1. Community composition is limited by a species’ ability to reach, establish, and survive on a site. Establishment and survival are constrained by both abiotic conditions and biotic interactions that operate together on local scales. They decide which species from the pool will form the community. For this reason, it is very important to clearly define the species pool, against which the community composition is compared. 
2. The effect of biotic and abiotic factors can be assessed experimentally, and the species pool by using estimation methods based on broad-scale observational data. We compared success of five species pool estimation methods in predicting establishment and survival in a seed/transplant addition experiment. 
3. In four different locations, we added resident and non-resident species to plots with and without competition and tested the ability of the species to thrive in both competition-free gaps (constrained mainly by abiotic conditions) and in intact vegetation (complete community filter). In these treatments, we studied the seedling recruitment and survival, and the establishment and survival of pre-grown transplants. The ability of species pool assessment methods to predict species performance in individual treatments was compared. The comparison of results from individual treatments indicates the importance of individual components of the community filter.
4. Species pool assessment methods, based on species co-occurrence patterns (Beals index, Favourability, and Unconstrained ordination), were the best predictors of species performance in the intact vegetation but were less successful in the competition-free environment. Methods based on co-occurrence patterns were the most effective for predicting seedling establishment, while seed germination alone and transplant survival were poorly predictable. 
5. The biotic filter was the principal factor defining our community composition, especially for the process of seedling establishment. The roles of biotic and abiotic filters are very difficult to distinguish without an experimental approach and the ratio of their importance changes during plant ontogenesis.

Study site

The seed/transplant addition experiment was conducted in four species-rich oligotrophic meadows in South Bohemia, Czech Republic. These four localities formed a moisture gradient. The wettest location, Ohrazení (48°57´N, 14°35´E, 510 m a.s.l.), is a wet meadow characterized as Molinion. Vrcov (48°55′N, 14°39′E, 510 m a.s.l.) is a mesic meadow characterized as Alopecurion with some elements of Molinion. Zvíkov (48°59′N, 14°36′E, 500 m a.s.l.) is a mesic meadow characterized as Arrhenatherion (association Poo-Trisetetum). The driest location, Závraty (48°56´N, 14°23´E, 460 m a.s.l.), is a relatively dry grassland with the lowest productivity (Table S2 in Appendix S1) characterized as Arrhenatherion (association Ranunculo bulbosi-Arrhenatheretum elatioris). All these localities were extensively managed with a single mowing term at the end of June with the exception of Vrcov, where the meadow was mown twice a year (June and September). The localities were a maximum of 20 km apart and experience very similar climatic conditions (Table S3 in Appendix S1). We did not need any permission for our fieldwork. 

Species

We selected 30 meadow plant species (their Ellenberg indicator values for light varied from 5 to 8, for moisture from 3 to 10 and for nutrients from 2 to 7), including both residents typical in target locations and non-residents typical in different habitats (Table S4 in Appendix S1). We chose those species which are not rare in the region, for which the seeds were available, and which provided high variability of their traits and ecological preferences. To increase the variability of ecological preferences, we also included two species of very wet habitats (Potentilla supina, Eleocharis palustris) and one mostly ruderal species (Geum urbanum) that are found only occasionally in very wet or disturbed meadows respectively. No forest species were included. On the other hand, we restricted our selection to herbaceous perennial species, to avoid the possible “successional” changes during the four years of the experiment. We also avoided Carex species that are known for their bad germination. We expected that values of indices expressing the species suitability (e.g., Beals index, Ellenberg indicator values, functional traits), which we used in further analyses, will be higher for resident than for non-resident species. Species residence was determined for each location individually: species were considered resident if present in at least one of the five phytosociological relevés (5x5m) recorded in each location in June 2016. These five relevés should be sufficient to thrive well in the locality – we expect that we might miss just very few transient species with rather erratic appearance in the locality. All non-resident species were present within a radius of 30 km from the locations (www.pladias.cz, access 13^th of November 2020), and are therefore part of the regional species pool. We chose only perennial species, with the exception of Potentilla supina, which could be both annual and perennial (Kleyer et al., 2008).

Seed introduction experiment

We used seeds from a commercial supplier (Planta Naturalis, Markvartice, Czech Republic) without any stratification. Seeds were sown into both the artificially created gaps (plots without competition from surrounding vegetation) and the intact vegetation (control plots with competition from surrounding vegetation) at the beginning of April 2016. In each location, two blocks, each containing 30 gaps and 30 control plots (0.2m x 0.2m each), were established (details about design are in Appendix S1, Fig. S1a). In each block, each species was sown separately in one gap and one control plot (200 seeds per species were sown evenly throughout the entire area of each plot). The germination of seeds and survival of seedlings was monitored from April 2016 to September 2019, two times per each year (in spring and summer) with exception of the last year when one extra monitoring was done before the end of the experiment. The number of surviving individuals was used as the response in statistical analyses, and we call this value “performance”. At the beginning of the observational period, this characteristic was affected mainly by germination and establishment and later modified by survival.

Transplant introduction experiment

Seeds (from the same source as in the case of seed introduction experiment) were germinated in the middle of April 2016 in a growth chamber (12 h light and 12 h darkness at 19°C) and subsequently, seedlings were replanted into a greenhouse, each individual in a separate pot (7x7x6.5cm). For the substrate, we used soil from the target locations (to ensure the natural composition of microbiota) mixed with sand in a ratio of 3:1 (to balance the level of nutrients increased in soil after its relocation from the field). After two months, we planted transplants, which had roughly the same size within the same species (Table S5 in Appendix S1), in the field. We used separate plots for transplant and seed introduction experiment. In each location (same as for the seed introduction experiment), the design of the transplant experiment was arranged in four random blocks with three treatments in each block (in addition to gap and control plots as in the seed addition experiment, clipped plots were added to exclude only aboveground competition; details about design are in Appendix S1, Fig. S1b). Two pre-grown transplants of each species were planted randomly in each plot in the middle of June 2016. The survival of transplants was monitored from June 2016 to April 2019, once in summer 2016, one spring and two summer monitoring in 2017, one spring and one summer monitoring in 2018 and one spring monitoring in 2019 before the end of the experiment.

Data analysis

To assess how well a species “belongs in a community”, we used five different methods of species pool assessment (Table S4 in Appendix S1). These indicators were subsequently correlated with measures of species performance in the field experiment. The methods used were Ellenberg indicator values (EIV; Euclidean distance between EIV and CWM, approach 1b.), species functional traits (Gower's distance between functional traits and CWM, group 2), and three methods based on co-occurrences (approach 1c.): Beals index (BI), its corrected form (Favourability), and unconstrained ordination (UNO).

We used EIV for moisture, light, and nutrients which were taken from a list of these values for the Czech Republic (Chytrý et al., 2018), and using the R function “dist”, we calculated the Euclidean distance for each species between three EIV and CWM of the target habitat (Appendix S2). Then, we used five species functional traits: canopy height determining competitive ability of species, specific leaf area (SLA) and leaf dry-matter content (LDMC) associated with leaf economic spectrum and thus trade-off between resource acquisition and conservation (all three traits taken from the LEDA database; Kleyer et al., 2008), lateral spread (with exclusion of freely dispersible organs) determining ability of clonal spread from the CLO-PLA 3.3 database (Klimešová et al., 2017), and seed mass which is related to reproductive ability. Average mass of one seed was derived from the real mass of 50 seeds used in the experiment. We computed the Gower’s distance of each species between five functional traits and CWM of the target habitat in R-package “gower” (van der Loo, 2022, Appendix S3).  In Appendix S1 (Fig. S2), we also presented results for raw values (i.e., not distance from CWM) of each functional trait separately showing which functional traits are generally good predictors for species establishment.

We used three methods of species pool determination based on species co-occurrence patterns calculated for each combination of sown species and experimental location: Beals index (BI), its corrected form (Favourability), and unconstrained ordination (UNO). We used the five phytosociological relevés recorded in each location as target vegetation, and the Czech National Phytosociological Database (Chytrý & Rafajová, 2013) as the reference database (stratified subsample, containing 31 512 relevés to reduce oversampling of some areas (Knollová et al., 2005)). We calculated abundance weighted form of BI, using the R-package “vegan” (Oksanen et al., 2019, Appendix S4). Favourability correction of Beals index (Real et al., 2006) was calculated using the R-package “fuzzySim” (Barbosa, 2015, Appendix S5). UNO was calculated using the R function “dark.pred.ca (Brown et al., 2019, Appendix S6). This method is based on the unconstrained ordination (Correspondence analysis) of the reference database. Then, the relevé of the target site is used as a supplementary sample, its coordinates are calculated, and the suitability of the location is determined based on the distance of the focal species’ position (i.e., its optimum) from the target site in the ordination space. We calculated BI, Favourability and UNO for each relevé separately and used the average value across the five relevés for each location for subsequent analyses. Because we consider the species pool to be fuzzy rather than a crisp set, we did not use any threshold to get species forming the pool, but we directly used the indicators calculated as a quantitative measure of species appropriateness for the community. 

In each treatment, we calculated Pearson's correlation for each indicator with performance of seedlings germinated from sown seeds in the field (seedlings) and performance of pre-grown transplants (transplants). Note that high correlation also means good predictive power (Lepš & Šmilauer, 2020). In each treatment type, we used the average performance (from two and four replications for seedlings and transplants respectively) for every combination of observation time, location, and species.

We also correlated the indicators with the ratios of seedling or transplant performance in different treatments, to reflect directly the effects of competition; “Clipped/Gap” represents the effect of belowground competition, “Control/Clipped” the effect of aboveground competition and “Control/Gap” the combined effect of both. Value 1 denotes no effect and 0 the strongest effect of competition.

To compare the predictive power of different methods of species pool assessments, we used obtained correlation coefficients in general linear models (GLM) with F test of Analysis of variance (Anova), where we used each measuring time and each location as replications (Appendix S7). To analyse the changes in predictive power of different methods of species pool assessment over time, we calculated a repeated measurement analysis of the variance of median of correlation coefficients between seedling/transplant performance and different methods of species pool assessment using Linear Mixed Effects Models (LMEM) in the R-package “nlme” (Pinheiro et al., 2021, Appendix S8). Both in GLM and LMEM, we used correlation coefficient values for EIV (Euclidian distance) and functional traits (Gower distance) multiplied by -1, because these are distances from CWM and thus, they are assumed to be negative predictors. Thus, the positive values in graphs always mean prediction in correct direction. For the values of correlation coefficients, we showed in the figures the limits of significance (p<0.05 in a two-tailed test) for a single correlation coefficient for the given number of species (30 in our case). It can be interpreted that if the average exceeds this limit, correlation coefficients for a given treatment are mostly significant. We also represented in Fig. 2 the interquartile ranges to show which percentage of correlation coefficients at different times falls into this range. To calculate the lower and upper quartiles, we used R-package “dplyr” (Wickham et al., 2022, Appendix S8).

Finally, we used BI as the species pool determination method with, generally, the best predictive power for experimental seedling/transplant performance to compare its results in detail. We used the values of correlation coefficients of seedling/transplant performance with BI in GLM in the program R (Appendix S9). In the case of transplants, the effect of the geotextile lining (Textile) was included.

For all figures, we used the R-package “ggplot2” (Wickham, 2016, Appendix S10).

R Studio, R, Microsoft Excel