2.3.1. Evaluation of ESV
In this study, we used spatial models and economic methods [
25,
26] to map the ESV in Dalian. Moreover, to eliminate the impact of price changes in each period, we adopted the constant price in 2015. The value of food production was calculated based on its vendibility [
27,
28]. The value of the water conservation service was quantified using the reservoir water storage cost method [
13] using the water yield model of Integrated Valuation of Ecosystem Services and Trade-offs (InVEST). We applied the carbon tax rate and industrial oxygen production method [
13] to quantify the value of carbon sequestration and oxygen release services, both of which were obtained from net primary productivity (NPP) based on the Carnegie–Ames–Stanford approach model (CASA) [
29]. The Revised Universal Soil Loss Equation (RUSLE) model, market value method, and shadow project approach were applied to map the value of soil retention services. To calculate the value of habitat support, we used the InVEST habitat quality model [
30]. Xie et al. improved Costanza et al.’s research in China based on expert knowledge methods. Their compilation of the Chinese ecosystem service equivalent value per unit area has been widely used. Referring to Xie et al., we mapped the value of landscape aesthetics to land use types [
31].
(1) Food production
Food production reflects human livelihoods and food security in countries. In this study, food production was quantified as the production of food from agriculture (e.g., grains, fruits, and vegetables), livestock (e.g., meat, eggs, and milk), and fishing (e.g., shrimp, crab, and others). The calculation formula is as follows:
where Sf is the value of the food production service (CNY/ha/a), Q
i represents the annual yield per hectare of the ith food (t/ha/a), P
i is the market price of the ith food (CNY/kg), and B is the average profit margin of food sales. The above factors were obtained from the Statistical Yearbook established by the Dalian Municipal Bureau of Statistics.
(2) Carbon sequestration and oxygen release
Based on the photosynthesis and respiration data of the vegetation, 1.63 units of carbon can be sequestered, and 1.2 units of oxygen can be released for every unit of NPP accumulated by the vegetation [
32]. The value of carbon sequestration and oxygen release services were calculated using Formulae (2) and (3).
where S
c is the value of carbon sequestration and oxygen release services (CNY/ha/a), C
c is the social cost of carbon [
33], and C
O is the cost of industrial oxygen production. NPP(x, t) is the NPP of position x at time t, APAR(x, t) is the available photosynthetic radiation absorbed by position x at time t (MJ·m
−2), and ε(x, t) is the light utilization efficiency of pixel x at time t (gC·MJ
−1).
(3) Water conservation
Water conservation services play a significant role in controlling soil desertification and reducing soil erosion. We used the InVEST water yield model to quantify the amount of water conservation. The fundamental formulae for the value of water conservation services are as follows:
where SW is the value of the water conservation service (CNY/ha/a), WR is the amount of water conservation (mm), and C is the reservoir construction cost (CNY/m
3). Y
x is the annual yield at pixel x (mm), P
x is the average annual precipitation at pixel x (mm), and AET
x is the actual annual evapotranspiration at pixel x (mm). The above factors of the InVEST water yield model are ensured by referring to the InVEST guidebook. The velocity is the coefficient, TI represents the terrain index, and Ksat means the soil saturated water conductivity (cm/d).
(4) Soil retention
The RUSLE model was used to estimate the potential and actual soil erosion in the study area. The spread between them is the amount of soil retention [
34]. The value of soil retention services due to vegetation was quantified based on the role of vegetation in reducing land loss and sediment accumulation. The formulae are as follows:
where S
s is the value of soil retention services (CNY/ha/a), S is the amount of soil retention (t/ha/a), and 24% represents the ratio of mud and sand accumulated in reservoirs, rivers, and lakes in China [
35]. C is the reservoir construction cost (CNY/m
3), ρ is the soil bulk density (g/cm
3), d is the average soil thickness (m), and F is the average forestry income (yuan/ha). R represents the erosion factor of rainfall, K represents soil erodibility, LS represents slope length and slope, C represents vegetation cover and crop management factors, and P stands for soil and water conservation measures. The process for calculating R, K, LS, C, and P factors is described in Remortel et al., Williams and Arnold, and Wischmeier and Smith [
36,
37,
38].
(5) Habitat support
Habitat support is the potential for an ecosystem to create conditions for species survival and reproduction, which is reflected in the habitat quality index [
39]. The distribution of habitat quality was obtained using a combination of landscape type sensitivity and external threat intensity based on InVEST habitat quality. Given the primary protected animals in Dalian (e.g., migratory birds, vipers, and harbor seals), we regarded the land use types of buildings, farmland, and aquaculture as external threat factors.
where S
h is the value of habitat support services (CNY/ha/a), Q
xj is the habitat quality of grid x in land use type J,
is the average annual habitat quality index, and VB is the baseline value of biodiversity (CNY/ha/a). H
j is the habitat suitability of land use type J, D
xj is the stress level of grid x in land use type j, K is the semi-satiety constant and is usually half of the maximum value of D
xj, and Z is the conversion factor. The ratio 1/7 represents the equivalent coefficient of the ESV, A
j represents the unit area value equivalent factor for the habitat support service provided by land use type j, Q is the average grain yield per unit area (kg/ha), and P
f is the average market price of grain in Dalian in 2015 (CNY/kg).
(6) Landscape aesthetics
Landscape aesthetics are embodied in the evaluation of landscapes with recreational, cultural, and artistic value. In addition, considering that there is some error in representing regional characteristics with a national parameter table, it is necessary to correct this. Since tourism can indirectly reflect landscape aesthetics, the service value is revised according to tourism income.
where S
l is the value of landscape aesthetic service (CNY/ha/a), t is the average tourism income of Dalian (CNY/ha), and T is the national average tourism revenue (CNY/ha). Additionally, 1/7, S
j, S, Q, and P
f are the same as above. L
j is the unit area value equivalent factor for the landscape aesthetic service provided by land use type j.
2.3.2. Calculation of Driving Factors
(1) Human active index
The human active index (HAI) has spatial variability, reflecting the influence of anthropogenic activities on land use and landscape composition changes. As a result of anthropogenic activities, the original natural characteristics of the landscape components are decreasing, and different types of landscape components represent different characteristics of anthropogenic activities or exploitation intensity. The following expression is used to calculate HAI:
where HAI is the anthropogenic influence index, N is the number of landscape types, A is the total landscape area, P
i is the anthropogenic influence intensity parameter reflected by the landscape component, and TA is the total landscape component area. P
i was determined using the Lohani checklist method, the Leopold matrix method, and the Delphi method [
40,
41,
42,
43]. To reduce the error, we took the average of the three coefficients as the coefficient (
Table 1).
(2) Land use intensity index
The land use intensity index (LUI) uses a discontinuous function to express land use intensity [
44], reflecting the degree of human development and utilization of land. Li et al. hypothesized that different land use types can be attributed to LUI with different utilization intensities [
42]. Mudflats and unused land are classified as LUI-1; forest, grassland, and open water as LUI-2; and aquaculture as LUI-3. The agricultural land type is LUI-4, and construction land is LUI-5. Therefore, this study divides the land use types in the study area into five types corresponding to different LUIs (
Table 2).
In the actual state, land use types are randomly combined in the same area, each with a different area weight, contributing to the local LUI according to their weights. Thus, the comprehensive quantitative indicators of LUI can be mathematically synthesized. The magnitude of their values integrally reflects the degree of land use:
where L is the composite index of LUI in the study area, P
i is the level LUI in the study area (i = 1, 2, 3, 4, and 5), Q
i is the percentage of the area occupied by the ith level landuse type in the study area, and n is the number of LUI classifications in the study area.
2.3.3. GIS Analysis and Spatial Statistics
Presently, when most scholars research ESV driving factors, they primarily select nations [
22], provinces/states [
24], cities, or particular regions [
18,
19,
20] as their research objects. As current research lacks grids and other small-scale studies, the ESV is spatially affected by different types of land and is diverse within a space. The methodological approach in this study is a primarily based on the grid method to process the ESV and its driving factors in space. Thus, factors are implemented in a smaller space, so the differences between different regions can be studied.
(1) Hot and cold spot analysis
A hot spot is a location or a small area within an identifiable boundary showing the concentration of incidents. This tool works by looking at each piece of data in the data environment of neighboring cells. To be a statistically significant hotspot, not only should the data itself have high values, but there also needs to be a certain number of clusters of high-value data around it. The Getis–Ord Gi* is expressed as follows [
45,
46]:
where x
j represents the observation in unit j, w
ij(d) represents the spatial weights between spatial units i and j, and n represents the number of units.
(2) Geographical detector method
A set of statistical methods used by geographical detectors can explore the consistency in spatial distribution between dependent and explanatory variables through spatial dissimilarity and reveal the driving forces behind the spatial distribution characteristics of the dependent variable [
47,
48]. The value of the q-statistic in the factor detector is used to measure the explanatory power of the explanatory variables for the spatial divergence of the dependent variable, detecting the extent to which a given driver X can be used to explain the spatially divergent characteristics of Y. The degree of spatial association was measured using the q-statistic [
49]:
where q is the determination power of X to the ESV; n is the number of sample units; Nh and N are the variances of the ESV of region h and Dalian, respectively; and σ
h2 and σ
2 are the variances of the ESV of sub-region h and the whole region, respectively. L denotes the total number of regions h.
(3) Geographically weighted regression
Due to the spatial complexity, autocorrelation, and variability in the ESV data, when exploring the influence of explanatory variables on dependent variables, their expression may be different in different regions. Therefore, based on the use of a geo-detector to determine the critical factors of the ESV, this study used GWR to analyze the spatial differentiation of driving factors to explore how each driving factor serves the ESV in different regions. The GWR model was expressed as follows:
where (u
i,v
i) are the spatial coordinates of the sample point i, β
k(u
i,ν
i) is the value of the continuous function β
k(u,ν) at the point i, x
ik is the explanation variable x
k in the location (u
i, v
i) and ε
i is the error term. The GWR model constructed has the highest model accuracy by drawing on the research results in the calculation and, through practice, the fixed Gaussian function is determined as the weight function, the bandwidth is determined by the AIC method, and the GWR4 is used for the regression calculation.