Vertebrate herbivore-induced changes in plants and soils: linkages to ecosystem functioning in a semi-arid steppe

Authors


Correspondence author. E-mail: yfbai@ibcas.ac.cn

Summary

  1. Large grazing herbivores have been reported to determine the structure and function of grassland ecosystems. However, the ecological linkages between structure and functioning components have yet been thoroughly explored.
  2. Here, we test the hypothesis of the impact of grazing on soil nematode community (e.g. structure and composition) and linkages to ecosystem functioning (e.g. soil N mineralization and ANPP) via changes in pathways of plant community, soil nutrients and soil environment using a field experiment maintained for 5 years with seven levels of grazing intensity in the Inner Mongolian grassland.
  3. A structural equation model (SEM) with nematode abundances as response variables showed that plant-feeding and fungal-feeding nematodes were driven by changes in the plant community, and bacterial-feeding nematodes were affected by soil abiotic nutrients and environment, while omnivorous + carnivorous nematodes were altered by soil environment and bacterial-feeding nematodes. This indicates that the top-down control by grazing leads to bottom-up control in the soil food web.
  4. We found that grazing affected the ecosystem functioning via different pathways. Grazing effects soil N mineralization by changing plant community, soil nutrients, soil environment and nematodes community structure, while it affects ANPP by altering soil N mineralization and soil environment.
  5. Our findings could provide a better understanding of the responses of plants and soils to grazing and the linkages between structure and functioning of above-ground and below-ground in the semi-arid steppe.

Introduction

In grassland ecosystems, above-ground grazing has far-reaching effects on the ecosystem structure and function (McNaughton 1985; Milchunas & Lauenroth 1993; Bardgett & Wardle 2003). Structurally, grazing modifies plant community, plant species diversity, soil nutrients and soil environment (McNaughton 1985; Milchunas & Lauenroth 1993), as well as soil fauna community in grasslands (Wardle et al. 2001; Bardgett & Wardle 2003, 2010). Functionally, grazing alters the soil nitrogen cycling (Ritchie, Tilman & Knops 1998; Frank et al. 2000) and above-ground net primary productivity (ANPP) (Robin & Rohweder 2000). However, the ecological linkages between the structural traits and the functional traits have not been thoroughly addressed (Wardle et al. 2004; Bardgett & Wardle 2010).

It is well documented that above-ground grazing has strong effects on the soil nematodes via their impacts on the plant community (e.g. structure and diversity) (Bardgett & Wardle 2003), soil nutrients (e.g. soil organic carbon and soil inorganic nitrogen) (Bardgett & Wardle 2003; Mikola et al. 2009) and soil environment (e.g. pH, soil bulk density and soil moisture) (Korthals et al. 1996; Mulder, Van Wijnen & Van Wezel 2005; Sorensen et al. 2009) in grassland systems. Considering the different trophic levels of nematodes in the soil (e.g. plant-feeding, fungal-feeding, bacterial-feeding, omnivorous and carnivorous nematodes), there are several pathways involved in determining the nematodes under grazing situation, such as bottom-up, top-down, competitive control factors (e.g. resource specificity) and soil environment (Bardgett & Wardle 2010; Veen et al. 2010). However, the underpinning mechanisms of grazing-induced effects on soil nematodes via plants, soil nutrients, soil environment or their interactions are still difficult to interpret (Mikola et al. 2009; Veen et al. 2010).

Previous independent studies have considered the grazing effects on ecosystem functioning (e.g. N mineralization and ANPP) through modification of the pathways of plant community (Ritchie, Tilman & Knops 1998), abiotic soil nutrients (Bardgett & Wardle 2010) and soil environment (Shan et al. 2011). As part of the food web, the soil nematodes drive ecosystem functioning (e.g. N mineralization and ANPP) (Ferris, Bongers & De Goede 2001; Bardgett & Wardle 2010), while the independent contributions of these pathways to ecosystem functioning have not been thoroughly explored under different grazing situations in grasslands. Moreover, in previous studies, the responses of plants and soils in grassland were mostly carried out under limited grazing treatments (e.g. not grazed, moderate grazing or heavy grazing) (Zolda 2006; McSorley & Tanner 2007; Mikola et al. 2009; Veen et al. 2010) and lacked an understanding under long-term and multi-level grazing field situations. Therefore, to better understand the responses of plants and soils to grazing, a multi-level grazing intensity experiment would be necessary.

In this study, we examined the causal pathways of grazing effects on plant and soil communities and linkages to ecosystem functioning by using structural equation modelling (SEM) in a semi-arid Inner Mongolia steppe ecosystem. Data on plant community, soil nutrients, soil environment, soil nematodes, soil N mineralization and ANPP were collected from a five-year field experiment maintained with seven levels of grazing intensity by sheep grazers. Our study addresses two questions in the semi-arid grasslands: (i) How does grazing affect all the measured factors, such as plant community, soil abiotic properties, soil fauna and ecosystem functioning (including soil N mineralization and ANPP)? (ii) Via which pathways does grazing affect soil communities and ecosystem functioning? To explore the underlying mechanisms of grazing impacts on plant and soil communities and thereby ecosystem functioning, we hypothesize that grazing influences soil nematode community structure and composition via alterations in hierarchical pathways (See Appendix S1). These pathways are controlled by both top-down (i.e. grazing influences nematode community by altering plant community) and bottom-up (i.e. grazing influences nematode community by altering soil nutrients and environment) forces.

Materials and methods

Experimental site

This study was conducted at the Inner Mongolia Grassland Research Station (IMGERS, 43°38′N, 116°42′E) of the Chinese Academy of Sciences, which is located in the Xilin River Basin of Inner Mongolia, China, at an altitude of approximately 1200 m above sea level (Bai et al. 2004). The semi-arid continental climate is characterized by mean annual precipitation of 336 mm and mean annual temperature of 0·5 °C (1970–2009). Precipitation mainly falls in the growing season (June–August), which coincides with high temperatures. The soil is a loamy sand texture (Calcic Chernozem according to ISSS Working Group RB, 1998). Leymus chinensis (Trin.) Tzvel. (perennial rhizomatous grass) and Stipa grandis P. Smirn. (perennial bunchgrass) are two dominant species in the study area, together accounting for 60–80% of total above-ground biomass.

Grazing experiment and sampling plots

Our experimental plots were located at the Sino-German grazing experimental site, which was established in June 2005 and occupies a total area of 160 ha (Schönbach et al. 2011). Seven levels of grazing intensity (GI = 0, 1·5, 3·0, 4·5, 6·0, 7·5 and 9·0 sheep ha−1) were set up. Each level of grazing intensity included one standard plot. To reduce high inherent spatial heterogeneity in plant community composition and soils between the plots and high management expense, we did not set replicated blocks in the grazing experiment (Schönbach et al. 2011). Each standard plot size was 2 ha except for treatment GI = 1·5, in which the plot size was 4 ha to achieve a minimum of six sheep per plot. All plots have been fenced since the beginning of the experiment to prevent sheep migration between plots. In the grazing area, sheep were allowed to graze continuously during the vegetative growth period from June to September. In each standard plot, five exclosure cages (2 m × 3 m) were randomly set up and moved on an annual basis, for measurement of above-ground net primary productivity (ANPP). Near each exclosure cage, one sampling quadrat (1 m × 1 m) was set up for sampling of plant community, below-ground biomass, soil and soil N mineralization.

Plant community and ANPP

In late August 2009, above-ground biomass, below-ground biomass, soil nutrients and soil environment were measured at each sampling quadrat (1 m × 1 m). Above-ground vegetation was sampled by clipping all plants at the soil surface. All living vascular plants were sorted to species, and all plant materials, including litter and standing dead matter, were oven-dried at 70 °C for 48 h and weighed. The dry mass of all living plants per quadrat averaged over the five replicates for each treatment was used to estimate the above-ground community biomass. We classified all plants into two functional groups based on life forms, including rhizome grasses and bunchgrasses as in Bai et al. (2004). For measuring below-ground root biomass, one soil core (diameter 5·6 cm, depth 40 cm) was collected from each sampling quadrat. Roots were washed over a sieve with 2 mm mesh size, dried at 70 °C for 48 h and weighed. The dry above-ground community biomass and below-ground root biomass samples were ground to analyse nitrogen content using the micro-Kjeldahl digestion, followed by colorimetric determination on a 2300 Kjeltec Analyzer Unit.

The above-ground biomass was measured using annually moved exclosure cages in late August 2009, approximating above-ground net primary production (ANPP), because the standing above-ground biomass of these steppe communities reached the annual peak at the end of August (Bai et al. 2004). The standing biomass was sampled using a 1 m x 1 m quadrant randomly placed inside and outside of each exclosure cage. Above-ground vegetation was sampled by clipping all plants at the soil surface in each quadrant. The plant material was oven-dried at 70 °C for 48 h and weighed. ANPP was calculated based on the biomass measurement inside and outside the annually moved exclosure cages.

Soil sampling and analysis

In late August 2009, we measured soil bulk density over 0–10 cm at each sampling quadrat using the core method (diameter 5 cm, depth 10 cm). Soil temperature (depth 10 cm) was recorded simultaneously at each sampling quadrat using a portable soil temperature recorder (SN2202, China). Soil sampling was collected using soil core (diameter 2 cm, depth 15 cm) from each sampling quadrat between the plants. One composite soil sample was taken from the upper 0–15 cm soil in each sampling quadrat, with each composite consisting of four randomly selected soil cores, so that a total of five composite soil samples were collected in each plot. After gentle homogenization and removal of roots, the moist soil was sieved through 2-mm mesh and separated into two parts. One part was maintained fresh for extraction of soil nematodes and the measurement of extractable dissolved organic carbon (DOCex), extractable dissolved organic nitrogen (DONex) and soil moisture. The other was air-dried to determine soil pH value, soil organic carbon and total soil nitrogen. We determined DOCex and DONex using K2SO4 extraction method (Chen et al. 2010). Briefly, 20 g of subsample of field moist soil was extracted with 60 mL of 0·5 mol L−1 K2SO4, and the filtered soil K2SO4 extracts were analysed for soluble organic carbon and soluble nitrogen using a total organic carbon analyzer (Shimadzu TOC-VCPH). Twenty gram of subsample of moist soil was oven-dried at 105 °C for 24 h to determine soil moisture. Soil pH was measured in a 1:2·5 (soil/water) suspension. Soil organic carbon content was determined using the dichromate oxidation method. Total soil nitrogen was determined by the micro-Kjeldahl digestion, followed by colorimetric determination on a 2300 Kjeltec Analyzer Unit. All results are expressed on a dry weight basis.

Soil nematodes

For each composite soil sample, nematodes were extracted for 48 h from 50 g of moist soil using the Baermann funnel method (Barker 1985). After fixation in 4% formalin solution, nematodes were counted under an inverted microscope, and the first 100 individuals encountered were identified to genus. All nematodes were identified when the nematodes were fewer than 100 individuals in each sample. Taxa are classified as five trophic groups (plant-feeding, bacterial-feeding, fungal-feeding, omnivorous and carnivorous nematodes) and their respective c-p (colonizer-persister) value (Yeates et al. 1993; Ferris, Bongers & De Goede 2001). As genera designated as the carnivorous trophic group were found infrequently in our experiment, we merged it to omnivorous + carnivorous trophic group.

To characterize nematode community structure in each sampling, we calculated nematode taxon ecological indices by the following approaches (Bongers 1990; Yeates & Bongers 1999): (i) genus richness R = (S−1)/ln(N), where S is the number of taxa and N is the number of total nematodes; (ii) maturity index MI = Σv(if(i) for non-plant parasitic nematode families, where v(i) is the c-p (colonizer-persister) of taxon i according to their r and K characteristics, f(i) is the frequency of taxon i in a sample; and (iii) plant parasite index PPI = Σv(if(i) for plant parasitic nematode families. The PPI is identical in concept to MI, but is based only on the life-history characteristics of the plant parasitic nematodes. MI and PPI are used to evaluate the functional responses of soil nematodes to resource and environmental change (Bongers 1990).

Soil N mineralization

The soil net N mineralization rates were determined using the in situ soil core incubation method (Hook & Burke 1995). During the 2009 peak growing season, soil samples were taken each month from July to September, and soil cores were incubated in the field for 1 month. Within each sampling quadrat, two sharp-edged PVC tubes (5·6 cm in diameter and 12 cm in length) were driven into the soil at a 10 cm depth. One PVC core was sealed by Parafilm membrane to prevent water penetration and allow gas exchange and then incubated in the field for about 1 month. The other core was taken back to the laboratory for measuring the initial inorganic N contents. Each soil core was sieved through 2-mm mesh and NH4+-N, NO3-N contents and soil moisture measured. For determination of NH4+-N and NO3-N contents, 10 g of subsample of field moist soil was extracted with 50 mL of 2 mol L−1 KCl, and the extract was subjected to colorimetric determination on a 2300 Kjeltec Analyzer Unit (FOSS, Höganäs, Sweden). Soil cumulative net N mineralization over the peak growing season was calculated by summing the net amount of nitrogen mineralized (or immobilized) for all of the incubation periods.

Response variables in categories

In the present study, we classified variables into five categories based on information in the literature: (i) plant community pathway, including below-ground biomass, nitrogen content of above-ground and below-ground biomass, plant richness, relative biomass of rhizome grasses and relative biomass of bunchgrasses (Bardgett & Wardle 2003; Mikola et al. 2009); (ii) soil nutrients, including soil organic carbon, total soil nitrogen, soil inorganic nitrogen, extractable dissolved organic carbon and extractable dissolved organic nitrogen (Bardgett & Wardle 2003; Szanser et al. 2011); (iii) soil environment, including soil bulk density, soil pH, soil moisture and soil temperature (Bouwman & Arts 2000; Mikola et al. 2009); (iv) soil nematode community components and structure, components including plant-feeding, fungal-feeding, bacterial-feeding, omnivorous and carnivorous nematodes; and structure including total nematode abundance, richness, maturity index and plant parasite index (Veen et al. 2010); and (v) ecosystem functioning including soil N mineralization (Ingham et al. 1985; Ferris et al. 1998) and ANPP (Robin & Rohweder 2000; Bai et al. 2007).

Statistical analyses

To address the responses of variables to grazing, linear and quadratic regression analyses were performed with grazing intensity levels as a continuous variable. Regression model analyses were performed using SigmaPlot 11.0 statistical software package (Systat Software Inc., Chicago, IL, USA). Structural equation modelling (SEM) was performed to analyse hypothetical pathways that may explain grazing effects on the soil nematode compositions and ecosystem functioning. SEM allows testing of multivariate hypotheses in which some variables can act as predictor and response variables at the same time (Grace 2006; Veen et al. 2010). Prior to the SEM procedure, we reduced the number of variables for plant community, soil nutrients, soil environment and soil nematode community structure separately through principal component analysis (PCA) (Grace 2006; Veen et al. 2010). For each group, the first principal component (PC1) was used in the subsequent SEM analysis. In our PCA results, the PC1 explained 62%, 66%, 58% and 41% total variance of plant community, soil nutrients, soil environment and soil nematode community, respectively (see Appendix S2). In the SEM analysis, via comparing the model-implied variance–covariance matrix against the observed variance–covariance matrix, data were fitted to the models using the maximum likelihood estimation method using Amos version 17.0.2 (Amos Development Corporation, Chicago, IL, USA) to parameterize the model. The chi-square goodness-of-fit statistic and its associated P value were used to judge the model fit to the data. A large P value (>0·05) associated with the chi-square value indicates that the covariance structure of the data does not differ significantly from the expected, based on the model (Grace 2006).

Results

Response of plant community, soil nutrients and soil environment

Grazing altered the plant community, soil nutrients and soil environment based on the results of responses of variables (Fig. 1). Grazing directly changed plant community with decreases in below-ground biomass, nitrogen content of below-ground biomass, plant richness and relative above-ground biomass of rhizome grasses, and with increases in nitrogen content of above-ground biomass and relative above-ground biomass of bunchgrasses (Fig. 1a–f). The decrease in relative above-ground biomass of rhizome grasses was 20–84%, and the increase in relative above-ground biomass of bunchgrasses was 38–156% compared to the ungrazed plots. Grazing directly changed soil nutrients as shown by quadratic decreases in soil organic carbon and total soil nitrogen, linear decreases in extractable dissolved organic carbon and extractable dissolved organic nitrogen, and quadratic increases in the soil inorganic nitrogen (Fig. 1g–k). Compared to the ungrazed plot, soil organic carbon, total soil nitrogen, extractable dissolved organic carbon and extractable dissolved organic nitrogen were decreased by 27–46%, 26–44%, 21–75% and 8–44%, respectively, while the soil inorganic nitrogen was increased by 101–182%. Grazing directly altered soil environment as shown by decreases in soil pH and soil moisture and increases in soil bulk density and soil temperature (Fig. 1i–o).

Figure 1.

The response of variables to grazing intensity in categories of plant community, soil nutrients and soil environment. Abbreviations: NAB, nitrogen content of above-ground biomass; NBB, nitrogen content of below-ground biomass; RABR, relative above-ground biomass of rhizome grasses; RABB, relative above-ground biomass of bunchgrasses; TIN, total inorganic nitrogen; DOCex, extractable dissolved organic carbon; DONex, extractable dissolved organic nitrogen. For each response variable, regression was estimated using a linear or quadratic model with grazing intensity as a continuous predictor.

Soil nematodes and ecosystem functioning

During the grazing experiment, soil nematodes were representatives of 33 genera (see Appendix S3). Bacterial-feeding and fungal-feeding nematodes were the dominant nematode trophic groups (Fig. 2a–d). Compared to ungrazed plots, grazing increased the abundance of plant-feeding and fungal-feeding nematodes by 14–213% and 11–87%, respectively, while it decreased the abundance of omnivorous + carnivorous nematodes by 18–92%. Responses of variables in soil nematode community structure to grazing intensity are shown by decreases in the maturity index and species richness, with 0·10–14% and 4–24% decrease compared to ungrazed plots, respectively, while the parasite index or total abundance of nematodes did not show response to grazing intensity (Fig. 2e–h). Our analysis further revealed that, compared to ungrazed control, grazing had no significant effects on species richness and maturity index at low (1·5 and 3·0 sheep ha−1) and moderate (4·5 sheep ha−1) levels of intensity (> 0·05), and it significantly decreased both at high levels of intensity (6·0, 7·5 and 9·0 sheep ha−1; < 0·05). Similarly, grazing altered the ecosystem functioning by decreases in soil N mineralization and ANPP by 69–150% and 3–62%, respectively (Fig. 2i,j).

Figure 2.

The response of variables to grazing intensity in categories of soil nematode community components, soil nematode community structure and ecosystem functioning (i.e. soil N mineralization and ANPP). Abbreviations: PF, plant-feeding nematodes; FF, fungal-feeding nematode; BF, bacterial-feeding nematodes; OC, omnivorous + carnivorous nematodes; Atotal, total nematode abundance; NM, cumulative net N mineralization; ANPP, above-ground net primary productivity. For each response variable, regression was estimated using a linear or quadratic model with grazing intensity as a continuous predictor.

Pathways determining soil nematodes and ecosystem functioning

Most variables examined in this study were correlated with one another, making this data set well suited for SEM analysis (see Appendix S4). The two SEM models suggested that grazing directly altered plant community, soil nutrients and soil environment, based on the significant standardized path coefficients by grazing (Figs 3 and 4). For the SEM model of grazing effects on soil nematode community compositions (Fig. 3), we found that the plant community determined the abundance of plant-feeding and fungal-feeding nematodes, and soil environment determined the abundance of omnivorous + carnivorous nematodes. The abundance of bacterial-feeding nematodes was determined by soil nutrients, soil environment and their interaction. Grazing indirectly affected the abundance of omnivores + carnivores nematodes by the bottom-up effect of bacterial-feeding nematodes. For the SEM model of grazing effects on ecosystem functioning (Fig. 4), we found that the soil environment was the most important pathway in determining the soil nematode community structure. The soil N mineralization was mainly affected by plant community, while it was partly affected by soil nutrients, soil environment and nematode community structure. Soil environment and soil N mineralization were the most important pathways in determining the ANPP, while the pathway of plant community or soil nutrients has no significant relationship with the ANPP.

Figure 3.

Final model results of structural equation modelling (SEM) analysis for the grazing effects on soil nematode community components via pathways of plant community, soil nutrients and soil environment. Square boxes denote variables included in the models. Plant community variables include NAB, NBB, BB (below-ground biomass), RP (richness of plant community), RABR and RABB; soil nutrient variables include SOC (soil organic carbon), TSN (total soil nitrogen), TIN, DOCex and DONex; soil environment variables include SBD (soil bulk density), soil pH, SM (soil moisture) and ST (soil temperature); soil nematode community components include PF, FF, BF and OC. Abbreviations are as in Figs 1 and 2. The symbols ‘↑’ and ‘↓’ indicate a significant increase or decrease, respectively, in the response of the variables to grazing. The value within each box indicates the mean response (as a percentage) to grazing over different levels of intensity (i.e. 1·5, 3·0, 4·5, 6·0, 7·5 and 9·0 sheep ha−2), compared to the ungrazed control. Plant community, soil nutrients and soil environment are PC1 using principal component analysis (PCA). Results of model fitting: (χ= 14·82, = 0·538, d.f. = 16). Solid arrows denote the directions and effects that were significant (< 0·05); dashed arrows represent the directions and effects that were not significant (> 0·05). r2 values associated with response variables indicate the proportion of variation explained by relationships with other variables. Values associated with solid arrows represent standardized path coefficients.

Figure 4.

Final model results of SEM for the grazing effects on soil nematode community structure and linkages to ecosystem function via plant community, soil nutrients and soil environment. Square boxes denote variables included in the models. Plant community variables include NAB, NBB, BB, RP, RABR and RABB; soil nutrient variables include SOC, TSN, TIN, DOCex and DONex; soil environment variables include SBD, soil pH, SM and ST; soil nematode community components include PF, FF, BF and OC; and soil nematode community (NC) structure includes Atotal, RN, MI and PPI. The dashed square box denotes ecosystem functioning variables (i.e. NM and ANPP). The abbreviations are adopted from Figs 1, 2 and 3. The symbols are as in Fig. 3. The plant community, soil nutrients, soil environment and nematode community structure are PC1 using principal component analysis (PCA). Results of model fitting: (χ= 14·241, = 0·198, d.f. = 10).

Discussion

Our multi-level grazing intensity experiment presents robust evidence of the directions of grazing-induced changes in soil fauna community and ecosystem functioning to grazing in a semi-arid grassland. Our results support the idea that grazing has negative effects on below-ground diversity based on soil nematode community. There has been a substantial debate about whether or not herbivores can alter soil fauna community (Zolda 2006; Mikola et al. 2009). Our findings in the semi-arid grassland were corroborated by results from steppe grasslands in eastern Austria, which found that grazing had consistent negative effects on soil nematode diversity (Zolda 2006). However, grazing by cattle and rabbits in sub-humid grasslands (Veen et al. 2010) or by goats and deer in forest ecosystems (Wardle et al. 2001) did not influence below-ground diversity (e.g. nematode communities). This may be mainly attributed to differences in grazing intensity, which ranged from light to moderate in the sub-humid grasslands and forests. Our results indicate that there exists a critical threshold at grazing intensity of approximately 4·5 sheep ha−1, where grazing at levels below this threshold might have no significantly negative effect on nematode community diversity. Moreover, our results demonstrate that grazing altered the abundance of nematode trophic groups (i.e. plant-feeding, fungal-feeding and omnivorous nematodes), although previous research in grazing systems observed different trends to our results (McSorley & Tanner 2007; Mikola et al. 2009). For example, Mikola et al. (2009) found that the abundance of bacterial-feeding, plant-feeding and omnivorous nematodes did not respond to grazing; however, there was a positive influence of grazing on the abundance of fungal-feeding nematodes.

The influence of grazing by herbivores on ANPP also remains largely debated (McNaughton 1985; Milchunas & Lauenroth 1993; Pastor et al. 1993). Grazing effects on ANPP can range from positive (McNaughton 1985), to neutral (Milchunas & Lauenroth 1993), to negative (Pastor et al. 1993) in different ecosystems. Our result showed that grazing decreased not only ANPP but also root biomass, contradicting the idea of grazing optimization, which states that ANPP should be maximized at some moderate level of grazing. These seemingly contradictory results and conclusions could be caused by a variety of abiotic and biotic factors, including climatic and soil conditions (semi-arid vs. sub-humid grasslands), grazing animals (livestock vs. wild ungulates), grazing intensity and nutrient input and output of the system (Bardgett & Wardle 2003). For example, the density and intensity of livestock (e.g. sheep, cattle and goats) are generally higher than those of wild ungulates in temperate grasslands. Our study demonstrated that the continuous sheep grazing at high intensities shifted plant functional group composition, elevated soil temperature, diminished soil moisture, reduced biomass N content, and thereby inhibited the recovery of defoliated grasses and caused the decrease in soil fertility, organic matter and water-holding capacity. This implies that the effects of grazing on ANPP and/or below-ground biomass are strongly mediated by soil water and N availability in the semi-arid Inner Mongolia grassland.

The results from soil nematodes confirmed that top-down control by large grazing herbivores leads to control in the basal part of the soil food web. In general, lower-trophic-level organisms in soil food webs are supposed to be controlled by top-down forces, whereas higher-trophic-level organisms are controlled by bottom-up forces (Bardgett & Wardle 2010; Veen et al. 2010). The increasing abundance of plant-feeding nematodes induced by grazing was probably due to that grazing directly altered plant community component, such as the increase in relative above-ground biomass of bunchgrasses which provides high-quality root exudates. Because plant-feeding nematodes may be host specific, particular plant species could alter the abundance of specific plant-feeding nematode types (De Deyn et al. 2004; Viketoft 2008). The increasing abundance of fungal-feeding nematodes induced by grazing was probably due to the fact that grazing directly decreased the below-ground biomass and nitrogen content, hence may enhance the fungi decomposition chain and influences the soil microbial community (Stanton 1988; Wardle et al. 2004).

The grazing effect on bacterial-feeding nematodes was bottom-up controlled by their resources (e.g. soil organic carbon and total soil nitrogen for microbes) and soil environment (e.g. soil pH, soil moisture, temperature and soil bulk density). Similarly, omnivorous + carnivorous nematodes were modified by their resources (fugal-feeding or bacterial-feeding nematodes) and soil environment (e.g. soil bulk density, soil pH and soil moisture). Soil organic carbon and nitrogen provide energy sources, and these nutrients can then be used for increasing bacterial biomass, which is consumed by bacterial-feeding nematodes (Ingham et al. 1985; Szanser et al. 2011). Abiotic soil environment, for example, soil pH, soil moisture and temperature, are key factors determining the physiological conditions for microbial and soil nematodes (Korthals et al. 1996; Mulder, Van Wijnen & Van Wezel 2005). The substantial soil compaction was the most plausible explanation for the decreasing abundance of omnivorous + carnivorous nematodes. The tramping, which restricts soil pore volume and permeability to air, causes a decrease in pore space and creates uninhabitable space for larger soil nematodes (Bouwman & Arts 2000; Mikola et al. 2009; Sorensen et al. 2009). Similarly, abundance of omnivores + carnivores nematodes was partly bottom-up controlled by their resources (bacterial-feeding nematodes).

Our research, based on the long-term grazing experiment, is the first to our knowledge that examines the grazing effects of soil nutrients, soil environment and plant community on soil fauna and ecosystem functioning in a semi-arid grassland ecosystem. The SEM analysis revealed that grazing influenced nematode community structure by directly altering soil environment (e.g. soil pH, bulk density, and soil moisture and temperature) but not by plant community or soil nutrients. The richness and maturity index of soil nematodes, which indicate the response of nematode community structure to resource and environmental changes, were also significantly related with soil environment (Yeates & Bongers 1999). Grazing influenced soil N mineralization by altering plant community, soil nutrients and soil environment. Our findings also support previous results that soil environment properties, especially soil pH, soil moisture, soil temperature and soil bulk density, directly regulate the soil N mineralization rates in grassland (Bardgett & Wardle 2003; Augustine & McNaughton 2006; Shan et al. 2011). The soil nutrients, especially soil organic carbon and total soil nitrogen, provide energy sources for growth of soil micro-organisms, resulting in the release of inorganic nitrogen and other essential nutrients by the soil micro-organisms (Ingham et al. 1985). The influence of grazing on plant community, for example, changes in plant species replacement, plant quantity and quality, and plant C and N allocation, can therefore be important in driving the below-ground soil N mineralization process (Bardgett & Wardle 2003).

The present research revealed that grazing influenced ANPP by altering soil N mineralization and soil environment (e.g. soil moisture, bulk density), while plant community, soil nutrients and nematode community structure only showed indirect influences on ANPP via altering soil N mineralization. It has been proposed that water and nitrogen are two of the most important factors controlling above-ground net primary production (ANPP) in grassland ecosystems (Chapin, Matson, & Vitousek 2011). This is particularly true in semi-arid grasslands where water and nitrogen both affect nutrient cycling and ANPP (Le Houérou, Bingham & Skerbek 1988; Bai et al. 2004, 2010). This indicates that grazing at high intensities, which diminished the soil net N mineralization, further intensified N limitation of ANPP.

It is important to note that some uncertainties exist on the ecological linkages between structure and functioning components due to the fact that plant and soil are responsive to climatic variations. For example, the intra- and inter-annual variations in precipitation and temperature could affect the strength of the linkages. It has been documented that the intra- and inter-annual variations in precipitation are key climatic factors controlling fluctuations in plant functional group composition and ANPP in semi-arid grasslands (Bai 1999; Bai et al. 2004). A previous study in the same grazing experiment also found that the directions of the effects of soil temperature and moisture on net N mineralization differed and they changed over seasons (Shan et al. 2011). Therefore, more research is needed to understand how grazing intensity affects the linkages between structure and functioning components under intra- and inter-annual variations in climatic conditions (e.g. dry vs. wet years). Our study, using the SEM analysis, provides new insights into the mechanisms and pathways of grazing impacts on ecosystem structure and functioning in the semi-arid grassland.

The semi-arid grassland ecosystems on the Mongolia plateau have been overused subjected to continuous grazing at high levels of stocking rates. Moreover, these grasslands are poorly managed, led to a widespread decline in ecosystem functioning in recent decades (Tong et al. 2004; Jiang, Han & Wu 2006). Our results show that grazing herbivores can change community composition, structure, diversity of plant and soil communities, followed by a decrease in soil fertility, water-holding capacity, organic matter and ANPP, thereby causing a decline in the capacity for grassland ecosystems to provide goods and services. The extent to which these changes occur may depend not only on the grazing intensity but also on the pattern of their grazing. Lowering grazing intensity and allowing for human herding of herbivores in the Inner Mongolia grassland could prove useful for recovering the degraded semi-arid grasslands. Fertilization or sowing legumes could increase soil nitrogen and nutrients, further increasing forage production, vegetation cover, water-holding capacity and ultimately the goods and services of ecosystems. Our findings have important implications for better understanding the effects of grazing on ecosystem services of the Inner Mongolia grassland, especially because it has been predicted that future climatic change will increase extreme droughts and intense rainfall events.

Acknowledgements

We thank Howard Ferris, G. F. Veen, Qingmin Pan, Yuanhu Shao, Jiaoyan Ying, Hong Wang and Hongwei Wan for their constructive comments on an earlier version of this paper, and James B. Grace for his help in structure equation modelling analysis. This work was supported by the State Key Basic Research Development Program of China (2009CB421102) and the Natural Science Foundation of China (30825008, 31030013 and 31100335).

Ancillary