Correlation between vegetation and environment at different levels in an arid, mountainous region of China

Abstract Vegetation patterns and spatial organization are influenced by the changing environmental conditions and human activities. However, the effect of environment on vegetation at different vegetation classification levels has been unclear. We conducted an analysis to explore the relationship between environment and vegetation in the land use/land cover (LULC), vegetation group, vegetation type, and formation and subformation levels using redundancy analysis with seven landscape metrics and 33 environmental factors in the upper reaches of the Heihe River basin in an arid area of China to clarify this uncertainty. Atmospheric counter radiation was the most important factor at the four levels. The effect of soil was the second determinant factor at three levels (except in vegetation formation and subformation level). The number of variables whose relationship to vegetation reached significant levels varied from 26 to 28, and 20 variables were the same at all four levels. The factors affecting vegetation were basically the same at vegetation group level and vegetation‐type level. It was sufficient to analyze the relationship between environmental and vegetation patterns only in LULC, vegetation group and vegetation formation and subformation level in mountainous regions; different factors should be considered at different vegetation levels.


| INTRODUCTION
The appropriate interpretation of the relationship between vegetation types and environment (climate) is one of the main tasks of plant ecologists (Manley, 1961;Motzkin, Wilson, Foster, & Allen, 1999).
Understanding the vegetation-environment relationship is essential, especially for improving knowledge of the effect of global change on ecosystems and the feedback of ecosystem to climate (Arneth, 2015;Austin, 2013;Huete, 2016). Understanding on responses of vegetation to climate change could improve predictions of the future consequences of climate change on ecosystems, biodiversity, and our own food security and welfare (Huete, 2016). The vegetation-environment relationship also can quantitatively disclose the interactions between driving factors and environmental processes and patterns, and thus help identify the main factors leading to environment changes (Sohoulande Djebou, Singh, & Frauenfeld, 2015).
Other research focused on fractal dimensions and their relationship with environmental factors that varied between plant and landscape with a focus on phytoecology (Burrough, 1981). The problem with this method is that fractal dimensions can just indicate one aspect of vegetation patterns. There is, therefore, a big gap in the systematic description of the vegetation-environment relationship (Burrough, 1981).
Vegetation is influenced by various ecological factors (e.g., precipitation, temperature, light, soil and site conditions) and human disturbance, such as cultivation activities, road traffic, and urban land use.
Therefore, investigating vegetation response to environment at multiscales instead of only at the LULC scale is necessary to meet different needs in resource management .
The analysis of some studies on the relation of soil and vegetation, precipitation and vegetation, elevation and vegetation, and environment and vegetation was based on a few selected factors, such as choosing mean annual precipitation to represent the factor "water" that may impact vegetation. However, the correlation between mean annual precipitation and vegetation may be not significant; alternatively, monthly average precipitation during the growing season or monthly evapotranspiration might be a much better way to analyze the effect of water on vegetation. It is reasonable to take as many environmental factors as possible into account, especially in highly heterogeneous environments and vegetation (Cheng et al., 2014).
The Qilian Mountains in north-western China are located in the ecotone of the Qinghai-Tibet Plateau, the Loess Plateau and the Central Asian desert (Chen, Peng, Huang, & Lu, 1994). The region's vegetation is typical of alpine vegetation in an arid area, and it is an ideal site for understanding the relationship between vegetation and a highly heterogeneous environment (Cheng et al., 2014). However, there has been no attempt to systematically understand the vegetation-environment relationship at different scales (Cheng et al., 2014).
In this study, we used a direct ordination approach to understand correlations between environmental factors and vegetation metrics in a highly heterogeneous environment at different scales and to find the differences in these correlations at different scales and their applications in resource management.

| Site description
The upper reaches of the Heihe River basin are located in the far north- The annual precipitation varies in the range 149-486 mm with more than 60% concentrated in the summer. The mean annual temperature (MAT) ranges from 6.9 to −9.8°C, with cooler averages at increasing elevation (climate data sourced from WorldClim, http://www.worldclim. org/). Precipitation increases from west to east and north to south in the study area, with temperature having the opposite trend (Gao, Hrachowitz, Fenicia, Gharari, & Savenije, 2014;Qin et al., 2013;Zhao, Nan, & Cheng, 2005).
F I G U R E 1 Location of study area

| Vegetation and environmental data
The vegetation data were modified and improved from the digital  Figure 2). The lowlands (1,600-2,400 m) are mainly desert, and the upper regions (2,400-2,800 m) are steppe consisting of Stipa spp., with needle-leaf forest in the north ranging from 2,400 to 3,200 m, shrub-meadows from 3,200 to 4,000 m and alpine vegetation, mainly Saussurea spp., in areas higher than 4,000 m. Glaciers form at the peaks of some mountains. The main land use is pasture; the forest is protected by the government; and logging has been forbidden in recent years. Some cultivated vegetation is grown near county towns, but cultivation covers an area of less than 1% in this region.
The environmental data used in this study comprised of 33 factors: Topography: altitude, slope, aspect.

| Correlation analysis
All landscape metrics and environment variables were submitted to the Shapiro-Wilk test for normality, a basic requirement for further application of parametric tests (Daniel, 1990), and all variables showed normal distribution.
The vegetation metrics and environmental data were analyzed using a direct ordination performed in the CANOCO for Windows program (version 4.5) (Braak & Smilauer, 2002;Hejcmanovā-Nežerková & Hejcman, 2006) at different scales including LULC, vegetation groups, vegetation types, and formations and subformations.
Detrended correspondence analysis (DCA) was conducted for landscape metrics data to detect the length of the species gradient. After DCA, redundancy analysis (RDA) was used because the length of gradients was smaller than 3 (Lepš & Smilauer, 2003

| Optional spatial extent analysis
A series of area extent was used to calculate vegetation landscape metrics and to determine the optional spatial extent. The influence of spatial scale was found to be highly significant. Within these spatial extents, at 15 × 15 km 2 , explained variability was more F I G U R E 3 Explained variance, number of samples, and scale to choose the optional spatial extent T A B L E 2 Results of redundancy analyses at LULC level. Variables abbreviations: soil moisture measured at 2 cm from surface (SM2), soil moisture measured at 100 cm from surface (SM100), topsoil gravel content (GRAVEL), topsoil bulk density (BD), reference soil depth (RD), topsoil reference bulk density (RBD), topsoil organic carbon (OC), topsoil pH (PH), cat ion exchange capacity of the clay fraction in the topsoil (CEC), topsoil base saturation (BS), topsoil salinity (ECE), soil texture (Texture), frozen soil (FS); mean annual precipitation (MAP), growing-season precipitation (GSP), actual evapotranspiration (AET), groundwater depth (ZWT), potential evapotranspiration (PET); active accumulated temperature (≥0°C) (AAT0), active accumulated temperature (≥10°C) (AAT10), mean annual temperature (MAT), growing-season temperature (GST), mean temperature of the coldest month (MTCO), mean temperature of the warmest month (MTWA), mean annual bio-temperature (MAB); solar radiation (RAD), atmospheric counter radiation (ACR), surface pressure (PSFC), annual average insolation duration (AST than 60%; sample's number was more than half of total vegetation patches; and these samples were distributed all over the study area, 15 × 15 km 2 being found to be the appropriate optional extent ( Figure 3).
The sum of all canonical eigenvalues was 0.63. The eigenvalue of axes 1 was 0.449 and was 0.086 for axes 2. Light and radiation, Temperature, soil, precipitation, topography, and human disturbances accounted for 49%, 25%, 24%, 5%, 3%, and 0% of all the explanatory environmental variables, respectively (p ≤ .05) (   Soil moisture is important for assessing water availability for plant growth in alpine prairies because it impacts nutrient uptake (He, Xing, & Bai, 2014). The distribution pattern of soil water is considered as the key factor to restrict the vegetation status (Xie et al., 2015). Although many soil factors were considered, only the effect of soil moisture measured at 2 and 100 cm from surface had significant effect on vegetation pattern in the upper reaches of Heihe River. Soil moisture measured at 2 cm from surface had significant relations to Mean Shape Index at levels except formation and subformation. Connectance Index was significantly explained by soil moisture measured at 2 cm from surface at the vegetation group level and by soil moisture measured at 100 cm from surface at the formation and subformation level, respectively. Human disturbance was less important in the study area, because the upper reaches of the Heihe River basin were protected by the Chinese Government, and human activity was weak (Wu, 2011), the farmland only 1% of the total area, and settlements and the road system were also limited.

3.
The sum of all canonical eigenvalues of each level was similar but the specific aspects of vegetation patterns explained by environmental factors were different.
There are some reports that environmental factors affecting vegetation distribution are different at different levels (Dias & Melo, 2010;Kariyeva, Van Leeuwen, & Woodhouse, 2012), and that it is insufficient to use the same environmental factors to analyze the relationship between vegetation and environment and then to predict the effect of environment changes on vegetation; consideration of suitable environmental factors should be based on the specific vegetation level (Kariyeva et al., 2012). Our research findings indicated that this was also true in the Heihe River basin, and that it is important for policy makers and local government to make appropriate policies for environmental conservation.

| Common environmental factors determining vegetation pattern at different levels
The upper reaches of the Heihe River basin are in an alpine region, which is part of the Tibetan Plateau. For the four levels, atmospheric counter radiation had the highest correlation with landscape metrics.
Atmospheric counter radiation was much more important for vegetation than precipitation, soil, and altitude. Because most of the study area was located in high mountains, energy should be the most important variable for maintaining vegetation, as reported by other studies (Qiu, Zeng, Chen, Zhang, & Zhong, 2013;Sohoulande Djebou et al., 2015).

| Uncertainty of relationship between environment and vegetation
It is not reasonable to expect environmental factors to explain all variations in vegetation, because other factors may affect it, such as the history of the vegetation and disturbance activities (Motzkin et al., 1999).
F I G U R E 7 Redundancy analysis diagram in the upper reach of Heihe River basin with respect to landscape metrics and environmental factors of vegetation formations and subformations. Other descriptions are same as Table 2 and Figure 4 Furthermore, in the present study, data matching was also a limiting factor. For example, the spatial resolution of temperature and precipitation was 0.05°; the spatial resolution of DEM, aspect, and slope was 30 m; and, additionally, the study area was a high mountain.
Mismatched data may have had a negative effect on our attempts to reach a reliable conclusion. This problem might be solved in the future with more accurate observations both on the ground and in remote sensing observations (Austin, 2013).

ACKNOWLEDGMENT
This work was supported by National Natural Science Foundation of China (No. 91225302). We also greatly appreciate the anonymous reviewers for the insightful comments.