Methods

Site

In 2015, we established PSF and diversity-productivity experiments in the Jena Experiment field site on the floodplain of the Saale River, Jena, Germany with eutric fluvisols (5 - 33 g C kg-1and 1.0 - 2.7 g N kg-1 soil; Roscher et al., 2004; Weisser et al., 2017). Long-term mean annual temperature and precipitation at the site are 9.8° C and 544 mm (2002 - 2018), respectively, and during the experiment (2015 - 2018) mean annual temperature and precipitation were 10.4°C and 499 mm, respectively (Kolle, 2020). The first and last year of the experiment (2015 and 2018) were drier than average, 459 mm and 395 mm, respectively, while 2017 was wetter than average (615 mm).

Field Experiments

The PSF experiment followed a two-phase, factorial bio-assay approach (Bever, 1994; Brinkman et al., 2010). This design is considered one of the most robust PSF experimental designs, because it measures plant growth in the field, on each soil type without mixing soils (Kulmatiski and Kardol 2008; Rinella and Reinhart 2018). In Phase I, monocultures of each plant species were grown for two years to create soils with a known plant cultivation history (nine soil treatments). Plants were then removed. In Phase 2, each plant species was grown for two years on replicate plots with each plant cultivation history (Bever, 1994; Brinkman et al., 2010; Rinella & Reinhart, 2018). The diversity-productivity experiment, which we will refer to as the current diversity-productivity experiment, was designed to replicate the dominance experiment established at the study site in 2002 which we will refer to as the pre-existing experiment (Roscher et al., 2004). In both the current and pre-existing experiments, plant communities with 1, 2, 3, 4, 6, and 9 species were grown for four years. Plant species included five grass species: Alopecurus pratensis ,Arrhenatherum elatius, Dactylis glomerata, Phleum pratense, Poa trivialis; two tall herbs: Anthriscus sylvestris, Geranium pratense; and two legumes: Trifolium pratense, Trifolium repens(Roscher et al., 2004).
In Fall 2014, a 75 x 22 m area was mowed, sprayed with glyphosate herbicide (Roundup ® 0.045 % v/v pelargonic acid; Evergreen Garden Care Österreich GmbH, Salzburg) and tilled using several passes to 30 cm with an agricultural cultivator. For the PSF experiment, a grid with 1,251 plots was created. To isolate each 35 cm-wide by 75 cm-long plot from each other, a 10 cm-wide by 35 cm-deep trench was dug around the outside of plots, and a custom-made, flat-bladed shovel was used to slice soils between plots to allow insertion of a 35 cm deep root barrier (RootBlock® 1 mm high density polyethylene; GreenMax, Netherlands). Each of the nine target species was randomly assigned to 139 replicate plots. In March 2015, seeds (4 g m-2) were applied by hand for one species in each plot. Prior to seeding, seeds of Anthriscus sylvestris were stored at -20 °C for two weeks (Roscher et al., 2004). Due to poor establishment, Anthriscus sylvestris and Geranium pratenseplots were reseeded in October 2015 with 2,000 germinating seeds [7.5 g m-2 and 28.3 g m-2, respectively; germination rates based on Roscher et al. (2004)]. Non-target species were removed by hand at least three times each growing season, and, consistent with other experiments at the site, aboveground biomass was harvested and removed each spring and fall as is typical for hay meadows in Central Europe (Roscher et al., 2004).
Phase 1 ended in September 2016, when standing biomass was removed and plots were treated with herbicide. Roughly two weeks later, plots were hand-tilled to prevent sprouting of Phase 1 species. Plots were randomly assigned so that each plant species was grown in 14 replicate plots that had grown the same species in Phase 1 (i.e., ‘home’ soils) and 15 replicate plots that had grown each of the other species in the experiment in Phase 1 (i.e., ‘away’ soils). Five replicate ‘home’ plots remained unseeded to assess the extent of resprouting growth in ‘home’ plots. It is important to distinguish new growth from resprouting, because resprouting growth would result in inappropriately positive PSF values. Mean resprouting growth varied in these control plots varied from 0 to 32 g m-2 and was removed from final biomass estimates in ‘home’ plots. On 15 March 2017, 2,000 pure live seeds m-2 were applied by hand to each PSF plot. In October 2017 and June-July 2018, biomass from Phase 2 plots was clipped to 5 cm above soil surface by hand, dried to constant weight at 70 °C, and weighed.
The current diversity-productivity experiment included 223 plots (1.5 m by 1.5 m), also lined with root barriers. Monocultures were replicated three times (9 species x 3 replicates = 27 plots). Each of the 91 plant communities grown in the pre-existing experiment was grown in one plot (91 plots; Supplemental materials; Roscher et al. 2004). Additionally, six randomly selected communities of two, three, four, and six species mixtures, were replicated in four additional plots (4 species richness levels * 6 communities* 4 replicates = 96). Communities with all nine target species were replicated in nine plots. In March 2015, target seed mixtures with 2,000 pure live seeds per m2, equally distributed among species, were applied by hand to each plot. Again, aboveground biomass was clipped to 5 cm above soil surface in November 2015, June and October 2016 and 2017. A subsample of 0.1 m2 per plot was sorted by plant species, dried at 70°C for three days and weighed. As in the PSF experiment, non-target species were removed by hand at least three times per year from 2015-2017.

Calculating Plant-Soil Feedbacks

PSF data were primarily used to parameterize plant community models, but to provide values that are comparable to other studies, PSF values are reported. PSFs were calculated as the difference of growth on ‘home’ and ‘away’ soils divided by the maximum of ‘home’ and ‘away’ soils. Similar to the log response ratio, this calculation produces values bound by -1 and +1, but these values are readily interpreted as the proportion of increased or decreased growth (Brinkman et al., 2010; Kulmatiski et al., 2012). Soil-level PSF values describe the growth of each plant species on each of eight ‘away’ soils resulting in 72 soil-level PSF values (i.e., eight values for each of nine species). Species-level PSF values describe the growth of each species across the other eight soil types resulting in nine PSF values. In both cases, PSF values and associated 95% confidence intervals were calculated using 2000 bootstrapped biomass on home and away samples (Schittko et al. 2016; Kulmatiski et al., 2017). PSF values with confidence intervals that do not overlap zero are considered positive or negative, as appropriate.

Calculating Relative Competition Intensity (RCI)

A goal of this research was to test whether or not PSFs improve predictions of plant growth in communities where other factors such as a plant’s competitive ability are also important (Lekberg et al. 2018). To better understand how PSF may interact with a plant’s competitive ability, we correlated PSFs with the relative competition index (RCI, Weigelt & Jolliffe, 2003) where RCI = (monoculture biomass – twice the two-species mixture biomass)/monoculture biomass. A low RCI indicates higher biomass production of a species in two-species mixtures than in monocultures (strong competitor). RCI was calculated for current and pre-existing diversity-productivity experiments.

Simulating Plant Growth in Communities

A suite of plant community growth models was parameterized with (PSF) or without (Null) PSF data (Kulmatiski et al., 2016). The models and modeling approach generally follow that of Kulmatiski et al. (2016), but briefly, the foundation of these models are logistic growth equations (equation 1). In addition to the effects of intrinsic plant growth rates r, total plant biomass in the community P, and a carrying capacity Κ, plant growth is also a function of soil conditions σ. Plants are assumed to change soil conditions as they grow, and plants grow at different rates on different soil conditions (Bever, 1994; Kulmatiski et al., 2011). In Null models, plant growth rates are the same across all soil conditions. Plants can ‘compete’ indirectly through carrying capacity, but competition coefficients were not included. Models were parameterized with different carrying capacities, data from different years (input data), and with different values for ‘neutral’ soils to produce a suite of simulations. Average biomass predictions from this suite of model parameterizations are reported. The goal of this modeling is to simulate relatively short-term plant growth in the field experiment and not to determine equilibrium species abundances (Kulmatiski et al. 2011; Feng et al. 2020).
In the PSF models, each plant species i conditions soil j, and therefore has a soil-specific growth rate ri,j. The biomass of plant species i at time t (Pi,t), depends on its growth rate at t (ri,t) and is limited by either community-level carrying capacity Κ alone (equation 1) or additional species-level carrying capacity κi (equation 2). At the community-level, Κ simulates interspecific competition, but ‘competitive strength’ is only defined by growth rates: Κ is defined as the maximum biomass a community can achieve. Whereas at the species-level, κi simulates intraspecific competition: κi is defined as the maximum biomass a species can achieve. The time- and plant-specific growth rate ri,trepresent the summed product of soil-specific growth rates ri,j and the proportion of soil at time\(t\)j,t; equations 3 and 4). Assuming gradual change of soil conditions as plants grow, we estimate growth rate on unconditioned soil (‘neutral’ growth rate, νi) and set the abundance of neutral soil to one (100%) at t=0 (equation 3). While plants grow, neutral soil is subsequently replaced by conditioned soil.