Reversal of multicentury tree growth improvements and loss of synchrony at mountain tree lines point to changes in key drivers


  • Alex Fajardo,

    Corresponding author
    1. Centro de Investigación en Ecosistemas de la Patagonia (CIEP) Conicyt–Regional R10C1003, Ignacio Serrano 509, Coyhaique, Chile
      Correspondence author. E-mail:
    Search for more papers by this author
  • Eliot J. B. McIntire

    1. Canada Research Chair, Center for Forest Research, Université Laval, Québec, QC G1V 0A6, Canada
    Search for more papers by this author
    • Present address: Canadian Forest Service, Pacific Forestry Centre, 506 Burnside Road W., Victoria, BC V8Z 1M5, Canada.

Correspondence author. E-mail:


1. Altitudinal tree line ecotones (ATE) are among the most sensitive plant formations facing global warming as the altitudinal decrease in temperature is considered the driver controlling the upper elevation limit of tree lines world-wide. In this study, we attempted to answer the following questions: (i) how have the conditions during the last 2–3 centuries affected ATE tree growth (physiology) and recruitment (demography)? and (ii) how strong is synchrony between these two processes at the ATEs?

2. We used spatial sampling grids at different ATEs in two ecosystems on two subcontinents: Nothofagus pumilio in the Andes of Chilean Patagonia (46° SL) and Pinus albicaulis in the Rockies of Western Montana, USA (46° NL). Basal increment cores were extracted from trees to estimate the growth and recruitment date. An annual detrended basal area increment was estimated for each tree and was modelled against elevation and time.

3. Tree growth improved over multiple centuries at all tree lines. Recently (c. 50 years), however, improvements are disappearing or reversing. The uppermost tree line trees showed moderate declines in Montana and incipient declines in Patagonia. The declines are most dramatic slightly below current tree line (c. 200 m). Tree recruitment patterns showed that tree lines have been moving uphill in both regions until at least 40–70 years ago. These movements occurred primarily through abrupt pulses upward with infilling occurring concurrently (Patagonia) or at some time thereafter (Montana).

4. Synchrony between growth and recruitment occurred in the 18th and 19th centuries in both regions. This synchrony was negative in Patagonia and positive in Montana, with varying lag periods. During the 20th century, these patterns of synchrony were lost at all sites. This loss of synchrony suggests that we could be entering a global period in which temperature is no longer the dominant driver of key features of tree lines.

5.Synthesis. Our study shows that at two structurally different tree lines, recent and initial declines in growth and losses of long-term synchrony are occurring in the latter part of the 20th century. These findings are opposite to simplistic expectations of global warming effects on tree line dynamics and call for a model reformulation that uncouples drivers of growth and recruitment.


There is extensive evidence that global temperatures have increased during the last century, particularly at high elevation (Díaz & Bradley 1997), with average reported increases of about 0.6 °C (Houghton et al. 2001; Jones, Osborn & Briffa 2001; Villalba et al. 2003). The ongoing global warming has been described as an important factor altering the range of species and performance, and its population-level effects are thought to be more readily detected or predicted in boundary areas of species distributions (MacArthur 1972; Innes 1991; Loehle 2000; Gamache & Payette 2004; Messaoud, Bergeron & Leduc 2007). Near the margin of its range, a species may develop on a relatively small variety of sites, and climate frequently becomes limiting for physiological and population processes, including growth and recruitment (Fritts 1976; Körner 2003; Holtmeier & Broll 2005). This is particularly true for alpine tree line ecotones (ATEs), which constitute relatively narrow and well-delineated landscape boundaries, and their maximum elevation is globally controlled by temperature (Jobbágy & Jackson 2000; Grace, Berninger & Nagy 2002; Körner & Paulsen 2004). As a consequence, ATEs have become key systems (e.g. bio-indicators) to assess, monitor and understand the mechanisms of current tree line dynamics associated with climate-dependent ecological processes, that is, climate change and variability (Körner & Paulsen 2004; Holtmeier & Broll 2005; Kullman 2007).

There are three main potential responses of tree line structure to changes in ecological drivers: changes in tree density via upslope advance or infilling of regeneration (population-level), increased tree growth rates (physiological-level) and changes in growth form from krummholz (crooked brushwood) to erect stems (physiological-level) (Innes 1991; Grace, Berninger & Nagy 2002; Daniels & Veblen 2004; Kullman 2007). Indeed, the response of trees at or just below the highest regional alpine tree line is expected to be a good indicator for tracking changes provoked by global warming because climate should constitute the main limiting factor for tree growth (Tranquillini 1979; Grace, Berninger & Nagy 2002; Körner 2003; Körner & Paulsen 2004). Physiologically, low temperature has been shown to limit tree productivity (Körner 1998) either by affecting photosynthesis or, with much more recent support, by limiting carbon usage (Fajardo, Piper & Cavieres 2011; Hoch & Körner 2012). Thus, an increase in temperature should increase tree productivity (Cullen et al. 2001). Consequently, a positive growth response has been evidenced by some dendroecological studies (Villalba et al. 1997; Motta, Morales & Nola 2006; Vila et al. 2008). On the other hand, at the population-level, some predictive models and empirical studies have documented upslope advance in seedling recruitment at the tree line in response to climate warming (Taylor 1995; Kullman 2002; Camarero & Gutiérrez 2004; Dullinger, Dirnböck & Grabherr 2004). But in both cases – tree growth and seedlings recruitment – there is some compelling evidence suggesting that warming is not creating the unidirectional expected changes. For example, upslope advance is not universal (Cuevas 2002; Daniels & Veblen 2004; Parmesan 2006; Harsch et al. 2009), and tree productivity has shown declines with warming (Paulsen, Weber & Körner 2000; Lloyd & Fastie 2002). In this respect, it is likely that in some places an increase in temperature may also lead to higher water deficit in plants (warmer conditions producing water limitations to plant performance), which could offset the expected positive effects from temperature increase, for example limiting tree growth or the upslope advance of tree line (Anfodillo et al. 1998; Lara et al. 2001; Daniels & Veblen 2004; Oberhuber 2004; Kullman & Öberg 2009). Furthermore, the negative impact of water stress could be caused by improved growth from higher temperature; the larger the increment of growth, the higher the water demands (negative feedback). Thus, global warming might affect tree growth and recruitment differently in distinct locations characterized by dissimilarities in climate-related factors, such as temperature and precipitation.

Synchronization of biological phenomena is widespread and requires either internally driven (some kinds of fruit masting; Kelly & Sork 2002) or externally driven (common weather drivers, such as those that affect spring bud burst) mechanisms. If temperature is indeed the primary and dominant driver for both recruitment and growth, these processes should be positively synchronized. However, some changes, such as a lack of moisture, will have a stronger effect on recruitment than on growth because germinants are more sensitive to drought stress. Also, tree lines are likely self-stabilized vegetation boundaries (Slatyer & Noble 1992) which might not follow climatic changes very rapidly (Hättenschwiler & Körner 1995). Thus, there may be climatic phases during which mature trees at the ATEs operate either persistently below or close to their physiological limits, with the overall position of the tree line remaining almost unchanged (Körner 2003). Inversely, there may be climatic phases where both tree growth and recruitment are affected in the same way, that is, a synchronization of processes in their response to abiotic drivers.

We interpret both these situations to mean that recruitment is more sensitive than mature tree growth to variations in climate and that recruitment should have dampened variation compared to growth, though still synchronous over a multiyear perspective. As both recruitment and tree growth at the ATE are thought to be affected by the same limiting factor (i.e. low temperature), changes at the tree line have been predominantly seen as a single-factor phenomenon. Periods in which they covary positively could be periods in which there is a single driver; periods in which they covary negatively or not at all suggest different drivers. Furthermore, in unstable climatic conditions, it is possible that the two processes will be lag-synchronous, even if they are driven by the same limiting factor. Long-term studies are needed to assess these processes.

Knowing that low temperature generally limits tree development (growth and recruitment), particularly at high elevations, we assessed here how three centuries of changing climate and more recent warming are affecting ATEs in terms of tree growth, recruitment and the synchronization of both processes. Chiefly, we were interested in determining (i) whether tree growth is changing through time and whether this change varies along the altitudinal gradient: as low temperature limits tree line formation, are tree line trees more affected by global warming than their counterpart trees at lower elevations? Likewise, we wanted to determine the same pattern for (ii) tree recruitment, and finally (iii) whether growth, a physiological individual-level process, is synchronized with tree recruitment, a population-level process, and does this synchrony vary with altitude and across the last three centuries. Using a spatio-temporal intensive sampling in four different ATEs in the southern Andes of Chile with Nothofagus pumilio and in the Rocky Mountains of Montana (USA) with Pinus albicaulis, we examined the expression of both mechanisms in time (last 300 years) and in space (altitudinal gradient). We note that we do not assess direct relationships with climate because we are interested in much longer time frame than is available from nearby (<500 km) meteorological stations; using the tree growth as an indicator of past climate would be circular logic in the present paper.

Materials and methods

Research Sites and Study Species Description

The first study sites are located in Patagonia, Chile; one is within the Cerro Castillo Natural Reserve (46°04′ S and 72°03′ W, c. 1340 m above sea level), and the other is within El Fraile mountain (45°37′ S and 72°00′ W, c. 1290 m above sea level), hereafter Patagonia sites, where N. pumilio (Poepp. et Endl.) Krasser (lenga, Nothofagaceae) is the only tree species at the tree line. This area belongs to the supra-temperate belt with humid climatic conditions (Amigo & Ramírez 1998). The annual precipitation is c. 1500 mm, mainly as snow. The precipitation during the growing season (November–March) is c. 500 mm (Informe Meteorológico de Chile, Dirección General de Aguas, 2008). The soil is derived from aeolian volcanic ash deposits. For central Patagonia, a widespread trend towards warmer and drier conditions has been reported, which has correlated with an upward trend in the Southern Annular Mode since c. 1950 (Aravena & Luckman 2009; Garreaud et al. 2009). More specifically, both instrumental and proxy reconstructions (primarily from tree rings) show a clear warming of 0.86 °C over the 20th century (compared with the 1640–1899 means) that accentuated in the mid-1970s (Kattenberg et al. 1996; Villalba et al. 2003). This warming has been accompanied by a decrease in precipitation in most of the Patagonian-Andean region (Aravena & Luckman 2009).

The second study sites are located in Western Montana, USA: one site is in the Pioneer Mountains range (45°40′ N and 112°58′ W, 2900 m above sea level) and the other is in the Gravelly Mountains range at some 120 km south-east of the first site (44°54′ S and 111°48′ W, 3000 m above sea level), hereafter Montana sites. Here, P. albicaulis Engelm. (whitebark pine, Pinaceae) is the dominant tree species at the tree line (with some association with Abies lasiocarpa and Pinus flexilis at lower elevation). Pinus albicaulis is a foundation and keystone species of subalpine forests and a major component of ATEs in the Northern Rocky Mountains of the United States and Southern Canada. Hemispheric records of 20th-century temperature indicate that temperatures in the northern hemisphere as a whole increased sharply from 1900 until c. 1940, remained constant or decreased from 1940 to 1970 and then increased again after 1970 (Houghton et al. 1996).

We selected tree line sites that have not been depressed by human activities, disturbed by mass movements or snow avalanches but represent the natural climatic (i.e. likely temperature controlled) tree limit of the respective region. Finding such places in the southern Andes and in Montana is less difficult than in Eurasia, where traditional human land use near the tree line is common and current upslope tree movement might be a recolonization rather than global warming–induced increases (Körner 2003). Our sites were selected from remote imagery – and subsequently confirmed – to be at upper positions on slopes with low angles (<30% slope) and with the presence of alpine vegetation above the tree line (Fig. 1). At each location, the forests selected were composed of erect, single-stemmed N. pumilio and P. albicaulis at the bottom of the altitudinal gradient, which gradually convert to more krummholz forms as elevation increases. At the tree line altitude, N. pumilio develops an abrupt transition where the ecotone ends sharply to alpine vegetation, whereas P. albicaulis develops a gradual transition from timberline to tree line and alpine vegetation and follows standard terminology more in accordance with north hemispheric tree lines.

Figure 1.

 Photographs showing the abrupt tree line of Nothofagus pumilio in Patagonia (Cerro Castillo site, left panel) and the diffuse tree line of Pinus albicaulis in Montana (Pioneer site, right panel).

Field Sampling

We sampled in February (Patagonia) and August 2007 (Montana) each site by setting a grid of 100 sampling points (five sampling points in each of 20 perpendicular-to-the-slope transects) covering the complete altitudinal gradient from erect, tall trees in a closed and relatively dense forest upward to tree line where krummholz-like trees (i.e. crooked) are common. Tree line is defined here as the uppermost limit of individuals having an upright growth form of at least 2–3 m (Körner 2003), which also includes, in some cases, krummholz that become erect after a period of horizontal growth. At each sample point, we measured canopy cover and extracted a core from the nearest largest, single-stemmed adult tree, avoiding those trees with obvious damage or dieback of major limbs. We determined the elevation of each sampling point with a multiple averaged global positioning system (GPS) sample. In order to reduce bias due to possible change of tree status over time, and therefore to avoid other factors altering tree growth other than climate itself (e.g. competition), we selected only dominant trees assuming that the social ranking remained stable over time; competition is low and thus climate signal is maximized (Stokes & Smiley 1996). For all trees cored, we measured diameter at breast height (1.35 m, DBH), diameter at coring height (c. 0.2 m, DCH) and total height and extracted an increment core to the pith at the base (c. 20 cm from the ground) to determine tree age and growth. Each tree was cored perpendicular to the slope using an increment borer (5.15 mm diameter; Haglöf, Långsele, Sweden). We cored repeatedly until a core through the pith (or as near as possible) was obtained so as to minimize age determination uncertainties. The extent of canopy openness for each sampling point, defined as the per cent of the upper hemisphere compared of open sky, was quantified using hemispherical photography. For this purpose, we used a 7-mm Nikon f 7.4 fisheye lens (the lens has an orthographic projection of 180° angle of view), mounted on a Nikon Coolpix 5000 digital camera (Nikon Corporation, Tokyo, Japan) and Gap Light Analyzer (gla ver. 2) software (Frazer, Canham & Lertzman 2000).

Tree Growth Determination

For each site, cores were prepared following the standard dendrochronological techniques (Stokes & Smiley 1996): cores were dried, mounted and glued firmly in grooved wooden sticks and sanded with successively finer grades of sandpaper until optimal surface resolution allowed annual rings to be distinguished under magnification. All samples were dated and visually cross-dated to detect the presence of either false or incomplete rings using marker rings. Following visual cross-dating, tree rings’ width was measured to the nearest 0.001 cm and assigned to calendar years using a microscope mounted on a dendrochronometer with a Velmex sliding stage and Accurite measuring system. For Patagonian sites, calendar dates were assigned to rings according to the southern hemisphere tree-ring dating convention that assigns an annual ring to the calendar year in which the annual ring formation begins (Schulman 1956). Only cores that either passed through the stem pith or close by (arc of innermost rings was visible) were retained. Among these, we used procedures described by Duncan (1989) to estimate the number of missing rings in cores that missed the pith of the tree. No correction was applied for time required to grow to coring height. Cross-dating was further validated using the program cofecha, which calculates cross-correlations between individual series and a reference chronology (Grissino-Mayer, Holmes & Fritts 1996). We elaborated a master series at each of four altitudes per site. Series showing portions of abnormal growth or low correlation with the reference chronology (i.e. r2 < 0.4) were either truncated or discarded to minimize the effect of atypical tree-ring series on the final chronology.

Estimating tree growth from one-dimensional increment core incurs two biases that can confound climate-driven responses inferred from annual tree growth rings. First, there is a geometry effect due to the difference between linear growth (ring widths) and two-dimensional (basal area) or three-dimensional growth, which are better indicators of tree growth, and second, tree vigour shows an ontogenetic increase–stabilize–decline growth trend with tree age. To minimize the geometry problem, we converted one-dimensional ring width increment to two-dimensional basal area increment (BAI), assuming perfect concentric circles:


where R is the radius of the tree and n is the year of the tree-ring formation. While this has a symmetry assumption, it nevertheless reduces the bias relating growth to ring widths. Assuming perfect concentric circles in coniferous species is less problematic than in broad-leaved species, like N. pumilio. To reduce this potential effect, and also to minimize compression and reaction wood, cores were taken on the ‘side’ of the trees (i.e. along an elevational contour line). To remove the ontogeny problem, we included smooth terms for tree age and log of basal area in the generalized additive model described below (Analysis section).

Recruitment and Tree Age

Estimating year of establishment (recruitment) for the current dominant trees from pith age may also have biases. We defined tree recruitment year as the calendar year of the cored pith. We minimized error in tree age determination by coring as low as technically feasible, usually <20 cm from the ground. This will likely create a slight underestimation of recruitment date (e.g. coring height, unreachable core’s pith); thus, we decided to manifest this uncertainty by using a window time (up to 10 years) around the estimated tree age.



To address the influence of time (i.e. linear or curvilinear changes over time due to drivers such as climate change), altitude and site on BAI, while removing the confounding effects of tree age and basal area, we fit the following three generalized additive mixed effect models using the GAMM function in the mgcv package (Wood 2011) in R (R Development Core Team 2009):

  • L: Linear only, inline image

  • LQ: Linear and quadratic, inline image

  • LQC: Linear, quadratic and cubic, inline image

where s() indicates a smooth spline fit (with default package settings), A is tree age, B is basal area, Y is calendar year, inline image is squared, centred calendar year, inline image is cubed, centred calendar year, E is plot elevation, and S is a categorical factor representing the four sites. We use the symbol ‘*’ for compactness to indicate all main effects and interaction effects; however, we did not include any redundant terms in the model as suggested by the equation above (e.g. the × S interaction only occurs once). We compared the support for these models using the Akaike Information Criterion (AIC). While we were interested in the main effects to examine whether there are global trends, we included an interaction term with site and elevation to examine further details. Also, because each increment core represents multiple repeated measures on each tree, we included a random effect of tree ID on the intercept to account for unwanted tree-specific noise. Since we are not interested in the two-first smooth terms (tree age, A, and basal area, B), as they are confounding factors, we present throughout the text the marginal means which are model predictions but with the actual age and basal area replaced with the overall grand mean age and the overall grand mean basal area, respectively (using the predict.gam function in R where we used the raw data for Y, E and S, but the mean values for A and B in the model above). To illustrate the variation around these model predictions, we show the marginal means plus model residuals. Given that this measure represents the BAI with the age and ontogeny effects removed, we call this detrended BAI (dBAI). In general, our procedure of obtaining detrended growth values is somehow similar to the more traditional method of applying a double detrending procedure using standardization functions (first, negative exponential or linear functions and then a cubic-smoothing spline function), through which one residual chronology per stand is created via the arstan software (Cook & Holmes 1986). By using model selection, our procedure allowed us to test for any linear trends through time, as well as quadratic (i.e. a change from increasing to decreasing), plus a cubic term which permitted for changes at the beginning or end of the time series. These are essential to examine for an effect of anthropogenic climate changes that has occurred only during the latest period of our data set.


To understand the role of altitude (linear and curvilinear) and site on recruitment year, we fit the following linear models in R:

  • L: Linear only, inline image

  • LQ: Quadratic and linear, inline image

  • LQC: Linear, quadratic and cubic, inline image

where Yrecrt is recruitment year, Ec is centred plot elevation (i.e. the mean elevation at each site is subtracted from the plot elevation), inline image is plot elevation squared, inline image is plot elevation cubed, and S is a categorical factor representing the four sites. The squared and cubed terms allowed us to determine whether there was a tendency for recruitment to move upslope in pulses with periods of infilling after the pulses or whether the recruitment tendency follows a more linear fashion. These three models were also compared using AIC. Because of the potential spatial dependency among sites, we also analysed this using generalized least squares regression, where we explicitly modelled the potential spatial dependency in the residuals using an exponential covariance structure. Since the parameter estimates and standard errors of the estimates using this approach were exceedingly close to the multiple linear regressions above (results not shown), and for reasons of ease of explaining, we present the results of the simple model. To limit the overly influential effect of a small number of trees on these quadratic and cubic models, we fit the model using only the data starting from the calendar year in which there were 10 or more trees with data. These years were 1797 and 1811 for the Patagonia sites and 1771 and 1788 for the Montana sites.


We were interested in patterns of synchrony between recruitment and growth and how this synchrony changed spatially (tree line vs. below tree line) and temporally (across centuries). Because of limited sample size and ease of interpretation, we analysed these two aspects separately. To test for temporal changes in synchrony, we divided our data set into three centuries (18th C.: 1706–1806, 19th C.: 1806–1906 and 20th C.: 1906–2006). To test for spatial changes in synchrony, we divided the data sets into the upper and lower plots, divided at 70 vertical metres below the highest tree with an increment core. We used 70 m at all sites to balance the sample sizes as much as possible. In each of these analyses, we took the subsampled data set, reran the GAMM model above, calculated the dBAI and then did simple temporal cross-correlations between counts of recruitment events and mean dBAI (‘stats’ package in R). Since there may also be lags in temporal correlations, we included lags from −50 to +50 years. We were also interested in periodic (i.e. repeating) patterns in the synchrony relationships. Consistent with visual evaluations of recruitment events and growth, cross-correlations of wavelet decompositions were not significant at any temporal scale, suggesting that whatever synchronizes recruitment and growth is not a periodic phenomenon (such as El Niño Southern Oscillation, ENSO). For this synchrony analysis, we used data from the 18th century, even though there were fewer than 10 trees at all sites at some point in that century. These models should be less sensitive than the recruitment and growth models above to single tree outliers because they are simple correlations.


General Trends

Patagonia and Montana ATEs differed markedly in structure across the altitudinal gradient (Fig. 1). At the Patagonia’s ATEs, we found that canopy openness remained quite similar along most of the altitudinal gradient and then increased abruptly (from c. 20% to 70% in few altitudinal metres, Fig. 2); in a similar pattern, tree height dropped abruptly once tree line altitude is reached, from an average height of about 4 to <1 m in under 25 m of altitudinal distance (Fig. 2). At the Montana’s ATEs, canopy openness increased monotonically to values of about 70% across the entire ecotone, whereas tree height was reduced also gradually from heights of c. 20 m at the lowest elevations down to c. 1 m at the tree line (Fig. 2).

Figure 2.

 Canopy openness and tree height means (±SE) as a function of altitudinal distance from the outpost tree line are shown for both tree line regions. The Patagonia tree line is composed of Nothofagus pumilio, while the Montana tree line is composed of Pinus albicaulis. The relationship between tree age and altitude is not linear.


The more complex and flexible linear, quadratic, cubic (LQC) model had unequivocal support (AICs: L = 3661, LQ = 3653, LQC = 3627). This LQC model for recruitment described well the variability in the data (R2 = 0.43). In general, higher elevation trees were younger (main effect of elevation, F = 14.4, = 0.0001; Fig. 3). Recruitment at El Fraile and Gravelly sites moved upwards with nonlinear pulses (main effect of elevation cubed, F = 4.4, = 0.037; Fig. 3; see Table S1 in Supporting Information for full anova table); at Pioneer, there was a band of oldest trees in the middle (Fig. 3). Only Cerro Castillo showed gradual upslope recruitment over time. At the Montana ATEs, the significant quadratic terms showed that there are younger trees below bands of older trees. In Pioneer site, however, the oldest trees (over 100 years older than trees at the top or bottom elevations) were found at middle elevations. The oldest individuals at both sites reached more than 600 years. The age of the oldest N. pumilio trees was about 300 years in Cerro Castillo and 350 in El Fraile sites, which is within the range for this species at other locations (e.g. Piper & Fajardo 2011); at the very (abrupt) tree line, tree ages were about 40 and 70 years for Cerro Castillo and El Fraile, respectively.

Figure 3.

 Estimated year of recruitment by elevation, showing fitted model and 95% confidence intervals. ‘Linear’, ‘Quad’ and ‘Cubic’ represent the particular site X linear elevation, quadratic elevation or cubic elevation interaction effect. < 0.1, *< 0.05, **< 0.01, ***< 0.001. CC, Cerro Castillo; EF, El Fraile; GR, Gravelly; PM, Pioneer.


A total of 348 trees and more than 500 cores were used to compute tree-ring chronologies that spanned from 13 (N. pumilio and P. albicaulis at the tree line) to 620 years (P. albicaulis in Pioneer site below tree line altitude). The LQC model was clearly superior to the other simpler models (AICs: L = 61621, LQ = 60931, LQC = 60409). The LQC was highly successful in describing the variation (R2 = 0.894), in part due to the smooth fits for ontogeny and size (See Table S2 for full anova table). From this model, detrended tree growth (dBAI, cm2 year−1) increased with time at all sites (a positive slope on the linear year term, β = 0.0038 ± 0.00031, < 1.0 × 10−16), indicating that over the past 2+ centuries the overall trend was for improving growth. However, the main effects of quadratic year and cubic year were negative (β = −0.027 ± 0.0078, P = 0.0006; β = −0.17 ± 0.016, < 1.0 × 10−16, respectively), indicating highly significant recent declines overall. Tree growth at lower elevations had even stronger improvements over time in Patagonia than in Montana, but has recently started to decrease quite dramatically (dark symbols in Fig. 4). Patagonia tree line trees have had improved growth over time, a pattern which is beginning to disappear (light symbols in Fig. 4). Growth at the Montana tree lines has begun to decline fairly dramatically in the past 50 years, with trees in lower elevations declining even earlier, with a possible lower plateau at Gravelly and a sharp decline in Pioneer. Looking at model predictions for tree line, middle- and low-elevation individual trees, these trends (and model residuals) are clearer in the longer time series of older trees (Fig. 5). Essentially, over two centuries, trees at and near tree line in two sites across in continents had improving growth over time; these improvements started to reverse around 50 years ago in Montana and at lower elevations in Patagonia first and then are spreading to all trees.

Figure 4.

 Trends of tree growth (detrended basal area increment, dBAI in cm2 year−1) of Nothofagus pumilio at abrupt (Cerro Castillo and El Fraile, c. 46° SL Patagonia, Chile) and Pinus albicaulis at diffuse (Gravelly and Pioneer, c. 46° NL Montana, USA) tree line ecotones through time and across altitude. Dots represent raw annual growth data and grey-scale colours go from the lowest tree line ecotone limit in dark towards tree line altitude in white in intervals of c. 25 for Patagonia and c. 40 years for Montana. Lines represent, in the same grey-scale significance, the increasing (positive) or decreasing (negative) slope of growth increment through time at different altitude ranges. Note the growth trend turning point for the diffuse P. albicaulis tree line ecotones around 1940. CC, Cerro Castillo; EF, El Fraile; GR, Gravelly; PM, Pioneer.

Figure 5.

 Predicted tree growth (detrended basal area increment, dBAI in cm2 year−1), with residuals and predicted confidence intervals, for a sample of the three longest lived trees at each site at three elevations on the transects (tree line, middle and lower). CC, Cerro Castillo; EF, El Fraile; GR, Gravelly; PM, Pioneer.

Process Synchronization

Synchronization between tree growth and recruitment at the upper ATE was negative at all sites with a 10- to 50-year lag, that is, recruitment events happened 10–50 years after poor growth years (Fig. 6, upper graphs). Taking into account the underestimate of recruitment date (due to coring above the root collar), this suggests that bad periods for growth are good periods for recruitment at tree line. At lower elevations, that is, in the closed canopy forests, a wide range of synchrony patterns are visible and include long-term positive and negative gradients (Montana sites) and periodic relationships with positive and negative lags (Patagonia sites). When examining all elevations together, but divided by century (Fig. 7), synchronizing patterns are shifting at all sites. Indeed, the synchronies that occurred in the 18th and 19th centuries appear to have changed in the 20th century. The fundamental pattern of synchrony in Montana in the 19th century shifted from a long-term gradient with positive synchrony between recruitment and growth to more or less independent processes. In Patagonia, a general loss of synchrony also occurred from the 18th and 19th centuries to the 20th century, although this is less marked than in Montana. The synchrony tended to be negative and lagged in the earlier centuries and was more or less absent in the 20th century.

Figure 6.

 Synchrony (cross-correlation coefficient) between recruitment years and mean detrended basal area increment (dBAI) at tree line (0 to −70 m) and below tree line (less than −70 m). dBAIs come from applying a gamma function (GAMM) on data subsets of trees from upper −70 m (i.e. below tree line trees, upper graphs) or all trees below −70 m (lower graphs). X-axis represents lags of tree recruitment from −50 to +50 years. We chose −70 m from tree line (0 m) as a trade-off between sample size and desire to only measure tree line trees. With time, the mean dBAI continues to increase in sample size as they establish. Blue horizontal dotted lines represent the significant correlation threshold (< 0.05). Number of recruitment events is shown in each graph. Negative correlation indicates that bad growth years tend to correspond to years with recruitment events. CC, Cerro Castillo; EF, El Fraile; GR, Gravelly; PM, Pioneer.

Figure 7.

 Synchrony (cross-correlation coefficient) between recruitment years and mean detrended basal area increment (dBAI) from each of the past three centuries. dBAIs come from applying a gamma function (GAMM) on data subsets of trees from each of the past three centuries (upper graphs: 18th C, middle graphs: 19th C, lower graphs: 20th C). X-axis represents lags of tree recruitment from −50 to +50 years. Seventy metres was chosen as a trade-off between sample size and desire to only measure tree line trees. With time, the mean dBAI continues to increase in sample size as they establish. Grey horizontal dotted lines represent the significant correlation threshold (< 0.05). Number of recruitment events is shown in each graph. Negative correlation indicates that bad growth years tend to correspond to years with recruitment events. CC, Cerro Castillo; EF, El Fraile; GR, Gravelly; PM, Pioneer.


Multicentury Tree Growth Increased Until Recently

Trees located at and near tree line of four structurally contrasting and geographically distant ATE sites showed increases in growth over the past 2+ centuries. Over the recent 50 years, however, these increasing growth rates reversed (Patagonia) or have begun to reverse (Montana). Other studies have reported similar increasing growth over longer time at ATEs including N. pumilio at mid-latitudes (c. 40° SL) in Chile and Argentina (Villalba et al. 2003; Daniels & Veblen 2004; Lara et al. 2005), Pinus cembra and Larix decidua in the Italian Piedmont (Motta, Morales & Nola 2006) and in the French Alps (Rolland, Petitcolas & Michalet 1998). Paulsen, Weber & Körner (2000), however, did not find historical tree growth trend differences for these two species in Switzerland and Austria, and Lloyd & Fastie (2002) even found a decrease in tree growth of Picea glauca in Alaska. There is no such a previous documentation for P. albicaulis. In general, increases in tree growth at European ATEs were considered to have begun after the so-called little ice age, that is, mid-19th century (Carrer & Urbinati 2006; Wieser et al. 2009) clearly not related to anthropogenic global warming.

As with growth increases, recent declines in tree growth have been reported across the globe (Boisvenue & Running 2006), some of which have been linked to temperature-induced drought stress (Allen et al. 2010). For example, Fagus sylvatica (a species belonging to a family phylogenetically closely related to N. pumilio) showed drought-induced growth declines in Mediterranean forests of Spain and Italy (Piovesan et al. 2005; Jump, Hunt & Peñuelas 2006), particularly at the lower elevation limit of the species’ distribution. Although we did not have a sufficiently long time period of data on temperature and precipitation to be worth considering in our multicentury study, a steadily increasing trend of mean annual temperatures since 1932 has been reported for southern Patagonia (Rosenblüth, Fuenzalida & Aceituno 1997; Villalba et al. 2003). In general, the recent growth declines at our sites could be caused by several, non-mutually exclusive drivers. A negative BAI trend is a strong indicator of a true decline in tree growth, which can be attributed to an increase in competition (Duchesne, Ouimet & Houle 2002) and climate change, by intensifying water stress in trees (e.g. drought), or through the increasing occurrence of disturbances such as insect outbreaks (Pedersen 1998; Hogg, Brandt & Kochtubajda 2002; Jump, Hunt & Peñuelas 2006). However, it has been found elsewhere (e.g. Pinus ponderosa) that older trees are even more responsive in growth to increasing CO2 than younger trees in the last century, particularly because older trees have exhibited higher intrinsic water-use efficiency (Knapp & Soulé 2011). Thus, in lower elevation trees, the general decline could be (i) due to changes in tree density over time, (ii) due to disturbance (e.g. white pine blister rust, Cronartium ribicola for P. albicaulis) (Tomback & Resler 2007) or (iii) directly due to anthropogenic climate change. We did observe some blister rust on some of the trees we examined in the Montana sites; however, the influence of these on growth is likely very recent (i.e. the last 10 years) as infected trees decline slowly and die within 20 years (Kendall & Keane 2001) and we only sampled vigorous trees. Since temperatures are likely increasing at our locations, and we removed age and size effects in our analysis, we speculate that drought may be an important driver of these recent decreases in growth.

Upward Increase in Tree Recruitment

For tree lines to be considered to be moving upward, tree ages must minimally become younger with elevation. We found evidence at all sites that there is an upward movement. In northern Patagonia, Daniels & Veblen (2004) found that an increase in temperature did not result in an upslope expansion of N. pumilio. We found no dead stems at any site except Pioneer, so this upward movement is new within the context of several centuries. The Pioneer site had some dead trees, but these were all below the band of old trees in the middle. This suggests that the younger trees at lower elevations were established after a mortality event, such as a fire or insect outbreak. We did not find very many seedlings or saplings above the tree line as we defined it (the highest coreable tree), but since we were not able to assess growth increment and recruitment dates on these, they were not included in our study. It is thus possible that recruitment is continuing upslope, although with apparently low densities. This pulsed, historical recruitment upslope indicates that it will be difficult to predict future recruitment dynamics, certainly without nonlinear relationships.

Synchronization of Processes

From a naive perspective that tree line recruitment and growth are limited solely by low temperature, both should respond in a common way to variation in temperature. We found little evidence of this. Only during the 18th and 19th centuries at one site in Montana was there a clear positive relationship between these two processes. More commonly, there was a negative synchrony indicating that these factors are responding to different cues. We are careful to not discuss the lag that we observed as we cannot be certain that it is true lag or a data lag due to the biased estimate of recruitment date. Thus, while it is plausible to expect a positive relationship between temperature and recruitment or growth at tree lines, when we examine these processes at the same sites, in non-experimental conditions, over the same time periods and with the same analysis, it is clear that these two processes only rarely respond in the same way to the same conditions. Both species, N. pumilio and P. albicaulis, are mast seeding species (Veblen et al. 1996; Crone, McIntire & Brodie 2011), meaning that they follow synchronous, episodic reproduction events (Kelly 1994). In this respect, we could expect a negative growth–recruitment synchronization given the trade-off between growth and reproduction, although with this mechanism, this synchronization would be at a shorter temporal scale, that is, interdecadal.

To the best of our knowledge, there is no previous study examining the synchronization of growth and recruitment at the ATE over such a long period of time (three centuries). At mid-latitude areas of Chile and Argentina, Daniels & Veblen (2004) found that radial growth of N. pumilio krummholz responded differently to interannual climate variation than seedling recruitment in the last half-century and that the temperature–growth relationship is not linear given that temperature interacts with moisture availability. There are some previous studies suggesting that a considerable lag time exists between changes in climate and the related pulse of tree establishment (MacDonald et al. 1998; Lloyd 2005). In our study, we did not really find this. Indeed, we found concurrent synchrony as well as lagged synchronies in both positive and negative directions. In the 18th and 19th centuries in Gravelly, we found long periods that were good for both growth and recruitment (large periods of positive synchrony). Essentially, this direct assessment of synchrony between processes – rather than two parallel assessments for each of growth and recruitment – shows that the dominant limiting factor at tree lines is most definitely not the same for growth and recruitment. This does not mean that a single factor, say temperature, would not be a detectable driver if we had done the separate analyses, but it does mean that, as a rule, it is not the limiting factor for both.

Tree Line Form: Little Support for Current Models

The N. pumilio tree lines are abrupt and have a relatively strong, oceanic climate in the southern hemisphere; the P. albicaulis tree lines located in the northern hemisphere are diffuse and have a strong continental climate influence. Although the objective of this study is not a trans-hemisphere comparison (we lack proper replicates), these structural ATE differences may contribute to some of the tree growth, recruitment and processes synchronization differences we observed. Regional analyses have revealed an increase in tree line elevation from oceanic to continental climates (Fang, Oshawa & Kira 1996; Daniels & Veblen 2003), which may convey a climatic differentiation between tree lines around the world (Jobbágy & Jackson 2000). At equivalent latitudes, summer heating is less pronounced in the southern hemisphere, and therefore, tree lines tend to be at lower elevations (Daniels & Veblen 2003). It has been suggested that strong abruptness of a tree line, like the one of Nothofagus, may indicate that it will respond slowly or not to climatic change (Harsch & Bader 2011). We found no evidence to support this idea. In contrast, north hemispheric tree lines are mainly composed of coniferous evergreen species and constitute diffuse tree lines where changes in structure are gradual with altitude (Körner 2003). According to Harsch & Bader (2011), it is expected that an increase in temperature should increase growth rates and recruitment in diffuse tree lines more than in other tree line forms. We found, at best, partial support for this as growth is declining. Recruitment appears to be moving upslope, but we could not confirm this over the last 40 or so years. In general, we found that tree growth appeared to be very dynamic and was responding negatively to global warming in all of our tree line sites, that is, it was independent of tree line form.


This work was supported by the Chilean Fondo Nacional de Desarrollo Científico y Tecnológico (FONDECYT) post-doctoral project 3070050 to A.F. Financial support to E.J.B.M. was received from the Natural Sciences and Engineering Research Council of Canada and the Canada Research Chair program. We thank the Corporación Nacional Forestal (CONAF) for facilitating access to Reserva Cerro Castillo. Finally, we thank A. Holz and J. Paritsis for commenting an earlier version of the manuscript.