Shifting with climate? Evidence for recent changes in tree species distribution at high latitudes

Abstract

Warming has been particularly strong at high latitudes in recent decades and bioclimatic models predict northern shifts in optimal conditions for most species. Climate is a strong predictor of site occupancy for trees at broad spatial scales and interacts with other drivers of forest dynamics. Recent changes in distribution and occupancy patterns should therefore provide the best evidence of a tree species' potential to shift in the direction predicted by bioclimatic models. Studies examining recent distribution changes for plants, however, have mostly done so along altitudinal gradients or have used the latitudinal position of juvenile trees relative to adult ones to infer range dynamics. This study provides rare evidence of latitudinal shifts for 11 northern tree species by assessing recent changes in distribution using globally significant inventories from 1970 to 2002. It also compares observed trends with those inferred from the position of juveniles relative to trees in a single survey. Samplings cover 6456 forest plots in temperate and boreal forests up to treeline in eastern North America. The average overall latitudinal shift was 3.07 ± 4.37 km northward although responses were species-specific. Shifts were detected more for juvenile than for adult trees and significant northward ones were detected more at northern range limits than at the median. All species demonstrated increased frequency of plot occupancy for saplings while occupancy generally decreased for adult trees. Five out of the 11 species examined (Acer rubrum, Acer saccharum, Betula papyrifera, Fagus grandifolia, and Populus tremuloides) showed significant distributional shifts consistent with northward migration. Saplings of Abies balsamea, Picea glauca, and Picea mariana, on the other hand, showed southward shifting trends. Natural and human disturbances undoubtedly interact with climate to determine forest dynamics; this study shows whether their combined effect can shift distribution in the direction predicted by bioclimatic models. Only continued monitoring will reveal whether these observations are just transient dynamics or indicative of shifting range in this century. Our study provides a benchmark against which to assess future observations of latitudinal shifts for trees.

Introduction

Climate warming is particularly strong at northern latitudes (Xu et al. 2013) and is projected to amplify through this century. For trees, climate variables are strong predictors of site occupancy (Canham and Thomas 2010), correlate with recruitment patterns (Elliott 2012), and determine northern range limits (Morin et al. 2007). Northward shifts of species optimal climate conditions of hundreds of kilometers are therefore projected for this century in the northern temperate and boreal forests (McKenney et al. 2011, Chambers et al. 2013, Berteaux et al. 2014, Périé et al. 2014). It is still doubtful, however, whether trees will establish at new sites fast enough to track changing climate conditions since most projections do not take into account constraints on dispersal and establishment which are likely to mediate actual shifts (Clark et al. 1998, Hampe 2011) and which can act differently on species. Paleorecords indeed suggest that differential migration rates will result in a major reorganization of forest communities as the pace of change quickens (Davis and Shaw 2001). Empirical evidence indicating whether tree species are currently shifting their distribution and whether that shift reveals the potential to track climate change is therefore crucial to improve predictions of the effects of climate change on forest biodiversity. That potential will arise from the interaction between species' traits and individual response to warming, short and long-distance dispersal events, site conditions, and disturbances providing opportunities for recruitment (Clark et al. 1998, Morin and Thuiller 2009, Cleland et al. 2012, Gray and Hamann 2013, Steenberg et al. 2013).

Given that warming is well under way at high latitudes, recent distribution changes in space and time should provide the most direct evidence of a species' potential to shift in the direction predicted by bioclimatic models even if a direct causal link with climate cannot be established (Beckage et al. 2008, Lenoir et al. 2008, but see Chen et al. 2011). Changes in recruitment patterns should be detectable especially for northern tree species, but tracking the movement of trees over broad spatial scales and large climatic gradients poses a challenge. For this reason, studies examining contemporary distribution changes for plants have usually done so along the compressed scale of altitudinal gradients, as opposed to latitudinal ones. In a meta-analysis of 26 studies examining recently reported range shifts, the seven studies that examined plants were all in relation to altitudinal shifts (Chen et al. 2011). These studies have generally shown upslope migration of alpine plant distributions consistent with observed warming, but not for all species (Beckage et al. 2008, Kelly and Goulden 2008, Lenoir et al. 2008). Alpine species, however, are restricted in distribution and preferences, and may not be representative of the responses of species with larger ecological tolerances and ranges or species under management. Lowland plant communities may in fact respond differently to warming compared to mountains' ones (Bertrand et al. 2011).

In spite of its importance to assess forest ecosystems' shifts, direct evidence of broad-scale latitudinal changes in plant species distribution is still limited. Studies have documented increasing plant abundance in the tundra (Elmendorf et al. 2012) or local migration at northern range margins (Asselin and Payette 2006). A compilation of local studies of treeline responses across 126 alpine and 40 latitudinal sites concluded that most treelines had advanced or remained stable (Harsch et al. 2009). Broad-scale assessment of tree latitudinal shifts over recent time remains difficult, however, because of insufficient data coverage (Shoo et al. 2006), lack of consistency in survey methodology (Woodall et al. 2008), and/or imprecise recording of survey locations (Tingley and Beissinger 2009). To circumvent the lack of temporal data, recent studies have compared the spatial distribution of juvenile trees (e.g., seedlings or saplings) with that of mature trees of the same species (Woodall et al. 2009, Zhu et al. 2012, Woodall et al. 2013, Zhu et al. 2014) for a given time period. In these studies, juvenile trees are seen as being indicative of recent and possibly future migration trends, whereas mature trees are indicative of past conditions. Contradictory patterns have been revealed by these studies depending on the species or the portion of the range examined. What has not been directly examined, though, is whether the spatial position of juveniles relative to trees actually translates into sustained range extension (or contraction) through time. Long-term monitoring data are invaluable in this context as they can contribute to validate observed spatial trends from the relative positions of trees and juveniles as well as those predicted from species distribution models or patterns of tree abundance (Murphy et al. 2010).

Our study fills a gap in the climate change/biodiversity literature by assessing changes in latitudinal tree distribution through time and across a broad geographical area, directly, using inventories from 1970 to 2002, and indirectly, using the position of juveniles relative to trees as an indication of recent and possibly future range dynamics. The strongest warming occurred in recent decades (Marcott et al. 2013) and is shifting optimum conditions northward for most tree species (Chambers et al. 2013). As a consequence, we expect to detect spatially-explicit changes in frequency of occurrences for a given species resulting in northern latitudinal shift. Such evidence would provide early signals of the potential to track climate change especially in northern forests where disturbances may interact with climate to provide opportunities for recruitment. Given the relatively short time scale of our observations, these signals should be detectable more for juvenile than for adult trees and at northern limits, where climate constrains establishment (Payette 1993), more than at the median of a species distribution within the study area. Finally, latitudinal distribution limits of juveniles relative to trees in a given time period are expected to predict the direction of observed shifts.

Methods

Study area

The study area covers the province of Québec, Canada, from 45° N up to the commercial treeline around 52° N, and from 80° W to 61° W (±761,100 km²; Saucier et al. 2003). This vast area is characterized by strong climatic gradients. We used interpolated weather station data (Regnière and Saint-Amant 2008) to characterize temperature and precipitation trends in the forest plots used in this study. Average annual temperature from 1965 to 2002 ranged from 6.6°C in the south to −4.7°C in the northern portion of the study area (Fig. 1). On average, mean annual temperature climbed by 0.48°C between the two time periods considered in this study (from 1965–1977 to 1987–2002; SD = 0.1813, paired t-test, p < 0.0001). Strongest warming occurred in the southwest and center-north areas of the province. Only 0.008% of plots demonstrated a decrease in annual average temperature with changes ranging from −0.2 to +1.19°C (Fig. 2). Annual precipitation ranged from 727-1512 mm, decreasing in the study area from south to north and from east to west (Fig. 3). Spatial patterns of precipitation change were more varied (Fig. 4): drier conditions were found in the center of the province and the Lower North Shore (south of Labrador) whereas wetter conditions were especially concentrated to the center-north. Climate change is expected to bring much warmer and wetter conditions in the winter with more moderate changes in the summer (Logan et al. 2011). Climate models project an increase of 1.8–2.7°C in the summer and 3–5°C in the winter for Quebec by 2050, as compared to 1971–2000 (Logan et al. 2011).

Figure 1

Mean annual temperature (at plot level) between 1965 and 2002. Values based on interpolated models from weather station data provided by Environment Canada (Regnière and Saint-Amant 2008). Colors represent interval level of mean temperature for a given plot, from the coldest temperatures (blue) to the warmest (red).

Figure 2

Difference in mean annual temperature (at plot level) between 1965–1977 and 1987–2002. Values based on interpolated models from weather station data provided by Environment Canada (Regnière and Saint-Amant 2008). Points represent one pair of plots (main and satellite). Colors represent interval level of differences for a given plot, from the negative differences (blue) to the largest positive differences (red).

Figure 3

Mean annual precipitation (at plot level) between 1965 and 2002. Values based on interpolated models from weather station data provided by Environment Canada (Regnière and Saint-Amant 2008). Points represent one pair of plots (main and satellite). Colors represent interval level of mean precipitation for a given plot, from the driest annual precipitation (dark red) to the wettest (dark blue).

Figure 4

Difference in mean annual precipitation (at plot level) between 1965–1977 and 1987–2002. Values based on interpolated models from weather station data provided by Environment Canada (Regnière and Saint-Amant 2008). Points represent one pair of plots (main and satellite). Colors represent interval level of mean precipitation for a given plot, from the largest negative difference (dark red) to the largest positive difference (dark blue).

Two major vegetation zones, each with two forest types, are part of the study area: the northern temperate zone and the boreal forest zone. The former includes Acer saccharum dominated forests to the south and mixed Abies balsamea-Betula alleghaniensis forests to the north, whereas the latter includes A. balsamea-B. papyrifera forests to the south and vast Picea-moss forests to the north (Saucier et al. 2003). The northern limit of the study area coincides with the ecotone between the end of continuous boreal forest and the beginning of the taiga. Historically, even-aged management has been the dominant harvesting method used in Quebec followed by uneven-aged management and pre-commercial thinning (Youngblood and Titus 1996, Canadian Council of Forest Managers 2013). A new forest regime became law in 2013 with the aim of sustainably managing forested areas. Management will have to take into account climate change.

Species data

We obtained species data from the Ministère des Forêts, de la Faune et des Parcs (Ministry of Forests, Wildlife and Parks; MFFP). They conducted exhaustive inventories of a network of circular permanent plots measuring 400 m² set up to monitor long-term forest dynamics throughout the study area (Duchesne and Ouimet 2008; Figs. 1 and 2). Quebec encompasses about 2% of all forested areas in the world and a significant proportion of all boreal forests (Saucier et al. 2003; MFFP, http://www.mrn.gouv.qc.ca/english/forest/understanding/understanding-forest.jsp). Data collection started in 1970 and is ongoing. Only plots for which tree species were completely surveyed in two distinct time periods (first period: 1970–1977; second period: 1992–2002) were retained for analysis. This corresponded to 6456 permanent plots out of 12,409 currently monitored. The number of years between plot sampling in the first and second time periods varied between 15 and 31 years, averaging 22.4 ± 2.7 years. Nearly 3/4 of plots were resampled after 21–25 years so if there is some temporal discrepancy over space, its effect on overall forest dynamics is expected to be minimal. These plots capture the northern part of tree species' range in eastern North America, where we would expect strong climate effect, and not their entire north-south range, which extends south in the United States. About a third of the plots (29.1%) had been affected by natural (windthrow, fire, insect outbreaks) and/or anthropogenic disturbances (selective or complete logging) when the plots were established increasing to 37.2% in the second period. Plots are usually paired for sampling on the landscape at an average distance of 425 m, as they represent one day of work for a field team. Plot density decreases from south to north, depending on the ecological domain sampled (1 pair per 26 km² in the A. saccharum forest, 1 pair per 104 km² in the A. balsamea forest and 1 pair per 259 km² in the Picea mariana forest), This is to take into account the fact that tree species diversity decreases northward through the domains until the large majority of stands are mostly dominated by black spruce. It is possible that reduced plot density going north introduced a sampling bias in our study, especially for the northernmost species, resulting in calculated limits south of their actual location. However, the geographic distribution of plots and sampling effort remained the same for the two time periods as the same plots were re-sampled in this study. The same bias would therefore have affected our calculated limits similarly.

According to the MFFP classification, a merchantable tree has a diameter at breast height (DBH) of at least 9.1 cm whereas saplings range from 1.1 to 9.0 cm DBH, grouped by 2 cm DBH classes, and are at least 1.3 m tall. The smallest class of saplings available (1.1–3.0 cm DBH; hereafter called “saplings”) was retained as the juvenile life stage in this study. The merchantable tree life stage (DBH ≥ 9.1 cm; hereafter called “trees”) was also analyzed. Saplings of the smallest size class are indicative of successful establishment events, whereas seedlings are susceptible to stochastic processes resulting in large annual variation and may be more susceptible to misidentification.

For each species, four combinations of time period/life stage classes were analyzed: first period saplings (S₁), second period saplings (S₂), first period trees (T₁), second period trees (T₂). Eleven species (Abies balsamea, Acer rubrum, Acer saccharum, Betula alleghaniensis, Betula papyrifera, Fagus grandifolia, Ostrya virginiana, Picea glauca, Picea mariana, Populus tremuloides, Thuja occidentalis) with the northernmost part of their range within the study area (under 53°N) and with at least 50 occurrences in each of S₁, S₂, T₁ and T₂ were selected (Little 1971, Soper and Heimburger 1990, Flora of North America Editorial Committee 1993, Farrar 1995). Only presence/absence data were used because climate predicts frequency of occurrence at different latitudes much better than it predicts local abundance in a plot (Canham and Thomas 2010, Chambers et al. 2013). Therefore change of occupancy of a plot (presence/absence) can provide a stronger signal for range shift than change in tree abundance in that plot. Note that change in occupancy in a plot can result from various demographic processes, including recruitment, aging, or death. For occupancy changes to affect latitudinal shifts, they need to be distributed non-randomly over the study area (i.e., show directional trends; see below).

Assessing latitudinal shifts

In this study, latitudinal shift, i.e., change in the location of latitudinal limits for a species, is seen as an indication of potential for range shift. The first step was to define latitudinal limits within the study area for each species. Distributional dynamics have been shown to vary depending on what part of the range is under examination and the method used to calculate range limits (Woodall et al. 2009, Zhu et al. 2012). Range edges are controlled by distinct demographic processes driving differential spatiotemporal changes (Purves 2009, Sexton et al. 2009). To assess whether northern distribution dynamics vary depending on the location examined, latitudinal limits were defined in the median (50th percentile of latitude) and northernmost (90th percentile of latitude) part of the distribution within the study area. The percentile approach was chosen to define limits because it reduces the potential bias of extreme latitudinal data points (Lenoir et al. 2009). We used two distinct methods to calculate the location of each of these limits. At the median, all occupied plots in the study area were used to calculate the 50th percentile of latitude for S₁, S₂, T₁ and T₂ (Woodall et al. 2009). At the northernmost limit in our study area, tree species tended to show distinct spatial patterns along an east-west gradient often resulting in distributional limits at greatly differing latitudes. To take these spatial patterns into account, the study area was divided into 38 0.5° longitude-wide bands between the western and eastern limit (80–61°W) of Quebec (Zhu et al. 2012). A window of two bands (1°) spanning the latitudinal breadth of the study area (45–53°N) was then used to calculate 90th percentile of latitude in 0.5° steps to smooth out sharp differences from one band to the next. Bands without plots were subsequently discarded. The average 90th percentile of latitude across the remaining longitudinal bands weighted by the number of occupied plots in each band was then calculated for S₁, S₂, T₁ and T₂. Note again that, because of the northern location of our study area, the 50th percentile of latitude does not reflect the median latitude of the entire north-south range of a species but the median for the study area. The 90th percentile represents the extreme northern edge inside the study area which, for most species, is their northern limit in eastern North America. Consequently, the 50th percentile analyses also give insight into northern spatiotemporal trends of our selected species but away from the leading edge. It is used further to determine consistency in changes in occupancy and overall distribution changes in the study area.

The second step was to assess latitudinal shifts, which was done (1) by comparing latitudinal limits in time for a given life stage (saplings or trees), and (2) by comparing, in space, latitudinal limits of saplings relative to trees of the same species for a given time period. In the first case, the differences between the latitudinal limits of the second time period and the first one were computed for each life stage and at each latitude percentile: sapling shifts (S_i+1,j–S_i,j) and tree shifts (T_i+1,j–T_i,j), where i is the time period (i = 1 (1970–1977) or i + 1 = 2 (1992–2002)) and j, the percentile of latitude (j = 50th or 90th). Positive values indicate that the latitudinal limit shifted northward between the two time periods whereas negative values indicate that the latitudinal limit shifted southward.

In the second case, the differences between the latitudinal limits of saplings and trees, S_i,j–T_i,j, were calculated for each time period i (i = 1 (1970–1977) or 2 (1992–2002)) and percentile j. Positive values in S_i,j–T_i,j occur when the latitudinal limit of saplings is north of the limit of trees, whereas negative differences result when the limit of trees is north of the limit of saplings. Moreover, if the spatial trends observed in the first time period (S_i,j–T_i,j) are indicative of ongoing or future range dynamics, they should predict the direction of the trends actually observed between the time periods for saplings (S_2,j–S_1,j) and/or trees (T_2,j–T_1,j).

For latitudinal shifts, significance was assessed depending on how latitudinal limits were calculated. At the 50th percentile, bootstrap resampling was used with n = 500,000 iterations (Efron 1979). Significance was assessed at the 0.05 level. To take into account the unequal sample sizes resulting from unoccupied bands and the lack of independence between the two samples at the 90th percentile, pooled variance was calculated on the computed latitudes from occupied bands and then a 95% confidence interval was calculated using a t distribution. A latitudinal shift was significant if the upper and lower bounds of the confidence interval did not include zero.

To examine whether spatial differences in latitudinal limits between saplings and trees predicted observed latitudinal shifts in time, simple linear regressions were performed for both latitude percentile (50th or 90th) with S_i,j–T_i,j as the predictor variable and either S_2,j–S_1,j or T_2,j–T_1,j as the response variable. Positive regression coefficients would indicate that sapling limits relative to trees predicted the direction of observed latitudinal shifts in time, whereas negative coefficients would indicate opposite trends.

Changes in occupancy

Detailing changes in plot occupancy provides further indication of the spatial processes that have led to changes in latitudinal limits as different processes may result in similar overall latitudinal patterns. To assess the contribution of these spatial processes to observed latitudinal shifts, we quantified changes in plot occupancy using latitudinal limits from the first time period. Latitudes of S₁ and T₁ for the 50th and 90th percentile were used to obtain the number of plots occupied (1) above the 50th percentile (>50th), (2) below the 50th percentile (<50th) and (3) above the 90th percentile (>90th) for S₁, S₂, T₁ and T₂. From these, absolute differences between the number of plots occupied in the first and second time periods and the ratio of change in the number of plots occupied relative to the first time period were calculated for both saplings and trees. McNemar's test for paired samples was used to see if the frequency of presence/absence was statistically different (P < 0.05) between the two decennials for a given life stage at each location.

Overall changes in distribution patterns

Finally, latitudinal shifts and plot occupancy were interpreted together to give a more complete picture of spatiotemporal trends in our study area. Distributional dynamics were categorized depending on how occupancy patterns varied in relation to observed latitudinal shifts through time. Relative to a species' distribution within the study area, positive latitudinal differences may result from several not necessarily mutually exclusive processes: (1) thinning (i.e., decrease in the number of occurrences) south of the latitudinal limit (at lower latitudes), (2) filling (i.e., increase in the number of occurrences) north of the latitudinal limit (at higher latitudes), or (3) expansion of range limit (or dispersal) beyond the current northern limit. In all cases, this suggests responses consistent with a northward migration of suitable conditions. In our study area, negative latitudinal differences may result from: (1) filling in the south, (2) thinning in the north, or (3) loss of the northernmost observations. A species demonstrated an overall significant change if both latitudinal shifts (50th and/or the 90th) and plot occupancy change (below 50th, above 50th, and/or above 90th) were significant.

Results

Latitudinal shifts

When measuring latitudinal shifts between the first and second time periods, results indicate mostly northward trends in distribution but vary depending on species, life stage, and latitudinal position. The average overall shift was 0.0276 ± 0.0393° (∼3.07 ± 4.37km, 1° latitude = 111.2 km at 49° N; Figs. 5 and 6) with 63% of calculated latitudinal shifts through time and 61% of the statistically significant ones being positive (7.79 ± 11.26 km; Fig. 5). Northward shifts were more pronounced through time at the northernmost range limits (average shift₉₀ = 4.1 ± 1.79 km) than at the 50th where, although the majority of calculated differences were positive, the average trend was southward (average shift₅₀ = −1.49 ± 1.64 km). This was mainly due to the significant southward shifts observed for northern species like Abies balsamea (S₅₀), Picea glauca (S₅₀, S₉₀), Picea mariana (S₅₀), and Populus tremuloides (S₅₀). Among the significant shifts, the northward ones were more often observed at the 90th, whereas the southward ones were more often at the 50th. For a given life stage, shifts were in the same direction (northward or southward) at the median and northernmost limits, with a few exceptions (P. tremuloides S₅₀ and T₅₀; Acer rubrum T₉₀). Several species, however, showed trends in opposite directions between saplings and trees (50th and 90th: A. balsamea, Betula papyrifera, Fagus grandifolia, P. glauca, P. mariana, P. tremuloides; 50th only: A. rubrum; 90th only: Ostrya virginiana) (Fig. 5). Overall, significant shifts were more often detected for saplings than for trees; eight species (A. balsamea, A. rubrum, A. saccharum, B. papyrifera, F. grandifolia, P. glauca, P. mariana, P. tremuloides) showed significant shifts through time for saplings at the median and/or northernmost limits, whereas for trees only P. glauca showed a significant northward shift (at the 50th).

Figure 5

Latitudinal shifts (in kilometers) between the first and second time periods for saplings (blue = at the 50th percentile of latitudinal distribution, red = at the 90th percentile) and trees (green = 50th, purple = 90th) of a given species. Columns with an asterisk at the top indicate statistically significant shifts. Latitudinal coordinates and differences can be found in Appendix: Tables A1 and A2 (50th) and A3 (90th), respectively.

Figure 6

Differences (in kilometers) in the latitudinal limits of saplings vs. trees for a given species during the first time period (blue = 50th percentile of latitudinal distribution, red = 90th percentile) and the second time period (green = 50th, purple = 90th). Columns with an asterisk at the top indicate significant differences. Latitudinal coordinates and differences can be found in Appendix: Tables A1 and A2 (50th) and A3 (90th), respectively.

When using latitudinal limits between life stages at a given time period to infer the direction of range shifts, we found responses not always consistent with the temporal patterns observed for a given species although relationships were significant (Table 1). Among the species for which the position of saplings at the first time period compared to trees predicted expansion (i.e., latitudinal limits of saplings and trees were significantly different in the first time period and saplings were north of trees), four were expected to expand their distribution at the 50th (A. balsamea, B. alleghaniensis, P. glauca, P. mariana), one of which also at the 90th (P. mariana) (Fig. 6). Contraction, on the other hand, (i.e., latitudinal limits of saplings and trees were significantly different in the first time period and saplings were south of trees) was predicted for five species (Betula alleghaniensis, B. papyrifera, F. grandifolia, O. virginiana, P. tremuloides) and always at the 90th. In all cases, the predicted and observed trends for saplings were in opposite direction (Table 1, Figs. 5 and 6), except for B. papyrifera at the 50th. Latitudinal differences between saplings and trees at t₁ were more consistent at predicting directional shifts for trees (Table 1). When considering both time periods, patterns showed that, in general, sapling limits were north of trees at the 50th (13.4 ± 9.72 km) but, conversely, tree limits were situated at more northern latitudes than those of saplings for the 90th (−3.75 ± 3.33 km). Interestingly, the average latitudinal gap between saplings and trees decreased between the first and second time periods both at the 50th and 90th (S_1,50–T_1,50 = 17.28 ± 14.36km, S2_,50–T_2,50 = 9.52 ± 5.74 km; S_1,90–T_1,90 = −6.04 ± 4.0 km, S_2,90–T_2,90 = −1.46 ± 2.9 km).

Table 1. Results of simple linear regression models predicting the latitudinal differences in range limits between (a) saplings over time (S₂–S₁) and (b) trees over time (T₂–T₁) from the observed spatial differences between the northern latitudinal limits of saplings and trees of the first time period (S₁–T₁) at the 50th and 90th percentiles of latitude (indicated in subscript).

Changes in occupancy

For saplings, nearly two-thirds (21 of 33) of all differences in plot occupancy between the first and second time periods were significant; all of these were positive indicating a trend towards increased frequency of occurrence (Fig. 7). All species had at least one portion of the range with a significant increase in sapling occupancy while three species had significantly higher sapling occupancy in all parts of their range (A. rubrum, B. papyrifera, F. grandifolia). Trees, on the other hand, demonstrated a general trend towards decreased frequency of occurrence (Fig. 8); nearly half (16/33) of the values were significant and all but two of these values were negative. All trees exhibited significant decreased frequency of occurrences in at least one portion of their range, except for A. rubrum, F. grandifolia, and O. virginiana. Two species (A. balsamea and P. tremuloides) exhibited significant tree occupancy losses in all parts of their range. Only one species for trees, A. rubrum, showed a consistently positive trend across its distribution in the study area, with significant increases below and above the median. The ratios of change were relatively smaller for trees than for saplings, indicating that fewer plots underwent occupancy change for trees than for saplings (Figs. 7 and 8).

Figure 7

Changes in number of sites occupied by saplings between first and second time periods below (<50th: blue column) or above the 50th percentile (>50th: red column) and above the 90th percentile of latitude (>90th: green column). Calculated percentiles from the first period (saplings or trees) were used to delineate the range. Columns with an asterisk at the top indicate statistically significant differences (P < 0.05). Number of plots occupied in each time period are found in Appendix: Table A4.

Figure 8

Changes in number of sites occupied by trees between first and second time periods below (<50th: blue column) or above the 50th percentile (>50th: red column) and above the 90th percentile of latitude (>90th: green column). Calculated percentiles from the first period were used to delineate the range. Columns with an asterisk at the top indicate statistically significant differences (P < 0.05). Number of plots occupied in each time period are found in Appendix: Table A4.

Overall changes in distribution patterns

Results from latitudinal shifts through time (Fig. 5) and occupancy trends (Figs. 7 and 8) across species were summarized qualitatively to provide an overall picture of range dynamics (Table 2; species maps for saplings and trees in Appendix). For saplings, four species had significant northward shifts in conjunction with increased occupancy (filling) in at least the northern part of their range: A. rubrum, A. saccharum, F. grandifolia and B. papyrifera. Three species showed significant southward shifts combined with filling in the south: A. balsamea, P. glauca, P. mariana. One species, P. tremuloides, had strongest filling below the northernmost limit, which translated into opposite shifting trends at the 50th (negative) and 90th percentiles (positive) of latitude. Only P. glauca demonstrated significant trends for trees, with reduced frequency (thinning) in the southern portion of its range resulting in a northward shift in calculated limits. Overall, when all the range limits at the 90th are considered for a species regardless of life stages, all but four species (F. grandifolia-tree, O. virginiana-tree, P. mariana-sapling, and P. tremuloides-tree) reach their northernmost limits in the second time period (Appendix: Table A1). For three of these species—A. rubrum, A. saccharum and B. papyrifera—significant latitudinal shifts in sapling distribution suggested range expansion in the study area.

Table 2. Summary of results for latitudinal shifts through time of a given life form (Fig. 5) and changes in number of occupied plots (Figs. 7 and 8). ‘Filling' indicates an increase in the number of occupied plots, whereas ‘thinning' indicates a decrease.

Discussion

This study provides rare empirical evidence for latitudinal shifts of tree species at broad spatial scale over recent time. Factors other than climate undoubtedly influenced forest dynamics (Chauchard et al. 2010). The trends observed, however, provide some indication of whether the combined effects of these factors are shifting species distribution in the direction predicted by bioclimatic models. As expected, responses were species-specific, more easily detected for juveniles than for trees, and varied with the portion of the range examined. Significant northward shifts were detected mostly at range limits, where climate constraints are expected to be determinant (Payette 1993, Beauregard and de Blois 2014). If sapling establishment provides any indication of the potential for species to track warming trends through increased recruitment in the northern portion of their range, at least five out of the 11 species examined (Acer rubrum, Acer saccharum, Betula papyrifera, Fagus grandifolia and Populus tremuloides) showed, over decades, significant distributional shifts consistent with northward migration. The first three species listed demonstrated northward expansion in saplings in the second time period beyond their calculated northernmost limits for trees. Saplings of conifers of the boreal forests such as Abies balsamea, Picea glauca, and Picea mariana, on the other hand, appear to regenerate better in the southern portion of their range in the study area, a pattern that would suggest contraction.

Although several studies have attributed distributional shifts to climate change (Beckage et al. 2008, Kelly and Goulden 2008, Woodall et al. 2009, Zhu et al. 2012), it may be hard to disentangle the effects of natural or human-induced disturbances from those related to climate without detailed examination of each species' distribution and disturbance history. To complicate things, climate directly or indirectly influences disturbance dynamics and interacts with other drivers of changes (Caplat et al. 2008). Since the 1970s, several major disturbance events—a major ice storm in the southern portion of the study area in 1998 (Beaudet et al. 1999), spruce budworm outbreaks in the boreal forests, northward progression of logging activities (Crête and Marzell 2006, Duchesne and Ouimet 2008), among others—have all influenced local regeneration and stand dynamics. Openings created by disturbances may facilitate migration especially for shade-intolerant or successional species. Increasing relative abundance of successional tree species from warmer areas over a 30 year period, for instance, has been reported in highly disturbed power line corridors just south of our study area (Treyger and Nowak 2011). However, migrating species cannot take advantage of disturbance if climatic conditions are at the limit of their physiological tolerance. Moreover, whereas disturbance frequency undoubtedly determines whether shade-intolerant species like B. papyrifera or P. tremuloides can colonize new sites, mid- or late successional species like A. rubrum, F. grandifolia, and A. saccharum also demonstrated consistent evidence of northward shifts. This suggests that observed patterns may not be explained by broad scale disturbances or forest management alone. Quebec forests are definitely undergoing rejuvenation (Crête and Marzell 2006, Duchesne and Ouimet 2008, Ministère des Ressources naturelles et de la Faune 2009) with consequences for species composition at temperate/mixed/boreal ecotones as supported by a trend towards increased frequency of occurrence of saplings and a general decline in site occupancy for trees in our study. The interplay between climate, broad-scale natural disturbances, and management can result in directional range shifts for some species if their combined effect is spatially structured. This may be more and more the case as human activities are being carried out farther and farther north in response to warming. A modeling study from Nova Scotia suggests warming combined with harvesting could influence stand dynamics, particularly benefiting early- to mid-seral deciduous species like A. rubrum and hindering boreal trees (Steenberg et al. 2013). Disturbances, however, had only a partial influence on tree range dynamics in eastern US forests when examined over a short (5-year) time period (Woodall et al. 2013).

Finer scale observation of forest dynamics is essential to understand some of the patterns uncovered from long-term surveys over large areas. Temperate species such as A. rubrum and A. saccharum, whose limits fall near the southern border of the boreal forest, have been shown to establish in boreal forests (Barras and Kellman 1998, Leithead et al. 2010). Competition between species such as A. saccharum and F. grandifolia in temperate forests could result in a decline in site occupancy, the latter appearing to be advantaged in the shared southern portion of their range where trees of A. saccharum are showing signs of stress consistent with our data. In younger age classes, dominant radial growth shifted from A. saccharum to F. grandifolia in recent decades in Quebec (Gravel et al. 2011) possibly due to warming (Beaudet et al. 1999). In boreal forests, co-occuring P. mariana and P. tremuloides show contrasting growth response to climate, the latter being favoured in warmer, drier conditions (Drobyshev et al. 2013). Warmer, drier conditions can also result in higher fire frequency which is expected to affect the flammable boreal conifers more than deciduous species such as P. tremuloides or B. papyrifera (Terrier et al. 2013). Whereas niche models predict, perhaps simplistically, a northward migration of suitable conditions for A. balsamea, its current disjunct distribution at its northernmost range limits is thought to represent residual stands that were once part of a larger, more favorable bioclimatic zones between c. 9000 and 5000 cal. yr B.P. (de Lafontaine and Payette 2011). Recent (1901–2002) increasing (in the northern portion of our study area) and decreasing (in the southern portion of our study area) fire danger (Girardin et al. 2013) could result in southward shifts through increased frequency of occupancy for fire-vulnerable conifers in the south and northward shifts for fire-tolerant successional deciduous species, as supported by our data. The northern part of the study area, where P. mariana approaches its northern limit, has in fact experienced in 2013 its largest fires in recent history.

Would the velocity of recent observed latitudinal shifts be sufficient for tree species to track climate change northward if that velocity was to remain constant? Species distribution models across eastern North America for A. rubrum, A. saccharum, B. papyrifera, and F. grandifolia suggest that latitudinal range displacement of the distribution centre between current and projected climate scenarios for 2050 of 1.6 km/yr, 4.9 km/yr, 2.6 km/yr and 4.5 km/yr, respectively, would be necessary to keep up with suitable conditions (Périé et al. 2014). Using significant 90th percentile latitudinal differences where we observed most northward movement, the observed annual velocity for saplings of these same species at their leading edge was 0.94 km/yr, 0.4 km/yr, 1.67 km/yr and 0.56 km/yr, respectively. While these are by no means precise estimations, they suggest that these species would lag behind their optimal climate conditions, although presumably less so for A. rubrum and B. papyrifera. In a study in eastern USA, A. rubrum was one of the species showing strong northern skew in abundance and a high proportion of cells occupied at the leading range edge, a pattern interpreted as being indicative of potential for range expansion (Murphy et al. 2010). As for rates for F. grandifolia and A. saccharum, they were comparable to annual post-glacial rates estimated for the Holocene (Clark et al. 1998), which are insufficient to keep up with ongoing climate change. Continued long-term forest monitoring is needed to validate whether the velocity observed will be sufficient for some species to track the climate or whether slow velocity or inertia will result in a mismatch between climatic optima and species distribution, as happened in the past. Our study provides a benchmark against which to assess future observations of latitudinal shifts for these species.

When temporal data are not available to assess directly latitudinal trends, information on the position of juveniles relative to mature trees has been used to infer past or recent range dynamics (Woodall et al. 2009, Zhu et al. 2012, Woodall et al. 2013). The trends uncovered, however, are not always consistent among studies for a given species likely because different parts of the range are examined. Our study provides an opportunity to quantitatively compare the findings from the static (relative latitudinal limits of saplings vs. tree) approach with those of the dynamic (temporal observations) approach for the same dataset. Zhu and colleagues (2012) found that 76 of 92 tree species at their 95th percentile of distribution in the eastern United States had sapling range limits at lower latitudes than trees, despite 62% of northern boundaries being positively correlated with temperature increase. These patterns were interpreted as indicative of range contraction, but it is not clear what processes resulted in apparent contraction at these northern edges. Since trees can integrate processes over nearly a century, one plausible scenario is that the northernmost sites for a species were colonized during periods of climatic conditions favorable for recruitment (i.e., pulse recruitment, Brown and Wu 2005, Fajardo and McIntire 2012), and populations of adult trees, being more resilient than juveniles (Bell et al. 2014), were maintained in spite of deficient recruitment in unfavorable years, resulting in leading tree fronts. Recent evidence suggests that contrasting climatic influences between life stages may prevent migration (Bell et al. 2014, Zhu et al. 2014). Also, disturbances may exacerbate disparities between tree and seedling limits (Woodall et al. 2013). In our study, static observations did not predict temporal trends observed for saplings, but did for trees. With recruitment becoming more frequent for all species in at least one part of their range, one of the consequences could be to reduce the latitudinal gap between adults and juveniles, or even for saplings to move northward of trees as time goes on. This pattern is corroborated by our observations between the first and second time periods at the 90th percentile of latitude. For example, at the 90th percentile of latitude the eight species for which sapling limits were south of tree limits in the first decennial all had positive shifts in sapling limits. In other words, contraction may not be the dominant process for some of these species, especially if their growth responds positively to new climatic conditions.

Latitudinal patterns for adult trees were relatively static at the time scale of our study in spite of adult trees of some species being managed. The distribution of juveniles, however, can be quite dynamic and responses to climate change can occur at ecological time scales with consequences for tree migration (Brown and Wu 2005, Beckage et al. 2008). By providing rare empirical evidence of latitudinal shifts for tree species over broad spatial scale, our study contributes to a growing body of literature highlighting recent distributional responses of species across taxa generally consistent with those predicted from bioclimatic models. The effects of disturbances are hard to disentangle from those of climate change, particularly when looking at a large study area. While our study, like several other ones discussing range shifts, does not demonstrate a causal link, the observed spatial patterns resulting from all the factors driving species dynamics are, at least for some species, consistent with current warming trends. We note as well that species showing significant shifts were early as well as late successional species. The nature of the MFFP dataset is fairly unique because of the large extent of forested area sampled starting in the 1970s when climatic warming was accelerating. Our results set the baseline for future studies. New data from the valuable long-term monitoring forest program conducted in Québec will soon be available when sampling for the fourth inventory is completed. Also, other jurisdictions instituted large-scale inventories more recently than Québec (e.g., Ontario, United States), so rates and magnitude of change can be compared between jurisdictions as data becomes available. Only continued monitoring will reveal whether our observations are just transient dynamics or indicative of range shifts in this century, but the velocities observed are also causes for concerns for our forests.

Acknowledgments

We thank the Ministère des Forêts, de la Faune et des Parcs (MFFP, Quebec) for providing access to forest survey data. This research was supported by grants from the MFFP and from the Natural Sciences and Engineering Research Council of Canada to Sylvie de Blois. The study benefited from interactions with a research group working on biodiversity and climate change in Quebec (CC-Bio). We thank Marie-Claude Lambert (MFFP) for providing statistical advice.

Appendix

Table A1. Latitudinal coordinates (in degrees latitude) for range limits. S₁ = first time period (1970–1977) saplings, S₂ = second time period (1992–2002) saplings, T₁ = first time period trees, T₂ = second time period trees.

Table A2. Latitudinal range shifts through time (in degrees latitude, saplings: S₂–S₁, trees: T₂–T₁) and latitudinal differences between life stages (in degrees latitude, first time period S₁–T₁, second period S₂–T₂) at the 50th percentile of latitude. The standard error was estimated using bootstrap resampling with replacement and is indicated in parentheses.

Table A3. Latitudinal range shifts through time (in degrees latitude, saplings: S₂–S₁, trees: T₂–T₁) and latitudinal differences between life stages (in degrees latitude, first time period S₁–T₁, second period S₂–T₂) at the 90th percentile of latitude. The pooled standard error is indicated in parentheses.

Table A4. Number of plots occupied for S₁, S₂, T₁, and T₂; <50th: below the 50th percentile of latitudinal distribution, >50th: above the 50th percentile of latitudinal distribution, >90th, above the 90th percentile of latitudinal distribution.

Figure A1

Species map for saplings of Abies balsamea. The lower pair of lines represent the 50th percentile while the upper pair represents the 90th.

Figure A2

Species map for trees of Abies balsamea. The lower pair of lines represent the 50th percentile while the upper pair represents the 90th.

Figure A3

Species map for saplings of Acer rubrum. The lower pair of lines represent the 50th percentile while the upper pair represents the 90th.

Figure A4

Species map for trees of Acer rubrum. The lower pair of lines represent the 50th percentile while the upper pair represents the 90th.

Figure A5

Species map for saplings of Acer saccharum. The lower pair of lines represent the 50th percentile while the upper pair represents the 90th.

Figure A6

Species map for trees of Acer saccharum. The lower pair of lines represent the 50th percentile while the upper pair represents the 90th.

Figure A7

Species map for saplings of Betula alleghaniensis. The lower pair of lines represent the 50th percentile while the upper pair represents the 90th.

Figure A8

Species map for trees of Betula alleghaniensis. The lower pair of lines represent the 50th percentile while the upper pair represents the 90th.

Figure A9

Species map for saplings of Betula papyrifera. The lower pair of lines represent the 50th percentile while the upper pair represents the 90th.

Figure A10

Species map for trees of Betula papyrifera. The lower pair of lines represent the 50th percentile while the upper pair represents the 90th.

Figure A11

Species map for saplings of Fagus grandifolia. The lower pair of lines represent the 50th percentile while the upper pair represents the 90th.

Figure A12

Species map for trees of Fagus grandifolia. The lower pair of lines represent the 50th percentile while the upper pair represents the 90th.

Figure A13

Species map for saplings of Ostrya virginiana. The lower pair of lines represent the 50th percentile while the upper pair represents the 90th.

Figure A14

Species map for trees of Ostrya virginiana. The lower pair of lines represent the 50th percentile while the upper pair represents the 90th.

Figure A15

Species map for saplings of Picea glauca. The lower pair of lines represent the 50th percentile while the upper pair represents the 90th.

Figure A16

Species map for trees of Picea glauca. The lower pair of lines represent the 50th percentile while the upper pair represents the 90th.

Figure A17

Species map for saplings of Picea mariana. The lower pair of lines represent the 50th percentile while the upper pair represents the 90th.

Figure A18

Species map for trees of Picea mariana. The lower pair of lines represent the 50th percentile while the upper pair represents the 90th.

Figure A19

Species map for saplings of Populus tremuloides. The lower pair of lines represent the 50th percentile while the upper pair represents the 90th.

Figure A20

Species map for trees of Populus tremuloides. The lower pair of lines represent the 50th percentile while the upper pair represents the 90th.

Figure A21

Species map for saplings of Thuja occidentalis. The lower pair of lines represent the 50th percentile while the upper pair represents the 90th.

Figure A22

Species map for trees of Thuja occidentalis. The lower pair of lines represent the 50th percentile while the upper pair represents the 90th.