Alternative transient states and slow plant community responses after changed flooding regimes

Abstract Climate change will have large consequences for flooding frequencies in freshwater systems. In interaction with anthropogenic activities (flow regulation, channel restoration and catchment land‐use) this will both increase flooding and drought across the world. Like in many other ecosystems facing changed environmental conditions, it remains difficult to predict the rate and trajectory of vegetation responses to changed conditions. Given that critical ecosystem services (e.g. bank stabilization, carbon subsidies to aquatic communities or water purification) depend on riparian vegetation composition, it is important to understand how and how fast riparian vegetation responds to changing flooding regimes. We studied vegetation changes over 19 growing seasons in turfs that were transplanted in a full‐factorial design between three riparian elevations with different flooding frequencies. We found that (a) some transplanted communities may have developed into an alternative stable state and were still different from the target community, and (b) pathways of vegetation change were highly directional but alternative trajectories did occur, (c) changes were rather linear but faster when flooding frequencies increased than when they decreased, and (d) we observed fastest changes in turfs when proxies for mortality and colonization were highest. These results provide rare examples of alternative transient trajectories and stable states under field conditions, which is an important step towards understanding their drivers and their frequency in a changing world.


| INTRODUC TI ON
Flood frequency worldwide is projected to increase in 42% and decrease in 18% of the global land area until the end of this century (Hirabayashi et al., 2013). At the same time, 90% of the global river volume is impacted by flow regulation (Grill et al., 2015). This has large consequences for riparian vegetation and associated ecosystem services like water purification, bank stability and biodiversity (Schulz et al., 2015). Natural flow regimes are therefore increasingly being restored, which often also induces large changes in the flow regime. In order to quantify and predict how and how fast changes in flooding affect plant community composition, more knowledge | 1359 SARNEEL Et AL. on the underlying processes is a priority. A difficulty in such studies (and similar ones in other systems) is determining the time needed for communities to adjust to new hydrological conditions (Walker et al., 2006), since this is often hampered by confounding factors (e.g. overruling effects of dispersal barriers) (Muller et al., 2014), insufficient time for follow-up studies (Hasselquist et al., 2015) or limited spatial setups of greenhouse and mesocosm studies (Garssen, Baattrup-Pedersen, Voesenek, Verhoeven, & Soons, 2015;Garssen, Verhoeven, & Soons, 2014;Webb, Wallis, & Stewardson, 2012).
Plants need specific adaptations, like adventitious root formation or aerenchyma to survive prolonged flooding (Voesenek, Colmer, Pierik, Millenaar, & Peeters, 2006). Differences between species traits determine their flooding tolerance and create a clear zonation along riparian elevation gradients (Fraaije, Ter Braak, Verduyn, Verhoeven, & Soons, 2015;Van Eck, Lenssen, Van De Steeg, Blom, & De Kroon, 2006;Webb et al., 2012). Therefore, after changes in flooding regime, drought-or flood-tolerant species will be replaced by others that are better adapted to the new conditions (Ström, Jansson, Nilsson, Johansson, & Xiong, 2011) and induce changes in the riparian zonation (Antheunisse & Verhoeven, 2008;Banach et al., 2009;Ström, Jansson, & Nilsson, 2012) and functioning (Lake, Bond, & Reich, 2007;Sarneel & Veen, 2017). However, while it is generally accepted that vegetation responses mostly are slow and that some intermediate stage may persist for a reasonable amount of time (Baastrup-Spohr, Sand-Jensen, Olesen, & Bruun, 2017;Hasselquist et al., 2015;Nilsson et al., 2015b), we lack understanding of how and how fast responses occur (Webb et al., 2012). Less is known about how riparian plants survive drought events, but root and leaf traits are thought to play a crucial role (Li et al., 2014). In a meta-analysis, Garssen et al. (2014) conclude that a drought events longer than one month affect riparian seedling survival and plant growth negatively. Furthermore, extreme drought events have been shown to induce sudden changes in biodiversity by causing plant mortality, which is thought more likely in warmer than in colder climates (Garssen et al., 2014). Some indirect empirical evidence indicates that the duration and species composition of transitional stages relate to the strength of environmental filtering of the new conditions (e.g. stress level) and to initial vegetation characteristics (historic factors). For instance, plant species distribution along the river Rhine was found to reflect an extreme flood 14 years in the past (Van Eck et al., 2006). This suggests (a) that the vegetation drastically and quickly changed after this strong disturbance, and (b) that recovery was slow during the low-stress period afterwards. This aligns well with the hypothesis that the magnitude of stress affects both the magnitude of change and the rate of recovery or convergence to new conditions, but longterm experimental studies are needed to shed light on how mortality and invasion could drive such changes. The rate of vegetation change can further be driven by other time and soil moisture-dependent processes such as the speed of nutrient release due to changed redox potentials (Beltman, Willems, & Gusewell, 2007) or the rate of groundwater-table drawdown in case of drought events (Froend & Sommer, 2010). In addition to abiotic conditions, vegetation composition can also determine the amount and rate of change. This was demonstrated within the Jena biodiversity experiment, where plant performance directly after flooding was less affected in high diversity plots (Wright et al., 2017).
Recent work, however, suggests that stochastic factors, like historic contingency, can interfere with the straightforward replacement of sensitive with tolerant species and can result in alternative transitional stages, or alternative stable states (Fukami & Nakajima, 2011;Maren, Kapfer, Aarrestad, Grytnes, & Vandvik, 2018;Sarneel, Kardol, & Nilsson, 2016;Stuble, Fick, & Young, 2017;Vannette & Fukami, 2014). For instance, in heathlands, replicate turfs followed different routes before converging to a similar vegetation type 7 years after fire disturbance (Maren et al., 2018). The rapidly developing theoretical framework behind alternative transitional stages suggests that the duration of the transient period could be correlated with mortality (Fukami & Nakajima, 2011). This is because mortality, induced by the changed stress level or by changed biotic interactions, will open patches which allow invasion of potentially better adapted species (Catford & Jansson, 2014), but field proof is lacking. To explore pathways of change after various changes in flooding regimes, we set up a long-term experiment and transplanted vegetation turfs full-factorial between three riparian elevations that correspond to different vegetation zones (upland border, middle and low elevation). Given the strong environmental forcing in riparian zones and the clear vegetation zonation within them, we hypothesize that the transplanted vegetation will eventually converge towards the new (target) vegetation, potentially via alternative transitional stages. Since target vegetation occurs directly adjacent to the transplanted turfs, dispersal will not strongly limit the rate of vegetation change. We hypothesize that due to the absence of strong dispersal barriers, the rate of vegetation change will be determined by factors that determine mortality (i.e. stress level) and or invasion (germination, establishment) and that response rates of individual species reflect their flooding tolerance, which we quantified using Ellenberg moisture values.

| Study area
The Vindel River is a ca. 455-km-long, free-flowing river that flows from the Norwegian border across Sweden into the Baltic Sea ( Figure 1a). The Vindel River has a mean annual discharge of

| Experimental setup
In June 2000, a total of 72 soil turfs were excavated, in eight repli-

| Measurements
In each turf, we determined species composition once a year, always maximum cover of 16). Species that were present in the quadrat but did not touch a pin were also recorded. We followed the taxonomy in Krok and Almquist (1994) and combined the following species to one taxon each to minimize observer effects: Agrostis stolonifera +

| Data analysis
Trends in vegetation development were analysed using nonmetric multidimensional scaling (NMDS) and the Vegan 2.4-3 package (Oksanen et al., 2012) in R version 3.2.5 (R Core Team, 2016). We used the pin-point scores of all years and all replicate turfs as input and species that were present in a turf but did not touch a pin were given the value 0.5. We used a Bray-Curtis distance and added "no. share = 0.1" to deal with the low overlap of species in some turfs. We extracted four dimensions and found convergence after 199 runs, with a stress of 0.105. We used two-way ANOVA and Tukey's post hoc tests to test if the NMDS scores of the transplanted turfs at the first two axes differed from the controls at initial and target elevation.
To be able to compare the movement in NMDS space and the degree to which convergence to the target conditions had occurred in the different treatments on a similar scale, we standardized the moved distances by total trajectory length of a treatment. To calculate this, we first calculated both the shortest Euclidean distance between a turf (i) and the centroids of controls at the initial (Dist INi ) and at target vegetation (Dist TAi ). These distance measures were summed to obtain the total trajectory length between initial and target vegetation. Subsequently, we divided Dist INi by the total trajectory length to obtain the relative distance (rDist) moved away from the initial vegetation composition. In summary, rDist = Dist INi /(Dist TAi + Dist INi ).
This allows comparing the degree of converging to target conditions in trajectories that differed in absolute distance in NMDS space.
The rate at which turfs moved along the total trajectory between initial and target elevation was expressed as the difference in rDist at two sampling occasions, divided by the number of years between those sampling occasions. Thus, the rate of change = abs(rDist t+1 − rDist t )/(Year t+1 − Year t ). Both rDist (%) and rate of change (% yr −1 ) were calculated per turf and averaged per treatment and sampling interval combination (n = 8).
For each species in each turf, we calculated the rate of cover change as the regression coefficient of the pin-point scores versus time (years). To deal with the infrequent and low occurrence of rare species, we selected species that (a) occurred in three or more years in a certain turf, and (b) increased to or decreased from a cover of three or more. We averaged values per species for the three treatments that were moved in the same direction along the elevation gradient and correlated those with the Ellenberg value for moisture, which we obtained from the Online British Flora (https://www.brc. ac.uk/plantatlas/). We could not obtain an Ellenberg value for only eight species, of which only one species was frequently occurring in the turfs (Salix spp). To not overlook this species, we gave this species the Ellenberg value of Salix phylicifolia (Ellenberg moisture value = 8) as this species was most frequently reported within this aggregated species group.
Colonization and mortality events drive vegetation change.
However, there may be a mismatch between the two and sometimes
The NMDS diagram shows clearly that directly after transplantation, the vegetation composition of the transplanted turfs resembled the vegetation at the initial elevation ( Figure 2 and Table S1 for statistics). Over time, major species shifts occurred in the trans-  Table S1). For these transplanted turfs, their movement in NMDS space during the last 4-5 years (Figures 2a, 3a), may even suggest the development of an alternative stable state, as it does not seem to move closer to the target centroid. Here, low-elevation species (mainly Carex acuta and Calamagrostis canescens) were still present with relative high cover ( Figure 4).

| Rate of change
Most turfs appear to change rather linearly, that is, with a constant rate over time, except for the turfs that were moved from high to low elevations where changes were faster during the first years of the study (Figure 3), and slower during later years. The turfs that were moved down in elevation generally changed faster and came to resemble the target vegetation earlier compared to the turfs that were moved in the opposite direction (Figure 3). The vegetation in the turfs that were moved from middle to low elevations were the first to convert from initial the vegetation to the target elevation (after six growing seasons). In some cases, we observed retrogression, where turfs became more similar to the initial vegetation (Figure 3b,c), but these changes did not last longer than 1 year.
Changes in the proportions of species with affinity for a certain soil moisture level confirmed that turfs transplanted from the upland border to low elevation adjusted more quickly to the target vegetation than turfs transplanted from low elevation to the upland border ( Figure 4). In 2018, species with a high affinity for soil moisture (Ellenberg values > 8) contributed on average 30.7% ± 10.1 S.E. of the total cover of the turfs that moved from low elevation to the upland border, whereas such species are almost absent in the control turfs at the upper border ( Figure 4). Cleary, the relative cover of species with intermediate affinity for soil moisture increased in turfs that moved from the upper border to low elevation, but after ca.
three to four growing seasons their proportion decreased to values comparable to the control turfs at the target elevation (Figure 4).

| Species driving changes in vegetation patterns
Across all the transplanted turfs we observed that species that were best adapted to the new (target) conditions were the ones that appeared and increased vigorously and fast in cover, while the least adapted species decreased. That is, in turfs that were moved to less flooded conditions, the cover of species with high Ellenberg values decreased quickly while species with low Ellenberg values increased (F 1,15 = 17.79, P = 0.001, R 2 = 0.54, Figure 5a). In contrast, in turfs that were moved to more frequently flooded conditions, species F I G U R E 4 Proportion of the plant community with specific moisture preferences (Ellenberg moisture values) in the control (left column) and transplanted turfs based on the pin-point scores. Panels are grouped per initial elevation (columns) and target vegetation (rows). Each bar is the average of eight replicates with low Ellenberg values disappeared and the cover of species with high Ellenberg values increased, but this correlation was not significant (F 1,20 = 3.16, P = 0.09, R 2 = 0.14, Figure 5b). We observed that species that are associated with succession changed disproportionally to their Ellenberg value, for example, Betula pubescens, Salix spp.
and Equisetum pratense increased, while Deschampsia cespitosa decreased. These trends were also observed in the control turfs where the Ellenberg value and the species-specific rate of cover change did not correlate (F 1,22 = 0.12, P = 0.735, R 2 < 0.01, Figure 5c).
At the community level, the cover loss per year of all the species that decreased in cover in a turf correlated with the rate of change in NMDS space (Pearson ρ = −0.59 t 46 = −4.92, P < 0.001; Figure 6).

| D ISCUSS I ON
We found that past flooding regimes can still affect community composition after a period of 19 growing seasons. Although individual species increased and decreased predictably following their preferences and adaptations for soil moisture, strikingly different vegetation communities were observed for intermediate transition stages.
These results provide rare experimental proof that alternative transient trajectories and stable states can occur under field conditions, which is an important step towards understanding the drivers behind community response to global change.
In many zoned ecosystems, such as riparian and tidal ecosystems, the lower boundary of a species is hypothesized to be determined by stress tolerance, while the upper border is formed by competition (Bertness & Callaway, 1994). We therefore assume that for turfs that were moved to more frequently flooded locations, vegetation changes could be driven by (sudden) plant death opening space for invasion, while when moving to the higher, drier locations, where mortality is lower, competition could be the driving force (Maren et al., 2018). Our observations support this idea, with vegetation responses to moving downward in elevation proceeding more rapidly compared to moving to higher elevations, and the correlation between rate of vegetation change and our proxies for mortality being steeper than with our proxies for invasion. Increased flooding stress levels likely had more direct and larger effects on vegetation composition than increased competitive pressure, thus leading to more gradual changes at higher elevations. The observation that the fastest convergence to target vegetation (six growing seasons), the turfs that moved from the middle to the low elevation, may be due to a combination of increased stress levels at the target elevation and the prior presence of low-elevation species at the middle elevation.
Since mortality appeared to be a driver of fast changes, it was somewhat surprising that the relationship between Ellenberg moisture values and species-specific cover change was not significant in the turfs that were moved to elevations with higher flooding stress levels. It may be that, rather than a gradual change, flooding stress may only allow species with certain moisture tolerance thresholds to persist (Garssen et al., 2015), which would result in nonlinear rela-  (Catford & Jansson, 2014). Hence, the slow response observed in our study could still be at the fast end of the continuum compared to other ecosystems with similar climatic conditions (Hasselquist et al., 2015). Such an interaction between climate and ecosystem properties on the slow rates of change is important to consider when evaluating environmental impact assessments or restoration efforts in other systems or climatic conditions (Garssen et al., 2014).
We not only observed differences in rate of change, we also ob- Only very few studies have investigated the effect of delayed or gradual vegetation response for the functioning and ecosystem services in riparian zones (Lake et al., 2007). For riparian soil functions, decomposition was shown to be intermediate for soils that had recently experienced a change in flooding regime (Sarneel & Veen, 2017), which is in line with the observed sensitivity of microbial communities to soil moisture legacies (Hawkes & Keitt, 2015). The effects of other vegetation-mediated changes in riparian functioning such as bank stabilization, water purification or litter subsidies to the aquatic food web remain to be investigated (Lake et al., 2007;Polvi & Sarneel, 2018).
When predicting the effect of environmental and climatic changes on the rate and trajectory by which convergence to the new conditions takes place, our study suggests that rates of initial mortality, and later invasion rates, are import drivers. Those, in their turn, depend on the initial vegetation composition and the difference between the old and new environmental conditions. (Lake et al., 2007;Nilsson et al., 2015a). An additional implication for restoration studies is that there may be multiple pathways of ecosystem recovery, which may differ from the trajectory of degradation. Such hysteresis effects (Fukami & Nakajima, 2011;Lake et al., 2007;Stuble et al., 2017) may be caused by differences in the rates of decline and extinction in relation to probability of colonization and establishment, as reported in this study. Finally, studies projecting vegetation responses to climate change face the triple challenges of predicting the time needed for vegetation adjustment, the likely pathways of change and the fact that controls may represent moving targets (Arft et al., 1999;Olsen, Topper, Skarpaas, Vandvik, & Klanderud, 2016).
These uncertainties can be minimized by knowledge of mortality risks in response to the abiotic conditions under change, competitive interactions among species in the community in focus and the probability of new species colonizing communities (Catford & Jansson, 2014;Fukami & Nakajima, 2011;Olsen et al., 2016).

ACK N OWLED G EM ENTS
Funding for the practical work was provided by The Gunnar and Ruth Björkman fund. JMS conducted the work within the (a) Vindel