Spatio-Temporal Characteristics and Influencing Factors of Ecological Risk in China’s North–South Transition Zone

Abstract

Ecological risks (ERs) are external threats to ecosystem structure and function, and the evaluation of ER is valuable for understanding the extent of the disturbance of ecosystems by human activities. An objective investigation of the spatio-temporal characteristics of ER and its affecting elements can contribute to the improvement of the ecological environment and the sustainable development of regional resources. The Funiu Mountains were chosen as the study area, the ecological risk index was built using landscape interference and landscape separation, and the spatial–temporal properties and scale effect of the ER were assessed. The findings of ER may be separated into four distinct zones to study the variations in ER levels from 2000 to 2015: modest, mild, moderate, and high. Furthermore, geographically weighted regression (GWR) was used to ascertain the degree to which environmental factors impact ER and regional disparities. The results show: (1) Ecological risk assessment was sensitive to scale transformation, and there are variations in the spatial pattern of ER on diverse scale units. With the scale following a 1 km grid to the watershed, Global Moran’s I values dropped from 0.53 to 0.04, and 1 km × 1 km was the most suitable unit. (2) The ecological and environmental quality of the Funiu Mountains improved from 2000 to 2015. The extent of the high ecological risk zone dropped by 398.07 km², but other types of ecological risk regions rose, owing to the execution of policies aimed at restoring cropland to forests. The ER presented a transitional distribution from a high risk in the surrounding area to low risk in the central area. (3) From the perspective of correlations among the influencing factors, altitude and slope had a negative correlation with other influencing factors, while others were positively correlated. The correlation coefficient between land-use degree and human activities intensity and ER was the highest. Each environmental factor had significant effects on the ERs and spatial variation existed, where the land-use degree and human activities intensity were the main obstacle factors limiting ER improvement. Population density, altitude, slope, and maximum temperature all displayed effects on ER with different directions and intensities.

1. Introduction

In the context of global change, uncertain factors in the environment have gradually increased, coupled with the increasing degree of interference of human activities on the ecosystem, leading to imbalances in the internal structure of the ecosystem, damage to self-repair capabilities, and deterioration of the quality of the ecological environment [1,2,3,4,5,6]. It is urgent to assess the disturbance of stress factors such as environment, society, and economy on the structure and function of the ecosystem, aim to identify ecological risk (ER) sources, regulate ER processes, and then improve ecosystem service efficiency [7,8,9,10]. In recent years, ER has become a hot topic of ecological and geographic research [11,12,13]. The ER concept is that an ecosystem, landscape, or region is subject to external threats such as natural- or human-induced, increasing the possibility of reduced productivity, environmental damage, and decreased ecosystem service function. At present, the “source-sink” method and landscape index method are used to evaluate ER [14,15]. The “source-sink” method focuses on the identification of “source-sink” properties of landscape risks and is widely used in research of non-point source pollution, soil erosion, and urban heat island effect [16]. The “source-sink” method applies to the evaluation of ERs with clear threats at the regional level [17]. Based on land use/cover data, the landscape index method focuses on the ecological environment effects caused by specific landscape assemblages and their spatial heterogeneity. The landscape index emphasizes the interconnectedness of patterns and processes [18,19]. Landscape disturbance index and landscape vulnerability index are the main indicators for constructing ER in landscape ecology; while the landscape disturbance indices consist of a landscape fragmentation index, a landscape separation index, and a landscape subdimensionality index [20,21]. Compared with the “source-sink” risk assessment, the landscape index method assesses ER at the landscape level. This method can identify the local spatial characteristics of ER, scientifically understand the interrelationship between landscape patterns and ecological processes and diagnose multiple ER sources and their cumulative effects in the landscape mosaic, which provides a theoretical reference for promoting the healthy and sustainable development of ecosystems.

Scale characteristics have been one of the main focuses of ecological and geographic research, and ecosystem structure and function tend to change with scale. Assessing the extent of the disturbance of ecosystems from external threats at different scale units is a major tool for developing ER management measures in the region. Currently, studies on the scale of ERs have been deepened, and ER assessment on different geographical units such as watersheds [22,23], mountainous areas [24], cities [25,26] and beaches [27] have been carried out. In recent years, an increasing number of scholars are beginning to assess ER at a fine scale. Zhang et al. [28] analyzed the ER of the high background area of heavy metals in the southwest from the township scale, making up for the shortcomings in precision of previous surveys at the regional scale, and meeting the needs of natural resource management. Zhang et al. [29] evaluated ER in southwest Guangxi-Beibu Bay based on the DPSIR model and OWA-GIS. Previous studies have shown that there are significant differences in the assessment of ER at different scales. If the assessment scale is too large, it is difficult to identify the local characteristics of ER. In contrast, if the assessment scale is too small, the overall characteristics of the region are often ignored. Therefore, the selection of the optimal assessment unit is a prerequisite for the accurate identification of the spatial differentiation characteristics of ERs and a basis for quantifying the impact of natural factors and the regional ecological environment.

The Funiu Mountains are the transition region from the north subtropical to the south warm temperate zone in China, and the area where the north–south divide passes through. It is also an important water-conserving forest area of the Danjiangkou reservoir, with climate regulation, water supply, and ecosystem service function. The ecological quality of the north–south transition region in China plays an important role in conserving biodiversity and enhancing ecosystem service capacity. Since the 21st century, the land-use pattern was profoundly changed by rapid urbanization, resulting in great pressure on the stability of the internal structure in the ecosystem. Given this, we constructed the ecological risk index by landscape disturbance degree and landscape vulnerability degree. The exploratory data analysis and spatial autocorrelation were used to diagnose the difference of ERs at different scales. Furthermore, correlation analysis and geographically weighted regression (GWR) were used to analyze the strength and spatial difference of the impact of various environmental factors on ER in 2015, to provide scientific reference for the ecological environment planning in the study area.

This article aimed to answer three questions: (i) What is the best scale for assessing ER in the Funiu mountain area—watershed, township, or grid? (ii) What changes in the ER and development trends of different grades of ER occurred from 2000 to 2015? (iii) At the scale of a 1 km grid, what are the intensity, and direction of influencing factors of ecological risk? The results of this study can provide relevant references for ecological quality assessment of other geographic transition areas in China and the world.

2. Data and Methodology

2.1. Study Area

Located in the transition zone from the north subtropical to the warm temperate zone in China, the Funiu Mountains are a typical ecological and geographically fragile area with rich biodiversity and important ecosystem services in the region. The vegetation type is the transition type of warm temperate deciduous broadleaf forest and north subtropical evergreen deciduous mixed forest. The temperature range is 13.6–15.1 °C, and the precipitation range is 700–1000 mm. In addition, this area is in the west of Henan Province with a total area of 19,711.75 km². The study area ranged from 110°30′ E to 113°30′ E and from 32°45′ N to 34°20′ N, reaching the border of Henan and Shaanxi in the west, north of Fangcheng city in the east, Xionger Mountain and Waifang Mountain in the north, and Nanyang Basin in the south (Figure 1).

2.2. Data Resources

The land cover/use data (2000–2015) were obtained from the Middle and Lower Yellow River Scientific Data Center, China Earth System Data Sharing Platform (http://www.geodata.cn/) (accessed on 2 May 2019). According to the “China land use/cover remote sensing monitoring data classification system”, the land-use types in each period were reclassified into six types, including construction land, forest, watershed, grassland, cultivated land, and unused land. The digital elevation model (DEM) data came from the Geospatial Data Cloud Platform (http://www.gsclod.cn) (accessed on 8 May 2019), with a resolution of 30 m. The slope was obtained by extracting DEM data. Climate data, population density, and watered zoning data were collected from the Data Center for Resources and Environmental Sciences, Chinese Academy of Sciences (http://www.resdc.cn/) (accessed on 12 May 2019). The land-use degree and human activity intensity were calculated according to the method of Zhuang et al. [30] and Yan et al. [31].

2.3. Methods

In this paper, we constructed the land use risk index using the landscape interference index and landscape vulnerability index in the landscape ecology. The exploratory data analysis was used to evaluate the optimal unit of ER. The GIS software was to analyze the spatio-temporal characteristics. The correlation analysis and geographically weighted regression models (GWR) were applied to estimate the influence degree of different environments on the ER. The overall framework of this paper is shown in Figure 2.

2.3.1. Construction of Land-Use Risk Index Model

(1) Landscape interference index

The landscape disturbance index (E_i) refers to the degree of external disturbance to different landscape ecosystems. The lower the intensity of landscape disturbance, the more stable the ecosystem, the more beneficial it is to the survival of organisms. In this paper, a landscape fragmentation index, a landscape separation index and a landscape subdimensionality index were used to construct the landscape disturbance index and were measured through the method of Wu et al. [32] and Xie et al. [33]:

$${\mathrm{E}}_{\mathrm{i}}=a{\mathrm{C}}_{\mathrm{i}}+b{\mathrm{S}}_{\mathrm{i}}+c{\mathrm{DO}}_{\mathrm{i}}$$

(1)

where C_i was the landscape fragmentation index, which can characterize the degree of ecosystem landscape heterogeneity. S_i was the landscape subdimensionality index used to describe the complexity of the landscape patches. DO_i was the landscape separation index; the larger the value, the more complex the landscape distribution and the greater the degree of external interference. a, b, and c represented the weights of each landscape index, and the sum of the weights was 1. Combined with previous studies, the values of a, b, and c were assigned as 0.502, 0.301, and 0.197, respectively [34,35].

(2) Landscape vulnerability index

Landscape vulnerability (F_i) was the degree of susceptibility in different ecosystems under the stage of the natural succession process of the landscape, which can characterize the integrated effect of human activities and the natural environment on land use. When the ecosystem is in the primary succession stage, the food chain structure is simple and has less ability to resist the risk of external disturbance, while the more complex the ecosystem structure, the greater the ability to resist the risk of external disturbance. According to the research of Zhao et al. [36] and Xie et al. [33], unused land was the most vulnerable, followed by watershed, while construction land was the most stable. Therefore, the unused land, grassland, watershed, cultivated land, forest land, and constructed land were assigned values of 6, 5, 4, 3, 2, and 1 in descending sequence, and the F_i values of each type were normalized to be 0.2857, 0.2381, 0.1905, 0.1429, 0.0952, and 0.0476, respectively [37].

(3) Land use ecological risk index

The land use ecological risk index (ERI_i) was constructed from the landscape disturbance index and landscape vulnerability index, and grid units were created to calculate the ecological risk value [33]. ERI_i was defined as follows:

$$ER{I}_k={\displaystyle \sum _{i=1}^n\frac{A_{ki}}{A_k}}(E_i\times F_i)$$

(2)

where ERI_k was the ecological risk index value of land use on the kth grid. A_k was the sampling area of the kth grid. A_ki was the area of land-use type i on the kth grid. E_i was the disturbance index of land use type I, and F_i was the corresponding vulnerability index.

2.3.2. Analysis of Ecological Risk Scale Characteristics

(1) Different scales of ecological risk assessment

To evaluate the spatial clustering characteristics of ER at different scales (Figure 3), the grid-scale (1 km × 1 km, 1.5 km × 1.5 km, 3 km × 3 km, 5 km × 5 km), township administrative unit scale, county administrative unit scale, and watershed-scale were constructed through resampling and magnitude conversion in Geographic Information Systems (GIS). By analyzing the change rule of ER under scale transformation, the best unit for evaluating ER can be determined.

(2) Exploratory data analysis

The exploratory data analysis was based on spatial statistics. The main object of the research was the spatial interaction and changes to the rules between events or phenomena and geospatial characteristic information, to analyze the randomness and structural characteristics, spatial correlation, dependence, and heterogeneity of variables in a specific region [38]. Spatial autocorrelation is one of the commonly used tools for spatial statistical analysis, which accurately represents the attribute association characteristics of a geographic element on a specific area and adjacent units [39]. In general, both global autocorrelation and local autocorrelation indicate the degree of spatial autocorrelation. The Global Moran’s I statistic and Getis–Ord Gi* index were used to measure the spatial autocorrelation of ERs in different scales with the software of Geoda and GIS. The Global Moran’s I was defined as follows:

$$\mathrm{I}=\frac{{\displaystyle {\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\displaystyle {\sum }_{\mathrm{j}\ne 1}^{\mathrm{n}}{\mathrm{w}}_{\mathrm{ij}}\left({\mathrm{y}}_{\mathrm{i}}{-\mathrm{y}}_{\mathrm{mean}}\right)\left({\mathrm{y}}_{\mathrm{j}}{-\mathrm{y}}_{\mathrm{mean}}\right)}}}{{\mathrm{s}}^2\cdot {\displaystyle {\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\displaystyle {\sum }_{\mathrm{j}\ne 1}^{\mathrm{n}}{\mathrm{w}}_{\mathrm{ij}}}}}$$

(3)

where n was the number of space units, indexed by i and j. Y_i and y_j were the attribute values of unit i and j, respectively. W_ij was the weight matrix. S₂ represented the sample variance. Y_mean was the mean of y. The Global Moran’s I was the spatial autocorrelation value of ER, the value of I ranged from −1 to 1. The Global Moran’s I value greater than zero indicated a spatial positive correlation. The larger the index, the stronger the positive spatial correlation, and an index close to 1 showed a smaller degree of spatial variation. A Moran’s I value less than zero indicated a spatial negative correlation. The lower the index, the stronger the negative spatial correlation, and an index close to −1.0 meant very large spatial differences. A Moran’s I value trending toward zero showed the random spatial distribution of ER. The Getis–Ord Gi* index was calculated as follows [38]:

$$G_{\mathrm{i}}{}^{*}=\frac{{\displaystyle {\sum }_{\mathrm{j}}^{\mathrm{n}}{\mathrm{w}}_{\mathrm{ij}}\left(\mathrm{d}\right){\mathrm{x}}_{\mathrm{i}}}}{{\displaystyle {\sum }_{\mathrm{j}}^{\mathrm{n}}{\mathrm{x}}_{\mathrm{j}}}}$$

(4)

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

(5)

where E(Gi*) denoted the mathematical expectation of Gi*. Var(Gi*) was the variance of Gi*, and W_ij was the spatial weight. The values of Z (Gi*) were classified into four classes: <−1.96, −1.96 to −1.65, −1.65 to 1.65, 1.65 to 1.95, and >1.96, representing the cold spot area, subcold spot area, non-significant change area, subhot spot area, and hot spot area, respectively.

2.4. Correlation Analysis

Correlation analysis was used to measure the correlation between ER and environmental variables and quantify the degree of correlation between variables. If the value was greater than zero, the correlation was positive. Otherwise, it was a negative correlation. The value represents the degree of correlation, and the higher the value, the stronger the correlation [39]. The definition of correlation analysis was shown as follows:

$$R_{xy}=\frac{{\displaystyle \sum _{i=1}^n\left(x_i-x_{mean}\right)\left(y_i-y_{mean}\right)}}{\sqrt{{\displaystyle \sum _{i=1}^n{\left(x_i-x_{mean}\right)}^2}}\sqrt{{\displaystyle \sum _{i=1}^n{\left(y_i-y_{mean}\right)}^2}}}$$

(6)

where R_xy was the correlation coefficient between two variables. N was the number of samples. X_i was the i value of x, and x mean was the average value of x. Also, y_i and y_mean represented y and the average value of y, respectively.

2.5. Geographically Weighted Regression Models

GWR models were improved based on the classical linear regression model (OLS). The spatial scale dependence and heterogeneity of the variables were reduced, and the local regional parameters and their influencing factor were estimated. The non-smoothness of correlation between the variables and the spatial location was found, and the results were more accurate [40,41,42].

$${\mathrm{y}}_{\mathrm{i}}={\beta }_0\left({\mu }_{\mathrm{i}},v_i\right)+{\displaystyle \sum _{\mathrm{k}=1}^{\mathrm{p}}{\beta }_{\mathrm{k}}}\left({\mu }_{\mathrm{i}},v_i\right)x_{ik}+{\epsilon }_{\mathrm{i}}$$

(7)

where y_i was the dependent variable, and x_ik was the independent variable. (μ_i,v_i) was the latitude and longitude of the sampling point i. βk (μ_i,ν_i) represented the value of the continuous function βk (μ₀,ν₀) on the space element of the k sample. ${\epsilon }_{\mathrm{i}}$ was the regression residual of point i. i = 1, 2, ..., n, was the number of positions in space.

3. Results

3.1. Scale Characteristics of Ecological Risk in Funiu Mountain Area

3.1.1. Cluster Analysis of Ecological Risk at Different Scales

As shown in Figure 4, with the evaluation scale gradually pushed down (from 5 km to 1 km), the spatial distribution pattern of ERs became more precise, and the distribution of hot spots and subhot spots become more and more, and the significance level became higher. When the evaluation unit increased from the 5 km grid scale to a watershed scale, the spatial correlation of the ER distribution became weak, and the distribution area of hot spots and subhot spots was insignificant. The results showed that the spatial agglomeration characteristics of ERs were different at various scales, and the larger the evaluation scale, the more obvious the impact of landscape composition and different types of land ecological subsystems on the results of ecological risk assessment. As a result, differences in the composition and structure of the interior landscape were filtered. Therefore, compared with administrative units and watersheds, grid units were more suitable for evaluating the spatial–temporal characteristics and influence mechanisms of ER.

3.1.2. Spatial Autocorrelation Analysis of Ecological Risk at Different Scales

It was clear that the spatial correlation degree of ER varied significantly at different scales (

Table 1. Statistical characteristics of spatial autocorrelation of ecological risk at different scales.

Scale	1 km	1.5 km	3 km	5 km	Township	County	Watershed
Moran’s I	0.53	0.49	0.24	0.24	0.07	0.06	0.04
Z-score	109.02	75.33	41.64	28.51	4.86	1.50	1.30

). The Global Moran’s I value of ER decreased from 0.53 to 0.04 when the assessment scale moved from the 1 km grid unit to the watershed scale. The correlation of ER decreased with the increase of scale, and the larger the assessment scale, the lower the spatial autocorrelation. The significance test showed that the Z-scores of 1 km, 1.5 km, 3 km, 5 km, and township scale were all greater than 1.96, indicating that both grid-scale and township scale passed the significance test, while the Z-scores of county scale and watershed scale were 1.50 and 1.30, respectively, failing the significance test (p > 1.96). The results of this study showed that the spatial characteristics and spatial correlation of the 1 km grid were high. Therefore, the spatial and temporal characteristics of the ER and its influencing factors in the study area were analyzed based on the 1 km grid.

3.2. Spatio-Temporal Differentiation Characteristics of Ecological Risk

As can be seen from Figure 5, the spatial distribution of ER in the study area did not change significantly from 2000 to 2015. The mild ecological risk zone was the most extensive, mainly concentrated in the southwest and central regions. The moderate ecological risk zone and high ecological risk zone were clustered and distributed in the low altitude area, mainly due to the influence of human activities in the low altitude areas, while the distribution of the mild ecological risk zone was small and scattered.

The ER levels in different periods were counted, and the area and proportion were calculated (

Table 2. Ecological risk (ER) distribution area (km²) and proportion (%) in Funiu Mountains from 2000 to 2015.

Ecological Risk Level	2000 Years	2005 Years	2010 Years	2015 Years	2000–2015
	Area/Proportion	Area/Proportion	Area/Proportion	Area/Proportion	Area/Proportion
Sl- ecological risk zone	2151.08 (10.91)	2641.54 (13.40)	2609.47 (13.24)	2295.56 (11.65)	144.48 (0.73)
Mi- ecological risk zone	13,549.08 (68.74)	13,101.06 (66.46)	12,958.77 (65.74)	13,753.82 (69.77)	204.74 (1.04)
Mo- ecological risk zone	2857.20 (14.50)	2351.22 (11.93)	2247.36 (11.40)	2906.53 (14.75)	49.33 (0.25)
Hi- ecological risk zone	1154.39 (5.85)	1617.92 (8.21)	1896.16 (9.62)	755.84 (3.83)	−398.55 (−2.02)
Total area	19,711.75 (100)	19,711.75 (100)	19,711.75 (100)	19,711.75 (100)	-

Note: Sl- ecological risk zone, Mi- ecological risk zone, Mo- ecological risk zone, Hi- ecological risk zone, respectively, Slight, Mild, Moderate, High ecological risk zone.

). In this study, the mild ecological risk zone was the most widely distributed and had the highest proportion in 2015 (69.77%) but the lowest proportion in 2010 (65.74%). The moderate ecological risk zone was the largest (2906.53 km²) in 2015, accounting for 14.75%, and the smallest (2351.22 km²) in 2005, accounting for 11.93%, reflecting the phase change characteristics of moderate ecological risk decreasing first and then increasing. The distribution area of the slight ecological risk zone was the smallest (2151.08 km²) in 2000 and the largest (2641.54 km²) in 2005, which showed the phase characteristics of first increasing and then decreasing. The highest level of ecological risk and ecological stress were found in the high ecological risk zone. In 2010, the high ecological risk zone was the most widely distributed (1896.16 km²), accounting for 9.62%; in 2015, it was the smallest (755.84 km²), accounting for 3.83%, and the overall high ecological risk area showed a downward trend. During the study period, the area of different levels of ER changed significantly, where the area of high ecological risk zone decreased by 398.07 km², and the proportion dropped by 2.02%, indicating that the high-risk level was controlled and the degree of risk stress on the ecosystem was reduced, which also reflected the trend of transferring from high ecological risk level to other ecological risk levels. The area of mild ecological risk increased by 204.74 km², and the area of slight ecological risk and medium ecological risk increased by 144.48 km² and 49.33 km², respectively.

3.3. Analysis of Ecological Risk Influencing Factors

3.3.1. Correlation Analysis

ArcGIS 10.3 was used to extract the value of ER in the 1 km grid in 2015, and the function of multivalue extraction to a point feature was used to extract the impact of human activities intensity, land-use degree, population density, maximum temperature, slope, and altitude. SPSS software was used for correlation analysis to calculate the pair correlation between the impact factors. The correlation among the influencing factors passed the significance test, except for precipitation and ER and maximum temperature and land-use degree, which did not pass the highly significant test (p < 0.01) (

Table 3. Correlation analysis of ecological risk and various impact factors.

Correlation Coefficient	ER	HAI	LUD	PD	HT	Slope	DEM
ER	1
HAI	0.491 **	1
LUD	0.768 **	0.696 **	1
PD	0.095 **	0.339 **	0.094 **	1
HT	0.082 **	0.351 **	0.006	0.390 **	1
Slope	−0.084 **	−0.438 **	−0.0034 **	−0.366 **	−0.483 **	1
DEM	−0.116 **	−0.492 **	−0.032 **	−0.411 **	−0.604 **	0.712 **	1

** Denotes significance at the 99% level of confidence. HAI represents human activity influence; HT represents highest temperature; LUD represents land-use degree; PD represents population density.

). In addition, the correlation coefficients among the factors were both positive and negative, indicating that the directing of action varied greatly among the influencing factors. Slope and altitude were negatively correlated with other influencing factors, and had the strongest influence on maximum temperature, with coefficients of −0.483 and −0.604, respectively. The correlation coefficients of land-use degree with ER and human activity intensity were 0.768 and 0.696, respectively, suggesting that the change in land-use degree significantly affected the ER and human activity intensity. The ER and human activity intensity were stronger for land-use types with higher development degrees. Both altitude and slope were negatively correlated with ER at the 0.01 level, the correlation coefficients were −0.116 and −0.084, respectively.

3.3.2. Model Selection

To select a suitable model for analyzing the influence of each environmental variable on ER, six environmental variables were selected as independent variables, with ER as the dependent variable in the OLS model to measure the influence of each explanatory variable on the explanatory variable and its significance level (

Table 4. OLS model estimation and diagnosis results.

Explanatory Variables	Coefficient	Standard Erro	t-Statistic	Probability	VIF	OLS
Constant term	0.0027	0.0018	1.5	0.133 **	-	R2 0.6444
HAI	−0.059	0.0012	−49.55	0 **	5.9	Adjust R2 0.6442
LUD	0.0004	0.000001	122.27	0 **	5.23	Sum of squares of residuals 2.29
PD	0.0001	0.000001	−0.78	0.435 **	1.42	Sigma 0.011
HT	0.0006	0.00005	10.74	0 **	1.68	Akechi information criterion AICc-122703.401113
Slope	−0.0001	0.000001	−6.27	0 **	2.16
DEM	−0.0001	0.000001	−24.57	0 **	2.92

** Denotes significance at the 99% level of confidence. HAI represents human activity influence; HT represents highest temperature; LUD represents land-use degree; PD represents population density.

). The variance inflation factor (VIF) of each parameter of the regression model was less than 7.5, demonstrating that the equation variables were set reasonably and there was no multicollinearity problem. However, the OLS model only considered the global characteristics of the regression coefficients and did not consider the local effects and spatial spillover of each influencing factor on ER, and there were spatial clustering characteristics of ER. In addition, the whole model had a strong spatial autocorrelation based on the results of Global Moran’s I, so there would be a large error existing in the OLS model which cannot explain the spatial heterogeneity of the risk factors well. Hence, the GWR model was further used to identify the spatial heterogeneity of ER influencing factors, and the parameters of the GWR model were obtained with the OLS model regression results as a reference (

Table 5. Analysis of GWR model results.

Model Parameter	Numerical Value
Bandwidth	809.87
Sum of squares of residuals	0.62
Akechi information criterion	−147,523.1174
Sigma	0.006
R2	0.90
Adjust R2	0.90

). Compared with the OLS model results, the GWR model fit was 0.9038, which was 40.25% better than the OLS model; the sigma and residual sum of squares were statistically significantly lower, with a decrease in the value of the deficit pool information criterion. Therefore, we concluded that the GWR model was significantly better than the OLS model.

The diagnostic results of the OLS model showed that all six explanatory variables passed the significance test at 0.01 level. The influence of land-use degree, population density, and maximum temperature on soil moisture was positively correlated, and the explanation level was maximum temperature, land-use degree, and population density from high to low. Though it remained unchanged under other conditions, the ER would increase by 0.0574% when the maximum temperature with the strongest explanatory ability increased by one unit. The values of the correlation coefficients of human activity intensity, slope, and DEM on the ER were negatively correlated, and the absolute value of the correlation coefficients in descending order were human activity intensity, slope, and DEM. The ER decreased by 5.85% with each increase of human activity intensity, and the ER increased slightly with the decrease of slope and altitude.

3.3.3. Spatial Differences of Impacts of Explanatory Variables on Ecological Risk

(1) Descriptive statistics of regression coefficients of GWR model

Descriptive analysis was conducted on the regression coefficients of each specific spatial unit of GWR, and the coefficient values of explanatory variables were obtained statistically, including minimum value, lower quartile value, median value, upper quartile value, maximum value, and average value (

Table 6. Descriptive statistics of regression coefficient of GWR model.

Independent Variable	Minimum Value	Lower Quartile Value	The Median Value	Upper Quartile Value	M	Average Value
Human activity intensity	−0.192	−0.152	−0.110	−0.0346	0.8282	−0.0551
Land-use degree	−0.002	0.0003	0.0005	0.000634	0.0007	0.0004
Population density	−0.001	−0.0001	−0.0001	0.0001	0.0001	−0.0001
Highest temperature	−0.0089	0.0001	0.0011	0.0038	0.0649	0.0028
Slope	−0.0006	0.0001	−0.0001	0.0001	0.0002	−0.0001
DEM	−0.0002	−0.0001	−0.0001	0	0.0001	−0.0001

). It showed that each explanatory variable was positive or negative with a large difference, indicating that each explanatory variable was not simply positively or negatively correlated with risk factors. There was obvious spatial heterogeneity in the degree of explanation, and the impact on the ER may be opposite in different spatial scopes. However, the direction of the median value and the mean value was consistent, suggesting that the GWR model had the same effect on ER in most spatial ranges, and the difference of its value indicated that each explanatory variable had different effects.

(2) Spatial differences of the impact of explanatory variables on ecological risk

The natural breakpoint method in ArcGIS 10.3 was used to visualize the results of model operation and reflect the influence characteristics of explanatory variables on the ER at the local spatial scale. Figure 6 presents the spatial coefficient distribution of six impact factors generated by the GWR model, indicating that there were clear spatial variations in the degree of influence of the ER impact factors: (1) In terms of human activity intensity, the regression coefficient decreased from central to peripheral, indicating that forest and grassland were the main land-use types in the central part, which were greatly affected by human activity intensity and had a weak ability to resist ER. The northwest and southeast regions were occupied land for urban residents, and the terrain was flat, and the ER decreased when other impact factors remained unchanged, and the human activity intensity increased. (2) In terms of land-use degree, there was a positive correlation between land-use degree and ER, displaying low levels in the middle and high surrounding areas. The high-value areas were mainly concentrated in the northwest, north, and south, mainly in areas with high land development, such as cultivated land and construction land. In addition, the interchange between various land-use types led to alterations in the ecological level. The government departments should implement some measures to protect the ecological land and relieve the external pressure on the ecosystem. (3) In terms of population density, the regression coefficient ranged from −0.61 to −1.05, showing a negative correlation. The low area gathered in an ellipse in the northwest, southwest, northeast, and high-value areas distribution in the northeast and southeast, located in the built-up areas of Lu, Nanzhao, and Zhenping county, the high population density, and flat topography was the main reason. If other factors remain unchanged, the population density increasing would aggravate ER. (4) In terms of elevation, the regression coefficients were negative in most regions, and the absolute values of the regression coefficients increased from the central region to the surrounding areas, indicating that the ER in the central region was less negatively affected by the average elevation than in other regions. (5) In terms of slope, the regression coefficients were concentrated in [−2.08, −1.53], exhibiting negative correlation characteristics, and the high absolute values of the regression coefficients were mainly distributed in the northeast, meaning the ER in the northeast was negatively affected by the average slope to a greater extent. (6) In terms of maximum temperature, the regression coefficients generally showed a trend of increasing from the northwest to the southeast, and a positive correlation was a dominant relationship. The ER in the southeast was more likely to be positively affected by raised maximum temperatures and was significantly affected by monsoonal climate, especially extreme climate changes that increased the threats to ecosystems. In addition, the flat topography of the southeast was less influenced by topographic factors than the northwest, which was a potential cause of increased ER from warmer temperatures in the east.

4. Discussion

4.1. Significance to Management

The Funiu Mountains, situated in the transition area between the north and the south of China, are one of the richest areas in terms of biodiversity conservation and the most sensitive ecological environment in China. We calculated the landscape pattern index by constructing a 1 km grid to evaluate the degree of ER. It found that the ER showed a trend of increasing and then decreasing, with the highest ER value in 2010 and the lowest ER value in 2015. The rapid urbanization in mountainous areas has changed the way and extent of land use, which has a significant impact on ER in the region. In addition, the ecological risk index showed a decreasing trend after 2010, and the ecological environment quality improved. The main reason being that the implementation of the policy of returning farmland to forest and the environmental protection concept of “Green water and green mountains are golden mountains” proposed by Xi Jinping has increased the protection of ecological land and curbed the trend of converting ecological land such as forest land to construction land. In the future, the implementation of the policy of returning farmland to the forest as well as ecological conservation should be implemented in the Funiu Mountains.

From 2000 to 2015, the ER showed an overall spatial characteristic of high and low in the middle, which is mainly because the topography of the Vernon Hills is high in the northwest and low in the southeast. The surrounding terrain is relatively flat, the degree of human activity influence is high. The construction land and cultivated land are widely distributed in low altitude areas, which are subject to a greater degree of human disturbance. The high degree of land fragmentation is an important factor contributing to the formation of the spatial distribution pattern of ER. Hence, government departments should pay attention to the protection of ecological land in the formulation of socio-economic development plans, strengthen the protection of national and regional ecological policies, plan ecological environmental protection well, and balance the relationship between socio-economic development and ecological environmental protection.

In the context of global changes, managers should propose policies that promote sustainable land use, focus on ecological land-use change in sensitive areas, implement red lines for farmland protection and ecological function zones, build buffer zones for the conversion of ecological land and construction land, set biodiversity protection zones, and implement the strictest land control system to promote the sustainable use of land resources in the region.

4.2. Scale Characteristics of Ecological Risk in Funiu Mountains

The selection of suitable risk evaluation units is the precondition for the accurate assessment of ER in the region, and how to determine ER evaluation units on different geographical units is a hot topic of academic interest [43,44]. In this study, the response characteristics of ER to scale were analyzed at grid, township, county, and watershed levels. The results showed that the hotspot areas and the Global Moran’s I values of ER increased as the scale of the 5 km grid was pushed down to 1 km, and both the distribution of hotspot areas and the Global Moran’s I values decreased as the scale was pushed from the grid to watershed. Therefore, the 1 km grid scale was determined to be the best unit for assessing ER, which was consistent with the research on the evaluation of ecosystem service values in the Dabie mountains [45]. In addition, the ecosystem is a very complex giant system, while this study only assessed the ER from the grid unit, which was not sufficient for the formulation of an ecological function zoning scheme and policy implementation. In future research, a region-wide assessment of ecoenvironmental quality and a quantitative analysis of capabilities and trends in ecosystem services, coupled with “water-soil-gas-biology-human” and other elements will be used to measure the external stress factors of ecosystems in a multilevel, comprehensive way.

4.3. Spatial and Temporal Evolution of Ecological Risks and Analysis of Influencing Factors

The landscape index method was used to assess the spatial and temporal characteristics of ER in the Funiu Mountains. The results displayed that the ecological environment quality improved from 2000 to 2015, and the area of high ER decreased the most, while the area of slight ER increased the most, which reflected the mutual conversion and development trend of different ecological risk levels in the region. Overall, due to the implementation of ecological and environmental protection policies such as returning farmland to forest, rural residential renovation, and mining land restoration, the ecological and environmental index of farmers showed a rising trend. From the viewpoint of space, the high ecological risk zone and moderate ecological risk zone were mainly located in the southeast and at the edge of the area, mainly because of the clustered distribution of construction and the strong degree of interference by human activities intensity. Hence, it required further strengthening the regulation and prevention of high-level ecological risk areas, controlling the disorderly expansion of construction land and strictly adhering to the ecological red line.

Topographical and geomorphological factors, construction of the settlement, policymaking, construction of protected areas, and the degree of anthropogenic disturbance can all have important effects on the strength of ER [46]. Ran et al. [47] discovered that land-use change is an important cause of ER. Wang et al. [48] constructed an ecological risk evaluation index system for Dongting lake, incorporating population density and gross regional product into the index system, based on the “pressure-response” model. In this paper, the aim was quantitatively exploring the degree of influence of external stress factors on ER and spatial differences, and indicators were selected from both natural factors and human disturbances. The natural factors were chosen as elevation, slope, and maximum temperature. Human activity intensity, land-use degree, and population density were selected as human disturbance factors. The topography was the key element in determining the quality of the mountain’s ecological environment. As a result, the elevation and slope were chosen as the representative factors. Existing research demonstrated that the maximum temperature has a significantly higher intensity of impact on ER than other climatic elements [45], and multiple covariances were found in this study when testing for covariance among environmental variables. Therefore, the maximum temperature was considered as a typical climatic element in this paper. Human activity intensity, land-use degree, and population density all indicated the degree of impact of anthropogenic disturbances on ecosystem structure and function, thus identifying the direction and degree of impact of anthropogenic activities on ER under different forms. The correlation analysis and GWR model were used to measure the intensity and spatial variation of each environmental variable on the ER, respectively. We found that the topographic and geomorphic factors had a strong influence on ER, which was consistent with the elevation, slope, and lithology being negatively correlated with the ER of landslide hazards in assessing ER in southwestern mountains [45]. Ni et al. [49] discovered that the ecological level was lower in areas with a high intensity of human engineering impacts and high population density, based on the assessment of geological risks in the Changbai mountains. The results of this paper indicate that there was spatial heterogeneity in human activities intensity, land-use degree, and population density with ER, which were manifested in differences in intensity and direction. In some areas, active protection measures may be taken by human beings so that the local ER will be reduced. ER is a joint natural-social-human influence on the ecosystem, and the selection of indicators is of great significance for indepth analysis of the formation mechanism of ER. Vegetation Normalized Difference Vegetation Index, extreme climatic environment (solar radiation, droughts, and floods), soil physical and chemical properties, soil erosion, and socio-economic factors (GDP, pesticides, mechanization level, urbanization level) should be considered to further research ER in the future. The application of high-resolution remote sensing data could improve the accurate analysis of temporal and spatial characteristics of ecological risk and provide the theoretical basis and empirical analysis for improving ecological environment quality.

5. Conclusions

This paper used spatial autocorrelation to identify the spatial aggregation characteristics of ecological risks at the grid, township, county, and watershed scales. The results showed that ecological risk assessment is scale-dependent, and there are differences in the spatial differentiation of ecological risk on different scale units. From the 1 km grid scale to the watershed scale, the correlation of the ecological risk spatial distribution grows weaker and weaker, the hot and cold areas gradually disappear, the Global Moran’s I value also drops from 0.53 to 0.04, and the Z-score is reduced from 109.02 to 1.30. The significance level also shows a downward trend, and it is further proved that the 1 km grid is the most suitable unit for evaluating the ecological risk. Therefore, we applied the landscape index model to calculate the ecological risk in the Funiu mountain area on the 1 km grid from 2000 to 2015. The results indicated that the ecological environment quality showed a good trend, mild ecological risk zone is the main ecological risk level, accounting for more than 60%. During the study period, the area of high ecological risk decreased by 398.07 km², and the proportion decreased by 2.02%, while the area of slight ecological risk increased the most (204.74 km²), the area of mild ecological risk area increased the most (144.48 km²), and the area of moderate ecological risk area increased the least (49.33 km²). The implementation of the policy measures to return farmland to the forest in mountainous areas is the main reason for the improvement of the ecological environment quality. In 2013, the introduction of the policy that green mountains are golden mountains increased the protection of ecological land and improved the ability of the ecosystem to resist external disturbances. Moreover, we took ER in 2015 as the subject of our investigation and analyzed the direction and degree of influence of different factors on ER. In terms of the correlation among various influencing factors, elevation and slope are negatively correlated with other influencing factors. The correlation coefficient between land-use degree and human activity intensity and ER is the highest, being 0.768 and 0.696, respectively. Human activity intensity, land-use degree, population density, altitude, slope, and the highest temperature had a significant influence on the ER, and there is space differentiation. Land-use degree and human activity intensity were the main obstacles ER to improvement; while population density, altitude, slope, and the highest temperature of ER revealed that there is an influence of different direction and strength.

In this study, we analyzed the spatio-temporal characteristics of ER and driving mechanisms based on multisource data, such as land use/cover data from 2000 to 2015, population density, elevation, slope, and maximum temperature, etc. The timeliness of the data may incur certain errors in the assessment results. Hence, the latest and higher resolution data should be collected to calculate ER, including land use/cover data, population density, and climate data in 2020. The accuracy of the data will be improved and provide suggestions for governmental departments to develop ecological environment planning.

In addition, this paper constructed ER through a limited landscape disturbance index and landscape vulnerability index, which may have some error influence on the results, and more indicators need to be selected to comprehensively evaluate ecological risk. Lastly, there is an urgent need to assess the impact of natural environment-socio-economic-human activities on ecological risk and to identify the driving factors of ecological risk.