Responses of community structure and diversity to nitrogen deposition and rainfall addition in contrasting steppes are ecosystem-dependent and dwarfed by year-to-year community dynamics.

BACKGROUND AND AIMS
Long-term studies to disentangle the multiple, simultaneous effects of global change on community dynamics are a high research priority to forecast future distribution of diversity. Seldom are such multiple effects of global change studied across different ecosystems.


METHODS
Here we manipulated N deposition and rainfall at levels realistic for future environmental scenarios in three contrasting steppe types in Mongolia and followed community dynamics for seven years.


KEY RESULTS
Redundancy analyses showed that community composition varied significantly among years. Rainfall and nitrogen manipulations did have some significant effects, but these effects were dependent on the type of response and varied between ecosystems. Community compositions of desert and meadow steppes, but not that of typical steppe, responded significantly to rainfall addition. Only community composition of meadow steppe responded significantly to nitrogen deposition. Species richness in desert steppe responded significantly to rainfall addition, but the other two steppes did not. Typical steppe showed significant negative response of species richness to nitrogen deposition, but the other two steppes did not. There were significant interactions between year and nitrogen deposition in desert steppe and between year and rainfall addition in typical steppe, suggesting that the effect of the treatments depend on the particular year considered.


CONCLUSIONS
Our multi-year experiment thus suggests that responses of community structure and diversity to global change drivers are ecosystem-dependent and that their responses to experimental treatments are dwarfed by the year-to-year community dynamics. Therefore, our results point to the importance of taking annual environmental variability into account for understanding and predicting the specific responses of different ecosystems to multiple global change drivers.


INTRODUCTION
Biodiversity is not only inherently precious, it also bolsters ecosystem functioning and reliable provisioning of ecosystem services on Earth Hautier et al., 2014). Global changes in climate and high nitrogen deposition levels due to human activity provoke biodiversity loss (Sala et al., 2000). To minimize biodiversity loss, there is an urgent need for understanding the response of biodiversity in contrasting ecosystems to global change (Stevens et al., 2004;Sullivan et al., 2016). Owing to the multi-faceted nature of global change drivers (Reich et al., 2001), forecasting community dynamics and distribution of diversity still remains an important challenge.
Intensive agriculture and fossil fuel combustion have increased nitrogen deposition globally (Stevens et al., 2004). Anthropogenic increases in nitrogen inputs increase above-ground biomass and subsequently reduce light, which changes the community through above-ground competition (Suding et al., 2005;Avolio et al., 2014;Hautier et al., 2014;Sun et al., 2016). Therefore, nitrogen deposition generally causes a decline in plant species diversity because of the asymmetrical competition between species, i.e. rare and slow-growing species being lost due to competition from fast-growing and weedy species. Thus, nitrogen deposition is predicted to be a major driver of global diversity loss (Hautier et al., 2009;Eisenhauer et al., 2013;Payne et al., 2013;Calvo-Fernández et al., 2018).
In addition, current and projected climate change will lead to altered rainfall patterns with a larger interannual variability and rainfall seasonality and thus change the water balance of terrestrial ecosystems (IPCC, 2012). The altered precipitation is expected to influence composition and diversity of natural communities by affecting plant growth and interspecific interactions, especially in arid and semiarid ecosystems (Loarie et al., 2009;Prevéy and Seastedt, 2014;Yu et al., 2017). Ecological theory suggests that plant community biomass and composition in some ecosystems may be jointly controlled by water availability and soil nutrient supply (Harpole et al., 2011;Eskelinen and Harrison, 2015). Alongside absolute changes in water availability and nutrient supply associated with global change, interannual climate variability may in itself impact plant community dynamics Collins et al., 2017). Therefore, nitrogen deposition and rainfall alteration will concurrently affect plant communities (Smith et al., 2016), yet to our knowledge few studies have simultaneously determined the effects of multiple global change drivers across different ecosystems and over longer time periods with high interannual climate variability (Tielbörger et al., 2014).
Grasslands cover more than a quarter of the total ice-free land area globally and provide ecosystem goods and services at a global scale (Gibson, 2009). Temperate grasslands are often both water-and nitrogen-limited, thus nitrogen deposition and altered precipitation are expected to have strong impacts on plant community composition and diversity (Morgan et al., 2007;Grime et al., 2008;Avolio et al., 2014). However, the biotic and abiotic factors driving community responses to global change have been difficult to determine, partly because of interactions among abiotic drivers and overall differences between different ecosystems and their abiotic settings (Hautier et al., 2009;Harrison et al., 2015). As a result, responses to global change can be community-specific (Reich et al., 2001;Bai et al., 2010), which might result from four community properties: previous exposure to climatic extremes, species richness, functional composition and successional status (Grime et al., 2000); and from current climate. Therefore, we need studies that simultaneously implement the same experimental environmental manipulation protocols in multiple communities to detect commonalities or differences in global change response (Luo et al., 2011;Tielbörger et al., 2014).
Here we tested for community composition and diversity in response to nitrogen deposition and rainfall addition in three ecosystems in Mongolia varying strongly in macroclimate (interannual temperature and rainfall variability): desert steppe, typical steppe and meadow steppe. We followed community dynamics for 7 years to test two hypotheses: (1) due to great differences in biotic and abiotic settings, community compositions of contrasting ecosystems vary in their response to rainfall and nitrogen manipulations (for example, being the driest steppe, desert steppe would response more significantly to rainfall addition than the other two steppes); (2) the ecosystemdependent responses to rainfall and nitrogen manipulations are also reflected in the diverse responses of species richness in the three steppes.

Experimental design
Desert, meadow and typical steppes represented the most typical vegetation types in the Mongolian steppe. We selected the three steppes in the east of Ulaanbaatar, Mongolia [see Liu et al. (2013) for the locations and Fig. 1A-C for photographs of the selected steppes]. The desert steppe was dominated by Allium polyrrhizum (perennial forb), Carex duriuscula (perennial sedge) and Stipa gobica (perennial grass). The dominant species in the meadow steppe were Potentilla acaulis (perennial forb), Festuca lenensis (perennial grass) and Carex pediformis (perennial sedge), while those in the typical steppe were Salsola collina (annual forb), Stipa grandis (perennial grass) and Artemisia palustris (annual forb). Annual average temperature was lowest in the meadow steppe and highest in the typical steppe (Fig. 1D). Desert steppe is the driest among the three vegetation types, and annual precipitation was broadly similar in meadow and typical steppes (Fig. 1E). Temperature and precipitation fluctuated strongly among years in the three steppes (Fig. 1D, E).
In each of the three selected steppes, three 14 m × 34 m blocks ~100 m apart were established on a gentle slope and fenced. Each block was divided into 36 quadrats (2 m × 2 m, 2 m from each other). Within each block, we applied full combinations of rainfall and nitrogen addition treatments. Nitrogen was applied at three levels of 0 (control), 0.44 and 0.88 g N m −2 year −1 [using 5 and 10 g ammonium nitrate (NH 4 NO 3 ) per 2 m × 2 m quadrat], and the added nitrogen was evenly broadcast by hand onto the vegetation. We added low nitrogen because the nitrogen deposition rate was low in this region (0.1-0.5 g N m −2 year −1 ; Dentener et al., 2006;Galloway et al., 2008). Rainfall addition also included three levels (ambient, +1/7 and +2/7 of ambient rainfall) to mimic realistic scenarios for an overall increase in precipitation in this region (Sato et al., 2007). Rainfall manipulation here was identical to an experiment in China [see Ye et al. (2017) for details]. Briefly, precipitation addition was done through the transfer of a certain area (equivalent to 1/7 or 2/7 of the experimental area) of intercepted precipitation. Tin troughs collected precipitation just outside each quadrat, and the area of the trough was related to the amount of collected precipitation. The collected precipitation drained in real time into the quadrat through a drip irrigation tube. The combination of nitrogen addition and rainfall manipulation resulted in nine treatments, which were randomly assigned to the quadrats and replicated four times within each block. Thus, there were 12 replicates in total per site (four per block and three blocks per steppe). The nitrogen and rainfall treatments were initiated in late July 2009 after the first plant community measurement (Year 0, see below). In subsequent years, nitrogen addition was always applied once after plant community measurement.

Response measurements
From 2009 (Year 0) to 2016 (7 years after treatments), we measured the plant community at peak biomass time (late July) each year. At each sampling time, we recorded species identity and density of each species in the central 1 m × 1 m area of each quadrat. Plant density was measured by counting the number of individuals (clumps of bunch grasses or clonal graminoids) per species per unit of surface area. For each steppe, density in all years was measured by the same specialist to minimize observer biases. In total, we collected 46 016 records of density by species data in eight consecutive years.
All analyses used R version 3.4.1 (R Core Team, 2017). Multivariate constrained ordination (redundancy analysis, RDA) was used to analyse community composition, i.e. changes in the abundances of all single species in our steppes under rainfall and nitrogen manipulations. RDA is a commonly used statistical technique to test clear hypotheses for multivariate data sets and has been used effectively to detect plant community change in response to global change (Tielbörger et al., 2014). The species abundance data of the four quadrats per block were pooled and square-roottransformed before RDA. RDA was done separately for each steppe using year (here a continuous predictor), rainfall addition, nitrogen deposition and their interactions as constrained factors, and block identity as covariates to control for potential differences between blocks before the onset of treatments. Monte Carlo tests were run with 999 permutations adjusted to the model structure. For visual illustration of the results of multivariate analyses, a non-metric multidimensional scaling (NMDS) analysis was performed on a Bray-Curtis dissimilarity matrix of total summed abundances within each treatment per steppe. Euclidean distances of the first two axes were calculated between NMDS values and averaged to display differences in community composition. Multivariate analyses were performed using the R package vegan (Oksanen et al., 2012). The univariate data and species richness [ln(x + 1)-transformed] were analysed by fitting linear mixed models (LMMs) for each steppe separately. We compared the data distribution before analyses. Shapiro-Wilk normality tests showed that species richness [ln(x + 1)-transformed] in two steppes did not deviate from normal distribution (meadow, W = 0.995, P = 0.76; typical, W = 0.991, P = 0.21), while desert steppe did deviate (W = 0.944, P < 0.001). However, none of the three steppes had species richness data that had a Poisson distribution (P < 0.001). Therefore, for consistency we modelled species richness using log(x + 1)-transformed data for all three steppes. The LMMs included rainfall addition, nitrogen deposition and year (here categorical) as fixed factors. Block intercepts were included as random error variables to account for our replicated measurements over time. The LMMs included all main effects and two-way and threeway interactions. In addition, some autocorrelation between sample years was identified, and an autoregression function of lag 2 (years) provided the best-fitting models and thus was included in all our LMMs. Furthermore, temperature and precipitation in the study years could have had an effect on community parameters and should have been included in the models. However, including temperature and precipitation in the LMMs had a larger Akaike information criterion than those described purely by categorical years in all three steppes for species richness (desert, 358.70 versus −1.22; meadow, 217.4 versus −130.66; typical, 367.07 versus −43.78). This was possibly because we had included year as a categorical variable, thereby not making any assumptions on precisely which (combination) of several possible environmental variables were causing the between-year variation in community parameters (Tielbörger et al., 2014). Therefore, we based our conclusions on the original and most powerful methods presented. LMMs were conducted in R packages nlme (Pinheiro et al., 2012) and MASS (Venables and Ripley, 2002).

Response of community composition to experimental manipulations
Redundancy analysis showed that community composition varied significantly and strongly among years and ecosystems (Table 1; Supplementary Data Fig. S1). Community compositions of desert and meadow steppes, but not that of typical steppe, responded significantly to rainfall addition. Only community composition of meadow steppe responded significantly to nitrogen deposition. None of the interactions was significant except that the interaction between rainfall and nitrogen manipulations showed a synergistic effect in meadow steppe ( Table 1). The NMDS confirmed that annual differences in community composition were much larger than the effects caused by the rainfall and nitrogen treatments, and trajectories of plant community across rainfall and nitrogen treatments were largely parallel, i.e. the treatments did not affect the temporal development of plant communities (Fig. 2).

Response of species richness to experimental manipulations
There were in total 169 species in the three steppes (Supplementary Data Table S1). The LMM showed that species richness in all three steppes was highly dynamic among years (Table 2; Fig. 3). Rainfall addition increased species richness in desert steppe, but not in the other two steppes (  3A). Only typical steppe showed a significant negative response in species richness to nitrogen deposition (Fig. 3F). None of the interactions was significant except that the interactions between year and nitrogen deposition amplified the year-to-year variation in desert steppe and that between year and rainfall addition buffered the year-to-year variation in typical steppe (Table 2), but these two significant interactions varied among years (Fig. 3A, F).

DISCUSSION
Ecosystems with different structures and functions are likely to respond differently to global change (Reich et al., 2001;Bai et al., 2010); thus, global change studies across multiple ecosystems are needed. Our selected three vegetation types, representative of contrasting climate zones, differed in overall species richness (Fig. 3), indicating they have different community structure. These differences in community structure between the three contrasting steppes may explain the differences in community composition change during our experiment ( Fig. 2; Grime et al., 2000). Our results support our first hypothesis, showing that the responses of plant community composition to global change drivers are ecosystem-dependent (Table 1)  and Allen-Diaz (1999) suggested that relatively moist steppes were more likely to be in equilibrium, while non-equilibrium dynamics should occur in steppes with <250 mm mean annual precipitation. Stumpp et al. (2005) reported a prevailing non-equilibrium situation for dry montane steppes in southern Mongolia. Therefore, in contrasting steppes such as ours, ecosystem equilibrium could be different and thereby determine the diverse responses among the steppes. Second, biotic interactions could be important in determining the responses. We found that meadow steppe was more responsive to both rainfall addition and nitrogen deposition than the other two steppes (Table 1). Liancourt et al. (2013) reported that intense competition existed in the Mongolian meadow steppe. Therefore, rainfall addition and nitrogen deposition might change community composition via altering biotic interactions, in particular enhancing competition, in this steppe. In addition, being the most water-limited ecosystem, community composition of desert steppe responded to rainfall addition (Table 1; Fig. 2A).
Our result is consistent with a study in in the northern Mongolia steppe showing that a roughly 25% increase in growing season precipitation causes rapid compositional changes (Spence et al., 2016). Taken together with the results discussed above, our study clearly demonstrates that the three ecosystems respond differently to rainfall and nitrogen manipulations owing to diverse mechanisms operating in the three steppes. Because changes in community composition indicate shifts in species composition and their relative contributions to the plant community (Morgan et al., 2007), we further determined the response of species richness to global change drivers. We found that rainfall addition only increased species richness in desert steppe (Table 2; Fig. 3A), confirming again that water is limiting in this ecosystem and this allows only the most droughttolerant species to occur. However, species richness in the other two steppes was not responsive to rainfall addition, which is consistent with some studies in other grasslands reporting that supplying water produces little or no change in species diversity (Grime et al., 2008;Holub et al., 2012). One possibility for the lack of response is that experimental rainfall addition intensifies existing rainfall events, but may not alleviate the drought stress caused by infrequent events. The different responses in species richness found in our study support our second hypothesis and are consistent with some studies in other grasslands. For example, increased variability in seasonal rainfall increased species diversity in tallgrass prairie (Fay et al., 2003) but not in a semiarid shortgrass steppe (Heisler-White et al., 2008).
The addition of nitrogen is generally accompanied by a loss of species richness as a result of the heterogeneous response to nitrogen among species and interspecific competition (Suding et al., 2005;Hautier et al., 2014;Sun et al., 2016). However, a significant reduction in species richness was only found in typical steppe (Table 2; Fig. 3F). The mechanism for the lack of response in desert steppe may differ from that in meadow steppe. For desert steppe, it may be because the vegetation is sparse and therefore interspecific competition is low, but for meadow steppe it is possibly due to the non-existence of nitrogen limitation in the steppe and/or the weak experimental treatments of nitrogen deposition (i.e. stronger treatments could have produced a different result). However, our study cannot distinguish between the two possibilities because we did not have the data on soil moisture and nutrient availability in the soil, in the plants or both.
Our results showed that the response of the three steppes to experimental manipulations was generally weaker than to interannual variation (Table 2). Our findings are consistent with a study in a moderately dry steppe of Mongolia, which showed that the addition of a low amount of nitrogen to simulate nitrogen deposition around the study area by the year 2050 did not significantly increase density in the grassland (Kinugasa et al., 2012). Our results are also consistent with several previous reports from grasslands in USA. For example, Zavaleta et al. (2003) reported that responses of the ten most common plant species in a Californian grassland to global change drivers were weak. Collins et al. (2017) found that the plant community structure of a desert grassland in central New Mexico was highly resistant to global change drivers (warming, nitrogen deposition, altered winter precipitation) prior to a wildfire. The lack of response to nitrogen deposition and rainfall addition is possibly due to fine-scale heterogeneity in environmental conditions, allowing intraspecific genotype selection instead of changes in mean abundance of species (Tielbörger et al., 2014) and/or nitrogen deposition and rainfall addition may drive change on longer time-frames (which were not measured), whereas interannual variation may be detected in community composition over the several years the study was conducted.
Furthermore, our results confirm that arid ecosystems are highly responsive to interannual variation in climate (Huxman et al., 2004;Collins et al., 2017). Community composition and species richness in all three steppes exhibited high annual dynamics (Figs 2 and 3), suggesting that interannual changes in biotic or abiotic factors may be very important in determining community dynamics in our steppes. The high interannual variation in environmental conditions, which is reflected in highly dynamic community structure, may have far more influence than the experimentally manipulated climate change. Therefore, the predicted changes in climatic conditions may lie well within the 'climatic comfort zone' to which the component species are adapted, resulting in the resistance of community structure and diversity to experimental manipulations (Tielbörger et al., 2014).
In conclusion, our multi-year experiment on multiple global change drivers across contrasting ecosystems demonstrates that responses of community structure and composition to global change drivers are ecosystem-dependent. The ecosystemdependent responses are not only reflected in different components of community structure but also in their differential responses to different global change drivers. Overall, the responses of community structure and diversity to experimental global change treatments were dwarfed by their year-to-year dynamics. Therefore, our results point to the importance of taking annual environmental variability into account for understanding and modelling the specific responses of different ecosystems to multiple global change drivers. This finding is also highly relevant to ecosystems in other biomes with strong interannual variation in environmental conditions.

SUPPLEMENTARY DATA
Supplementary data are available online at https://academic. oup.com/aob and consist of the following. Figure S1: RDA biplots showing the effect of year, nitrogen and water manipulations on community composition in the three study steppes. Table S1: species list for the three steppes.