Plant functional diversity and nutrient availability can improve restoration of floating fens via facilitation, complementarity and selection effects

*Both authors contributed equally. 1Ecology & Biodiversity Group, Department of Biology, Utrecht University, Utrecht, The Netherlands 2Dutch Botanical Research Foundation (FLORON), Nijmegen, The Netherlands 3Department of Aquatic Ecology, Netherlands Institute of Ecology (NIOOKNAW), Wageningen, The Netherlands 4Deltares, Marine and Coastal Systems, Delft, The Netherlands 5Aquatic Ecology and Water Quality Management Group, Wageningen University, Wageningen, The Netherlands

Introducing a wide selection of plant species typically stimulates development of target communities during ecosystem restoration, especially if this includes keystone species with important ecosystem functions (Geist & Hawkins, 2016;Lunt et al., 2013). At a given nutrient availability, species-rich plant communities are generally more productive than species-poor communities, which is known as a positive diversity-productivity relationship or overyielding (Balvanera et al., 2006;Cardinale, 2011;Hooper et al., 2005;Spehn et al., 2005). Overyielding can be caused by two primary mechanisms. First, species complementarity can increase productivity if intraspecific competition in monocultures exceeds interspecific competition in mixed communities; either because species or functional groups partition available resources (niche partitioning) or because of positive species interactions (i.e. facilitation) (Cardinale et al., 2012;Hooper et al., 2005;Loreau & Hector, 2001). Second, productivity can increase due to a selection effect: at higher diversity the chance increases that a highly productive species or functional group is present (Loreau & Hector, 2001;Wardle, 1999).
Overyielding is well-documented for terrestrial ecosystems (Hooper et al., 2005), in which its strength can depend on abiotic conditions including nutrient availability (del Río, Schütze, & Pretzsch, 2014;Schmid, 2002). In wetland ecosystems, plant functional diversity has been shown to increase plant biomass (Engelhardt & Ritchie, 2001), decrease methane effluxes (Bouchard, Frey, Gilbert, & Reed, 2007;Schultz, Andrews, O'Reilly, Bouchard, & Frey, 2011) and change nitrogen cycling (Schultz, Bouchard, & Frey, 2012). However, for wetland ecosystems, current knowledge on the diversity-productivity framework and potential mediating effects of nutrients on relations within this framework is still limited (Giller et al., 2004). Which keystone plant functional groups should be introduced, and at which nutrient levels, is largely unclear. Especially for nutrient levels, the question remains whether it is better to restore oligotrophic conditions typical for peat-forming wetland systems (Verhoeven, 1986) and the associated (red-listed) plant species (Rydin, Jeglum, & Jeglum, 2013); or to stimulate plant biomass production by providing more nutrients to initiate succession and peat formation (Lawlor, Schulze, Beck, & Müller-Hohenstein, 2010). Nutrients change many processes within plants and interactions among plants (Kraiser, Gras, Gutiérrez, González, & Gutiérrez, 2011), and may therefore also affect mechanisms such as complementarity and selection effects.
Here, we aim to provide recommendations for the restoration of peat-forming, floating fen wetland communities by evaluating the effectiveness of manipulated functional diversity of introduced plants during the initial phase of restoration across a gradient of nutrient availability. We hypothesized that (a) increasing functional diversity of introduced species would stimulate the formation of peat-forming target communities, their biomass accumulation and expansion onto open water; and that (b) nutrient availability would affect the underlying mechanisms and mediate the relative contribution of specific functional groups to these effects. We expected most biomass accumulation in the vegetation at very high nutrient levels (mediated by rapid biomass production of fast-growing, competitive helophytes such as Phragmites australis), but most expansion of the communities onto open water at low-to-intermediate nutrient levels (mediated by rapid clonal expansion of rhizomatous helophytes such as Comarum palustre). Furthermore, we anticipated that a third functional group, nonclonal helophytes, would be facilitated by the floating fen formation by either of the two other groups. We investigated this by manipulating functional diversity of experimental wetland plant communities for 2 years, after which we measured the accumulation of biomass as a proxy for vegetation carbon storage, and the formation of plant cover and rhizomes as proxies for colonization of open water.

| Experimental design
We experimentally studied the influence of functional plant diversity on biomass accumulation, cover and rhizome formation by wetland plant species over a nutrient gradient in 36 artificial outdoor ponds in Loenderveen, the Netherlands (52°12′41″N, 5°2′18″E).
The ponds were square, 1.5 m deep, 5.0 × 5.0 m wide at the top and 3.0 × 3.0 m wide at the bottom and lined with waterproof foil. Each pond was filled with a 0.3 m layer of sand-clay mixture (10:1), and a 0.7 m water column from a nearby lake ("Waterleidingplas"). This lake water was used to initially fill the ponds, and subsequently used to control the water level via an overflow mechanism. This water was oligotrophic due to phosphate removal for drinking water (measured monthly during the 2-year experimental period (n = 24): mean total N = 2.71 ± 0.42SD mg/L, mean total P = 0.008 ± 0.005SD mg/L; Waternet, unpubl. data). No fish were present in the ponds. To prevent variation among ponds in possible nutrient uptake, submerged vegetation was removed each July from ponds with a submerged plant cover >60%. The artificially created ponds enabled us to experimentally manipulate nutrient availability and the functional diversity of introduced plants without interference of existing vegetation and/or environmental conditions. We studied growth of nine typical wetland plant species in these ponds from September 2012 to September 2014. Seedlings of all species were grown from seeds in potting soil. When their aboveground parts measured ~0.05 m in height, the seedlings were divided over 126 artificial mats or rafts ("Röhrichtmatten," Bestmann Green Systems, Tangstedt, Germany) that were randomly distributed across all the ponds. Each pond received either three or five mats to ensure complete randomization of the experimental treatments. The mats (1.0 × 1.0 m) were made of a base layer of floating polyethylene strings with a coconut fibre mat fixed on top ( Figure S1). The polyethylene strings were tied together, creating a floating mat with an open structure through which roots could grow into the water. The mats did not contain any nutrients, and were kept apart by a fixation to the bottom of the ponds to prevent plant interactions between mats.
At the start of the experiment, 24 young plants were inserted into small holes cut into the coconut fibre top of each mat. Each mat received 24 individual plants belonging to one, two or three functional groups based on the classification method as initially proposed by Boutin and Keddy (1993) (Figure 1). These three functional groups differ in their functional morphology: clonal dominants (represented by Typha latifolia, P. australis and Phalaris arundinacea), clonal stress-tolerators (Calla palustris, C. palustre and Menyanthes trifoliata) and interstitials (Alisma plantago-aquatica, Iris pseudacorus and Acorus calamus). All species (except A. plantago-aquatica) are rhizomatous to some extent, but the tall clonal dominants typically have much longer creeping rhizomes than the shorter interstitials, while the much shorter clonal stresstolerators proliferate via rhizomes and creeping stems extensively.
The optimal habitat of the clonal stress-tolerators is more oligotrophic than that of the other two groups (species traits presented in Table S1). We manipulated functional diversity by regulating functional richness (i.e. the number of functional groups) instead of species diversity to make our results more widely applicable, and avoid the implicit assumption that all species are equally different from each other (Hooper et al., 2005). Our selection of three different species within each functional group makes the results per group more representative for species with that same functional role.
To investigate interactions among the three functional groups,  Table S2). Planting densities on the mats resembled realistic and cost-effective planting schemes as applied locally in restoration projects in the Netherlands.
To investigate possible effects of nutrient availability on the developing plant communities, we manipulated nutrient loadings in the ponds throughout the experiment. Nine different loadings of both nitrogen and phosphorus were applied by weekly additions of NO 3 NH 4 and KH 2 PO 4 to the water column in the ponds (range 0-5.0 mg N and 0-0.5 mg P/L, details in Table S3) from April until October in both study years. Loadings mimicked a full range from oligotrophic to hypertrophic waters. To avoid stoichiometric effects, KH 2 PO 4 and NO 3 NH 4 were added in a ratio of 1:12.6 to obtain a N:P ratio of 10:1 g/g throughout all treatments. All combinations of functional groups were exposed to the whole nutrient gradient, with two replicates per combination in every nutrient treatment ( Figure S2).
Few species other than those selected for the functional groups spontaneously colonized on the mats. In total these were <5 species, dominated by Mimulus guttatus and Bidens frondosa, with a combined fresh weight always <10% of the total vegetation fresh weight. These species were removed in May, June, August and October 2013 and April, May, June and July 2014.

| Data collection
Data were collected per plant species per mat at the end of the experiment. First, we estimated the percentage of cover on each mat.
Second, we recorded the presence (yes/no) and length (total in m) of rhizomes growing onto the open water surface (measured starting from the edges of the mats). Third, as an indicator of fen formation and ultimately vegetation carbon storage, all above-ground biomass per plant species that formed on the mats (so excluding rhizome biomass) was destructively harvested by collecting all plant material growing above the mats. All plant material was dried for at least 48 hr at 70°C, and weighed on a scale (d = 0.1 g). Most roots had grown into the polyethylene of the mat, which made it impossible to harvest root biomass representatively.

| Data analysis
We analysed how the development of vegetation on the mats was af- The mixed-effects models were fitted with one of four possible dependent variables: (a) biomass, (b) cover, (c) rhizome presence and (d) rhizome length. Residuals were normalized by natural logtransformations of biomass, cover and rhizome length, and analysed using package "nlme" (Pinheiro, Bates, DebRoy, & Sarkar, 2015). The presence of rhizomes was analysed as a binomial dependent variable using package "lme4" (Bates, Mächler, Bolker, & Walker, 2015).
Nutrient loading into the ponds (ranging from 0 to 5.0 mg N/L and 0 to 0.5 mg P/L) was included as continuous predictor variable numerically ranging from 0 to 5, hence, estimated effect sizes in the models are presented on the scale of nitrogen loadings. Nutrient loading was centred by subtracting the mean from all values to improve interpretability (Raudenbush & Bryk, 2002). Intercepts in all models were allowed to vary by pond by including individual pond (36 levels) as random factor.
In Model I we tested whether cumulative values per mat calculated for either one of the four vegetation variables responded In addition to the mixed models, we applied randomization tests to quantify the effects of functional groups expressed in standard deviation units on three variables: total plant biomass, total plant cover and total rhizome length per mat (Gotelli, Ulrich, & Maestre, 2011). Our working hypothesis was that the presence or absence of a particular functional group significantly affected all three of these variables. The null hypothesis was that mats with and without a particular functional group would not differ more than expected by chance. This method has been developed for species, but we here applied it analogously to functional groups. More details can be found in Supporting Information Methods.
To disentangle possible complementarity and selection effects as underlying mechanisms explaining observed net diversity-productivity effects, we used the additive partitioning technique as described by Loreau and Hector (2001). For every mixture and for every nutrient level we calculated the net effect, the complementarity effect and the selection effect. More details can be found in Supporting Information Methods.

| Effects of functional diversity and nutrient loading
Mean biomass (dry weight in g), cover (in %) and the probability of rhizome formation on the mats increased with functional richness and nutrient loading (Table 1, Figure 2, Table S4). The positive effects of increasing functional diversity were consistent across different levels of nutrient loading. The vegetation transgressively overyielded, that is, the maximum biomass and cover produced by the best mixed community (1,894 g dry weight and 100% cover per mat for mats with all three functional groups) exceeded the maximum biomass and cover produced by the best monoculture (1,288 g for the monoculture with interstitials and 90.5% cover per mat for the monoculture with clonal stress-tolerators). Biomass production, cover formation and the probability of rhizome formation on the mats increased with the number of functional groups that were present (Figure 2a-c, Model Ia, b and c in Table 1, respectively). Rhizome length did not increase due to functional diversity (Figure 2d, Model Id in Table 1).
Nutrient addition had a much stronger effect than functional diversity on final plant biomass (Figure 2a-c, see also Figure S1).
Mean biomass of all mats was 40 g when no nutrients were added, which increased 26-fold to 1,048 g at the highest nutrient loading.
Nutrients affected coverage less than biomass, but again had a stronger effect on coverage than functional diversity. Mean cover ranged from 15% at the lowest nutrient level to 72% cover at the highest nutrient level (a fourfold increase). At the lowest nutrient level, rhizomes formed on 7 of 14 mats (50%), and at the highest nutrient level on 11 of 14 mats (79%); a relative increase by 58%.
Furthermore, nutrient loading increased rhizome length >19-fold, from 0.90 m ± 0.45SD at the lowest, to 17.9 m ± 17.0SD at the highest nutrient level (Figure 2d).

| Relative contributions of the three functional groups
The three functional groups differed in their relative contributions to the vegetation that had formed on the mats after 2 years, and this relative importance depended on nutrient loading (Figure 3).
Summed over all nutrient levels, biomass formed by clonal dominants was 11.7% of all formed biomass, which was less than the biomass formed by the clonal stress-tolerators (35.2%) or the interstitials (53.1%). At low nutrient loadings clonal stress-tolerators

| Complementarity and selection effects among the functional groups
The positive effect of functional richness on biomass accumulation and cover formation could almost entirely be attributed to a species complementarity effect (Figure 4a,b). The positive TA B L E 1 Final model-averaged parameter estimates (β), their standard errors (SE), 95% confidence intervals and p-values for significant terms remaining in the models after model selection (for details on model selection see   Table S5). Effects were in most cases stronger on mats with three functional groups than on mats with two functional groups (Table S5). For biomass production, the mean proportion of the net effect that was explained by the complementarity effect was 1.00 ± 0.07SD, the selection effect was close to zero, and this pattern was largely consistent over the different nutrient loadings (Table S5). For cover formation, the proportion explained by the complementarity effect was 0.99 ± 0.04SD and the selection effect was again close to zero. For the presence of rhizomes, the selection effect did explain a large proportion of the net effect: the net effect consisted of 0.25 ± 0.24SD selection, and 0.15 ± 0.50SD complementarity effect across all nutrient loadings and diversity levels (details in Table S5).
A more detailed investigation of possible positive interactions among functional groups underlying the complementarity effect indicated both facilitation and competition (Models II, III and IV in Table 1 and Table S4). Cover formation by interstitials increased by 18% if clonal dominants were present. Clonal stress-tolerators facilitated cover formation by clonal dominants (+109%) and interstitials (+29%). The presence of interstitials increased cover formation by clonal dominants (+52%), but decreased the chance that this functional group formed rhizomes (−50%). Our observation of overyielding in mixed wetland communities is in line with the general diversity-productivity framework in terrestrial ecosystems (Balvanera et al., 2006;Cardinale, 2011;Hooper et al., 2005;Spehn et al., 2005) and confirms previous studies in wetland ecosystems (Engelhardt & Ritchie, 2001;Schultz et al., 2011Schultz et al., , 2012. Our study expands the available knowledge for wetland systems by using different species and a different approach, and-most importantly-specifically tests the mediating role of nutrient availability in the diversity-productivity F I G U R E 4 This figure visualizes the net, complementarity and selection effects of functional diversity on (a) biomass production in grams per mat, (b) percentages of mats covered by vegetation and (c) the presence of rhizomes per mat (1 = rhizomes, 0 = no rhizomes), and how they change over the gradient of experimental nutrient loadings (in mg/L). In general, an increase of diversity to three functional groups (grey lines) affected the vegetation parameters stronger than an increase to two functional groups (black lines); and the strength of the effects increased with nutrient loadings. For biomass and cover formation, the net effect could almost completely be explained by the complementarity effectover the entire gradient of nutrient loadings. For rhizome presence, the selection effect was the most important mechanism. The vertical axis is natural log-scaled for biomass and cover (a and b), and a binomial scale for rhizome presence (c). Statistical details are available in Table S5 0

| Complementarity and selection effects among the functional groups
Productivity in our experimental wetlands increased with functional diversity due to both complementarity and selection effects.
Complementarity was the dominant mechanism causing overyielding for biomass accumulation and cover formation in the mixed communities, and the selection effect for the presence of clonal stress-tolerators was the most important for rhizome formation.
Observing a complementarity effect suggests that there was interspecific resource partitioning among the three functional groups, or that the groups facilitated each other (Hooper et al., 2005). Because positive interactions (facilitation) were much more common than negative interactions (competition), facilitation provides a likely explanation for the observed complementarity effect.
Observing facilitation is in line with some previous studies (Le

| Floating fen restoration
Difficulties with restoration of floating fens are a problem because of the rapid losses of this important ecosystem type world-wide (Chimner et al., 2017;Lamers et al., 2015). Even after abiotic conditions have been restored in degraded systems, propagules of target plants need to arrive (Soomers, Karssenberg, Soons, et al., 2013) and establish (Sarneel & Soons, 2012) at suitable shallow shorelines for rhizomatous growth to expand onto open water (Sarneel, Huig, Veen, Rip, & Bakker, 2014;Sarneel et al., 2011). Our study shows the added value of introducing multiple, carefully selected target species during restoration, and the relevance of nutrient availability for these plants during the initial years of restoration projects. This knowledge can be applied when selecting plant species for introduction and when determining whether or not to manipulate nutrient levels during wetland restoration practises. For example, the process of nutrient reduction need not necessarily be finished before species are reintroduced at a site, as the availability of nutrients appears not to be a limiting factor for restoration success during the establishment phase.
After this phase, nutrient levels should be more tightly managed to ensure development and persistence of target species known to respond negatively to nutrient-rich conditions (e.g. Lamers et al., 2015).
Our experiment did not fully resemble the natural field situations. Two important differences between our experiments and field situations are (a) the absence of herbivores such as waterbirds, which can severely reduce expanding vegetation in fen systems (Dingemans, Bakker, & Bodelier, 2011;Sarneel et al., 2014), so that the effect of any species reintroductions is likely to be strongly reduced; and (b) a lack of water flow around the mats. In most field situations, debris, sediments and plant seeds will become trapped in expanding rhizomes analogous to the way described in tussocks (Ervin, 2009). This could provide a suitable substrate for new seedlings or other species to establish and expand, thereby contributing to the developing community. Addressing these additional aspects in long-term field evaluations could further improve wetland restoration practises.

| Conclusions and implications for wetland restoration
This study implies that for restoration of the initial succession stages of peat-forming ecosystems, increasing plant functional diversity in peat-forming ecosystems can accelerate community development during restoration of the initial stages, both as establishment of new ecosystems or expansion of existing systems.
Community biomass accumulation, cover and rhizome formation all increased with functional group richness. Peat-forming communities can benefit from facilitation among different functional groups, and initially develop under a surprisingly wide range of nutrient availabilities. The observed facilitation effects suggest that increasing functional diversity can stimulate terrestrialization and peat formation. However, apart from facilitation we also observed a positive selection effect. In our experiment, the impact of adding clonal stress-tolerators to mats was very large. Clonal stress-tolerators acted as a keystone functional group for the colonization of open water, and importantly contributed to the early phase of fen restoration.