BACKGROUND

At alpine treeline ecotones in the Rocky Mountains, closed forest dominated by subalpine fir (Abies lasiocarpa) and Engelmann spruce (Picea engelmannii), often break up into patches before giving way to tundra vegetation (Allen and Walsh, 1996; Baker et al., 1995), a pattern that is common for ecotones generally (Gosz, 1993). In windswept locations, these patches may consist of elongated islands of krummholz, oriented parallel to the prevailing wind. Larger linear strips composed of both krummholz and dwarfed upright trees have been described as hedges (Holtmeier, 1982). Although many different terms have been used to classify alpine treeline ecotones, “timberline” usually refers to the upper limit of continuous, closed forest, “treeline” or “tree limit” are used to indicate the highest elevation of patches in which trees still maintain an upright, tree-like growth form, and “species limit” refers to the upper limit of krummholz growth forms (Wardle, 1974; Hustich, 1979; Hansen-Bristow and Ives, 1985; Slatyer and Noble, 1992; Kullman, 2002).

The ecotone on Lee Ridge contains linear patches of forest that differ from typical krummholz islands and hedges in several respects. First, they are dominated by lodgepole pine (Pinus contorta var. latifolia) rather than subalpine fir or Engelmann spruce. Second, many of the patches actually extend like “fingers” from closed forest into tundra. Finally, while some of the fingers contain krummholz and dwarfed trees, they primarily consist of separate individuals that have established from seed rather than by layering, and that are near the same height as those in the closed forest below.

The fingers are most clearly defined in the upper portion of the tree limit zone of Lee Ridge, but they are also visible at lower elevations, where the tree limit grades into the timberline zone (Fig. 1). These lower fingers extend to the top of the ridge and contain trees that are generally taller and more upright than in the upper fingers. Interestingly, however, several of the largest individuals have a pronounced upslope basal sweep, some growing along the ground in the direction of the prevailing wind for as much as a few meters before turning upright. Many of the trees exhibiting basal sweep also have a krummholz base (cf. Earle, 1993) characterized by a dense array of dead lower branches. These observations suggest that forests at this site formerly contained a mix of krummholz and dwarfed trees that have been able to change their growth form over time. Moreover, it suggests that the lower fingers were formerly similar in appearance to the upper fingers, and thus may provide insight into how the upper portion of the ecotone on Lee Ridge may change in the future.

The presence of such unusual patterns on Lee Ridge raises questions regarding their origin, as well as their potential effects on environmental processes. The concept of positive feedback switches (Wilson and Agnew, 1992) illustrates the potential for pattern to affect processes at ecotones. Positive feedbacks involve the modification of the environment by a community (used here to denote a general vegetation state that may include only one species), which makes the environment more suitable for that community. The term “switch” is used for these processes because they can switch the development of vegetation from one pathway to another, thus producing alternative stable states (e.g., Zimov et al., 1995).

It is important to note that “positive feedback” does not necessarily imply a “positive interaction” between communities (e.g., Bertness and Callaway, 1994). A community may have a beneficial or detrimental impact on itself or an adjacent community, depending on the type of feedback involved. Four types of switches have been identified, resulting from different combinations of one or both communities modifying an environmental factor in patches where they are or are not present. Each of these switches may produce one or more of the following vegetation conditions: (1) a stable vegetation mosaic, where the two communities persist in different patches over time; (2) a sharpening of the boundary between communities across a gradual environmental gradient; (3) delayed vegetation change; or (4) accelerated vegetation change (Wilson and Agnew, 1992).

The consistent windward to leeward alignment of linear forest patches suggests the importance of directional, wind-mediated switches in producing the pattern. Moreover, Alftine and Malanson (2004) found that spatial patterns of the treeline in the upper portion of Lee Ridge could be approximated in a cellular automaton if such directional positive feedback was included. Although wind is involved in several processes affecting tree growth, the deposition of wind-blown snow as a result of trees reducing wind speeds has been shown to play a particularly important role in treeline patterns and dynamics. Walsh et al. (1994) found that an intermediate amount of snow is required to allow tree growth to occur at treeline locations in Glacier National Park. Too much snow can cause direct, mechanical damage (e.g., Kullman, 1989; Homma, 1997), or indirect effects such as a reduced growing season or desiccation due to frozen soil (Billings, 1969; Hansen and Klikoff, 1972; DeLucia, 1986). Conversely, snow has been shown to be important in protecting plants from damaging or desiccating winds, along with blowing snow and ice (Tranquillini, 1979; Frey, 1983; Hadley and Smith, 1983, 1986), and snowdrifts provide additional moisture to plants after peak stream flow in the spring and summer (Berg, 1987). Finally, snowdrifts increase the input (Bowman, 1992) and availability (Williams et al., 1998) of nutrients through atmospheric deposition and increased soil temperatures, respectively. Thus, in arid treeline sites with a moderate level of snow deposition, the leeward environment of existing patches may be enhanced for seedling establishment and survival. Moreover, if the feedback is sufficiently strong, the switch may produce an acceleration of vegetation change.

The potential for feedback to accelerate vegetation change has important implications for the sensitivity of ecotones to climate change. Because ecotones often represent the distributional limits of the communities they divide, it has been suggested that changes in their positions may be useful in detecting climate change (e.g., Turner et al., 1988; Neilson, 1993; Risser, 1995), but others have argued that the movement of most ecotones is too slow to make them useful as climate change indicators (e.g., Noble, 1993; Kupfer and Cairns, 1996; Cairns and Malanson, 1997). However, ecotones that are controlled by positive feedback may respond more rapidly to changes in climate.

In this paper, I evaluate the effects of directional positive feedback on the pattern and dynamics of the ecotone on Lee Ridge by reconstructing the development of the lower elevation fingers. I use dendrochronology to reconstruct and map spatio-temporal patterns of forest advancement, in order to determine whether positive feedback between forest pattern and tree establishment can explain the observed patterns, and can produce an acceleration of vegetation change. I also evaluate the relative importance of feedback and climatic controls on the ecotone.

FOREST ADVANCEMENT PATTERNS AND RATES

I determined tree ages by preparing and cross-dating the tree cores using standard dendrochronological techniques, including ring width (Stokes and Smiley, 1968), latewood width and density (Filion et al., 1986), and frost rings (LaMarche, 1970). I used geometric approximation to correct samples that did not include the pith. I then determined tree ages, and mapped tree locations and establishment dates in 10-yr age classes.

To determine whether the fingers represent advancement of the forest limit on Lee Ridge, I examined the relationship between the distance from the base of a transect and tree age. Advancement should be identifiable as a negative relationship between these variables. However, infilling at the base of a finger would weaken the relationship without representing a reduction in advancement. A scatter plot of age and distance in this case would appear as a lopsided cloud of points, with a negatively sloped edge. Ecological phenomena frequently show such “factor-ceiling” distributions, because a given ecological factor may act as a constraint on the upper bounds of organism responses, but other (often unmeasured) factors may control the lower bounds (Thomson et al., 1996). Changes near the center of these distributions may be statistically indistinguishable from zero, while changes near the extremes of distributions are nonzero. In such situations, traditional correlation analyses are unable to convey the ecologically meaningful information contained in the edge of the distribution.

Quantile regression has been used for several applications in ecology (see Cade and Noon, 2003, for a review) to estimate rates of change along or near the upper boundary of response distributions. In this technique, regression lines are fit to different quantiles of the cumulative distribution function so that a certain proportion, τ, of the sample observations are less than or equal to the estimate. For example, the 50th (percentage) quantile represents the median of the distribution, such that 50% of the observations are less and 50% greater than the estimate; for a 60th quantile estimate, 60% of the observations are less and 40% greater, and so forth. Although quantiles at or near the upper boundary of factor-ceiling distributions will best represent a limiting relation, Cade et al. (1999) suggest testing a range of quantiles in order to maximize the trade-off between increasing response and loss of precision in the estimate associated with wider confidence intervals at extreme quantiles. I used the 20th, 40th, 60th, and 80th quantiles to determine whether the slope of the relationship between distance from the base of a transect and tree age was significantly different from zero, and whether this relationship changed at different levels of the response distribution.

To determine whether an acceleration of vegetation change has occurred at the study site, I calculated advancement rates at two scales. At the transect scale, I determined the distance from the base to the top of each transect, and divided this value by the number of years between the establishment of the first tree and the furthest upslope tree in the transect. In addition, because the purpose of this study was to examine the finer-scale feedback between existing trees and subsequent tree establishment, I also calculated advancement based on the establishment of new trees within 5 m upslope of existing trees. Any tree that established upslope of the 5-m limit of an existing tree was considered to provide a separate base for calculating advancement in subsequent years. Trees that established downslope of an existing tree were considered to represent infilling, and were not used to calculate advancement.

ADVANCEMENT PATTERNS

The quantile regression analyses showed negative slopes for all quantiles in all transects (Table 2, Fig. 3), and slopes were steepest for the 80th quantile estimate in four of the five transects (1, 2, 4, and 5; 60th quantile for transect 3). Slopes were significantly (P < 0.05 or less) different from zero in transects 1, 2, 3, and 4, and the significantly different slopes occurred most frequently in the 80th quantile.

Several similarities in spatial patterns of advancement may be identified among the transects. For example, most of the oldest trees established at or near the base of the fingers. Although a general trend of decreasing tree age leeward of the oldest trees is evident, some of the transects (particularly 3 and 4) show a repeating windward to leeward pattern of old trees followed by successively younger trees. In most cases, each repetition began leeward of boulders, and the locations of the oldest trees coincided generally with groups of dead trees.

Temporal patterns of establishment also varied among the transects (Fig. 4). Mean tree ages between transects were significantly different (t-tests, P < 0.05) for all pairs except transects 2 and 5, and transects 3 and 4. Several trees established during the 1700s in transect 1, while none to a few established in the other transects. In addition, peaks of establishment occurred during the 1940s or 1950s in transects 2, 3, and 4, while the peaks for transects 1 (1880s) and 5 (1900s) occurred earlier. Similarities among the transects include higher overall establishment during the 1900s compared to the previous two centuries, and low establishment during the 1910s. Trees in all transects were significantly older (t-tests, P < 0.001) than those in the cove.

ADVANCEMENT RATES

Patch-scale advancement rates were similar among the transects, while transect-scale rates were more variable (Table 3). Transects 3 and 1 had the highest overall rates, each greater than 0.5 m/yr. Transect 4 followed closely at 0.48 m/yr, and transects 5 and 2 had the lowest rates at 0.32 and 0.28 m/yr, respectively. Rates at the patch scale fell in the same order by transect, but they were 2–4 times smaller than those at the transect scale.

TRANSECT CHARACTERISTICS

The similarities in the dimensions of the transects suggest that similar processes were responsible for the development of the fingers. For example, the fact that all of the transects were oriented generally west to east suggests the importance of the dominant westerly winds in controlling the ecotone patterns. The similarity in width among the transects, combined with the fact that tree ages were consistently older in the transects compared to the sampled cove, suggests that these winds may be funneled around existing trees at high speeds, thus preventing expansion in the intervening areas, as suggested by Wilson and Agnew's (1992) reaction switch, and as observed in other treeline environments (Scott et al., 1993).

The structural and compositional characteristics of the transects also exhibit some important similarities and differences. For example, lodgepole pine had the highest basal area in all of the transects, and the highest density in all but one. This is unusual for treeline locations in Glacier National Park, which are usually dominated by subalpine fir (Cairns, 2001). Carter and Smith (1988) found that lodgepole pine had greater water-use efficiency then either subalpine fir or Engelmann spruce on open, dry sites at low elevations. The windy, low-elevation (75 m below the park average) treeline conditions on Lee Ridge may therefore give lodgepole pine a competitive advantage. Alternatively, although there was no direct evidence of recent fires near my study site, extensive crown fires may have occurred in the past, killing most trees and increasing the abundance of lodgepole pine seeds relative to other species.

FEEDBACK AND SPATIAL PATTERN

The fact that the slopes of the regression lines for all transects were negative illustrates that advancement has taken place within the study site, and the general increase in slope steepness toward the tops of the response distributions indicates that infilling has also occurred. Moreover, the regression analyses suggest the importance of feedback in the advancement process. In particular, the windward to leeward pattern of successively younger trees suggests that established trees improve the microenvironment for new seedlings. Variations in this expected pattern are not surprising, given the fact that the movement of ecotones is typically heterogeneous over both time and space (e.g., Payette and Filion, 1985; Kullman, 1986; Malanson, 1993; Lloyd and Graumlich, 1997; Payette et al., 2001). Nevertheless, many of the differences among the transects may be explained by the fact that the feedback-induced pattern is superimposed on other, spatially variable factors.

The influence of boulders, for example, is evident in transects 3 and 4, where trees were able to establish tens of meters upslope from the closed forest or from existing patches. Although trees were occasionally able to establish well in advance of existing trees in the other transects, which did not contain boulders, such jumps were generally not as common or as large. The boulders thus appear to have provided safe sites for advanced seedling establishment, as has been suggested in the development of krummholz islands and hedges (Klikoff, 1965; Marr, 1977; Holtmeier, 1982). Therefore, when several boulders are present within a finger, the resulting pattern is a repeating sequence of old trees followed by successively younger trees leeward.

In addition to boulders, differences in microtopography also explain some of the differences in the spatial pattern of advancement among the fingers. Although I did not measure microtopographic variations, relict solifluction terraces cover most of Lee Ridge upslope from my study site (Walsh et al., 2003). The terraces are oriented at acute angles (roughly 20°; see Fig. 1 inset) relative to fingers on this upper slope. In some areas, the terraces appear to be providing shelter from wind, just like the boulders, in which seedlings can successfully establish, and then produce hedges or krummholz islands. The edges of these terraces may also produce turf exfoliation sites (Butler et al., 2004), which are more penetrable and thus more favorable microsites for seedling establishment. Although these terraces are not as evident in my study site, they are visible in a few areas, particularly in the higher elevation transects (1 and 5). This may explain why some of the first trees to establish in these fingers are located near the middle or top of these transects, despite the fact that climatic conditions were presumably harsher compared to the other fingers, and that no boulders were present.

A final factor that could explain variations in spatial patterns among the transects is caching of limber pine seeds by Clark's nutcracker (Nucifraga columbiana). Although limber pine was rare, clusters of saplings were present in the study site, indicating regeneration from seed caches (Casey and Tomback, 1994). Moreover, most of the dead trees identified in the transects were limber pine, which also grew in clusters. It is possible that past establishment of these clusters also allowed the development or extension of fingers.

FEEDBACK MECHANISMS

Direct observations, and snow surveys conducted by the USGS–Biological Resources Division on Lee Ridge, show that snow deposition is enhanced leeward of existing trees. Thus, it is likely that wind-snow feedback plays an important role in the observed patterns. Given the fact that trees within the fingers maintain canopies well above the snowpack level and show little evidence of wind-damaged foliage, it seems most likely that increased spring and summer moisture, rather than protection from winter winds, is the primary feedback mechanism. The importance of soil moisture in moderating the influence of temperature over the position of treeline ecotones has been suggested by other studies (e.g., Weisberg and Baker, 1995; Hessl and Baker, 1997; Lloyd and Graumlich, 1997; Cairns and Malanson, 1998).

The leeward environment of existing trees may also be improved through the increased input and availability of soil nutrients, again tied to snow deposition. Sediment traps placed within and leeward of krummholz patches on Lee Ridge show little deposition (D. R. Butler, personal commun.), but krummholz soils throughout Glacier National Park are generally deeper and more nutrient-rich compared with tundra soils (Malanson and Butler, 1994; Cairns, 1999). At other treeline sites in the Rocky Mountains, migrating krummholz islands have been shown to reduce soil carbon in areas unaffected by enhanced snowpack (Pauker and Seastedt, 1996), but moderate to heavy snowdrifts increased the deposition and cycling of nutrients, particularly carbon and nitrogen (Bowman, 1992; Holtmeier and Broll, 1992; Williams et al., 1998).

Recent studies in the Medicine Bow Mountains in Wyoming have suggested that existing trees may also improve conditions for seedling establishment and survival during the summer by reducing temperatures and intense UV radiation during the day, and increasing soil temperatures at night (Smith et al., 2003; Johnson et al., 2004). However, the intensity of UV radiation is probably less important on Lee Ridge, which is 1000 m lower in elevation than the Medicine Bow sites. Moreover, the increase in night-time soil temperatures around existing trees in the summer should be isotropic, and thus cannot explain the directional advancement of the fingers.

ACCELERATION OF VEGETATION CHANGE

The overall pattern of advancing fingers with intervening coves implies the presence of an acceleration of vegetation change. Where existing trees, boulders, microtopographical variation, or other conditions made initial seedling establishment possible, trees have been able to replace tundra vegetation via positive feedback. In contrast, the coves, where either no feedback is operating, or a reaction switch may actually be delaying the change from tundra to trees, appear to be undergoing a more gradual and more recent invasion by trees. Such a reaction switch would explain why advancement in the coves is taking place more slowly than in the fingers, the opposite pattern than is expected from research on the shape and dynamics of landscape boundaries (e.g., Hardt and Forman, 1989).

The possibility of an acceleration of vegetation change could also be assessed through comparison of advancement rates with other studies of changes in alpine treeline ecotones. Such comparison is difficult, however, because other studies either do not report advancement rates, or base advancement on a different component of the ecotone, such as the establishment of new seedlings and saplings rather than mature trees, or expansion resulting from krummholz stems changing to upright trees. Nevertheless, Kullman (1986) did calculate advancement rates based on newly established tree-size individuals in the Swedish Scandes that are similar to those calculated for the fingers (mean 0.52 m/yr). The overall rates for the fingers are sufficiently high that changes could be detected in the field, or through high-resolution aerial photography or remote sensing.

ESTABLISHMENT PATTERNS AND CLIMATIC CONTEXT

The oldest trees identified in four of the five sample transects established between 1720 and 1750. However, the presence of dead trees, which were also sampled but could not be cross dated, raises questions regarding the timing of tree establishment. It is possible that these dead trees established before the oldest live trees, especially considering that most of the dead trees were limber pine, a common pioneer species (Burns and Honkala, 1990; Rebertus et al., 1991). Nevertheless, the oldest live trees were lodgepole pine rather than limber pine, and some regeneration of limber pine is occurring at the study site. The locations of dead trees, especially if combined with unsampled trees, may, however, explain why some live trees were apparently able to establish more than 5 m upslope of sampled trees. If all trees were dated, it is possible that a more clear sequence of progressively younger trees from west to east would be evident.

Regardless of whether the oldest live trees were the first to establish, the commencement of forest advancement in the study site pre-dates the approximate end of the Little Ice Age in Glacier National Park, in the mid-nineteenth century (Carrara and McGimsey, 1981; Carrara, 1989; Hall and Fagre, 2003). Nevertheless, advancement and increases in numbers of trees were very gradual through the end of the eighteenth century, suggesting that climatic conditions were indeed still very harsh during this time.

It is also interesting to note that more than half of the trees that established during the eighteenth century exhibited basal sweep and/or a krummholz base. In a study of the arctic treeline, Payette et al. (1989) suggested that “creeping spruces” (krummholz) were unable to enhance snow drifting at exposed sites, thus preventing seedling establishment at these sites. However, if krummholz individuals were able to change their growth form, as is suggested by the upright trees with basal sweep at my study site, then they would be able to reduce wind speed sufficiently to deposit snow immediately around and leeward of their stem (e.g., Earle, 1993). Thus, when the first trees in the transects established, feedback effects may have been very small. But as they began to change to a more vertical orientation under improving climatic conditions, feedback, and thus tree establishment and advancement rates, would have been enhanced.

The increase in advancement and numbers of trees during the early to mid-1800s suggests that climatic conditions were becoming more favorable during this period. Nevertheless, almost all establishment and advancement during this period occurred either adjacent to or within 5-m upslope of trees or boulders, suggesting that feedback was necessary for the survival of new trees.

Between 1850 and 1880, in contrast, trees in transects 1, 4, and 5 established more than 5 m upslope of existing trees or boulders. These patterns combined with the doubling or tripling of establishment rates between 1880 and 1900 in transects 1, 3, and 4 suggest that climatic conditions in the latter half of the nineteenth century were becoming sufficiently favorable to reduce the importance of feedback.

Patterns and rates of establishment among the transects were more variable in the first half of the twentieth century. However, rates dropped considerably for all transects between 1910 and 1920, raising questions regarding extreme or anomalous climatic events that may have restricted establishment during this time period. Many samples at my study site recorded a frost ring early in the growing season of 1910. An extreme late frost event may have caused reduced reproduction in subsequent years. Many of the newly established trees during this time represented increases in density rather than advancement, but in several cases they continued to establish beyond 5 m upslope of existing trees. Over the next 20 yr the distances between existing and newly established trees increased even more, and establishment rates either increased or remained the same in all transects.

Establishment generally continued to increase toward the middle of the twentieth century, reaching an overall peak between 1950 and 1960. Most new trees were scattered evenly, indicating that most of this establishment represented infilling. However, trees were concentrated near the top of transect 1, indicating that conditions had improved sufficiently to allow establishment at the highest elevations of the study site.

The decline in establishment rates in most of the transects over the last 30 yr, especially between 1980 and 1990, and the concentration of most new trees near the tops of the transects seem to represent another abrupt change. This may be explained by the fact that seedlings and saplings, which were present in all transects, were not sampled. However, data from seedlings sampled higher on the ridge also indicate an abrupt drop in establishment beginning in 1980, possibly in response to a switch to a positive phase of the Pacific Decadal Oscillation cycle (Alftine et al., 2003).

EFFECTS OF CLIMATE VS. FEEDBACK

The lack of synchroneity in the timing of advancement among the transects, particularly prior to the twentieth century, suggest that local-scale processes, including biotic and abiotic feedback effects, were more important than regional-scale climate in determining pattern and process. However, spatio-temporal patterns of advancement within each transect suggest that thresholds may exist, in which the relative importance of climate and feedback in controlling establishment and advancement rates switches over time, causing differences in the rates. For example, conditions between the eighteenth and early nineteenth centuries were apparently too harsh for much establishment and advancement to occur. Feedback then appears to have played an important role in increases in establishment and advancement during the early to mid-nineteenth century. The dramatic increase in establishment toward the end of the nineteenth century seems to represent another threshold, since advancement occurred over a broader scale, suggesting that continually improving climatic conditions had reduced the importance of feedback. Variation in the strength of feedback in response to climate variability has been identified previously in the facilitation of tree establishment by nurse shrubs in Argentina (Kitzberger et al., 2000), and threshold effects of temperature on tree growth rates have been identified at the treeline in the Alps (Paulsen et al., 1999). These results are also consistent with the prediction that positive interactions between plants will be most common in areas of high physiological stress (e.g., Bertness and Callaway, 1994; Hacker and Gaines, 1997), in that the importance of these interactions seems to have diminished over time as physiological stress decreased.

KRUMMHOLZ ISLANDS, HEDGES, AND FINGERS

The importance of feedback in producing and maintaining krummholz islands and hedges (Marr, 1977; Holtmeier, 1982; Benedict, 1984; Alftine and Malanson, 2004), and the influence of krummholz islands on seedling establishment (Camarero and Gutiérrez, 1999, 2002) have been identified previously. However, only one other study (Minnich, 1984) suggests that feedback may play a role in producing linear patterns of upright forest near treeline. This study mentioned the existence of trees in “row configuration” on Mount San Gorgonio in the San Bernardino Mountains of California, and although the specific cause of this pattern was not investigated, the rows occurred near ridgetops in the most wind-exposed areas. In addition, the orientation of these rows changed according to spatial differences in local storm winds, suggesting that established trees improved conditions for downwind individuals. Interestingly, the dominant tree species in these rows was also lodgepole pine (although var. murrayana). It may be that as a generalist species, lodgepole pine is more tolerant of the dry conditions produced by strong winds and is able to respond well to slight improvements in microclimate. In addition, lodgepole pine does not reproduce vegetatively through layering, as subalpine fir and Engelmann spruce do. This helps to explain why the fingers on Lee Ridge and the rows on Mount San Gorgonio differ from the hedges described by Holtmeier (1982).