Landscape context and long-term tree influences shape the dynamics of forest-meadow ecotones in mountain ecosystems

Forest-meadow ecotones are prominent and dynamic features of mountain ecosystems. Understanding how vegetation changes are shaped by long-term interactions with trees and are mediated by the physical environment is critical to predicting future trends in biological diversity across these landscapes. We examined 26 yr of vegetation change (1983–2009) across 20 forest-meadow ecotones spanning a range of landforms/hydrologies and elevations (montane and subalpine) in the Three Sisters Biosphere Reserve, Oregon (USA). We quantified changes in tree structure (cover, density, and basal area) and in the abundance and diversity of ground-layer vegetation based on species’ habitat associations and growth forms. To explore the contributions of tree structure, landscape context, and initial vegetation to changes in ecotonal communities, we used a combination of NMDS, PCA, and multiple regression. Despite a long history (50–100 yr) of tree invasion, ecotones were still dominated by meadow species in 1983. Ecotones exhibited significant but varying patterns of change over the study period while adjacent forest and meadow habitats remained stable. Despite a significant increase in summer temperature, we found little evidence of a direct influence of climate on ecotonal changes. Declines in total richness, and in the cover and richness of meadow species, were greater where soil moisture was seasonally limiting (montane mesic slopes and subalpine early snowmelt sites). Forest species showed much greater increases in montane than subalpine ecotones; limited colonization of the latter reflects the depauperate nature of subalpine forest understories in this region. Vegetation changes were related to initial tree structure but not to changes in structure over the study period. Past tree invasions, a legacy of both climate variation and disturbance history, continue to exert strong influences on ecotonal ground-layer communities. However, the consequences for local diversity vary across the landscape. Quantifying the nature of this variation through long-term observations is a critical step toward predicting future changes in the biological diversity of these and other mountain ecosystems.


INTRODUCTION
Ecotones are ubiquitous features of most biomes (Gosz 1991) that regulate the movement of organisms, materials, and energy (Fagan et al. 2003, Yarrow andMarin 2007), and can have profound influences on the dynamics of populations and communities (Fagan et al. 1999).They are vegetation zones across which compositional or structural changes are steep relative to neighboring communities (Lloyd et al. 2000), and can reflect underlying gradients or discontinuities in environmental factors, sharp contrasts in plant physiognomy (herbs vs. trees), or past disturbances (van der Maarel 1990, Wilson andAgnew 1992, Cadenasso et al. 2003).Ecotones are often viewed as ''tension zones'', sensitive to changes in climate or other extrinsic factors (van der Maarel 1990).Recent work has focused on detecting and characterizing the spatial structure of ecotones (Fagan et al. 2003, Strayer et al. 2003, Yarrow and Marin 2007).However, major challenges remain in understanding why and how ecotones change over time and what these changes imply for other ecological attributes or functions (e.g., Hufkens et al. 2009).Answers to these questions are fundamental to predicting future changes in vegetation at both local and landscape scales (Cadenasso et al. 2003, Strayer et al. 2003, Hufkens et al. 2009).
Among terrestrial systems, forest-grassland ecotones are among the most striking, dynamic, and widely studied (Archer 1995, Scholes and Archer 1997, Coop and Givnish 2008).Globally, grass-and forb-dominated ecosystems are experiencing encroachment by trees (Van Auken 2000, Bowman et al. 2001, Duarte et al. 2006), spurring concern over the loss of biological diversity and other ecosystem services.Given the time spans necessary to detect change, the ecological effects of encroachment are often inferred retrospectively (Jackson et al. 2002, Briggs et al. 2005, Haugo and Halpern 2007).Direct long-term observations that are critical to documenting and interpreting these changes are rare.
Climatically defined ecotones, where tree establishment or growth is controlled by temperature or precipitation (e.g., alpine/arctic or lower arid treelines), are a common research focus (Allen and Breshears 1998, Bond 2008, Batllori et al. 2009, Harsch et al. 2009, Virtanen et al. 2010).Many forest-grassland ecotones, however, are shaped by other factors such as landform, soils, hydrology, and disturbance (Franklin andHalpern 1999, Gibson 2009).In these contexts, tree establishment and growth are not climatically limited and ecotones can change rapidly.Within the Pacific Northwest, for example, many such ecotones in the montane and subalpine zones have shifted dramatically during the 20th centu-ry (Miller and Halpern 1998).The patterns and causes of these changes have been characterized (Franklin et al. 1971, Rochefort et al. 1994, Woodward et al. 1995, Miller and Halpern 1998, Takaoka and Swanson 2008, Zald 2009), but few studies have addressed the community-level consequences of encroachment either at local (Magee and Antos 1992, Haugo and Halpern 2007, 2010) or landscape scales.
Tree establishment in herb-dominated systems can effect two types of compositional change: competitive exclusion of resident meadow species and facilitation of forest understory species (Duarte et al. 2006, 2007, Haugo and Halpern 2007, 2010).Both processes can be influenced by the structural characteristics of the ecotone (spatial distribution, density, and size of trees) and the rates at which these change.Patterns of species' loss and gain are also likely to vary with landscape context (Jones and Callaway 2007) in response to resource or environmental variation that can affect the outcomes of tree-herb interactions (Bertness and Callaway 1994), or to variation in the sizes or traits of species' pools.Understanding the role of landscape context is particularly important for predicting future responses to climate warming (Cannone et al. 2007, Damschen et al. 2010, Harrison et al. 2010, Virtanen et al. 2010), whether these are driven by direct effects on plant performance (De Valpine and Harte 2001, Morgan et al. 2007, Forrest et al. 2010) or indirect effects mediated by tree establishment and growth (Franklin et al. 1971, Miller and Halpern 1998, Batllori et al. 2009, Harrison et al. 2010).
In this paper, we explore long-term changes in the structure and composition of forest-meadow ecotones in the Three Sisters Biosphere Reserve, a largely undisturbed mountain landscape in the central Cascade Range of Oregon.Building on earlier studies in this landscape (Halpern et al. 1984, Miller andHalpern 1998), we analyze three decades of vegetation change  across ecotones representing diverse physical environments (both montane and subalpine), vegetation types, and tree invasion histories.We are unaware of similar studies of comparable duration, ecological breadth, or spatial extent.
We address the following questions: (1) How has vegetation (species composition, richness and abundance) changed across forest-meadow eco-tones over the past 26 yr? (2) Do the magnitude and direction of change in vegetation differ among forest, ecotone, and meadow habitats?
(3) How have richness and abundance changed for groups of species defined by habitat association (meadow vs. forest understory) or growth form (grasses, sedges/rushes, herbs, and shrubs)?(4) How are patterns within the ecotone related to landscape context, local tree influences, and initial vegetation?

Study area
The Three Sisters Biosphere Reserve (TSBR) is a federal wilderness area of nearly 100,000 ha in the Cascade Range of central Oregon, USA.It encompasses two physiographic provinces, the Western and High Cascades.The former are older, more deeply dissected landforms, comprised of volcanic flows and pyroclastic deposits that originated during the Oligocene and Miocene epochs (Orr et al. 1992).Soils derive from basalts, andesites, and pyroclastic tuffs and breccias.Primary ridge crests approach 1500 m in elevation.The High Cascades province lies to the east at higher elevations, dominated by younger (Quaternary) shield and composite volcanoes.Topography is more gentle and rolling.Soils are deep and well drained, derived from recent deposits of pumice, ash, and cinders (Orr et al. 1992).
The climate is maritime, with cool, wet winters and warm, dry summers.Temperature and precipitation vary with elevation and topography, reflecting strong orographic effects.Within the study area, mean annual precipitation ranges from 1700 to 2700 mm and falls primarily as snow (Daly et al. 2008).However, depth and duration of snowpack-thus length of growing season-differ markedly in montane and subalpine environments.In the latter, snow can persist into late July or August in some years.Average maximum July temperature ranges from 19 to 238C and minimum January temperature ranges from À9 to À48C (Daly et al. 2008).
TSBR supports a broad range of forest, meadow, and ecotonal communities, whose structure and composition vary with elevation, topography, soils, and hydrology (Halpern et al. 1984).We consider those that are characteristic of montane and subalpine environments (Franklin andDyrness 1988, Franklin andHalpern 1999; Appendix A), but not those of upper timberline or treeline environments.In the montane zone (,1600 m), forests are dominated by Abies grandis, Pseudotsuga menziesii, A. amabilis, and Pinus contorta, and graminoid-, herb-and shrub-dominated meadows occur on south-facing slopes, in hydric basins, and on poorly drained flats.In the subalpine zone (1600-2000 m), forests of Tsuga mertensiana and A. lasiocarpa are distributed among a mosaic of graminoid-, forb-, and heath-dominated meadows (Halpern et al. 1984, Franklin andHalpern 1999).Descriptions of the composition and environmental correlates of these forested and non-forested communities have been reported previously for TSBR (Halpern et al. 1984) and other mountain landscapes in western North America (Fonda and Bliss 1969, Brooke et al. 1970, Kuramoto and Bliss 1970, Douglas 1972, Henderson 1973, Hickman 1976, Franklin and Halpern 1999).
Fire, the primary disturbance agent in this system, occurs infrequently.Disturbance histories suggest fire-return intervals of 100-150 yr in the montane zone (Teensma 1987, Cissel et al. 1999), and many hundreds of years in the subalpine (Halpern et al. 1984).Although aboriginal burning of meadows likely occurred-primarily at lower elevations (Burke 1979, Boyd 1999)direct evidence is lacking in TSBR.None of our study locations have experienced recent fire.Sheep grazing was common in TSBR from 1880-1910(Rakestraw and Rakestraw 1991), as in most meadows of the Cascade Range, until deteriorating range conditions resulted in the closure of most grazing allotments between 1920 and 1938 (Rakestraw andRakestraw 1991, Miller andHalpern 1998).Sheep last grazed in TSBR in 1947 (Johnson 1985).

Field measurements
In 1983, 21 transects were established across forest-meadow ecotones spanning a range of elevations, landforms, and vegetation types (Tables 1 and 2).Transects were 2 m wide and ranged in length from 50-220 m (Table 1, Fig. 1).Each originated in closed-canopy forest and extended into open meadow, beyond any tree invasion.In 1983 and1993, all trees (.10 cm tall) and tree seedlings ( 10 cm tall) within each transect were spatially mapped, measured for diameter (basal or breast height, depending on size), and aged to enable reconstructions of tree invasion history (Miller and Halpern 1998).In addition, 1 3 1 m quadrats were established on alternating sides of each transect (n ¼ 39-108 quadrats per transect).In each quadrat, the cover (%) of each vascular plant species was estimated visually.Cover of tree species was estimated

Habitat definitions and species classification
We delineated three habitats along each transect-forest, ecotone, and meadow-based on the presence, cover, and ages of trees (Fig. 1).Forest was defined by the presence of trees that established prior to 1900 (typically much earlier; Miller and Halpern 1998) and by continuous (occasionally patchy) canopy cover at the initial measurement in 1983.Meadow was defined as the terminal portion of the transect in which trees were absent or present only as seedlings in 2009.At one site (Obsidian Creek), we removed from the analysis a 10-m section of transect where a small, isolated tree island, .30m from the forest, interrupted otherwise contiguous open-meadow habitat.Ecotone was defined as the zone between forest and meadow, representing former meadow into which conifers had established during the 20th century.Differences in the timing, intensity, and spatial pattern of tree invasion (Miller and Halpern 1998) led to considerable variation in the cover, density, and spatial distribution of trees among ecotones (Fig. 1).Between 1983 and 2009, trees recruited past the endpoints of two transects (Rebel Rock and Olallie Meadow) resulting in complete loss of meadow habitat.
We classified all ground-layer taxa as characteristic of meadow (n ¼ 195) or forest understory (n ¼ 73) (Appendix B).Trees were not classified, nor were 23 other taxa (ruderal species and those identified only to genus).Assignments were based on regional floras (Hitchcock et al. 1969, Hitchcock andCronquist 1973), phytosociological studies in TSBR (Halpern et al. 1984), and previous chronosequence-based studies of conifer encroachment (Haugo andHalpern 2007, 2010).Although this classification simplifies the distributions of some species (e.g., those tolerant of edge environments), it captures the distinct habitat associations of most.To assess physiognomic changes in the ground layer, we also classified species by growth form: grass, sedge/ rush, herb (including ferns), and shrub (including sub-shrubs).

Climatic variation
To assess vegetation changes within the context of recent trends or inter-annual variation in climate, we examined climate records from 1940 to 2009.We examined four climatic variables known to influence plant phenology or productivity: mean summer (June-August) temperature, mean annual temperature, total summer precipitation, and spring snowpack (snow water equivalent [SWE] on 1 April, a proxy for soil moisture availability and length of growing season) (Walker et al. 1993).Temperature and precipitation data were taken from the closest meteorological station with long-term records (McKenzie Bridge, 450 m; U.S. Historical Climatological Network; hhttp://cdiac.ornl.gov/epubs/ndp/ushcn/ushcn.htmli).Although models such PRISM (Daly et al. 2008) can be used to estimate temperature and precipitation in complex mountainous terrain-including our sites, which occur at higher elevations-projections for the period of interest  are susceptible to inhomogeneities in temporal data (reflecting undocumented changes in climate stations or station locations; Luzio et al. 2008).Moreover, previous regional scale analyses of long-term climatic trends in the Pacific Northwest suggest strong coherency among climate stations at varying elevations (Mote 2003(Mote , 2004)).Snowpack data were taken from the nearest SNOTEL site (McKenzie Pass, 1454 m; USDA Natural Resources Conservation Service; hftp://ftpfc.sc.egov.usda.gov/OR/snow/snowcourse/or_datai).For each climatic variable, we tested for significant linear relationships with time for the full record  and for the period of study .

Measures of vegetation structure and diversity
We computed several measures of tree structure and ground-layer diversity and abundance for each habitat (forest, ecotone, and meadow) 3 transect 3 measurement period.Measures of tree structure included mean tree density (.10 cm tall) and basal area (m 2 /ha) within the 2-m-wide belt transect.Tree cover (sum of individual tree species) and the coefficient of variation (CV) in tree cover were computed from quadrat values; the CV was included to characterize the patchy distribution of tree cover in many ecotones.Measures of ground-layer diversity and abundance included total species richness, richness and total cover of meadow and forest species, and total cover of each growth form.We do not report patterns of growth-form richness, which were dominated by herbaceous species at most sites.

Comparing rates of change between sampling intervals
Prior to more comprehensive analyses, we tested whether rates of vegetation change were similar between sampling intervals (1983-1993 vs. 1993-2009).Comparable rates would support simpler analyses of change over the full study period.For each response variable 3 habitat combination, we computed the annualized change for each sampling interval, then compared intervals using paired t-tests (n ¼ 16).Intervals differed (P , 0.05) in only two of 38 tests (change in richness of forest species in ecotone and meadow habitats).Given the consistency between intervals, we based all further analyses of change on the 26-yr study period .

Temporal changes among forest, ecotone, and meadow habitats
To compare temporal changes in vegetation among habitats (questions 1-3), we employed repeated measures Permutational Analysis of Variance (PERMANOVA; Anderson 2001, McArdle andAnderson 2001).We used PERMANOVA because of heterogeneous variance among many of the response variables.Analyses used a blocked, split-plot design with transects as blocks, habitat (forest, ecotone, or meadow) as the main plot, and time (1983 or 2009) as the subplot.Models also included the time 3 transect and time 3 habitat interactions.For significant habitat effects or time 3 habitat interactions, posthoc pairwise comparisons were made using Fisher's LSD (Zar 1999).Models were run using Euclidean distances with 9,999 permutations in the PERMANOVAþ (v.1.0.2) add-on for PRIMER 6.0 (Anderson et al. 2008).
To visualize temporal changes in species composition among habitats (questions 1 and 2), we used non-metric multidimensional scaling (NMDS; Kruskal 1964).Ordinations were based on the average species composition of each habitat 3 time, with rare species (present in ,5% of habitat 3 time samples) removed and Bray-Curtis distance measures.To prevent floristic variation among the full set of transects from masking temporal trends, each of the four landscape contexts (see Methods: Comparing changes among ecotones across the landscape) was ordinated separately.Ordinations had two-di-mensional solutions, were initiated from random starting configurations for a maximum of 400 iterations, and were rerun up to 40 times or until an instability criterion of 0.00001 was met.Final solutions were rotated with Principal Components Analysis (PCA; Hotelling 1933) to maximize the variation explained by the first axis.

Comparing changes among ecotones across the landscape
To assess how patterns of change within ecotones varied with landscape context (question 4), we qualitatively defined the environmental (landscape) context of each transect based on NMDS ordination of the composition of meadow habitats in 1983.Using this method of indirect ordination, relationships among transects are based on floristic dissimilarity.Environmental or landscape gradients are not defined a priori, but are inferred from the distributions of species in ordination space.We ran NMDS (as described above) with each transect represented by the average species composition of its meadow quadrats in 1983, thus minimizing effects of trees.For the two transects without meadow habitat in 2009 (see Methods: Habitat definitions and species classification, above), we used the average composition of ecotonal quadrats lacking tree cover in 1983 and 2009.The final twodimensional solution had a stress value of 13.9.We subjectively assigned transects to four distinct ''landscape contexts'' (Fig. 2A) based on their positions in ordination space and our knowledge of the primary environmental gradients by which communities are structured in this system: elevation (montane vs. subalpine) and landform (affecting soil moisture availability) (Halpern et al. 1984, Franklin andHalpern 1999).The four landscape contexts are: (1) montane hydric basins: lower elevation basins with seasonally high water tables; (2) montane mesic slopes: lower elevation, south-facing slopes or benches; (3) subalpine early snowmelt sites: higher elevav www.esajournals.orgtion benches and south-facing slopes with longer growing seasons; and (4) subalpine late snowmelt sites: higher elevation basins and northfacing slopes with shorter growing seasons (Appendix A).Each supports a distinct flora (Fig. 2B; Halpern et al. 1984, Franklin andHalpern 1999).
Results of NMDS were used in two ways to explore the importance of landscape context for changes in the ecotone.First, positions in ordination space served as a template for comparing changes in tree structure and ground-layer vegetation over the study period.For each response variable, we produced a ''bubble plot'' quantifying the direction (positive or negative) and magnitude of change at each site.Second, scores on NMDS1 and NMDS2 were used as surrogates for landscape context (landform/hydrology and elevational zone, respectively) in models of vegetation change (see Methods: Modeling the importance of tree influence, landscape context, and initial vegetation, below).

Modeling the importance of tree influence, landscape context, and initial vegetation
We used multiple regression models to assess the joint contributions of tree influence, landscape context (NMDS scores, as defined above), and initial vegetation to changes within ecotones from 1983-2009 (question 4).Tree influence was quantified by two types of variables.The first summarized initial (1983) tree structure: tree cover, CV of tree cover, tree density, and basal area.These served as measures of cumulative tree influence over the study period, with the expectation that vegetation responses would be greater where the initial (and thus cumulative) effects of tree structure were greater (e.g., Pugnaire et al. 1996, Haugo andHalpern 2010).The second represented changes in each structural measure over the study period (2009 minus 1983 values), with the expectation that vegetation responses would be greater where changes in tree structure were greater.We used PCA to reduce the dimensionality of these structural variables, using the function prcomp in R v2.10.0 (R Development Core Team 2009).The first principal component (PC1) explained 37% of the total variation and correlated positively with initial tree structure (tree cover, homogeneity of tree cover, and basal area) (Appendix C).PC2 explained 26% of the total variation and correlated negatively with measures of change in tree structure (change in density, basal area, and cover).
Predictors in regression models included scores on NMDS1 and NMDS2 (landscape context), scores on PC1 and PC2 (tree influence), and the initial (1983) value of each response variable.The latter was used to test whether the magnitude of change was correlated with the initial condition (e.g., if sites with greater richness experienced greater loss of species).For each measure of vegetation change, we started with a ''full'' model and removed predictors in reverse v www.esajournals.orgorder of strength (sums of squares) to minimize the Akaike Information Criterion (AIC; Akaike 1974).We used standard diagnostics to confirm normality and homogeneity of variance (Zar 1999).Models were developed in SPSS 17.0 (SPSS 2008).

Climatic trends and inter-annual variation
In the long-term record , we detected small but statistically significant increases in mean annual temperature (0.018C/yr; R 2 ¼ 0.105, P ¼ 0.006) and mean summer temperature (0.038C/yr; R 2 ¼ 0.312, P , 0.001) (Fig. 3A), but no trends in summer precipitation or spring snowpack (Fig. 3B).During the period of study , we detected a significant increase in summer temperature (0.0648C/yr; R 2 ¼ 0.354, P ¼ 0.001; Fig. 3A), but no changes in mean annual temperature, precipitation, or spring snowpack.Among sampling years, 1983 and 1993 had relatively cool/wet summers, with average to below-average temperatures, above-average summer precipitation, and average spring snowpack (Fig. 3A, B).In contrast, 2009 had a warm, dry summer (Fig. 3A) but above-average spring snowpack (Fig. 3B).(1983, 1993, and 2009).The dashed regression line represents the significant increase in summer temperature during the period of study.

Temporal changes in vegetation among habitats
Vegetation changed to varying degrees among habitats (questions 1-3).All measures of tree structure were stable in forest habitats (Fig. 4).In contrast, tree cover increased and heterogeneity of cover decreased in ecotonal habitats (Fig. 4A,  B), although tree density and basal area did not change significantly (Fig. 4C, D).Tree cover and basal area remained greater in the forest and heterogeneity of cover remained greater in the ecotone over the study period.
For most ground-layer variables, values in the ecotone were intermediate to those in the forest and meadow (Fig. 5).Exceptions included total richness (ecotone ¼ meadow; Fig. 5A), sedge/rush cover (ecotone ¼ forest; Fig. 5G), and shrub cover (similar in all habitats; Fig. 5I).Ecotones were more dynamic than adjacent forest and meadow habitats.In the ecotones, total richness (Fig. 5A), richness of meadow species (Fig. 5B), and cover of grasses (Fig. 5F) declined, whereas richness of forest species (Fig. 5D) increased.In contrast, cover of shrubs did not differ among habitats, but increased over time (Fig. 5I).
Species composition in the ecotone was intermediate between forest and meadow (Fig. 6), except at Green Lake (Fig. 6D), where tree structure was minimal (Table 2).Over time, the composition of most ecotonal habitats trended toward adjacent forests.By comparison, compositional changes within meadows were small and non-directional.

Changes among ecotones across the landscape
Among ecotonal habitats, changes in vegetation varied with landscape context (question 4).Increases in tree structure (basal area, cover, and homogeneity of cover) were generally greater where soil moisture was more limiting-on mesic slopes in the montane zone and on early snowmelt sites in the subalpine zone (Fig. 7B-D).By comparison, tree density was generally stable (Fig. 7A).Landscape-scale variation in the ground-layer was more complex.With one v www.esajournals.orgexception (Corral Flat), declines in total richness were greatest in subalpine early snowmelt sites (Fig. 8A), reflecting similar declines in richness of meadow species (Fig. 8B).In contrast, increases in richness and cover of forest species were greater in montane than in subalpine sites (Fig. 8D, E).Among growth forms, grass cover declined in montane hydric basins (Fig. 8F) and shrub cover increased in all landscape contexts except subalpine early snowmelt sites (Fig. 8I).Changes in sedges/rushes and herbs were highly variable across the landscape (Fig. 8G, H).

Modeling the importance of tree influence, landscape context, and initial vegetation
Multiple regression models, incorporating tree influence, landscape context, and initial vegetation (question 4), explained significant variation in all measures of vegetation change in the ecotone (R 2 of 0.30-0.81,P , 0.05; Table 3).Change in tree structure (PC2) was not a significant predictor in any model.However, Fig. 5. Changes in ground-layer vegetation (mean 61 SE) over the study period for forest (black), ecotone (gray), and meadow (white) habitats.For each variable, P-values for the habitat (H), time (T), and habitat 3 time (H 3 T) terms are from a univariate repeated measures PERMANOVA.For significant habitat terms, the results of post-hoc comparisons using Fisher's LSD are reported after the P-value.v www.esajournals.orginitial tree structure (PC1) was significant in four of nine models (Table 3).Sites with greater initial tree structure showed larger declines in cover and richness of meadow species, in total richness, and in herb cover.
Landscape context was significant in models of richness, but not of cover.Where moisture was more limiting (montane mesic and subalpine early snowmelt sites; high scores on NMDS1), declines in meadow and total species richness were greater.At lower elevations (montane sites; low scores on NMDS2), meadow species experienced greater declines and forest species experienced greater increases in richness.Fig. 6.Temporal changes in species composition in the forest, ecotone, and meadow habitat at each site, as portrayed in NMDS ordination space.A separate ordination was run for each landscape context (Fig. 2A) to avoid masking of temporal trends by floristic differences among sites.Samples represent the average species composition of each habitat 3 sampling date (1983, 1993, and 2009).Lines connect sampling dates; arrows indicate directions of change over time.Transect codes are defined in Table 1.CF, WP, LM2, OM, and GL were not sampled in 1993.
v www.esajournals.org The initial value of each response variable was a significant predictor of change in all models (Table 3).Cover and richness of meadow species and cover non-woody growth forms declined more where initial values were greater.In contrast, cover and richness of forest species and cover of shrubs increased more where initial values were greater.

DISCUSSION
Forest-meadow boundaries are prominent features of the montane and subalpine zones of this and other mountain landscapes.Conifer encroachment into meadows has occurred for over a century in TSBR (Miller and Halpern 1998), yet ecotonal habitats remain structurally distinct from adjacent forests and retain similarities to adjoining meadows.It is within these historical and ecological contexts that we explore the contributions of landscape context and tree influence to recent changes in ecotonal vegeta-tion.

The dynamics of forest-meadow ecotones
The structure, composition, and diversity of ecotonal habitats changed significantly over 26 yr of study, whereas forest and meadow habitats remained stable.Climate warming has been implicated in compositional changes in other mountain ecosystems of North America and Europe (Damschen et al. 2010, Harrison et al. 2010, Lenoir et al. 2010), yet we saw little evidence of a direct effect of warming on ground-layer vegetation in this system.If warming was having a direct effect, it likely would not be limited to ecotonal habitats.Although tree canopies can buffer microclimatic changes in the forest (e.g., Damschen et al. 2010), changes in snowpack, timing of snowmelt, or summer drought stress should be as apparent in open meadow as in the ecotone.Alternatively, recent warming could have an indirect effect limited to ecotones if it were mediated by effects on tree Fig. 7. Changes (1983Changes ( -2009) ) in ecotone tree structure across the TSBR landscape, as portrayed in NMDS ordination space (see Fig. 2A).Gray and white circles represent positive and negative changes, respectively.Magnitude of change is proportional to symbol size, scaled to the maximum for each variable: (A) tree density ¼ 3,000 trees/ha; (B) tree basal area ¼ 70 m 2 /ha; (C) tree cover ¼ 60%; and (D) CV of tree cover ¼ 1.4.v www.esajournals.orggrowth.This also seems unlikely, as regression models consistently showed a lack of response to variation in tree growth (PC2; changes in cover, homogeneity of cover, basal area).However, climate variation during the 20th century has played a major role in initiating conifer invasions Fig. 8. Changes (1983-2009) in ecotone ground-layer vegetation across the TSBR landscape, as portrayed in NMDS ordination space (see Fig. 2A).Gray and white circles represent positive and negative changes, respectively.Magnitude of change is proportional to symbol size, scaled to the maximum for each variable: (A, B, and D) species richness ¼ 5 species/m 2 ; and (C and E-I) cover ¼ 60%.
v www.esajournals.org of these and similar meadows in both montane and subalpine settings throughout western North America (Brink 1959, Taylor 1995, Miller and Halpern 1998, Rochefort et al. 1994, Woodward et al. 1995).It is these legacies of past changes in climate that continue to exert their influences.

The role of landscape context
Understanding how landscape context affects the dynamics of species' interactions within ecotones is critical for predicting the broader ecological consequences of forest encroachment or expansion (Baker and Weisberg 1995, Jones and Callaway 2007, Harrison et al. 2010).Despite considerable variation among ecotones, two important patterns emerged across the landscape.First, declines in richness of meadow species were greater where soil moisture is seasonally limiting-on mesic slopes in the montane zone, and in early snowmelt sites in the subalpine.A similar relationship was not observed for cover, suggesting that declines reflect local extirpation of subordinate species.It is possible that under conditions of moisture stress, these subordinate taxa may be more susceptible to below-ground competition from trees.Where soil moisture is less limiting (hydric and late snowmelt sites), root competition may be less intense and declines are more likely to be driven by competition for light (Tilman 1988) and thus related to ecotone structure (tree influence) rather than landscape position.
Second, despite the expansion of trees in ecotonal habitats, increases in the richness and cover of forest herbs were largely limited to montane sites.Studies from old fields, secondary forests, and forest-grassland mosaics have demonstrated that dispersal opportunities largely determine the development of forest understory communities (Matlack 1994, Brunet and von Oheimb 1998, Verheyen and Hermy 2004, Duarte et al. 2007).Montane forests in this region support a rich diversity of shade-tolerant species (66 taxa among the nine montane transects in TSBR) that can readily disperse into the understories of invading forests (Haugo andHalpern 2007, 2010).Subalpine forests are depauperate in comparison: the 11 subalpine transects in TSBR supported only 37 forest species, and 22 of these were restricted to the Rebel Rock transect.Deep and persistent snow packs, short growing seasons, and young, infertile soils contribute to this floristic condition (Halpern et al. 1984, Franklin andHalpern 1999), which limits the potential for forest species to colonize newly forming forest habitats.Together, these two trends can have profound, local effects.Tree invasion into subalpine, early-snowmelt sites leads to loss of meadow species, but there are few forest species to colonize these newly forming habitats, leading to marked declines in plant diversity.In more general terms, our results illustrate the diverse ways in which landscape context can mediate compositional changes in structurally dynamic environments: environmental or resource variation can affect the outcomes of species' interactions (Bertness and Callaway 1994) or pose Notes: Results include adjusted R 2 (variation explained) and level of significance (P) for the final model, and standardized coefficients (Coeff.)and significance (P) of predictors.Significant (P , 0.05) predictors are in bold font.Starting with a full model, final models were derived by sequential removal of predictors to minimize AIC.Predictors included two measures of landscape context (scores on NMDS1 and NMDS2; Fig. 2A); initial tree structure (PC1 score); change in tree structure over the study period (PC2 score); and the initial value of the response variable.

The nature and strength of tree influences
Trees substantially alter their local environments, particularly in non-forest settings.They create shade, moderate air and soil temperatures, affect precipitation and snow accumulation, compete for soil resources, and alter litter inputs and nutrient cycles (Belsky et al. 1993, Scholes and Archer 1997, Coomes and Grubb 2000, Hibbard et al. 2001, Musselman et al. 2008).Thus, we were surprised to find that changes in the ground-layer vegetation were unrelated to changes in tree structure (PC2), but strongly related to initial structure (PC1).These seemingly contradictory results can be explained by considering the conditions and processes that lead to vegetation change in this system.First, despite decades of tree presence, ecotones were still dominated by meadow species when initially sampled in 1983 and thus had significant potential for change.Second, multiple decades of change in the ground layer reflect the cumulative effects of trees already present (Pugnaire et al. 1996, Haugo andHalpern 2010).Variation in initial structure was large among sites, reflecting diverse invasion histories (Miller and Halpern 1998).In comparison, changes in structure were small and less variable.Thus, ecotones with initially greater tree cover or basal area had the potential to effect larger changes in the ground layer.Interestingly, despite the significance of initial tree structure in these models, declines in the cover and richness of meadow species were also correlated with initial cover and richness.This relationship may reflect the statistical property that communities with greater richness or cover have ''further to fall.''However, it also reveals significant variation in the ground layer that was unrelated to initial tree structure, but indicative of the potential for change.
In contrast to meadow species, forest species did not respond to variation in tree structure.Although we anticipated greater colonization (change in richness) and growth (change in cover) of forest species in ecotones with greater initial tree structure or greater change in structure, neither predictor was significant in regression models.We attribute this general lack of response to patterns in the subalpine zone, where poorly developed understories result in limited potential for dispersal and colonization of the ecotone (e.g., Matlack 1994, Brunet andvon Oheimb 1998).Similar constraints do not exist at lower elevations where forest herbs can disperse rapidly into ecotonal habitats formed by invading trees (Haugo andHalpern 2007, 2010).

Implications and future changes
Long-term observations of forest-meadow ecotones in TSBR allow us to place recent changes into the broader historical context of forest expansion, and to consider what these changes imply for the future.Despite nearly a century of forest encroachment, ecotones remain structurally distinct from adjacent forests and retain elements of the meadows from which they originate.Ecotone characteristics also vary across the landscape in response to past invasion patterns (Miller and Halpern 1998) and to underlying gradients in resource availability.Vegetation dynamics in ecotonal habitats are the cumulative effect of how these factors interact to cause species' losses and gains.
Our study was not designed to quantify the rate or extent of recent meadow loss, or the susceptibility to encroachment of meadows representing different landscape contexts-larger-scale questions that are being addressed through remote sensing and change-detection analysis (e.g., Takaoka andSwanson 2008, Zald 2009).However, we can describe the communitylevel consequences of encroachment where it does occur.In this sense, our work has important implications for the use of remote sensing to quantify landscape-scale changes in ecosystem state.First, even in the absence of detectable increases in forest extent, changes in the groundlayer along existing forest-meadow boundaries can be significant.Second, simple estimates of change in forest cover may not capture the variation in biological response (e.g., loss of meadow species or establishment by forest species) that occurs in different portions of the landscape.Integrating our results with remote sensing tools could provide a more complete understanding of the extent and ecological consequences of forest encroachment in this and similar landscapes.
Future changes in climate are likely to alter the dynamics of forest-meadow ecotones through effects on tree establishment or growth.Current predictions are for warmer drier summers, warmer wetter winters (Mote and Salathe 2009), and shifts in the form of precipitation (snow to rain; Elsner et al. 2009), suggesting the potential for increasing summer drought and longer growing seasons.These effects are likely to be manifested to varying degrees and in complex ways across the elevational, topographic, and edaphic gradients that define mountain landscapes (Daly et al. 2009).In the montane zone, for example, warmer drier summers may reduce rates of tree invasion and growth on mesic upland slopes, but enhance them in hydric basins, where waterlogged soils currently limit both the extent of tree establishment and associated changes in ground-layer vegetation (Miller and Halpern 1998).Similar ''switches'' may occur in early-vs.late-snowmelt sites in the subalpine zone in response to changing snowpack and growing-season length.Climate change also has the potential to elicit indirect or secondary effects in the form of increased frequency or intensity of insect outbreaks or wildfire (Littell et al. 2009), disturbances that can dramatically alter the characteristics of existing ecotones or create new boundaries.Where invading trees are killed, ecotonal areas are likely to revert quite rapidly to dominance by meadow species (C.B. Halpern, unpublished data).
To our knowledge, this analysis represents the first long-term, large-scale assessment of recent responses of mountain meadows to 20th-century forest expansion.Understanding how tree influence and landscape context shape the dynamics of forest-meadow boundaries represents a first critical step toward predicting future changes in the biological diversity of these and other mountain landscapes.

v
www.esajournals.orgseparately for seedlings and trees.In 2009, 26 yr after establishment, we recensused tree populations and recorded post-1993 recruitment on each transect, and repeated cover estimates in each quadrat.Sixteen transects were resampled in 1993 and 20 transects in 2009.

Fig. 1 .
Fig. 1.Total canopy cover (summed cover of all tree species) in 1983 (gray) and 2009 (black) for select forestmeadow ecotones in the (A-D) montane and (E-H) subalpine zones of the Three Sisters Biosphere Reserve, Oregon.Values are 3-m running averages.Habitats are delineated by vertical dashed lines and coded as f ¼ forest, e ¼ ecotone, m ¼ meadow.
Final solutions had stress values of 4.7 to 19.3.Stress is rough measure of fit-dependent, in part, on sample size-with values ,5 generally considered excellent and values ,15, satisfactory (McCune and Grace 2002).NMDS was run using the metaMDS function of the Vegan v1.11-0 package in R v2.10.0 (R Development Core Team 2009).

Fig. 2 .
Fig. 2. NMDS ordination illustrating (A) the landscape context of each transect and (B) characteristic meadow species coded by growth form.Transect positions are based on the average species composition of meadow habitats in 1983.The dashed lines are included as interpretive guides distinguishing the four landscape contexts (also see Appendix A).

Fig. 3 .
Fig. 3. Climatic variation between 1940 and 2009 (expressed as standard deviations from the 1940-2009 mean).(A) Temperature and (B) precipitation data are from McKenzie Bridge, Oregon (450 m elevation).Snowpack data are from McKenzie Pass, Oregon (1453 m), expressed as snow water equivalent (SWE) on 1 April.The summer period for temperature and precipitation is June through August.Vertical arrows denote sampling dates(1983,  1993, and 2009).The dashed regression line represents the significant increase in summer temperature during the period of study.

Fig. 4 .
Fig. 4. Changes in tree structure (mean 61 SE) over the study period for forest (black) and ecotone (gray) habitats.For each variable, P-values for the habitat (H), time (T), and habitat 3 time (H 3 T) terms are from a univariate repeated measures PERMANOVA.For significant habitat 3 time terms, the results of post-hoc comparisons using Fisher's LSD are shown between times for each habitat.

Table 1 .
Physical characteristics and transect length at each study site in the Three Sisters Biosphere Reserve, Oregon.

Table 2 .
Initial (1983)tree composition and structure (.10 cm tall) in the ecotone at each study site in the Three Sisters Biosphere Reserve, Oregon.

Table 3 .
Results of multiple regression models explaining changes in species richness and cover in ecotonal habitats from1983-2009.