From canopy to seed: Loss of snow drives directional changes in forest composition

Abstract Climate change is altering the conditions for tree recruitment, growth, and survival, and impacting forest community composition. Across southeast Alaska, USA, and British Columbia, Canada, Callitropsis nootkatensis (Alaska yellow‐cedar) is experiencing extensive climate change‐induced canopy mortality due to fine‐root death during soil freezing events following warmer winters and the loss of insulating snowpack. Here, we examine the effects of ongoing, climate‐driven canopy mortality on forest community composition and identify potential shifts in stand trajectories due to the loss of a single canopy species. We sampled canopy and regenerating forest communities across the extent of C. nootkatensis decline in southeast Alaska to quantify the effects of climate, community, and stand‐level drivers on C. nootkatensis canopy mortality and regeneration as well as postdecline regenerating community composition. Across the plot network, C. nootkatensis exhibited significantly higher mortality than co‐occurring conifers across all size classes and locations. Regenerating community composition was highly variable but closely related to the severity of C. nootkatensis mortality. Callitropsis nootkatensis canopy mortality was correlated with winter temperatures and precipitation as well as local soil drainage, with regenerating community composition and C. nootkatensis regeneration abundances best explained by available seed source. In areas of high C. nootkatensis mortality, C. nootkatensis regeneration was low and replaced by Tsuga. Our study suggests that climate‐induced forest mortality is driving alternate successional pathways in forests where C. nootkatensis was once a major component. These pathways are likely to lead to long‐term shifts in forest community composition and stand dynamics. Our analysis fills a critical knowledge gap on forest ecosystem response and rearrangement following the climate‐driven decline of a single species, providing new insight into stand dynamics in a changing climate. As tree species across the globe are increasingly stressed by climate change‐induced alteration of suitable habitat, identifying the autecological factors contributing to successful regeneration, or lack thereof, will provide key insight into forest resilience and persistence on the landscape.


| INTRODUC TI ON
Climate change is altering the conditions for tree recruitment, growth, and survival, and range shifts are a widely anticipated consequence of novel temperature and precipitation regimes. Local adaptation to historical climate is already creating mismatches between species' current distributions and suitable habitat conditions (Aitken & Bemmels, 2015;Aitken, Yeaman, Holliday, Wang, & Curtis-McLane, 2008), and these disparities are likely to be most extreme for seedlings, which have a narrower range of tolerance to climate conditions than mature individuals (Niinemets, 2010). As climate regimes shift, habitat suitable for survival of mature trees and the conditions necessary for germination and establishment may no longer correspond with each other, potentially leading to simultaneous canopy mortality and declines in regeneration, and, ultimately, shifts in species distributions (Walck, Hidayati, Dixon, Thompson, & Poschlod, 2011). The key to regeneration success and long-term survival will be continued synchronization of tree germination, establishment, and growth with local climate (Aitken & Bemmels, 2015).
While the driving processes are difficult to disentangle, species range contractions consist of concurrent or sequential canopy mortality and regeneration failures. Climate change-induced canopy mortality is already a globally documented phenomenon (Allen, Breshears, & McDowell, 2015;Allen et al., 2010). Altered precipitation regimes (i.e., change in timing, amount, frequency, type) combined with simultaneous increases in temperature are leading to both drought mortality (Guarín & Taylor, 2005;Peng et al., 2011) and, at the other extreme, mortality attributed to earlier snowmelt and spring freezing events (Bourque, Cox, Allen, Arp, & Meng, 2005).
Drought-induced canopy mortality is well-established (Allen et al., 2015;Anderegg et al., 2013), and recent studies have quantified the impacts of prolonged drought (Redmond, Weisberg, Cobb, & Clifford, 2018) and postfire drought conditions (Stevens-Rumann et al., 2017;Young et al., 2019) on regeneration. However, relatively little is known about the consequences of reduced snowpack and early season frost damage on canopy mortality and regeneration response, although these phenomena may become more common as the climate warms (Woldendorp, Hill, Doran, & Ball, 2008)-leading to root mortality and nutrient loss (Decker, Wang, Waite, & Scherbatskoy, 2003), needle and bud injury (Man, Kayahara, Dang, & Rice, 2009), canopy mortality (Buma, 2018), and seedling mortality (Camarero & Gutiérrez, 2004). Regardless of the climatic stressor, widespread canopy mortality is likely to be ongoing under the more extreme conditions predicted for the future, and under such conditions, range contraction potential will be governed by regeneration success or failure.
Not all climate-driven mortality will result in a range contraction, as regeneration after extensive mortality is dependent upon two conditions: (a) available sources of seed and/or vegetative reproduction and (b) the establishment environment (climatic, abiotic, and biotic). Seed supply generally decreases with a loss of mature trees (Tepley, Veblen, Perry, Stewart, & Naficy, 2016), and this loss can result in reductions in seedling abundances, lower recruitment into mature tree size classes, and the potential for local extirpation due to competition with nondecline-affected species (Oakes et al., 2014).
Mortality of mature trees may, conversely, create more favorable establishment conditions by increasing available light and releasing advanced regeneration (Macek et al., 2017;Zeppenfeld et al., 2015) or leading to more successful germination and subsequent survival (Whitmore, 1989). Such increases in favorable microsites may partially offset canopy declines; under such conditions, regeneration of species declining in the canopy may increase despite a reduction in available seed-similar to pulses of recruitment following windthrow events (Dunn, Guntenspergen, & Dorney, 1983;Peterson & Pickett, 1995). Comparing the relative strength of seed source versus community competition and abiotic changes is therefore important in predicting the net effects of mortality and the potential for range shifts at broader scales-thus, resilience.
The effects of canopy mortality on regeneration are, however, difficult to isolate, as climate change-induced canopy mortality often leads to concurrent decline of multiple species. Concurrent declines challenge our ability to identify the factors driving reductions in suitable habitat for mature tree survival versus those leading to regeneration failures for individual species, as a variety of interspecific relationships are changing simultaneously. Simplified systems in which a single species undergoes climate-related mortality in isolation of climate effects on co-occurring species provide a means of parsing out the effects of climate change on mortality, regeneration, and the resultant ecological community.
Mortality drivers in this complex pathway include the limited cold tolerance of roots, a reduction in insulating snowpack due to warmer winters, soil freezing due to lack of insulation, and a positive feedback loop in which canopy gaps in declining forests lead to further reductions in springtime snowpack (Beier et al., 2008;Schaberg et al., 2008). Given predicted future climate conditions, C. nootkatensis is likely to experience continued decline over most of its range (Buma, 2018) and be replaced by other regionally dominant conifers (Oakes et al., 2014). Co-occurring Tsuga heterophylla Raf. (Sarg) (western hemlock) and Picea sitchensis Bong. (Carr) (Sitka spruce) are not considered sensitive to snow loss (Buma & Barrett, 2015) and may increase in dominance when C. nootkatensis fails to regenerate (Oakes et al., 2014). Yet, despite extensive research on the drivers of decline (Barrett, Latta, Hennon, & Eskelson, 2012;Buma et al., 2017;Hennon et al., 2012;Hennon & Shaw, 1997;Schaberg et al., 2008), little is known about C. nootkatensis regeneration following canopy mortality, and the fate of C. nootkatensis and long-term dynamics of affected forests remain unknown.
To address these knowledge gaps and isolate the conditions differentiating habitat suitable for mature tree survival versus those key to successful regeneration in a climate mortality-affected system, we sampled declining C. nootkatensis forests over a five-degree latitude range in the NPCTR of southeast Alaska to ask the following: (a) "Does climate-induced mortality occur across all size classes of the affected species, and, specifically, does regeneration response correspond with the same climate conditions driving mortality?", (b) "If response is differential, what climatic and community factors drive tree mortality versus postdecline composition of the regenerating community?", and (c) "Is community composition stable or in the process of a decline-induced shift?". Our analysis provides new insight into stand dynamics in a changing climate by increasing understanding of forest ecosystem response and rearrangement following the decline of a single species. Obtaining information on life stage response to climate-induced mortality, postdecline community composition, and stand dynamics in the NPCTR and beyond will be essential to scientifically based forest management and vital to supporting conservation efforts in the face of climate change. are common in areas of poor drainage and low forest productivity, decreasing in abundance with increasing slope and increasing depth to groundwater (Bisbing, Cooper, D'Amore, & Marshall, 2016;Neiland, 1971).

| Study area and species
Callitropsis nootkatensis is locally distributed across the NPCTR's hydrologic gradient from emergent wetlands to upland forests, which corresponds to a gradient of low to high forest productivity (Hennon et al., , 2016. This gradient drives the distribution, abundance, and biomass of the region's dominant tree species (Bisbing et al., 2016), with low-lying saturated peatlands limiting the success of most species but providing low-competition environments for stress-tolerant species, such as P. contorta (Bisbing et al., 2016) and C. nootkatensis (Caouette et al., 2015;. Background mortality rates in healthy C. nootkatensis forests average <25% . Callitropsis nootkatensis regeneration occurs via seed but also through vegetative reproduction. Vegetative layering is particularly common on lower-productivity peatlands with poor drainage and/ or high snow cover; lower limbs will produce adventitious roots when depressed by accumulating Sphagnum and snow (Hennon et al., 2016). Individuals recruited through vegetative layering often persist on these lower-productivity peatlands despite mature tree mortality.

| Plot design and sampling
A total of 67 plots were compiled from published (Oakes et al., 2014) and ongoing research by the authors in the perhumid NPCTR subregion of southeast Alaska ( Figure 1). This plot network represents all compatible studies in the region, with compatible defined as those including both regeneration and canopy community data, precise locations, comparable methodologies, and comparable scales. Due to aggregating across sources and studies, each with their own goals, plots were not randomly distributed across the region. However, the data products were compatible and collectively allowed for analysis of regeneration response to the canopy decline severity gradient.
Plots installed by Oakes and colleagues (2014, n = 50) were randomly selected from areas stratified by C. nootkatensis status-live forests (n = 20) and three time-since-mortality classes (n = 10 each in recent, mid-range, and old) on Chichagof Island and in Glacier Bay National Park (installed in 2011-2012, Figure 1). These plots represent the northernmost extent of the contiguous species distribution while also approaching the northernmost extent of documented decline. Data were collected in nested, fixed-radius plots: (a) 10.3m radius (~333 m 2 ) for mature trees (>25 cm diameter at breast height, DBH) and (b) 6m radius (~113 m 2 ) for saplings (<2.5 cm DBH and >1 m in height), treelets (2.5-9.9 cm DBH), and small trees (10-24.9 cm DBH, see Oakes et al., 2014 for details). In addition, conifer germinants and seedlings (<10 cm height) were sampled using eight one-m 2 quadrats (8m 2 total area sampled) installed in the cardinal directions at five and eight meters from plot center.
To extend inference across C. nootkatensis' distribution in southeast Alaska, we leveraged 17 additional nested, fixed-radius plots on Kupreanof and Prince of Wales Islands (installed 2015-2016, Figure 1). Mature and small trees (>10 cm DBH) were sampled over a 400-m 2 plot. Treelets, saplings, and advanced regeneration (>10 cm but <1.37m in height) were sampled in a nested 100-m 2 plot.
Germinants and seedlings (<10 cm height) were tallied by species in four one-m 2 quadrats at the corners of the 400-m 2 plot (16 m 2 total area sampled). Hereafter, the term regeneration refers to the combined germinant and seedling classes, including all recently emerged germinants (i.e., cotyledons still visible) to seedlings less than 10cm in height and, presumably, <2 years of age. Across all plots, species, DBH, tree status (live/dead), live crown condition, and snag class were sampled for all trees. Snags were identified to species, if possible, and C. nootkatensis snags classified into time-since-mortality classes according to . Unidentifiable snags were classified as unknown species.
Additionally, regeneration was identified to the genus for Tsuga germinants (heterophylla and mertensiana) but to the species for all other species and size classes. Disturbed areas (e.g., windthrow, landslides) were eliminated from plot selection to avoid the confounding influence of disturbance on community composition.

| Environmental variables
Climate data were extracted from ClimateWNA v5.51 (http://www. clima tewna.com, Wang et al., 2016), which provides spatially interpolated, locally downscaled climate data and derived biologically relevant climate variables. Data were obtained for three date ranges to capture drivers of ongoing decline and regeneration-the last few F I G U R E 1 Map of Callitropsis nootkatensis plot network distributed across southeast Alaska. Previously established plots were installed by Oakes and colleagues in 2011 and 2012 in Glacier Bay National Park (GBNP) and Chichagof Island (CHICH), and additional plots were installed by the authors in 2015 and 2016 on Kupreanof Island (KUP) and Prince of Wales Island (POW). Inset plots for illustration of the numerical distribution of plots only (scale varies). For a complete range and decline map, see Buma et al. (2017) or Buma (2018) decades (30-year normal, 1981-2010), the five-year period prior to and including sampling year (five-year average, range 2007-2016), and the year of sampling (annual, range: 2011-2016). Annual data were used to evaluate the influence of current year conditions on the regenerating community, while data from the five-year period were used to quantify the influence of multiyear climate on canopy mortality, potential seed availability and viability, and regenerating community composition. A five-year window was selected for two reasons: (a) C. nootkatensis cones take two to three years to mature and produce viable seed (Bonner & Karrfalt, 2008), and (b) we focus regeneration analysis on individuals presumed to be up to two years of age.
Prior to analysis, we ran a correlation analysis to reduce the full set of ClimateWNA variables to a minimally correlated set (<0.65) while retaining those considered biologically important in C. nootkatensis decline, including winter and spring temperatures and precipitation, length of the growing season, and frost period (Buma et al., 2017;Hennon et al., 2012). The following variables were selected as potential predictors: winter (December-February) and spring (March-May) minimum and maximum temperatures (°C); winter, spring, summer (June-August), and average annual precipitation (mm); number of frost-free days in winter and spring (days); average annual length of growing season (frost-free days), date of first frost (Julian date); and annual precipitation as snow (mm; Table 1).
Local topographic data were obtained from the NASA ASTER mission (30m resolution, LP DAAC 2017) and used to derive elevation, slope, and aspect. To assess the role of local soil drainage, which influences competition and rooting, we selected two metrics operating at different scales. At the local scale, we used Sphagnum coverage. To do so, we sampled bryophyte (n = 67) and Sphagnum (subset, n = 12) cover on replicate 1-m 2 quadrats (8-10 per plot).

| Data analyses
To control for the possibility of general forest declines, which would affect all species and obscure C. nootkatensis-specific mortality, we used simple linear regressions to assess the correlation between the proportion of dead C. nootkatensis (log-transformed) as compared to that of the other dominant co-occurring tree species, and Kruskal-Wallis tests for nonparametric, ranked data to assess variation in proportional mortality among species, locations, and size classes (as defined above). There were no significant concurrent declines in any co-occurring species (p > 0.05), so we focused subsequent mortality analyses on C. nootkatensis alone. We also used a Kruskal-Wallis analysis to test for significant differences in live tree abundances among species and locations. We then tested for significant differences in canopy and regenerating community composition among locations and as related to C. nootkatensis mortality with multivariate analysis of variance (MANOVA) tests using Bray-Curtis distances in the vegan package. The proportion of dead C. nootkatensis in the canopy (dead C. nootkatensis out of total C. nootkatensis) was also compared to regeneration density using a negative log-transformed linear regression.
Next, we compared drivers of C. nootkatensis canopy mortality versus regeneration. Two random forests analyses were run-one to model the proportion of dead C. nootkatensis in the canopy and another the abundance of C. nootkatensis regeneration. Both were based on potential drivers related to climate, disturbance exposure, topography, drainage, and competition (Table 1). Random forests, an extension of regression tree analysis (Breiman, 2001), are well suited for complex, nonlinear interactions between variables and generally perform better than other methods in predictive accuracy (Prasad, Iverson, & Liaw, 2006).
We took a two-step process similar to importance-based variable selection procedures (Evans & Cushman, 2009

| Community composition, canopy mortality, and regeneration
Tree diversity is generally low in the region (Caouette et al., 2015;Neiland, 1971), and despite the latitudinal range, all plots were similar in species composition. The canopy was a mixed-conifer forest of Tsuga (heterophylla and/or mertensiana), P. sitchensis, and C. nootkatensis. Picea sitchensis, although common in the region, was rare on the plot network, with only one tree documented on many of the plots.
Pinus contorta occurred in peatlands on Chichagof Island, while T. plicata was found only at higher elevations on Prince of Wales Island (other plots fell outside T. plicata's range). Total basal area ranged from 9 to 87 m 2 /ha, and higher C. nootkatensis basal areas were generally found on higher productivity, upland forests or in areas north of the decline (Table A1). Callitropsis nootkatensis regeneration abundance ranged from 0 to 9.6 per m 2 (mean 1.9/m 2 , median = 1/m 2 ; Table A1).
Mortality was documented in each of the most common conifers on the plot network (C. nootkatensis, P. sitchensis, T. heterophylla, T. mertensiana; Figure 2) and across all size classes, but, with the exception of C. nootkatensis, proportions were in line with or lower than expectations of snag abundances for the region (Deal, Oliver, & Bormann, 1991;Hennon & McClellan, 2003). Proportional mortality in mature P. sitchensis was high on Kupreanof Island (Figure 2), but this was driven by a lack of trees on the plot network (two dead of three total trees on eight plots).
Mortality of less common P. contorta occurred only on Chichagof Island (mean = 30% ± 3% for mature and small trees), and no mortality was documented for T. plicata.
Callitropsis nootkatensis mortality did occur across all tree size classes (mature, small, and treelet; Figure 2), and estimated time since C. nootkatensis mortality ranged from 0 (in healthy stands) to 75 years (Table A1)

| Climate and community drivers of mortality and regeneration
In the random forests model, higher C. nootkatensis mortality was correlated with cool winter temperatures and lower winter precipitation as well as two metrics of soil drainage-moderate Sphagnum percentage and higher slopes (mean squared residuals = 0.06, variance explained = 0.40; Figure 4). As a result, the highest proportional mortality on the plot network was found in cooler, relatively drier regions and on higher productivity upland forests.  This was corroborated by the direct comparison between canopy mortality and regeneration; regeneration (log-transformed) was negatively correlated with increasing mortality (p < 0.001, r 2 = 0.38, F = 40.64 on 1,65 df; Figure 6).

| Response and stability in regenerating community composition
Four regenerating community types were identified in the cluster and indicator species analysis. Community types were largely determined by extent of C. nootkatensis decline, with communities assigned to mixed-conifer (Mixed), C. nootkatensis (CANO), Tsuga-P.
The Tsuga-PISI type was found on steeper slope, higher productivity plots and consisted of Tsuga (ISS = 81, p < 0.01) and P. sitchensis In areas of high C. nootkatensis decline, C. nootkatensis regeneration was low and replaced by Tsuga spp. (Figure 7), and community F I G U R E 4 Change in predicted percent of mortality (modeled as proportion of dead Callitropsis nootkatensis) on the plot network as driven by the top four most significant uncorrelated variables in the final model. Y-axis is the modeled mortality percentage as a function of the overall random forest model while varying the top for variables, respectively. For example, mortality generally declines in areas of higher winter precipitation and lower slopes

F I G U R E 5 Change in predicted
Callitropsis nootkatensis regeneration densities on the plot network as driven by the top four most significant uncorrelated variables in the final model. There is a clear relationship between forest health, seed source, and regeneration, with higher densities seen in areas of lower C. nootkatensis mortality, higher live C. nootkatensis (presumably seed source), and higher overall basal area. Higher densities are also found in wetter landscapes, with a higher contributing area

| D ISCUSS I ON
As the climate changes, successful tree regeneration is threatened by novel climate and disturbance conditions and associated shifts in canopy community composition. Our study highlights how species-specific sensitivity to climate change can lead to restructuring of the forest community following canopy mortality.
We demonstrate that reductions in both mature tree and regeneration abundances after single-species mortality events occur via the same mechanism-death of mature, seed-producing trees. In the case of C. nootkatensis, our work suggests that climate-induced forest mortality is driving alternate successional pathways in forests where C. nootkatensis was once a major component, which is likely to lead to long-term shifts in community composition and stand dynamics. As suitable habitat conditions shift with ongoing global climate change, both mature trees and regeneration will be increasingly exposed to novel conditions, and identifying the autecological factors contributing to successful regeneration, or lack thereof, is an essential first step toward predicting forest response and resilience to climate change.

| Climate impacts on mortality and regeneration in Callitropsis nootkatensis
In the NPCTR, climate change-induced canopy mortality is restricted to a single species, C. nootkatensis, allowing for a focused examination of the effects of climate, seed source, and local conditions on regenerating community response and subsequent stand dynamics. Mature tree mortality across the study area corresponded with a reduction in C. nootkatensis regeneration ( Figure 6) and a shift to surviving canopy species, predominantly Tsuga (Figure 3), a widespread species known to dominate the regenerating community following disturbance (Alaback and Tappeiner II (1991); Deal & Farr, 1994). Callitropsis nootkatensis regeneration constituted less than 20% of the regenerating community in decline areas ( Figure 3) and appears to be limited by a lack of available seed and vegetative source, although drainage conditions, as determined by slope, and maximum winter temperatures also contribute to structuring regenerating community composition ( Figure 7).
Local hydrologic regime and associated ecosystem productivity are key factors in both species' distributions (Bisbing et al., 2016) and in the extent of C. nootkatensis decline (D'Amore & Hennon, 2006).
Mortality across this plot network was most extensive on wetter, lower-productivity peatlands; however, proportional mortality was greatest in productive upland forests. The apparent lack of congruity between these results and previous research, which reports higher mortality on lower slopes (D'Amore & Hennon, 2006), is explained by our use of proportional mortality rather than total. Although mortality is most extreme on saturated peatlands, it is likely that surviving individuals will allow for ongoing perpetuation of the species on lower-productivity peatlands.
Nonetheless, the loss of C. nootkatensis from both the canopy and the regenerating community across the extent of decline in southeast Alaska indicates that canopy trees and regeneration are in sync in their responses to climate change. Cold, low-snow winters have led to widespread mortality over our study area ( Figure 4) and across the species' range (Buma et al., 2017), and local drainage conditions (i.e., slope, hydrologic regime) amplify or mitigate response of the canopy (Figure 4) and regenerating community (Figures 5 and 7). While this study cannot conclusively determine the mechanism for the lack of regeneration, the close correspondence between the canopy and regenerating communities is striking. These plot-level findings are consistent F I G U R E 6 Callitropsis nootkatensis (CANO) regeneration densities (individuals per square meter) per plot (black, filled circles) as a function of the proportion of dead C.nootkatensis in the canopy. Regeneration densities decline with increasing severity of canopy mortality (p < 0.001) with recent modeling work on drivers of canopy mortality that identified winter temperatures and slope as conditions leading to decline at broad scales (Buma et al., 2017). Mortality is predicted to be ongoing in areas above the winter snow threshold, and up to 50% of current suitable habitat is expected to experience climate conditions favorable to decline (Buma et al., 2017). Continued mortality of this conifer is likely to lead to more widespread shifts in community composition (Oakes et al., 2014;data presented here). The close relationship between seed source and composition of the regenerating community suggests that no compensation in the form of increased recruitment will make up for the loss of the C. nootkatensis canopy.

| Single versus multiple species decline
Mortality events specifically attributed to climate change (Breshears et al., 2005;Van Mantgem & Stephenson, 2007;Williams et al., 2010) do not typically discriminate but, instead, lead to decline or mortality of multiple species in the affected ecosystem. Sierra Nevada mixed-conifer forests have, for example, seen concurrent declines in shade tolerant Abies and shade-intolerant Pinus following years of temperature-driven drought stress (Van Mantgem & Stephenson, 2007), while extreme droughts in the arid southwest led to mortality in Pinus edulis and Juniperus monosperma (Mueller et al., 2005).
In both cases, no immediate impacts on regenerating community F I G U R E 7 Nonmetric multidimensional scaling (NMS) of the regenerating community by classification of the severity of Callitropsis nootkatensis canopy mortality (shapes) and community type (colors), with each point representing individual sample plots. Mortality severity classes: low = 1%-25%, moderate = 25.1%-69.9%, high = 70%-99%, and all = 100%. Ellipses represent the mean scores of each NMS community cluster. Species labels represent the most abundant species in each NMS community cluster. CANO, Callitropsis nootkatensis; PICO, Pinus contorta spp. contorta; PISI, Picea sitchensis; THPL, Thuja plicata; and Tsuga, Tsuga heterophylla and mertensiana. Variables and associated vector arrows indicate the direction and magnitude of effects of the canopy community, local stand conditions, and local annual and five-year climate on regenerating community composition. Arrow length is proportional to the magnitude of correlation. Variables identified as both significant (p < 0.05) and minimally correlated include Max.Wt.Temp = maximum winter temperature (°C), Five.Yr.PAS, fiveyear average precipitation as snow (mm), average plot slope (°), PICO, Pinus contorta spp. contorta basal area; and PISI, Picea sitchensis basal area composition were evident, although higher P. edulis mortality indicated that long-term shifts are likely (Mueller et al., 2005). The climate change-induced C. nootkatensis mortality in the NPCTR is unique in its isolated effect on a single species but also in its observable, and now documented, postdecline shift in species dominance.
Regenerating community response in other single-species mortality events has been highly variable and largely driven by the severity of a biotic disturbance, such as insect attack (Burr & McCullough, 2014), on the canopy community and variability in local establishment conditions (Kayes & Tinker, 2012). In some cases, regeneration proceeds successfully following the loss of canopy species (Diskin, Rocca, Nelson, Aoki, & Romme, 2011;Macek et al., 2017), while, in others, species experience loss of dominance (Pelz & Smith, 2012), increasing the probability of ecosystem type conversions (Burr & McCullough, 2014;Klooster et al., 2014). Self-replacement of a single species following canopy morality has been documented following extreme drought (Suarez & Lloret, 2018) and bark beetle attack (Diskin et al., 2011), where co-occurring species established but failed to dominate. These instances of ongoing success, however, occurred in forest types where the impacted species dominated the canopy prior to the disturbance and either did not experience wholesale mortality, as is the case in Nothofagus drought mortality (Suarez & Lloret, 2018), or possessed a canopy seed bank, as occurs with Pinus contorta ssp. latifolia forests (Diskin et al., 2011).
Regeneration declines or failures have also been documented in cases of single-species mortality (Burr & McCullough, 2014;DeRose & Long, 2007;Klooster et al., 2014;Pelz & Smith, 2012). Laminated root rot in the Northwest has led to similar compositional changes as seen in C. nootkatensis forests, where T. heterophylla assumes dominance as canopy mortality of Pseudotsuga menziesii proceeds (Hansen & Goheen, 2000). Additionally, the 1990s Dendroctonus rufipennis attack on Picea engelmannii led to Abies lasiocarpa and Populus tremuloides dominance in the regenerating community, precluding P. engelmannii from returning to dominance (DeRose & Long, 2007). As with C. nootkatensis response to canopy mortality, the primary limiting factor in all cases is an available seed source.
Regenerating community response to canopy mortality is clearly influenced by a number of factors, such as establishment environment conditions, postdisturbance climate, seed availability, and herbivory-but available seed source is regularly identified as a primary limiting factor, regardless of disturbance type (e.g., fire vs. beetle attack) and magnitude (e.g., single vs. multiple species). Consistent with studies in other forest types and under different climate-induced disturbances, our findings suggest that seed source (DeRose & Long, 2007;Redmond et al., 2018;Urza & Sibold, 2017) and microsite/establishment environments (Harvey, Donato, & Turner, 2016;Redmond et al., 2018;Urza & Sibold, 2017)  Documenting changes in communities following canopy mortality will be crucial to identifying the ecological, and potentially economic, consequences of these losses. Identifying successional trajectories, in particular, will reduce the large uncertainty around the long-term impacts of climate change-driven forest mortality events. Forest response to novel disturbance conditions will be hard to predict, and the lessons learned from these cases can help shift our expectations of postdisturbance stand dynamics, particularly in climate-impacted systems.

| Ecosystem resilience and transition to alternate stable states
The consequences of rapid climate change on forest resilience remain uncertain and are likely to be highly variable, based on a particular forest's ecological memory (i.e., information legacies of ecological adaptations to disturbance; Johnstone et al., 2016). For forests to be resilient and resist transitions to alternate states, there must be synchrony between both information legacies (i.e., genetic adaptations to disturbance, like sprouting in avalanche-prone ecosystems) and material legacies (i.e., the physical legacies, such as seedbanks, that are present after a disturbance event; Johnstone et al., 2016).
Historically, C. nootkatensis' shallow root system and early response to spring warming were beneficial, allowing for early, rapid growth each growing season (Hennon et al., 2016); however, this once advantageous legacy is now a deleterious adaptation. Mortality in mature trees limits the potential for biological inertia (Young et al., 2019) due to a lack of available seed. Instead, advanced regeneration via vegetative reproduction remains a successful strategy even in areas of severe decline (unpublished data/personal observation).
This may help the species maintain a presence on the landscape during periods of mature tree mortality until periods more favorable for sexual reproduction, similar to the "orphaned cohort" example of Fraxinus in response to Agrilus planipennis invasion (Klooster et al., 2014). The immediate loss of seed-producing trees and seedlings ( Figures 2 and 3), as well as competition from faster growing species, means these forests are lacking the biological inertia needed for resilience and are likely to transition to T. heterophylla-dominated forests even if C. nootkatensis maintains some temporary or disjunct presence on the landscape. For species or forest types undergoing similar climate-induced mortality events, ecologists will need to determine which legacies contribute to ecosystem resilience, or if novel climate or disturbance will remove these legacies that were historically critical.
Ecosystem resilience may also vary as a function of regional climate fluctuations over space and time, known as transitional climate mortality (Buma, 2018), where mortality is highest within a particular range of climatic conditions but decreases above or below this range. Currently, the mid-range of the C. nootkatensis' distribution is experiencing mortality, but, if emissions scenarios continue toward worst-case trajectories, it is possible that the mortality "donut-hole" will be relatively short-lived (Buma, 2018

| CON CLUS I ON S AND NE X T S TEPS
Widespread mortality of tree species due to changing climate is a major concern in forests worldwide, but the potential for resilience is rarely assessed during an ongoing mortality event, a gap we have attempted to fill here. Species' responses to climate change-induced mortality will vary widely based on species-specific traits, sensitivity to climatic extremes, biotic stressors, and abiotic conditions of the establishment environment, thus requiring autecological studies on factors limiting versus promoting success. In C. nootkatensis forests, there is no increase in C. nootkatensis regeneration abundances to offset canopy mortality. As a result, this forest type is not resilient to mortality associated with ongoing snow loss, and a type change appears to be underway. This example of climate change-driven mortality in a single species highlights how species-specific sensitivity can lead to shifts in community composition and stand dynamics following canopy mortality via the same mechanism-death of mature seed trees.
Few strategies or solutions exist for forests vulnerable to ecosystem type conversion due to mature tree mortality and associated loss of seed source, and numerous knowledge gaps remain.
In the case of C. nootkatensis, common garden studies using seed sources from across the species' range could allow for identification of genotypes with fine-root frost tolerance. Planting gardens across decline severity gradients would also allow for targeted research on regeneration response to concurrent canopy mortality and associated establishment conditions. Given the potential for a wave of transitional mortality across the C. nootkatensis range (Buma, 2018), however, survival of local versus foreign seed source is hard to predict. Long-term monitoring in our plot network will allow us to track forest response to the predicted transitional mortality phenomenon and provide demographic information on the impacts of climate on different size classes as well as on the growth and survival rates of regeneration across the decline severity gradient. Demographic studies will also resolve the confounding effects of seed availability, canopy mortality, and establishment conditions on regeneration. Simultaneously investigating climate-induced tree mortality and subsequent postmortality resilience gives a clearer view of the long-term consequences of climate change on forest health, and this observational study has just scratched the surface of filling the critical knowledge gaps essential to understanding and predicting long-term forest resilience to climate change. TA B L E A 1 (Continued)