The experiment took place between 2002 and 2005 in a 0.3 ha experimental wetland on the campus of Southeastern Louisiana University (30° 31.4' N, 90° 28.4' W; 11 m elevation; Campbell et al. 2016). Mean annual temperature was 19.3oC, with January and July means of 9.5oC and 27.6oC, respectively, and mean annual precipitation was 159.1 cm (1981–2010 normals, Hammond 5E, LA, 10 km E; NCEI 2020). From 2002 to 2005, annual temperature and precipitation were within one standard deviation of these normals, except for 2003, which received 193.6 cm of rain.
We smoothly graded the northeast and southeast shorelines to a 33% slope (Fig. 1). We built 75 0.5 m wide lanes with lumber, perpendicular to the shoreline, grouped into five blocks. Each lane extended 3.66 m along the slope into the pond and was separated from adjoining lanes by two 4 m by 25 cm by 8 cm planks and wooden walkways. Since many wetlands along the Gulf Coastal plain arise in floodplain sediments deposited by rivers, and also occur in loess landscapes, we aimed for relatively fertile conditions, and therefore also spread a ~ 10 cm layer of topsoil, bringing the surface ~ 5 cm below the timber framework. We divided each lane into 0.5 by 0.25 m cells along the elevation gradient into which we planted plants.
We selected ten species of herbaceous perennial emergent plants classified as obligate wetland species in the Atlantic and Gulf coastal plains (USACE 2018; Table 1), so they occur over 99% of the time in wetlands. These species have wide geographical distributions and are also abundant locally in wetlands (Penfound and Hathaway 1938; Chabreck 1972). We purchased plants from Louisiana Growers (Amite, LA) in 1 L containers or as bare root seedlings (Sagittaria lancifolia and Pontederia cordata) and planted them from June-November 2002 into each cell. We planted them as (1) monocultures (ten species, n = 5); (2) mixtures with all ten species planted in each cell, with two multi-species mixture treatments per block (n = 10) and (3) three mixtures of two species (n = 5). For pairs, we chose (i) Cladium/Panicum, two graminoids that we judged to be likely ecological dominants; (ii) Peltandra/Pontederia, two clump-forming species that we judged were likely interstitial species (sensu Boutin and Keddy 1993); and (iii) Panicum/Pontederia, a mixture of a presumed dominant and a presumed interstitial species.
All plants remained non-flooded until spring 2003 to allow them to establish. We weeded the lanes periodically to remove non-planted species.
We began flooding cycles in April 2003. The lowermost cell was continuously flooded during the growing season, while the uppermost cell remained unflooded. We artificially manipulated water levels over the eight remaining cells in a cyclic fashion along the elevation gradient by progressively raising water levels by stages, each stage lasting approximately one week. In a typical year, this meant three flooding pulses over the growing season. Cells consequently received 0, 11, 22, 33, 44, 55, 67, 78, 89, and 100 percent flooding over approximately 6.5 months during the growing season. Our flooding regimes created a strong redox potential gradient, with a rapid transition from aerobic to anaerobic soil conditions at the pond’s water line (Campbell et al. 2016). At the end of the growing season in October, we left water levels permanently low until the following spring. On occasion, large precipitation events altered water levels, but the normal cycle was re-established after a few days. The experimental pond had low fetch (< 60 m), so there was no wave erosion associated with our flooding gradient. In late August 2005, a large oak fell into the experiment during Hurricane Katrina, destroying three of the lanes in one block and our ability to manipulate the water level, so we ended the experiment in the autumn 2005.
To avoid destructive sampling, we visually estimated the cover of each species in each cell prior to flooding (November 2002) and for three growing seasons after flooding was initiated (September 2003, October 2004, September 2005). In November 2005, we harvested the aboveground biomass, sorted by species and elevation, dried it to a constant mass at 80oC and weighed it.
We analysed the data with R (version 4.0.2) using linear or generalized linear mixed-effects models with the packages lme4 and glmmTMB (Bates et al. 2015; Brooks et al. 2017). We evaluated competing models with AIC and checked residuals using DHARMa (Hartig 2020; code available in Supplemental Materials).
We first examined the number of surviving species along the flooding gradient for each year separately. We compared the number of surviving species per block in the ten monoculture lanes grouped together against the two mixture lanes as a function of the presence and absence of neighbours, with the extent of flooding as a repeated measure. We used a generalized Poisson distribution because of problems of over or under dispersion.
We then evaluated the changes in distribution of species with flooding using four metrics, based on our data for cover in each cell in autumn 2003, 2004 and 2005: (i) the minimum and (ii) the maximum flooding limit at which species survived in a lane, (iii) the range of flooding from the difference in maximum and minimum flooding tolerance, and (iv) the mode of the flooding distribution at which peak cover occurred. We minimally rescaled these variables, so they fell just above 0 and below 1 (Smithson and Verkuilen 2006), and we tested the overall effect of neighbours on each of these rescaled metrics, with the presence/absence of neighbours as the main effect and year as a repeated measure within lanes, while ignoring the identity of species. If a species did not survive in mixture, we removed the species from this overall test, leaving ten species in 2003, eight in 2004 and only five in 2005. We evaluated models with a beta distribution with or without zero augmentation and with constant or variable dispersion (Douma and Weedon 2019). We then examined for effects of the presence and absence of neighbours and species on these metrics, separated by year because of the decreasing survival of species in the mixture plots, with blocks as the random factor, and with the same beta distribution models. We used Tukey tests to examine post hoc differences.
To further evaluate the importance of competition versus facilitation along the flooding gradient, we also determined the additive interaction importance index (NImpA; Diaz-Sierra et al. 2017) using the final biomass data. NImpA is a standardized and symmetric index that examines the difference of a species’ performance between mixtures and monocultures, relative to the species’ best performance. The largest biomass per cell across all blocks was used as to determine a species’ best performance. We analysed NImpA separately for each species against flooding levels, with block as a random variable.
The analysis of the two species mixtures was similar. We could not conduct analyses of survival because of problems of model convergence or lack of variance at many flooding levels. We analysed the flooding statistics (minimum, maximum, range and mode) for each pairing by the presence or absence of neighbours, species and year within lanes, again using beta distributions. We also again calculated and analysed NImpA along the flooding gradient.