Conflict of interest
The authors have no competing interests.
Abstract: Nitrogen enrichment affects and changes the community structure of terrestrial ecosystems. Recently, intransitivity competition networks have been widely considered to in maintaining coexistence and biodiversity. However, whether the structure and function of the species competition network changes under nitrogen enrichment remains unclear. In this paper, we applied the Apriori algorithm to find dominant species assemblies to simulate competition matrices. We found that (1) intransitive competition networks derived by a species-dominant assembly accounted for a relatively large proportion; (2)Under N enrichment, the complexity of the networks varied from an intransitive network to a sub-competition structure, and transitive networks was a transition structure between these two structures; (3) the negative effect of N-enrichment on species network was counteracted by mowing and rainfall; and (4) increasing the number of loops enhances the diversity in a community, while high competition among species is a representation of an intransitive competition network.

1 Introduction

The proportion of nitrogen (N) in nature is relatively stable, but the process of sequestering N from the atmosphere into the biosphere is accelerated by excessive emission (Vitousek etal . 1997; Erisman et al . 2013). Significant changes in N cycles (Canfield et al . 2010) have a negative impact on richness at both local and spatial scales (Chalcraft et al . 2008; Lan et al . 2015). Competition is an inherent driver of species richness and symbiosis (Stevens et al . 2004; Bobbink et al . 2010), so as understanding competition among species is important for comprehending community structure and symbiosis. However, the dynamic effects of environmental factors are still poorly understood.
Competition can be an interaction of pairwise species or indirect effects among species in a group (Levine 1976; Wootton 1994), and these species and their effects constitute a species-competition network (SCN). Competition is generally divided into two categories: transitive (hierarchical) competition and intransitive competition (Solivereset al . 2018). Intransitivity is figuratively similar to the rock-paper-scissors game (such as A>B>C>A), and there is no best competitor and worst loser (Gilpin 1975; May & Leonard 1975). The theoretical hypothesis of intransitivity (Gilpin 1975) has received increasing attention from ecologists and greatly promoted “modern coexistence theory” (Soliveres et al . 2011; Ulrich et al . 2014). Perfect transitivity has distinctly hierarchical competition. Plant competition has always been considered to be a clear hierarchy, among which superior species will eventually win over inferior species (Tilman 1988; Paul et al . 1989). However, the fact that inferior species are always found in sample plots indicates that this theory is still open to discussion (Allesina et al . 2011).
Competition is not strictly hierarchical (Aarssen et al . 1992); in particular, outcomes of coexistence after environmental disturbances are unpredictable (Carsten 2007). Intransitivity can also occur frequently in plant communities (Mayfield & Stouffer 2017; Solivereset al . 2018) and promote the coexistence of species (Huismanet al . 2001; Reichenbach et al . 2007; Laird & Schamp 2009). The intransitivity level might be influenced by environmental heterogeneity (Allesina & Levine 2011) and successional stages (Worm & Karez 2002).
Intransitivity has important effects on the stability of species coexistence on an ecological time scale (Landi et al . 2018). Many architectural patterns found in real food webs tend to enhance community stability and species coexistence (Thebault et al . 2010). Theoretically, if a loop contains an odd number of species, it can induce population growth and promote species coexistence (Allesina & Levine 2011; Huisman & Weissing 1999). If a loop includes an even number of species, minor disturbances will be amplified by the cycle and lead to coexistence instability and even local extinction of one or more species (Gallien et al . 2017).
In addition, climate variation may change the temporal heterogeneity of resources and further affect intransitivity (Soliveres et al . 2015), especially in arid areas, where water availability tends to be more volatile than in humid areas (Whitford 2002). Under N enrichment, the effect of climate on richness cannot be generalized but may depend on the indirect effect of intransitive competition. However, few studies have examined the effects of N enrichment and climate factors on SCN structures and their ecological functions.
Relative biomass is a direct manifestation of differences in interspecific competitivity (Meserve et al . 1996; Levine & Rees 2002). The assembly rule and biomass are the final manifestations of interactive mechanisms, but the assembly mode and species number in assemblies are still under discussion. Based on the hypothesis that dominant species were more likely to produce intransitivity, the strategy of a fixed number of species for assembly (Soliveres et al . 2015; Ulrich et al . 2018) is limited to superior species, which led to a relatively small proportion of the inferred intransitivity and ignored the fact that inferior species can survive in long-term fierce competition (Fox 2013).
Here, we tested the following hypotheses: (1) species with small total biomass (inferior species) but high-frequency occurrence in an assembly play a crucial role; (2) under different N addition rates, SCNs may vary from intransitive competition networks (ICNs) to transitive competition networks (TCNs) and finally degenerate to a sub-competitive structure (SCS) dominated by environmental factors; (3) precipitation and mowing can alleviate the negative effect of heavy N-enrichment and increase the probability of formation of intransitive structures embedded in SCNs; and (4) an ICN is favourable to maintain richness, while a TCN is maintained for a longer period during grassland degradation.

2 Materials and methods

2.1 Experimental sites and sampling

Field experiments were carried out on a temperate steppe (43°32’51”N, 116°40’23”E) in Xilin, Inner Mongolia Autonomous Region, China. From 2008 to 2020, average annual temperature was 1.45°. The average annual precipitation was 329.9 mm. This simulated N-deposition experiment began in September 2008 following a randomized complete block design. Each plot was 8 m× 8 m, and sampling was carried out in the same 0.5×2 m2 area. Purified NH4NO3 (>99%) was added to plots at nine different rates (0, 1, 2, 3, 5, 10, 15, 20, and 50 g N m-2 year-1, of which concentrations of 0 to 10 g N m-2year-1 were considered low nitrogen addition and concentrations of 15 to 50 g N m-2year-1 were considered high nitrogen addition) at two frequencies (monthly and twice a year). In the mowing groups, no-mowing groups, and two control groups, there were 38 experimental treatments, each of which was replicated 10 times. A total of 4,930 plots of field samples were collected over 13 years. These data were used to mine the SDA of above-ground species and construct SCNs.

2.2 Methods

Interactions, such as competition or mutualism, are often considered to be essential drivers of ecosystem services, but directly measuring them is challenging (Olff et al . 2009). Interaction is usually measured by the co-occurrence number (Wang et al. 2020) or correlation coefficient (Feng et al . 2017) based on repeated experimental data. Repeated experiments play an important role in obtaining reliable indirect metrics of competitivity determined from field experimental data. The Apriori algorithm was used to mine SDAs from species presence-absence data, and then the RECC model (Ulrichet al . 2014) was used to simulate SCNs of the SDAs with abundance data. The complexity and structure of the SCNs, centrality of important nodes, their ecological functions, their variation with biotic or abiotic factor changes, and their significance were explored, as shown in the schematic experimental diagram of this study (Fig. 1).

2.2.1 Excavation of species-dominant assembly

Although spatial heterogeneity in adjacent areas may increase opportunities for niche differentiation and enhance intransitive competition, it also makes it more complex and difficult to estimate levels of competition based on the observed abundance of species (Ulrichet al . 2014). Therefore, sample plots with matching species assembly should be as environmentally homogeneous as possible. N enrichment can reduce environmental heterogeneity in small communities (Western 2001; Fraterrigo et al . 2005), and a similar environment is likely to promote similar community dynamics (Western 2001; Wescheet al . 2012), reducing spatial asynchrony. Therefore, we believe that under N enrichment, SDA can occur with high probability in multiple sample plots. For this reason, we used the Apriori algorithm to find SDAs (Wang et al . 2020).
The Apriori algorithm is an unsupervised learning algorithm that can mine species assemblies frequently occurring in 10 repeated experimental plots. When the support degree exceeds the threshold, a species assembly with a high occurrence frequency is considered a SDA (Agrawal 1994; Wanget al . 2020). Species absent from a plot will be excluded from the SDA, so it will naturally be out of the numerical simulation. We set 0.5 as the threshold for SDA selection, ensuring that the estimation of species competition was based on at least 5 plots. Species numbers in some SDAs can be up to 9. Considering the efficiency of calculation and Apriori algorithm theory, an SDA comprising the largest number of species was selected as the numerical base.

2.2.2 Measurement of intransitivity

The RECC model proposed by Ulrich et al. (2014) was used to simulate the competition matrix of SDAs. Many competition matrices C (100,000, m×m matrices, where m is a species number) were generated then transformed into transfer matrix P by the colonization competition model. Since the abundance orders ofPU and U are completely positively correlated (r2=1, P<0.05), the Spearman test was performed between PU and U . The average rank correlation coefficient (rs) was used as an index to select the optimal matrices C and P .
The rs is mainly used to measure the relative essentiality between interspecific competition mechanisms and other factors acting on communities. In the field experimental data, the value is usually low, indicating that interspecific competition mechanisms play a relatively small role, possibly due to environmental heterogeneity (Dufour et al . 2006) or other more important mechanisms (Ulrich et al . 2018). Therefore, if rs≥ 0.6, competition is thought to occur, and C is a competition matrix (or optimal matrix).
If competitions among a group are completely transitive in an SCN, there will be a species whose competitivity is significantly higher than that of others or that will rank first in C , suggesting that the species will eventually outcompete the others. An intransitive loop marked by matrix C indicates that there is an intransitive competition structure derived from an SDA. According to the level of complexity, three types of SDNs are defined as follows (Fig. S1).
  1. An ICN (or loop network) is an SCN containing at least one intransitive structure.
  2. A TCN (or chain network) is an SCN with a perfectly hierarchical competition structure rather than comprising any loop structure.
  3. The SCS is a community (in a plot) in which rs is less than 0.6 and might be dominated by environmental factors such as N enrichment. How to vividly describe its structure needs further research.

2.2.3 Complexity measurement of network structure

Luo and Magee (2011) proposed flow hierarchy (simplified as h) , which was defined as the percentage of edge number in chain paths to the total number of all edges in links maintaining their overall direction, that is, the percentage of edges not in any loop paths to total edges (Luo & Magee 2011). If the network is a pure structure, such as a tree structure, then h =1. Therefore, the complexity of an SCN was redefined as follows,
\begin{equation} C\text{om}plexity=\left\{\begin{matrix}0,if\frac{h*c}{N_{\text{node}}}\geq 1\\ 1-\frac{h*c}{N_{\text{node}}},if\ 0\leq\frac{h*c}{N_{\text{node}}}<1\\ \end{matrix}\right.\ \nonumber \\ \end{equation}
where h is the measure of the flow hierarchy,\(N_{\text{node}}\)is the species number of the network, and c is a constant. The numerical experiment proves that c = 10 is optimal for complexity to the interval [0,1].

2.2.4 Betweenness centrality

The betweenness centrality(BC) of a node refers to the number of shortest paths through the node in a network (Linton 1977) and is used to estimate its importance. A node with high BC can effectively raise the complexity boundary of a network. If the node was removed from the network, the structure would be segmented into several disconnected subnetworks.
In this paper, The Apriori Algorithm in the mlxtend package in Python was used to mine SDAs. The NetworkX package in Python was applied to construct SCN diagrams and calculate the complexity of networks and BC of each species.

3 Results

3.1 Statistical description of species competition networks

In this paper, plants in the experimental plots were structuralized and divided into three structure types: ICNs, TCNs and SCSs. Environmental changes were the main inducer of transformation in competitive networks. At N = 0 g N m-2 year-1, 95.84% of the plots were determined by interspecific competition. With the increase in N addition rates from 0 to 50 g N m-2 year-1, the proportion of the plots dominated by competition decreased by approximately half to 44.23%. When N = 50 g N m-2 year-1, the proportion of ICN decreased to 9.61% from 51.92%, while the SCS eventually increased to 50%. The variation in the TCN proportion was relatively stable because the proportion of TCNs degenerating to the SCS might be supplemented by the proportion of ICNs degenerating to TCNs. With increasing N addition rate, some ICNs transformed into TCNs, while some TCNs transformed into SCSs, which basically maintained the distribution of TCNs in a dynamic balance (Fig. S2). The proportion of ICNs under mowing was significantly higher than that without mowing. The high and low frequencies of N addition had no significant effect on the distribution of the three structures.
Species(18 kinds) in SCNs were divided into three kinds species based on total biomass: inferior species(< 500 g)、sub-superior species(500≤total biomass ≤10000 g)、superior species(> 10000 g)(Tab.1). Frequent two-species and three-species assemblies were also mainly composed of superior and sub-superior species (Tab. S1&S2). Inferior species usually more easily survive in ICNs and rarely in TCNs, although they are infrequently in SCNs (Tab. 1). A three-species intransitive structure is considered the minimal loop. There were 168 different minimal loops in all SCNs. Among all the minimal loops, there were different 94 pairs and 17 single species.
BC emphasizes the important role of a species in controlling the structural complexity of an SCN. Once the species with high BC were eliminated from an SCN, the SCN was broken into several disconnected small networks. There was a significantly positive correlation between the complexities of SCNs and the maximal BC of nodes in themselves (r2=0.772, P<0.001 in Fig. S3). The inferior species had fewer opportunities to appear in an SCN, and once they appeared in an SCN, the opportunity to play an intermediary role increased. Therefore, these species played a significant bridging role in their SCNs (Tab. 1).

3.2 Effects of species competition network on community attributes

Species richness were similar in the control environment. Here were significant differences in inter-annual variation trends of species diversity, especially at high nitrogen (Tab. S4). The lower the N addition rate was, the greater the species richness (Fig. 2). Their polygonal lines ran in a compact way. However, for TCNs, the increasing richness under each N addition rate was not obvious, and their polygonal lines ran in a dispersed way. There were more plots labelled by TCNs than by ICNs under high N concentrations (N ≥ 15 g N m-2 year-1), while ICNs at low nitrogen had better maintaining biodiversity. The N-cumulative effect accelerated the decomposition of TCNs and increased the randomness of species assembly (Steiner & Leibold 2004; Chase & J.M. 2010). Eventually, this destroyed species co-occurrence and competitive structure, resulting in the interspecific ecological function of the community being entirely lost. Especially in 2020, the biodiversity was not only maintained but also promoted, but that of TCNs had the opposite trend.
Biomass and the number of RP are two essential attributes portraying the plant community and also directly embody interspecific competition. We found that network structure affected the attributes of communities and their associations in a sample plot. In the plots identified as an ICN, there was no significant correlation between the average biomass and number of RPs (r2=0.116). While for the plots identified as a TCN, there was a significant positive correlation (r2= 0.482***). Interspecific competition in TCNs is not only for biomass but also for the number of RPs, indicating that the network was strong overall.
With regard to ICNs, 49.36% of samples in the 95% confidence region were the plots under low N addition rates and mowing (Fig. 3). A low N addition rate is beneficial to the number of RPs under intransitive competition, and mowing is beneficial to biomass. The 95% confidence region for TCNs was dominated by a low N addition rate and no mowing, accounting for 40.10% of the 95% confidence region. The sample plots outside the confidence region were mostly those treated with high N addition rates and mowing (Fig. 3). In conclusion, there was more ICNs than TCNs at low N addition rates, while more TCNs at high nitrogen had a strong correlation between biomass and the number of RPs. The area outside of the 95% confidence region cannot provide any information about the effect of network structure on the attributes of communities under high or low N addition frequencies.

3.3 Responses of species competition network to N addition

A TCN is a kind of simple competitive structure, also known as a hierarchical network, the complexity of which is labelled as 0. As the number of loops in an ICN increases, complexity will increase from 0 to 1. As shown in Fig. 4, the complexities gradually decreased from the upper left to the lower right with the nitrogen accumulation effect, suggesting that the network structure evolved from a complex ICN to a simple TCN and finally to NaN. The trend of complexity was similar to that of precipitation over time (Fig. S4), which was also verified by logistic regression (Tab. 2). With the N cumulative effect, complexities decreased several steps and then reached zero, even those for SCSs. The peak complexities were mainly stabilized at 2 g N m-2year-1, indicating that it may be the most beneficial to form ICNs or maintain richness under the accumulative effect of N addition.
Fig. 4 shows that more intransitive competition existed under N addition rates less than 15 g N m-2 year-1. With increasing duration and N addition rate, the number of loops in the SCNs gradually decreased from the upper left to the lower right and gradually changed to a simpler transitive mode. In particular, when the N addition rate was 50 g N m-2year-1, TCNs lasted 8 years. Species competition in 2019 and 2020, after long-term environmental filtering, became sparse, and species assemblies were more random than in other cases. Therefore, an intransitive network was beneficial to maintain richness but susceptible to nitrogen with a short periodicity, while a TCN was maintained for a longer period during grassland degradation under high nitrogen addition rates.

3.4 Logistic regression analysis of species competition network

Through univariate logistic regression analysis, we found that the community attribute variable, number of RPs, and three environmental variables (N addition rate, mowing, and precipitation) were significant predictors of TCN and ICN formation, and the accuracy for TCNs or ICNs is shown in Tab. 2. When the community attributes and environmental variables were combined to perform multivariate logistic regression, the four above-mentioned variables were still significant. The number of RPs was an important community attribute variable to identify network structure (Tab. 2). As shown in Fig. 3, under the same average biomass, the number of RPs in ICNs was greater than that in TCNs.
With increasing N addition rates, the probability of ICN formation gradually decreased, while the opposite was true for TCN formation (Tab. S5). The regression coefficient gradually changed from positive to negative, and the probability of ICN formation decreased from 0.595 to 0.217, while that of TCN formation increased from 0.405 to 0.783. When N ≥15 g N m-2 year-1, the significance of the regression coefficient became increasingly stronger, which was why we grouped the sample plots at 15 g N m-2year-1 in Fig. 4. The difference in probabilities between the two network structures intensified under high N- addition rates (Tab. S5).

4 Discussion

4.1 Selection of species assembly

Species assembly is a reliable prerequisite for detecting species interactions and a necessary step for constructing SCNs in this study. Currently, there are two main rules for the selection of species assembly: (1) based on the assumption that intransitivity is more common among superior species (Soliveres et al . 2015; Ulrich et al . 2018), and (2) all species were brought into the network to represent complex relationships in the natural state to the greatest extent (Hugo Saiz et al . 2018). The two strategies of selecting species assemblies had shortcomings for acquiring network structures and their corresponding ecological functions and information.
It is well known that the null model is a powerful tool for empirical and theoretical studies for community isolation or aggregation (Dullinger et al . 2007; De Bello et al . 2009), but nonrandom species assembly cannot be determined (Gotzenberger et al . 2012). Based on species co-occurrence in repeated plots, frequent assemblies can be screened out by the Apriori algorithm (Wang et al . 2020), which breaks the limitation of only using biomass to investigate species assembly and emphasizes co-occurrence frequency. Moreover, the Apriori algorithm maintained nonrandomness in species assembly, which motivated species with a small biomass but frequent occurrence in the SCN with a certain probability.
Therefore, SDAs with a high occurrence probability were taken as the starting point of this study and overcame the shortcomings in the above selection rules. We concluded that 78.95% of the test plots were dominated by interspecific competition (rs ≥ 0.6), 49% of which contained intransitive structures. Compared with prior studies (Soliveres et al . 2015). Two main reasons for this observation are that the selection rule does not limit superior species, and other species with high occurrence frequencies are also considered; environmental filtering causes plants on the grassland to evolve and substitute, resulting in species assemblies that also change constantly.

4.2 Inferior species are more likely in an intransitive competition network

In general, it is difficult for inferior species to adapt widely to grasslands. Their persistence depends on the interactions between species (Chler, Michalet & Callaway 2001). There may be differences in the importance of intransitivity between superior and inferior species (Hugo Saiz et al . 2018). Intransitivity can inhibit the superior utilization efficiency of resources and urge interspecific-competition hierarchical differences to be small, providing more occurrence opportunities for the inferior group (Maynard et al . 2017). Inferior species in a plot always play a role as a defeater that only occasionally appears; therefore, inferior species are more likely to survive in ICN. If an inferior species was in an SCN, it commonly promoted network complexity as an intermediary, transforming the TCN into a more complex ICN. For example, an ICN including Potentilla acaulis had 6 loops, 5 of which covered Potentilla acaulis (Fig. S7), which had the largest BC (Tab. 1). Under suitable environments or sufficient resources, inferior species can play a crucial role and can even directly determine the complexity of an SCN.

4.3 Effect of species competition network on community attributes

Intransitivity focuses more on symbiosis than on resource constraints (Huisman et al . 2001; Allesina & Levine 2011), and Soliveres (2015) proposed that intransitive structures embedded in species networks can better explain coexistence, which is also the reason why networks with intransitive structures are defined as ICNs. We found that ICNs could better maintain species richness under low N addition. Intransitive structures are a sign of complex competition, and changes in competition intensity are key drivers of interspecific interactions (Tilman 1988; Carsten 2007).. An increase in independent loops in a community will increase the number of species falling into loops, which has a stabilizing effect on the community (Gallien et al . 2017). TCN is also a common interaction pattern among plant species (Freckletonet al . 2000), and with great differences in interspecific competition, inferior species will be eliminated (Keddy & Shipley 1989). As species disappear, complex ICNs gradually degenerate into simpler TCNs. TCNs are a transition pattern of the species network from ICNs to SCSs, from a competition-dominated to environment-dominated.
Our results also confirmed that the species number in the plots labelled by TCNs was greater than that labelled by ICNs. This difference became greater with an increase in the N addition rate. TCNs can better maintain interspecific dominance of a species, while the main ecological function of ICNs was to maintain species symbiosis. In a plot labelled by ICNs, the individual biomass was relatively small, but the number of RPs was large, which was very different from that plots labelled by TCNs. This can explain why the number of RPs and not biomass can be used as a predictor to distinguish the type of SCN.

4.4 Environmental effects on network structure

Prior studies have shown that adding nitrogen to grassland ecosystems through fertilization or atmospheric N enrichment is beneficial to primary productivity (Lebauer & Treseder 2008; Zhang et al . 2015). However, a continuous increase in nitrogen to plant communities may have unforeseen consequences, including loss of richness (Stevenset al . 2004; Payne et al . 2017), a decrease in the degree of intransitivity (Soliveres et al . 2018), and threats to community stability (Hautiler et al . 2014; Zhang et al . 2016).
Environmental disturbance can affect competitive relationships (Goldberget al . 1999). N-enrichment events significantly reduce richness (Zhang et al . 2019). An increase in N enrichment reduces network complexity and causes species in a community to gradually compete for a single nutrient resource, enhancing the competition hierarchy among species and decreasing competition intransitivity at the same time. This is mainly because excessive perturbations in environmental factors increase adaptivity differences, leading to more asymmetrical competition (such as the transition from nutrient to light competition (De Malach et al . 2017)) or more transitive competition (Soliveres et al . 2018).
We found that much precipitation was beneficial to enhance the probability of ICN formation in the Inner Mongolia steppe, which seems to contradict the prior results that drought aggravates intransitivity (Soliveres et al . 2015; Gallien et al 2018). The reasons for this may be as follows: (1) the interannual differences in the drought gradient in the experimental area are smaller than those in global vision. Moreover, high temperature accelerates evaporation of surface water (Berget et al . 2014; Diro & Sushama 2017);(2) high N enrichment causes excessive nutrition in the experimental area, but a large amount of precipitation will dilute the N- concentration, thus enhancing the probability of ICN formation; and (3) a lack of water resources in the area makes rainwater a priority resource for species to compete for, tending to promote intransitive competition in a community.
Mowing reduces the temporal heterogeneity of biomass (Osem et al . 2002; Grman et al . 2010), variation in the heterogeneity of community richness and species asynchrony under N- enrichment (Hautileret al .,2014; Zhang et al . 2017). Nutrients and water are important resources for plant growth in temperate steppe, but mowing will aggravate the loss of soil nutrients (such as N, P, K, etc.), resulting in nutrient imbalance (Giese et al . 2013; Liu et al . 2015). Redistribution of nutrients intensifies competition for resources after mowing. Our numerical results indicated that under N enrichment, mowing contributed to enhance species complexity from an environment-dominated SCS to a TCN or from a TCN to an ICN. Because the negative effect of N enrichment on plant diversity is partially counteracted by infrequent annual mowing after the reproductive phase of plants (Collins et al . 1998; Storkey et al . 2015; Zhanget al . 2017). In addition, mowing may impair the potential balance among species, such as resource uptake and pollinator attraction (Aarssen 1992), but it also reclassifies competitive hierarchy and increases intransitive competition, such as the co-existence of inferior species.
Low frequency will lead to low colonization (acquisition of new species) (Zhang et al . 2015). The low colonization rate may be due to soil ammonium poisoning under low fertilization frequencies (Britto & Kronzucker 2002; Berg et al . 2005). Although N addition frequency is not the main factor affecting network structure and community attributes, ICN at low frequencies is slightly higher than that at high frequencies (top right panel in Fig. 3). The cause of this is that low colonization could not break down the adaptive relationship of the original species as the result of stabilization of ICN. High frequency destroyed the competitive balance of the original community, resulting in decreasing intransitivity competition.

5.Conclussion

In this paper, we first used the Apriori algorithm to mine SDAs from presence-absence data of plants. Inferior plants were included in the dominant assembly due to their high co-occurrence (Tab. 1). ICNs included not only superior but also inferior both existing in SDAs, which is one of the successes of this paper. Superior species are usually the main component of TCNs, while inferior species are more likely to enter ICNs. TCNs can be regarded as transition modes from more complex ICNs to SCSs mainly dominated by environmental factors. Environmental changes can affect variations in network structure. An increase in the N addition rate caused the network structure to change from an ICN to a TCN (Fig. 4). However, low N addition rates, mowing and rainfall were external drivers of ICN formation, which could maintain biodiversity better than TCNs.

Acknowledgements

This work was supported by the National Key Research and Development Project of China (2016YFC0500705) and the Fundamental Research Funds for the Central Universities (2015ZCQ-LY-01).

Reference

Aarssen, L.W. (1992). Causes and consequences of variation in competitive ability in plant communities. J. Veg. Sci. , 3, 165-174.
Agrawal, R. & Srikant, R. (1994). Fast Algorithms for Mining Association Rules and Sequential Patterns. Morgan Kaufmann Publishers, San Francisco CA.
Aschehoug, E. T., Brooker, R., Atwater, D. Z., Maron, J. L. & Callaway, R. M. (2016). The mechanisms and consequences of interspecific competition in plants. Annu.Rev.Ecol.Evol.S. , 47,263-281.
Atmar, W. & Patterson, B. D. (1993). The measure of order and disorderin the distribution of species in fragmented habitat.Oecologia , 96,373-382.
Berg, A., Lintner, B.R., Findell, K.L., Malyshev, S., Loikith, P.C. & Gentine, P. (2014). Impact of soil moisture–atmosphere interactions on surface temperature distribution. J. Climate , 27,7976-7993.
Bobbink, R., Hicks, K., Galloway, J., Spranger, T., Alkemade, R., Ashmore, M., et al . (2010). Global assessment of nitrogen deposition effects on terrestrial plant diversity: a synthesis.Ecol. Appl. , 20, 30-59.
Canfield, D. E., Glazer, A. N. & Falkowski, P. G. (2010) The evolution and future of earth’s nitrogen cycle. Science , 330: 192-196.
Carsten, F.& Dormann .(2007). Competition hierarchy, transitivity and additivity: investigating the effect of fertilisation on plant–plant interactions using three common bryophytes. Plant Ecol .191, 171-184.
Chalcraft, D.R., Cox, S.B., Clark, C., Cleland, E.E., Suding, K.N., Weiher, E. et al . (2008). Scale-dependent responses of plant richness to nitrogeN-enrichment. Ecology , 89, 2165–2171.
Chase, J.M. (2010). Stochastic community assembly causes higher richness in more productive environments. Science , 328, 1388-1391.
Choler, P., Michalet, R. & Callaway, R. M. (2001). Facilitation and competition on gradients in alpine plant com‐munities. Ecology , 82, 3295-3308.
Collins, S. L., Knapp, A. K., Briggs, J. M., Blair, J. M. & E. M. Steinauer. (1998). Modulation of diversity by grazing and mowing in native tallgrass prairie. Science, 280,745-747.
de Bello, F., Thuiller, W., Leps, J., Choler, P., Clement, J.C., Macek, P. et al (2009). Partitioning of functional diversity reveals the scale and extent of attribute convergence and divergence.J. Veg. Sci , 20, 475-486.
DeMalach, N., Zaady, E., & Kadmon, R. (2017). Light asymmetry explains the effect of nutrient enrichment on grassland diversity. Ecol. Lett. , 20, 60–69.
Diro, G.T.& Sushama, L. (2017). The role of soil moisture–atmosphere interaction on future hot spells over North America as simulated by the canadian regional climate model (CRCM5). J. Climate , 30, 5041-5058.
Dufour, A., Dagallah, F., Wagnr, H. H., Guisan, A., & Buttler, A. (2006). Plant species richness and environmental heterogeneity in a mountain landscape: Effects of variability and spatial configuration.Ecography , 29, 573-584.
Dullinger, S., Kleinbauer, I., Pauli, H., Gottfried, M., Brooker, R., Nagy, L. et al . (2007). Weak and variable relationships between environmental severity and small-scale co-occurrence in alpine plant communities. J. Ecol. 95, 1284-1295.
Durrett, R., & Levin, S. (1998). Spatial aspects of interspecific competition. Theor. Popul. Biol. , 53, 30-43.
Erisman, J. W., Galloway, J. N., Seitzinger, S., Bleeker, A., Dise, N. B., Petrescu, A. M. R.et al .(2013). Consequences of human modification of the global nitrogen cycle. Philos. T. Roy. Soc. B, 368, 20130116.
Feng, K., Zhang, Z.J., Cai, W.W., Liu, W.Z., Xu, M.Y., Yin, H.Q.et al .(2017). Biodiversity and species competition regulate the resilience of microbial biofilm community. Mol. Ecol. 26, 6170-6182.
Fox, J. W.(2013). The intermediate disturbance hypothesis should be abandoned. Trends Ecol. Evol. , 28, 86-92.
Fraterrigo, J.M., Turner, M.G., Pearson, S.M. & Dixon, P. (2005). Effects of past land use on spatial heterogeneity of soil nutrients in southern Appalachian forests. Ecol. Monogr. , 75, 215-230.
Freckleton, R.P., Watkinson, A.R., Dowling, P.M. & Ley, A.R. (2000).Determinants of the abundance of invasive annual weeds: community structure and non-equilibrium dynamics. Proc. R. Soc. B , 267, 1153-1161.
Gallien, L., Zimmermann, N. E., Levine, J. M., & Adler, P. B. (2017). The effects of intransitive competition on coexistence. Ecol. Lett. , 7, 791-800.
Giese, M., Brueck, H., Gao Y. Z., Lin S., Steffens M., Kögel-Knabner I.et al. (2013). N balance and cycling of Inner Mongolia typical steppe: a comprehensive case study of grazing effects. Ecolo. Monogr. , 83,195-219.
Gilpin, M. E. (1975). Limit cycles in competition communities. Am. Nat . 109,51–60.
Goldberg, D.E., Rajaniemi, T., Gurevitch, J.&Stewart-Oaten, A. (1999). Empirical approaches to quantifying interaction intensity: competition and facilitation along productivity gradients. Ecology , 80,1118-1131.
Gotzenberger, L., de Bello, F., Brathen, K.A., Davison, J., Dubuis, A., Guisan, A. et al. (2012). Ecological assembly rules in plant communities-approaches, patterns and prospects. Biol. Rev. , 87, 111-127.
Grman, E., Lau, J.A., Schoolmaster, D.R. & Gross, K.L. (2010). Mechanisms contributing to stability in ecosystem function depend on the environmental context. Ecol. Lett. , 13, 1400-1410.
Hautier, Y., Seabloom, E.W., Borer, E.T., Adler, P.B., Harpole, W.S.,Hillebrand, H. et al . (2014). Eutrophication weakens stabilizing effects of diversity in natural grasslands. Nature , 508, 521-525.
Hugo S.,Yoann L. B. P.,Nicolas G.&Fernando T. M..(2018) Intransitivity increases plant functional diversity by limiting dominance in drylands worldwide.Journal of Ecology .107, 240-252.
Hui, C., Li, Z. Z., & Yue, D. X. (2004). Metapopulation dynamics and distribution, and environmental heterogeneity induced by niche construction. Ecol. Model. , 177, 107-118.
Huisman, J., Anna M.,Johansson, Eelke O. Folmer&Franz J. W.(2001). Towards a solution of the plankton paradox: the importance of physiology and life history. Ecol.Lett . 4, 408-411.
Huisman, J. & Weissing, F. J. (1999). Richness of plankton by species oscillations and chaos. Nature , 402, 407-410.
Keddy, P.A. & Shipley, B. (1989). Competitive hierarchies in herbaceous plant communities. Oikos , 54, 234-241.
Kendall, M. G. & Smith,B.B. (1940). On the method of paired comparisons. Biometrika , 31,324-345.
Kerr, B., Margaret A. R., Feldman, M. W. & Bohannan, B. J. M. (2002). Local dispersal promotes richness in a real-life game of rock–paper–scissors. Nature , 418,171-174.
Laird, R. A. & Schamp, B. S. (2009). Species coexistence, intransitivity, and topological variation in competitive tournaments.J. Theor. Biol. , 256,90-95.
Laird, R. A. & Schamp, B. S. (2006). Competitive intransitivity promotes species coexistence. Am. Nat. , 168,182-193.
Laird, R. A.& Schamp, B. S. (2015). Competitive intransitivity, population interaction structure, and strategy coexistence. J. Theor. Biol. , 365,149-158.
Lan, Z.C., Jenerette, G.D., Zhan, S.X., Li, W.H., Zheng, S.X. & Bai, Y.F. (2015). Testing the scaling effects and mechanisms of N-induced richness loss: evidence from a decade-long grassland experiment.J.Ecol. , 103, 750-760.
Landi, P.,Minoarivelo, H.., Brännström, Åke, Hui, C.&Dieckmann, U.(2018). Complexity and stability of ecological networks: a review of the theory. Popul. Ecol. , 60, 319-345.
LeBauer, D.S. & Treseder, K.K. (2008). Nitrogen limitation of net primary productivity in terrestrial ecosystems is globally distributed.Ecology , 89, 371-379.
Levine, J. M. & Rees, M. (2002). Coexistence and relative abundance in annual plant assemblages: the roles of competition and colonization.Am. Nat. , 160, 452-467.
Levine, S. H. (1976). Competitive interactions in ecosystems. Am. Nat. , 110, 903-910.
Liu, N., Kan, H. M., Yang, G. W., & Zhang Y. J. (2015). Changes in plant, soil, and microbes in a typical steppe from simulated grazing: explaining potential change in soil C. Ecolo. Monogr. , 85,269-286.
Luo, J. & Magee, C.L. (2011). Detecting evolving patterns of self-organizing networks by flow hierarchy measurement. Complexity , 16, 53-61.
May, Robert M. & Leonard, Warren J. (1975). Nonlinear Aspects of Competition Between Three Species. SIAM J. Appl. Math. , 29, 243-253.
Mayfield, M. M., & Stouffer, D. B. (2017). Higher-order interactions capture unexplained complexity in diverse communities. Nat. Ecol. Evol. , 1,621-627.
Maynard, D. S., Bradford, M. A., Lindner, D. L., van Diepen, L. T. A., Frey, S. D., Glaeser, J. A. et al . (2017). Diversity begets di‐versity in competition for space. Nat. Ecol. Evol , 1, 1561-1568.
Meserve, P. L., Gutierrez, J. R., Yunger J.A., Contreras L.C.& Jaksic, F.M. (1996). Role of biotic interactions in a small mammal assemblage in semiarid Chile. Ecology , 77,133-148.
Osem, Y., Perevolotsky, A. & Kigel, J. (2002). Grazing effect on diversity of annual plant communities in a semi-arid rangeland: interactions with small-scale spatial and temporal variation in primary productivity. J.Ecol. , 90, 936-946.
Olff, H., Alonso, D., Berg, M.P., Eriksson, B.K., Loreau, M., Piersma, T. et al. (2009). Parallel ecological networks in ecosystems. Philos. T. Roy. Soc. B. 364, 1755-1779.
Paul A. Keddy & Bill Shipley. (1989). Competitive hierarchies in herbaceous plant community. Oikos , 54, 234-241.
Payne, R.J., Dise, N.B., Field, C.D., Dore, A.J., Caporn, S.J.M. &Stevens, C.J. (2017). Nitrogen deposition and plant richness: past,present, and future. Front. Ecol. Environ. , 15, 431-436.
Reichenbach, T., Mobilia, M. & Frey, E. (2007). Mobility promotes and jeopardizes richness in rock–paper–scissors games. Nature , 448 ,1046-1049.
Wang, S., Zhang,G.M. & Li, J.S.(2020). Effects of simulated nitrogen deposition on the species networks of inner Mongolia grassland.Glob. Ecol. Conserv. , 24,1-13.
Shizuka, D. & D. B. McDonald. (2012). A social network perspective on measurements of dominance hierarchies. Anim. Behav. , 83, 925-934.
Slater, P. (1961). Inconsistencies in a Schedule of Paired Comparisons.Biometrika, 48, 303-312.
Soliveres, S., Eldridge, D. J., Maestre, F. T., Bowker, M. A., Tighe, M., & Escudero, A. (2011). Microhabitat amelioration and reduced competition among understorey plants as drivers of facilitation across environmental gradients: Towards a unifying framework. Perspect. Plant Eco. , 13, 247-258.
Soliveres, S.& Eric Allan.(2018). Everything you always wanted to know about intransitive competition but were afraid to ask. J. Ecol. , 106, 807-814.
Steiner, C.F. & Leibold, M.A. (2004). Cyclic assembly trajectories and scaledependent productivity-diversity relationships. Ecology , 85, 107-113.
Stevens, C.J., Dise, N.B., Mountford, J.O. & Gowing, D.J. (2004). Impact of nitrogen deposition on the species richness of grasslands.Science , 303, 1876-1879.
Storkey, J., Macdonald, A. J., Poulton, P. R., Scott, T., Köhler, I. H., Schnyder, H., et al . (2015). Grassland richness bounces back from long-term nitrogen addition. Nature 528,401-404.
Tilman, D. (1994). Competition and richness in spatially structured habitats. Ecology , 75, 2-16.
Thebault, E.& Fontaine, C. (2010). Stability of Ecological Communities and the Architecture of Mutualistic and Trophic Networks.Science , 329(5993), 853-856.
Ulrich, W., Soliveres, S., Kryszewski, W., Maestre, F. T., & Gotelli, N. J. (2014). Matrix models for quantifying competitive intransitivity from species abundance data. Oikos , 123, 1057-1070.
Vitousek, P. M., Aber, J. D., Howarth, R. W., Likens, G. E., Matson, P. A., Schindler, D.W. et al. (1997). Human alteration of the global nitrogen cycle: sources and consequences. Ecol. Appl. , 7, 737-750.
Ulrich,W., Kubota, Y., Piernik,A.& Gotelli,N.J.(2008). Functional attributes and environmental characteristics drive the degree of competitive intransitivity in European saltmarsh plant communities. J.Ecol. , 106,865–876.
Wesche, K., Krause, B., Culmsee, H. & Leuschner, C. (2012). Fifty years of change in Central European grassland vegetation: large losses in species richness and animal-pollinated plants. Biol. Conserv. , 150, 76–85.
Western, D. (2001). Human-modified ecosystems and future evolution.Proc. Natl Acad. Sci .,98, 5458–5465.
Whitford, W.G. (2002). Ecology of Desert Systems. J. Mammal. ,84,1122-1124.
Wootton JT. (1994). The nature and consequences of indirect effects in ecological communities. Annu. Rev. Ecol. Syst. , 25,443-466.
Worm, B. & Karez, R. (2002). Competition, coexistence and diversity in rocky shores. Springer , 161, 133-163.
Zhang, Y.H., Feng, J.C., Isbell, F., Lü, X.T., Han, X.G. (2015).Productivity depends more on the rate than the frequency of N addition in a temperate grassland. Sci. Rep. , 5, 12558.
Zhang, Y., Loreau, M., He, N., Zhang, G., & Han, X. (2017). Mowing exacerbates the loss of ecosystem stability under nitrogeN-enrichment in a temperate grassland. Funct. Ecol ., 31, 1637-1646.
Zhang, Y.H., Loreau, M., Lu, X.T., He, N.P., Zhang, G.M. & Han, X.G. (2016). NitrogeN-enrichment weakens ecosystem stability through decreased species asynchrony and population stability in a temperate grassland. Global Change Biol. , 22, 1445–1455.
Zhang, Y.H., Lu, X.T., Isbell, F., Stevens, C., Han, X., He, N.P. et al. (2014). Rapid plant species loss at high rates and at low frequency of N addition in temperate steppe. Global Change Biol. , 20, 3520-3529.

Figure Legends:

Figure 1 Schematic diagram of this experimental study.