Shrub patch configuration in relation to precipitation and soil properties in Northwest China

2.1 Study area

The Inner Mongolia Plateau is located in Northwest China and is characterized by a continental temperate climate (Figure 1). From the northeast to southwest, it is characterized by an increase in solar radiation and air temperature, and a decrease in precipitation, resulting in a strong climatic drought gradient (Xie, Guo, Gabler, Li, & Ma, 2015). This climatic gradient provides an ideal transect for studying the changes in shrub patch configuration. An east–west transect was established along the climatic gradient in the study area (Figure 1). Six sample sites (S1–S6) were selected from the precipitation gradient in sequence. This transect covers semiarid, arid, and extremely arid areas associated with landscapes of typical steppe, desert steppe, and Gobi Desert, respectively. All sites have a long history of grazing except S1 which located in Gobi desert with extremely arid environment. Mean annual precipitation (MAP) of the six study sites ranged from 54 to 322 mm, whereas mean annual temperature (MAT) ranged from 2.7 to 12.9 °C (Table 1). All meteorological data from the weather stations were obtained from the China Meteorological Data Sharing Service System. The altitudes varied from 995 to 1521 m. The main shrub and grass species found at each site are listed in Table 1. The dominant shrubs were Reaumuria songarica, Nitraria sphaerocarpa, Caragana tibetica, and Caragana microphylla, whereas the associated dominant grasses were Cleistogenes squarrosa and Stipa capillata. Soil types included Hapli-Orthic Aridisols, Sali-Orthic Aridisols, and Hapli-Ustic Cambisols. Soil texture changed from sand (S1), loamy sand (S2, S3, and S4), and sandy loamy (S5) to sand (S6) again, along the MAP transect.

2.2 Field survey

The field plant and soil investigations were carried out from 12 to 24 August 2011 across the Gansu province, Ningxia Hui Autonomous Region, and Inner Mongolia Autonomous Region. In the 2000 km route (Figure 1), we investigated the vegetation quadrats, patch patterns, and individual shrubs, and carried out soil sampling at the above six sample sites, separately as S1 (54 mm; 12 to 13 August), S2 (116 mm; 14 to 15 August), S3 (145 mm; 18 to 19, Aug.), S4 (191 mm; 21 August), S5 (210 mm;16 to 17 August), and S6 (322 mm; 23 to 24 August). All these regions are characterized by dry climate, low and erratic precipitation, strong solar radiation, and wind.

To avoid biases from community heterogeneity, we set up three randomized 25 m × 25 m plots (625 m²) at each sample site. In each plot, three 5 m × 5 m quadrats for shrub patches and three 1 m × 1 m quadrats for interspace grass patches were established to analyse patch configuration, biomass, and herb community. We pursued the plot investigation as below. First, we recorded the elevation and geographical coordinates of each plot. Second, we recorded the shrub names, measured the crown diameters, height, biomass, and position of all shrub patches in three plots of each sample sites. Third, we recorded species names and measured the mean cover, height, density, and biomass of each species in three 1 m² herb quadrats. In detail, aboveground biomass of shrub (AGB-s) and grass (AGB-g) were sampled by clipping live vegetation rooted within the quadrat to the ground surface. Belowground biomass of shrub (BGB-s) was determined by digging within the estimated area of shrub roots at depths ranging from 0 to 100 cm at 10 cm intervals. The roots were washed and oven-dried at 65 °C to constant weight for determination of biomass. BGB-s/AGB-s, which reflects the biomass allocation strategies of shrubs in vertical direction, is the ratio of shrub belowground to aboveground biomass. The average shrub height was assessed by measuring 6 times at each quadrat. The areas and spatial patterns of shrub patches were recorded on paper in the field in each 25 m × 25 m plot and then digitized using ArcGIS 10; shrub coverage, size, and the spatial distribution of shrub patches were estimated using Fragstats 4. The edges of the shrub patches in the field were visually identified, on the basis of the projective geometry of crown shape and size of shrubs growing in isolation or in dense clusters. The average distance between shrubs was calculated using Conefor Inputs, an expanded procedure of ArcGIS. Shrub density refers to the number of shrubs within a unit area, whereas shrub cover refers to the ratio of the whole vertical projection area of aboveground parts to the quadrat area. Species richness was calculated using the Shannon–Wiener index:

(1)

where H′ is Shannon–Wiener index; S is the number of species in the quadrat; P_i is the relative importance index of species i.

The soil samples were collected at depths of 0–100 cm at 10 cm intervals. Afterwards, they were air-dried and then sieved to pass through a 2-mm screen. For grain size distribution, the sand fraction (2–0.05 mm) was separated by sieving, and the silt (0.05–0.002 mm), and clay (<0.002 mm) fractions were separated using the pipette method. The soil organic matter (SOM) was determined using the K₂Cr₂O₇ titration method after digestion. Total nitrogen (TN; %) was measured using a Nitrogen/Carbon Analyser (NA 1500 NC elemental analyser, Fisons Instruments, Italy) with Eager 200 software. Total phosphorus (TP) was measured by molybdenum blue colorimetry, after H₂SO₄-HClO₄ digestion. TN, TP, and SOM of 0–20 cm soil profile under shrub patch were combined to represent the soil fertility conditions. The aridity index (AI) is the ratio of mean annual potential evapotranspiration (PET) to MAP (PET/MAP) from 2000 to 2010 (Peel, Mcmahon, & Finlayson, 2010). In detail, PET was calculated by the Hargreaves equation (Hargreaves & Allen, 2003):

(2)

Where R_a is the extraterrestrial radiation (MJ m⁻² day⁻¹), which can be estimated according to the equations provided by FAO (Food and Agriculture Organization) for each day of the year and for different latitudes (Allan, Pereira, Raes, & Smith, 2005); TC means daily temperature (°C); TR is the daily temperature range (°C).

2.3 Data analysis

Curve fitting and correlation of shrub patch configuration and biomass was conducted using version 19.0 of the SPSS software (SPSS Inc., Chicago, IL). Then we used a forward stepwise regression method to drop non-significant variables before they were included into the general linear models (GLMs) to explore the explanations of climatic, soil, and vegetation factors on shrub patch configuration. Landscape index analysis was conducted using Fragstats 4 (Cover, 1982). Principal component analysis (PCA), redundancy analysis (RDA), and variation partitioning were conducted using Canoco 5.0 (Parkhurst, 2003). PCA constructs two theoretical variables that best fits the data according to a linear model. The theoretical variable is the first and second ordination axis (PC1 and PC2). The case scores are the values that the theoretical variable takes in the cases. The theoretical variable/ordination axis does not need to be related to any observed explanatory variable. The ordination diagram was used to display the result of PCA. RDA is a multiple regression for all responses variables simultaneously (i.e., a multivariate regression) with linear constraints on the regression coefficients. Because of the constraints, RDA uses less parameters than multivariate multiple regression. In summary, RDA is thus both a constrained form of PCA and a constrained form of multivariate multiple regression. In PCA ordination diagram and RDA figures, a positive correlation is expected when the arrows of two variables point to same direction, and the angle between two arrows is inversely proportional to the degree of their correlation. The longer the arrows, the better relationship indicated between variables and the canonical axes. The variation partitioning analyses partition the variation (or variance) in the response variables into parts attributed to sets of explanatory variables. The analysis uses the forward selection of group members to select a set to explain response variables and also perform a permutation test of the joint effect of all the candidate members of each group as recommended (Blanchet, Legendre, & Borcard, 2008). All patch configuration, climate, soil, and vegetation variables were transformed as log (x + 1) for the normalization in ordination diagram, RDA and variation partitioning analysis. All statistical analyses were carried out using the p = .05 confidence level (^*p < .05; ^** p < .01), coefficient of correlation analysis is represented by r.

3.1 Overall climatic, vegetation, and soil characteristics

Meteorological, plant, and soil characteristics are listed in Table 1. AI was negatively correlated with MAP (r = −0.841^**). S1 was the driest site and S6 was the wettest in the transect. Main shrub species varied from R. songarica to C. microphylla, showing an increasing trend in coverage with more MAP. Among them, R. songarica was widely distributed (in S1, S4, and S5) and its AGB-s increased with increasing MAP. However, BGB-s of this species was higher in S4 and S5 but extremely low in S1 (8.95 g/m²). BGB-s of other three sites was close to each other. Interestingly, BGB-s/AGB-s ratio also exhibited an increasing trend along rising MAP gradient. Shrub height, density, and size all showed a parabolic relationship with MAP, indicating higher, sparser, and larger shrub patches in S1 and S6 and shorter, denser, and smaller ones in S3 and S4. Interspace developed from bare ground to grassland, grass cover, AGB-g, and H′ increased with MAP except in S3. AGB-g was extremely high (134.6 g/m²) in S6, but much lower in other sites. Remarkably, the interspace/shrub radius exhibited a significant decreasing trend along with precipitation (p = .064). Third, TP remained stable across study sites, but TN and SOM were positively correlated MAP, increasing from west to east in spatial distribution (p = .006). Soil particle size (sand, silt, and clay) did not show obvious variation patterns along the MAP gradient.

3.2 Variation of shrub patch configuration along the climatic gradient

It was observed that the shape of shrub patch tended to change from spot in the site with low precipitation to strip in the site with high precipitation (Figures 1 and 2). In Figure 2, the size of shrub patch was relatively large in S1 (MAP = 54 mm) and S6 (MAP = 322 mm). R. songarica was the only shrub species in the S1 without grass in the interspace and shrub patch, whereas C. microphylla coexisted with grasses. In other sites, two or more shrub species coexisted with the mosaic of shrub patches and interspace grass patches.

Changes in shrub cover, patch size, height, and density along with increased gradient of MAT, MAP, and AI are presented in Figure 3. MAP had the strongest relationship with these indicators. Shrub cover significantly increased with increasing MAP. Other indicators all showed parabola fitting with precipitation (Figure 3). There were obvious trends in the spatial traits of shrub patches with AI. For instance, shrub cover and patch size decreased with increasing AI at a significant level (p < .001). These trends were consistent with results shown in Table 1. However, temperature showed a weaker effect on these indicators. For example, shrub height had no correlation with MAT; shrub cover increased with MAT after first decreasing significantly.

Interspace/shrub radius decreased with increasing MAP following a linear function (Figure 4a). Shrub patches were distributed more closely (Figure 2e), and patch size increased (Figure 2f) with increasing MAP. Interspace/shrub radius was approximately 10, 5, and 3 under the gradients of 0–100, 100–200, and 200–400 mm, respectively. The index was more than threefold in S1 (MAP = 54 mm) of the S6 (MAP = 322 mm) at a decrease rate of 2.2/100 mm. In contrast, BGB-s/AGB-s showed a linear positive correlation with MAP (Figure 4b). BGB-s/AGB-s of S6 was 4.6-fold that of S1 and increased at the rate of approximately 0.7/100 mm. Generally, shrubs had less interspace among them and allotted more percentage of biomass to belowground (0–1 m) part under more MAP conditions.

Shrub patch configuration (shrub cover, density, height, patch size, BGB-s/AGB-s, and interspace/shrub radius) of 18 plots in six sites were used to do PCA ordination diagrams (Figure 5). Relative distances, relative directions or relative ordering of sample plots, and arrow lines were used to interpret these results. In this figure, configuration dissimilarities between two plots are positively correlated with their relative distances. The arrow points in the direction in which the configuration variable value increases at the largest rate across the PC axis. The rate of change in the orthogonal direction is zero. The arrow can be extended on either side to form a line. By projecting the plot points on the line and ranking the projection points, a ranking can be obtained for that particular configuration variable. Two theoretical variables constructed by PCA are the first and second ordination axis (PC1 and PC2), which does not relate to any observed explanatory variable. The case scores in the brackets are the values that the theoretical variable takes in the cases (40.1% and 29.6%). First, three replicates of each study site showed a close distribution, suggesting its internal similarity. S2, S3, and S4 (Figure 5) overlapped due to some similarities in shrub cover and interspace/shrub radius (Table 1). Second, the distance between the sample points approximated their differences in configurations. Thus, the study sites could be divided into four groups according to their spatial traits as S1 (MAP = 54 mm), S2 to S4 (116–191 mm), S5 (210 mm), and S6 (322 mm). Among them, S1 greatly differed in interspace/shrub radius and shrub cover, thus it was isolated from the other sites. Lastly, some of these variables were correlated. For instance, shrub cover and interspace/shrub radius were negatively correlated (r = −0.797^**). Shrub height and density also showed a negative relationship (r = −0.510^*), with low shrub patches usually linked with dense distribution. Analysis of the above results showed that shrub patch tended to occupy a larger percentage of landscape and had less interspace under more MAP conditions. Shrub patch varied from larger, higher, and sparser shrub patches in S1 to smaller, shorter, and denser ones from S2 to S4 and then to higher and larger ones in S5, but sparser in S6.

Table 2 shows landscape indexes of the six sites. Number of patches (NP) was used to describe landscape heterogeneity. NP was low in the S1 and S6 (NP = 45, 172; MAP = 54, 322 mm, respectively) sites whereas the maximum values were observed in S4 (NP = 656, MAP = 191 mm), showing the trend of decrease after first increase along with increasing precipitation gradient. In addition, S1 and S6 showed many similarities in other indices that differed from the corresponding values in other sites. The ratio of the largest patch to total patch area (LPI) was significantly larger in S1 (20.9%) and S6 (15.8%) than in the other sites (1.5–3.8%). Contiguity index (CONTIG), which was an important index in landscape pattern, showed the aggregation extent or extending trend. CONTIG was a little higher in the S1 and S6 sites, reflecting more small patches from S2 to S5, whereas the dominant patches in S1 (R. songarica) and S6 (C. microphylla) already formed better connectivity. The results of connectance index (CONNECT) and patch cohesion index (COHESION) also reflected similar trends. Interspersion juxtaposition index was used to determine the overall dispersal and gathering conditions on a landscape scale. IJI was low in all sites, indicating that shrub patches connected few other patch types. In conclusion, S1 and S6 had many similarities in all landscape indices and formed scattered distribution with better connectivity. Other sites were characterized by dense distribution with small shrub patches.

3.3 Combined effects of climate, vegetation, and soil on shrub patch configuration

Multivariate analysis methods, such as RDA, GLMs, and variation partitioning were applied to analyse the combined effects of climate, vegetation, and soil on shrub patch configuration. In these three analysis, MAT, MAP, and AI were selected to represent climatic factors, H′, and AGB-g to represent vegetation factors and TN, TP, SOM, and sand to represent edaphic conditions, to assess their contribution to the changes in shrub patch configuration, which was represented by shrub height, cover, patch size, density, interspace/shrub radius, and BGB-s/AGB-s (Table 3).

Results of the RDA analysis showed a comprehensive relationship between explanatory variables and indicators of shrub patch configuration (Figure 6). As introduced before, the RDA is interpreted by the line length and the angles between two arrows. Red arrow lines represent environmental/explanatory variables, including climate, soil, and vegetation variables, whereas blue arrow lines stand for response variables of shrub patch configuration. Many of these variables or indicators were strongly correlated, such as MAP and shrub cover (r = 0.647^**), AGB-g and patch size (r = 0.958^**), AI and interspace/shrub radius (r = 0.913^**), and sand and density (r = −0.634^**). Climate (e.g., MAP and AI), vegetation (e.g., H′ and AGB-g), and soil (e.g., TN, SOM, and sand) all had significant contributions to these two axes. However, there was also strong correlation between the explanatory variables. Therefore, GLMs was used to explore the explanatory variables for each configuration indicator.

GLMs results revealed that climatic factor could explain 84.2% of the total variance in shrub cover, 55.7% in height, 61.9% in interspace/shrub radius, and 20.0% in BGB-s/AGB-s (Table 4). AGB-g and MAT were the primary factor affecting BGB-s/AGB-s, whereas MAP and MAT combined was the major factors determining shrub cover and interspace/shrub radius. Shrub height was positively correlated with AI. GLMs also revealed that soil texture (sand) was the key factor in determining shrub density (Table 4). Soil could explain 14.0% of the variance in shrub height, 40.2% in density, and 15.9% in interspace/shrub radius. In general, shrub patches showed high aggregation patterns but were separated from each other in space with coarse soil texture, such as S1 and S6. In contrast, shrubs dispersed to low patches but were close to each other in sites with fine soil texture, such as S2 and S4 (Figure 2). The results also showed that vegetation factors (AGB-g) could also improve the explanatory power for patch size and interspace/shrub radius to 91.8% and 19.7%, respectively. AGB-g was the primary factor affecting patch size. According to the stepwise regression, there were significant negative correlations between shrub height and SOM, density and sand, AGB-g, MAP, MAT and interspace/shrub radius, AGB-g MAT, and BGB-s/AGB-s (Table 4, coefficient). Other factors were all positively correlated.

Furthermore, variation partitioning analysis was made to quantitatively analyse the total effects of climate, vegetation, and soil on shrub patch configuration and their combined effects. The variation explained figure (Figure 7) displays the computed explanation of the seven variation fractions and compares them, as percentages, with respect to the total variation in the response data. In total, climate, soil, and vegetation explained 72.9% of shrub patch configuration at a significant level. Edaphic conditions made the largest contribution (5.8%), whereas climate accounted for just 1.8%, but their combined effects explained 31.6% of the shrub patch configuration. Vegetation alone contributed 4.6%, and the combined effects of climate and vegetation were 27.9%. The interaction of climate, soil, and vegetation factors accounted for 5.5% of configuration variations.