Extreme Climatic Events as Drivers of Ecosystem Change

The ecosystems present a great diversity worldwide and use various functionalities according to ecologic regions. In this new context of variability and climatic changes, these ecosystems undergo notable modifications amplified by domestic uses of which it was subjected to. Indeed the ecosystems render diverse services to humanity from their composition and structure but the tolerable levels are unknown. The preservation of these ecosystemic services needs a clear understanding of their complexity. The role of research is not only to characterise the ecosystems but also to clearly define the tolerable usage levels. Their characterisation proves to be important not only for the local populations that use it but also for the conservation of biodiversity. Hence, the measurement, management and protection of ecosystems need innovative and diverse methods. For all these reasons, the aim of this book is to bring out a general view on the function of ecosystems, modelling, sampling strategies, invading species, the response of organisms to modifications, the carbon dynamics, the mathematical models and theories that can be applied in diverse conditions.

populations to stressful abiotic conditions is, in many cases, non-linear and sensitive to discrete thresholds Beniston & Stephenson 2004). As such, accurate quantification of the likelihood and intensity of extreme climatic events is essential. Unfortunately, this task is made difficult by the fact that no single definition of what constitutes an extreme event actually exists, and the impacts of unusual climatic situations on ecosystems have not traditionally been studied in a systematic manner (Gutschick & BassiriRad 2003;Smith 2011). As noted by Beniston & Stephenson (2004), extremes can be defined in terms of rarity, intensity or impact, although individually these definitions may fail to capture the critical features of climatic variability that impact on plant and animal populations. Smith (2011) suggests that an extreme climatic event should be defined as a rare or unusual climatic period that alters ecosystem structure and/or function well outside what is considered normal variability, and many of the studies discussed below inherently use this concept, despite the fact that the rarity of the climatic conditions being studied is often not formally described.
From a quantitative standpoint, a range of approaches are used to characterise extreme events. The Intergovernmental Panel on Climate Change (IPCC 2011) defines a climate extreme as "the occurrence of a value of a weather or climate variable above (or below) a threshold value near the upper (or lower) ends of the range of observed values of the variable" (p. 2). Indices derived using this approach usually quantify the duration or frequency of events which exceed a specific temperature or precipitation threshold, for example total number of frost days, growing season length, heat wave duration, and number of consecutive dry days (e.g., Tebaldi et al. 2006). Another commonly-used method is to define extreme events as those occurring within a certain percentile range (often the 5 th , 10 th , 90 th or 95 th percentiles) of a climatological distribution within a given timeframe (see Bell et al. 2004). Hennessey et al. (2008) define exceptional droughts as those of one year in duration and occurring, on average, once every 20 years (i.e., a 5% probability of occurring within a given year; Katz et al. 2005).
The standard statistical approach for quantifying climatic variation is to fit a probabilistic model to a given climatic data set and then to evaluate the likelihood (and severity) of specific climatic events based on the associated probability density function. In recent years a cohesive statistical theory of extreme events has emerged (Coles 2001). These techniques, and their application to ecological problems, have been discussed in detail elsewhere; readers are directed to Katz & Brown (1992), Gaines & Denny (1993), Katz et al. (2005), and Resnick (2007) for details. The basic statistical approach can be visualised in Fig. 1; for further discussion see Meehl et al. (2000) and IPCC (2011). Consider a normally-distributed climatic variable, such as temperature (Fig. 1a). For this probability density function (PDF), the top 5% of values, which may be classified as extreme, lie 1.65 standard deviation (σ) units above the mean (PDF I; Fig. 1a). A shift in climate alters the frequency and severity of extreme events, with the magnitude of change depending on the location and shape of the new distribution (Fig. 1b). In PDF II (Fig. 1b), the distribution undergoes only a mean shift with variability (σ; the standard deviation) remaining constant, this results in an increase in the frequency of climatic events of a given magnitude (from 5% in PDF I to 26% in PDF II). Similarly, the incidence of extreme events at the other end of the distribution declines. A change in the shape of the distribution (increase in σ in PDF III, Fig. 1b) results in a further increase in the frequency of extreme events (PDF III, Fig. 1b), indeed more than an equivalent change in the mean (Katz & Brown 1992;Meehl et al. 2000 Frequency of extreme events for a given hypothetical climatic variable showing variation in probability related to change in shape and location of the probability density function (PDF). a) normal distribution (PDF I; µ = 0, σ = 1) with the 5% most extreme events occurring 1.65 or more standard deviation units from the mean (vertical dashed line). b) Increase in frequency of extreme events following increases in the climatic mean (PDF II; µ = 1, σ = 1) and standard deviation (PDF III; µ = 1, σ = 1.5). c) Generalised extreme value (GEV) distribution with location parameter (μ) = 0, scale parameter (σ) = 1, and shape parameter (ξ) = 0.1 (PDF IV) showing the larger tail that characterises the GEV PDF, with the 5% most extreme events lying above the vertical dashed line. An increase in the parameter σ to 1.5 (PDF V) increases the spread of the distribution and the probability of extreme events that lie in the upper tail. The shape of the GEV function is determined by ξ (see Katz et al. 2005 for details).

Density function
www.intechopen.com While some climatic variables can be approximately normally distributed, many others have heavy tails with significant skew towards large values (Gaines & Denny 1993;Gutschick & BassiriRad 2003). For example, precipitation-related variables tend to be non-normal, and are often modelled using gamma (Watterson & Dix 2003) or lognormal (Cho et al. 2004) distributions. Model selection is critical, since the objective is to adequately characterise the distributional tails. One family of distributions widely used for modelling extreme values, especially maxima, is the generalised extreme value (GEV) probability density function, which produces light-, heavy-and bounded-tailed distributions (Katz et al. 2005). A typical heavy-tailed (Fréchet) GEV distribution is shown in Fig. 1c (PDF IV). Shifts in location (µ), scale (σ) and shape (ξ) parameters that define the GEV distribution all impact on the frequency and severity of extreme events (Fig. 1c, PDF V).

Demographic change and plant mortality
Probably the most important mechanism by which extreme climatic events (ECEs) can drive rapid ecosystem change is by causing extensive mortality in dominant or keystone plant species. The loss of such species can result in rapid, cascading compositional change in plant communities, in turn affecting the population viability of faunal and floral community associates (Gitlin et. al. 2006) and the ecosystem services that they provide (Walker et al. 1999;Kremen 2005). Unfortunately, ECEs are difficult to study within a pre-planned experimental and statistical framework (Buckland et al. 1997), and so little or no information exists on the responsiveness of most plant communities to ECEs of varying severity. However, a range of studies provide insight into the dramatic and persistent impacts of prolonged abiotic stress on plant community composition and structure.
The classic studies of Weaver (1944, 1945) on prairie and woodland ecosystems in North America were among the first, and most comprehensive, to document the impacts of protracted, multi-year drought on plant communities. During the 1930-1940 "dustbowl" years, much of the central US experienced record-or near-record low rainfall, high temperatures, high evaporation, and declining soil moisture. In the most extreme years (e.g., 1934, 1936, 1939), rainfall was up to 40% below normal, summer maximum temperatures were 3-6°C above normal, evaporation exceeded that of non-drought years by up to 33%, and the water table fell by one metre or more (Albertson & Weaver 1945). This drought was one of the three most extreme to affect North America since 1900 (Cook et al. 2004), although not as severe as previous megadrought periods that occurred in the 12 th to 16 th centuries (Woodhouse & Overpeck 1998;Stahle et al. 2007).
The most striking effect of this drought was the high level of mortality that occurred in virtually all plant species across the landscape. Mortality of dominant and subordinate tree species (e.g., Ulmus americana, Populus sargentii, Celtis occidentalis, Salix spp.) exceeded 50% across a broad range of topoedaphic habitats with distinctly different hydrological regimes, most notably in dry ravines and along intermittent creeks (Albertson & Weaver 1945). In prairie communities perennial grasses such as Andropogon scoparius, Koeleria cristata and Poa pratensis suffered up to 80-90% mortality, while subordinate grasses and forbs were almost eliminated (Albertson & Weaver 1944). The catastrophic loss of groundcover resulted in intense wind erosion, with dust accumulating to depths of 2 feet (60 cm) in sheltered locations (Albertson & Weaver 1945). This dust exacerbated the severity of water stress experienced by plants during the drought by preventing rain infiltration into the soil, effectively blocking moisture from reaching the rhizosphere (Albertson & Weaver 1945). Insect attack also increased the impact of drought on tree populations (Albertson & Weaver 1945;Mattson & Haack 1987).
Other studies report similar effects during prolonged periods of exceptionally severe drought. Record dry conditions experienced during 2006-2007 at a semi-arid grassland site in south-eastern Australia resulted in 90% or higher mortality of the dominant grass species Austrostipa aristiglumis (Godfree et al. 2011). Here, the most acute period of rainfall deficiency occurred during the middle of a decade-long drought. Similarly, Edwards & Krockenberger (2006) reported 64% mortality among seedlings of rainforest species during the 2002 ENSO event in north-eastern Australia, during which rainfall at their study site was only 36% of average. On a much larger spatial scale, mortality of piñon pine (Pinus edulis) in western North America varied between 40% and >90% in response to drought during 2000(Breshears et al. 2005. The severity of this drought, which was among the three driest in the past century, was exacerbated by high temperatures (Breshears et al. 2005).
Similar levels of drought-induced mortality have been observed in Canadian aspen forests (Hogg et al. 2008), beech forests (Peterken & Mountford 1996), and in forests globally (for review see Allen et al. 2010). Indeed, forest mortality in response to drought and heatwaves is so common that Allen et al. (2010) concluded that no forest biome is invulnerable to climate change, even in systems that are not thought to be water-limited. However, while droughts do not have to be of unprecedented or record severity to cause significant mortality in plant communities, it is also clear that not all droughts result in high levels of plant mortality (e.g., Condit et al. 1995;Fensham and Holman 1999;references in Allen et al. 2010); and not all generate detrimental, lasting effects on plant populations (e.g., Morecroft et al. 2002;Yurkonis & Meiners 2006).
Briefly, other extreme climatic events can also have strong, direct impacts on ecosystems by inducing injury or mortality in plants. High temperatures, especially those exceeding 55°C, when Rubisco activity, electron transport and overall photochemical performance becomes impaired (Kappen 1981;Musil et al. 2009) are known to be detrimental to plants. Simulated heatwaves in which temperatures approach or exceed these temperatures have been shown to result in canopy decline and mortality of succulent plant species (Musil et al. 2005); in this study mortality in heated treatments ranged from 33-74% compared with 7-38% in controls that experienced 5.5°C cooler daytime temperatures. Such extremes are probably most likely to occur in arid and semi-arid systems where lower-canopy species lack cover, and may in part explain the loss of low-stature mesophyllic grasses and forbs following death of overstorey plants in drought-affected grasslands and woodlands Weaver 1944, 1945). Studies conducted in tundra environments have also resulted in physiological impairment and mortality in cold-adapted species (Marchand et al. 2006); numerous examples exist in other systems (e.g., Van Peer et al. 2001;Groom et al. 2004, Larcher et al. 2010.

Community composition and structure
During drought, a range of physiological, demographic and environmental factors interact to determine the impact of extreme water deficiencies on specific plant individuals and species. McDowell et al. (2008) recently reviewed these mechanisms; briefly, the key drivers of plant mortality under moisture stress are thought to be carbon starvation, the activity of biotic agents, and hydraulic failure, with the relevance of each depending on the intensity and duration of stress. Plants adopt a range of mechanisms to tolerate or resist drought, with intraspecific and interspecific variation found in water use efficiency (Farquhar et al. 1989), dormancy (Oram 1983), dehydrin expression , extraction of water at low soil water potential , senescence of aerial tissue (Volaire et al. 1998;Bolger et al. 2005), root structure (Van Splunder et al. 1996) and resource allocation (Aronson et al. 1993) to name a few. Indeed, most species use different physiological and anatomical mechanisms to protect tissue during periods of moisture stress (Scott 2000).
Plant survival also depends strongly on spatial heterogeneity in the landscape. Variation in soil moisture occurs at a range of spatial scales (Buckland et al. 1997;Gitlin et al. 2006;Dobrowski 2011;Godfree et al. 2011), and because plant mortality is non-linearly related to soil moisture content, microscale variation in water availability can critically influence plant survival under extreme drought (Godfree et al. 2011). At larger spatial scales, hydrological variation and the frequency of soil drought is a key driver of species assortment and community composition (Oberbauer & Billings 1981;Buckland et al. 1997;Yurkonis & Meiners 2006), a process which reflects variation in the ability of species to recover from drought (e.g., Tilman & El Haddi 1992;Stampfli & Zieter 2004) as much, or more, than tolerance of drought itself (Gutschick & BassiriRad 2003).
Given these sources of variation, it is not surprising that plant species tend to show highly differential mortality when placed under extreme drought in natural settings. During drought in the 1930's, Weaver (1944, 1945) observed that survival among tree species in ravine environments ranged from 64% (Celtis occidentalis) to only 30% (Ulmus americana), and drought hardy species, such as Juniperus virginiana, had much lower mortality (1-37%) than all other co-occurring species, even persisting as monospecific stands on the most xeric sites (Albertson & Weaver 1945). In prairie communities, all species were affected but some, including Andropogon furcatus, suffered much lower mortality than others (e.g., Andropogon scoparius, Stipa spartea) which experienced over 80% mortality. Survivorship was linked strongly to depth of the rooting system, since during the drought soil water deficiencies gradually moved lower in the soil profile (Albertson & Weaver 1944). Interestingly, over the entire drought period the composition of the prairie community was in constant flux (Albertson & Weaver 1944). Interspecific variation in mortality has also been observed in numerous other studies (Tilman & El Haddi 1992;Condit et al. 1995;Gitlin et al. 2006), and in response to other climatic stressors (e.g., Stiles 1930; Barua et al. 2003;Henry & Molau 2003;Marchand et al. 2006; many others).
Under extreme, prolonged drought, rapid changes in vegetation composition can be persistent or effectively permanent. Perhaps the best documented example occurred during extreme drought in the 1950's, when the ecotone separating Pinus ponderosa forest from Pinus edulis-Juniperus monosperma woodland moved by 2 km or more within only five years. This change has persisted for at least 40 years (Allen & Breshears 1998). One of the most significant persistent changes observed by Albertson & Weaver (1944) was the expansion of wheat grass (Agropyron smithii), which was favoured by moist springs and dry summers, at the expense of other species such as Andropogon furcatus. Elsewhere, Gitlin et al. (2006) showed that continuation of extreme drought conditions in the southwestern United States experienced during 2002 would likely result in significant change to the composition and structure of entire plant communities and ecosystems. Similarly, Tilman & El Haddi (1992), in a controlled experiment, showed that post-drought recovery of grassland vegetation was not accompanied by a significant recovery in species richness, and suggested that postdrought recruitment limitation may determine the richness of prairie plant communities.
Here, drought with a return interval of approximately 50 years was sufficient to significantly alter local species richness (Tilman & El Haddi 1992). Drought can also alter competitive relationships among species, resulting in stratification of plant communities along hydrological gradients (Buckland et al. 1997) or promotion of invasion by fast-growing annuals (White et al. 2001).

Case study: Change in composition of an Australian semi-arid grassland during and following extreme drought
The case studies above provide overwhelming evidence that extreme climatic events, and especially drought, can drive rapid changes in plant community composition by causing differential rates of mortality and recovery among plant species. A central remaining challenge to ecologists, however, is to understand when, and under what threshold conditions, abiotic stress will lead to changes of large magnitude (McDowell et al. 2008), knowledge that will be essential if we are to accurately predict the impact of climate change on natural vegetation globally. In this section I aim to improve our understanding of such processes by drawing conclusions from a study in which I investigate the impact of a multi-year period of exceptional drought on a semi-arid Australian grassland ecosystem.

Background: The "Millenium Drought" in Australia
Between 1997 and 2009 much of south-eastern Australia was affected by an extremely severe and protracted drought known as the "Millennium Drought" (Whitaker 2006;Bond et al. 2008). During this period, annual rainfall was well below average, especially during autumn (March-May), while maximum and minimum temperatures were at or near record highs (Murphy & Timbal 2008). Over the past half century mean maximum and minimum atmospheric temperatures have increased over most of Australia Nicholls 2006) and extreme temperature events have become more frequent (Collins et al. 2000). Recent droughts also may have become more severe due to increased evaporation associated with warmer temperatures (Nicholls 2004;Cai & Cowan 2008). Severe drought is a regular phenomenon in Australia ), and there is a strong suggestion that rising atmospheric greenhouse gas concentrations have played a role in the development of rainfall deficiencies and elevated temperatures across southern parts of Australia in recent decades (Timbal et al. 2006;Murphy & Timbal 2008).
The most extreme droughts in Australia impact on regional agriculture, hydrology, ecosystem function, and the population dynamics of flora and fauna. The 1997-2009 drought was no exception: extreme low rainfall and high temperatures (Cai & Cowan 2008) during the drought resulted in the near total loss of surface water and a persistent decline of groundwater (Leblanc et al. 2009), which in turn reduced agricultural productivity significantly (Horridge et al. 2005;Pook et al. 2009). Large impacts were observed on aquatic flora and fauna (Bond et al. 2008; references therein), with extensive tracts of the key riparian tree species Eucalyptus camaldulensis dying in the lower Murray-Darling Basin (Cunningham et al. 2009). While Australian ecosystems are commonly perceived to be resilient to drought (e.g., Bond et al. 2008), their ability to recover from events of this severity and duration, or the increasingly extreme events of the future , remains open to question (Godfree et al. 2011).

Study objectives
The objective of this study was to improve understanding of the role of extreme climatic events as drivers of rapid ecosystem change by comparing the landscape-level responses of different plant species to drought across multiple grassland habitats, and investigate whether the observed phytosociological changes could be explained by simple predictors including statistical quantification of drought severity, topography, and knowledge of the broader range distributions and habitat affinities of the study species. I specifically addressed the following hypotheses: 1. prolonged drought of unprecedented severity will result in high mortality across a range of grassland plant species; 2. mortality will be highest in the most xeric sites and mesic low-lying habitats will act as refugia during drought; 3. post-drought recruitment and recovery will be the primary drivers of post-drought community composition and structure; and 4. species with ranges that extend further into drier regions will have higher survival and recruitment than species with more mesic distributions.
Evidence supporting these hypotheses might indicate that at least some basic principles could be generally applied to the study of extreme events to improve prediction of their impacts on vegetation systems.

The study system
The study was conducted in a high quality remnant semi-arid grassland located approximately 30 km to the east of West Wyalong in central NSW (Fig. 2). The choice of a semi-arid biome reflects the general view that these ecosystems are highly susceptible to shifts in climate (Allen & Breshears 1998;Holmgren et al. 2006). The topography of the 34 hectare site is mainly flat with extensive treeless plains dissected by a series of small creeks that incise up to 2-3 m below the surrounding terrain. While low, this topographic heterogeneity does generate a range of habitat types characterised by different floral assemblages (illustrated in Fig. 2 prolutum and L. panaetioides all grow on the western slopes and plains regions of NSW, Queensland and Victoria, but P. prolutum and L. panaetioides extend further into the drier, semi-arid zone than A. aristiglumis, and L. panaetioides occurs in arid habitats in far western NSW (Fig. 3).    (Fig. 4a). Temperatures during this period were also at or near record levels (see below), which exacerbated drought severity (see Nicholls 2004).

Field surveys
The

Demographic change in response to drought
Prior to the major mortality event that occurred in late 2006, A. aristiglumis dominated the more mesic and flat habitats at the study site, while P. prolutum was most abundant in the Panicum flats habitat (Fig. 5a). In drier habitats (south-and north-facing slopes) both grasses had much lower densities, being largely replaced by the more xerophytic shrub L. panaetioides (Fig. 5a). By late 2007, however, the population density of A. aristiglumis and P. prolutum had changed considerably (Fig. 5b), reflecting drought-induced mortality followed by a major recruitment event in autumn and winter 2007.
During the most extreme phase of the drought mortality of A. aristiglumis and P. prolutum occurred in all habitats, with populations in the more xeric, sloping habitats suffering losses of 90% or more (Fig. 6a). P. prolutum suffered >65% mortality in all habitats and was eliminated from north-facing slopes (the most xeric habitat), while A. aristiglumis survival exceeded 30% only in the most mesic terrace environments (Fig. 6a). In contrast, L. panaetioides survival was at least 70% in all but one habitat (north-facing slopes) and no plants died in the terrace and gully habitats (although density was low to start with). Mean site survival, averaged across habitat types, was 21% for A. aristiglumis, 17% for P. prolutum, but 79% for L. panaetioides.
www.intechopen.com  Habitat types are arranged from most mesic (terraces) to most xeric (north-facing slopes). np = species not present in the habitat type; n/a = not applicable.
Spatially, each of the habitats accounted for a different proportion of the area of the study site, and the impact of the drought on the total population size of each species reflected both mortality rate and the spatial extent of each habitat in which it occurred. For A. aristiglumis, only 16% of all plants at the site survived the drought, since high mortality (96%) was observed in the extensive Austrostipa flats habitat. P. prolutum survival was highest (34%) in habitats where its original density was also highest (2.0 m -2 ), and so overall survival (27%) exceeded that of A. aristiglumis. The relatively high total survivorship of L. panaetioides (73%) also strongly reflected generally high survivorship across habitat types.
Following rain in autumn 2007, significant recruitment of A. aristiglumis and P. prolutum, but not L. panaetioides, occurred across the study site. Recruits exceeded the number of droughtkilled plants in gullies and terraces (A. aristiglumis) and gullies, terraces, Austrostipa flats and south-facing slopes (P. prolutum). Recruitment was minimal in L. panaetioides, and in no habitats did recruits fully replace plants that succumbed to the drought (i.e., adult plant replacement rate < 1; Fig. 6b). These differences in mortality and recruitment over the 2006-2007 study period resulted in a significant spatial redistribution of plant species across the study site, and a landscape-level change in community composition. By late 2007, A. aristiglumis was predominantly restricted to terrace and gully habitats, where post-drought populations were actually larger than pre-drought populations (Fig. 5b), and was virtually absent from xeric habitats (Fig. 5b). P. prolutum, in contrast, increased in abundance in most habitats, and became co-dominant with A. aristiglumis in Austrostipa flat and south-facing slopes habitats, and increased its dominance in the Panicum flats habitat (Fig. 5b). L. panaetioides declined in all habitats, but maintained dominance on the most xeric northfacing slope environments (Fig. 5c). In reality, however, the decline of all species in this habitat (Fig. 5c) left it essentially bare (Fig. 2), a condition which has been largely maintained for at least 3 years since (R. Godfree, personal observation).

Implications for predicting the impacts of extreme events
The results of this work, and those published previously (Godfree et al. 2011), support the hypothesis that extreme climatic events can significantly reconfigure landscape-scale vegetation mosaics within relatively short timeframes via direct mortality of established plants. While other studies have reported very high plant mortality during extreme drought (e.g., Weaver 1944, 1945;Allen & Breshears 1998;Breshears et al. 2005;Gitlin et al. 2006;Edwards & Krockenberger 2006), the >90% mortality rates observed in this study in two community dominant species do appear to be unusually high. Perhaps this ultimately reflects the magnitude of rainfall deficiencies observed at the study site -2006 was the driest year in at least a century, and, surprisingly, 20% drier than the next driest year (2002). Heavy mortality can have a range of important demographic and genetic consequences for the long-term fitness of plant populations and species, and if the frequency of events such as the one described here increase under projected climatic change Hennessy et al. 2008;Planton et al. 2008), the consequences for the conservation of native vegetation are likely to be significant.
As a result of the 2006-2007 drought, populations of two of the three study species (A. aristiglumis and L. panaetioides) shifted lower in the landscape, with the dominance of A. aristiglumis declining in all but the most mesic terraces and gully environments. Consistent with hypothesis 2, all three species suffered the greatest mortality in the more xeric environments, and at the height of the drought, live plants were restricted almost entirely to mesic refugial habitats (with the exception of L. panaetioides, which although completely defoliated, did survive in modest numbers in drier areas). Interestingly, however, there was also some evidence that pre-drought habitat suitability was not a good predictor of drought survival or post-drought recruitment. For example, Austrostipa and Leiocarpa both declined most in the habitats in which their pre-drought populations were most dense (Figs. 5a, 6a), and recruitment of P. prolutum was poorer in the Panicum flats habitat than in most other habitats. In the case of A. aristiglumis and P. prolutum it is possible that terrace habitats were actually more suitable for these species (the plants present, although of lower density, were much larger than in drier habitats). This did not seem to apply to L. panaetioides -plants were large and abundant on the most xeric sites. Perhaps L. panaetioides is competitively excluded from mesic habitats by A. aristiglumis and P. prolutum and drought alters this competitive hierarchy, a pattern that has been observed elsewhere (White et al. 2001). Regardless, mesic refugia, albeit small in spatial scale (Godfree et al. 2011) clearly play a key role in ensuring survival of a range of species in a given plant community during extreme drought, including those that are adapted to drier conditions.
The pattern of change among species observed at the study site did not support the hypothesis that post-drought recruitment and recovery are the primary drivers of postdrought community composition for all species, the demographic responses of which were highly idiosyncratic. The final distribution of P. prolutum mainly reflected strong postdrought recruitment across multiple habitats, but A. aristiglumis was most abundant in terrace habitats due to significant post drought recruitment and high mortality. Leiocarpa density depended almost solely on high drought survivorship, and the final composition of the vegetation found in the more xeric habitats primarily reflected the drought hardy nature of this species. The presence of such complex patterns is perhaps not surprising given the diversity of strategies displayed by plants for ensuring survival through drought and other abiotic stresses (e.g., Barrett 1998;Mal & Lovett-Doust 2005;McDowell et al. 2008) but it does indicate that post-drought community composition jointly reflects the processes of mortality and recovery in heterogeneous environments.
Finally, from a practical point of view, the species-level responses observed here do partly support the hypothesis that population behaviour in response to drought can be predicted by their broad climatic envelopes (see McDowell et al. 2008;references therein). Drought survival of A. aristiglumis was much lower than that of the more arid-adapted (see Fig. 3) L. panaetioides in all habitats, with differences in survivorship between the two species being greatest in the more xeric, sloping habitat types (Fig. 6a). Differences between A. aristiglumis and P. prolutum were not as clear, since survival of A. aristiglumis was actually higher than that of P. prolutum in three habitats (terraces, gullies and north-facing slopes; Fig. 6a), which does not appear to be consistent with the fact that P. prolutum is capable of surviving in much drier regions than A. aristiglumis (Fig. 3). On the other hand, P. prolutum populations did perform better overall than A. aristiglumis, mainly as a result of a higher rate of postdrought recruitment from the seedbank (Fig. 6b). Perhaps the presence of a large persistent seedbank, rather than high drought survivorship, explains why P. prolutum grows in areas that are considerably drier than A. aristiglumis can tolerate (e.g., far western NSW and Queensland; Fig. 3). Such traits are known to be linked to population fitness and reproductive assurance in arid environments (Auld et al. 1995;Facelli et al. 2005).

Drought severity and species responses
This study has shown that extreme drought drives changes to ecosystem structure and composition by impacting on mortality and recruitment of plant populations, and that these processes may be broadly predictable given an understanding of drought severity and community composition. But this observation begs the questions: how extreme do droughts need to be to result in changes of this magnitude, and how much might drought severity increase under anthropogenic global warming?
Let us return to the study system at hand. As mentioned, drought conditions at the study site d u r i n g 2 0 0 6 , w h e n o n l y 1 8 1 m m o f r a i n f e l l , c a n r e a s o n a b l y b e d e s c r i b e d a s b e i n g o f unprecedented severity with respect to the 1900-2009 instrumental record (Fig. 4a). Based on historical annual rainfall data for Wyalong, NSW, fit with a gamma probability distribution (Fig. 4b), an annual rainfall of 181 mm has a predicted return interval of 453 years, and although care needs to be taken when such low probabilities are involved, it is obviously an exceptionally rare occurrence indeed. As noted previously, the mortality rates observed in this study are consistent with the impacts of exceptional drought observed in other systems. We simply do not understand the exact conditions that resulted in the observed changes at the study site, beyond the fact that many weeks of dry weather occurred during spring and summer at the end of a very dry year in the middle of a decade-long drought. Perhaps we can at best speculate that substantial mortality of natural, minimally disturbed grassland in the study region is most likely to occur when extremely dry years (roughly 250 mm, around 50% of average; return interval 40 years) occur during an extended period of below-average rainfall. There might have been as many as 3 to 5 instances of such conditions over the past 110 years (Fig. 4a). However, it may require extremely rare events, like 2006, to generate the high levels of mortality observed in this study.
A further complication is that the actual level of water stress experienced by a plant population is a function of evapotranspirational demand relative to water availability. The extremely dry conditions experienced in central NSW during the Millenium Drought were exacerbated by high temperatures (Cai & Cowan 2008), which suggests that consideration of rainfall deficiencies alone would underestimate the severity of the drought. Data from Wagga Wagga AMO (S 35.16°, E 147.46°; Fig. 7), the nearest station with suitable observations for estimating potential evapotranspiration (ETo), show the severity of 2006 in terms of low rainfall (Fig. 7a), high ETo (Fig. 7b), and very low atmospheric water balance (AWB; calculated here as annual rainfall -ETo). Indeed, the AWB was considerably lower in 2006 than any year since at least 1970 (Fig. 7c)

Predictions under climate change
We may conclude this study by making some very rough guesses as to the possible impacts of drought on grassland vegetation at the study site under climate change. First, we can estimate changes in drought severity at the study site based on the projections of global circulation models (here I use 50 th percentile, medium emissions projections; see www.climatechangeinaustralia.gov.au). Current estimates for changes in precipitation in the study region by the year 2070, relative to the 1980-1990 baseline period, are for declines of 0%, 3.5%, 15%, and 15% for summer, autumn, winter and spring respectively, with changes in ETo of 6%, 10%, 14%, and 3%. Historically, mean annual AWB at Wagga Wagga ) is -720 mm (Fig. 7c). If we use the extremely simple approach of modifying the observed 1970-2009 data according to these projections, we obtain a 2070 estimate for annual AWB of -860 mm, a 19% increase. Under the current climate regime ) a year like 2006 (AWB = -1229 mm) has a return interval of 62 years, but by 2070, the return interval becomes 19 years. This suggests that, if a similar condition holds at Wyalong, the frequency of years in which significant mortality might occur in native plant communities could increase by around threefold, possibly resulting in persistent shifts in vegetation composition similar to those observed in this study. However, given the low mortality of the three study species in terrace habitats even under the conditions experienced during 2006, their long-term persistence at the study site, albeit in a possibly restricted manner, seems virtually certain.
A final line of evidence supports this prediction. Nyngan, NSW, which lies approximately 450 km north of Wyalong, is an approximate 2070 climate analogue for the study site. All three study species occur at, or near Nyngan (although A. aristiglumis is restricted to very mesic riverine habitats), and P. prolutum and L. panaetioides occur much further west in drier areas. As mentioned, however, it is exceedingly difficult to make accurate predictions of this kind, and to account for mitigating factors such as atmospheric CO 2 enrichment (Koch et al. 2004), population-level evolution for drought tolerance, competition (White et al. 2001) or many other potentially important factors (see Godfree et al. 2011) that are known to affect the response of plant species to drought.

Conclusions
Understanding the role of extreme climatic events as drivers of contemporary and future vegetation change is one of the greatest challenges that ecologists face today. Extreme events are difficult to study, and the responses of plants, species and communities to abiotic stress are contingent on a broad array of physiological, demographic and landscape-scale process that are often nonlinear in nature. In this paper I have provided evidence that extreme climatic events, and especially drought, have the capacity to rapidly alter the structure and composition of plant communities, with the magnitude of change roughly reflecting the statistical severity of the conditions. The majority of droughts that cause the highest rates of mortality among extant plant populations seem to be exceptionally rare events, occurring only on multi-decadal or century timescales, although quantifying the exact relationship between drought severity and plant mortality clearly needs further research.
The results of surveys conducted in a semi-arid grassland site in Australia suggest that changes in community structure and composition following drought reflect the processes of both mortality and post-drought recruitment and recovery, and that the demographic responses of species to extreme water stress are highly idiosyncratic. Nonetheless, drought performance of individual species does appear to be at least partly predictable based on the nature of their climatic envelopes. The data also suggest that, if drought is severe enough, topoedaphic refugia are crucial for the survival of a broad suite of species, not just those that favour mesic habitats. Finally, quantification of the statistical distribution of rainfall and atmospheric water balance in the semi-arid study region suggests that relatively modest changes in rainfall and evaporation could lead to large changes in the frequency and severity of extreme drought in coming centuries.
Much remains to be understood about the mechanisms and conditions under which extreme climatic events act as drivers of vegetation mortality, and the specific characteristics of populations, species and communities that predispose them to rapid abiotically-driven change. The overall objective of ecology is to develop theory that usefully predicts phenomena in nature, and working towards development of a theory that improves our understanding of the relationship between extreme climatic events and vegetation change is certain to be a fruitful area of ongoing ecological research.

Acknowledgments
I would like to thank Brendan Lepschi for assistance with data collection and acquisition of herbarium records.