A plant's perspective of extremes: terrestrial plant responses to changing climatic variability

Abstract

We review observational, experimental, and model results on how plants respond to extreme climatic conditions induced by changing climatic variability. Distinguishing between impacts of changing mean climatic conditions and changing climatic variability on terrestrial ecosystems is generally underrated in current studies. The goals of our review are thus (1) to identify plant processes that are vulnerable to changes in the variability of climatic variables rather than to changes in their mean, and (2) to depict/evaluate available study designs to quantify responses of plants to changing climatic variability. We find that phenology is largely affected by changing mean climate but also that impacts of climatic variability are much less studied, although potentially damaging. We note that plant water relations seem to be very vulnerable to extremes driven by changes in temperature and precipitation and that heatwaves and flooding have stronger impacts on physiological processes than changing mean climate. Moreover, interacting phenological and physiological processes are likely to further complicate plant responses to changing climatic variability. Phenological and physiological processes and their interactions culminate in even more sophisticated responses to changing mean climate and climatic variability at the species and community level. Generally, observational studies are well suited to study plant responses to changing mean climate, but less suitable to gain a mechanistic understanding of plant responses to climatic variability. Experiments seem best suited to simulate extreme events. In models, temporal resolution and model structure are crucial to capture plant responses to changing climatic variability. We highlight that a combination of experimental, observational, and/or modeling studies have the potential to overcome important caveats of the respective individual approaches.

Introduction

Although the spatial and temporal extent of future climatic changes is still partly uncertain (IPCC, 2007a), it is likely that the adaptive capacity of terrestrial plants and ecosystems will be exceeded in many regions (IPCC, 2007b). Already today, responses to climate change can not only be observed for individual species and ecosystems (e.g. Allen & Breshears, 1998; Gitlin et al., 2006) but also across species and organizational scales (e.g. Walther et al., 2002; Allen et al., 2010; Lindner et al., 2010). Climate change may manifest itself in two fundamentally different ways: in a change in the mean of for example temperature or precipitation, and in a change in their variability (i.e. variance and/or distribution, Fig. 1; Rummukainen, 2012; Seneviratne et al., 2012). It is important to note that these terms relate to steady-state systems. The climate system and ecosystems, however, are in permanent transition and therefore the term ‘mean’ and ‘variability’ only make sense relative to well-defined spatial and temporal scales. Moreover, mean and variability may not be fully independent (e.g. an increasing mean value often implies increasing standard errors). Here, we still treat changes in mean and variability as two separate aspects, defining changes in the mean as changes over longer time periods (e.g. inter-annual changes) and changes in variability as changes over medium/short-term periods (e.g. inter-daily changes) of climatic variables. We define extreme events from this climatological perspective as increasing climatic variability (i.e. increasing variance and/or changing distribution) in contrast to changes in mean climate. Our aim is to emphasize the generally unrecognized distinction between impacts of changing mean climate and changing climatic variability on terrestrial ecosystems.

Figure 1. The two theoretical cases of changing climatic drivers: (1) changes in the mean but not the variance (upper panel), (2) changes in the variance, but not the mean of a variable (lower panel). A third case is conceivable where both the variance and the mean remain comparable, but rare, very extreme events occur, changing essentially the nature of the distribution. Importantly, any discussion of means vs. extremes requires a temporal reference, as a short-term increase in the mean may turn out to be a long-term increase in the variance.

We center, but do not limit our synthesis on a plant's perspective of temperature and precipitation extremes, since these are the most important climatic determinants of plant growth and survival globally (e.g. Boisvenue & Running, 2006). Observations since 1950 show that the length of warm spells and heat waves increased (e.g. Barriopedro et al., 2011; Rahmstorf & Coumou, 2011; Seneviratne et al., 2012). More intense and longer droughts are observed, but at the same time the number of heavy precipitation events increased (Seneviratne et al., 2012 and references therein). Future projections on changes in climatic variability show strong spatial and temporal heterogeneity (Giorgi et al., 2004; Orlowsky & Seneviratne, 2012) and are highly uncertain (Seneviratne et al., 2012). Using multi-model experiments, Barriopedro et al. (2011) for instance found that the probability of summer heatwaves may increase by a factor of 5–10 in the future although Schär et al. (2004) predict that temperature variability will increase by a factor of 2 in Europe. Projected changes in extreme precipitation events (droughts or flooding) are even more uncertain. Orlowsky & Seneviratne, 2011 derived from their simulations with an ensemble of general circulation models (GCMs) robust projections on increasing droughts over the Mediterranean and increasing heavy precipitation over the Northern high latitudes.

Although changes in the mean values are important, there is evidence that plant distribution (Chapin et al., 1993; Bokhorst et al., 2007), survival (Van Peer et al., 2004) or net primary productivity and species diversity (Knapp et al., 2002) respond to extreme rather than to average conditions (Jentsch & Beierkuhnlein, 2008). In addition to that, different physiological processes such as photosynthesis, water relations or nutrient uptake at the species, community or ecosystem level affect the response of plants to climatic variability (Fig. 2). To account, for example, for changing precipitation distributions, Knapp et al. (2002) decreased precipitation frequency, but not its total amount in a mesic grassland leading to more intense precipitation events. They found reduced carbon turnover, but increased species diversity. Drier conditions also tend to decrease evapotranspiration, which leads to lower evaporative cooling (Teuling et al., 2010). In combination, warming and drought can therefore lead to additional warming of an ecosystem (Seneviratne et al., 2006; Fischer et al., 2007; Kuster et al., 2012).

Figure 2. Conceptual overview of the different processes and scales affected by extremes and the study designs to assess them.

In addition to the impacts of changing climatic variability, the physiological response of terrestrial plants depends also on interactions between species (Thorpe et al., 2011) and their ability to adapt and acclimate. The water available for plants depends on the water holding capacity of the soil (Kramer & Boyer, 1995; Porporato et al., 2004; Leuzinger & Körner, 2010; Raz-Yaseef et al., 2010), competition with other plants (Casper & Jackson, 1997) and precipitation patterns (Knapp et al., 2008). The latter has different effects on soils with high or low water holding capacity (i.e. a stronger or weaker buffer against drought; Knapp et al., 2008) or on flood occurrence, which is an important driver of plant distribution (Crawford, 1992; Colmer & Flowers, 2008; Parolin & Wittmann, 2010). Furthermore, interactions between changing climatic variables as well as thereby induced community shifts may affect the response of plants to new conditions (Langley & Megonigal, 2010; De Boeck et al., 2011). For example, a drier and warmer climate will exert stronger constraints on plant growth than a warmer, but also wetter climate; or rising CO₂ may alleviate the impact of drought (Morgan et al., 2004; Holtum & Winter, 2010). Moreover, more prolonged dry periods will alternate with more intensive rainfall events, both within and between years, which will change soil moisture dynamics (Weltzin et al., 2003; Porporato et al., 2004; Fay et al., 2008; Knapp et al., 2008; Bartholomeus et al., 2011a). Eventually, it is also crucial how quickly plant communities adapt genetically to the imposed changes. The IPCC (2007b) concluded that the rate of natural adaptation will be slower than the rate of climate change. Natural adaptation differs between species: although species with short generation times may adapt within years, Rehfeldt et al. (2001) for example estimate that it will take 2–12 generations (an equivalent of 200–1200 years) for a coniferous trees species to show genetic adaptation in response to climatic change. All these factors determine whether plants at a specific site will experience changing climatic variability as extreme or not.

Thus, the vulnerability of terrestrial plants to climate change will, besides changes in the mean, largely depend on the changes in the climatic variability and the occurrence of extreme events. The understanding of this difference in experiments and model simulations requires very good knowledge of the baseline or control climate (especially the background variability to which plants are adapted to). This complies with the fact that extreme conditions per se have shaped ecosystems for a long time (Körner, 1998, 2003) and may also foster adaptation and thus decrease sensitivity (Hegerl et al., 2011). A plant's response to specific environmental conditions produces its specialized set of traits, which allows it to prevail over competitors and occupy a specific habitat (Körner, 1998, 2003). We use the term ‘stress’ throughout this review according to Lortie et al. (2004) to refer to situations in which plants experience critical environmental conditions beyond what they experience normally (Chapin, 1991) such that damage to vital function occurs (see Gaspar et al., 2002).

In this article we strive to answer the following questions:

• Which plant processes are vulnerable to changes in the variability of climatic drivers rather than to changes in their mean?
• How can we quantify responses of plants to changing climatic variability?

We present evidence from experiments, observations, and modeling studies that help to understand the current and future responses of individuals and communities to changing variability, with a particular focus on temporal and spatial patterns. These examples also help to identify important research gaps. We do not aim to cover the literature on these topics systematically.

Which plant processes are vulnerable to changes in the variability of climatic drivers rather than to changes in their mean?

The vulnerability of plants refers to their susceptibility to adverse effects of environmental change (IPCC, 2007b). Estimates of vulnerability depend on the definitions (e.g. the definition of death (Zeppel et al., 2011)) and the spatiotemporal scale considered. The ultimate limit to withstanding environmental stress from an individual plant's perspective is mortality due to physiological failure (‘You can only die once’), but at the community level, already reductions in growth and subsequently competitiveness may constitute a limit to species fitness. For commercial crops, it may even be a critical reduction in productivity so that cultivation is discontinued.

In the following sections, we discuss the vulnerability of phenological and (individual and interacting) physiological processes to changes in the climatic variability rather than the mean of climatic drivers and we highlight how these processes play out at the species and the community level (see schematic overview in Fig. 2). Our list of examples is not exhaustive, but is meant to illustrate this important difference between changes in climatic variability rather than the mean.

Phenological processes

One of the well-studied responses of plant species or communities to environmental change is phenology, which tracks seasonal events in generative and vegetative plant growth. Given the predominant influence of climate (with the important exception of photoperiodism, see Körner & Basler, 2010), phenology has emerged as a key tool in identifying fingerprints of anthropogenic climate change in nature (Menzel et al., 2006). Observed large-scale phenological changes such as an earlier onset of leaf unfolding/flowering (Menzel & Fabian, 1999; Walther et al., 2002; Parmesan & Yohe, 2003; Root et al., 2003; Menzel et al., 2006) are mainly driven by changes in mean climatic conditions especially temperature (Vitasse et al., 2009; Polgar & Primack, 2011; see also Table 1).

Table 1. Examples of observed plant vulnerabilities to changes in the mean climate and climate variability

Process	Changing mean	Effect/Response	Reference	Changing variability	Effect/response	Reference
Phenology	Increase in mean temperature	Prolongation of growing season, earlier onset of leaf unfolding and first flowering, delay of leaf senescence	Menzel & Fabian (1999); Walther et al. (2002); Parmesan & Yohe (2003); Root et al. (2003); Menzel et al. (2006); Polgar & Primack (2011); Vitasse et al. (2009)	Early and late frosts, warm spells, drought, heavy rain	Frost damage, possibly fatal damage to opportunistic species, second or extended flowering, advanced mid-flowering, decreased flowering length	Leuzinger et al. (2011a); Luterbacher et al. (2007); Jentsch et al. (2009)
Soil organic matter decomposition	Increase in mean temperature	Potentially increase in soil organic matter decomposition	Saxe et al. (2001)	Droughts/heatwaves	Increase in soil water repellency leads to reduced decomposition of soil organic matter	Goebel et al. (2011)
Water relations	Increase in night-time warming (and mean temperature)	Slight increases in stomatal conductance	Albert et al. (2011)	Drought	Stomatal closure and carbon starvation (isohydric plants), hydraulic failure (anisohydric plants)	Bréda et al. (2006); McDowell et al. (2008)

Phenological changes in response to changing climatic variability are much less studied although they clearly interact with phenological changes induced by changing mean climate. For example, in the temperate and boreal zones, which are often temperature limited, a central trade-off revolves around maximizing the vegetation period while avoiding frost damage (Kramer et al., 2010). An untimely response to early warm spells may be fatal, but can bring enormous advantages for early successional or opportunistic species (r-strategists, Leuzinger et al., 2011a). In contrast, long-lived, late successional species often have chilling requirements and photoperiodic safety mechanisms (Heide, 1993) and thus may be in a position to avoid increasing risks of late frost due to changing climatic variability, but would also benefit less from early warm spells. This is supported by the fact that the risk of damage due to late frost events has not increased so far for several coniferous and broad-leaved species in Central Europe (Menzel et al., 2003; Scheifinger et al., 2003). Besides this example, there is further evidence, that extreme events may alter phenological responses depending on their timing and strength (e.g. Jentsch et al., 2009; Menzel et al., 2011). This can lead to unexpected effects such as second flowering in autumn or extended flowering until the beginning of winter for some species (Luterbacher et al., 2007). Moreover, extreme warm spells decreased the differences in spring phenology between urban and rural sites (Jochner et al., 2011). Furthermore, only half of the trees reached leaf maturity in an extreme drought experiment in the Mediterranean (Misson et al., 2011). Overall, the response of phenology to climatic variability seems to be less well understood than to changing mean climate although increasing climatic variability may have a strong damaging potential.

Physiological processes

We here focus on the response of plant water relations such as transpiration to climatic variability (drought/heat waves and excess water). Increasing temperatures and/or heat waves combined with less precipitation or more variable precipitation events lead to prolonged dry periods and high atmospheric demand for plant transpiration, which determine drought stress of plants beyond changes in mean climate (Schimper, 1903; Porporato et al., 2004). Barriopedro et al. (2011) predict such an increase in drought events for the 21st century and the consequences for plant water relations are well documented (e.g. Leuzinger et al., 2005; Bréda et al., 2006; Granier et al., 2007) although not all mechanisms are fully understood. There is an ongoing debate about two competing response strategies to drought: Isohydric plants may respond by closing their stomates thus reducing their water loss, but eventually facing carbon starvation, whereas anisohydric plants keep their stomates open thus running the risk of hydraulic failure (McDowell et al., 2008; Sala et al., 2010; Zeppel et al., 2011). Furthermore, Craine et al. (2012) highlighted the importance of the timing of an extreme event for grassland productivity. The response of plants to drought is of such importance that Hartmann (2011) refers to it as a ‘change of evolutionary forces’ from competition for light to competition for water and carbon. The responses of plants to climatic variability and particularly drought have important consequences for net primary productivity (NPP) and hence carbon cycling even at large spatial scales such as Europe (Ciais et al., 2005; Dury et al., 2011). Thus, plant responses to increasing drought events and heat waves influence plant functioning across spatial and temporal scales.

Also climatic variability resulting in excess water (i.e. flooding or waterlogging), can induce important physiological responses by terrestrial plants. Due to waterlogging, O₂ diffusion and supply to the roots is reduced, and the oxygen demand of plant roots (that is root respiration – oxygen consumption in the roots), cannot be fulfilled (Lloyd & Taylor, 1994; Blom & Voesenek, 1996; Kozlowski, 1997; Amthor, 2000). This results in waterlogging/oxygen stress, i.e. lack of oxygen due to high soil moisture contents (Bartholomeus et al., 2008). Both the oxygen supply and demand may be affected by a more extreme climate, due to more intense precipitation and higher temperatures (respiration increases with temperature), respectively. Therefore, to analyze the effects of low soil oxygen availability on species performance, it is necessary to integrate the soil physical and plant physiological processes, thus accounting for both the oxygen supply to and oxygen demand of plant roots (Bartholomeus et al., 2011b). Besides reduced root respiration rates, the decrease of water absorption due to waterlogging stress causes sensitive plants to wilt in a similar way to drought (Jackson & Drew, 1984). Many species already growing in flood-prone habitats have developed different strategies to survive hypoxia, by producing aerenchyma and/or adventitious roots in response to an increase in the concentration of ethylene and auxin (Blom & Voesenek, 1996). Flooding can also give rise to detrimental effects at leaf level, by inducing stomatal closure and, consequently, limiting gas exchange and plant growth (Kramer, 1951; Chen et al., 2005; Rengifo et al., 2005; Fernandez, 2006). Thus, similarly to drought, extremes of excess water, in combination with higher temperatures, strongly alter plant physiological processes such as carbon uptake and transpiration.

In conclusion, we note that plant water relations seem to be very vulnerable to increasing variability in temperature and precipitation and that changing heatwaves and flooding have stronger impacts on physiological processes than changing mean climate (see also Table 1).

Interacting physiological processes

The interaction of physiological processes such as photosynthesis, nutrient uptake, and water relations may strongly affect the response of plants to changing climatic variability. Furthermore, interactions among several global change drivers or between global change drivers and other environmental variables, may result in other growth-limiting factors (e.g. soil type) becoming less important. Drought periods, for example, may have the potential to not only determine growth or mortality in an ecosystem but also to cause shifts in growth-limiting factors such as nutrient limitations. For example, in an experiment of Kuster et al. (2012) oaks were grown on two different soil types with different nutrient availabilities. Under well-watered conditions, growth on one soil was lower due to nutrient-limiting conditions, whereas under repeated drought periods these differences disappeared. This shows that growth-limiting factors such as nutrient availability can become less important under changing climatic variability, although they may be overlooked if only changes in mean climate are considered. There are many other examples of interacting processes under changing climatic variability such as ozone stress during periods of high temperature (Matyssek et al., 2010; Pretzsch & Dieler, 2011).

The interactions of physiological processes can, however, be even more intriguing. In coastal habitats (i.e. the interface of terrestrial and aquatic habitats), which are not only saline but are also prone to flooding (e.g. mangroves and salt marshes) (Colmer & Flowers, 2008) Tamarix africana Poiret, for example, showed a reduction of CO₂ assimilation rates only in young Tamarix africana Poiret leaves after 45 days under continuous flooding with saline water (200 mm), whereas old leaves and the aboveground relative growth rate were not affected by the treatment (Abou Jaoudé et al., 2012). Thus, although parts of the plants actually responded to flooding, this was not the case for the entire plant. This example is rather related to changes in mean climatic conditions (i.e. temperature-induced rising sea levels), but it highlights that changing climatic variability is likely to interact with an already complex interplay of physiological processes.

Species-level processes

At the species level, responses of different genotypes to climate provide information on how a species may react to changing climatic variability. Since genotypic variation results in different sensitivity thresholds of distinct ecotypes to changing climatic variability it can partly substitute lacking data of changing climatic variability for a specific genotype. In an ecotype study (Klein et al., under review) that included all three climate types (meso-Mediterranean (MM), thermo-Mediterranean (TM), and semiarid (SA) within the natural distribution of the forest tree Pinus halepensis Mill. (and hence three very different combinations of mean climate and climate variability), two major physiological adjustments identified were as follows: (1) shortening of the growing season length (from 165 to 100 days) to match a shorter rainy season and (2) increasing water use efficiency (from 80, to 95, to 110 μmol CO₂ mol-1 H₂O under MM, TM, and SA climates, respectively). However, the sensitivity threshold differed in between ecotypes: Northern ecotypes mainly responded to the change MM to TM, whereas Southern ecotypes responded to the change TM to SA. At the species level, the study showed that higher xylem sensitivity to embolism in specific ecotypes matched previous reports (Atzmon et al., 2004; Schiller & Atzmon, 2009) of significantly higher mortality rates in these ecotypes under yet harsher conditions. These observations suggest that while hydraulic constraints in response to climatic variability limited the distribution of a tree species, plasticity in water use efficiency and growth phenology enabled its success under a wide range of climatic conditions.

Community-level processes

At the community level, phenological, physiological, and species-level processes as well as their interaction culminate in complex responses to changing mean climate and climatic variability (Fig. 2). Species range shifts not only have been associated with changes in mean climate (Lenoir et al., 2008; Harsch et al., 2009) but also with changing climatic variability (Kelly & Goulden, 2008; Doak & Morris, 2010). They lead to a disruption of ecological communities and species interactions due to different dispersal speed and success. These processes differ between the trailing and the leading edge of a population (Doak & Morris, 2010; Kramer et al., 2010). From a community's perspective such range shifts may entail positive (e.g. release from competition) and negative (e.g. loss of important pollinator) consequences. Despite these important consequences of range shifts, it is yet not clear whether changing mean climate or changing climatic variability will be a more important driver of range shifts.

At the community level, for annual plants, the variability of rainfall is important for the success of germination. Increasing climate variability can have both negative and positive effects on species persistence and thus plant population dynamics (Levine et al., 2008). Climatic fluctuations, for example, may enable species to avoid interspecific competition if species differ in the years in which they perform (e.g. reproduce or grow) best (Levine & Rees, 2004). Dormancy and germination biology determine whether temporal variability favors or inhibits species persistence (Levine & Rees, 2004) and can thus be limiting for a species (Godefroid et al., 2011). Temporal variation in resource availability as induced by climatic variability may reduce the effects of competitive exclusion, allowing more species to coexist (Knapp et al., 2002).

A combination of extremes/multiple stresses may not only hamper performance but may also drive extinctions (Smith & Huston, 1989; Niinemets & Valladares, 2006). As functional trade-offs exist in adjusting to multiple environmental limitations (Holmgren et al., 1997; Silvertown et al., 1999), adapting to one stressor may go at the cost of adapting to another (Holmgren et al., 1997; Niinemets & Valladares, 2006). This trade-off among the tolerances to multiple environmental limitations hampers niche differentiation (Niinemets & Valladares, 2006). Bartholomeus et al. (2011a) demonstrated that the interaction between both the wet and dry extremes of plant water stress (oxygen/waterlogging and drought stress) is particularly detrimental to the survival of specialists and of endangered plant species. Both wet and dry weather extremes may increase due to changing climatic variability, thus increasing the risk of a combination of these stressors to occur at a site (Knapp et al., 2008; Bartholomeus et al., 2011a). This may favor generalists over specialists and rare species and thus influence vegetation dynamics and associated ecosystem services in response to changing climatic variability at the community level.

How can we quantify responses of plants to changing climatic variability?

Just as responses to global change in general (Rustad, 2008), the responses of plants to changing climatic variability can be assessed in observational, experimental, and modeling studies and combinations of these approaches (Fig. 2). All these approaches have their limitations in assessing a plant's perspectives of extremes: on the one hand, observational studies are by definition ‘opportunistic’ in the sense that extreme conditions such as a long-lasting drought cannot be planned (Smith, 2011). On the other hand, scaling and higher order interactions are an important issue in experimental and modeling studies (Leuzinger et al., 2011b; Wolkovich et al., 2012). Furthermore, it is crucial for any type of study that claims to assess climate variability to report whether changing mean climate and/or changing climatic variability have truly been measured and what the background variability of the system is over a well-defined time period. We qualitatively show this in Table 2 for a number of studies cited above as a first attempt to foster consistent reporting of studies dealing with climatic variability.

Table 2. Are we measuring the impact of mean climate or climate variability? Non-exhaustive list of the studies cited in the text and their testing amplitude in comparison to the background variability of the respective study system. The last column indicates in a qualitative way how well the testing amplitude accounts for climatic variability in terms of the background variability

Study system	Testing amplitude	Background variability	Study type	Reference	Testing climate variability?
European grassland & heath species	Drought: 32 days	Local 100-year extreme drought (number of days with precipitation <1 mm), 33 days of drought in 1976	Experiment	Jentsch et al. (2009)	Yes, 100-year-event
European grassland & heath species	Precipitation: 170 mm over 14 days	Local 100-year rainfall extreme, 152 mm of precipitation over 14 days in 1977	Experiment	Jentsch et al. (2009)	Yes, 100-year-event
European plant phenology	+1.5 (warm), +3 (very warm), −1.5 (cold) and −3 (very cold) standard deviations from the long-term mean at the respective grid point to classify warm and cold spells	Long-term mean	Observation	Menzel et al. (2011)	Yes, ±3 standard deviations from mean
Grasslands	6–8 large precipitation events per growing season (mean per event = 42 mm)	The size and frequency of precipitation events in the treatment are within the documented range of precipitation regimes of the past 100 years.	Experiment	Knapp et al. (2002)	Yes, but less than 100-year-event
Young oak stands (3 species (Quercus robur, Quercus petraea, Quercus pubescens), 4 provenances each)	Amount of irrigation water in drought-treated stands was 60% lower than the long-term mean precipitation (728 mm during the growing season from April to October) in 2007 and 43% lower in 2008 and 2009. Experimental droughts were imposed by stopping irrigation for several consecutive weeks during selected periods in the growing season.	In comparison to the long-term mean of the site, irrigation in the control was 16% lower in 2007, 26% higher in 2008 and 30% higher in 2009.	Experiment	Kuster et al. (2012)	Unclear, but testing amplitude much larger than variability in control
Mixed broadleaved forest in Central Europe	Seasonal precipitation: 50% of the 10-year mean from 1989 to 1999 Spring precipitation: below the mean Mean monthly temperatures: exceeded the 10-year mean from 1989 to 1999) (e.g., +6.8 °C for June)	Long-term mean	Observation	Leuzinger et al. (2005)	Unclear (background variability not further specified) but likely
Pinus halepensis stands (3 contrasting sites, 5 provenances)	Precipitation in Rome, 766 ± 156 mm; Tel Aviv, 557 ± 184 mm, and Yatir (semiarid), 279 ± 88 mm	Long-term mean (differences in mean climate are very large hence testing amplitude equals high background variability but no explicit testing of climate variability)	Transplantation	Klein et al. (under review)	Locally unclear but over the species distribution range probably yes
Tamarix africana Poiret	Continuous soil flooding with fresh and saline water during 45 days.	Not explicitly mentioned, plants survived 45 days of flooding	Experiment	Abou Jaoudé et al. (2012)	Unclear

Observational studies

Observational studies elucidate a plant's perspectives of extremes, if by chance they cover extremes. This makes them inherently opportunistic (Smith, 2011) unless they involve some retrospective elements such as dendrochronology. Observations from ‘extreme’ (from a plant's perspective) sites (e.g. from the leading and trailing edge of a population (Doak & Morris, 2010)) can help us to learn about the limits and coping range of plants. To this end, GIS mapping of ‘extreme’ sites within a species’ distribution requires careful interpolation of weather/climate data collected at appropriately distributed climate stations. However, ‘extreme’ sites are sometimes only poorly studied since they represent marginal ecosystems, whose services are not fully valued by society and have thus been outside the main focus of researchers. The psammophilic plants and vegetation of the beaches and dunes of the Portuguese coast, for example, are highly adapted to very specific environmental conditions and directly exposed to sea level rise, storms, and severe erosion processes. Unless their ecological requirements, functioning as communities and most influential physical drivers are understood, it will be difficult to study their responses to future climate change (Martins et al., accepted). It is, however, important to note, that in some disciplines there is a strong focus on extreme sites (such as on cold, high elevation or very dry sites in dendrochronological studies (e.g. Gruber et al., 2012)), which in turn may complicate studying mean climate impacts.

Generally, observational studies are well suited to study plant responses to changing mean climate, since long-term ecological data can be matched with increasingly available climatic observations. They are less suitable to gain a mechanistic understanding of plant responses to climatic variability since usually too many factors are involved and not all are measured.

Experimental studies

Experiments allow for controlled conditions and factorial experiments in the field and laboratory, have a long history in ecological research and are of crucial importance for global change studies (Luo et al., 2011)). When quantifying climate change impacts, however, field experiments can usually only test a limited number of factors and their combinations due to financial and logistic constraints (Templer & Reinmann, 2011). Therefore, interactions can often not be fully assessed (e.g. Wolkovich et al., 2012). Furthermore, to provide answers to the question of how extreme climatic events impact on ecosystems, experimenters should ensure that the applied treatment is indeed ‘extreme’ beyond the current background variability of the system over a well-defined time period, running the risk of killing plants (Leuzinger & Thomas, 2011; Beier et al., 2012).

Also, the temporal scale influences the outcome of an experiment. A comparable set of factors and a minimal experimental duration, for example, for all drought experiments would therefore be desirable. However, even then, most experiments would have to stop after few years. This raises the question whether the experiment actually simulates extreme situations or long-term change and whether or not the system recovers after the experiment ends. The high diversity in the response of growth parameters of oaks to drought as discussed in Kuster et al. (2012), shows that in experimental conditions, for example treatment duration and intensity, tree age or experimental set up, have to be considered in the evaluation of drought effects on trees. Thus, it is crucial to assess what degree of change and what temporal scale experiments cover if we want to evaluate whether they actually simulate responses to changing climatic variability, or rather to changing mean climate.

In a transplantation study, for example, the effect of a drying and warming trend was obtained by comparing tree performance in Rome (Italy), Tel Aviv (Israel), and Yatir (Israel) along a precipitation gradient (Klein et al., under review). The sites differed significantly in their mean annual precipitation, each representing a different climate type, but the responses were interpreted as drought acclimation. Results from this study captured many plant adjustments that were induced by both phenotypic plasticity and locally adapted ecotypes. Such transplantation experiments along altitudinal or latitudinal gradients do not require manipulation of the environment and may be an alternative to laboratory/greenhouse experiments. So far, transplantation experiments have not been considered in comparative studies of different artificial warming methods (e.g. Aronson & Mcnulty, 2009). However, such experiments seem to be well adapted especially for long-term experiments, as they project a realistic simulation of future climate conditions considering also the length of the growing period, one of the most important limiting factors in alpine plant growth (Jonas et al., 2008). Similar to laboratory/greenhouse experiments it is crucial that the results are interpreted in terms of changing mean climate and changing variability over well-defined temporal scales.

Modeling

Models can be used as diagnostic and predictive tools that integrate results from experiments and observation to gain mechanistic understanding and allow testing hypotheses generated from field data, experiments, and theory (Leuzinger & Thomas, 2011; Luo et al., 2011). Models have to be designed for a specific purpose and here we discuss which ones are suitable to simulate plant responses to changing climate variability. This is a highly relevant question, since models that account for extremes may require a different structure, for example an appropriate time resolution, to capture an extreme precipitation event. Many forest models for example use monthly input data and are thus unable to account for short-term extreme events (e.g. Bugmann, 2001). Forcing such a model with daily weather instead with monthly climate data improved its performance (Stratton et al., 2012). Zimmermann et al. (2009) argue that for capturing some ecosystem responses even daily climate data may be insufficient since they smooth meteorological extremes.

Generally, effects of climate change on ecosystems are analyzed by driving simulation models with output from GCMs and regional climate models (RCMs). To account for the uncertainty of climate change projections, besides different scenarios, also several GCMs (e.g. Buisson et al., 2010) and different realizations of a scenario may be used. Many models do not use the original GCM/RCM data at hourly resolution (which may also not always be available), but only daily or monthly aggregations and thus strictly speaking miss some of the meteorological variability. The CARAIB dynamic vegetation model (Otto et al., 2002; Laurent et al., 2008; Dury et al., 2011), for example, derives daily values of meteorological variables, as usual in large-scale simulations, from monthly mean outputs from GCM/RCMs using a stochastic weather generator (Hubert et al., 1998). The sequences of daily temperature or precipitation produced by the stochastic generator are renormalized to the monthly values generated by the RCMs. Thus, the precise day-to-day sequence of an extreme event in the model, such as a drought period or a succession of heat wave days (Beniston et al., 2007; Déqué, 2007), depends on the distribution functions used in the stochastic generator, although the monthly values of the climate model are not altered. While evidently it is challenging for such large-scale modeling efforts to integrate high-frequency climate variability, these studies are necessary to assess different feedbacks of vegetation types (e.g. feedbacks of ecosystem response to drying on near-surface temperature differ between forest and grassland ecosystems (Teuling et al., 2010) at the global scale).

Also, species distribution models face the challenge of including changing climate variability. Usually, they use information on species distribution (both potential from expert knowledge or forest communities, and actual from inventories and landcover-data) together with climate data to construct bioclimatic ranges (also called climate envelopes). They show a two dimensional frequency distribution of, for example temperature and precipitation, indicating the mean climatic range, in which the analyzed species (potentially) exist. Extrapolation of this information allows identifying regions with comparable climate to, for example, estimate a (extended) potentially occupied habitat (Guisan & Zimmermann, 2000) or new growing areas outside the recent (actual or potential) distribution (Miller et al., 2004; Peters & Herrick, 2004). Also the match of actual and future suitable ranges can be identified, classifying species into tolerant or intolerant to expected climatic conditions (Dunk et al., 2004; Gibson et al., 2004). This provides further understanding about expanding or shrinking habitats under changing climate (Erasmus et al., 2002; Midgley et al., 2006). Usually, climate envelopes are derived from mean values (e.g. mean temperature) and are thus designed to assess impacts of changes in mean climate. Consequently, especially regions at the edge of the distribution range may appear suitable, but in reality maximum or minimum precipitation or temperature may determine the distribution range (or other, non-climatic factors such as soil type or herbivory). This can partly be circumvented by including standard deviations as variables (Zimmermann et al., 2009), and species distribution models could also be built with extremes (e.g. maximum temperature or minimum precipitation) to enhance the predictive power. Zimmermann et al. (2009) for example found that incorporating climatic extremes slightly improved models of species range limits, since it corrected local over- and underprediction, but they also argue that climate variability rather complements the response to mean climate. Thus, including climate variability is one uncertainty of species distribution models that has to be considered to assess compliance of climate envelopes (P. Gloning, S. Taeger, H. Seifert, U. Schäffler, C. Kölling, M. Schilcher & A. Menzel, in preparation).

Although generally process-based modeling is required to derive climate-robust relationships to predict vegetation characteristics (Franklin, 1995; Guisan & Zimmermann, 2000; Schwalm & Ek, 2001; Botkin et al., 2007; Suding et al., 2008; Hajar et al., 2010), this is even more evident when considering changing climate variability. Bartholomeus et al. (2011b) demonstrated that, in contrast to process-based relationships between site factors and vegetation characteristics, relations based on indirect site factors produce systematic prediction errors when applied outside their calibration rate, and so cannot be used for climate projections. Mean groundwater level, for example, is only an indirect site factor related to plant performance, as it is the interaction between soil-water-plant-atmosphere that essentially determines if plants suffer from drought stress or oxygen/waterlogging stress. When, for example, soil moisture availability is too low to meet the water demand for transpiration, a plant suffers from drought stress (Schimper, 1903; Reddy et al., 2004). This so-called physiological drought (Schimper, 1903), implies that not only water availability but also vegetation's demand for water has to be considered. Instead, more process-based explanatory variables are needed to predict the effects of changing climate variability on the species composition of the vegetation. These explanatory variables should consider the interacting meteorological, soil physical, microbial, and plant physiological processes in the soil-plant-atmosphere system. Bartholomeus et al. (2011a) did so for water-related stressors, by simulating respiration reduction (reflecting the combined effect of high temperature and low oxygen availability), and transpiration reduction (reflecting the combined effect of high atmospheric water demand and low water availability) for a reference vegetation. The simulated stress for reference vegetation acts as a habitat characteristic, that is a measure for the moisture regime of the soil to which the actual vegetation will adapt. The use of reference vegetation improves the applicability of models in which stress measures are implemented, especially in predicting climate change effects (Dyer, 2009).

Combined approaches

Combined approaches unite experimental, observational, and/or modeling studies. A recent meta-analysis shows that the temperature sensitivity of phenology in warming experiments is underestimated in comparison to observations (Wolkovich et al., 2012). It highlights that observational studies are crucial to test whether or not experimental results match observations in natural systems. A combination of laboratory and field studies is necessary to determine whether or not thresholds detected in the laboratory are also likely to occur in the field. This is especially relevant when calculating the effects of changing climatic variability. We take leaf gas exchange and ecosystem flux measurement data from Brilli et al. (2011) as an example of how to link experiments and observation at different scales and how an experiment can complement observations to study plant responses to climate variability. Figure 3 shows that evapotranspiration measured in the field with the eddy covariance method, was insensitive to soil drying over the range of soil water contents occurring in the field. The leaf gas exchange measurements during the laboratory drought experiment when extended to much drier conditions showed that the plant species occurring at this site start to down-regulate stomatal conductance at soil water contents close to the wilting point – conditions that have never been reached in the field during the observational period of 2001–2009. Back-of-the-envelope calculations suggest that ca. 10 additional rain-free days would have been required even during the 2003 and 2006 droughts in order for plants at this site to experience gas exchange limitations. Such information is crucial to assess whether responses to changing mean climate or to changing climate variability are measured.

Figure 3. Evapotranspiration measured in the field with the eddy covariance method (black filled dots) over the range of soil water contents (gray bars) occurring in the field and stomatal conductance measured in a laboratory experiment (black open dots). Data and further descriptions are available in Brilli et al. (2011). SWC, soil water content.

Moreover, results can be extended to a larger spatial scale, by combining simulation models with research tools like raster GIS (Minacapilli et al., 2009; Bonfante et al., 2011) and Digital Elevation Model (DEM) derived analysis (Macmillan et al., 2000). Furthermore, studies that combine observational or experimental results – at field scale – with simulation models of hydro-thermal regime – at landscape scale – allow to quantify the effects of changing climate variability (Bonfante et al., 2010). Riccardi et al. (2011) assessed the adaptive capacity of olive cultivars to future climate by means of a database of cultivars’ climatic requirements, combined with a spatially distributed model of the soil–plant–atmosphere system. They set up a database on climatic requirements and defined critical environmental conditions using two quantitative indicators of soil water availability (the relative evapotranspiration deficit, i.e. the ratio of actual to maximum evapotranspiration of the crop, and the relative soil water deficit, i.e. the ratio between the actual and the maximum volume of soil water available to plants taking into account the water retention characteristics, to get a comparable indicator across soil types). The response in terms of yield of several olive cultivars to these indicators was determined through the re-analysis of experimental data derived from scientific literature (Moriana et al., 2003; Tognetti et al., 2006). This database on cultivars’ requirements was used in combination with a plant-soil-atmosphere model (SWAP, Van Dam et al., 2008). The model was used to describe the soil water regime at landscape scale under future climate scenarios from statistically down-scaled GCMs, resulting in several realizations (Tomozeiu et al., 2007). The indicators of soil water availability were thus determined in different soil units, and were compared with the limits set for each cultivar. A cultivar was considered tolerant to expected climatic conditions when the indicator values resulted above critical values in at least 90% of realizations. Although Riccardi et al. (2011) did not further specify the climate scenarios and realizations in terms of changing mean or climate variability, such analysis could be easily linked to the soil water availability indicators and the related limits for cultivars under climate change.

Conclusions

In this review, we have emphasized that changing climatic variability and the resulting extreme (climatic) conditions are highly relevant for different plant processes at different scales in comparison to changes in mean climate (although mean and variability may not be fully independent of each other). We have also shown how to quantify responses of plants to changing climate variability: Although experiments seem to be well suited to study the effects of changing climatic variability it is important to remember that they only control a limited number of factors. For modeling studies we stress that the model structure should allow integrating extreme events (e.g. by having the appropriate temporal resolution). These points highlight the importance of linking experiments, observations, and modeling studies as well as assessing study results in light of the background variability of the system and the temporal scale considered. We also identified several research gaps. Although knowledge of plant responses to changing climatic variability for individual processes has to be consolidated, we still lack knowledge on how interactions of these processes and other environmental variables play out at different hierarchical levels and in combination with changing mean climatic conditions. Similarly, although there is room to improve individual methods to study changing climatic variability, there is a particular need to integrate observations, experiments and model results across scales.

Ultimately, the information on extremes and corresponding vulnerability of plants are crucial to identify which species and regions (and thus which ecosystem services and functions) are most at risk from climate change. Moreover, designing ecosystem-based adaptation strategies to climate change relies on understanding the interactions between species’ natural adaptive capacity and climate change. Analyzing plant responses to climate variability is important to determine drivers of ecosystem dynamics over time (slow vs. fast processes) and highlights the importance of extremes to assess the impacts of environmental change on socio-ecological systems.

Acknowledgements

This review synthesizes and expands the results from a session, which was held during the 2011 European Geoscience Union (EGU) general assembly (BG2.7). We are grateful to all participants of this session for the valuable discussions. CR acknowledges funding from the EC FP7 MOTIVE project (grant agreement no. 226544). SL was funded by EC FP7 ACQWA. AR acknowledges funding from the EC FP7 project CARBO-Extreme (grant agreement no. 226701). FDL acknowledges funding from the MIPAAF-IT project AGROSCENARI. We are grateful to one anonymous reviewer for intelligent comments on an earlier version of this article.