habitat quality remains relatively high in this region from 2000 to 2024. This indicated that ecosystems in these regions are stable and sustainable [37]. Specifically, EPZ contains mountainous areas and ecological sensitive regions, such as regions near Helan Mountain. The primary function of EPZ is to maintain and restore ecological balance while safeguarding vital ecological resources such as biodiversity, water sources, and natural vegetation. Land use within these zones prioritized environmental conservation, with strict restrictions on development and utilization. Key conservation efforts focuses on protecting water sources and biodiversity, preventing negative impacts from excessive development on the ecological environment.

Priority Conservation Zone (PCZ): PCZ mainly distributes in central regions of Ningxia, particularly in Yinchuan, Shizuishan, and adjacent regions. Land use and land cover type in EPZ are mainly crop land and built-up land. Habitat quality is characterized by poor and exhibits a decline trend in PCZ. Land planning strategy for PCZ is to balance ecological conservation and economic development. In PCZ, agricultural activities and infrastructure construction need strict environmental control by government.

Buffer Zone (BZ): BZ is located between the PCZ and the EPZ. Land use and land cover types in BZ are mainly crop land and vegetation. Habitat quality in BZ is characterized by medium and good. Change in habitat quality in BZ is moderate, reflecting moderate disturbance. BZ is served as recovery area to alleviate the ecological pressure of EPZ and provide transitional space for ecological restoration.

Ecological conservation strategy based on change patterns of habitat quality can provide scientific basis for sustainable development and land planning in Ningxia. Considering ecological conditions Ningxia’s ecological background and socioeconomic development, Yinchuan and Shizuishan in PCZ need constrain control on land use practices to mitigate negative impacts on ecosystem. Conversely, grassland-dominated areas should implement appropriate management measures to preserve grassland ecological functions and prevent issues such as desertification or overgrazing.

6. Conclusions

This study proposed an effective framework to conduct a mechanistic analysis of habitat quality for Ningxia, including habitat quality assessment, evolutionary process identification, and spatial heterogeneity clarification. Zone-based strategies determined by spatiotemporal evolution of habitat quality provide scientific basis for long-term ecological conservation and land planning. There are three main conclusions for this study:

(1)

From 2000 to 2024, the habitat quality of Ningxia is dominated by good habitats. The proportion of poor habitat area increased from 29.26% to 24.63%, with excellent habitats largely recovered and a net positive change of 5.12%.

(2)

From 2000 to 2024, the habitat quality transformation is remarkably distinct across both spatial and temporal dimensions. Spatially, the transformation is more pronounced in the northern and southern regions. Temporally, short-term variations exhibit localized characteristics, while long-term trends show a gradient-based interaction.

(3)

From 2000 to 2020, habitat quality patterns are influenced by natural and socioeconomic factors. Over the past two decades, vegetation factors (i.e., NDVI and NPP) are the primary driver of habitat quality evolution. Furthermore, a widespread nonlinear enhancement exists among those drivers. Strong interactions between natural factors, such as NPP, NDVI, and terrain, serve as the key mechanism that determines the current spatial differentiation of habitat quality.