Tree mortality from drought, insects, and their interactions in a changing climate

Introduction

Forests cover c. 30% of the globe's land surface area, provide numerous ecosystem services to human societies, and play a central role in global biogeochemical cycles (Bonan, 2008). Yet the future of forest ecosystems given expected changes in climate and other environmental drivers is uncertain. Warming and drought, sometimes co-occurring with insect outbreaks, have been linked to tree mortality in many regions, and future changes in climate are expected to drive more extensive, severe, or frequent tree mortality events (Allen et al., 2010). Forest mortality can have manifold consequences for biodiversity, ecosystem function and services, and feedbacks to climate change through biophysical effects and loss of carbon sinks (Adams et al., 2010; Anderegg et al., 2013). The mechanisms that lead to forest mortality are complex and include plant physiological responses to climate, climate influences on insect pests and pathogens, and their interactions. Efforts to model climate-driven forest mortality, however, most often focus on tree physiological responses to drought (Tague et al., 2013; Anderegg et al., 2015) and, very rarely, on tree responses to insect attack (Dietze & Matthes, 2014), but no model exists that incorporates both drought and insects and their interaction.

Approaches to modeling drought-induced tree or forest mortality typically define drought from an ecosystem water-use perspective, using such measures as an extended period of above-average climatic water deficit, that is, unusually large or persistent excess of water demand by evaporation and transpiration relative to supply. A rapidly evolving plant physiology literature that examines tree responses to drought and how these can lead to tree mortality provides a basis to incorporate the effects of drought on trees in models. Research on tree responses focuses on failure of the hydraulic system to conduct water (Anderegg et al., 2012b; Nardini et al., 2013; Urli et al., 2013) and reductions in available carbohydrates (e.g. O'Brien et al., 2014).

Insect outbreaks are often driven by drought, which stresses host trees (Raffa et al., 2008), but also by many other factors. Temperature directly affects insects through developmental rates and survival that drive population success (Regniere & Bentz, 2007; Powell & Bentz, 2009; Weed et al., 2013). Characteristics of host trees, such as abundance, density, size, and physiology, and their spatial pattern across the landscape, are known to influence the capacity for insect population growth and spread (Raffa et al., 2008). Community associates (e.g. natural enemies and symbiotic associates) influence population outbreaks and subsequent tree mortality and are also affected by climate (Hajek, 1999; Régnière & Nealis, 2007; Raffa et al., 2008; Six, 2013). Clearly, the interaction of drought and insects may lead to increased insect populations, with disproportionate consequences for tree mortality that may not be accounted for by drought or insects alone.

While mechanistic and conceptual models of insect population dynamics (Dietze & Matthes, 2014; Powell & Bentz, 2014) and ecosystem carbon cycling in response to climate (Fisher et al., 2010; Powell et al., 2013; Tague et al., 2013) have been used to estimate forest mortality, a key question is: how much detail in drought-induced tree physiological responses as well as representation of insect outbreaks that respond to both climate and host (tree) condition is needed in models to predict mortality in a changing climate?

In this paper, we argue that an integrated approach that considers both drought and insects is needed to accurately predict forest dynamics in a changing climate. We first present a framework that provides conceptual models of drought, insects, and their interactions that can serve as guides for implementing a fully coupled insect–drought physiology model of tree mortality. We then briefly summarize the mechanisms by which plants respond physiologically to drought. We further illustrate the mechanisms by which two major categories of insects, bark beetles and defoliators, kill trees, and show the linkages between insects and drought. Our framework suggests a continuum of forest mortality drivers, from mortality mostly dominated by drought effects on tree physiology independent of insects, to mortality dominated by insect dynamics independent of drought. We highlight the measurements that might contribute to identifying where on this spectrum a given mortality event occurs, and identify events where both insects and plant drought response must be considered. Finally, we review the evidence for cross-biome patterns in the relative importance of insect- and drought-triggered mortality and highlight promising avenues for future research.

Partitioning the roles of drought, insects, and their interactions in recent tree mortality

The coupling between drought and insect outbreaks in many past events suggests that a simple approach for including the effects of insects might be based on drought metrics. However, this approach is often overly simplistic. We posit that it is critical to separate and understand the relative contribution of insects and drought tree responses when: (1) insect and tree physiology exhibit different responses to climate drivers (e.g. winter warming is likely to have a disproportionate effect on overwintering insect population dynamics (e.g. Bentz et al., 2010) relative to their host trees); (2) nonlinear thresholds can be passed where drought processes can favor initial insect population increments but thereafter no longer drive tree mortality (e.g. populations of aggressive bark beetles rely on stressed trees at low beetle densities, but at high beetle densities healthy trees are killed; Raffa et al., 2008); or (3) interactions between insects and tree physiology lead to nonadditive effects in tree mortality rates (e.g. moderate (but nonlethal) drought can make trees more vulnerable to dying from defoliation (often nonlethal alone) or more susceptible to subsequent lethal attack by wood borers, secondary bark beetles, and latent pathogens; Davidson et al., 1999; Wallin & Raffa, 2001; Muilenburg & Herms, 2012).

We show later that one or more of these three cases are met for multiple major mortality events affecting millions of hectares of forest in western North America (e.g. Pinus contorta and Pinus edulis). Partitioning of the relative contribution of insects and host drought is useful for separating causalities from correlations and to understand and predict these large-scale mortality events in a changing climate.

Although existing broad frameworks include drought, insects, and their interactions in driving tree mortality (e.g. Manion, 1981; McDowell et al., 2008), there is as yet no general method for attributing their relative importance in tree death. We present a general two-axis framework for attributing the relative roles of drought, insects, and their interactions. Our framework revolves around two central questions. How many trees would have died in a given mortality event with drought but without insects? Conversely, how many trees would have died in a given mortality event with insects but without drought? This gives rise to two hypothetical numbers of dead trees in a focal region. In practice, these quantities can be difficult to determine but can be estimated from sensitivities observed in field experiments (Gaylord et al., 2013; Netherer et al., 2015) or from empirical relationships derived from previous insect outbreaks or mortality events (Chapman et al., 2012; Creeden et al., 2014). Then, based on inventory plots, aerial surveys, or satellite remote sensing, the observed number of dead trees can be estimated. By comparing the relative sizes of the insect-free and drought-free hypothetical cases and the observed mortality rates, it should be possible to determine the relative influence of each stressor on mortality rates.

Several techniques and lines of evidence have been used to determine where mortality events fall along a continuum, from where the dominant driver of mortality was climate stress on the tree to where it was climate influence on insect populations. Postmortem assessments of trees that died using field plots or inventory networks provide the most basic pieces of information – were dead trees attacked by insects based on evidence such as bore holes or resin exudates and, if so, at what densities and by which insect species? Fig. 1 presents such information as collected by the US Forest Inventory and Analysis plot network. This network assigns a proximate mortality agent based on visible evidence (e.g. insect galleries, fire scars) (Fig. 1), although drought-kill can be difficult to determine visually and is thus more uncertain (Supporting Information Notes S1). System knowledge of the ecosystem from previous mortality events can also be informative. Are the insect species known to be primary agents that kill trees outright, or secondary agents that typically require stressed trees to be lethal? For example, several major die-offs were strongly associated with physiological impacts of drought stress on trees, such as widespread juniper, oak, and pine mortality in Texas in 2011, with relatively few signs of insect attack (Twidwell et al., 2014). In recent widespread trembling aspen (Populus tremuloides) mortality in the Rocky Mountains and boreal forest in Canada, all observed biotic agents for aspen are considered to be secondary (Marchetti et al., 2011) and up to 20% of dead ramets had no signs of biotic attack (Anderegg et al., 2012b), indicating a major role of drought. Finally, spatial patterns of mortality in relation to drought stress patterns or insect outbreak and dispersal patterns also provide an insight into the dominant drivers of mortality. For example, high mortality at lower elevations, south-facing slopes, and in regions of the highest drought stress all support a prominent role of drought stress driving mortality – although improved conditions for tree-killing insects and pathogens on these warmer sites might also affect these patterns. On the other end of the drought–insect continuum (Fig. 2), climate influences on insect populations may be the dominant driver in other mortality events. For example, in aggressive bark beetles, the combination of warm temperatures that directly benefit beetle population success, and a supply of susceptible, nutritionally optimal host trees is often necessary to shift insect populations across the endemic to epidemic threshold (Raffa et al., 2008). Warming has also allowed the mountain pine beetle's northward expansion into Canadian jack pine (Cullingham et al., 2011) and enhanced beetle survival in western high-elevation pine forests (Bentz et al., 2013), in the absence of any major drought effects (Fig. 2).

Drought and insect stresses may also interact and lead to compounding effects on mortality (Fig. 2). The timing of mortality relative to climate anomalies is a crucial piece of information. For example, in lodgepole pine (P. contorta), the correlation with drought extent and mortality illustrates an important role for drought stress (Chapman et al., 2012; Hart et al., 2013), but the mortality event continued long after the observed drought stress had abated (Raffa et al., 2008; Creeden et al., 2014). This indicates that the presence of higher insect populations prolonged the mortality event, and thus might be categorized as a situation with compounding stress. Experimental evidence can also reveal when and where drought and insects interact. For example, a manipulative drought experiment revealed that drought stress predisposed piñon pine (P. edulis) trees to attack by piñon ips beetles (Ips confusus) (Gaylord et al., 2013), highlighting that drought–insect interactions were critical in this recent widespread mortality event (Breshears et al., 2005; Meddens et al., 2015).

Thus, in recent tree mortality events in western North America, we found examples of drought-driven, insect-driven, and drought–insect interaction-driven tree mortality (Fig. 2). This brief survey illustrates the importance of understanding not just the role of plant physiological stress but also the wider context of insect population dynamics and tree physiology pertaining to insect attack when studying climate-induced tree mortality.

Drought and insect mortality pathways

Overview

Most vegetation models do not explicitly consider the effects of insects, but instead implicitly assume that the impact of insects on plant death is accounted for in the temporally and spatially averaged ‘background’ mortality rate (McDowell et al., 2011). Moving beyond this assumption will require both theoretical and empirical advances, ideally operating in tandem. To this end, Dietze & Matthes (2014) propose a theoretical framework for modeling drought–insect interactions in trees. The framework examines the effects of different insect functional groups via changes in leaf area, phloem flow, xylem flow, stem turnover, and root biomass (Dietze & Matthes, 2014). This useful framework could be extended by including a limited set of additional processes for different insect feeding guilds (Koricheva et al., 1998). First, in addition to direct effects of climate on tree hosts, it could incorporate a mechanistic description of insect population dynamics as a function of climate (direct effects of climate on insects; Fig. 3). Second, and most importantly, it could incorporate the interaction between insects and tree physiology and their consequences on mortality (indirect nonadditive effects; y-axis in Fig. 2). We propose that such an interaction may be modeled with a basic physiological framework of the physiology of plant defense, such as the growth-differentiation balance hypothesis (GDBH; see later) (Herms & Mattson, 1992).

We begin with a baseline pathway that attributes mortality solely to drought stress on tree physiology. As discussed earlier and previously (McDowell et al., 2011; Anderegg et al., 2012a; Sala et al., 2012), changes in plant hydraulics and carbon metabolism probably mediate the multitude of drought mortality pathways. In ecosystem models, whole-plant hydraulic conductivity (K_p), the hydraulic conductance of water within plant xylem vessels from the root surface to the stomata, has frequently been used to represent both reductions in productivity with drought and hydraulic failure (Martínez-Vilalta et al., 2002). This concept integrates the combined effects of water availability on biomass allocation to conducting tissue, transpiration, and inherent xylem hydraulic properties. Similarly, nonstructural carbohydrate (NSC) reserves may provide a time-integrated measure of the carbon status of a plant (Tague et al., 2013). K_p and NSC, however, are interdependent: changes in K_p will affect carbon assimilation, and NSCs have been implicated in xylem repair (Salleo et al., 2009; Secchi et al., 2011; Trifilò et al., 2014), as well as in overall plant hydraulic integrity (O'Brien et al., 2014). Future research should focus on the interaction between K_p and NSC to better understand their respective influence on each other and the subsequent specific thresholds that lead to mortality. In addition to capturing tree physiological status, these two variables allow incorporation of several important interconnections between trees and insect populations, particularly through secondary metabolites (Fig. 3a), although other interconnections may also be important.

Cross-biome patterns in insects’ roles in tree mortality

How the mechanisms articulated in Fig. 3 operate in different biomes is uncertain. Given the importance of insects as herbivores in tropical forests, a better understanding of drought–insect interactions in tropical forests is probably crucial if we are to represent global biogeochemical dynamics more realistically, because of the significant contribution of these forests to the global carbon cycle – but most recent literature on insect outbreaks has focused on temperate forests. How do the relative roles of insects and drought vary across ecosystems and across the globe? Cross-biome patterns of the relative roles of insects vs drought in causing tree mortality are hampered by available data in tropical forests, but some generalizations can be made across biomes. Compared with the insect outbreaks in temperate forests previously described, tropical outbreaks usually affect fewer trees, largely because tropical forests are more diverse in species composition and most herbivores are limited to one or a few related tree genera (Dyer et al., 2012). Thus, we hypothesize that the relative severity of climate-sensitive, insect-caused mortality should increase with latitude as tree species diversity declines and the number of host trees increases. Insects at high latitudes are also predicted to have relatively greater fitness as climate warms, because they are currently living below their thermal optima, compared with insects at low latitudes that currently live very close to their thermal optima and face greater extinction risks (Deutsch et al., 2008). Although tropical outbreaks are usually most severe in areas of highest host tree density (Dyer et al., 2012), they can still occur in diffusely distributed host species that comprise 1% or less of trees in a forest (Wong et al., 1990). Importantly, generalized outbreaks also do sometimes occur, with several different insect species simultaneously attacking a taxonomically diverse array of tree species (Van Bael et al., 2004; Dyer et al., 2012).

Although our understanding of outbreaks in species-rich tropical forests is less than in the temperate and boreal zones (Dyer et al., 2012), available information hints at some important contrasts between tropical and temperate forests. For example, although several species of tropical bark beetles and wood borers are known to undergo outbreaks (Nair, 2007; Dyer et al., 2012), most published information on tropical outbreaks focuses on defoliators. The dominance of publications on defoliators may be a consequence of defoliator outbreaks being more common or it may reflect a reporting bias as a result of easier detection of defoliator outbreaks.

Many outbreaks in tropical forests – at least of defoliators – appear to be triggered by drought. However, outbreaks often occur during the rainy season immediately following the drought, perhaps in response to postdrought leaf flushes or delayed postdrought recovery of the outbreaking herbivores’ enemies (predators, parasitoids, and pathogens; Coley, 1998; Van Bael et al., 2004). Tropical outbreaks may generally be shorter than temperate outbreaks (weeks or months rather than years; Wong et al., 1990; Van Bael et al., 2004), perhaps as a result of rapid top-down control by enemies (Van Bael et al., 2004). Tropical outbreaks might also, on average, be more spatially restricted than temperate outbreaks. For example, Van Bael et al. (2004) reported localized outbreaks along the Pacific Coast of Panama following a drought, but found no outbreaks among the same host species in moister interior forests. A few tropical outbreaks have been documented as long-lived and spatially extensive, particularly when host-tree densities are high. Nair (2007) summarized a century's worth of outbreaks of Hoplocerambyx spinicornis (a cerambycid beetle) in one of its host trees in India and found that the largest outbreak spanned > 6 yr and 500 000 ha, killing millions of trees.

Finally, the role of higher trophic levels as negative feedbacks that control insect outbreaks appears to be more important in tropical forests than in extratropical forests. Predicting the effects of climatic changes in these systems will thus depend on an improved understanding of climatic effects on trophic cascades. For example, if climatic changes disrupt the synchronization of top-down control of outbreaking insects by their enemies (Coley, 1998; Stireman et al., 2005), we might expect more frequent, extensive, generalized, and long-lasting outbreaks in the future.

Tools and approaches for moving forward

Multiple experimental, observational, and modeling approaches appear promising for attributing tree mortality to drought, insects, and their interactions, although each approach has its limitations. Factorial experiments that manipulate both drought and insects have good potential to improve understanding of tree drought response, insect population dynamics, and tree–insect interactions, but to date these have been rare and expensive. Rainfall exclusion experiments have been widely used to impose drought stress on forest ecosystems (Beier et al., 2012; Zeppel et al., 2014) and have, in some cases, induced mortality in concert with insect attacks (Gaylord et al., 2013; Netherer et al., 2015). Insecticide sprays or manipulated insect introductions can control for insect attack on certain trees, and factorial drought experiments with and without insect attack may be a useful tool to disentangle the relative roles and climatic underpinnings of tree stress and insect attack in mortality (Raffa & Berryman, 1983; Netherer et al., 2015).

Observational data that include both tree mortality and insect abundances or presence in dying trees, especially across different locations and ecosystem scales over longer time periods, will help to advance our understanding. A few data sets are available that cover landscape and broader scales over multiple years (Shaw et al., 2005; Meddens et al., 2012) (Fig. 1). Satellite remote sensing, particularly time series of imagery, can be especially useful when accuracy is evaluated against other observations, such as field data, and when attribution of tree mortality to a disturbance agent is identified (Hansen et al., 2013; McDowell et al., 2014). Quantification of insect abundance along with tree mortality is a critical data gap, and would be especially useful if collected in concert with large-scale forest inventory networks. Empirical studies of the causes of tree mortality that consider multiple factors, including climate, stand structure, soil characteristics, and insect populations, will be particularly valuable.

Mechanistic models exist that simulate the internal physiology of tree hydraulic and carbon status (McDowell et al., 2013; Powell et al., 2013) but will probably require substantial empirical calibration (Fisher et al., 2010) to account for uncertainty in meteorology and soil physical properties that affect plant water supply and therefore accurate representation of drought (Lichstein et al., 2014).We posit that better modeling of internal plant physiology, such as plant hydraulics, along with coupling to mechanistic insect dynamics models (Fig. 2), will improve higher-scale modeling of mortality resulting from drought, insects, and their interactions. For example, a hydrological model that incorporated a plant hydraulic threshold was able to predict spatial patterns in drought-driven mortality of P. tremuloides with 75% accuracy and was a big improvement on prediction from soil moisture, precipitation, or temperature alone (Anderegg et al., 2015).

Mechanistic models have also incorporated insect outbreaks in several studies in different ways (for a review, see Hicke et al., 2012). One way involves prescribing insect-caused tree mortality to study the effects on vegetation structure and function by bark beetles (Kurz et al., 2008; Edburg et al., 2011) and other feeding guilds (Hogg, 1999; Keith et al., 2012). Sensitivity studies are useful for identifying key processes missing from models, such as the inclusion of snags that have a large effect on heterotrophic respiration fluxes (Edburg et al., 2011). Studies of past events document the impacts of insect outbreaks on vegetation dynamics, including influences on carbon fluxes that may affect national policy (Kurz et al., 2008), although adequate information about the extent, severity, and duration of tree mortality is needed. Investigations have illustrated the advantages of detailed tree-based models run at stand scales (Pfeifer et al., 2011), as well as less detailed growth and yield models (Kurz et al., 2008) or global land surface models (Edburg et al., 2011) that permit large-scale assessments and/or linkages to Earth system models. Effects of future outbreaks can be assessed by prescribing the timing and severity of these events using scenarios, which is particularly effective with defoliator outbreaks given their cyclical nature (Dymond et al., 2010; Hennigar & MacLean, 2010). A disadvantage of the scenario approach is that any coupling with climate may be limited.

A second means of incorporating insect outbreaks in mechanistic vegetation models is by developing and including a prognostic insect outbreak model. Insect models to predict the probability, extent, and/or severity of tree mortality that incorporate climate variables to predict insect-caused tree mortality have used a mix of temperature and precipitation variables (Jewett et al., 2011; Preisler et al., 2012). More recent mechanistic insect models also include spatially explicit host tree densities (Powell & Bentz, 2014) and fungal associates (Addison et al., 2014). Inclusion of insect models in vegetation models allows us to assess the degree of ecological understanding (when results are evaluated against historical events) as well as to estimate the impacts of future climate change and management decisions. However, such studies are rare given the state of knowledge. A notable exception is the inclusion of European spruce bark beetle (Ips typographus) in LPJ-GUESS (Jönsson et al., 2012). Thus, more detailed coupling of tree and insect models and additional applications to case studies should be possible, at least in some ecosystems, in the near future (Fig. 2).

A key decision is when and how best to model insect population dynamics. The status quo approach for typical vegetation models is that insect-induced death is implicitly included in background mortality rates, but in reality that insect-caused mortality is not constant, but rather often aggregated in space and time and nonlinear. One approach relies on the dominance of temperature as a significant driver of insect life cycles, and information on temperature-dependent development times in outbreaking insect populations that are well studied (Powell & Bentz, 2009). These climate-driven insect population models can be coupled with ecosystem models (Jönsson et al., 2012). Further complexity can be added by considering diffusion or spread, where the probability of infestation varies as a function of proximity to other currently infested trees (Preisler et al., 2012; Powell & Bentz, 2014). Advantages of spatially modeling insect population dynamics include the capacity to incorporate density-dependent factors and the potential to better capture spatial and temporal variation in tree mortality that results from climate-driven insect population growth. Disadvantages include the need for additional parameters and model complexity, the lack of scientific understanding for many processes, and the possibility of introducing erratic and unpredictable model behavior as a result of greater model complexity.

Conclusion

Advancing our understanding of drought, biotic disturbance agents, and their interactions is critical to developing and applying models that predict future patterns of tree mortality in a changing world. Recent developments in tree mortality research have focused separately on the physiology of trees under stress and on insects as tree-killing agents, yet these processes are linked. Insects play many critical roles in physiology, demography, and disturbance of forest ecosystems, although they are often not explicitly included in ecosystem models used to make predictions of vegetation change or carbon cycling under climate scenarios. Furthermore, insect outbreaks are strongly influenced by climate (both warming and drought), and the role of drought varies by insect species. An important first step in predicting future mortality is to identify situations where considering both agents is necessary. We argue that a fully coupled approach is needed when: insect and tree physiologies exhibit different responses to climate drivers; nonlinear thresholds can be passed where drought processes may incite but not be needed to continue tree mortality: or interactions between insects and tree physiology lead to nonadditive effects on tree mortality rates. We present examples of different types of events throughout western North America, and argue that it will be essential to study mortality in other biomes, particularly the tropics, to improve mortality modeling under climate change. Our conceptual model that links internal tree physiology with insect feeding guilds offers a pathway for modeling drought-induced tree mortality, insect outbreaks, and their interaction in a fully coupled approach. The representation of insect diversity according to their functional interaction with trees (feeding guilds) offers a useful way to focus effort on understanding plant–insect interactions at large spatial and temporal scales, and provides a much-needed common point of interaction for cross-disciplinary science.