A trait-based approach to assessing resistance and resilience to wildfire in two iconic North American conifers

Abstract

1. Ongoing changes in fire regimes have the potential to drive widespread shifts in Earth's vegetation. Plant traits and vital rates provide insight into vulnerability to fire-driven vegetation shifts because they can be indicators of the ability of individuals to survive fire (resistance) and populations to persist (resilience) following fire.
2. In 15 study sites spanning climatic gradients in the southern Rocky Mountains, USA, we quantified variation in key traits and vital rates of two co-occurring, widely distributed conifers (Pinus ponderosa Douglas ex. P. Lawson & C. Lawson and Pseudotsuga menziesii (Mirb.) Franco). We used mixed-effects models to explain inter- and intraspecific variation in tree growth, survival, bark thickness and seed cone production, as a function of species, tree life stage (i.e. diameter, height and age), average climate, local competition and site conditions.
3. Pinus ponderosa was predicted to survive low-severity fire at a 23% earlier age than P. menziesii. Pinus ponderosa had thicker bark and more rapid juvenile height growth, traits conferring greater fire resistance. In contrast, P. menziesii was predicted to produce seed cones at a 28% earlier age than P. ponderosa. For both species, larger individuals were more likely to survive fire and to produce cones. For P. ponderosa, cone production increased where average actual evapotranspiration (AET) was higher and local competition was lower. More frequent cone production on productive sites with higher AET is an important and underappreciated mechanism that may help to explain greater resilience to fire in these areas.
4. Synthesis. Our analyses indicated that many plant traits and vital rates related to fire differed between Pinus ponderosa and Pseudotsuga menziesii, with trade-offs between investment in traits that promote individual defence to fire and those that promote recolonization of disturbed sites. Future changes in fire regimes will act as a filter throughout North American forests, with our findings helping to infer which individuals and populations of two iconic species are most vulnerable to future change and offering a framework for future inquiry in other forests facing an uncertain future.

1 INTRODUCTION

Fire has acted as a key selective force in terrestrial ecosystems for millions of years, shaping plant traits and the global distribution of vegetation (McLauchlan et al., 2020; Pausas & Bond, 2019). Whereas existing vegetation communities may be well-adapted to conditions of the past (Keeley, Pausas, Rundel, Bond, & Bradstock, 2011; Miller & Safford, 2020), anthropogenic climate change and other human activities have led to shifts in fire activity (Abatzoglou & Williams, 2016; Balch et al., 2017), with uncertain future outcomes for many woody plant species. Population-level demographic processes (i.e. vital rates) and individual-level plant traits offer crucial insight into the vulnerability of woody plant communities (Enright, Fontaine, Bowman, Bradstock, & Williams, 2015; Johnstone et al., 2016). Determining the extent to which vital rates and plant traits are adaptive under expected future conditions is an important goal of many mechanistic models in ecology, which have the potential to provide realistic and transferable projections of future vegetation dynamics (Cabral, Valente, & Hartig, 2017). However, our ability to project future changes in ecosystems is limited by an insufficient understanding of processes such as demography, dispersal and species responses to environmental variation (Urban et al., 2016); many of these processes are likely to vary both among and within species and have important implications for individual survival (i.e. resistance) and population persistence (i.e. resilience). Improved knowledge of factors that drive inter- and intraspecific variation in plant traits that influence vulnerability to fire will enhance the ability to understand and predict vegetation conversions.

Species life-history strategies in fire-prone ecosystems are thought to relate to selective pressures of the fire regime (Keeley et al., 2011); that is, the typical ranges of fire frequency, severity, seasonality and spatial pattern in a relatively uniform space and time (Baker, 2009; Sousa, 1984). For example, in tropical savannas, frequent fires limit forest cover through a demographic bottleneck (i.e. the ‘fire trap’; Hoffmann et al., 2012) that inhibits tree recruitment. Tree species that can resprout following topkill, limit injury due to thick bark, or exploit fire-free periods with rapid height growth have a higher likelihood of survival (i.e. resistance; Werner & Prior, 2013). Similarly, in conifer forests throughout western North America, communities with a history of more frequent, low-severity fire have traits such as thick bark and self-pruning that promote fire resistance (Stevens, Kling, Schwilk, Varner, & Kane, 2020). Allocation to bark and other fire-defensive traits is less advantageous when fires burn at higher severity. In plants that cannot resprout, severe fire can overwhelm bark protection and there is a low probability of individual survival (Pellegrini et al., 2017). Thus, fire regimes characterized by severe fires are often associated with species life-history strategies and plant traits that allow for rapid post-fire recovery (Loehle, 2000; Noble & Slatyer, 1980). These include frequent and abundant seed production, on-site propagule storage and long-distance seed dispersal that facilitate recolonization of disturbed areas (Pausas, Bradstock, Keith, & Keeley, 2004). Differential allocation in early life stages provides insight into interspecific differences in life history. For example, greater early-life investment in traits promoting resistance (e.g. bark tissue; Jackson, Adams, & Jackson, 1999) is expected in species inhabiting frequent-fire systems, likely resulting in greater time to reproductive maturity (Clark, 1991; Obeso, 2002).

Intraspecific variation is an increasingly recognized component of population- and community-level processes (Bolnick et al., 2011; Violle et al., 2012), and can relate to local adaptations to fire and climate (Laughlin, Fulé, Huffman, Crouse, & Laliberté, 2011). Individual-level trait expression in plants reflects inherent trade-offs among competing needs of growth, defence and reproduction (Lauder, Moran, & Hart, 2019; Redmond, Davis, Ferrenberg, & Wion, 2019). It follows that intraspecific variation (i.e. among and within populations) in vital rates and plant traits is influenced by factors such as individual life stage (e.g. tree diameter, height and age) and the resources available for tree physiological processes. Across a range of species and forest ecosystem types, larger trees are more likely than small individuals to survive fire (Harmon, 1984; Woolley, Shaw, Ganio, & Fitzgerald, 2012) and to produce abundant seeds (Greene & Johnson, 1994). Environmental gradients promote intraspecific variation by influencing tree growth and resource availability (Buechling, Martin, & Canham, 2017), and may also indirectly affect fire resistance traits such as bark thickness by influencing the fire regime (Laughlin et al., 2011). A relatively higher reproductive allocation has been associated with plant populations at dry sites within a species' range (Santos-del-Blanco, Bonser, Valladares, Chambel, & Climent, 2013; Sheth & Angert, 2018). Alternatively, prioritizing drought defence allocation over reproduction might maximize fitness in populations of longer-lived plant species occupying marginal sites (Lauder et al., 2019). Fine-scale resource availability, driven in part by local competition, influences tree growth (Buechling et al., 2017) and is related to individual-level reproductive output (Flathers, Kolb, Bradford, Waring, & Moser, 2016; Greene, Messier, Asselin, & Fortin, 2002; Nygren, Rissanen, Eerikäinen, Saksa, & Valkonen, 2017).

In the present study, we propose a framework for assessing vulnerability to fire-driven conversions that incorporates inter- and intraspecific variation in vital rates and plant traits (Figure 1), and we examine this framework through a case study of two iconic and widely distributed North American conifers—Pinus ponderosa Dougl. ex. P. Lawson & C. Lawson and Pseudotsuga menziesii (Mirb.) Franco (Figure 2). Pinus ponderosa and P. menziesii are useful study species because they co-occur across vast areas and exhibit important differences in physiological tolerances and life history. For example, P. menziesii has intermediate shade and fire tolerance, whereas P. ponderosa is highly shade-intolerant and fire-tolerant (Bigelow, North, & Salk, 2011; Buechling et al., 2017; Stevens et al., 2020). Within our study area in the southern Rocky Mountains, USA, historical fire regimes of P. ponderosa var. scopulorum and P. menziesii var. glauca forests ranged from high-frequency surface fires to mixed-severity regimes in which less frequent, higher-severity fires played key roles (Baker, Veblen, & Sherriff, 2007; McKinney, 2019). While there is broad spatial and temporal variation, historical fire frequency tends to decrease and severity to increase with the increasing abundance of P. menziesii (Swetnam & Baisan, 1996; Tepley & Veblen, 2015). Both species are non-serotinous obligate seeders with high interannual variability in seed cone production (i.e. mast seeding; Krugman & Jenkison, 2008; Stein & Owston, 2008), and post-fire establishment may continue for many years (Rodman, Veblen, Chapman, et al., 2020). We assessed the extent to which important plant traits and vital rates differed between species (reflecting different species life history), across gradients of resource availability (as influenced by broad-scale climate and effects of local competition) and throughout individual life stages. Specifically, we asked:

1. How do two co-occurring and widespread conifers differ in survival following recent wildfires and plant traits associated with fire resistance and resilience? We expected that P. menziesii would be less likely to survive fire than P. ponderosa, showing more limited allocation to traits promoting fire resistance (i.e. thinner bark, slower growth in early life), but having an earlier age of reproductive maturity.
2. How do tree life stage (i.e. tree diameter, tree age), climate and local competition influence intraspecific variation in traits promoting fire resistance and resilience? We expected that larger and older trees would be more likely to survive fire and produce cones. We also expected that variation in resource availability (expressed through average climatic conditions and local competition) would mediate these relationships by influencing physiological processes.

FIGURE 1

(a) Conceptual representation of the factors that influence variation in traits promoting fire resistance (individual tree survival) and resilience (post-fire reproduction and population persistence) in North American conifers. Over time, fire regimes have selected for individuals within a community that have traits that are most adaptive. For selected individuals, trait expression is influenced by species identity, life stage and resource availability, as well as resource allocation trade-offs. Individual trait expression then determines the traits that are present at the community-level. Under (b) past climatic conditions and selective pressures, traits were often aligned with the fire regime maintaining resilient communities within the safe operating space for survival and persistence (adapted from Johnstone et al., 2016). (c) Future changes in fire activity may cause increased misalignment between community traits and the fire regime, and vulnerability to vegetation conversion will be determined by the extent to which these two components overlap. Thus, a key question is: What factors influence individual trait expression in fire-prone forests, and under what conditions are these traits most adaptive?

FIGURE 2

The distributions of Pinus ponderosa and Pseudotsuga menziesii in (a, b) western North America (Little, 1971) and in (c) the study area in the southern Rocky Mountains, USA (Wilson, Lister, Riemann, & Griffith, 2013). We surveyed plant traits and vital rates related to fire resistance and resilience in 15 sites within and adjacent to recent (1988–2010) wildfires. In burned sites, we obtained records of individual tree survival following fire, post-fire juvenile height growth and juvenile bark thickness. In burned and unburned sites, we also collected data describing bark thickness of mature trees, seed cone production, site index and radial growth

2 MATERIALS AND METHODS

2.1 Study area and field sites

For P. ponderosa and P. menziesii, we synthesized data describing plant traits and vital rates in 15 study sites surrounding recent (1988–2010) wildfires in the southern Rocky Mountains, USA (Figure 2; Appendix S1). All fire events occurred in the April–September growing season (fire start dates range 29 April to 17 September; Eidenshink et al., 2007), and fires burned at a range of severities (i.e. 0%–70% high-severity area at fire-level; Rodman, Veblen, Chapman, et al., 2020; Figure S2 in Appendix S1). Tree age structures in unburned and low-severity areas indicated that many forest stands were multi-aged before fire, but the majority of live trees established in the 20th century (Figure S3 in Appendix S1). Climatic conditions vary substantially among study sites and span much of the climatic range of P. ponderosa and P. menziesii in the study area (January min: from −14.9 to −6.8°C; July max: from 18.1 to 30.2°C; annual precipitation: from 351 to 1,049 mm; PRISM Climate Group & Oregon State University, 2018).

Within each of the 15 fire perimeters, we established field plots to broadly characterize pre-fire forest structure and post-fire forest dynamics (hereafter ‘burned plots’). In these burned plots, which were located in stands spanning a range of fire severities (i.e. 0%–100% mortality at the stand-scale), we reconstructed tree survival and quantified characteristics of post-fire juveniles (i.e. individual stems establishing following the surveyed fire). Burned plots were restricted to areas that burned once between 1984 and the time of field surveys (i.e. 2016–2018). While tree die-off events and insect-caused tree mortality have been widespread in other forests throughout the southern Rocky Mountains (Breshears et al., 2005; Hart, Veblen, Schneider, & Molotch, 2017), these events have been less prevalent in P. ponderosa and P. menziesii-dominated forests (Chapman, Veblen, & Schoennagel, 2012). Therefore, we assumed that the studied fires were the primary disturbance events that recently affected these sites. We also established a separate set of field plots in a subset of eight fires (selected to span the geographic and climatic range of the original 15; though two of these sites solely included data on seed cone frequency) to sample a greater number of established (i.e. taller than 1.37 m) trees in areas that were unburned or burned at low severity in the respective fire events. In these unburned or low-severity plots, we quantified bark thickness of established trees, seed cone production and interannual variation in radial growth. Sampling protocols for each trait and vital rate followed a hierarchical structure with multiple individuals within each plot and multiple plots within each fire site. Additional site and sampling details are provided in Appendix S1.

2.2 Response variables—Plant traits and vital rates

As response variables in our analyses, we selected several readily measurable traits and vital rates hypothesized to influence resistance and resilience to wildfire (Table S1 in Appendix S1). As indicators of fire resistance, we quantified: (a) individual tree survival following wildfire events, (b) the relative bark thickness (i.e. mean bark thickness relative to total stem diameter) of post-fire juveniles and established trees and (c) height growth of post-fire juveniles. As indicators of resilience to wildfire, we quantified: (a) seed cone presence at the time of field sampling for individuals of each species (quantified by noting the presence of mature or developing cones or evidence of past cone production) and (b) the frequency of P. ponderosa seed cone years from 2006 to 2017 (reconstructed using cone abscission scars; the period in which multiple branches from all surveyed trees could be reliably dated). We focused specifically on longer-term P. ponderosa cone year frequency because the cone abscission scar method, used here to survey past cone production, has been primarily used in Pinus species (Redmond et al., 2016) and has not been previously used or validated for the genus Pseudotsuga. Here, we define cone years for P. ponderosa individuals as events in which abscission scars were found on at least two branches and at least 25 cones were produced. The two-branch threshold is meant to minimize the effects of dating errors, while the 25-cone threshold acts as a conservative indicator for potential self-replacement (Table S1 in Appendix S1). Survey methods for each trait and vital rate are described in Appendix S1.

2.3 Predictors of plant traits and vital rates—Species, life stage, climate and local competition

To quantify factors influencing inter- and intraspecific variation in resistance and resilience to wildfire, we analysed the selected plant traits and vital rates in relation to species (P. ponderosa vs. P. menziesii), individual tree life stage (i.e. juvenile height, mature tree diameter and height) and variables associated with resource availability (i.e. average climate and metrics of local competition; Table S3 in Appendix S1). To quantify life stage in post-fire juveniles, we recorded the total stem height (i.e. root collar to the tip of the terminal bud) and aged individuals by destructively sampling and dating multiple cross-sections near the root-shoot boundary (Telewski, 1993). For established trees, we recorded the diameter at breast height of all individuals (DBH; 1.37 m above ground level). Because bark thickness was measured at a lower point on the stem (20–40 cm above ground level), we also recorded stem diameter at this location for use in statistical models of bark thickness. We quantified the age of established trees using increment cores collected at 20–40 cm above ground level and corrected tree age estimates for the offset from pith and time to reach coring height (Duncan, 1989; Table S3 in Appendix S1). To summarize variation in climate among sites, we used average actual evapotranspiration (AET) and climatic water deficit (CWD) for the 1981–2010 period, developed at a 250-m spatial throughout the southern Rocky Mountains (Rodman, Veblen, Battaglia, et al., 2020). AET is the total evapotranspiration constrained by moisture availability, and higher values are thought to be an indicator of greater net primary productivity (Stephenson, 1998). CWD is the unmet evaporative demand of the atmosphere, and higher values are an indicator of greater plant moisture stress (Stephenson, 1998). For post-fire juvenile height growth, we quantified the effects of competition for resources using separate site-level metrics of: (a) post-fire density of resprouting angiosperm trees (primarily Quercus spp. and Populus tremuloides), (b) post-fire density of juvenile conifers, (c) post-fire per cent cover of graminoids and (d) surviving overstorey basal area in a stand. To quantify the effects of local competition on established trees, we developed a competition index (a continuous numeric predictor ranging 0–2 from low to high) using fixed- and variable-radius plots centred on each tree (Appendix S1).

2.4 Analytical methods—Statistical models

To analyse factors influencing each plant trait and vital rate, we used linear mixed models (LMMs) and generalized linear mixed models (GLMMs) in the lme4 package (Bates, Mächler, Bolker, & Walker, 2015) in r version 3.5.0 (R Core Team, 2018), where error structures were selected based on the form of each response variable (Table S4 in Appendix S1). Prior to model fitting, continuous numeric predictors were scaled using the series-wide mean and standard deviation (i.e. z-score transformed) to standardize model coefficients. We selected variables for inclusion in candidate statistical models by independently adding predictors from sets of correlated variables (r > 0.7; i.e. tree diameter vs. age, average AET vs. CWD) to determine the strongest predictor in each set based on the sample size-corrected Akaike's information criterion (AICc; Bartón, 2018). We then retained predictors in final models if their inclusion reduced overall model AICc. Relevant non-linear terms (i.e. second- and third-order polynomials) and two-way interactions were also tested and included if they reduced AICc and made realistic predictions that matched existing relationships in the published literature. We tested for multicollinearity in final models using variance inflation factors (VIFs). Except for interaction terms, all VIFs for final model terms were below 2. Because of the hierarchical structure of data collection and dependence among samples, we included a random intercept or nested random intercept term in each LMM and GLMM (e.g. ‘site’ or ‘plot within site’; Table S4 in Appendix S1). To assess model assumptions, we performed residual diagnostic tests using the DHARMa package (Hartig, 2018) and checked for spatial autocorrelation of residuals using spline correlograms in the ncf package (Bjornstad, 2019) in r (Appendix S2). When necessary, we transformed predictor and response variables in LMMs and selected different error structures in GLMMs to improve distributions of residuals. To summarize explanatory power of final models, we calculated the marginal (hereafter ‘R²m’) and conditional (hereafter ‘R²c’) coefficients of determination (R² and R²glmm) using the MuMIn package (Bartón, 2018) in r.

2.5 Analytical methods—Thresholds of fire-resistance and reproductive maturity

Using Bernoulli GLMMs of tree survival and cone presence, we determined the minimum diameter at which an individual tree of each species could be expected to survive fire or be cone-bearing by maximizing Youden's J statistic (Youden, 1950). Youden's J evenly weights sensitivity and specificity and is commonly used to identify classification thresholds in Bernoulli models. Next, we determined the average age at which an individual tree would reach these diameter thresholds under open post-fire growing conditions using an LMM of DBH as a function of plant age and local competition (R²m = 0.65; R²c = 0.68; Figure S4 in Appendix S1; species identity and climate variables did not improve model fit). For tree survival, we also incorporated the effect of stand-scale fire severity by identifying fire-resistant diameter and age thresholds across a range of severities: low (<30% basal area mortality within a stand), moderate (30%–70% mortality) and high (>70%). The midpoints of each severity class (i.e. 15%, 50% and 85% for low, moderate and high, respectively) were used in predictions of tree survival. To assess the classification performance of GLMMs for tree survival and cone presence, we calculated the area under the receiver operating curve (AUC) statistic using the rocr package (Sing, Sander, Beerenwinkel, & Lengauer, 2005) and balanced accuracy with the caret package (Kuhn et al., 2014) in r.

2.6 Analytical methods—Testing the effects of climate on tree growth

To determine the extent to which our selected climate variables (average AET and CWD) were connected to tree physiological processes, we performed two additional analyses. First, we quantified site index, a species-specific measure of site quality that estimates the average height of a dominant tree at an age of 100 years. Because P. menziesii individuals meeting the necessary criteria were absent in some stands, we restricted these calculations to P. ponderosa. We used the site index curves of Minor (1964) which were developed for P. ponderosa in Arizona and New Mexico, USA. We calculated the Spearman correlation coefficient (ρ) between site index and average AET and CWD. Next, we used a subset of increment cores (see Potential Predictors), stratified by DBH and plot, to quantify interannual variation in tree growth (using the Gini coefficient of detrended ring widths from 1970 to 1999; Appendix S1). We expected that growth variation would relate to climate (less variable growth on wetter and more productive sites), tree diameter and age (more variable growth in slow-growing individuals) and local competition (less variable growth in open-grown trees), and tested for these effects using LMMs.

3 RESULTS

3.1 Plant traits and vital rates related to fire resistance

The final GLMM of tree survival following fire included terms for species, stand-scale fire severity (i.e. percent basal area mortality), DBH and the interaction between species and severity (Figure 3; R²m = 0.19; R²c = 0.28; AUC = 0.80; balanced accuracy = 74.5%). Pre-fire stand basal area, average AET and average CWD did not improve predictions of tree survival and thus were not included in the final model (Table S4 in Appendix S1; Appendix S2). On average, P. ponderosa individuals were more likely to survive fire than P. menziesii individuals, and fire resistance for each species increased with DBH. Furthermore, interspecific differences in survival increased with fire severity (i.e. a two-way interaction). In low-severity fire (<30% mortality at the stand-scale), P. ponderosa were predicted to survive when larger than 6.9 cm DBH (25 years in open-grown conditions) as compared to P. menziesii at 10 cm DBH (32 years; Figure 3a). In moderate-severity fires (30%–70% mortality), P. ponderosa and P. menziesii were predicted to survive at 25.1 cm DBH (72 years) and 38.1 cm DBH (111 years), respectively (Figure 3b). High-severity fire (>70% mortality) most clearly differentiated the two species, with predicted survival at 43.4 cm DBH (127 years) for P. ponderosa and 66.2 cm DBH (204 years) for P. menziesii (Figure 3c).

FIGURE 3

Predicted probabilities (from final generalized linear mixed model) of individual-tree survival following (a) low-, (b) moderate- and (c) high-severity fire based on DBH (tree diameter at breast height; 1.37 m above ground level) for two North American conifers. Low severity indicates <30% mortality of total stand basal area, moderate severity is 30%–70% and high severity is >70%. Predicted values, conditional on the mean value of random effects, are shown with solid (Pinus ponderosa) and dashed (Pseudotsuga menziesii) curves, and shaded regions approximate ±1 SE (i.e. 68% interval) of the prediction based on 1,000 simulations of the fitted model. Vertical lines give the diameter threshold at which an average tree of each species is predicted to survive a fire event at a given severity. Points (vertically jittered for clarity) show the observed data for trees located within stands that burned in each severity class

The final LMM of relative bark thickness (i.e. mean bark thickness divided by total stem diameter) included stem diameter (at 20–40 cm above ground level; the location of bark measurements), species and a two-way interaction between these terms (R²m = 0.63; R²c = 0.69; Figure 4). Bark thickness varied between species, with the greatest interspecific differences in small individuals (Figure 4). For example, in individuals with stem diameters <5 cm, mean bark thickness in P. ponderosa was 13.5% of stem diameter, as compared to 11.2% in P. menziesii. In individuals with diameters >5 cm, mean bark thickness was 7.6% in P. ponderosa and 7.2% in P. menziesii. Relative bark thickness did not vary across climatic gradients of average AET and CWD (Table S4 in Appendix S1; Appendix S2).

FIGURE 4

Results of the final linear mixed model predicting relative bark thickness (i.e. mean bark thickness divided by stem diameter) across a range of stem sizes for two North American conifers. Bark thickness and stem diameter for this analysis were both recorded at 20–40 cm above ground level to better capture the location at which bark would most influence tree survival following low-intensity surface fires. Predicted values, conditional on the mean of random effects, are given with solid (Pinus ponderosa) and dashed (Pseudotsuga menziesii) curves, points represent the observed data and shaded regions approximate ±1 SE (i.e. 68% interval) of the prediction based on 1,000 simulations of the fitted model

The most parsimonious LMM of post-fire juvenile height included terms for total age, species and surviving overstorey basal area (R²m = 0.52; R²c = 0.53; Figure 5). On average, P. ponderosa juveniles were predicted to grow 20% faster than P. menziesii juveniles (Figure 5a). Post-fire juveniles that established in plots without surviving trees were predicted to have height growth that was 28% faster than juveniles in plots with 10-m² basal area/ha of surviving tree cover (Figure 5b). Average AET, average CWD and predictors that were proxies for fine-scale competition (such as the density of juvenile conifer and angiosperm trees and per cent cover of graminoids), did not improve LMMs of juvenile height (Table S4 in Appendix S1; Appendix S2).

FIGURE 5

Results of the final linear mixed model of post-fire juvenile height for two North American conifers based on (a) plant age and (b) surviving overstorey density. Predicted values in each panel, conditional on the mean values of random effects and other predictors, are given with solid (Pinus ponderosa) and dashed (Pseudotsuga menziesii) curves, where points represent the observed data and shaded regions approximate ±1 SE (i.e. 68% interval) of the prediction based on 1,000 simulations of the fitted model. Note that the y-axis on panel (b) has been transformed for ease of interpretation and differs from the original response variable (shown in a)

3.2 Plant traits and vital rates related to fire resilience

Cone presence was best predicted by a Bernoulli GLMM that included species, DBH and an interaction between the two terms, with clear size thresholds that were consistent across sites (R²m = 0.89; R²c = 0.92, AUC = 0.97; balanced accuracy = 92.6%; Figure 6). Although DBH and age were highly correlated, DBH was a much stronger predictor of cone presence (ΔAICc = 175.6). The inclusion of local competition, average AET and average CWD did not improve model fit (Table S4 in Appendix S1; Appendix S2). The average P. menziesii individual began to produce cones at 12.4 cm DBH (38 years under open growing conditions), which was smaller than P. ponderosa at 18.1 cm DBH (52 years). There was, however, notable intraspecific variation and minimum diameters (ages) of cone production were 6.2 cm (26 years) for P. menziesii and 6.7 cm (28 years) for P. ponderosa. All surveyed individuals with DBHs exceeding 35 cm were cone-bearing (Figure 6).

FIGURE 6

Predicted probabilities (from final generalized linear mixed model) of seed cone presence based on DBH (tree diameter at breast height; 1.37 m above ground level) for two North American conifers. Predicted values, conditional on the mean values of random effects, are given with solid (Pinus ponderosa) and dashed (Pseudotsuga menziesii) curves, points represent the observed data (jittered vertically for clarity) and shaded regions approximate ±1 SE (i.e. 68% interval) of the prediction based on 1,000 simulations of the fitted model. Vertical lines give the diameter threshold at which an average tree of each species is predicted to produce cones. For interpretability between 0 and 30 cm on the x-axis, 59 trees larger than 50 cm in diameter, all bearing cones, were excluded from plotting

The frequency of seed cone years in P. ponderosa from 2006 to 2017 was negatively related to local competition and positively related to DBH and average site AET (R²m = 0.31; R²c = 0.37; Figure 7). In the final Poisson GLMM, the strongest single predictor was local competition (Table S4 in Appendix S1; Appendix S2), and trees without competitors were predicted to produce cones five times more often than those with high competition (i.e. competition indices of 0 and 1, respectively; Figure 7b). Cone year frequency was more closely related to DBH than to tree age (ΔAICc = 1.5) and average AET was a much stronger predictor than average CWD (ΔAICc = 3.8). Trees on sites with relatively high AET (i.e. 650 mm/year) were predicted to produce cones nearly twice as often as trees on sites with low AET (i.e. 400 mm/year; Figure 7a).

FIGURE 7

Results of the final generalized linear mixed model predicting seed cone frequency for Pinus ponderosa at the tree level (i.e. the number of seed cone years recorded in 2006–2017 scaled from 0 to 10) as a function of (a) average AET (actual evapotranspiration), (b) local competition intensity and (c) DBH (tree diameter at breast height; 1.37 m above ground level). Predicted values in each panel, conditional on the mean values of random effects and other predictors, are given with solid curves, and points show the observed data. Shaded regions approximate ±1 SE (i.e. 68% interval) of the prediction based on 1,000 simulations of the fitted model

3.3 Effect of climate on P. ponderosa growth

Of the variables used to represent the site-level climate, AET was most strongly associated with growth in P. ponderosa. Site index (i.e. the average height of a dominant, unsuppressed P. ponderosa at 100 years in age) was strongly related to average AET (Spearman's correlation [ρ] = 0.58; n = 15) and only weakly related to average CWD (ρ = −0.10; n = 15). Similarly, final LMMs of tree-scale variability in interannual growth (as determined from the Gini coefficient of detrended ring widths) indicated that growth was less variable on sites with high average AET (Figure 8a; R²m = 0.16; R²c = 0.22). Interannual variation in radial growth declined with DBH (Figure 8b) and increased with plant age (Figure 8c). In general, young, fast-growing trees on sites with high AET had the lowest interannual variation in radial growth.

FIGURE 8

Results of the final linear mixed model of variation in Pinus ponderosa interannual radial growth (i.e. the Gini coefficient of inequality in detrended ring widths 1970–1999) for individual trees as a function of (a) average actual evapotranspiration (AET), (b) DBH (tree diameter at 1.37 m above ground level) and (c) tree age in 1985 (i.e. the middle of the analysis period). Predicted values in each panel, conditional on the mean values of random effects and other predictors, are given with solid lines, and points show the observed data. Shaded regions approximate ±1 SE (i.e. 68% interval) of the prediction based on 1,000 simulations of the fitted model

4 DISCUSSION

Plant traits and vital rates for P. ponderosa and P. menziesii, two iconic, widely distributed and co-occurring tree species in western North America, showed pronounced variation, much of which could be explained by species life history, individual life stage (i.e. diameter, height and age) and resource availability (i.e. climate and local competition). As hypothesized, P. ponderosa demonstrated greater allocation to fire resistance traits; having thicker bark, more rapid juvenile height growth and achieving resistance to low-severity fire at a 23% earlier age than P. menziesii. In contrast, P. menziesii had a greater allocation to fire resilience traits that facilitate recolonization of disturbed sites following fire, as evidenced by a 28% earlier age of reproductive maturity (i.e. the point at which an average tree is most likely to show seed cone presence) than P. ponderosa. While many plant traits related to resistance and resilience to fire have been described qualitatively and at a coarse scale (i.e. range-wide) for North American conifers (e.g. Baker, 2009), the present study provides a framework for the assessment of factors that influence individual-level trait expression. This individual-level variation is particularly important because myriad ecological processes operate at this scale (Clark et al., 2011). We hypothesized that individual-level trait expression (which in turn influences the pool of traits available in a community) would relate to tree life stage and resource availability, and indeed these factors were strong predictors of each trait and vital rate. Further, the observed inter- and intraspecific trait variation found in the present study can be used to infer potential vulnerability of individuals, populations and communities to altered fire regimes and to inform mechanistic models of future change.

4.1 Interspecific variation in traits promoting resistance and resilience to fire

Across frequent-fire ecosystems globally, thicker bark helps trees to survive surface fires, while more rapid juvenile height growth provides the opportunity for individuals to ‘escape’ (i.e. grow above) more quickly, minimizing the number of surface fire events through which they must survive (Hoffmann et al., 2012; Jackson et al., 1999; Werner & Prior, 2013). Our finding of high relative bark thickness in juvenile P. ponderosa and P. menziesii individuals is indicative of preferential early-life allocation to fire resistance (Jackson et al., 1999) and suggests that both species have adaptations to frequent, low-severity fire. Nevertheless, when compared to P. menziesii, P. ponderosa has thicker bark and more rapid height growth early in life (i.e. <30 years of age), as well as a higher probability of survival across a range of fire severities; all evidence of greater fire tolerance. Our findings related to height growth and bark thickness are generally consistent with prior research on these species in western North America (Bigelow et al., 2011; Laughlin et al., 2011; Littlefield, 2019), although one study in California, USA found that bark thickness for the two species was similar (van Mantgem & Schwartz, 2003), perhaps due to variation across each species’ range. While both P. ponderosa and P. menziesii were present in open, frequent-fire forests historically (c. 1800s), fires were generally less frequent in sites with a greater prevalence of P. menziesii (Swetnam & Baisan, 1996; Tepley & Veblen, 2015). Our findings further elucidate the effects of fire frequency on species composition in Rocky Mountain forests, suggesting that fire-free periods exceeding 25 years (for P. ponderosa) and 32 years (for P. menziesii) may be sufficient for individual trees to establish and reach a size at which they are resistant to the next low-severity surface fire.

While P. menziesii showed relatively less allocation to fire resistance, we found that this species achieved reproductive maturity at an earlier age than did P. ponderosa. Seed production has a high resource cost including substantial inputs of nitrogen, carbon and phosphorous (McDowell, McDowell, Marshall, & Hultine, 2000; Sala, Hopping, McIntire, Delzon, & Crone, 2012), and reproductive investment is often at the expense of investment in other life-history traits (Obeso, 2002; Redmond et al., 2019). Greater allocation to fire resistance during early life stages would be expected in species that occupy sites typified by frequent, low-severity fire (Jackson et al., 1999), whereas species that are more often fire-killed are likely to allocate more resources to reproduction, thereby attaining reproductive maturity at an earlier age (Clark, 1991). Several traits not measured here are also likely to influence interspecific differences in fire resistance and resilience for the two species in our study. Filled seed mass in P. menziesii is one-third that of P. ponderosa (McCaughey, Schmidt, & Shearer, 1986), and low seed mass may facilitate long-distance dispersal of P. menziesii into recently burned areas. Pinus ponderosa has a more open crown structure (i.e. lower whole-tree leaf area) and a lower specific leaf area than P. menziesii (Monserud & Marshall, 1999), factors that reduce shade tolerance (Falster, Duursma, & FitzJohn, 2018), but likely improve fire tolerance (Baker, 2009). Also, the self-pruning of lower branches in larger P. ponderosa reduces the chances of fire spreading from the ground surface to the tree crown (Stevens et al., 2020).

4.2 The influence of life stage and resource availability on individual-level trait expression

We found that large-diameter trees were more likely than small individuals to survive fire, to be reproductively mature, and to be frequent cone producers. Indeed, large trees are known to play a prominent role in fire-prone ecosystems (Keeton & Franklin, 2005; Ordóñez, Retana, & Espelta, 2005), and in driving tree species population dynamics more generally (Harper, 1977). These individuals are more fire-resistant, in part, because total bark thickness increases with tree diameter, and thicker bark increases the time required for lethal heating of the cambium (van Mantgem & Schwartz, 2003). Further, large trees are capable of tolerating more substantial stem and crown damage during a fire than are small trees (Wyant, Omi, & Laven, 1986). Across species, larger individuals are also the primary seed producers (Davi et al., 2016; Greene & Johnson, 1994) because they have a greater crown volume and greater access to resources. So, while plant traits and vital rates vary among species, individual-level traits are strongly dependent upon life stage. Considering the species identities and life stages of individuals within a community helps to paint a more detailed picture of community interactions (Rudolf, Rasmussen, Dibble, & Van Allen, 2014), functional diversity (Díaz & Cabido, 2001) and forest vulnerability to wildfire.

Climatic conditions, described in the present study using water balance metrics of AET and CWD, were not strongly predictive of all plant traits and vital rates. However, P. ponderosa site index, ring complacency (i.e. low Gini coefficients of detrended ring widths) and cone frequency all increased with higher average AET. Together, these results demonstrate that sites with high AET—generally warm and wet—have a more consistent availability of resources for P. ponderosa growth and reproduction. Climate also appears to play an important role in the reproductive output of other Pinus species. For Pinus edulis, average cone crops are smaller on more arid sites (Wion, Weisberg, Pearse, & Redmond, 2019) and increases in growing season temperature have been associated with declines in cone production since the 1970s (Redmond, Forcella, & Barger, 2012). In the context of a warming future, reproduction in P. ponderosa and other plant species inhabiting moisture-limited regions may be strongly influenced by climatic conditions because of inherent trade-offs in allocation to reproduction, growth and defence.

The frequency of seed cone production is an important indicator of forest resilience to wildfire. Across recent fires in the western United States, forest recovery has been more rapid on mesic and productive sites (Korb, Fornwalt, & Stevens-Rumann, 2019; Stevens-Rumann & Morgan, 2019). This is, in part, due to beneficial growing conditions which likely enhance juvenile growth and survival on these sites. However, our finding that P. ponderosa cone production was more frequent in areas with high AET suggests that increased seed availability is one additional mechanism for variation in post-fire recovery across abiotic gradients. Juvenile establishment is episodic for many North American conifers (Andrus, Harvey, Rodman, Hart, & Veblen, 2018; Hankin, Higuera, & Dobrowski, 2019) and establishment years are becoming less frequent in dry forests (Davis et al., 2019). Frequent cone production increases the likelihood that seed availability will align with climatic conditions that are suitable for establishment (e.g. Feddema, Mast, & Savage, 2013; Petrie et al., 2017). Over the next century, AET is projected to increase across the southern Rocky Mountains, particularly at higher elevations and northern latitudes within the region (Rodman, Veblen, Battaglia, et al., 2020). Whereas these projected increases in AET have the potential to benefit growth and reproduction for some plant species and populations, substantial increases in moisture deficit may temper these potential benefits, especially at lower elevations and on south-facing slopes (Rodman, Veblen, Battaglia, et al., 2020).

Competition was an important predictor of juvenile height growth, radial growth in mature trees and cone year frequency. While we did not find evidence that P. ponderosa and P. menziesii responded differently to competition, other studies have demonstrated that juvenile height growth (Bigelow et al., 2011) and mature tree radial growth (Buechling et al., 2017) in P. ponderosa is quite sensitive to the local competitive environment. Reproductively mature P. ponderosa located in open areas had substantially elevated levels of cone production, a finding consistent with prior studies of this species and other conifers (Flathers et al., 2016; Nygren et al., 2017). One likely mechanism for this relationship is the increased availability of light, which helps to explain whole-tree and within-tree variation in cone production (Despland & Houle, 1997; Greene et al., 2002), though local competition for moisture and soil nutrients may also play an important role.

The influence of competition on juvenile height growth, radial growth in mature trees and seed cone production implies that there are potential compensatory effects that can occur with the removal of trees during moderate- and high-severity fire events. Mature trees that survive fire are more likely to produce frequent and abundant cones due to reduced local competition, thereby contributing propagules for new seedling establishment. However, isolated individuals of each species may also be more likely to self-pollinate, and inbreeding depression can reduce total seed production, individual seed mass and rates of seedling height growth (Sorensen & Miles, 1974). Juveniles that establish in open post-fire conditions will grow more rapidly, reaching sizes of fire resistance and reproductive maturity more quickly. But while open conditions may benefit growth, post-fire juvenile densities for these species are highest near surviving trees (Korb et al., 2019; Rodman, Veblen, Chapman, et al., 2020) and these surviving trees also moderate microclimatic conditions (Davis, Dobrowski, Holden, Higuera, & Abatzoglou, 2018). Therefore, severe fires may promote compensatory effects through competitive release but also trigger decreases in gene flow, declines in available seed sources and reduced microclimatic buffering from the overstorey canopy. The spatial patterns of fire severity influence how this balancing act plays out in post-fire environments.

4.3 Forest vulnerability to changing fire activity

Local plant traits and vital rates, and the extent to which they are adaptive in the context of future fire regimes, will help to determine the vulnerability of plant communities to fire-driven conversions (Figure 1c). Over the next century, climatic conditions are expected to become increasingly suitable for wildfire occurrence and spread throughout western North America (Abatzoglou, Williams, & Barbero, 2019). However, projections of future fire activity are also contingent upon fuel dynamics and post-fire vegetation flammability (Hurteau, Liang, Westerling, & Wiedinmyer, 2019; Tepley et al., 2018), and future fire regimes are as yet undetermined. In dry forests dominated by P. ponderosa and P. menziesii, an increase in fire frequency could be expected to favour the most fire-resistant individuals in a community—larger trees, faster-growing individuals and those with relatively thick bark. As a species, P. menziesii could be expected to be more vulnerable than P. ponderosa to increases in fire frequency because of more limited allocation to fire resistance traits. A future with increasingly severe fire would bring more uncertain outcomes. Through both an earlier age of reproductive maturity and greater seed dispersal distance (based on a much smaller individual seed mass; McCaughey et al., 1986), P. menziesii may be able to more quickly colonize large burned areas and those that have suffered early post-fire regeneration failure. However, future climate warming is likely to limit post-fire recovery for these two species (Rodman, Veblen, Battaglia, et al., 2020), and P. menziesii juveniles may be particularly vulnerable to drought-induced mortality (Rother, Veblen, & Furman, 2015).

More broadly, our findings and those of other studies suggest that conifer-dominated forest communities in North America are likely to respond to shifting fire regimes in many different ways (Coop et al., 2020). Interspecific differences in fire-adaptive traits can help to narrow the range of expected responses (Davis, Higuera, & Sala, 2018) and these traits can now be mapped across broad spatial extents (Stevens et al., 2020). The increasing prevalence of high-severity fire would favour species with life-history traits such as resprouting, seed banking and long-distance seed dispersal (Pausas et al., 2004). Short-interval high-severity fires would select for resprouters or herbaceous vegetation over seed-banking species that can require many years to develop seed stores (Enright, Fontaine, Lamont, Miller, & Westcott, 2014; Keeley, Ne'eman, & Fotheringham, 1999; Turner, Braziunas, Hansen, & Harvey, 2019). Where increased fire frequency buffers forests by reducing severity in subsequent events (e.g. Walker, Coop, Parks, & Trader, 2018), species that can resist fire (either through survival of the above-ground stem or resprouting) could be expected to benefit.

While the persistence strategies of woody plants are often described at the species level (e.g. Loehle, 2000; Noble & Slatyer, 1980), additional insight can be gained from considering the factors that shape individual-level trait variation. Indeed, intraspecific trait variation is an increasingly recognized component of many ecological processes (Bolnick et al., 2011; Violle et al., 2012). Overall, community traits are shaped by traits of the component individuals, which are influenced by species identity, individual life stage and resource availability. This study provides a framework for understanding the drivers of inter- and intraspecific variation in plant traits promoting resistance and resilience to wildfire, but improved knowledge of the factors driving trait variation is needed in other regions and species to inform the next wave of ecological models that will be used to assess forest vulnerability in a rapidly changing world.

ACKNOWLEDGEMENTS

This work was funded by Australian Research Council award DP170101288, Joint Fire Science Program Graduate Research Innovation award 17-2-01-4, National Science Foundation award DEB-1833529, the John Marr Ecology Fund, the Colorado Mountain Club Foundation and the University of Colorado Boulder. The authors would like to thank Nicole Batron, Kelsey Bohanon, Aaron Broughton, Kristina Cowell, Molly Day, Katherine Feldmann, Ana Miller-ter Kuile, Rachael O'Neill, Halina North, Josh Vonloh, Kathleen Weimer, Brenna Wider and Jake Zatz for assistance with field data collection and laboratory work. Helpful feedback from two anonymous reviewers led to improvements of this manuscript.

AUTHORS' CONTRIBUTIONS

Authors 3–8 are ranked alphabetically rather than in order of contribution; K.C.R. and T.T.V. designed the study; K.C.R., A.D.G. and A.P.W. conducted fieldwork; K.C.R. performed statistical analyses with input from the other authors; K.C.R. wrote the first draft of the manuscript and all authors contributed to manuscript revisions.

Open Research

PEER REVIEW

The peer review history for this article is available at https://publons.com/publon/10.1111/1365-2745.13480.

DATA AVAILABILITY STATEMENT

All data and statistical analyses presented in this study are available through the Dryad Digital Repository https://doi.org/10.5061/dryad.cz8w9gj1b (Rodman, Veblen, Andrus, et al., 2020).