Plant community feedbacks and long-term ecosystem responses to multi-factored global change

If you add CO2 or nitrogen to a single plant it will likely grow more, but the amount by which each resource stimulates growth differs widely across species. When you add either resource to a whole ecosystem, total plant growth will likely also increase, but there will be winners and losers, causing a change in the relative abundance of plant species, and therefore altering the way the whole ecosystem responds to the added resource, a “community feedback”. These feedbacks are very difficult to predict, especially when multiple resources are added, but a lot of recent experimental evidences suggests that community feedbacks will determine how future ecosystems operate.


Introduction
Unpredictability in the functioning of terrestrial ecosystems underlies large uncertainties in predictions of future atmospheric CO 2 content (Friedlingstein et al. 2006;Denman et al. 2007;Matthews 2007;Dolman et al. 2010;Piao et al. 2013). The uncertainty among projections of the terrestrial carbon sink over the next century is equivalent to 50 years of greenhouse gas emissions at current rates (Sitch et al. 2008). Some Earth System models incorporate biogeochemical constraints such as those imparted by the nitrogen (N) cycle on plant productivity and respiration and thus ecosystem carbon exchange (Thornton et al. 2007;Sokolov et al. 2008;Bonan and Levis 2010;Bala et al. 2013), but these models may still overestimate the ability of ecosystems to offset human emissions (Piao et al. 2013). Long-term carbon balance may be sensitive to other, slow-acting ecological feedbacks (Friedlingstein and Prentice 2010) that affect the magnitude of the CO 2 fertilization effect in terrestrial ecosystems.
Though Earth System models have incorporated more sophisticated dynamic global vegetation models to represent vegetation shifts, relatively short-term, physiological mechanisms still underpin plant responses (Smith and Dukes 2013). A major source of uncertainty in these models is their ability to predict changes in plant communities in ways that exert a strong influence on ecosystem response to global change (Scheiter et al. 2013), recently referred to as the 'community effect' (Polley et al. 2014). Plant community shifts in response to global changes could initiate key feedbacks that ultimately overwhelm short-term physiological responses (Chapin et al. 1997;Shaver et al. 2000;Suding et al. 2008;Smith et al. 2009). Moreover, the community shifts observed in global change experiments may not correspond to those expected based on short-term physiochemical responses of individual species or functional groups (e.g. Kardol et al. 2010;Langley and Megonigal 2010;Niu et al. 2010;Dawes et al. 2013;Isbell et al. 2013;Ward et al. 2013).
Previous theoretical work has established how changes in plant community structure (defined herein as changes in species identity, richness or evenness), either from shifts of dominance within an ecosystem, immigration or extinction, may drive ecological feedbacks on longterm time scales (Smith et al. 2009) that are relevant for global change predictions. Here, we apply this idea to multi-factored change by (i) surveying recent literature that highlights the importance of community dynamics in long-term, multi-factor field manipulations, (ii) presenting new data from a brackish marsh as a case study, and (iii) using a simple community-net primary production (NPP) model to illustrate how community shifts can mediate ecosystem responses to multiple global changes. We emphasize the marsh as a case study because the plant community is simple (richness ¼ 3) and dynamic so that it responds rapidly to new conditions. The responses observed in the marsh may therefore illustrate mechanisms that apply in other ecosystems that have more complex communities such as grasslands, or ecosystems that require much more time to equilibrate, such as forests. We emphasize the challenge of predicting the responses of future ecosystem productivity to elevated CO 2 in combination with other drivers; however, the concepts may be generally applicable to other combinations of drivers, as well as other ecosystem responses.

Community Effects in Response to Single Factors
Single-factor scenarios provide a simple starting point to evaluate the importance of the community effect. Rising CO 2 should shift communities towards dominance by plant species that respond most positively to elevated CO 2 (Purves and Pacala 2008;Polley et al. 2012). Typically, such a unidirectional community response will have an amplifying effect on the ecosystem response and an ameliorating effect on the initial resource perturbation. For instance, models that allow for vegetation dynamics yield stronger negative feedbacks on atmospheric CO 2 rise than those that have static communities (Purves and Pacala 2008). As species that respond strongly to CO 2 increasingly dominate, enhanced ecosystem CO 2 uptake will tend to counteract the resource perturbation, which in this case is rising atmospheric CO 2 concentrations.
Examples from the first generation of global change experiments showed in single-factor studies that plant community structure responded predictably to one perturbation and had predictable consequences for ecosystem processes. For instance, N addition strongly shifted plant dominance towards nitrophilic species, those with a greater capacity to take advantage of extra N, enhancing total ecosystem productivity in a Minnesota grassland (Tilman 1987). In a brackish marsh, 4 years of exposure to elevated CO 2 yielded increasing plant dominance by C 3 plants over C 4 grasses in mixed plots, enhancing the overall ecosystem response (Arp et al. 1993). These types of findings supported the idea that plant community shifts tend to strengthen ecosystem response to resource addition.
However, as global change studies increased in duration, encompassing more background environmental variability and others included multiple factors, the responses became more idiosyncratic. After 18 years of exposure, the same brackish marsh plots showed a weak response of plant community to elevated CO 2 compared with the community responses to background variability in other abiotic factors (Erickson et al. 2007). Similarly, elevated CO 2 initially drove a shift in species dominance in a New Zealand grassland, but after 11 years, an additional factor of grazing negated that shift and the ecosystem CO 2 response (Newton et al. 2014).
The short-and long-term effects of global change drivers may differ for several reasons. For instance, the effects on vegetative production may differ from effects on reproduction or recruitment (including floral production, seed production, seed quality, seedling establishment), leading to longer-term demographic shifts that are not predictable from short-term growth responses (Williams et al. 2007). Moreover, the future will bring multiple changes simultaneously, and some factors, like N eutrophication, may have stronger effects on plant community structure than others, like rising CO 2 (Reich 2009). What happens to the community feedback when multiple factors important to plant physiology and growth change at once?

Community Shifts in Response to Multiple Factors
Relatively few long-term, in situ global change experiments have manipulated multiple factors (Norby and Luo 2004), but several have recently found that the combined effect on ecosystems differs substantially from the sum of single-factor effects ( Table 1). For example, a global change experiment in a Californian grassland found a CO 2 stimulation of plant growth when CO 2 was applied in isolation but not in combination with warming or N addition (Shaw et al. 2002;Dukes et al. 2005), and similarly, CO 2 effects on N 2 O emissions were dampened with other factors (Brown et al. 2012). In a hardwood forest, high CO 2 enhanced growth but not in combination with ozone (Karnosky et al. 2003) or climatic variability (Kubiske et al. 2006). In a tidal marsh, the CO 2 stimulation of productivity diminished with added N (Langley and Megonigal 2010). More generally, the magnitude of global change responses decreases with an increasing number of drivers (Leuzinger et al. 2011).
Plant community feedbacks may help explain this pattern in ecosystem response. Elevated CO 2 alone tends to favour dominance by CO 2 -responsive species yielding a more CO 2 -responsive ecosystem (Arp et al. 1993;Newton et al. 2014;Polley et al. 2014), but addition of N can cause unexpected effects mediated by plant community shifts ( Fig. 1; Langley and Megonigal 2010). While elevated CO 2 and N addition tend to have direct and positive effects on the productivity of plants in isolation, each resource has different effects on plant communities by favouring different subsets of species. Moreover, N addition generally imparts a much more profound influence on the community than CO 2 (Reich 2009;White et al. 2012). In the tidal marsh, N addition promoted dominance by C 4 grasses that respond poorly to elevated CO 2 , thereby reducing ecosystem responsiveness ( Fig. 1, second factor). In years when N had a strong effect on dominance by grasses, it also reduced the response of NPP to elevated CO 2 (Fig. 2).
A new generation of decadal-scale global change studies is converging on the finding that the strongest effects of global change may be mediated by plant community shifts, eclipsing other more direct effects on plant physiology (Kardol et al. 2010;Langley and Megonigal 2010;Niu et al. 2010;Yang et al. 2011;Wu et al. 2012;Isbell et al. 2013;Newton et al. 2014;Polley et al. 2014). In a Tasmanian grassland, plant functional type had a large effect on ecosystem carbon cycling . A recent analysis of N cycle response to warming and clipping found that the strongest effects of N on productivity were driven by species shifts (Niu et al. 2010). In an old-field ecosystem, Kardol et al. (2010) found that differences in soil processes varied more strongly among plant species than between any global change treatments such as watering, added N and elevated CO 2 , implying that in the long term, even small changes in species abundance would hold more sway than the sum of physiological changes imparted on individual species. Warming enhanced NPP in an arid grassland until community shifts negated the stimulation (Wu et al. 2012). Even though N addition strongly increased plant productivity in a Minnesota grassland initially, the effect diminished in the long term owing to changes in plant diversity (Isbell et al. 2013).
While community shifts may take decades to occur, direct physiological responses can occur within seconds but may diminish through time because of acclimation (Isbell et al. 2013;Smith and Dukes 2013). For instance, the short-term physiological response and ecosystem productivity response to CO 2 often diminish in the longer term (Seiler et al. 2009;Lee et al. 2011). Likewise, warming of soil causes a short-term increase in respiration that wanes over time (Bradford et al. 2008). Therefore, while experiments are likely to overestimate the direct influence of global change, they also may underestimate community effects, particularly in plant communities that change slowly (Smith et al. 2009). Forest community dynamics are particularly difficult to capture owing to long individual lifespans. Tree species shifts in response to CO 2 have not been adequately examined (Norby and Zak 2011), though chronic N pollution is beginning to alter tree species in some forests (e.g. Zaccherio and Finzi 2007).
AGB, aboveground biomass; ANPP, aboveground net primary production. Figure 1. The pathways of influence by which elevated CO 2 and N may drive an ecosystem function. Symbols represent the effect of added resources on each relationship. The black arrows represent short-term, physiological effects of N addition on the CO 2 fertilization effect, which are generally found to be positive for individual plants. The red arrows represent the effect of N-driven community shifts, which may counteract or enhance the physiological effects depending on how they covary with the other pathways of influence. Elevated CO 2 may also affect community change, but the effects are typically not as pronounced as with added N. treatments counteracting each other (i.e. warming may dry soil while water addition will restore moisture) (Luo et al. 2008;Leuzinger et al. 2011), mounting evidence indicates that in the long term, changes in plant community structure could ultimately govern ecosystem responses (Figs 1 and 2; Smith et al. 2009;Langley and Megonigal 2010;Niu et al. 2010). Yet, accurately predicting these phenomena requires understanding community responses mechanistically.

Tradeoffs in Plant Responsiveness to Different Perturbations
A cornerstone of modern ecological theory is that plants exhibit tradeoffs in resource strategy (Tilman 1988;Craine 2009), such that optimization for acquisition and use of one resource precludes optimal acquisition and use of another. A simple example of a tradeoff is that of plant allocation among plant organs. Energetic limitations dictate that the same resources cannot be allocated simultaneously to multiple organs, while phylogenetic limitations appear to dictate that a single plant has limited plasticity in shifting allocation-that is, plant species are hardwired to a narrow range of allocation strategies These differences in allocation strategy could help explain how plants respond to global changes, particularly those that alter resource availability. For instance, plants that exhibit low root-to-shoot biomass ratios are optimized for relatively high nutrient conditions, and should be better positioned to respond positively to soil resource addition than plants that have high rootto-shoot ratios. Other tradeoffs in physiology or allocation may play a role in resource strategy. For instance, a tradeoff may exist in plant metabolic pathways such that a positive response to elevated CO 2 may interrupt the efficient incorporation of nitrate-N into a usable organic form (Bloom et al. 2010). If these tradeoffs result in a predictable relationship between the magnitude of plant growth response to elevated CO 2 and that to added N, then this knowledge could help constrain our predictions of ecosystem responses to global change.
Evidence for a tradeoff between CO 2 and N responses exists across functional groups. For instance, legumes respond most strongly to CO 2 compared with other functional groups (Jablonski et al. 2002) and are among the weakest responders to N addition (Xia and Wan 2008). At the other end of the spectrum, grasses tend to exhibit a very weak CO 2 response (Jablonski et al. 2002;Ainsworth and Long 2005), but respond strongly to N enhancement in many ecosystems (Pennings et al. 2002;Langley and Megonigal 2010). Slow-growing plants generally respond more positively than fast-growing ones to elevated CO 2 (Ali et al. 2013), while the opposite is true for responses to N addition. Still, enormous variability exists within functional groups, making it difficult to generate useful functional group parameterizations for models. Again, these tradeoffs in response may occur among other resources or conditions as well. Recent evidence, for example, suggests differential responses to CO 2 and water, and the same responses to CO 2 and warming (Dijkstra et al. 2010;Dieleman et al. 2012).

A Simple Plant Community Model
To examine the potential importance of plant community shifts for ecosystem response we simulated a simplified ecosystem with 30 species of initially equal biomass. We assigned realistic, randomized distributions of biomass responses to each plant species modelled after observed distributions of responses. Carbon dioxide response ratios (R ¼ treated/control) across species are generally normally distributed with a mean of 1.25 and a standard deviation of 0.35 (Poorter and Navas 2003). Nitrogen stimulations are typically log-normally distributed (Xia and Wan 2008) with a mean of 1.29 and a standard deviation of 0.37 (LeBauer and Treseder 2008). (Note that model outputs are qualitatively insensitive to the effect sizes used.) We assigned a random value (P) from 0 to 1 to each of 30 species. We used P as a cumulative probability to calculate the corresponding CO 2 response according to a normal distribution (m ¼ 1.25, s ¼ 0.35).
Then, we generated combined (CO 2 + N) responses for three types of relationships between CO 2 and N responses: negative, positive and independent, representing three ways that responses to each driver may covary. For 'negative', the randomly generated probability was inverted (12P) to generate the N response of each species using a lognormal distribution (m ¼ 1.29, s ¼ 0.37). Therefore, a species with a large CO 2 response would have a relatively small N response and vice versa. The negative relationship represents a tradeoff wherein plants that respond strongly to CO 2 do not respond strongly to N. For 'positive', each species probability (P) was converted to an effect size for N using the lognormal distribution of N responses (m ¼ 1.29, s ¼ 0.37) such that a species that responds most strongly to CO 2 also responds most strongly to N. A positive relationship could occur if the capacity to respond to both N and CO 2 was related to some other factor. Some have suggested, for instance, that plants with high relative growth rates should be more suited to take advantage of increased resource availability, though this has not been observed for elevated CO 2 (Poorter and Navas 2003). The 'independent' scenario assumes that the N response of a species is unrelated to the CO 2 response, so random probabilities were generated independently for the N responses and then converted using the N response distribution (m ¼ 1.29, s ¼ 0.37). The response ratios were maintained for each species throughout the simulation.
We calculated a species biomass response ratio (R) for each species in each scenario. For single-factor scenarios (CO 2 only and N only), we used R for each species based only on either CO 2 or N responses. For the CO 2 + N scenarios (independent, positive and negative) we calculated additive and multiplicative effects of CO 2 and N. Additive was estimated as R X = R CO 2 + R Nitrogen . Multiplicative stimulation was calculated as R X = R CO 2 × R Nitrogen . These two methods represent two ways CO 2 and N may interact in ecosystems . Because the additive and multiplicative simulations yielded qualitatively similar responses, we limit further discussion to the multiplicative. We weighted each plant response according to its relative abundance (A X ) to calculate total ecosystem biomass response (ER): Therefore, the first-generation ecosystem response is equivalent to the average of individual plant responses because the distribution of species was initially even. For subsequent generations, we allowed species relative abundance in a given generation (t) to shift according to its relative biomass response in the previous generation (t 2 1): Here we assumed that community shifts follow biomass responses. This assumption may not hold as, for instance, reproductive response, which may influence long-term demography, can differ from the growth response (Williams et al. 2007;Way et al. 2010). Still, there is evidence that long-term plant community shifts may respond consistently over time to some drivers like N eutrophication more strongly than to others like elevated CO 2 (Reich 2009;Langley and Megonigal 2010). To account for the possibility that community shifts follow N responses but not the combined CO 2 × N responses, we conducted a simulation using just the single-factor R X representing the N response. Because no species are added and none go extinct, the model represents physiological responses and shifts within the community, or species reordering (Smith et al. 2009), but not exchange with external species pools. However, species relative abundance is allowed to drop infinitesimally close to zero, so that species may effectively go extinct in some cases.
To constrain the upper limit of ecosystem response we imposed a light limitation according to an established relationship between leaf area index (LAI) and canopy assimilation given a high light environment given the following equation (Bonan 2008): where A canopy is the ecosystem level CO 2 assimilation per ground area used as a proxy for plant growth; A max is the light saturated leaf-level assimilation (20 mmol m 22 s 21 ); 1 is the light use efficiency (0.06); K is the light extinction coefficient (0.5); F 0 is the solar radiation at the top of the canopy (1000 mmol m 22 s 21 ); and L is the LAI, which was set initially at 2.5 and allowed to vary through time. Then, we allowed shifts to occur for 14 additional generations to look at the trajectory of ecosystem response with shifting community structure (Fig. 3). We repeated the simulation with re-randomized response distributions 20 times to estimate mean response and error.
Though this exercise makes many simplifying assumptions, the results illustrate some important points. If the ultimate community shifts follow initial, short-term responses, then the community feedback effect will amplify individual physiological responses to an enhanced resource in isolation to other global change factors or other disturbances ('CO 2 only' and 'N only' scenarios; Fig. 3, top panel). The ultimate ecosystem NPP response roughly doubles the initial stimulation, intuitively approaching the stimulation of the strongest responding species. This finding agrees with initial results from many ecosystem studies described above.
With multiple interacting factors, the magnitude of the initial ecosystem response (at Generation 1) and especially the community effect (apparent in subsequent generations) depends on how plant responses to one factor relate to that of another factor. The existence of a negative relationship between CO 2 response and N response across plant species, which would arise if tradeoffs exist in the ability of plants to respond to enhancement of each resource, will reduce total ecosystem productivity in the long term below what would be predicted if CO 2 and N responses are independent or positively related. Interestingly, the model indicates that with negative covariance, community feedbacks in response to CO 2 and N in combination could lead to an ecosystem productivity stimulation that is smaller than with CO 2 only (Fig. 3). The short-term responses do not AoB PLANTS www.aobplants.oxfordjournals.org depend as strongly on the type of covariance (from 42 to 56 % stimulation, Generation 1; Fig. 3) as the long-term responses (from 46 to 86 % stimulation, Generation 15; Fig. 3).
We have made the point above, however, that longterm community shifts do not always follow initial physiological responses. If the ultimate community shifts occur randomly and independently of the physiological stimulation, then the ultimate stimulations would resemble the initial stimulation (Generation 1), and community shifts would be unimportant for predicting long-term ecosystem response. It is more likely that community shifts are related to physiological responses to global change drivers, though the shifts may not occur in a manner as simple as represented in this model. For instance, N addition tends to have a stronger influence on community structure than does elevated CO 2 (Reich 2009;Langley and Megonigal 2010). Therefore, we may expect that plant communities exposed to both CO 2 and N would follow the initial N response more than the initial CO 2 response. When community shifts followed N response, the results were dampened but qualitatively similar to when the community followed the combined CO 2 × N response (results not shown).
The model indicates that failing to properly account for plant community shifts could engender large errors in estimates of long-term ecosystem productivity. In the case of multi-factored global change, the community feedback effect depends strongly on the type of covariance plant responses exhibit among different factors. For instance, failing to account for negative covariance between CO 2 and N response could lead to a drastic overestimate in ecosystem response. Recognizing these patterns could help improve predictability of long-term plant responses. Indeed several examples from the recent literature point to such shifts in community structure driven by one factor diminishing the ecosystem response to another (Table 1) indicating that capturing these types of interactions in models will help predict future response of ecosystems to global change.

Predicting the Global CO 2 Fertilization Effect
A decade ago, global climate models systematically overestimated the ameliorating effects of terrestrial ecosystems on the future carbon cycle by excluding the nutrient limitation of the CO 2 fertilization effect. Several more recent models have now incorporated this stoichiometric limitation (Sokolov et al. 2008;Thornton et al. 2009;Bala et al. 2013). Interestingly though, as models refine representation of the N cycle, some show that N may become less limiting in the future, owing to accelerated mineralization in response to warming and projections of increasing N deposition. Accounting for enhanced N availability, therefore, partially restores the large CO 2 fertilization effect (Sokolov et al. 2008;Thornton et al. 2009;Zaehle and Dalmonech 2011). Evidence from longterm CO 2 studies suggests that N scarcity limits the CO 2 stimulation of productivity in temperate ecosystems (Norby et al. 2010;Reich and Hobbie 2012). However, N addition exerts a strong influence on plant diversity and species composition (Tilman 1987;Isbell et al. 2013) that must be considered when extrapolating to a future with widespread and heterogeneous increase in N availability.
It has been suggested that Earth System models, many of which represent biodiversity coarsely, overestimate the negative effects of global change on ecosystem processes because they exclude the mitigating effects of plant species shifts (Purves and Pacala 2008) apparent in diversity experiments . For instance, higher plant biodiversity diminished the negative effects . Ecosystem productivity response (top panel) and relative difference in productivity (bottom panel) for simulated plant communities for single factors (CO 2 and N) and combined responses (CO 2 + N) according to the relationships between CO 2 response and N response across species (independent, positive or negative). Ecosystem productivity is expressed as a factor of the control with ambient CO 2 and no added N, which is set to 1. The first-generation response (highlighted in grey box) represents the physiological effect while subsequent generations represent the influence of dynamic communities. For instance, while CO 2 alone here yields a response factor of 1.24 in the first year, ensuing community shifts favouring CO 2 -responsive species could increase the response to 1.49 by year 2015.
of drought on plant productivity Reich 2009). However, in predicting an enhanced terrestrial carbon sink in the future (Pinsonneault et al. 2011;Bala et al. 2013), the expectation is reversed-rather than relying on stable ecosystem functioning, society stands to benefit from a dynamic ecosystem function, specifically, increasing net carbon uptake through time. If this dramatic change in ecosystem behaviour does not persist, then the atmospheric CO 2 concentration will rise more than expected, and climate change will be more severe than presently predicted.

Conclusions
We have focused on how the interactive effects of two added resources may affect plant communities to drive ecosystem function. Yet, the recognition of community shifts in global change response applies to other factors besides enhanced resource availabilities (De Marco et al. 2013). For instance, if ozone tolerance were negatively related to CO 2 responsiveness across plant taxa, then ozone exposure will favour species that exhibit a slighter CO 2 response, thereby dampening the ecosystem-level CO 2 response. The combination of CO 2 and N could likewise favour a unique subset of species that can take advantage of the combination of high CO 2 and high N. For example, some invasive species have been shown to respond more strongly to both elevated CO 2 and N enrichment than non-invasives (Dukes et al. 2011;Mozdzer and Megonigal 2012). This general propensity for invasives to take advantage of high resource availability (Davidson et al. 2011) may indicate that they are not as constrained by the same tradeoffs as other plants. The degree to which community will shift and ecosystem productivity will respond to future change will partly depend on the degree to which individual plant responses to global change drivers covary across taxa.

Conflicts of Interest Statement
None declared.