Species loss due to an increasing number of added nutrients has been explained by both light competition through biomass increase and by niche dimension reduction as a result of species specific limiting soil resources trade-offs. Disturbances, by reducing community biomass, species dominance and increasing light availability, may counteract above ground nutrient effects. However, it is unknown if diversity loss at local or spatial scales generated by increasing number of added nutrients can be redressed with canopy disturbance.

We evaluated if local (alpha) and spatial scale (beta) diversity loss generated by the number of added nutrients can be reverted by disturbances in Flooding Pampa grasslands, Argentina. In a 4-yr replicated field experiment, we added soil resources combining nitrogen, phosphorus and potassium to obtain 0, 1, 2 or 3 nutrients and manipulated the regime of canopy disturbance by seasonal mowing and biomass removal.

We found that the increasing number of added nutrients strongly reduced local and spatial plant diversity, despite biomass and light changes generated by mowing. In mown plots, nutrient driven local diversity loss was intensified along time, thus increasing species dominance. While mowing did not affect dominant species loss, increasing number of added nutrients promoted rare species loss and reduced spatial dissimilarity. Furthermore, mowing increased local and spatial diversity regardless light or biomass effects, suggesting alternative pathway effects for disturbance.

Synthesis: Our results demonstrated that even when disturbance generated a positive effect on local and spatial diversity, it did not completely counteract the negative effect of number of added nutrients. Thus, the relative importance of above and belowground resource competition may change when chronic disturbances alter community dominance. Under low light availability, above-ground competition may drive species richness loss but when disturbance reduces light limitation, the increasing number of added nutrients may reduce niche dimensionality and thus species coexistence. In sum, faced with the need to manage eutrophized grasslands, our study showed that disturbance may not completely mitigate the negative effect of multiple nutrient inputs on local and spatial grassland diversi

In 2012 we established a long-term experiment of nutrient addition in six fenced areas excluded from grazing and incorporated mowing treatment in 2014 because canopy disturbances were absent. Our experiment consisted of a split-plot design, replicated in six blocks, with the addition of limiting soil resources in the main plot (5 x 5 m2) and mowing treatments as frequent canopy disturbance in the sub-plots (2,5 x 2,5 m2; Supplementary Fig. 1S). The nutrient addition experiment is part of the Nutrient Network (https://nutnet.org/https://nutnet.org/ https://nutnet.org/https://nutnet.org/www.nutnet.org), however mowing as treatment is exclusive of our site. Blocks consisted of six 20 x 25 m² enclosures separated by ~100m from each other and were established in 2004, as part of a long-term experiment of cattle exclusion (Longo et al., 2013). A factorial arrangement of nitrogen [N], phosphorous [P] and potassium [K + micronutrients, hereinafter K] were randomly assigned to each of eight 25 m² main plots to represent the minimum number of potentially limiting elemental nutrients added (Harpole & Tilman, 2007). In each block, nutrient addition followed a scheme of increasing number of limiting resources (hereafter, number of added nutrients), with zero nutrients added (1 plot/block), one nutrient added (N, P, or K; 3 plots/block), two nutrients added (NP, NK or PK; three plots/block), and three nutrients added (NPK; 1 plot/block). Each nutrient was added at a rate of 10g m² year¹ to ensure a substantial increase in nutrient availability and remove potential limitations (Fay et al., 2015). Doses were split and equally applied three times a year (1/3 in early spring, 1/3 in early summer and 1/3 in early autumn). The nutrients were applied in commercially available granules of urea (N), triple superphosphate (P) and potassium sulphate (K) (Borer et al., 2014). From the six blocks, three blocks contained the entire nutrient range (0-1-2-3) and the other three only contained the extreme values (0 and 3). We incorporated all blocks to increase statistical power. Mowing treatments involved mechanical slash and removal of aerial biomass to a height < 5 cm in one sub-plot. Thus, each main plot was divided in 4 subplots (2.5 x 2.5 m); mown subplot was opposed to the “intact” subplot to ensure less contact and thus more independence between mowing treatments (Fig. 1S). As for nutrients, mowing was applied three times a year at the beginning of spring, summer, and autumn [i.e. September, December and April, respectively].

Data collection

Every year, between 2014 and 2017, we measured vegetation in late spring (i.e. November) and late summer (i.e. March) in order to assess the effect of nutrient addition and mowing on diversity. Plant percentage cover in the permanent 1x1 m² was visually estimated to the nearest 5% using a modified Daubenmire method (Daubenmire1959). In all analyses we used the maximum cover value recorded for each species in late spring or summer to accurately represent species’ performance in the entire year (Perelman et al., 2001).

            We measured peak plant biomass and light penetration simultaneously with plant cover measurements (i.e. November-March). Live and standing dead plant aerial biomass was harvested annually at peak biomass (i.e. early March; Sala, Deregibus, Schlichter, & Alippe, 1981), within two frames of 0.2 x 0.5 m², randomly located in mown and intact subplots, avoiding the area used for plant cover estimation. Samples were dried at 60° C for 72 h. and weighed to the nearest 0,1 g. Photosynthetically active radiation (PAR, μmol m² s¹) was measured above and below vegetation at noon (11:00 and 14:00) on sunny days using a 1m-long ceptometer (Cavadevices, Buenos Aires, Argentina). Proportion of PAR reaching the soil (pPAR) was calculated as the ratio of below/above PAR. We used the minimum value from two readings to account for of light limitation.

Diver.csv contain cover of each species and columns with each treatment. With this data we calculated richness, Simpson inverse index and Bray-Curtis dissimilarity to estimated beta diversity. We used linear mixed effects models (R package nlme, ´lme´ function, Pinheiro et al., 2018). In all cases, we modeled response variables as a function of number of added nutrients (ID: NUM), mowing (ID: Mow) and time (ID: Years) as fixed effects. For random effect structure, we assumed that there were 1 to 3 plots within a block accounting for the same nutrient level, thus mown subplots were nested in plots (ID: PLOT), plots nested in number of added nutrients, and the number of added nutrients nested in blocks (ID. BK). We modeled variance (e.g. Varident) and/or a correlation structure (e.g. auto-regressive moving average; ARMA) and we used Akaike´s Information Criteria (AIC), following Zuur et al. (2009). Predicted values and confidence intervals extracted of these models were used to build Fig. 1a, b c. Further, this data file includes a column with the data of PAR (light availability, ID: PAR). This variable has been analyzed through generalized linear effect models (R package MASS, ´glmmPQL´ function, Riplley et al., 2018) and predicted value and confidence intervals (95%) of this model were used to build Fig. 4S.

Abundance_time.csv contain the data for cover (as estimator of abundance) of each species at the beginning and at the final of experiment, 2014 and 2017 respectively. In this case, only considerate species present in the initial time and at the final time; thus, we did not consider the species that was gained during the years of the experiment. With this data, we evaluated if the probability of species loss depended on effect of number of added nutrients (ID: Num) and initial species abundance (ID: cover:initial) (Suding et al., 2005; Avolio et al., 2014). Here, we used a generalized linear mixed effect model (glmer function in ‘lme4’ package, Bates et al., 2014), with species presence-absence (ID: presence) as response variable, and initial cover, number of added nutrients and mowing (ID: Mow) as fixed predictors. We considered that a species was lost if present in initial time but absent after 4 years of treatment in the same plot (Avolio et al., 2014) and so we did not consider the species that was gained during the years of the experiment. This approach permitted to evaluate if species were lost due to treatments and not because they were not present at the beginning of the experiment. Predicted value and confidence intervals (95%) extracted of this model were used to build Fig. 2.

Sem.csv contain data about richness (ID:Riqueza), beta diversity (ID: distancias), gamma diversity(ID: gamma), biomass (ID: biomasa) and PAR (light availability, ID: PAR). These data allowed identify direct and indirect effects of the number of added nutriens (ID: NUM) and mowing (ID: Mow) along time (Years) on richness and beta diversity. For this, we used Structural Equation Modeling (SEM; Grace 2006; sem.fit function in ‘piecewiseSEM’ package, Lefcheck 2016). This technique allows to include biomass (log-transformed) and light (PAR) as both responses of treatments and predictors of plant diversity, facilitating the identification of direct and indirect effects (Lefcheck, 2016). With these results we built the Fig. 3.