Different Causal Factors Occur between Land Use/Cover and Vegetation Classification Systems but Not between Vegetation Classification Levels in the Highly Disturbed Jing-Jin-Ji Region of China

Land use/cover and vegetation patterns are influenced by many ecological factors. However, the effect of various factors on different classification systems and different levels of the same system is unclear. We conducted a redundancy analysis with 10 landscape metrics and ecological factors in four periods (1986–2005/2007, 1991–2005/2007, 1996–2005/2007, 2001–2005/2007) to explore their effects on the land use/cover system, vegetation group and vegetation type, and formation and subformation levels of the vegetation classification system in the Jing-Jin-Ji region. Soil, temperature and precipitation from 1986–2005, 1991–2005, and 2001–2005 were the important causal factors, and anthropogenic disturbance and atmospheric factors in 1996–2005 were causal factors at the land use/cover level. The total explained variance from 1996–2005 and 2001–2005 was higher than that from 1986–2005 and 1991–2005 at the land use/cover level. Causal factors and the variance explained by causal factors at the vegetation group, vegetation type, and formation and subformation levels were similar but different in the land use/cover system. Geography, soil and anthropogenic disturbance were the most important causal factors at the three vegetation levels, and the total explained variance from 2001–2007 was higher than that from 1986–2007, 1991–2007, and 1996–2007 at the three vegetation levels. In environmental research, natural resource management and urban or rural planning, geographic factors should be considered at the vegetation group, vegetation type and formation and subformation levels while atmospheric and temperature factors should be considered at the land use/cover level.


Introduction
Vegetation is an important component of ecosystems that can provide various service functions for human beings. Vegetation classification mainly focuses on the vegetation itself; types of vegetation in different classification levels must be taken into account in environmental research, natural resource management and urban or rural planning [1]. Closely related to vegetation, land use/cover is a complex formed by the interaction of natural and artificial elements on the Earth's surface, including various types of natural covers (such as soil and natural vegetation) and artificial covers (such as buildings and roads). Therefore, vegetation classification is much different from land use/cover system, and one type in land use/cover includes several types in the vegetation classification, such as woodland in the classification of land use/cover, which includes needleleaf forest, broadleaf forest, and highly concentrate landscape pattern information and reflect some structural composition and spatial configuration characteristics. These metrics at different levels often serve as indicators to quantify vegetation distributions [28,31,32]. For a long time, they have been used in similar studies to objectively describe the patterns of vegetation and land use/cover [4,31]. They are usually used in combination with ordination analysis and related statistical models to investigate these relationships, which is essential for the next or future steps [19,28]. However, most of the associated studies are only performed at one level, such as land use/land cover [14,19] and certain vegetation types [4,32], using one or a few metrics. Because a few metrics can only reflect the vegetation pattern in limited aspects, many more metrics that can cover the overall picture of vegetation at the different levels are necessary. Furthermore, with economic development, understanding the vegetation pattern and its causal factors in highly disturbed regions is of cardinal significance.
The Beijing-Tianjin-Hebei region, also called the Jing-Jin-Ji region, is an important political, cultural, and economic region in China. However, due to rapid economic development since the reform and opening up of China, the environment in the Jing-Jin-Ji region has been heavily disturbed. In addition, the development of this region is extremely uneven. Large cities, including Beijing and Tianjin, are very developed, while other cities are not. Beijing presents the most developed economy, the best social security and the most complete infrastructure; however, it also has the most crowded population, the least per capita possession of natural resources and the worst environmental bearing capacity [33]. To achieve sustainable development in the Jing-Jin-Ji region, the Chinese government led major afforestation projects over the last years [34]. It is necessary to break up the administrative divisions and understand the relationships between vegetation as well as land use/cover and their causal factors from the perspective of the entire region to provide support and insights for further planning and management of natural resources, such as afforestation [34,35]. Moreover, as a world-level metropolitan area, both vegetation and land use/cover have been heavily affected by human disturbance in the Jing-Jin-Ji region; thus, understanding the causal factors for vegetation and land use/cover will provide a reference to balance socioeconomic development and environmental sustainability worldwide.
In this study, we performed a redundancy analysis to determine the causal factors in four time intervals for the pattern of land use/cover, which is often used in land resource management, and vegetation patterns in different class levels, which are often used in vegetation and biodiversity management. We hypothesized that short-term factors have greater impacts on land use/cover pattern than on vegetation pattern. The objective was to (1) investigate the correlations between ecological factors in four time intervals and land use/cover and vegetation metrics on three levels; (2) determine the main causal factors for the change in vegetation and land use/cover; and (3) determine whether short-or long-term average ecological factors have a greater effect on land use/cover and vegetation patterns. The findings would be useful for achieving sustainable development by eliminating the effect of human disturbance on the environment.

Study Area
The Jing-Jin-Ji region extends from 113 • 04 to 119 • 53 E and 36 • 01 to 42 • 37 N and is located in the northern area of the North China Plain, with the Yanshan Mountains to the north, the North China Plain to the south, the Taihang Mountains to the west and Bohai Bay to the east. The Yanshan-Taihang mountain system in the northwest transforms gradually into a plain in the southeast, which shows the characteristics of higher elevation in the northwest and lower elevation in the southeast (Figure 1). It has a temperate monsoon climate, with annual precipitation ranging from 305 to 711 mm and mean annual temperature ranging from −3 to 14 • C (climate data were obtained from WorldClim, http://www.worldclim.org/, accessed on 13 October 2019). The precipitation gradually increases from northwest to southeast, while the temperature gradually declines from northwest to southeast. The southeast is a coastal alluvial plain dominated by agricultural and urban ecosystems, while forests are mainly distributed in the Yanshan and Taihang Mountains in the northwest. The region covers an area of approximately 216,000 km 2 and has a population of approximately 110 million, and it includes two provincial cities, Beijing and Tianjin, and 11 prefecture-level cities in Hebei Province [35].
gradually into a plain in the southeast, which shows the characteristics of higher elevation in the northwest and lower elevation in the southeast (Figure 1). It has a temperate monsoon climate, with annual precipitation ranging from 305 to 711 mm and mean annual temperature ranging from −3 to 14 °C (climate data were obtained from WorldClim, http://www.worldclim.org/, accessed on 13 October 2019). The precipitation gradually increases from northwest to southeast, while the temperature gradually declines from northwest to southeast. The southeast is a coastal alluvial plain dominated by agricultural and urban ecosystems, while forests are mainly distributed in the Yanshan and Taihang Mountains in the northwest. The region covers an area of approximately 216,000 km 2 and has a population of approximately 110 million, and it includes two provincial cities, Beijing and Tianjin, and 11 prefecture-level cities in Hebei Province [35].

Land Use/Cover and Vegetation Category Data and Landscape Metrics
Land use/cover data for the study area in 2005 were obtained from the Resource and Environment Data Cloud Platform (http://www.resdc.cn/Default.aspx, accessed on 13 October 2019), and the vegetation distribution data in 2007 were obtained from the Vegetation Map of the People's Republic of China [1]. Three classification levels of the vegetation distribution data from coarse to fine are vegetation groups (such as needleleaf forest), vegetation types (such as cold-temperate and temperate mountains needleleaf forest), and formations and subformations (such as Larix principis-rupprechtii forest). This region contained six land use/land covers, nine vegetation groups, 16 vegetation types and 58 formations and subformations (Table A1). Considering the suitable spatial scale and amount of sample based on the actual situation, land use/cover and every vegetation level were  [36]. These metrics included four aggregation characteristics (connectance index, patch density, landscape shape index, and mean Euclidean nearest neighbor distance), three area-edge characteristics (largest patch index, mean patch area, and edge density), two diversity characteristics (patch richness and Shannon's diversity index), and one shape characteristic (mean shape index) [31].

Geographic, Soil, Precipitation, Temperature, Atmospheric, and Anthropogenic Disturbance Data
The ecological factors used in this study involved geographic, soil, precipitation, temperature, atmospheric, and anthropogenic disturbance factors (Table A2). Geographic variables were obtained from Zhao et al. (2018) [37]. The soil data were provided by the China soil map-based harmonized world soil database (HWSD). All attribute data, including aspect, texture class, available water storage capacity, drainage class and soil depth, were changed into numeric data, e.g., 1, 2, 3, etc.
Population, gross domestic product, gross domestic product from primary industry, gross domestic product from secondary industry, and gross domestic product from tertiary industry . The maximum temperature, minimum temperature, and precipitation for every month in 1986-2007 were provided by WorldClim (https://www. worldclim.org/, accessed on 13 October 2019). Other variables associated with precipitation and temperature in 1986-2007 were calculated from the maximum temperature, minimum temperature, and precipitation in every month.

Suitable Spatial Scale and Redundancy Analysis in Four Periods
We calculated the mean values of ecological variables, including precipitation, temperature, and atmospheric and anthropogenic disturbance factors, at different times. To reveal variables in how a long period of time can affect the pattern of vegetation and land use/cover, the variables in four time intervals were used for land use/cover analysis, i.e.  (Table A2).
All variables were natural logarithmically transformed for a normal distribution [2,19]. Landscape metrics for land use/cover and vegetation classification systems served as response variables, and ecological factors served as explanatory variables. Detrended correspondence analysis (DCA) was performed for landscape metrics to detect the length of the species gradient. Because the length of the species gradient was less than 3, a redundancy analysis (RDA) was performed using CANOCO 4.5 to detect the correlation between species variables and environmental variables. As a common and effective method, RDA is a constrained ordination technique that combines regression analysis and principal component analysis, which can find the best explanatory variables showing the changes of species variables based on the existing environmental variables [2,23]. The suitable spatial scale in land use/cover and each vegetation level were selected according to the explained variability and the sample number at 15 spatial scales [2,19]. Ecological factors with variance inflation factors greater than 10 were removed to avoid correlations among all ecological variables [38]. The significance of the canonical axes was assessed by the Monte Carlo permutation test, and the relative importance of environmental variables and variance explained by them were determined by forward selection [23].

Suitable Spatial Scale
The number of samples ranged from 107 to 2150 at all scales in the land use/cover system and three levels of the vegetation classification system. Landscape metrics were calculated at a series area extent to determine the optional spatial scale in the four classifications. The 15 × 15 km distance was found to be the suitable optional extent, with more than 40% of the explained variability and of the total sample number in land use/cover. The 12.5 × 12.5 km distance was found to be the suitable optional extent in three vegetation classifications with more than 60% of the explained variability and of the total sample number in three vegetation classifications ( Figures A1-A4).

Causal Factors for Land Use/Cover Patterns and Correlations between Landscape Metrics and Environmental Factors
The sums of all canonical eigenvalues were 0. 17 (Table 1). The largest patch index, Shannon's diversity index, and patch richness were highly correlated with drainage class. The connectance index and mean Euclidean nearest neighbor distance were highly negatively correlated with soil depth. Patch density, connection index, and mean patch area were highly correlated with annual precipitation

Causal Factors for Vegetation Group Patterns and Correlations between Landscape Metrics and Environmental Factors
The sums of all canonical eigenvalues were 0.49, 0.50, 0.50, and 0.56 for 1986-200

Causal Factors for Vegetation Group Patterns and Correlations between Landscape Metrics and Environmental Factors
The sums of all canonical eigenvalues were 0.49, 0.50, 0.50, and 0.56 for 1986-2007, 1991-2007, 1996-2007 Table 2). The largest patch index and connectivity index were highly correlated with slope and soil depth in the four periods. The correlations of drainage class to landscape metrics were similar in the four periods ( Figure 3).

Causal Factors for Vegetation Type Patterns and Correlations between Landscape Metric and Environmental Factors
Similar to vegetation group, the sums of all canonical eigenvalues were 0.49, 0.50, and 0.56 in 1986-2007, 1991-2007, 1996-2007 (Table 3 Generally, the correlations between landscape metrics and environmental fac were similar in 1986-2007, 1991-2007, and 1996-2007; the largest patch index

Causal Factors for Vegetation Type Patterns and Correlations between Landscape Metrics and Environmental Factors
Similar to vegetation group, the sums of all canonical eigenvalues were 0.49, 0.50, 0.50, and 0.56 in 1986-2007, 1991-2007, 1996-2007, and 2001-2007, respectively. The causal factors in each of the four time intervals were the same at the vegetation group level, and only the explained variance of each factor was different. The total explained variance in 2001-2007 was also higher than that in 1986-2007, 1991-2007, and 1996-2007 (Table 3). Generally, the correlations between landscape metrics and environmental factors were similar in 1986-2007, 1991-2007, and 1996-2007; the largest patch index and connectivity index were highly correlated with slope at the four time intervals; and the correlations of gross domestic product from primary industry for 2001-2007 were similar to those for 1986-2007 ( Figure 4).   Table 4).

Causal Factors for Formation and Subformation Patterns and Correlations between Landscape Metrics and Environmental Factors
The correlation between gross domestic product from primary industry and drai class in 1986-2007 was similar to that in 1986-2007 at the vegetation type level;

Factors Affecting Landscape Patterns in Different Classification
The combination of various ecological factors contributes to the distributio natural vegetation communities and then changes the landscape [24]. Geography usu serves as an important impact factor in vegetation and is used to predict the vegeta distribution [10,39,40]. In this study, geography was an important impact factor fo vegetation pattern at the three vegetation classification levels during four periods (1 2007, 1991-2007, 1996-2007, 2001-2007), and geography could much better explain pattern at the three vegetation classifications than land use/cover during four per (1986-2005, 1991-2005, 1996-2005, 2001-2005) (Tables 1-4). Many researchers have sh

Factors Affecting Landscape Patterns in Different Classification
The combination of various ecological factors contributes to the distribution of natural vegetation communities and then changes the landscape [24]. Geography usually serves as an important impact factor in vegetation and is used to predict the vegetation distribution [10,39,40]. In this study, geography was an important impact factor for the vegetation pattern at the three vegetation classification levels during four periods (1986-2007, 1991-2007, 1996-2007, 2001-2007), and geography could much better explain the pattern at the three vegetation classifications than land use/cover during four periods (1986-2005, 1991-2005, 1996-2005, 2001-2005) (Tables 1-4). Many researchers have shown that the spatial heterogeneity of soil characteristics, such as soil moisture, soil organic carbon, total nitrogen, electrical conductivity, pH, soil depth, soil drainage, and C:N ratios, is highly related to vegetation [8,29,41,42]. In this study, drainage class and soil depth were found to have a significant impact on vegetation patterns at the three vegetation classification levels at four time intervals (1986-2007, 1991-2007, 1996-2007, 2001-2007). The drainage class of soil also had a significant impact on landscape pattern at the land use/cover level during four periods (1986-2005, 1991-2005, 1996-2005, 2001-2005), and the soil depth only had a significant impact on landscape pattern at the land use/cover level during three periods (1986-2005, 1991-2005, 2001-2005) (Tables 1-4). Our results are partly consistent with the studies of Franz et al. (2011) [41] and   [8]. The slope, drainage class, and soil depth, as the top three impact factors in vegetation classification, had a relatively stable influence on vegetation patterns at different times, which may be related to the large area of farmland in the Jing-Jin-Ji region. Although many studies have highlighted the importance of human disturbance, such as land use, the development of agriculture and animal husbandry, to vegetation [8,24,42,43], some studies have shown that within an agro-pastoral ecotone in semiarid regions, the impact of human activities on landscape metrics is weak [19]. In this study, anthropogenic disturbance had a significant effect on the three vegetation levels during four periods (1986-2007, 1991-2007, 1996-2007, 2001-2007). Anthropogenic disturbance had a significant effect on the landscape pattern of land use/cover only in 1996-2007 (Table 1-4). Many studies have shown that climate factors, including temperature, precipitation, and atmospheric factors, are highly related to vegetation [9,19,24]. It is emphasized that temperature and precipitation, especially in spring, are the trigger factors for vegetation growth and phenology [16]. Temperature was closely related to the landscape pattern of land use/cover except for 1996-2005 in this research. Contrary to geography, temperature can much better explain the pattern at the land use/cover level than at the three vegetation classification levels (Tables 1-4). This may be related to the effect of urban heat islands with rapid urban development. Precipitation had a significant effect on land use/cover at the four time intervals in this study and a significant effect on the three vegetation levels in 2001-2007 (Tables 1-4). Some studies have also shown the importance of clouds, light, and wind [9,24]. However, atmospheric factors did not significantly affect the vegetation patterns of the three vegetation levels in this study, and wind only significantly affected land use/cover in 1996-2005 (Tables 1-4).
The characteristics of the same vegetation type living with several identical ecological factors differed in different spaces and different periods [44]. In this study, the causal factors, especially environmental factors, were the same for different vegetation classifications and different periods but different for land use/cover in different periods, and the causal factors were different for land use/cover and vegetation classifications. Classification levels should be considered based on the needs of researchers and policy makers when environmental research, natural resource management, and urban or rural plan making are conducted.
Vegetation, especially vegetation net primary productivity in a specific year, can be affected by ecological factors within the same year and over short or long times up to a year. Many researches generally analyzed directly the changes in vegetation or land-use/cover over different periods; however, the influence of factors in different periods on vegetation or land use/cover are relatively rare [12,13,45]. In this study, the environmental factors affecting vegetation patterns at the three levels in 2007 were the same at different times when the vegetation map was published, indicating that shorter or longer time averages of environmental factors had similar effects on vegetation patterns. The reason might be that the three environmental factors slope, drainage class, and soil depth changed less during the last 20 years. However, the causal anthropogenic factors were different during the four periods, which may indicate a phenomenon for vegetation pattern changes in highly disturbed regions due to socio-economic development. Short-term factors had greater impacts on land use/cover and vegetation patterns than long-term factors in this study. These results were partly different from our hypothesis, i.e., short-term factors have greater impacts on land use/cover pattern than on vegetation pattern. The possible reasons might be the dominant vegetation and land use/cover are farmlands, which tend to be influenced by short-term factors.

Different Classification
Various plant communities are integrated into a certain hierarchical classification level according to their inherent natural characteristics, which makes the similarities in one type of vegetation and the differences in different types of vegetation significant [1]. The functional and structural properties of the types of vegetation at different classification levels are different, and these types of vegetation at different levels may respond differently to the same ecological variables. Compared with vegetation classification systems, land is divided into several different types according to the form and purpose of land use. Therefore, the land use/cover system is a system designed for the planning of land use and is more closely related to human activities.
Many vegetation classification levels, such as plant function types [46], Holdridge life zones [47], International Geosphere-Biosphere Program (IGBP) systems [48], and types of vegetation from the vegetation map of the People's Republic of China [10], were used to simulate vegetation distribution according to their environment. In this study, we also selected types of vegetation from the vegetation map of the People's Republic of China, i.e., vegetation groups reflecting similar appearances of communities, vegetation types reflecting similar appearances of communities and climates, and formations and subformations reflecting similar dominant species, as vegetation classification levels [1]. The explained variances in the levels of vegetation group, vegetation type, and formation and subformation were similar. The total variances explained by four similar variables, including slope, drainage class, soil depth, and gross domestic product per capita or gross domestic product from primary industry or Defense Meteorological Satellite Program/Operational Linescan System night light, reached 45% in the three vegetation classification levels in 1986-2007, 1991-2007, and 1996-2007 and even 53% in 2001-2007 (Tables 2-4). However, the total variances explained by the top four factors were lower than 22% in the land use/cover in the four periods. Temperature and precipitation factors played an important role in the land use/cover system but did not significantly affect the landscape pattern in most vegetation classification systems, indicating that different factors should be used in different classification systems when terrestrial carbon cycles are simulated and predicted under global warming.

Other Possible Factors Not Considered in This Research
Vegetation patterns are not only caused by geography, climate, soil, and human disturbance but are also influenced by other factors, such as crops planted in particular fields, duration of cultivation, government policy, and historical disturbance from humans, animals, and nature. However, complexity, quantitative limitations and the lack of historical information make it difficult to analyze the impact of these factors on vegetation [8]. Precipitation is a commonly used ecological factor in many studies, but in fact, altering rainfall time and interval with no change in total rainfall quantity led to changes in vegetation and other environmental factors [44]. Carbon dioxide fertilization, nitrogen deposition and even extreme weather events, including abnormal temperature and precipitation, can affect vegetation or land use patterns [12]. These climate factors may need to be considered in future research. Some soil characteristics, such as electrical conductivity, pH, very fine sand content, CACO 3 , soil saturation, soil organic carbon, soil organic nitrogen, and C:N ratios, have been shown to be closely related to vegetation [17,27,49]. Among these soil factors, the same or similar factors, such as calcium carbonate, pH, and electrical conductivity, were adopted in this study, although they were not closely related to the vegetation pattern. However, this does not mean that they do not significantly affect vegetation, because these soil factors were removed due to their high correlation with other ecological factors (variance inflation factors greater than 10). Many ecological factors cannot be considered in isolation because of their strong interactions with one another [12].
In addition, both landscape metrics and vegetation-environment relationships are sensitive to the scale of sampling, i.e., the extent of observations in the study [17,50,51]. Scale-dependent relationships among many environmental variables, including biological and biophysical variables, have been demonstrated [28,30]. Understanding the scaledependent relationship and considering scale is important when interpreting vegetation patterns. Finally, errors, such as algorithm errors of data preprocessing and survey errors of vegetation categories, are observed when generating vegetation categories and ecological factor data.

Conclusions
The causal factors for the land use/cover and vegetation classification systems were different, and the variance explained by causal factors was much higher in the vegetation classification system than the land use/cover system. Causal factors were different at different times for the land use/cover system, and soil, temperature, and precipitation were the top three causal factors in 1986-2005, 1991-2005, and 2001-2005; in addition to precipitation and soil, the systems were also affected by anthropogenic disturbance and atmospheric factors in 1996-2005. For the three vegetation classifications, slope, drainage class, and soil depth were the top three impact factors in 1986-2007, 1991-2007, 1996-2007, and 2001-2007; in addition to the top three factors, anthropogenic disturbance was also a causal factor. Short-term factors had greater impacts on land use/cover and vegetation patterns than long-term factors. Different ecological factors should be considered in different classification systems and different levels in vegetation classification systems when natural resource management and urban or rural planning are conducted.   Gross domestic product Gross domestic product from primary industry Gross domestic product from secondary industry Gross domestic product from tertiary industry Gross domestic product per capita Gross domestic product per unit area Defense meteorological satellite program/operational linescan system night light Figure A1. The number of samples and explained variance at all scales in land use/cover in four periods. Figure A1. The number of samples and explained variance at all scales in land use/cover in four periods.