Forest vulnerability to hotter drought: observations from the Southwest USA

Recent events in the mountainous landscapes of the Southwest USA (Arizona, New Mexico, and southern portions of Utah and Colorado) highlight the effects of hotter drought on forest stress, insect outbreaks, wildfire, and forest die-offs. Since ca. 2000 the Southwest has been subject to large increases in tree mortality in response to the combination of protracted drought and early 21st century warmth (Breshears et al. 2005, Williams et al. 2010, 2013, Allen 2015), as part of a broader sub-continental pattern of hotter drought that has been driving historically unprecedented insect outbreaks (Raffa et al. 2008) and wildfire (Westerling et al. 2014) in western North America. While past severe droughts are documented to have caused substantial tree mortality in the Southwest during the 1950s (Allen and Breshears 1998) and likely even as far back as the 1580s “megadrought” (Swetnam and Betancourt 1998), the recent hotter “global-change-type” drought (Breshears et al. 2005) has caused particularly pronounced tree mortality. High levels of tree mortality have been observed across broad elevational and landscape gradients, from the mesic spruce-fir (Picea-Abies) forests above 3500 m elevation down through mixed-species forests of Douglas-fir (Pseudotsuga), aspen (Populus) and pine (Pinus) to the lowest woodlands of juniper (Juniperus) at 1500 m (Gitlin et al. 2006), with substantial dieback and mortality of still lower-elevation shrubs and grasses (C. D. Allen, unpublished data), and even riparian trees along some ephemeral watercourses during the peak drought stress periods of 2000–2004 and 2011–2013 (C. D. Allen, personal observations).

The Southwest USA has been a productive setting for a diverse range of tree mortality research that is providing insights into the linked roles of drought and heat stress in driving Southwest forest productivity and health, physiological thresholds of tree mortality, and forest disturbance processes (e.g., Breshears et al. 2005, 2009, Allen 2007, Adams et al. 2009, 2015, McDowell et al. 2011, Clifford et al. 2013, Garrity et al. 2013, Sevanto et al. 2014, Anderegg et al. 2015b, Limousin et al. 2015). Notable recent work by Williams et al. (2013) derived a forest drought-stress index (FDSI) for the Southwest region using a comprehensive tree-ring growth dataset representing AD 1000–2007, determined by both warm-season temperature and cold-season precipitation. Substantial warming over the past 25 years has significantly amplified regional forest drought stress, likely to a large degree by increasing atmospheric vapor pressure deficits during the growing season months. Strong correspondence exists between FDSI and forest productivity, tree mortality, bark-beetle infestations, and wildfire in the Southwest (Williams et al. 2013), illustrating the powerful interactions among climate, land use history (livestock grazing, fire suppression), disturbance processes, and forest dynamics in this region (Allen 2015). After nearly a century of fire suppression, a wet period highly favorable to tree growth occurred from ca. 1978–1995 that supported maximal development of dense high-biomass forests in the Southwest, followed by the onset of warmer drought conditions that have driven historically extreme levels of drought-induced tree mortality, bark beetle outbreaks, and high-severity fires (Williams et al. 2010, 2014, Allen 2015). Tree rings document 2002 as the worst year for regional tree growth since at least AD 1000, due to hotter drought conditions (Williams et al. 2013). If regional temperatures increase as projected by climate models, the mean forest drought stress by the 2050s will exceed that of the most severe droughts in the past 1,000 years (Williams et al. 2013; cf. Cook et al. 2015), suggesting high vulnerability of current forests to extreme levels of tree mortality within just a few decades.

The observed emergence of extensive, ongoing, and diverse episodes of tree mortality and forest die-offs related to hotter droughts in the Southwest USA may be an early regional harbinger of broader forest vulnerability to hotter drought worldwide. Few scientists explicitly predicted the rapidity and broad extent of the forest changes that have occurred in the region over the past two decades. Indeed, note that if the only time window available with observations on Southwest forest growth and health was the favorable wetter years of the 1980s (Williams et al. 2013), the conclusion could be “what a great place to be a tree”—whereas today the continued existence of old-growth trees and historical forest ecosystems appears increasingly at risk in this region. Whether this region is particularly sensitive to tree mortality, given its climatology and forest history, could be debated. Nonetheless, rapid recent changes in Southwest USA forests provide cautionary insights about the vulnerability of other wooded ecosystems globally, illustrating that tree mortality can escalate rapidly beyond historical levels once critical drought and heat stress thresholds are exceeded. Overall, a global perspective illustrates that similarly dramatic changes in forest growth, productivity, and mortality can (and do) occur abruptly and pervasively elsewhere in response to changes in climate drivers, as exemplified by recent tree mortality events linked to combinations of drought and heat events in all major forest biomes around the world (Fig. 2; Allen et al. 2010).

The role of open access publication and emerging technologies in facilitating recent progress in tree mortality research

Research on forest vulnerability to tree mortality from drought and heat stress has grown rapidly in recent years, aided in part by open access publication and emerging technologies; highlighting such advances is part of the general focus of this set of ESA Centennial articles. Widespread e-publication and electronic communication technologies have: (1) fostered the rapid development of regional, national, and global networks and collaborations of researchers in this topic area (e.g., DIRENet (http://www7.nau.edu/mpcer/direnet/), Allen et al. 2010, Choat et al. 2012, Büntgen et al. 2013, Zeppel et al. 2014, Doughty et al. 2015, Frank et al. 2015a, b, Hartmann et al. 2015, Sitch et al. 2015); (2) increased communication through new media (e.g., the interactive website corresponding with the 2014 release of the US National Climate Assessment http://nca2014.globalchange.gov/); (3) supported electronic archival publication of datasets, enabling subsequent extension and re-analysis of datasets (e.g., Breshears et al. 2009, McDowell 2011); and (4) led to increasing numbers of syntheses from widespread research activities and networks, ranging from tree-growth data from the International Tree-Ring Data Bank (ITBRD, http://www.ncdc.noaa.gov/data-access/paleoclimatology-data/datasets/tree-ring) that enable broad-scale analyses of tree growth responses to climate (e.g., Williams et al. 2013, van der Sleen et al. 2014, Martin-Benito and Pederson 2015) to large data syntheses from permanent forest plot demography and growth networks that address fundamental questions of relative forest growth versus mortality in an increasingly greenhouse world (Carnicer et al. 2011, Luo and Chen 2013, Brienen et al. 2015).

Emerging and improving technologies also are supporting recent progress in climate-induced tree mortality research. Many of the tools currently used to assess how trees die are decades old but still are being continuously refined, including measurements of growth via dendrometers, photosynthesis and respiration via infra-red gas analyzers, measurements of plant water potential with pressure chambers, and water use via sapflow. New techniques also are making major contributions to understanding tree mortality, such as distributed sensors that can evaluate how soil temperature changes with changes in canopy cover (Royer et al. 2012), phloem function sensors (Sevanto et al. 2014), on-line isotopic measurements (Hartmann et al. 2013), highly controlled chamber systems for experimental manipulations of both [CO₂] (e.g., Quirk et al. 2013, Duan et al. 2014), and water availability (e.g., Limousin et al. 2012), and revolutionary “omics” advances that enable understanding the interplay between physiological responses and regulatory control during stress and mortality (reviewed in McDowell et al. [2011] and Thomas [2013]). Various emerging technologies have promise to provide breakthrough understanding of mortality processes, including improvements in our understanding of the roles of lipids as a storage reserve in trees (Hoch 2015) and micro-analysis of tree anatomy and physiological dynamics via imaging with nuclear magnetic resonance (NMR; Defraeye et al. 2014) and positron emission topography (PET; de Schepper et al. 2013). Observational capabilities of tree mortality are rapidly increasing as exemplified by globally-widespread forest inventory monitoring networks (e.g., Lewis et al. 2009, Anderson-Teixeira et al. 2015, Brienen et al. 2015, Malhi et al. 2015) and advances in airborne and satellite-based remote sensing techniques (e.g., Garrity et al. 2013, Hansen et al. 2013, Meddens and Hicke 2014, Asner 2015, McDowell et al. 2015). Ultimately, much of this work serves not only to provide better fundamental understanding, but importantly, supports the predictive capability of next-generation models that are being developed to better simulate tree mortality (e.g., McDowell et al. 2013, Fisher et al. 2015).

Observed and projected climate changes of relevance to tree mortality from drought and heat

Observed and projected climate changes could indicate lesser (Table 1, “Climate” category) or greater (Table 2, “Climate” category) vulnerability of forests to mortality from hotter drought. Implying lesser vulnerability (Table 1, CL), arguably the greatest benefits of global climate change to forest growth and resilience are rising atmospheric CO₂ concentrations, and increases in growing season length in high-latitude and other cold regions from rising temperatures (IPCC 2014, Xia et al. 2015). Rising [CO₂] benefits plants by increasing the essential C substrate available for photosynthesis, and simultaneously increases the ratio of CO₂ uptake to water lost (water-use efficiency; Ainsworth and Rogers 2007, Keenan et al. 2013). Additionally, increases in growing season length in high-latitude and other cold regions from rising temperatures could result in greater overall plant growth. Another factor suggesting lesser vulnerability is that increased evaporation of ocean water into a warmer atmosphere should increase average atmospheric humidity and increase rainfall in some regions, partially buffering rising vapor pressure deficit (VPD) due to rising temperature (Held and Soden 2006), and supporting greater tree growth and resilience in water-limited regions (e.g., Liu et al. 2015a). Indeed, despite warming temperatures in recent decades, pan evaporation measurements indicate a global trend of declining atmospheric evaporative demand, largely attributed to declines in near-surface wind speeds (McVicar et al. 2012); such a trend suggests potential amelioration of forest drought stress from projected warming-driven increases in atmospheric moisture demand. Also, precipitation has been rising in some regions and globally overall (Chou et al. 2013, IPCC 2013), with continued increases projected for some regions of the world; this could promote larger on-site water stores for trees to survive droughts when they do occur. Increases in the proportion of precipitation from large events can somewhat compensate for decline in total precipitation amounts (Dorman et al. 2015b). Warming has markedly increased growing season length across extensive temperate and boreal regions (Richardson et al. 2013), promoting additional growth in many temperature-limited forests (Keenan et al. 2014, Xia et al. 2015). As a consequence of these atmospheric changes, overall forest growth and carbon storage in many locations has increased in recent decades (Ballantyne et al. 2012, Graven et al. 2013), as further documented below.

In contrast, other climatic studies (Table 2, CL) indicate greater levels of forest vulnerability to projected hotter drought conditions. An increase in drought frequency and duration is predicted for much of the globe as climate change progresses (IPCC 2013). There is both observational and model evidence of the ongoing global emergence of historically unprecedented large and rapid increases in global temperature this century (Diffenbaugh and Scherer 2011, IPCC 2013). Additionally, the area impacted by drought is both observed and projected to increase globally (Dai 2013). Even if the increased heating from global warming does not directly cause increases in drought frequency, hotter conditions will result in droughts intensifying more quickly once they do occur and generally being more severe—the essence of “hotter drought” (Trenberth et al. 2014). Projected warming also is anticipated to reduce tropospheric relative humidity in the tropics and subtropics, in conjunction with a widening of the subsiding branches of the Hadley Cell, resulting in increased frequency of dry events in many geographic locations worldwide (Lau and Kim 2015)—observations show these predicted changes already emerging (Marvel and Bonfils 2013). Additionally, there is evidence of increasing drought severity caused by temperature rise and associated VPD in southern Europe and elsewhere (Vicente-Serrano et al. 2014). This is critical because nonlinear increases in atmospheric moisture demand (VPD) associated with hotter temperatures during drought are a key driver of forest physiological and ecological vulnerabilities to drought stress and increased mortality risk (e.g., Breshears et al. 2013, Eamus et al. 2013, Williams et al. 2013, Hart et al. 2014). When the effects of warmer temperature and greater VPD during drought are isolated from one another, model predictions highlight that it is the increased atmospheric moisture demand component that is most important in amplifying tree drought stress that could drive associated mortality (Eamus et al. 2013).

On longer time scales, the earth system is now moving into an altered regime of multi-decadal rates of climate warming that are unprecedented over at least the past 1,000 years (Smith et al. 2015), and perhaps over the past 65 million years (Diffenbaugh and Field 2013). Further, models project increasingly extreme precipitation (drought) and temperature (heat wave) events (IPCC 2012, 2013, Cai et al. 2014, 2015), including more frequent swings between opposite precipitation extremes from one year to the next (Cai et al. 2015), with an increased range between wet and dry season precipitation already observed (Chou et al. 2013). Recent extension of dryness-controlled areas of limited terrestrial carbon sequestration, driven by warming, already may be triggering a positive feedback that is further accelerating global warming by suppressing vegetation productivity in these areas (Wei et al. 2014). Finally, changes in tree cover can cause broad-scale ecoclimatic teleconnections, whereby vegetation change in one area affects not only regional climate but also climate and subsequently vegetation in another region or even another continent (Swann et al. 2012), including vegetation change associated with widespread forest die-off.

Observed spatial patterns of tree growth, forest stress and productivity, and drought- and heat-induced tree mortality and forest die-off

Substantial research is being generated on spatial patterns of tree growth, productivity, and hotter-drought-induced tree mortality and forest die-off using diverse approaches and providing perspectives that sometimes differ markedly between study regions, methodologies, and time periods—ranging from implications of lesser (Table 1, “Patterns” category) to greater vulnerability (Table 2, “Patterns” category). Regarding lesser vulnerability, the development of spatially extensive networks of permanent forest plots, where individual trees are repeatedly measured through time for ecological or forestry purposes, provide direct field measurements on forest growth and mortality patterns that are beginning to span sufficient time (decades) to determine significant trends in tropical, temperate, and boreal regions (Pan et al. 2013). In some portions of the world, forests are observed to be growing as well as, or faster than, ever measured before (e.g., McMahon et al. 2010, Hember et al. 2012, Fang et al. 2014,), particularly where water has not been limiting recently—consistent with predictions of strong CO₂ fertilization effects, and of the benefits of longer growing seasons in some colder regions (Pan et al. 2013, Keenan et al. 2014). For example, plot measurements have generally shown pan-tropical increases in forest growth and carbon storage in recent decades (Baker et al. 2004, Lewis 2006, Lewis et al. 2009).

In contrast, (Table 2, PN) findings of declines in forest growth and productivity are emerging in some regions due to episodic or growing drought and heat stress, particularly in historically hot or dry regions (Carnicer et al. 2011, Dorman et al. 2013), but also sometimes in wet or cool areas such as the Amazon (Brienen et al. 2015) and boreal forests (Chen and Luo 2015). Growth of most natural forests is water limited in various ways (Gedalof and Berg 2010, Jenerette et al. 2012, Bernacchi and VanLoocke 2014). Recent work in the Amazon (Malhi et al. 2015) highlights that to accurately assess forest biomass cycling relationships with climate it is necessary to address the significant spatial and temporal variability in poorly understood processes such as carbon-use efficiency, allocation of net primary productivity (NPP), and biomass turnover times (i.e., mortality rates, rooting depths, and phenological patterns and drivers). From forests worldwide there is increasing evidence that bigger, taller trees are most vulnerable to drought stress and mortality in a warming world (Phillips et al. 2010, McDowell and Allen 2015, McIntyre et al. 2015).

Shifting focus to tree-ring data, aggregated networks allow geographically extensive syntheses of annually-resolved tree growth-climate relationships that can extend back centuries to millennia. These dendrochronological analyses confirm variability in regional growth trends and climate drivers (Gedalof and Berg 2010, Peñuelas et al. 2011, Silva and Anand 2013, Vicente-Serrano et al. 2014), ranging from observations implying lesser vulnerability (Table 1, PN) with historically-unprecedented surges in growth found in some boreal (Juday et al. 2015) and high-elevation temperate (Salzer et al. 2009) forests; to nearly stable tree growth in diverse tropical forest sites (van der Sleen et al. 2014); to observations suggesting greater vulnerability (Table 2, PN) with significant declines in tree growth attributed to greater drought/heat stress in many dry regions, including the Southwest USA (Williams et al. 2010, 2013), Mediterranean Europe (Tognetti et al. 2000) and North Africa (Touchan et al. 2011a), other parts of Europe (Lévesque et al. 2014), interior Asia (Liu et al. 2013), and boreal Canada (Girardin et al. 2014) and Alaska (Juday et al. 2015, Walker et al. 2015). A subset of global tree-ring chronologies, mostly boreal, shows a recent divergence of tree-growth responses to climate in the 20th and 21st centuries from previous historical patterns, due to a complex of changing climate/hydrological/soil/physiological drivers (D'Arrigo et al. 2008, Juday et al. 2015); this “divergence problem” further emphasizes the risk of sweeping global over-generalizations from tree-ring growth patterns. Studies of carbon isotopes from within tree-rings also provide historical evidence regarding levels of tree water stress (e.g., McDowell et al. 2010) and widespread increases of intrinsic water-use efficiency without corresponding increased tree growth (Peñuelas et al. 2011, Silva and Anand 2013). Also implying greater vulnerability (Table 2, PN) are recent dendroclimatic studies that have increasingly linked warmer temperatures to poor tree growth and higher levels of forest drought stress (e.g., Williams et al. 2010, 2013, Girardin et al. 2014, Juday et al. 2015, Walker et al. 2015), with the most recent hot droughts appearing as the most severe relative to tree growth in some tree-ring records extending back for at least 800 years in North Africa (Touchan et al. 2011a), 1000 years in the Southwest USA (Touchan et al. 2011b, Williams et al. 2013), and 1200 years in California (Griffin and Anchukaitis 2014).

The development of networks of biogeochemistry flux tower sites (e.g., FLUXNET) and Free-Air CO₂ Enrichment (FACE) experiments have allowed ever-more detailed and broader-scale assessments of forest stand-scale photosynthesis, respiration, gross primary productivity, water-use efficiency, etc. in response to temperature and precipitation variability (e.g., Tang et al. 2014) in a global-change world of elevated CO₂ concentrations. Recent analyses highlight complicated and sometimes conflicting forest vegetation responses, including trends of alternating increasing and decreasing growth (Gatti et al. 2014), evidence for and against strong and persistent CO₂ fertilization effects (De Kauwe et al. 2014, Zaehle et al. 2014), and changes in water-use efficiency (Tang et al. 2014).

Rapid advancements in remote-sensing techniques (often in concert with advances in ground observations) provide increasingly high-resolution data on spatial and temporal patterns of global temperature and precipitation, drought severity (atmospheric moisture demand, soil moisture and associated plant water availability), water-use efficiency (Tang et al. 2014), forest growth and productivity (Hilker et al. 2014), and forest disturbance patterns (Espírito-Santo et al. 2014). Similar to patterns from ground observations, remote-sensing analyses show diverse responses to recent climate variability and change. Implying lesser forest vulnerability (Table 1, PN), many studies show widespread “greening” and biomass accumulation (Jong et al. 2012, Pan et al. 2013), from uncut tropical moist forests (Liu et al. 2015a) to semiarid savanna ecosystems (Buitenwerf et al. 2012, Fensholt et al. 2012, Liu et al. 2015a), consistent with strong CO₂ fertilization paradigms, and perhaps in part through increases in growing season length (Keenan 2015, Xia et al. 2015). Remote-sensing (and confirming ground observations) also document substantial invasions, expansions, and densifications of woody vegetation in many regions, consistent with several hypotheses regarding the potential roles of: fire suppression (Andela et al. 2013); vegetation recovery since historical land clearing in regions from Mediterranean Europe (Lloret et al. 2012) and the eastern USA (Nowacki and Abrams 2014) to Russia and China (Liu et al. 2015a); and atmospheric drivers of greening in semi-arid landscapes from increased water-use efficiency with greater CO₂ concentrations (Buitenwerf et al. 2012) and wetter periods in some regions (Liu et al. 2015a).

Yet implying greater vulnerability (Table 2, PN), other remote-sensing work suggests CO₂ fertilization effects are starting to be limited by increasing drought and heat stresses, ranging from forests in boreal (Beck and Goetz 2011) and temperate (Ciais et al. 2005, Potter et al. 2012) regions to tropical forests in the Congo (Zhou et al. 2014a) and Amazon (Hilker et al. 2014) basins. More broadly there are globally widespread indications of slowing or declining forest growth in response to episodic drought and heat events (Zhao and Running 2010) as well as from chronic rises in heat-related drought stress with extensive zones of reduced NPP (Yi et al. 2014).

Finally, also implying greater vulnerability (Table 2, PN) growing interest in drought- and heat-related tree mortality issues has accompanied an overall increase in documentation of both increasing background tree mortality and forest die-offs in many regions globally (Fig. 2), with methods ranging from direct ground observations to synoptic remote-sensing (Allen et al. 2010, IPCC 2014). Studies from forest biomes in many areas show increased background tree mortality rates that have been linked to: (1) warmer temperatures that increase plant water stress (van Mantgem et al. 2009, Carnicer et al. 2011, Peng et al. 2011); (2) warmer temperatures that can amplify mortality from biotic agents (Raffa et al. 2008, Logan et al. 2010, Das et al. 2013, Anderegg et al. 2015a); and (3) increased competition resulting from CO₂-enhanced tree growth rates (Zhang et al. 2015, Doughty et al. 2015). Two emerging global patterns are noteworthy: (1) the most significant forest die-off events are associated with hotter droughts (i.e., “global-change-type” droughts; Breshears et al. 2005, Allen et al. 2010, Matusick et al. 2013), where the warming is thought to drive higher levels of forest drought stress (consistent with Williams et al. 2013); and (2) larger trees seem to be at greater risk of mortality from hotter drought (Nepstad et al. 2007, Phillips et al. 2010, Zhou et al. 2013, McDowell and Allen 2015, McIntyre et al. 2015). Comprehensive documentation of global forest health and definitive determination of tree mortality trends currently is lacking due to the absence of an adequate global monitoring system (Allen et al. 2010); however, the technical capability has now been demonstrated (Vogelmann et al. 2009, Hansen et al. 2013, Mascaro et al. 2014, Meddens and Hicke 2014, Asner 2015, McDowell et al. 2015)—what is lacking yet is the global vision and will to support such a worldwide monitoring system (Mascaro et al. 2014).

Overall these wide-ranging studies document highly variable patterns and trends in tree growth, forest productivity, and tree mortality. Unprecedented strong forest growth in some portions of the world implies lesser vulnerability and is consistent with predictions of strong CO₂ fertilization effects and of the benefits of longer growing seasons in some colder regions, particularly where water has not been limiting recently. In contrast, declines in forest growth and productivity, and also widespread increases in background tree mortality and forest dieoff, are emerging in many regions due to growing drought and heat stress—often in historically hot dry regions where warming temperatures are increasing drought stress, but also occurring in cooler and wetter areas such as boreal forests and the Amazon. These latter studies consistently imply greater forest vulnerability to hotter drought. The current broad range of observed forest responses to climate variability in the Amazon, and the associated range of interpretations about the relative vulnerability of these forests to drought and heat stress, is emblematic of the ongoing challenges in sorting out seemingly conflicting observations using diverse methodologies.

Physiological, morphological, and genetic mechanisms and processes that affect tree vulnerability to drought- and heat-induced mortality

Research has proliferated recently on diverse physiological, morphological, and genetic mechanisms and processes that affect tree vulnerability to drought and heat mortality, yielding findings that could imply either lesser vulnerability (Table 1, “Mechanisms” category) or greater vulnerability (Table 2, “Mechanisms” category). Beginning with physiological studies that imply lesser vulnerability (Table 1, MC), there are many physiological mechanisms at the tissue and stand-scales that may partially compensate for rising temperatures, drought stress, and associated impacts on mortality. These mechanisms are generally not included in process models, which may lead to a conservative bias in regards to simulation of forest vulnerability to climate change. At the plant scale, acclimation and adaptation is known to occur for a wide range of physiological processes (Mencuccini 2003), such as: down-regulation of respiration (e.g., Atkin and Tjoelker 2003); upregulation of photosynthesis (e.g., Chaves et al. 2009); changes in carbon allocation to plant chemical defenses (Herms and Mattson 1992); maintenance of positive C balances by coordination of nonstructural carbohydrate carbon reserves to meet demand even when stress reduces photosynthate C supply (Klein et al. 2014b); shifts in embolism resistance (e.g., Kolb and Sperry 1999); and xylem refilling to reverse embolism (Klein et al. 2014a). Trees have many mechanisms to avoid drought stress (Klein et al. 2014a), ranging from leaf-scale to whole-tree level, including phenological adjustment of growth processes to avoid drought periods (Klein et al. 2013, Adams et al. 2015).

There are many ways in which hotter temperatures, particularly if they accompany drought, have negative biological effects that imply greater vulnerability of trees to mortality under hotter drought (Table 2, MC). One of the most important impacts of rising temperature is upon VPD and subsequent impacts on transpiration and photosynthesis (Eamus et al. 2013). VPD is nonlinearly dependent upon temperature such that a small rise in temperature causes a relatively larger rise in VPD (discussed in the context of tree mortality in Breshears et al. 2013). This induces greater water loss through the stomata and from the soil surface, increasing water stress. The risk of hydraulic failure, or the loss of water transport capacity, is thus enhanced by rising VPD (McDowell et al. 2008). To compensate for this greater risk of hydraulic failure, plants typically close their stomata to limit water loss; however, this comes at the cost of reduced photosynthesis (Martínez-Vilalta et al. 2002) and increased risk of carbon starvation (the process of failure to maintain metabolism and defend against biotic agent attacks; McDowell et al. 2011). Thus, rising temperature increases the risk of both hydraulic failure and carbon starvation. Additionally, warmer droughts increase the risk of mortality through a myriad of additional mechanisms that can accelerate the processes of hydraulic failure and carbon starvation, or even bypass these mechanisms. Respiration is nonlinearly (positively) related to temperature (Atkin and Tjoelker 2003), potentially resulting in greater consumption of energy stores at higher temperatures in the absence of adequate down-regulation (acclimation) of respiratory biochemistry. This should accelerate the carbon starvation process, and indeed it has been shown that respiration rates are higher, and death occurs more rapidly, in experimental warm-drought scenarios (e.g., Adams et al. 2009).

Regarding phenotypic plasticity and morphological adjustments, a variety of studies support lesser vulnerability to mortality (Table 1, MC). Trees compete for light and growing space during favorable climate conditions when water is not limiting, investing in above-ground leaf and stem tissues and building up high levels of live biomass, but under drought and heat stress individual trees and forest communities can adapt through diverse morphological responses at multiple time scales (Nicotra et al. 2010, Richter et al. 2012, Bussotti et al. 2015). Note that these morphological adjustments are closely interrelated with the physiological responses discussed above (Mencuccini 2003). Such phenotypically plastic morphological adjustments can include rapid short-term reductions in leaf area through early senescence or partial dieback of stems and leafy canopies (Rood et al. 2000, Mencuccini 2003, Ciais et al. 2005, Carnicer et al. 2011, Limousin et al. 2012, Filewod and Thomas 2014), which can be followed by post-dieback resprouting of woody tissues and leaves from stems or roots (Zeppel et al. 2014). Other phenotypically plastic responses include longer-term growth-mediated transformations of hydraulic architecture, wood density (Britez et al. 2014), and overall tree morphological architecture, emerging through altered relative growth investments in the size, number, and longevity of leaves, stems, roots, and mycorrhizal symbionts (Nicotra et al. 2010, Limousin et al. 2012, Zanetti et al. 2015). These morphological compensatory responses all lessen vulnerability to tree mortality, so for these compensatory responses the issue of lesser versus greater vulnerability largely hinges on whether these are sufficient to overcome the risk factors listed in other categories.

Regarding genetic variation, lesser vulnerability can also be implied (Table 1, MC) from studies documenting drought and heat resistance within tree species populations at multiple spatial scales (local, landscape, whole population), allowing survival of pre-adapted individuals in the short term, which also promotes natural selection of genotypes better adapted to survive warmer and drier future conditions (Gutschick and BassiriRad 2003, Alberto et al. 2013, Alfaro et al. 2014, Liepe 2014). Tree species have optimal climate zones, such that populations in the colder portions of their distributions are expected to have significant genetic acclimative capacity for warmer temperatures, whereas populations from warmer range-limit portions of the species' distribution are generally expected to be more vulnerable to stress from warming climate (Rehfeldt et al. 2002, 2004, 2014). Tree populations from warmer outlying localities can be better adapted genetically to handle drought conditions (Chen et al. 2010, Carsjens et al. 2014), although they may become subject to reduced genetic diversity at such “trailing edge” sites (Borovics and Mátyás 2013). Overall, higher levels of genetic diversity foster adaptive responses to climate change stresses (Jump et al. 2009a, Harter et al. 2015), including drought and heat stress (Mátyás et al. 2009, Sthultz et al. 2009). Nonetheless, greater vulnerability (Table 2, MC) is implied for those cases where large rapid climate changes exceed evolutionary tipping points (Botero et al. 2015).

Ecological factors and feedbacks at forest, landscape, and earth system scales that affect forest vulnerability to drought- and heat-induced tree mortality

There also are many ecological factors and feedbacks at forest, landscape, and earth system spatial scales that affect tree vulnerability to drought and heat mortality. Again, several of these imply lesser vulnerability of forests to hotter drought (Table 1, EF). Forests with higher levels of tree diversity have more species options to respond to both climate stresses and post-disturbance opportunities, resulting in relatively incremental adjustments at the forest scale (Fauset et al. 2012, Peters et al. 2015). In addition, more diverse forests are less likely to experience mortality events driven by outbreaks of biotic agents, whereas low diversity systems (e.g., monocultures) can be far more susceptible (Dyer and Letourneau 2013). In some ecosystems, climate change may actually hurt the insect pest agents that attack trees either through direct climate impacts on population growth or through increases in the abundance of predators which control those agents (Hicke et al. 2006, Raffa et al. 2008, Jamieson et al. 2012). At landscape scales the diversity of topographic (Adams et al. 2014), soil (Peterman and Waring 2014, Twidwell et al. 2014, Dorman et al. 2015a), and hydrological (Silvertown et al. 2015) settings and microsites provides relatively buffered refuge locales where trees have cooler-moister conditions to survive hot drought stresses (e.g., Allen and Breshears 1998), as well as favorable sites to recover more readily after mortality events. Such landscape diversity can allow “climate relict” populations of trees to persist as climate conditions become less favorable (Hampe and Jump 2011). Note, however, that “microrefugia” have limited buffering capacities (Hylander et al. 2015), implying greater vulnerability to increasingly severe hotter droughts (Table 2, EF).

Lesser vulnerability (Table 1, EF) is also suggested by a number of forest-scale “stabilizing processes” in response to tree mortality which can support the retention and recovery of original species or of new species, thereby buffering the system from a complete loss of forest (Lloret et al. 2012). Tree mortality is often strongly enhanced by competition for growth-limiting resources (water, light, nutrients; Ruiz-Benito et al. 2013), so forest-scale canopy defoliation, dieback, and elevated whole-tree mortality rates reduce competition between surviving trees, thereby reducing stress and limiting further mortality (Lloret et al. 2012), while surviving plants can facilitate new regeneration (Lloret and Granzow-de la Cerda 2013, Kane et al. 2015). Meanwhile, forests have strong effects on local climate by modulating evapotranspiration and albedo (Peng et al. 2014), with tropical and temperate forests cooling the local climate (Li et al. 2015); reductions in forest cover in these biomes could be expected to amplify climate warming. Changes in forest densities and canopy cover can have both positive and negative feedbacks on tree mortality processes by altering understory microclimates. For example, increased forest densities and canopy cover in European and eastern USA temperate forests moderate the impacts of macroclimatic warming on understory microclimates, fostering resilience of forest understory plants, including young trees (De Frenne et al. 2013)—whereas in tropical moist forests of the Amazon, reductions in tree canopy cover from drought mortality (as well as by timber harvest or fire) causes more open, drier understory conditions that can lead to greater drought stress and increasingly flammable fuel conditions (Brando et al. 2014). Alternatively, in some temperate coniferous forest types, moderate levels of drought-related tree mortality can sometimes reduce risks of high-severity stand-replacing fire by reducing ladder fuels, increasing the height-to-live-crown and canopy spacing, and decreasing crown bulk density, similar to mechanical forest thinning treatments (e.g., Hicke et al. 2012b).

For a general review of the effects of increasing atmospheric [CO₂] and changing climate on the dynamics of forest recovery after disturbances, including drought-induced tree mortality, see Anderson-Teixeira et al. (2013).

Also implying lesser vulnerability at broader spatial scales (Table 1, EF), tree populations in the past naturally have responded to environmental changes with adjustments in their geographic distributions (Corlett and Westcott 2013). With global warming the expectations are for range retractions through mortality at the warmer “trailing edge” margins of species distributions, with range extensions through migration anticipated at the colder “leading edge” (Hampe and Jump 2011). However, implying greater vulnerability (Table 2, EF), in some regions tree populations do not appear to be tracking recent climate changes fully or at all (Feeley et al. 2011, 2013, Zu et al. 2013, Fensham et al. 2014, Nowacki and Abrams 2014; but see Pederson et al. 2014a), and overall there are questions about whether natural tree migration rates will be fast enough to keep up with projected rates and magnitudes of climate change (Feeley et al. 2012, Corlett and Westcott 2013, IPCC 2014, Zhu et al. 2014), with habitat fragmentation a growing impediment to species migration and colonization (Haddad et al. 2015).

Implying lesser vulnerability at global scales (Table 1, EF), biome-scale resilience in water-use efficiency to interannual precipitation variability has been observed (Ponce-Campos et al. 2013), which can somewhat buffer hotter drought effects. There also are reasons to doubt the existence of global tipping points related to vegetation change and atmospheric dynamics (e.g., Brook et al. 2013). Further implying greater vulnerability (Table 2, EF), however, it is likely that prolonged droughts could exceed hydro-climate thresholds and trigger considerable vegetation mortality (Ponce-Campos et al. 2013). Additionally, recent work (Vicente-Serrano et al. 2013, 2014) finds that forest growth declines occur in humid biomes in response to shorter-duration drought stress relative to forests in more semi-arid and subhumid conditions where longer-duration drought drives growth declines; these biome-level differences likely reflect the relative drought vulnerability of the dominant tree species in current forest communities that have been subject to differing durations and severities of historical drought stress.

Drought and heat also have major direct and indirect amplifying effects on multiple other tree-killing disturbance processes, including fire (Pechony and Schindell 2010, Flannigan et al. 2013, Williams et al. 2014, Jolly et al. 2015), insect outbreaks (Raffa et al. 2008, Weed et al. 2013), and pathogens (Desprez-Loustau et al. 2006). These climate-related disturbances, sometimes also including other human disturbances such as logging or fire ignitions, can interact in “disturbance complexes” (McKenzie et al. 2008), often synergistically amplifying tree mortality (Allen 2007, van Mantgem et al. 2013, Brando et al. 2014). Individually, many of these forest disturbance processes involve nonlinear threshold responses to drought and heat drivers (Williams et al. 2014, Anderegg et al. 2015a) at multiple spatial scales, from local and regional (Allen 2007, Brando et al. 2014) to global (Hughes et al. 2013, IPCC 2014). In addition, cover loss of tropical moist forest can have significant feedback affects on local and regional climate (amplifying both drought and heat; Brando et al. 2014, Lawrence and Vandecar 2014), with emerging indications of inter-hemispheric teleconnections linking forest cover change and climate between North and South America (Swann et al. 2012). Indeed, there is significant concern about the potential existence of forest-related tipping points at global scales in response to projected climate changes (Lenton et al. 2008, Barnosky et al. 2012, Scheffer et al. 2012b, Hughes et al. 2013, IPCC 2014). In particular relative to forests, if growing drought and heat stresses reduce forest productivities and cause massive forest die-offs, there are concerns that forests, which currently sequester about 25% of the human atmospheric carbon emissions annually (Pan et al. 2013), could switch to become a net source of carbon back to the atmosphere (Bonan 2008, Kurz et al. 2008, Phillips et al. 2009). However, overall ecosystem carbon dynamics also depend on post-die-off soil respiration responses that strongly affect net release of forest-sequestered carbon back to the atmosphere (e.g., Moore et al. 2013).

Broad-scale modeled projections of forest growth, productivity, and vulnerability to drought- and heat-induced tree mortality with climate change

Regarding broad-scale modeled projections of vegetation change, studies also vary regarding support for lesser or greater levels of vulnerability of forests to hotter drought. Implying lesser vulnerability (Table 1, PJ), process models of vegetation responses to projected climate (e.g., DGVMs) generally project increasing future forest growth and resilient forest carbon stocks (e.g., Huntingford et al. 2013, Sitch et al. 2015). Most of the models include substantial positive effects of CO₂ fertilization and associated increased water-use efficiency, reflecting core physiological knowledge (e.g., Ainsworth and Rogers 2007, Arora et al. 2013, Keenan et al. 2013,). For example, recent modeling work “… indicates a much lower risk of Amazon forest dieback under CO₂-induced climate change if CO₂ fertilization effects are as large as suggested by current models” (Cox et al. 2013). Also, note that many of the compensatory physiological mechanisms reviewed above generally are not included in these process models of forest responses to climate change, which could lead to over-estimating simulated forest vulnerability to projected drought and heat stresses (e.g., Wythers et al. 2013).

In contrast, without strong CO₂ fertilization processes, models would tend to show vegetation “browning” sooner rather than “greening” in response to future warming, implying greater vulnerability of forests (Table 2, PJ). There are many questions about the actual strength and duration of CO₂ fertilization effects in varying “real world” situations (e.g., Peñuelas et al. 2011). Additionally, realistic projections of future tree mortality response to anticipated climate changes likely are greatly limited currently because few DGVMs or earth system models mechanistically represent physiological tree mortality processes (McDowell et al. 2011; cf. Betts et al. 2015). Further, other important tree-killing disturbance processes (fire, insect outbreaks, diseases) currently are missing or poorly represented in DGVMs (Lindner et al. 2014, Sitch et al. 2015). Each of these forest disturbance processes involves nonlinear threshold responses to drought and heat drivers (e.g., Williams et al. 2014, Anderegg et al. 2015a) that are difficult to realistically represent individually in process models, much less collectively, despite their importance as interactive disturbance complexes in real-world forest dynamics (Allen 2007, McKenzie et al. 2008). The incomplete and uneven inclusion of realistic mortality processes likely is one major reason that tests of diverse vegetation models result in poor performance when predicting change in vegetation carbon storage from elevated [CO₂] versus FACE experimental data (De Kauwe et al. 2014, Zaehle et al. 2014). Improving the representation of mechanistically realistic tree mortality processes in DGVMs increasingly is acknowledged as an important strategy to more accurately predict the tree mortality rates (i.e., “carbon turnover times” [Brienen et al. 2015] or “woody biomass residence times” [Galbraith et al. 2013, Malhi et al. 2015]) needed to better project future changes in ecosystem biomass (e.g., Sitch et al. 2015); however, pursuing greater realism of tree mortality processes drives associated increases in model complexity and more challenging “requirements for model specification using data that are difficult to acquire” (Joetzjer et al. 2014). As models incorporate more realistic mortality functions, some are showing greater vulnerability of forests to mortality from projected future hotter droughts (e.g., Jiang et al. 2013, Tague et al. 2013, Vicente-Serrano et al. 2015). Another significant limitation of most current DGVMs is inadequate representation of the effects of extreme climatic events on vegetation (Zimmermann et al. 2009, Kitzberger 2013, Reyer et al. 2013, Zhang and Cai 2013, Bahn et al. 2014, Niu et al. 2014, Orsenigo et al. 2014), particularly including extreme droughts and heat waves which drive major pulses of forest die-off that can filter out particular and significant components of tree populations and forest species compositions via rapid mortality.

Another modeling challenge may be inherent when using localized experimental results (e.g., CO₂ enrichment or drought mortality experiments) to scale up from individual trees to earth system models, as there is evidence of “a general trend for the magnitude of the responses to decline with higher-order interactions, longer time periods and larger spatial scales” (Leuzinger et al. 2011); if so, DGVMs may over-estimate both positive and negative impacts of climate change. Meanwhile, the current absence or inadequate representation in DGVMs of many ecologically-fundamental forest disturbance processes that generally are amplified by warmer and drier conditions (e.g., drought-induced tree mortality, fire, insect attacks), suggests that current broad-scale model projections of forest vulnerability to hotter droughts may be too conservative.

In addition to process-based mechanistic models of climate-induced tree mortality, a variety of other modeling approaches instead are based upon empirical relationships between climate/environmental factors and tree mortality or forest biome transitions, here lumped as “empirical models” (Adams et al. 2013). Widely used examples are “climate-envelope” species-distribution models based upon observed geographic distributions of individual tree species (Jackson et al. 2009, Iverson et al. 2011, Rehfeldt et al. 2012, Iverson and McKenzie 2013) or forest biomes (Gonzalez et al. 2010); note Feeley (2015) addresses two major unresolved assumptions that underlie these climate envelope models. If solid experimental and observational data could be used to determine climate envelopes that reflect mortality in response to extreme events (rather than envelopes based on species presence as related to mean climate conditions), climate envelopes potentially could be powerful predictive tools. Other empirical models of tree mortality risk include: forest demography models (e.g., Wunder et al. 2008); climate-growth-mortality response models (Williams et al. 2010, Williams et al. 2013, Macalady and Bugmann 2014, Huang et al. 2015); various models of tree mortality as a function of climate ranging from Australia (Mitchell et al. 2014) to Arizona (Clifford et al. 2013) and California (Das et al. 2013); a hybrid empirical-process model of a climatic water deficit threshold for Populus tremuloides mortality in Colorado USA (Anderegg et al. 2015b); and a model based upon joint climate-competition interactions for eastern US forests (Clark et al. 2014).

Management actions relative to forest vulnerability to drought- and heat-induced tree mortality

In general, land management has large earth system effects, including significant feedback interactions with climate (IPCC 2014, Luyssaert et al. 2014). With growing recognition of climate change risks to forests, there has been increasing interest in the potential for management actions to reduce vulnerability of trees to mortality from drought and heat effects (Table 1, MG). At a broad level, a practical conceptual framework to manage forests in the face of climate change uncertainties includes supporting a multitude of flexible approaches utilizing incremental and reversible actions and an emphasis on adaptive learning (Millar et al. 2007). Three major categories of adaptation options include forest management adaptation actions, new approaches and tools for decision-making with stronger researcher-practitioner partnerships, and policy arrangements to support adaptation in forest management (Keenan 2015). Other recent overviews of adaptation options to address forest drought and heat stresses from climate change address similar themes (e.g., for Europe see Lindner et al. 2010, Kolström et al. 2011, Hlásny et al. 2014, Lindner et al. 2014). Five key management actions to address hotter drought stressors on forests (presented relative to European forests but potentially much more widely applicable) include: use resilient plant species; increase forest carbon storage; manage disturbance impacts; manage forests as renewable energy resources; and value and marketize forest ecosystem benefits and services to society (Fares et al. 2015). Another strategy is to start incrementally with historically proven management practices, and use adaptive management learning to gradually utilize more novel transformational practices as needed to accompany anticipated more extreme climate change progresses (Pinkard et al. 2014). Numerous studies support various historically-proven forest harvesting and thinning practices to improve the resilience of post-treatment tree mortality from drought and heat stresses by directly reducing resource competition and increasing tree growth, vigor, and defenses against pests (e.g., D'Amato et al. 2013, Giuggiola et al. 2013, Yaussy et al. 2013, Tarancón et al. 2014). Management can change species composition and genetics of tree populations to promote resistance to hotter droughts by selective cutting, planting (including assisted migration beyond historical ranges; Williams and Dumroese 2013), and breeding. To maintain valued ecosystem services (Bonan 2008), forest management can target retaining more water onsite to ameliorate forest vulnerability despite growing stress from warmer global-change-type droughts (Grant et al. 2013). There also are opportunities to shift traditional perspectives and work with the new biotic assemblages which are necessarily emerging in response to changing environmental conditions in the Anthropocene, “applying adaptive conservation to all human activities”, including forest management (Lugo 2015). Fitting local actions into a global perspective on global change risks, Scheffer et al. (2015) suggest “positive, action-oriented framing of a safe operating space for the world's iconic ecosystems” such as the Amazon rainforest, to muster societal support to manage local stressors to promote ecological resilience at local-to-global scales (cf. Steffen et al. 2015). Networking within a region also could buffer against impacts of forest die-off (Breshears et al. 2011), with effectiveness depending on the degree of patchiness (López-Hoffman et al. 2013). Collectively, then, many management actions have been identified that could contribute to reducing forest vulnerability to hotter drought.

The effectiveness of potential management responses, however, also depends on the relative expression of many other risk factors that drive greater tree vulnerability (Table 2, MC, EF). As considered throughout this article, the diverse impacts of hotter drought could overwhelm the effects of forest management actions. For example, one key management strategy, forest thinning, likely becomes insufficient to buffer trees against mortality when droughts become severe enough (Williams et al. 2013), just as it may be insufficient to prevent wildfire spread under hotter drought conditions (Tarancón et al. 2014). As another example, sustaining historical forests through management of the genetics of native tree populations depends on genetic variation being sufficiently large to buffer against hotter drought events, which generally is uncertain currently. Overall, today's forests inevitably will become more vulnerable to amplified tree mortality if climate warming proceeds to extreme enough levels to drive effects of the diverse risk factors associated with hotter drought (Table 2) to exceed the compensatory capacities of the various lesser vulnerability factors (Table 1), including the mitigating potential of management actions.