Data analysis
We quantified both functional (i.e., biomass based) and structural resistance and recovery of plant communities to examine their responses to our experimental treatments. We focused on absolute biomass changes as measures of functional resistance and recovery because they are directly relevant for agriculture and animal husbandry. Biomass resistance (B rst) was determined as the difference in AGB between each of the drought years (2015-2017) and the year before the drought (2014) (Pfisterer & Schmid, 2002). Biomass recovery (B rc) was determined as the difference in AGB between the post-drought year (2018) and each of the drought years (2015-2017) (Jasper & Frank, 2010). Structural resistance (S rst) was defined as the Bray-Curtis similarity (Bray & Curtis, 1957) between communities in the pre-drought year and each of the drought years, which was calculated as:Srst = 2Σi min(AGBi ,pre-drought,AGBi ,drought)/(ΣiAGBi ,pre-drought+ ΣiAGBi ,drought), where AGBi ,pre-drought andAGBi ,drought are the AGB of species i in the community before and during drought, respectively. S rst equals 1 if there is no structural difference between drought and pre-drought communities and equals 0 when drought and pre-drought communities have no species in common. Structural recovery (S rc) was measured as the Bray-Curtis dissimilarity (Bray & Curtis, 1957) between communities in each of the drought years and the post-drought year, which was calculated as: Src = 1-2Σi min(AGBi ,post-drought,AGBi ,drought)/(ΣiAGBi ,post-drought+ ΣiAGBi ,drought), where AGBi ,postdrought is the AGB of species i in the community after drought. Maximum recovery values approach 1 (i.e., when post-drought AGB is substantially greater than AGB during the drought), whereas values <<1 reflect low recovery. We also calculated Srst and Src of dominant species (hereafter dominant structural resistance and dominant structural recovery, respectively) using the above formulas by considering only the five most common species (L. chinensis ,S. baicalensis , C. squarrosa , C. duriuscula andB. scorzonerifolium ), whose summed AGB accounted for 79.2% of community AGB across treatments and sampling years. To determine the role of species asynchrony in regulating community stability, we calculated species asynchrony as , where is the variance of community AGB and is the standard deviation of the AGB of the i th species in a community (Loreau & de Mazancourt, 2008). Species asynchrony was calculated separately for resistance and recovery by using data from the years relevant to the calculation of resistance and recovery, respectively.
Linear mixed-effects models were used to assess the effects of block, N, mowing, and their interactions on soil pH, inorganic nitrogen concentration, community AGB, species richness, biomass resistance, biomass recovery, structural resistance, and structural recovery; year was included as a random effect in the models. Two-way ANOVA was used to test the effects of block, N and mowing on species asynchrony. One-way ANOVAs with Duncan’s multiple range tests were used to evaluate differences among treatments. Relationships between community resistance/recovery and their potential abiotic and biotic drivers were explored with bivariate regressions. Bivariate regressions were also used to test for potential trade-offs among stability properties. Based on the bivariate relationships, we constructed a prioristructural equation models (SEMs) to understand the direct and indirect effects of abiotic and biotic factors on community biomass resistance, biomass recovery, structural resistant and structural recovery (Fig. S2). Overall fit of the SEM was evaluated using the chi-square test (the model has a good fit when 0.05 < P ≤ 1.00 for χ2 test) and Akaike information criteria (AIC; lower AIC indicating a better fit); final models were obtained by eliminating non-significant pathways and state variables based on regression weight estimates.
The data on community AGB and biomass recovery were ln-transformed to meet the assumptions of normality. SEM analyses were performed using AMOS 18.0 (Amos Development Co., Greene, Maine, USA). The remaining statistical analyses were conducted using SPSS 13.0 (SPSS, Inc., Chicago, IL, USA).
RESULTS
Nitrogen addition reduced soil pH and increased inorganic nitrogen concentration, while mowing had no significant effects on these two variables (Table 1, Fig. S3). The three-year (2015-2017) drought significantly reduced species richness and community AGB (all P< 0.01) across treatments when compared to 2014, the year before the drought (Fig. S4). Nitrogen enrichment increased community AGB, but had no effect on species richness; mowing resulted in decreased community AGB, but increased species richness (Tabe 1; Fig. S4).
As the most dominant species, L. chinensis accounted for approximately 46.5% of plant AGB, which was much greater than the contribution of any other species (none exceeded 12.3%). Nitrogen enrichment consistently increased the AGB of L. chinensis (allP < 0.05), while mowing consistently reduced the AGB ofL. chinensis , across the five years of the experiment (allP < 0.001, Table S1; Fig. S5). The treatment effects on the summed AGB of the five dominant species mirrored those for L. chinensis (Table S1; Fig. S5).
Nitrogen enrichment significantly reduced community biomass resistance (Fig. 1a), but increased community biomass recovery (Fig. 1b). Mowing marginally significantly increased community biomass resistance, but did not affect community biomass recovery (Fig. 1a,b). Mowing, however, mitigated the negative effect of nitrogen enrichment on biomass resistance (significant nitrogen × mowing term; Table 1; Fig. 1a). For structural stability, nitrogen enrichment had no discernable effect on community structural resistance or recovery, whereas mowing significantly decreased community structural resistance, and increased community structural recovery (Table 1; Fig. 1c,d). Nitrogen enrichment, but not mowing, decreased species asynchrony for both resistance and recovery (Table 1).
Nitrogen enrichment decreased biomass resistance of L. chinensisbut increased its biomass recovery (Table 1; Fig. 2a,b), paralleling its effect on community-level biomass resistance and recovery. By contrast, mowing increased biomass resistance of L. chinensis but decreased its biomass recovery (Table 1; Fig. 2a,b). Mowing, but not nitrogen addition, reduced dominant structural resistance; both nitrogen addition and mowing had a positive effect on dominant structural recovery (Table 1; Fig. 2c,d).
Structural equation modelling (SEM) revealed that nitrogen deposition decreased community biomass resistance by reducing biomass resistance of the most dominant species L. chinensiss and, in turn, its positive effect on species asynchrony (Fig. 3a). Mowing promoted community biomass resistance by increasing biomass resistance of L. chinensis and resultant species asynchrony (Fig. 3a). Nitrogen deposition had no detectable effect on community structural resistance, as its positive effect through decreasing L. chinensis biomass resistance was offset by its negative effect through decreasing species asynchrony and dominant structural resistance (Fig. 3a). Mowing decreased community structural resistance mainly through reducing dominant structural resistance, and this negative effect was partially offset by increased L. chinensis biomass resistance which increased dominant structural resistance (Fig. 3a).
Nitrogen addition promoted community biomass recovery by increasing biomass recovery of L . chinensis and a host of other less dominant species (the latter captured as decreasing species asynchrony in the SEM; Fig. 3b). However, mowing showed no significant effect on community biomass recovery, as its negative effect through decreasing biomass recovery of L. chinensis was largely offset by its positive effect through increased dominant structural recovery (Fig. 3b). Whereas nitrogen deposition had little effect on community structural recovery, mowing promoted community structural recovery mainly by increasing dominant structural recovery (Fig. 3b).
Bivariate regressions revealed that community biomass resistance and recovery were significantly positively associated with their structural resistance and recovery, respectively (Fig. S6a, b). SEMs also revealed a positive effect of community structural recovery on biomass recovery, but no significant relationship between community biomass and structural resistance (Fig. 3).