Data analysis
We used detrended correspondence analysis (DCA) for the ordination of samples. DCA is an effective method in community analysis. In our study, we conducted DCA using a site-order matrix with relative abundance data to analyze the similarity of order compositions among samples. Kruskal-Wallis method was used for analysis differences in the richness of soil faunas in the three climatic regions.
We performed a redundancy analysis (RDA) based variation partitioning analysis to assess the relative effects of environmental and spatial variables on soil fauna community composition. Before the RDA, the environmental variables with high variance inflation factor (VIF) >10 were eliminated to avoid collinearity among factors (Singh et al., 2019). The importance of environmental and spatial variables in explaining order compositions was determined by an RDA analysis using Monte Carlo permutation tests (999 unrestricted permutations) followed by forward selection to remove the non-significant variables from each of the explanatory sets. The “envfit” function in the R software with “vegan” package was used to test the significance of each environmental factor and orders distribution (Oksanen et al., 2007). The pH value results were not significant in overall, phytophage, and predacity species composition, so the pH value was excluded in the subsequent analysis.
For environmental variables, climate and soil factors (MAP, MAT, MTCM, EMT, SOC, and SBD) were used to determine the environmental divergence between pairs of sampling sites. All environmental variables were normalized as: x ’ = (x - mean(x )) / standard deviation (x ), where x is a variable. Differences of latitude values (Table 1) of the sampling sites were used to obtain the spatial variable as a response variable. Relationships between turnover rate of order compositions and environmental and spatial variables were determined with linear regression. We used the dissimilarity coefficient (βj and βs) as the response variable and three sets of explanatory variables which included climate variables (MAP, MAT, MTCM, and EMT), soil factors (SOC, and SBD) and spatial variables (geographical co-ordinates for sampling sites), respectively. Where necessary, values were log (x + 1) transformed in order to meet assumptions of normality of residuals.
To further evaluate the relative importance of each environmental variable and spatial distance on the order turnover rates, we used a Partial RDA (pRDA) approach. This method can analyze the effects of pivot variables and covariables on order distributions respectively (Lososová er al., 2004). Partial RDA analysis divided the variance in order turnover index into eight parts, which were pure spatial effects, pure climate effects, pure soil effects, spatially structured climate effects, spatially structured soil effects, climatological soil effects, spatially structured environmental effects, and the unexplained part.
All statistical analyses were carried out with R v.3.4.3 (R Core Development Team, 2017). DCA and PCA were performed using the “vegan” package (Oksanenetal et al., 2007).