Analysis of Spatial-Temporal Differentiation and Influencing Factors of Ecosystem Services in Resource-Based Cities in Semiarid Regions

Abstract

The spatial-temporal differentiation characteristics and driving mechanisms of ecosystem services are of great significance for optimizing the pattern of land spatial protection and realizing regional sustainable development. Existing studies seldom consider the segmental influence mechanism of various influencing factors on different levels of ecosystem service value (ESV). Therefore, this paper analyzes the temporal and spatial differentiation evolution characteristics of ESV in semiarid regions through an improved ESV evaluation model. The spatial panel quantile regression (SPQR) model was introduced to explore the relationship between various types of influencing factors and ESV in different intervals. The results showed the following: (1) The changes in ESV in Baotou City from 2000 to 2018 tended to be stable, but the spatial differentiation of ESV intensified. The aggregation feature of the low-ESV region is significant and gradually expanding. (2) Precipitation was the dominant factor increasing the ESV in each interval, and temperature had a significant negative impact on the low-ESV area. (3) Higher land use integrity accelerates the decline of ESV in the surrounding areas of built-up areas. The high-ESV area was more sensitive to the intensity of human activity. The direction of human activities should be effectively controlled, and the structure of comprehensive land use should be optimized to enhance the service function of regional ecosystems. This research provides new thinking for the ecological restoration zoning of regional territorial spatial planning and the sustainable development of resource-based cities.

1. Introduction

Ecosystem services are defined as the benefits derived from ecosystems [1], which are the foundation of human survival and sustainable socioeconomic development. Ecosystems provide a variety of services for human life and production, such as raw material production, climate regulation, soil conservation, water conservation, and waste disposal through ecological functions [2]. These services are necessary for maintaining ecosystems and urban development.

However, with dramatic climate change, rapid urbanization, and overexploitation of natural resources in recent years, the ecosystem services of resource-based cities have been severely threatened [3]. Increasing haze pollution weakens climate regulation and gas regulation, making it difficult to effectively circulate and purify air in cities [4,5], water pollution aggravates the contradiction between supply and demand of water resources, and rapidly expanding built-up areas occupy increasing amounts of ecological space [6]. The high intensity of human exploitation of natural resources intensifies soil erosion and weakens the regional water supply capacity [7]. Therefore, to more systematically understand these changes in ecosystem services in resource-based cities from a global perspective, it is urgent to explore and analyze the spatiotemporal evolution characteristics of ecosystem services and their internal influencing mechanisms. It is of great significance to maintain ecological balance and realize regional sustainable development.

In response to the series of changes in ecosystem service functions, many studies have explored and analyzed the factors influencing ecosystem services. Climate factors have significant impacts on gas regulation and soil erosion [8], and climate change may lead to ecosystem degradation and affect the provision of ecosystem services [9,10]. The combined intensity of human activities had a significant impact on ecosystem services [11]. The physical conditions and chemical properties of soils have also been shown to have a significant impact on the maintenance of biodiversity [12]. Furthermore, changes in land use can also affect the intrinsic structure of ecosystems and their functions [13,14] and lead to intuitive changes in ecosystem service value (ESV). For example, rapid urban expansion significantly changes the spatial pattern of regional habitats and poses a threat to habitat quality [15,16]. Ecological restoration policies have also been proven to effectively enhance the value of regional ecosystem services and improve the regional ecological environment [17]. Research from different perspectives suggests that human activities are important drivers of ESV changes [18,19]. However, the change in ESV is the result of the combined action of multiple factors, and it is necessary to fully consider various natural and human factors to conduct an in-depth analysis and interpretation of the complex ESV change.

Existing studies often use the ordinary least squares (OLS) model for the analysis of influencing factors. This global analysis model can analyze the influence direction of factors, but cannot accurately capture the spatial location information of each influence factor [20]. The spatial dependence of ESV violates the basic assumption of observation independence of the OLS model, which tends to affect the validity and unbiasedness of OLS results [21]. The spatial autoregressive model introduces spatial correlation based on the OLS to explain the spatial dependence of the respective variables and dependent variables at spatial points [22,23]. However, the model still needs to follow a strict normal distribution assumption, which is difficult to meet with actual space observation data. Furthermore, the spatial autoregressive model is unable to deeply analyze the role of spatially heterogeneous associations between the independent and dependent variables. Zhao et al. (2020) addressed spatial heterogeneity with a geographically weighted regression model (GWR) but failed to identify spatial dependencies [24]. Some studies use correlation analysis and geographic detectors to identify the role of influencing factors [25,26]. Although these methods can partially identify the relationship between influencing factors or explore fixed influencing mechanisms, they all ignore differential effects of factors on different levels of ESV.

In summary, the current research methods mainly focus on ESV influence mechanisms based on the assumption of fixed relationships between variables and independent variables, while ignoring the segmented dynamic influence effects of different factors with each level of ESV. Moreover, as a typical region where human activities strongly interact with ecosystems, resource-based cities urgently need to explore the segmented influence relationships and influence mechanisms between various influencing factors and ESV changes in the regions. Therefore, we imported long time-series, multi-source remote sensing data to improve the ESV assessment method and conducted an in-depth analysis of the correlation and influence relationship between multiple impact factors and ESV for each zone in a resource-based city in a typical semiarid region, Baotou City, through the spatial panel quantile regression (SPQR) model. We discuss the segmented effects of various influencing factors on ecosystem services and provide more reasonable and differentiated reference suggestions for optimizing the pattern of territorial spatial development and protection.

2. Materials and Methods

2.1. Study Area

Baotou (109°51′–110°26′E, 40°40′–42°43′N) is located in the western part of the Inner Mongolia Autonomous Region, adjacent to Hohhot and Ordos, and borders Mongolia in the north (Figure 1). The city is the hinterland of the Bohai Rim Economic Circle and the Yellow Economic Belt. Baotou is located at the southern end of the Mongolian Plateau and mainly consists of three major landform types: the northern hilly plateau, the central mountainous zone, and the southern plain. The study area is a typical semiarid area, and the climate type is a typical semiarid, mid-temperate continental monsoon climate, which is dry and windy in spring, hot and rainy in summer, cool in autumn, and dry and cold in winter, with little rain and snow. The annual average temperature is 8.4 °C. The annual precipitation of the region gradually increases from northwest to southeast. In the southeast, Shiquan District and Tumert Right Banner have more annual precipitation, better vegetation cover, and less evaporation; in the north, Bayan Obo and Darhan Muminggan United Banner have low annual precipitation, poor surface vegetation cover, high aridity, and strong evaporation. Most of the rivers in Baotou are valley seasonal rivers that belong to the Yellow River system and inland river system. Natural water resources are in short supply, and the per capita water resources in the city are only 16% of the average level in China. Baotou City is an important resource-based city in China and a global light, rare-earth industry center. Its mineral resources have the characteristics of large reserves, high grade, concentrated distribution, and easy mining. It is rich in iron, rare-earth, niobium, gold, limestone, dolomite, coal, oil shale, and other mineral resources, among which iron and rare-earth resources are the most abundant. The rare-earth ore resources in Baotou account for 37.8% of the global supply. It is known as “grassland steel city” and the “rare-earth capital”.

2.2. Data Sources

The main data used in this study are as follows: (1) Land use remote sensing monitoring data provided by the Resource and Environment Science and Data Center (http://www.resdc.cn, accessed on 19 November 2021), including five periods of data from 2000, 2005, 2010, 2015, and 2018, with a spatial resolution of 30 m. (2) Digital elevation model (DEM), provided by Geospatial Data Cloud (http://www.gscloud.cn/, accessed on 19 November 2021), with a spatial resolution of 30 m. (3) NDVI data, that are annual NDVI maximum values obtained by series data preprocessing and data smoothing, based on the Google Earth Engine cloud computing platform using Landsat 5/7/8 remote sensing data provided by the National Ecosystem Science Data Center (http://www.nesdc.org.cn/, accessed on 17 February 2022), including five periods of data in 2000, 2005, 2010, 2015, and 2018, with a temporal resolution of 1a and a spatial resolution of 30 m. (4) NPP data acquired from the terrestrial observation product MOD17A3 released by the National Aeronautics and Space Administration (http://lpdaac.usgs.gov, accessed on 17 February 2022), including five periods of data in 2000, 2005, 2010, 2015, and 2018, with a temporal resolution of 1a and a spatial resolution of 500 m. (5) Variable data: average annual precipitation (PRE), average annual temperature (TEM), economic density (GDP), and population density (POP), provided by the Resource and Environment Science Data Center of the Chinese Academy of Sciences (http://www.resdc.cn, accessed on 25 February 2022), including four periods of data in 2000, 2005, 2010, and 2015, with a spatial resolution of 1 km. (6) The statistical data of the average market price of grain per unit of production obtained from the Statistical Yearbook of Baotou City, the National Compilation of Information on Costs and Benefits of Agricultural Products, and the China Agricultural Products Price Survey Yearbook.

2.3. Variable Selection

The factors influencing ESV are complex and diverse. Soil, climate, and socioeconomic factors are often used as influencing factors of ecosystem services [8,27,28]. Considering the availability of influencing factor data and the actual characteristics of the study area, we selected climate factors, human activities, topography factors, and land use structure factors as explanatory variables and ESV as the response variable. The detailed descriptions of each explanatory variable are presented in the following text and in

Table 1. Classification and description of ESV influencing factors.

Types	Variable	Description
Climate factors	PRE	Average annual precipitation, mm
	TEM	Average annual temperature, °C
Human activities	GDP	Economic density, 10,000 CNY/km2
	POP	Population density, people/km2
Topographical factors	ELE	Average elevation, m
	SLO	Average slope, °
Land use structure factors	MPI	Mean proximity index
	LUI	Land use intensity index

.

2.3.1. Climate Factors

Changes in climate affect the abundance, productivity, distribution, and quality of terrestrial ecosystems [29], and all ecosystem services may be affected by climate change. Water yield and soil conservation are affected by temperature increases or changes in precipitation patterns [30,31]. Therefore, the average annual precipitation (PRE) and average annual temperature (TEM) were chosen to represent climatic factors.

2.3.2. Human Activities

With urban development, human activities and economic flows in the region are highly concentrated, which not only changes the structure of land use but also affects various processes of the ecosystem [32]. From an integrated perspective, urbanization development has a long-term impact on ecosystem services, such as hydrological regulation [33], waste disposal [34], and soil conservation [35], through population growth, industrial development, and regional expansion. Resource-based cities have long relied on natural resources for economic development, making their economic and human activities more interactive with ecosystems. Therefore, economic density (GDP) and population density (POP) were selected to represent human activities.

2.3.3. Topography Factors

It was suggested that low elevation and mild topography are beneficial for food supply [36], while the strength of soil conservation and hydrological regulation at high elevation is closely related to topographic conditions and vegetation cover [37]. Slope size also directly affects soil water content and surface runoff [38]. In resource-based cities, elevation and slope can also have an indirect effect on ESV changes by limiting or facilitating resource development, so we chose elevation (ELE) and slope (SLO) to represent topographic factors.

2.3.4. Land Use Factors

Land use factors include land use structure, quantity, and use type. The influence of these factors on ESV has been extensively studied [39,40,41]. Considering that resource-based cities occupy and partition ecological land more strongly than traditional cities, we used the mean proximity index (MPI) and the land use intensity index (LUI) to represent land use factors. The MPI reflects the proximity and fragmentation of land use patches, which can better reflect the fragmentation of land use structures [42]. The LUI represents the degree of intensification of various land uses and measures the degree of human activities on various land systems through the hierarchy of land use degrees [43]. The MPI was calculated by Fragstats4.2, and the LUI index was calculated by the method of Reference [44].

2.3.5. Descriptive Statistics of Variables

Descriptive statistics for explanatory variables are shown in

Table 2. Descriptive statistics of explanatory variables.

Variables	Observations	Unit (BL)	Max (BL)	Min (BL)	Mean (NL)	Std. (NL)
PRE	109,657	mm	507.2	74.0	7.681	0.351
TEM	109,657	°C	9.4	0.9	3.943	0.242
GDP	109,657	104 CNY/km2	179,735	0	4.062	1.815
POP	109,657	people/km2	11,990	0	2.741	1.453
ELE	109,657	m	2242.95	941.30	7.226	0.170
SLO	109,657	°	33.41	0.74	1.565	0.454
MPI	109,657	/	9.93	0	0.455	0.542
LUI	109,657	/	415.44	88.98	5.362	0.216

Note: BL represents the variables before taking the natural logarithm, and NL represents the variables after taking the natural logarithm.

. Standard deviations of explanatory variables ranged from 0.17 to 1.815. Among them, the standard deviations of GDP and POP were the largest, 1.815 and 1.453, respectively, indicating that the economic intensity and population distribution of different regions greatly varied. After the correlation test, there was no high correlation coefficient between the variables, so there was no multicollinearity problem between the variables (the correlation coefficients are shown in Table S5). The spatial distribution of variables is shown in Figure 2 (due to the limited space, only the spatial distribution of variable data in 2015 is shown).

2.4. Methods

We used the ecosystem service value (ESV) assessment model and the spatial panel quantile regression (SPQR) model to analyze the spatial-temporal evolution and influence mechanism of ecosystem services in the study area. First, we explored the evolution of ESV in Baotou and the characteristics of ESV distribution changes in various regions through the revised ESV assessment model and hotspot analysis tool. Then, we explored the segmental effects of climate, human activities, topography, and land use factors on ESV in different intervals through the SPQR model.

2.4.1. Ecosystem Service Value Assessment Model

Ecosystem service value (ESV) is the sum of all service functions and natural assets provided by all ecosystems in a region and varies with the type, size, and quality of different ecosystems. Costanza proposed an ESV assessment method based on unit area [1], and XIE Gao-di improved this method and it has been widely used in various ecosystem services research [45]. Although this method can systematically calculate regional ESV, the accuracy of the calculated ESV is often insufficient, especially at the municipal and smaller scales. Vegetation cover and NPP have important influences on ecosystem services, such as soil conservation and climate regulation in various ecosystems [42,46]. Therefore, we introduced NDVI and NPP remote sensing data to improve the traditional ESV evaluation model to quantitatively calculate the ESV in the study area through remote sensing. The specific formula is as follows:

$$ESV={\displaystyle \sum }_{t=1}^{c}Vt$$

(1)

where $ESV$ is the total ESV (CNY) in the study area, $t=1,2,\dots ,n$ is the type of ecosystem (ecosystems in this study include cropland, grassland, woodland, water, and unused land, but building land is not considered [32]), and $Vt$ is the ESV of ecosystem type $t$.

$$Vt={\displaystyle \sum }_{i=1}^p{\displaystyle \sum }_{j=1}^{q}V{t}_i\times M_{ij}\times R_{ij}$$

(2)

where $i=1,2,\dots ,p$ is the type $i$ ecosystem service function of ecosystem $t$, $V{t}_i$ is the value per unit area of the type $i$ ecosystem service function of ecosystem $t$, $j=1,2,\dots ,q$ is the number of pixels in the ecosystem-like system, and $M_{ij}$ is the area of each pixel. $R_{ij}$ is the adjustment coefficient, which is determined by the quality of the ecosystem. The calculation formula is as follows:

$$R_{ij}=\left(\frac{f_j}{f_{mean}}+\frac{NP{P}_j}{NP{P}_{mean}}\right)/2$$

(3)

where $f_j$ is the $f$ value of the $j$ pixel, $NP{P}_j$ is the net primary production of the $j$ pixel, and $f_{mean}$ and $NP{P}_{mean}$ are the average values of $f$ and NPP, respectively, of the $t$ ecosystem in the study area. The $f$ value is calculated by the pixel dichotomy model, and the calculation formula is as follows:

$$f_j=\frac{NDV{I}_j-NDV{I}_s}{NDV{I}_m-NDV{I}_s}$$

(4)

where $NDV{I}_j$ is the vegetation normalization index of the $j$ pixel, $NDV{I}_m$ is the NDVI value of the pixel completely covered by vegetation, and $NDV{I}_s$ is the NDVI of the bare land ecosystem or the pixel without vegetation cover. $NDV{I}_m$ and $NDV{I}_s$ are often replaced by the maximum and minimum NDVI values in the region, respectively.

$V{t}_i$ was based on the ESV equivalent table [45] and was revised based on the average grain yield price in Baotou City from 2000 to 2018 to obtain the equivalent table of ESV per unit area in the study area (

Table 3. Ecosystem service value equivalent per unit area in the study area (CNY/hm² · year).

Ecosystem Service Functions	Ecosystem Type
	Forestland	Grassland	Cultivated Land	Waters	Other Land
Supply services	9427	3920	1702	11,258	916
Regulation services	40,226	16,694	11,301	64,874	1988
Support services	25,048	9570	7496	53,029	830
Cultural services	2975	1245	243	6352	343

).

2.4.2. Hotspot Analysis

The hotspot analysis tool can visually reflect the spatial aggregation of high and low values within a region through hotspot maps [47]. In this study, the study area was divided into 1 km × 1 km grid cells, and the spatial aggregation characteristics of ESVs in Baotou city over the years were analyzed by the index of the Hotspot Analysis module in ArcGIS 10.5. The $Getis-Ord G_i^{*}$ index was calculated as follows:

$$G_i^{*}=\frac{{{\displaystyle \sum }}_j^n{W}_{ij}\left(d\right)x_i}{{{\displaystyle \sum }}_j^n{x}_j}$$

(5)

$$Z\left(G_i^{*}\right)=\frac{G_i^{*}-E\left(G_i^{*}\right)}{\sqrt{Var\left(G_i^{*}\right)}}$$

(6)

where $E\left(G_i^{*}\right)$ is the mathematical expectation of $G_i^{*}$, $Var\left(G_i^{*}\right)$ is the variance of $G_i^{*}$, and $W_{ij}$ is the spatial weight. When $Z\left(G_i^{*}\right)<-2.58$, it is the high cold-spot area of ESV; when $-2.58<Z\left(G_i^{*}\right)<-1.96$, it is the intermediate cold-spot area of ESV; when $-1.96<Z\left(G_i^{*}\right)<-1.65$, it is the low cold-spot area of ESV; when $-1.65<Z\left(G_i^{*}\right)<1.65$, it is the nonsignificant area of ESV; when $1.65<Z\left(G_i^{*}\right)<1.96$, it is the low hotspot area of ESV; when $1.96<Z\left(G_i^{*}\right)<2.58$, it is the intermediate hotspot area of ESV; when $Z\left(G_i^{*}\right)>2.58$, it is the high hotspot area of ESV [48].

2.4.3. Sensitivity of Ecosystem Service Values

As the value coefficients of the ESV valuation model were corrected, we verified the impact on the uncertainty of the results by using the ecosystem service value sensitivity analysis [11], and the formula is as follows:

$$CS=\frac{\left(ES{V}_b-ES{V}_a\right)/ES{V}_a}{\left(V{C}_{bi}-V{C}_{ai}\right)/V{C}_{ai}}$$

(7)

where $CS$ is the sensitivity coefficient, $V{C}_{bi}$ is the value coefficient after adjustment for land use type $i$, $V{C}_{ai}$ is the value coefficient before adjustment for land use type $i$, and $ES{V}_b$ and $ES{V}_a$ are the total ESV of the study area after and before adjustment of the value coefficient, respectively. If $CS$ > 1, the ESV is elastic relative to the $VC$, and the results are less credible. If $CS$ < 1, the ESV is inelastic relative to the $VC$, and the results are more reliable [18]. We adjusted the $VC$ of each land use type in the study area by (±)50% to calculate $CS$.

2.4.4. Spatial Panel Quantile Regression (SPQR) Model

Compared with the traditional spatial econometric model, the SPQR model can effectively explain the spatial dependence and heterogeneity of the independent variable on the dependent variable [21], and show in detail the correlation between each level of ESV and the influencing factors. This method does not have strict requirements on the data distribution. To build the SPQR model, we first accepted its basic model—the quantile regression model. The quantile regression model is a linear function regression model that fits the independent variable X based on the distribution conditions of the explanatory variable Y [49]. Compared to the OLS model, the quantile regression model evaluates the parameters from any quantile point, makes no specific assumptions about the distribution of the error term, and is much less sensitive to outliers than mean regression, which makes the model yield more robust results [21].

Panel data can better control the heterogeneity of individual data, ensure unbiased analysis results, and provide a rich amount of information [50]. Therefore, with the advantage of panel data, quantile regression model calculation by panel data can accurately measure the marginal effects of dependent variables on independent variables at each quantile to more objectively analyze the effects of independent variables on the conditional distribution of dependent variables at different quantile points. The basic form of the panel quantile regression model is as follows:

$$Q_{y_{it}}\left(\tau |x_{it}\right)=x_{it}^T\beta \left(\tau \right)+{\alpha }_i , i=1,\dots ,p;t=1,\dots ,q$$

(8)

where $y_{it}$ is the observed value of the dependent variable $i$ at period $t$, $x_{it}$ is the corresponding observed value of the independent variable, and ${\alpha }_i$ represents the difference between individuals that do not change with quantile τ and are not controlled by other variables.

To evaluate Formula (7), the penalty term ${{\displaystyle \sum }}_{i=1}^p\left|{\alpha }_i\right|$ was added, and the estimates of the panel quantile regression parameters were obtained by solving the following objective function with a penalty term:

$$\begin{array}{c}min\\ \left(\alpha ,\beta \right)\end{array}{\displaystyle \sum }_{j=1}^n{\displaystyle \sum }_{i=1}^p{\displaystyle \sum }_{t=1}^q{\omega }_j{\rho }_{{\tau }_j}\left(y_{it}-{\alpha }_i-x_{it}^T\beta \left({\tau }_i\right)\right)+\lambda {\displaystyle \sum }_{i=1}^p\left|{\alpha }_i\right|$$

(9)

where ${\omega }_j$ is the weight of the $j$ quantile, which is used to control the relative influence of the $n$ quantile ${\tau }_i$ on parameter ${\alpha }_i$. $\lambda$ is the penalty factor, and when $\lambda \to 0$, the fixed effect is estimated at this time, and the estimated value corresponding to the fixed effect can be obtained; when $\lambda \to \infty$, the fixed effect is eliminated, and the estimated value after the fixed effect is eliminated can be obtained.

To analyze the effect of multiple influencing factors on ESV in regions with different ESV levels and how this relationship changes, the following model was developed based on panel quantile regression:

$$\begin{gathered} Q_{ESV}\left(\tau |x_{it}\right)={\beta }_1\left(\tau \right)PR{M}_{it}+{\beta }_2\left(\tau \right)TE{M}_{it}+{\beta }_3\left(\tau \right)GD{P}_{it}+{\beta }_4\left(\tau \right)PO{P}_{it}+{\beta }_5\left(\tau \right)EL{E}_{it}+ \\ {\beta }_6\left(\tau \right)SL{O}_{it}+{\beta }_7\left(\tau \right)MP{I}_{it}+{\beta }_8\left(\tau \right)LU{I}_{it}+{\alpha }_i\left(\tau \right)+{\mu }_{it} \end{gathered}$$

(10)

where $Q_{ESV}\left(\tau |x_{it}\right)$ is the $\tau$ conditional quantile of the dependent variable ESV, $\tau$ is the quantile, $\tau \in \left(0,1\right)$, and $\beta \left(\tau \right)$ is the regression coefficient at the $\tau$ quantile. $PR{M}_{it}$, $TE{M}_{it}$, $GD{P}_{it}$, $PO{P}_{it}$, $EL{E}_{it}$, $SL{O}_{it}$, $MP{I}_{it}$, and $LU{I}_{it}$ are the independent variables, ${\alpha }_i\left(\tau \right)$ is the fixed effect at the $\tau$ quantile, and ${\mu }_{it}$ is the residual term. The 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, and 90% quartiles of the model were selected to provide a clearer understanding of the dynamics of ESV as the values of the influencing factors increased.

Limited by data availability, we used four periods of spatial panel data for 2000, 2005, 2010, and 2015 for the analysis of the ESV influence mechanism in Baotou. In the model calculation, the presence of heteroskedasticity affects the actual regression efficiency, so to reduce the impact of data heteroskedasticity on the analysis results, we applied the natural logarithm treatment for all the data [51].

3. Results

3.1. Change of Land Use Pattern

The spatial distribution pattern of land use in Baotou City in 2000, 2005, and 2019 is shown in Figure 3. The main land type of Baotou is grassland and cultivated land. Grassland is mainly distributed in the northern region and the Daqingshan area in the southern region, while cultivated land is distributed in the central region and the southernmost part of Baotou City. From 2000 to 2018, various types of land underwent intricate changes, but the most significant changes were in grassland, cultivated land, and construction land (Figure 4). Grassland and cultivated land decreased by −428.68 km² and −355.26 km², respectively, and the reduced grassland and cultivated land were mainly transformed into construction land. The construction land increased by 78.67%, and the expansion area was 620.08 km². The expansion areas were mainly located in Kundulun District, Qingshan District, Donghe District, and Tumote Right Banner. Changes in forestland, waters, and other land types were minor (detailed land change data are shown in Tables S1–S4).

3.2. The Spatiotemporal Evolution of Ecosystem Service Value

The spatial distribution pattern of ESV in Baotou City from 2000 to 2018 is shown in Figure 5. In general, the overall ESV in Baotou City has declined in the past ten years, with a decrease of 2.43 × 10⁶ CNY from 2000 to 2018. The regional differentiation of ESV was very significant. The northern region of Baotou City was generally low in ESV, although it is mainly a grassland ecosystem. The ESV in the northern region gradually improved after 2010 and reached stability in 2018. The central region, which is dominated by grassland ecosystems and agroecosystems, benefited from the stable support of regional ecosystem services, and the ESV fluctuated less during the decade. In the southern region, except for Kundulun District, Qingshan District, Shiguai District, and other main urban areas that maintained a relatively low ESV level, the ESV levels in other regions were maintained at a relatively high level, and the distribution was relatively concentrated.

The spatial change pattern of ESV in Baotou from 2000 to 2018 is shown in Figure 6. The total area of ESV negative growth area in the study area during the decade was 6821.13 km², and the ESV decreased by 3979.52 × 10⁶ CNY in total, accounting for 24.74% of the total area of the study area, mainly in the southeastern part of Darhan Muminggan United Banner, eastern Guyang County, and southern Tumote Right Banner. The ESV positive growth area in the study area increased by a total of 3971.01 × 10⁶ CNY, involving an area of 8000.20 km², accounting for 29.02% of the area of Baotou City, mainly located in north–central Darhan Muminggan United Banner and the northern part of Tumote Right Banner. From the trend of the evolutionary characteristics of ESV, the evolutionary process of ESV in Baotou can be divided into four stages. In the first stage (2000–2005), with the rapid rise of industry and the mining industry, the ESV in most regions of Baotou City showed a decreasing trend. Due to the intensive distribution of the mining industry, the ESVs of Darhan Muminggan United Banner and Bayan Obo in the northern region declined most significantly, by 7.16% and 12.08%, respectively. In the second stage (2005–2010), the evolution of ESV in Baotou City showed an east–west divergence, with a slight increase in ESV in the western region and a slight decrease in the eastern region. In the third stage (2010–2015), the ESV of the southern region around the central city and Daqingshan area was significantly reduced due to the frequent human resource extraction activities in the pre-urbanization period. In the fourth stage (2015–2018), ESV increased around the central urban area of the southern region, which may be related to the increase in environmental protection and ecological restoration during the urban development process.

In terms of ESV composition, the relative proportions of the four ecosystem services were relatively stable (

Table 4. Proportion of various ecosystem services in Baotou.

	Year	Supply Services	Regulation Services	Support Services	Cultural Services	Total
Ecosystem services value (106 CNY)	2000	3622.64	22,612.74	14,145.06	3041.90	43,422.34
	2005	3672.57	22,855.63	14,243.74	3079.72	43,851.66
	2010	3616.42	22,573.60	14,012.48	3039.92	43,242.43
	2015	3671.08	22,990.93	14,319.36	3103.39	44,084.76
	2018	3594.38	22,735.30	14,020.35	3069.89	43,419.92
Mean of ESV (106 CNY)		3635.42	22,753.64	14,148.20	3066.96	43,604.22
Mean percentage (%)		8.34	52.18	32.45	7.03	100.00

). Regulation services accounted for the highest proportion, at 52.18% (the average proportion from 2000 to 2018), indicating that the ecosystem in Baotou City mainly plays the role of climate regulation, gas regulation, and hydrological regulation. This was followed by support services (32.54%), indicating that soil conservation and maintaining biodiversity are also the main service functions to maintain the stability of ecosystem services. The supply services only accounted for 8.34% of the total ESV, which is closely related to grassland degradation and the overall low quality of cultivated land in Baotou.

3.3. Hotspot Analysis of Ecosystem Service Value

We further explored the spatial aggregation characteristics and change patterns of ESV in Baotou from 2000 to 2018 through hotspot analysis (Figure 7). The high-ESV areas were mainly clustered in the eastern and southeastern regions, while the low-ESV areas were mainly clustered in the northern grasslands and the southwestern main urban areas, while the spatial aggregation characteristics of ESV in most of the remaining areas were not significant. From the perspective of temporal changes, the high-ESV aggregation areas changed significantly between 2000 and 2010, among which the high-ESV aggregation area in the southern region significantly expanded and showed a continuous trend. The high-ESV aggregation areas in the eastern part of Darhan Muminggan United Banner significantly shrank. The low-ESV aggregation areas significantly decreased, and the decreased area was mainly distributed in the grassland area in the northern region. From 2010 to 2018, the high-ESV aggregation areas in Baotou City transitioned from “high aggregation” to “scattered distribution”. From the changes in the area of cold spots and hotspots (Figure 8), the area of cold spots and hotspots in Donghe, Jiuyuan, Kundulun, Qingshan, and Shiguai significantly increased, while the area of Bayan Obo continued to increase. In general, the distribution of ecosystem services shows significant differences. Urbanization gradually expanded the ESV cold-spot areas around urban areas, while the core hotspot areas of ESV remained stable, providing a guarantee for maintaining the stability of regional ecosystem services.

3.4. Sensitivity Analysis of Ecosystem Service Value

After adjusting the value coefficient of ESV by (±)50%, the CS values from 2000 to 2018 were less than 1 (

Table 5. Coefficient of sensitivity of ESV.

	Forestland	Grassland	Cultivated Land	Waters	Other Land
2000	0.09	0.56	0.16	0.24	0.01
2005	0.06	0.53	0.16	0.21	0.01
2010	0.08	0.56	0.16	0.26	0.00
2015	0.08	0.59	0.18	0.21	0.01
2018	0.06	0.58	0.17	0.21	0.01

). This indicates that ESV is inelastic to VC and that the ESV calculation is credible. The ESV coefficients for various land use types used in the calculations of this study are appropriate and consistent with the actual situation in the study area.

3.5. Segmental Influence Effects of Influencing Factors on ESV

According to the Hausman test, the p-value is significant at the 1% level, so we chose the fixed-effect model to analyze the influencing factors of the ecosystem service value (ESV) in Baotou City. The regression results of the fixed-effect model show (

Table 6. The regression results of SPQR of influencing factors under different quantiles.

	PRE	TEM	GDP	POP	ELE	SLO	MPI	LUI
Fixed effects	0.997 (0.000 ***)	−0.384 (0.000 ***)	0.21 (0.000 ***)	−0.249 (0.000 ***)	−0.501 (0.000 ***)	0.332 (0.000 ***)	0.006 (0.062 *)	0.542 (0.000 ***)
Quantile 0.10	0.162 (0.000 ***)	−0.559 (0.000 ***)	0.037 (0.000 ***)	−0.111 (0.000 ***)	−0.606 (0.000 ***)	0.756 (0.000 ***)	−0.021 (0.003 ***)	1.248 (0.000 ***)
Quantile 0.20	0.321 (0.000 ***)	−0.291 (0.000 ***)	0.052 (0.000 ***)	−0.073 (0.000 ***)	−0.055 (0.144)	0.559 (0.000 ***)	−0.109 (0.000 ***)	0.778 (0.000 ***)
Quantile 0.30	0.377 (0.000 ***)	−0.123 (0.000 ***)	0.040 (0.000 ***)	−0.031 (0.000 ***)	0.314 (0.000 ***)	0.407 (0.000 ***)	−0.115 (0.000 ***)	0.494 (0.000 ***)
Quantile 0.40	0.401 (0.000 ***)	−0.056 (0.000 ***)	0.028 (0.000 ***)	0.003 (0.174)	0.446 (0.000 ***)	0.317 (0.000 ***)	−0.087 (0.000 ***)	0.320 (0.000 ***)
Quantile 0.50	0.429 (0.000 ***)	−0.018 (0.231)	0.021 (0.000 ***)	0.023 (0.000 ***)	0.466 (0.000 ***)	0.262 (0.000 ***)	−0.046 (0.000 ***)	0.165 (0.000 ***)
Quantile 0.60	0.457 (0.000 ***)	0.003 (0.838)	0.016 (0.000 ***)	0.037 (0.000 ***)	0.429 (0.000 ***)	0.222 (0.000 ***)	0.000 (0.949)	−0.003 (0.640)
Quantile 0.70	0.496 (0.000 ***)	−0.006 (0.682)	0.004 (0.000 ***)	0.063 (0.000 ***)	0.361 (0.000 ***)	0.174 (0.000 ***)	0.039 (0.000 ***)	−0.192 (0.000 ***)
Quantile 0.80	0.522 (0.000 ***)	−0.001 (0.968)	−0.005 (0.000 ***)	0.088 (0.000 ***)	0.353 (0.000 ***)	0.127 (0.000 ***)	0.079 (0.000 ***)	−0.389 (0.000 ***)
Quantile 0.90	0.560 (0.000 ***)	0.023 (0.350)	−0.005 (0.000 ***)	0.113 (0.000 ***)	0.382 (0.000 ***)	0.082 (0.000 ***)	0.106 (0.000 ***)	−0.570 (0.000 ***)

Note: There were 109,657 observations. ***, **, and * represent significance levels of 1%, 5%, and 10%, respectively.

) that average annual precipitation (PRE) had a significant positive effect on ESV, indicating that higher average annual precipitation can effectively promote ESV. In contrast, the regression coefficient of average annual temperature (TEM) was negative, implying that the stability of the ecological environment requires a suitable temperature, and a higher temperature is not conducive to the stability of ecosystem services. Furthermore, population density (POP) had a significant negative correlation with ESV.

The fixed-effects model gives a basic judgment of the effects of the respective variables on ESV, while the results of spatial panel quantile regression (SPQR) help analyze the differences in the effects of the factors at different ESV levels and the patterns of change. The quantile regression results show (Table 6) that the regression coefficients of PRE were significant at each quantile and gradually increased from low to high (Figure 9). It shows that the dependence of ESV from low to high values on precipitation is increasing, and PRE is the dominant factor to enhance ESV. Adequate precipitation can effectively improve the raw material production, climate regulation, hydrological regulation, soil conservation, and other ecosystem service functions of various ecosystems, such as forestland, grassland, and cultivated land, thereby maintaining the overall stability of the ecological environment. The quantile regression results of TEM show that the regression coefficient of TEM was only significant in the quantile 0.1–0.4 range, indicating that TEM only had a negative effect on the low-ESV areas. Since the low-ESV area is mainly located in the main urban areas, the population concentration, industrialization, and urban expansion cause the ESV to continue to decline and form a typical “heat island effect”, which makes the TEM and low ESV form a negative correlation effect.

Although the regression coefficients of economic density (GDP) were significant in all quartiles, the regression coefficients were all less than 0.06, which indicates that GDP had little effect on ESV. Different from GDP, the regression coefficient of POP from the low quantile to the high quantile changed from negative to positive and showed a significant upward trend (Figure 9). In the low-ESV range, POP had a certain inhibitory effect on ESV, but this negative effect gradually diminished and had a positive influence as the ESV rose. This result fully indicates that ESV is generally low in densely populated areas with a high intensity of human activities, while in high-ESV areas, human activities are low, and the ecosystem tends to be stable.

The regression coefficient of elevation (ELE) changed from negative to positive, reaching a maximum value of 0.466 at quantile 0.5, showing that at higher ESV levels, the influence of mean altitude on ESV tended to stabilize at higher altitude levels. The regression coefficient of slope (SLO) was always significantly positive at the 1% level, but its effect on ESV was more obvious in the mid–low-ESV range.

The quantile regression coefficient of the mean proximity index (MPI) was negative in the quantile 0.1–0.5 interval and positive in the 0.7–0.9 interval. In areas with low ESV, higher land use integrity was not conducive to the stability of ESV. In contrast, higher land use integrity can promote ESV improvement in areas with high ESVs. The reason for this is that low-ESV areas are mostly located in built-up urban areas, where the expansion of built-up urban areas gradually compresses the natural ecosystem space and weakens the supply of ecosystem services. In areas with high ESVs, intact ecosystem land can better-maintain natural ecosystem services.

The magnitude of the land use intensity index (LUI) visualizes the effects of human activities on ecosystems. The results of SPQR of LUI showed that LUI had a significant positive effect on low ESV and medium ESV. In the high-ESV interval (quantile 0.7–0.9), the regression coefficients of LUI were all negative, and the negative correlation gradually increased. This shows that human development activities can significantly affect the balance of the ecological environment, and the high-ESV areas where the development activities are less active are more likely to suffer from the coercion of land development and utilization activities.

4. Discussion

4.1. Different Contributions of Natural and Human Factors to ESV

The study of the evolutionary characteristics of ecosystem services and their influence mechanisms is an important indicator for understanding the relationship between humans and the natural environment and whether urban development and ecosystems are in harmony, as well as an important part of urban sustainable development research. The change in ecosystem service function is not a simple process, and the change in various factors may directly change the original change trend of ecosystem service function, so grasping the intrinsic mechanism of its action has profound research significance.

In semiarid regions, natural and human activities have different effects on ecosystems. Some researchers found that rainfall and the normalized vegetation index (NDVI) were determinants of the trade-offs of ecosystem services in the Yellow Huaihai Plain through quantile regression models [52], while others found that high-intensity underground coal mining did not lead to a decline in regional ecosystem services through ecosystem evaluation models in coal mining areas [53]. However, the influence mechanisms of ESV in different regions of resource-based cities in the semiarid region are still unclear, so we focused on exploring the segmental influence effects of ESV in the region. From the segmental influence mechanism of ESV change, it was found that the high-ESV areas, which are mainly ecological lands, have dense vegetation, strong water connotation functions, and perfect ecological circulation systems, and play an important role in maintaining ecological stability, climate regulation, soil conservation, and biodiversity [54]. These areas have a stronger ability to utilize natural factors, and natural factors have a good promotion effect on the high ESV. However, these areas are also more vulnerable to the stress of intensive development activities. The increase in land use intensity will lead to fragmentation and separation of landscape patterns in the area and damage to the ecosystem structure [55], which in turn lead to a decline in the ability of climate regulation and hydrological regulation. Maintaining the coherence and integrity of regional ecosystems is the main way to consolidate and enhance ecosystem service functions.

In addition, densely populated low-ESV regions are less influenced by natural factors and more influenced by socioeconomic activities such as land development and urban expansion. In contrast, low-ESV regions with sparse population are dominated by natural factors such as low vegetation cover, severe sandy land, and grassland degradation, which dominate the process of ecosystem service changes in the region [56,57]. Therefore, enriching the ecosystem types of densely populated low-ESV areas and strengthening human ecological management in sparsely populated low-ESV areas are high-quality strategies to improve the overall ESV of the region.

4.2. Implications for the Sustainable Development of Resource-Based Cities

As a typical resource-based city, Baotou has sustainable influences on regional ecosystems from its anthropogenic mineral extraction activities, but these influences greatly vary in different regions. Although the high-ESV areas provide the most significant ecosystem service function security in the study area, they are more sensitive to the impact of mining activities and have weaker self-protection capabilities, while the medium-ESV and low-ESV areas have stronger adaptability and resilience. Therefore, high-ESV areas with relatively concentrated mining activities should be focused on, and it is particularly critical to maintain the balance of mineral resource exploitation in these areas and maintain the connectivity between ecosystems (Figure 10). Vulnerable or sensitive areas with high ESV should be classified as restricted development areas or key protection areas to avoid further damage to the regional ecological environment. Furthermore, traditional studies suggest that economic development affects the balance of ecosystem services to varying degrees [58,59]. However, as a typical semiarid city in northern China, Baotou has an overall low economic density that has not reached the conflict line with the ecosystem, and the ecosystem has a considerable range of resilience [32]. Therefore, from the perspective of space, economic density and ecological systems are not completely opposite.

4.3. Limitations

In this paper, the improved ESV assessment model based on multi-source remote sensing data and the SPQR model were used to intuitively discuss the changes of ecosystem services in various parts of the study area, as well as the segmental effects of various factors on different levels of ESV. However, there are still some limitations in this paper. Firstly, the factors affecting ESV are diverse, but due to the constraints of the study model and data availability, it was not possible to include all other influencing factors in this study. Secondly, the research analysis of the ESV segmental influence mechanism is not deep enough, and it is not yet possible to specifically analyze the combined influence effect of multiple factors in each interval and how the influence is spatially expressed. Therefore, in the next step of research, we will explore or calculate more feasible relevant influencing factor data, improve the framework of influencing factors, and try to combine other analysis methods to deepen the research on the influence mechanism of ESV.

5. Conclusions

This study revealed the distribution characteristics and spatial evolution of ecosystem service value in typical semiarid regions through an improved ESV assessment model based on multi-source remote sensing data. The SPQR model was used to explore the subsection influence mechanisms of natural factors, human activities, and land use factors on various levels of ESV. The research results showed that:

(1)

From 2000 to 2018, the spatial differentiation of the ESV in Baotou intensified. The ESV cold-spot region had significant aggregation characteristics and a trend of gradual expansion. Regulation services dominated the fluctuation of ESV throughout the study area, and the forestland ecosystem provided the core ecosystem services for the region.

(2)

Among all the driving factors, PRE was the dominant factor that promoted the elevation of ESV. Due to the heat island effect, TEM had a significant negative impact on the low-ESV areas. Appropriate human activities had a positive effect on ESV enhancement in areas with a better natural substrate and stronger ecosystem service functions. The single land use structure restricted the ESV enhancement in areas with weak ecosystem services, and the enrichment of land use types in the region is an effective way to improve the low ESV.

(3)

Natural ecological areas with higher ESVs are more sensitive to the intensity of human activities and vulnerable to human development activities. These areas should strictly limit the direction of human activities and adopt stricter ecological control measures to reduce the risk of over-exploitation. It is also crucial to maintain the connectivity and integrity of natural ecosystems, which is important for the sustainable development of resource-based cities.