Tree regeneration following wildfires in the western US: a review

Wildfires, like many disturbances, can be catalysts for ecosystem change. Given projected climate change, tree regeneration declines and ecosystem shifts following severe wildfires are predicted. We reviewed scientific literature on post-fire tree regeneration to understand where and why no or few trees established. We wished to distinguish sites that won’t regenerate to trees because of changing climate from sites where trees could grow post fire if they had a seed source or were planted, thus supporting forest ecosystem services for society and nature, such as timber supply, habitat, watershed protection, and carbon storage. Our literature review showed that little to no post-fire tree regeneration was more common in low-elevation, dry forest types than in high-elevation forest types. However, depending on the region and species, low tree regeneration was also observed in high elevation, moist forests. Regeneration densities varied by species and seedling densities were attributed to distances to a seed source, water stress or precipitation, elevation, slope, aspect, and plant competition. Our findings provide land managers with two primary considerations to offset low tree regeneration densities. First, we supply a decision support tool of where to plant tree seedling in large high severity burned patches. Second, we recommend possibilities for mitigating and limiting large high severity burned patches to increase survival of trees to be sources of seed for natural regeneration. Few or no tree seedlings are establishing on some areas of the 150+ forest fires sampled across western US, suggesting that forests may be replaced by shrublands and grasslands, especially where few seed source trees survived the wildfires. Key information gaps on how species will respond to continued climate change, repeated disturbances, and other site factors following wildfires currently limit our ability to determine future trends in forest regeneration. We provide a decision tree to assist managers in prioritizing post-fire reforestation. We emphasize prioritizing the interior of large burned patches and considering current and future climate in deciding what, when, and where to plant trees. Finally, managing fires and forests for more seed-source tree survival will reduce large, non-forested areas following wildfires where post-fire management may be necessary.


Introduction
Continuing climate change, droughts, and extreme weather (IPCC 2013), coupled with associated changes in wildfire activity (Westerling et al. 2006, Westerling et al. 2011, Abatzoglou and Williams 2016 are resulting in landscape ecosystem changes and shifts in community composition (Stevens-Rumann et al. 2018). Climate change is altering the mountainous ecosystems of the western US and affecting the people who depend on them for ecosystem services and livelihoods. With rapid biophysical changes already occurring in these forests, land managers are increasingly seeking to understand and mitigate the effects of a changing climate. Effective action depends on understanding regional and local implications of climate science and ecological effects (Blades et al. 2016), which can directly affect fire extent, tree mortality, and post-fire ecosystem recovery.
Legacies of prior disturbances and land use history, in addition to climate, play a prominent role in disturbance severity and subsequent recovery following disturbances (Hessburg et al. 2005, Westerling et al. 2011. While ecosystem shifts are concerning in any ecosystem type, the transition from forests to grasslands and shrublands is often particularly alarming due to the loss of carbon storage capacity in forests versus grasslands or shrublands (Liang et al. 2018), habitat loss for many wildlife species (e.g., Hobson and Schieck 1999), and the potential economic loss in timber industries (Thomas et al. 2017). In some settings, shifts to grassland or shrubland may be long-persisting as alternate stable states that do not transition back to forested ecosystems (e.g., Savage and Mast 2005). Although some studies forecast a reduction of conifer-dominated ecosystems in the coming century due to a combination of climate and disturbances, these studies also point to locations where conifer regeneration may be abundant (Baker 2018, Serra-Diaz et al. 2018.
Forest ecosystem transitions are precipitated by both mortality of mature trees and reduced recruitment. Extensive mortality of mature trees due to wildfires, bark beetles, and drought is of growing concern. Severe wildfires contribute to millions of hectares of mature forest loss in the western US every year (NIFC 2018). While there is some debate about whether the proportion of area burned at high severity within large wildfires has increased in recent years (Dillon et al. 2011;Flannigan et al. 2013, van Mantgem et al. 2013, Parks et al. 2016a, area burned in large wildfires has increased over the past 30 years in the western US (Dennison et al. 2014). Further, mature tree mortality induced by drought stress and the compounding susceptibility to insects has contributed to reduction of forest cover worldwide (Allen et al. 2010, Hicke et al. 2012, Anderegg et al. 2015. Conifer tree species in the western US exhibit many different life history strategies. Some have traits that aid in regeneration after high-severity fire, including bradspory, or serotiny, in which cones only open after heat is applied, thus protecting seed during a high-intensity fire and releasing viable seeds immediately following wildfires (Lotan 1967). Species like lodgepole pine (Pinus contorta Douglas ex Loudon) are often identified as serotinous; however, the proportion of serotinous cones and individuals is highly variable and largely unknown outside of the greater Yellowstone area (Lotan 1967). Populations of lodgepole pine outside of the Rocky Mountains may be primarily non-serotinous, depending on site and disturbance history (Lotan 1967, Alexander 1974. Most conifers in western North America rely on other seed dispersal and reproduction mechanisms. For example, species like whitebark pine (Pinus albicaulis Engelm.) and limber pine (Pinus flexilus James) can rely on dispersal by animals (Agee 1993). Many other species, including ponderosa pine (Pinus ponderosa Lawson & C. Lawson) and Douglas-fir (Pseudotsuga menziesii [Mirb.] Franco), rely on nearby seed trees for wind or gravity dispersal (Agee 1993). Surviving trees are critical sources of seed for tree replacement after fires, thus the severity and extent of mature tree mortality is critical, although less so for serotinous trees. For many species, living seed trees and the number and viability of their seeds plays a critical role in tree establishment. Drought and temperature stress impact the youngest life stages of many plants (Bell et al. 2014, Petrie et al. 2016. Thus, even with available seed, germination and establishment may be controlled by climatic conditions (Petrie et al. 2016), and temperature and moisture requirements vary significantly by species (Petrie et al. 2016, Davis et al. 2018. In many ecosystems, site conditions such as aspect, slope, and microsite conditions can influence success, with fewer tree seedlings often found on harsh sites (e.g., Bonnet et al. 2005).
However, it is unclear how seed source, climate, competition or facilitation, and micro-climates may interact to either promote or inhibit successful reestablishment post fire.
Managers commonly plant trees or plan for natural regeneration from local seed sources (Johnson et al. 2010). On federal and state lands in the US, policies require reforesting after logging and fires within areas that are actively managed for timber supply (e.g., 16 U.S.C. § 475, 16 U.S.C. § 551; 81 FR 24785). On private lands, reforestation is typically required after a disturbance by state forest practices acts (e.g., Idaho IDAPA 20.02.01, Washington Title 222 WAC). While some planting may be contested from a historical fire regime perspective or when regeneration is expected to be slow and episodic (e.g., Owen et al. 2017, Baker 2018, managers of many public and private lands are mandated to regenerate by law and policy. Thus, there is a need to strategically target their limited money, personnel, and time where they will be effective. In a regional study of the Rocky Mountains of the US, Stevens-Rumann et al. (2018) found decreases in tree regeneration in recent decades partly as a result of increasing temperatures and low moisture availability and long distances to seed sources. They detected a significant change in post-fire climate for fires that burned after 2000 in comparison to fires that burned between 1984 and 2000. However, Stevens-Rumann et al. (2018) did not cover a full range of forest types across their study areas, with only subalpine and upper montane forests represented in the northern Rockies, and this warranted an additional analysis of the literature. Here, we built upon this regional dataset with a literature review to address the following questions for all western US forests: 1) To what extent do recent studies of post-fire tree regeneration indicate patterns of low tree regeneration densities post fire? 2) What causes are attributed to low tree regeneration densities post fire? 3) What are the ecological and management implications for the future throughout the West, particularly where extensive mature tree mortality due to fire is coupled with low regeneration densities?
We conducted a systematic review of recently published literature that reported field data on natural tree regeneration following wildfires in the western US. We discuss the ecological and management implications for the future, particularly when a lack of regeneration is coupled with extensive mature tree mortality. We outline key information gaps and scientific uncertainties that currently limit our ability to determine trends in forest regeneration and predict locations of future climate-induced ecosystem transitions. Finally, we discuss possible management actions that may offset the effects of low regeneration densities, and on which sites.

Methods
We conducted a systematic search for published accounts of post-fire tree regeneration using the ISI Web of Science (https://login.webofknowledge.com) and Google Scholar (https://scholar.google.com/). We used different three-word combinations of these key words: "wildfire" and "forest" or "tree regeneration," or "seedling" and "failure" or "lack of" or "alternate stable states" or "transition" in the searches on both ISI Web of Science and Google Scholar. In addition, we used Google Scholar to find more recent publications that had cited the publications found in our search. To identify potential explanatory factors for tree seedling density, we additionally searched for terms "distance to seed source," or "climate," or "repeated fire" in conjunction with "wildfire." Only peer-reviewed papers published since 2000 that included field measurements of natural (not planted) post-fire tree seedling density in the western US were included in our analysis. Many researchers included statistical modeling, and a few had manipulative experiments or simulation modeling, but such papers without field observations were excluded because we wanted to understand observed tree regeneration patterns.
Data presented here are from studies that focused on natural regeneration. Additionally, we excluded post-fire studies that gave densities of seedlings but did not analyze factors potentially influencing the observed tree regeneration. We limited our inferences to only those studies or parts of studies that focused on areas not treated post fire. Forests worldwide are faced with similar concerns over low regeneration densities (e.g., Retana et al. 2002, Pausas et al. 2008, Paritsis et al. 2015; however, we felt the potential influencing factors across the world would be many and thus were excluded from our review. While we focused primarily on conifer species, all tree species mentioned in each paper were considered. We focused on papers published since the year 2000 for several reasons. First, if regeneration densities are changing due to climate and increased area burned, as demonstrated in multiple studies (e.g., Littell et al. 2009), then we expected these effects to be evident in more recent literature. However, multiple wildfires included in studies presented here burned prior to 2000, with the earliest wildfires in the review dating from the 1940s. Second, while localized climate data dates back to 1979, the availability of these data and use in post-fire studies is only found in the literature in the last 5 to 10 years. While we included papers in the discussion that focused on >100-year-old fires that employed dendrochronological methods for tree establishment, we intended to focus on patterns of recent regeneration in our review. Additionally, there are many potential disturbances that interact with wildfires to change post-wildfire recovery, such as wind events, pathogens, seed predation, human disturbances, etc. While papers that discussed these topics were included, we focused on the other driving factors of regeneration for this review as well as on the two most prevalent repeated disturbances in the literature: repeated wildfires and bark beetle-fire interactions.
In our management recommendation, we excluded discussions of other post-fire and pre-fire management strategies besides wildfires and planting, as these other aspects of management were not addressed directly in this review. We did not incorporate any studies of regeneration after prescribed fire or forest treatments and did not synthesize data on salvage logging, mulching, or other post-fire management actions. Thus, we did not provide recommendations around these actions.

Results
We found more than 200 publications in our search. After those not meeting our criteria were excluded, we included the remaining 49 in our synthesis. These publications documented tree seedling presence or density by species for 1 to 64 years following more than 150 wildfires in forests across the western US (Fig. 1).
Post-fire regeneration from 20 conifer tree species were measured in these studies, the most common being ponderosa pine, lodgepole pine, Douglas-fir, subalpine fir (Abies lasiocarpa [Hook.] Nutt.), and Englemann spruce (Picea engelmanii Parry ex Engelm.). Other conifers found in at least one study were twoneedle pinyon (Pinus edulis Engelm.), knobcone pine (Pinus attenuata Lemmon), sugar pine ( Tree seedling density was highly variable, with some plots within a given fire having an abundance of seedlings and other sites having few to none. The warmest and driest sites near lower timberline had the highest probability of none or very few seedlings. Researchers attributed the lack of post-fire tree regeneration success to multiple factors, including distance to seed source, burn severity, moisture stress, increased temperatures, repeated disturbances, topographic variables, and competing vegetation. Areas with high mortality of mature trees had few tree seedlings. In the studies assessed here, seed availability was often inferred from either distance to a living tree seed source or burn severity. In 26 studies, researchers measured and provided statistical significance for distance from plots or transects to living trees, although whether this was to a single living tree or distance to a low-severity burned or unburned patch with many trees varied among the studies (Table 1). In 24 of the 26 studies, tree regeneration density decreased significantly at distances of 40 to 400 meters from a living mature tree, regardless of which conifer species was dominant.
Multiple researchers analyzed tree seedling density by burn-severity class (low, moderate, and high severity) or some other ground-truthed version of severity (Table 2). For example, Coop and Schoettle (2009) used "partial burn" or less than 100% versus "complete burn" or 100% tree mortality, while Welch et al. (2016) used five severity classes. Additionally, multiple studies measured both severity and distance to seed source (e.g., Coop et al. 2010, Rother andVeblen 2016). Proximity to potential seed source trees is often assessed using burn-severity classes, and these burn-severity categories and distance to seed Fires presented in all selected US studies listed in Additional file 1 that were reviewed as a part of this study. In many cases, a single study spanned a fairly large geographic range, thus the location of the author's name and year are centrally located among multiple fires. Some wildfires were approximated either because a map of wildfires samples was not provided in the original publication or the names of the wildfires were missing source can be highly correlated ). In several studies, tree regeneration was not significantly correlated with burn severity, but tree regeneration did vary with distance to seed source (Coop and Schoettle 2009;Rother and Veblen 2016). Most studies examined here found either no relationship between regeneration and burn severity or a significant decline in regeneration at high-severity burned sites. In contrast, Coop et al. (2010) and Shive et al. (2013) found the highest tree regeneration densities at high-severity burned plots, although distance to a living tree was still important.
Numerous additional studies stated that distance to a living tree was measured but did not report if this metric was significant in predicting regeneration. For example, Ouzts et al. (2015) recorded distance to the nearest seed source up to 36 m away but no farther, and Harvey et al. Table 1 The US papers from 2005 to 2018 chosen for our literature review that measured and discussed the effect of distance to seed source on tree regeneration following wildfires. Papers, region, signficant relationship, and distance are described. A plus sign (+) indicates a positive relationship between distance to seed source and tree regeneration, a minus sign (−) indicates a negative relationship, and a zero (0) indicates no relationship. The major findings describe the distance at which tree regeneration begins to decline, and the distance beyond which no regeneration was detected in parentheses, if identified in individual papers. All distances are in meters. In some cases, a distance was not specified, although a significant relationship was detected. Additionally in some cases, different metrics or descriptions were used and are included here (2013) included distance to seed source in a regression tree, but neither study reported these results.  stated the range of distance to seed sources measured, but did not present analysis on this variable. Similarly, Savage and Mast (2005), Shive et al. (2013), and Morgan et al. (2014) stated that patterns of tree regeneration were influenced by distance to seed source, but results of analysis on this factor were not presented. Only six studies explicitly analyzed climate as a factor in post-fire tree regeneration. The significant climate metrics included moisture deficit of some kind as the most common influence on regeneration, followed by either annual or seasonal precipitation (Table 3). While all studies identified here found that water stress influenced regeneration, one study demonstrated a relationship to degree days for one species (Urza and Sibold 2017). , , and Dodson and Root (2013) discussed potential site climates through topographic proxies and metrics like heat load index, forest type, or elevation-precipitation-temperature gradients, but did not analyze climate directly.
Repeated disturbances, topography, and competing vegetation also influenced tree regeneration. In the five studies that examined tree seedling response to repeated wildfires, short intervals between wildfires (1 to 30 years) resulted in a decline in post-fire tree regeneration density compared to once-burned areas ( Table 4). The four studies on the effect of bark beetle outbreaks and wildfires were less consistent across species and studies. Topographic factors including elevation, aspect, and slope, were common predictors in many of the studies assessed here, as were potential competition of shrubs, non-native grasses, and resprouting deciduous trees. At least 27 studies presented in this review found one of these factors to be significant (Additional file 1).

Discussion
Fires are catalysts for forest change Several factors dominated in explaining tree regeneration patterns post fire. First, seed supply is required for seedling establishment unless planted; thus, proximity to living trees was important across many of the tree species in the studies presented here, including species that are serotinous, animal dispersed, and wind or gravity dispersed. Tree regeneration density decreased at distances of 40 to 400 meters from a living mature tree, Table 2 Literature that tested the effect of severity on tree regeneration density. Region, severity at which lower regeneration density was observed, and the major findings in relation to severity and its relationship to tree regeneration especially for ponderosa pine, Douglas-fir, and true firs across the western US. However, several studies of often animal-dispersed species such as whitebark pine and limber pine also experienced a decline in density at longer distances from seed trees. Both Coop and Schoettle (2009) and Leirfallom et al. (2015) found tree seedling densities declined with increasing distance from mature trees unaffected by white pine blister rust (Cronartium ribicola J.C. Fisch), although this decline was more gradual than for species that only relied on wind or gravity dispersal. Studies that examined the relationship between lodgepole pine seedling density and distance to a living lodgepole pine found variable relationships.  In this one case, the influence of climate was analyzed separately in two different climatic periods and is described instead of individual species influenced Table 4 Literature that discussed the impact of interacting disturbances on tree regeneration. Studies, forest type, and described interacting disturbances identified. Fire-fire indicates that a study examined two wildfires that reburned the same area; bark beetlefire describes studies that examined bark beetle outbreaks that preceded a wildfire. The influence of disturbances column explains the differences observed between areas that only experienced a wildfire and those that experienced either a previous bark beetle outbreak or a previous wildfire between distance to live seed source and relative abundance of lodgepole pine seedlings. However, this could be explained by the proximity to nearby burned, serotinous individuals and lack of other species regenerating at far distances from living trees, rather than a relationship to live lodgepole pine trees. To some degree, the lack of consistency in the importance of proximity of a burned location to live lodgepole pines may be due to the proportion of lodgepole pines that are serotinous, which was demonstrated through the differing predictive factors of serotinous versus non-serotinous lodgepole pine by . In this case, non-serotinous lodgepole pine was negatively correlated to distance while serotinous lodgepole pine was not   Bonnet et al. (2005) in the Black Hills, which only observed ponderosa pine regeneration. A burned site's distance to living trees to living trees is also simply a proxy for explaining the availability of seeds. However, the assumption that living nearby trees equals adequate available seed is not always accurate. For example, Leirfallom et al. (2015) demonstrated that it was the proximity to healthy, non-blister-rust affected, whitebark pine trees that was important, not just proximity to simply living trees. Similarly, ponderosa pine trees in the southwestern US and in the southern Rocky Mountains have episodic regeneration events (Savage et al. 1996). Thus, proximity to a living, mature tree does not always guarantee seed availability. More research is needed to identify locations where the lack of tree regeneration is due to lack of seed production or seed viability instead of proximity to trees.
Second, changing climatic conditions are influencing regeneration densities, as climate has for centuries. With the increasingly warm springs and summers in recent decades throughout the western US (IPCC 2013), conditions for seedling survival are changing. On the driest sites, even a small increase in water deficit could negatively influence tree regeneration (Stevens-Rumann et al. 2018). On colder, more mesic sites, these changing climatic conditions could promote regeneration where previously limited by cold or snow (Stevens-Rumann et al. 2018). Water deficit, low precipitation, or drought reduced regeneration success in all studies. The correlation between increased regeneration and available moisture is supported by Petrie et al. (2017), who found this to be an important predictor of seedling success in greenhouse experiments. Additionally, in a multi-century analysis of tree establishment, Brown and Wu (2005) found strong links between seedling establishment and moister climatic conditions. However, in the case of Brown and Wu (2005), the pattern of establishment during cooler and wetter climatic periods correlated with periods of decreased fire activity; thus, the growth of individual trees to a fire-resistant age between wildfires may contribute to the observed cohort of tree establishment, as others have found (Meunier et al. 2014). Modeling exercises similarly found correlations between tree seedling response and both moisture gradients and temperature following wildfires (Hansen et al. 2018). Thus, changes in precipitation patterns may play a large role in the vulnerability of regenerating trees in a warmer climate, and abnormally moist years following wildfires may allow for germination and establishment in otherwise non-regenerating sites. The relationship between climate and regeneration is an aspect of post-fire seedling establishment that warrants extensive additional research.
Individual tree species respond differently to environmental stressors due to their varying environmental requirements and species characteristics , Davis et al. 2019. In dry forests dominated by ponderosa pine and Douglas-fir trees, ponderosa pine appears to be more susceptible to regeneration failure (Davis et al. 2018), but current observed patterns could also be related to the longer seed dispersal distances for Douglas-fir. However, these studies focus on lower elevational ranges of these species; thus, it is important to consider how species ranges may expand as higher elevations and higher latitudes become more favorable to tree growth (Lenoir et al. 2008). In this review, only three studies explicitly tested for species shifts following wildfires (Buma and Wessman 2011, Kulakowski et al. 2013, and only one examined potential climate-induced range shifts following wildfires . Understanding how species ranges may expand, in addition to contract, will be critically important for forest management of burned areas in the coming decades. As discussed previously, most studies presented here combined species for analyses even though it is well understood that species respond differently to environmental and climatic conditions. This is further demonstrated by climate envelope modeling (e.g., Rehfeldt et al. 2014), which suggests that different species will have variable responses to climate change. Rehfeldt et al. (2014) predicted a 50% decline in the range of ponderosa pine by 2060. Alternatively, in higher elevation forests, most observed tree regeneration studies demonstrated consistent and often abundant regeneration of lodgepole pine, even where there were few seedlings of other tree species (e.g., . While documented declines in lodgepole pine regeneration have not been observed, there is modeling evidence that lodgepole pine may see substantial declines in the coming decades due to both an increase in fire frequency (Westerling et al. 2011) and continued changes in climatic conditions (Coops and Waring 2011). All of this speaks to the large degree of uncertainty around individual species responses to climate variability following wildfires.
Third, multiple disturbances such as bark beetles and fire and repeated fire are playing an increasingly important role in tree establishment. As the extent of wildfires increases, in part due to climate (Westerling et al. 2011;Dennison et al. 2014), so will the area repeatedly burned (Prichard et al. 2017). Similarly, interactions between wildfires, drought, insects, and pathogens are expected to increase (Hicke et al. 2016). Repeated wildfires or the combination of bark beetle infestation and wildfire may influence ecosystem transitions. Repeated high-severity wildfires in short succession may be precipitating vegetation changes (Table 4). This change is likely due to the repeated disturbances themselves, not increasing distance to seed source, as at least in several studies, distance to seed source did not vary between once burned and repeatedly disturbed areas (e.g., Stevens-Rumann and Morgan 2016). However, repeated wildfires at low to moderate severity that result in the survival of at least some overstory trees, especially in low elevation forests that would have historically burned at more frequent intervals, may be maintaining more open forests rather than precipitating a transition to non-forested ecosystems (Larson et al. 2013;Morgan 2016, Walker et al. 2018). In studies of tree seedling response after bark beetle outbreaks and wildfires, the findings are less consistent and these two disturbances, in many cases, do not seem to be decreasing tree regeneration (Harvey et al. 2014a(Harvey et al. , 2014bStevens-Rumann et al. 2015). Conversely, Harvey et al. (2013) did find significant declines in tree regeneration follow bark beetle outbreaks and high-severity wildfires in Douglas-fir-dominated systems. Given the relatively small number of studies on repeated disturbances and lack of agreement among studies, future research is needed, especially in the face of increasingly common overlap of disturbances in both time and space. Fourth, tree regeneration was less successful post fire in particular site conditions, with the most common influences being elevation, slope, aspect, and competing vegetation. Steeper slopes and more westerly or southerly aspects often resulted in lower regeneration density compared to shallower slopes and northerly and easterly aspects (Lydersen and North 2012, Ziegler et al. 2017). In some locations, highly competitive and potentially flammable non-forest vegetation may result in positive feedbacks by burning readily and thus continuing dominance by non-forest vegetation (Wilson andAgnew 1992, Donato et al. 2009). In many locations, forests are being replaced by non-forest vegetation, but the replacing vegetation varies. In California, the primary concern, especially in drier forest sites at low elevations, is that the chaparral shrubland vegetation replacing forests is highly flammable (e.g., Collins and Roller 2013). In the US northern Rocky Mountains, forests are more commonly replaced by grass or shrubs (e.g., , while in the southwestern US and some places in the southern Rocky Mountains, conifers are often being replaced by resprouting trees or shrubs (e.g., McGarigal 2010, Roccaforte et al. 2012). All these potential ecosystem transitions to grasslands or shrublands should be assessed with an understanding of the historical heterogeneity. In some places, wildfires may be restoring the historical vegetation structure where tree establishment only occurred as a result of fire suppression (Hessburg et al. 2005, Nagel andTaylor 2005). Thus, there is a need to identify where the lack of regeneration is creating new and novel conditions on a site versus where little or no regeneration is promoting historical heterogeneity. More research is needed in this area to promote sound management intervention in areas of concern.
Understanding the relative importance of the variables presented here will be critical for adaptive fire management in the future. Climate may drive changes in some forested systems, while the disturbance, topography, or competing vegetation may drive changes in other forests. Scientists and managers are implementing experiments to understand the contributing factors (e.g., Tercero-Bucardo et al. 2007, Rother et al. 2015, Petrie et al. 2017, but more analysis of species-specific vulnerabilities is needed to fully understand responses to climate, landscape settings, and biological interactions.

Limitations and research needs
Comprehensive understanding of the drivers of natural regeneration and thus areas where true regeneration "failures" occur is limited by three prominent limitations that warrant further research: (1) understanding spatial variability across burned landscapes, (2) quantifying temporal variability of regeneration patterns, and (3) identifying the controlling mechanisms of regeneration success. First, while tree seedling density may be low in many burned areas, it is highly variable spatially and, as such, researchers should be careful to conclude that large, high-severity burned patches are not regenerating at all without sampling larger plots to characterize spatial variability of fire effects and tree regeneration. The plots used to sample tree regeneration were small across most studies (generally less than 600 m 2 ). However, in studies using plot sizes of 4 ha, tree seedlings were always detected (Owen et al. 2017). Fully understanding spatial and temporal variability of fire effects and vegetation post fire is important to ecosystem function, yet we often do not capture this with a smaller plot size and short periods of study.
Second, time since fire has long been a concern for making inferences about ecosystem trajectory and recovery. Some long-term studies suggested continued recruitment through decades to even centuries following wildfires (MacKenzie et al. 2004, Tepley et al. 2013, Freund et al. 2014. While many of the sites studied here may continue to see tree regeneration in the coming decades, short-term tree regeneration patterns are often highly correlated with long-term regeneration patterns (e.g., Coop et al. 2010). Thus, the patterns of regeneration in the first few years post fire will likely influence the trajectory of that ecosystem . Most of the studies only presented data for <10 years post fire, with some studies as short as 1-2 years post-fire (e.g., Strom andFule 2007, Larson et al. 2013), although some studies presented did span a large time since fire, of 25-64 years post fire (e.g., Passovoy andFule 2006, Haire andMcGarigal 2010). Further, past regeneration patterns may be very different from future patterns given the increasingly unfavorable climate for tree regeneration, especially at lower elevational ranges. Thus, discussion of persistent non-forest vegetation shifts is more prevalent and the level of uncertainty around continued gradual regeneration through time is an ongoing concern. That said, some of these vegetation shifts may be offset by one or several consecutive moist or cool and moist years that could promote tree establishment and growth, even in a warmer climate (Serra-Diaz et al. 2018), thus there is need for additional research into the mechanisms controlling tree germination and regeneration.
Finally, the processes driving seedling establishment and survival in burned areas specifically may vary from unburned areas and greenhouse experiments (Petrie et al. 2017). The interaction among all potential influencing drivers in post-fire environments are complex, and measurements like distance of a site to living trees or fine resolution climate variables may not adequately explain the processes involved in germination and survival. Some alternative explanations for a lack of regeneration that were not considered here but could be important include competing vegetation, other pre-fire disturbances not often described (e.g., Buma andWessman 2011, Leirfallom et al. 2015), highly variable cone production, seed viability that is linked to climate and many other factors (e.g., Buechling et al. 2016), and slow regeneration rates that may be natural for some species and harsh sites. Authors of studies presented here often attributed slow regeneration to changing climate, but there are relatively few studies on physiology of cones, seeds, and seedlings in these settings. To truly determine how a lack of regeneration in the short term post fire corresponds to long-term persistence of non-forest communities, further considerations of these mechanistic influences, temporal variability, and spatial variability are necessary.

Management in a world of more fires and less tree regeneration
Multiple management strategies can be employed to help forests adapt to more fires and more burned area in a changing climate. Specifically, we propose several management strategies to promote tree survival during fires and higher likelihood of tree regeneration success, especially of tree species that are not serotinous. We generally break these strategies into two categories that correspond to two of the dominant drivers of tree regeneration presented here: (1) how to adapt post-fire management strategies, such as planting, to changing climate, and (2) how to mitigate large high-severity burn patches and, thus, long distances to living tree seed sources.

Planting in a changing climate
Strategies for managing landscapes after wildfire must consider climate suitability now and in the future. The many factors interacting to influence seedling germination and establishment warrant consideration prior to post-fire management. We made a decision tree (Fig. 2) based on the literature presented here and in consultation with managers. This decision tree framework can be used to guide planting and other management considerations within wildfire perimeters. We propose that, along with considerations of the forest plans, site objectives, and funding availability, current and future precipitation quantity and patterns and temperature are carefully evaluated along with forest type that can indicate general climatic conditions. While science may suggest that climate consideration or forest type should be the first factor in decision making, we acknowledge that management of burned landscapes is often influenced by more than science; thus, our decision tree begins once an area has been prioritized for potential reforestation. In some cases, forest plans and general guidelines for site objectives should be refined to incorporate climate change more effectively, although this is beyond the scope of our decision tree.
As climate becomes increasingly unfavorable to tree establishment for many species within portions of their current distribution, we must identify areas that may be favorable for target species, given the current and future climate. As multiple studies presented here demonstrated, the climate on some sites may not be currently, or in the near future, compatible with the establishment and survival of those tree species present pre fire (e.g., Davis et al. 2019). Planting trees should be limited to prime locations for tree regeneration success, now and in the future climate, that are not expected to regenerate without management intervention. Thus, promoting productive and diverse non-forested ecosystems or new, novel forested ecosystems may be the best course of action in some locations. For example, many studies showed that those areas at the lowest elevations, or often in the hottest and driest forests were regenerating poorly (e.g., Dodson andRoot 2013, Donato et al. 2016;Stevens-Rumann et al. 2018), thus it may be beneficial to identify whether these sites could continue to support similar species assemblages before planting, and perhaps prioritize those sites that are slightly cooler and wetter over the harshest sites. Alternatively, planting diverse tree species or a single species from a genetically diverse pool may help overcome some of the site productivity limitations, especially if current individuals are no longer suitable but a forest ecosystem is still desired.
Our second broad-scale, science-based consideration is proximity to a seed source. As the literature presented here demonstrated, there are multiple locations where tree regeneration is likely to be abundant, such as within small high-severity burned patches or within 50 to 100 m of living trees; thus, these areas may not need to be replanted. Conversely, planting may be necessary to reforest in large, high-severity burned patches with long distances to surviving seed-source trees. Distance to living trees was an important predictor of post-fire tree seedling density; thus, should be altered by region and management unit and informed by local experience and monitoring. Aspect indicates the aspect that a hillside is facing: S-W indicates south-through west-facing slopes; N-E indicates north-through east-facing slopes. The triangle at the top indicates the gradient of forest types and climatic conditions that vary between the hottest and driest locations and the coolest and wettest, with increasing favorability as one moves to cooler and moister conditions. Dashed lines indicate a need to assess all locations for more localized important factors listed to the right after more broad scale assessments, prior to planting understanding the shape and size of high-severity burned patches is critical in understanding the potential for tree regeneration and when management intervention may be desired (Shive et al. 2018). Mapping high-severity burned patches through satellite imagery can begin to address which areas are large enough to warrant concern, and where competing vegetation may pose a threat to delayed planting success. In many cases, no planting may be recommended due to the size and shape of those large patches, even within large fires. Even in large patches, the irregular shape and unburned islands leave much burned area close to surviving trees that can be potential seed sources. For instance,  found that, on 21 large fires, >75% of the burned area was within 95 m of surviving trees. Thus, planting efforts should be focused on those largest, high-severity burned patches in which large areas within the patches are more than 100 to 400 m, depending on region and species, from a lesser burned edge or unburned island of substantive size (North et al. 2019).
We propose not planting trees close to the edge of high-severity burned patches to avoid areas that will likely experience high fuel accumulations through time (e.g., Roccaforte et al. 2012; Eskelson and Monleon 2018) and thus experience higher potential for reburning (Prichard et al. 2017). Prior fires can limit fire extent or burn severity for~5 to 20 years, depending on region, forest type, and individual fire events (Parks et al. 2015;Parks et al. 2016b). Thus, providing a buffer and potential containment area within a high-severity burn perimeter will decrease the likelihood of burning planted seedlings. With the increasing likelihood of reburns, managers should be conscious of where and how subsequent fires may burn and how tree seedlings may contribute to the fuel loading. We propose planting only within the interior of large, high-severity burned patches to avoid loss of money and effort invested in planting tree seedlings.
After these first two broad-scale considerations of climate and seed source availability, we focus on those finer-scale topographic or topo-climate variables that are important for regeneration success across the western US. Elevation, aspect, and slope all play roles in identifying suitable locations for planting. We provide general guidelines for these different metrics, but local knowledge and adaptations will be necessary. For example, , , and many others demonstrated that less regeneration was naturally occurring at the lowest elevation sites on south-to southwest-facing slopes, thus indicating that these sites may be outside of their climatic tolerance. Conversely, a south-facing slope at a high elevation location could promote regeneration at the upper elevational range of a particular species. Slope may be more important in combination with soil properties. Shallow soils on steep slopes may not promote regeneration establishment, but in deep, rich soils, perhaps the slope percentage is of less concern.
Finally, multiple studies found local site conditions to be important and should always be considered prior to planting. Identifying if and where competing vegetation may either deter natural regeneration or compete against planted seedlings may alter planting timelines. For instance, promoting immediate post-fire planting in areas of fast-growing, resprouting species may be more important in some areas, while in other areas where competing vegetation is less of a concern, allowing a couple of years to observe natural regeneration densities may be warranted before planting. Soil types and soil depths may influence the density or species planted, and a plethora of microsite conditions, such as the presence of nurse structures, may be important considerations (Haffey et al. 2018). All of these are very site specific and may even vary across a treatment unit, thus requiring manager discretion, drawing on local experience.
Using fire to promote natural regeneration Hessburg et al. (2015) outlined seven principles for managing landscapes strategically. They emphasized the need for thinking about where to do what treatments, including doing nothing in some locations. Increasing landscape-level treatments, including thinning, fuels treatments, prescribed fire, or managed wildfires, could curtail the area that burns under extreme conditions, and thus increase the potential for survival of trees that could provide seed sources for regeneration. Currently, multiple researchers and managers are suggesting increasing strategic planning pre fire that focuses on limiting risk to firefighters, managing the very high costs of fire suppression, and developing a preemptive strategic plan for managing wildfires Calkin 2011, Schoennagel et al. 2017). Managing fires often includes aggressive fire suppression that will likely continue when human values are at risk. However, in places where risk to human values is low and ecological need is high, we join others in advocating for less aggressive suppression actions (Schoennagel et al. 2017, Halofsky et al. 2018. Likely, this means delaying, herding, and otherwise managing ongoing fires to foster survival of seed source trees and desirable patch size distribution (Hessburg et al. 2015). The National Cohesive Fire Management Strategy (USDA and USDI 2018), which is supported by federal, state, and other managers, calls for fire suppression when possible and desired, as well as using fire as part of natural resource management. Part of adapting to a future of more frequent and larger fires will be appreciating when wildfires are doing "good work" to further vegetation management goals.
With respect to post-fire tree regeneration, the broader goals of these managed wildfires include fostering survival of more adult trees to rain seed through time, and creating less severe, patchier fire effects. Both effects are more likely when fires burn under less extreme conditions-the very conditions in which fire suppression is most effective. Increasing the proportion of "fire refugia," unburned or low-severity burned sites, could provide more ecological services (Kolden et al. 2012, Meddens et al. 2018, and while the promotion of fire refugia is important under any burning condition, less extreme weather conditions promote the creation and maintenance of "fire refugia" (Krawchuk et al. 2016). Some forest openings, especially when smaller in size, may be ecologically beneficial and increase the heterogeneity of landscapes.
Ultimately, these approaches blend together for, in most areas, fires will occur in the future and what we do post fire will affect the forest conditions prior to the next fire and how that next fire will burn. Decisions need to be strategic for, in years of widespread fire, there are more areas to plant than there are trees, and there are many sites that will likely not support similar forests to those found pre fire. All management strategies, including no action, have consequences that can be evaluated to inform management decisions.

Conclusions
Management actions that could offset the effects of low tree regeneration densities in the first years following a wildfire on some sites include prescribed burning and managing fires to burn more area under less extreme conditions to favor more seed source trees surviving, and planting post fire in locations where seedlings are most likely to survive in terms of microsite and site conditions and risk of future fires. Triage among sites is needed to strategically differentiate those sites for which management actions are most likely to be worth the effort and cost. For instance, post-fire tree seedling densities are high on many mesic sites, and on many sites close to seed source trees where regeneration may eventually occur, while others are so warm and dry that even with intensive effort, neither planted nor naturally regenerated trees are likely to survive (Stevens-Rumann et al. 2018). Baker (2018) aptly points out that some sites will regenerate slowly without intervention and management actions may not always be warranted. Fire suppression has, in many ecosystems across the western US, increased tree density, forest continuity, and homogeneity, and allowed encroachment of many trees into non-forested ecosystems (e.g., Gruell 2001, Nagel andTaylor 2005). As a result, some loss of forest area may restore historical landscape heterogeneity of different ecosystem types. Thus, while we have focused on the lack of tree regeneration and possible pathways for restoring forest ecosystems, in some cases promoting a non-forested ecosystem may be recommended even if the site is climatically suitable for forests.
While reactive treatments will no doubt occur, proactive management informed by assessment of forest vulnerability is sorely needed. Such an approach could mean that we prescribe more fires and manage more wildfires to foster future resilience (Schoennagel et al. 2017). However, with the projection for larger fires and longer fire seasons, vulnerability of forests for which disturbance-induced tree mortality overlaps with and contributes to lack of tree regeneration, we will also have to accept that some forests will be replaced by shrublands, woodlands, and grasslands as we adapt to a future with more fire.

Additional file
Additional file 1: All US studies from 2005 to 2018 that were included in the literature review, showing region, time since fire, and significant variables found to influence tree regeneration. (DOCX 22 kb)