Recent observations of divergence between tree growth at high-latitude sites and temperature, as well as the decline of yellow cedar [Callitropsis nootkatensis (D. Don) Oerst. ex D.P. Little] in southeast Alaska due to warming, emphasize a need to investigate nonstationary climate response of Alaskan coastal forests to warming in other tree species. Comparison of annual tree growth in mountain hemlock [Tsuga mertensiana (Bong.) Carrière] to mean monthly temperature and precipitation data from Sitka, Alaska, from A.D. 1830s to 1990s along an elevational transect reveals nonstationarity in tree growth response to climate that suggests an ongoing biome shift. We observe a marked weakening in the positive relationship between annual growth and warmer growth season temperatures at low-elevation hemlock sites, and a concurrent increase in growth and sensitivity at higher elevations coupled with increased correlation between growth and April precipitation at all sites along our transects. As previously observed with yellow cedar, the mechanism of the hemlock biome shift may be due to an increased susceptibility of roots to damage from late frosts resulting from earlier seasonal loss of protective snowpack.
Tree rings are an important paleoclimatic proxy record that has been used to increase our understanding of climatic variability and to reconstruct past climate (Jansen et al., 2007). Dendroclimatology depends on the stationarity of the climate response of trees over time in order to confidently reconstruct past climate. In recent decades divergence, the dissociation between trends in tree growth parameters (e.g. ring width) and climate (Esper and Frank, 2009), has been observed in northern sites that were previously tracking temperature (D'Arrigo et al., 2008). This loss of sensitivity to temperature results in the underestimation of recent warming in reconstructions and presents the possibility of underestimation of similarly warm periods prior to meteorological observation (D'Arrigo et al., 2008; Esper and Frank, 2009).
In addition to divergence, increased temperature and drought conditions have contributed to increased tree mortality in the past decade around the world (Adams et al., 2010). Understanding the response of forests to climatic and environmental changes, especially the possibility of widespread tree mortality, is important for predicting changes in ecosystem functioning, global carbon cycling, regional hydrology, and regional climatology (Adams et al., 2010). One example of contemporary widespread tree mortality in coastal northeast Pacific is the recently observed decline in yellow cedar [Callitropsis nootkatensis (D. Don) Oerst. ex D.P. Little], which has been attributed to warming (Hennon et al., 2005; Beier et al., 2008). Using long-term temperature and precipitation data from the Sitka Magnetic Observatory in Sitka, Alaska, Wiles et al. (2012) demonstrated that this decline can be detected as a decreased correlation between ring widths and temperature, beginning with the period of warming marking the end of the Little Ice Age (LIA).
Mountain hemlock [Tsuga mertensiana (Bong.) Carrière] is an important late-successional species that provides wildlife habitat and watershed protection in subalpine environments and is under increasing harvest pressure (Ally et al., 2000; Edwards et al., 1993). It is commonly used in tree ring reconstructions due to its temperature sensitivity (e.g., Wilson et al., 2007; Wiles et al., 1998; Peterson and Peterson, 2001; Laroque and Smith, 2005). The northernmost extent of the range of this species is south-central, coastal Alaska, where it occurs from sea level to treeline in pure stands or associated with Sitka spruce [Picea sitchensis (Bong.) Carrière] (Gedalof and Smith, 2001).
Although inhabiting different elevations in southeast Alaska, mountain hemlock and yellow cedar share certain physiological characteristics, particularly low root hardiness, that have been implicated in the yellow-cedar decline (Coleman et al., 1992; Schaberg et al., 2008). By comparing growth to summer temperature and winter snowpack, Smith and Laroque (1998) determined that high growth in mountain hemlock is associated with either cool summers and shallow winter snowpack or warm summers and deeper (<4 m) snowpack. This is consistent with what would be expected if trees relied on snowpack for protection against late frosts, as has been suggested for yellow cedar (Beier et al., 2008). A combination of low temperatures and shallow snowpack results in snowpack remaining until the growing season begins, as typically does the combination of high temperatures with a deep snowpack. High temperatures, however, can cause a shallow snowpack to melt before frosts end, resulting in root damage (Schaberg et al., 2008), while low temperatures and high snowpack delays the growing season (Graumlich and Brubaker, 1986). Here, long-term climate data from Sitka, Alaska, is used to investigate climate-response patterns of mountain hemlock in southeast Alaska at multiple elevations during the duration of the Sitka climate data, from the latter part of the LIA to the 1990s, to determine if the beginning of a yellow-cedar—like shift in climate response can be detected.
Tree-ring sites are located in Glacier Bay National Park and Preserve, Alaska, and in the Tongass National Forest of the Coast Mountain Range near Juneau (Fig. 1). Climate patterns in the area are dominated by the Pacific Decadal Oscillation (PDO), the interdecadally fluctuating first principle component of North Pacific sea surface temperature described by Mantua et al. (1997) and Shulski and Wendler (2007). The mountainous terrain, combined with high precipitation from the Gulf of Alaska, has contributed to extensive glaciation (Turtle, 1990). The most recent glacial advance in the region occurred during the broadly defined cold period known as the Little Ice Age (LIA) and ended in the late 1800s (Motyka et al., 2007; Mann et al., 1998). Since the LIA, glacial retreat has been prevalent across the Gulf of Alaska region (Conner et al., 2009; Larson et al., 2005; Barclay et al., 2009). In the glaciated areas during the LIA, patches of forest were left on mountains above the glaciers, called glacial refugia (Cooper, 1923). Other areas in the Tongass National Forest were located below the LIA ice limit and are several-hundred-year-old continuous forests.
DATA AND ANALYSIS
We developed six mountain hemlock ring-width chronologies from sites 200 to 800 m above sea level (Fig. 1). Cores were prepared using standard dendrochronological techniques (Stokes and Smiley, 1968; Cook and Kairiukstis, 1990). Each core was sanded with progressively finer sandpaper, and rings were measured to the nearest 0.001 mm using a Velmex measuring system and the program MeasureJ2X. Dating and quality control was confirmed using COFECHA (Holmes, 1983). Visual and graphical comparisons of cores were used to locate locally absent rings in the series. Descriptive statistics for each chronology (Table 1) are reported from the COFECHA chronologies.
To meaningfully compare and average individual series, exactly dated ring-width series were standardized in ARSTAN (Cook, 1985) using the interactive detrending option. A negative exponential curve, which follows the expected growth function of a radially expanding tree, was used to detrend individual series where appropriate. Where an age-related decrease in ring widths with time was not detected, a horizontal mean line was used to avoid adding or removing long-term growth trends. Standardization results in a series of ring width indices that are then averaged to develop a mean site chronology. A Mann-Whitney test was performed to determine significant changes in ring width between the four time periods. As part of this statistical analysis, Levene's Test for Equality of Variances was also performed to detect changes in variability of yearly growth (Levene, 1960).
Monthly temperature and precipitation records from Sitka, Alaska, were obtained from the Global Historical Climatology Network (GHCN) (Peterson and Vose, 1997; Jones and Bradley, 1992). These records cover the years 1832–1993 for temperature and 1842–1990 for precipitation, with incomplete data from 1876 to 1900. Monthly series were divided into four time periods for analysis: LIA: 1832/1842–1876; post-LIA: 1900–1993/1990; first half of the 20th century: 1900–1950; and second half of the 20th century: 1951–1993/1990.
Standardized tree ring chronologies were correlated to temperature and precipitation data from Sitka for the different time periods. For temperature comparisons, the 20-month dendroclimatic year was used (Fritts, 1976), comparing annual ring growth to each month's temperature from the previous March to the current October. The hydrologic year of the previous January to the current August was used for precipitation analysis.
Average mountain hemlock ring-width indices (Fig. 2) increased significantly (p < 0.05) from the LIA (pre-1876) to the post-LIA (20th century) at the Beartrack (BT) and Juneau Mountain (JM) sites. A significant decrease in growth from the first to the second half of the 20th century was observed at BT, Son of Repeater (SR), Lemon Creek (LC), and McGinnis Trail (MT), while an increase in ring width during this time was observed at JM. The mid-elevation site Repeater Station (RS) experienced no significant change in growth over either time step. Average ring indices increased in variability from the LIA to the post-LIA at high-elevation sites SR and JM (Levene's Test for Equality of Variances: p ≤ 0.05) (Table 2).
The significant positive correlation to temperature at low-elevation mountain hemlock sites during the LIA is present at high elevation sites during the second half of the 20th century (Fig. 3). This trend is especially visible for the Glacier Bay sites, where the high-elevation site SR shifts from no significant correlation with February-August temperature during the LIA to a strong positive correlation during the post-LIA interval. In contrast, the lower elevation site BT shows the opposite, while the mid-elevation site RS is significantly correlated to temperature throughout all time periods. We also examined the responses of tree growth to monthly and seasonal precipitation. A significant trend in growth response to seasonal precipitation through the record was not observed. However, an increasing positive response to April precipitation over time is evident at all sites (Fig. 4).
The low-elevation mountain hemlock sites (LC and BT; Fig. 3) that were positively correlated with temperature during the LIA show a steadily weakening response to temperature over the past 100 years. While trees at low elevations are becoming less sensitive to temperature, trees at mid-elevation sites RS and MT remained consistently positively correlated with temperature for each time period. Those at the highest elevation site, SR, have transitioned from no significant correlation during the LIA to strong positive correlations toward the end of the 20th century (Fig. 3). This trend is more apparent in Glacier Bay than Juneau, a difference that may be attributed to local climatic conditions and is perhaps associated with the ongoing 120-km retreat of ice from the bay (Larson et al., 2005). The increase in annual growth variation at SR and JM (both treeline sites), coupled with a strengthening relationship with warmer temperatures and an increase in growth at JM, suggest that trees at these sites have been released from growthstunting stresses (i.e. low temperature) and are now tracking the rising air temperature (Fritts, 1976; Holtmeier and Broil, 2007).
Wiles et al. (2012) found a shift in the relationship between yellow-cedar growth and January—July temperature since the LIA. A slight negative correlation to temperature was present at the lowelevation site at the end of the LIA, while the higher-elevation site was significantly positively correlated to temperature. By the second half of the 20th century both sites had become strongly negatively correlated to temperature. This strong negative response to warming by the yellow cedar together with a positive relationship with total March and April precipitation suggested that the yellowcedar sites may be susceptible to decline and is consistent with the hypothesis that the yellow-cedar decline is linked to decreased snowpack (Wiles et al, 2012).
While no mountain hemlock sites have become negatively correlated to temperature as the yellow cedars have, the patterns in growth response to climate observed over the 20th century at the low-elevation mountain hemlock sites resemble those observed by Wiles et al. (2012) from the LIA to the first half of the 20th century in the higher elevation cedar site in Glacier Bay. This suggests that the change in growth response of both species is related and elevationally determined, the effects traveling upslope with time and warming. The yellow-cedar decline has been linked to decreased snowpack since the LIA, which exposes roots to damaging spring frosts (Beier et al., 2008; Schaberg et al., 2008). As mountain hemlocks also have low root hardiness (Coleman et al., 1992), it is likely that they are also being affected by a decrease in snowpack with warming. This mechanism is supported by the patterns in growth response to temperature and also by the change in growth response to precipitation (Fig. 4). April precipitation has become more important for annual growth at all six sites since the end of the LIA, suggesting an increased reliance on late spring snowfall.
It is expected that treeline would be advancing with warming. Although a release in growth was found at the high elevation sites, no regeneration data were collected. If mountain hemlocks are in the beginning of a cedar-like decline at the lower extent of their range, as suggested by our findings, and not advancing at the upper extent, then the range of this species will diminish and the species as a whole may begin to decline. If, however, regeneration is occurring and treeline is advancing at a rate equal to or greater than the rate of decline at low elevations, then a biome shift will occur in response to climatic warming. If this is the case, the habitat range of hemlocks will move up in elevation, potentially resulting in habitat fragmentation as the lower extent of the range moves out of valleys and up mountain sides and in habitat loss as the upper habitat range extends beyond the elevation of the mountains.
This study provides data that are consistent with the modeled findings of Laroque and Smith (2003). Using forecasted temperatures and a biologically based deterministic tree growth model, Laroque and Smith (2003) modeled future growth of multiple tree species in British Columbia (mountain hemlock, yellow cedar, western red cedar (Thuja plicata), Douglas fir (Pseudotsuga menziesii), and western hemlock (Tsuga heterophylla). Their predictive model indicates that, of these species, mountain hemlock will be most negatively affected by predicted warming, beginning a sharp decline in growth around 2030. They attributed this decline mainly to the projected decrease in future spring precipitation, to which growth in mountain hemlocks is positively correlated at higher temperatures (Smith and Laroque, 1998).
Our results also bear importantly on the application of tree rings as a proxy record of paleoclimate and its application to understanding the projected future changes in climate (Jansen et al., 2007). The basic assumption of dendroclimatology is that trees are stationary in their climatic response and are responding to climatic variables in the same way today as they have in the past. While many of the studies describing diverging chronologies discussed in D'Arrigo et al. (2008) point to drought stress as a probable cause, the timing of divergence in those studies (mid-20th century) is consistent with the decreased correlation between temperature and growth response described in this study. Since few climatic records begin before the 20th century, our understanding of the relationship between climate and tree growth is limited to the relatively warm 20th century, making it difficult to accurately predict past or future climate without more understanding of the growth dynamics of the species being used. For climatic reconstruction, growth response should be relatively stable to decrease the potential for inaccuracy of reconstruction due to changes in the climate-growth response. While high elevation sites are desired for their climatic sensitivity, they are more likely to have experienced a change in dominant stress over the extent of the chronology than sites at a lower elevation. In this study, the mid-elevation sites RS and MT appear to be the most consistently correlated to growth season temperature and would thus be the most appropriate for growth-season temperature reconstruction (Fig. 3).
Mean ring-width indices, rounded to the nearest 0.001 mm, for each time period, and Mann-Whitney test results (p value) for Little Ice Age (LIA)/Post-LIA and 1st/2nd half of 20th century annual growth comparisons. Shaded rows indicate a significant (p < 0.05) change in ring width.
Mountain hemlocks in southeast Alaska have been shown to be non-stationary in their growth response to climate over the past 200 years. Stands at elevational treeline are becoming more variable in annual growth and positively correlated to temperature, suggesting a growth release at these previously stunted sites, whereas at low-elevation sites the positive correlation of the past between temperature and growth is declining. This negative trend in correlation between temperature and growth at low-elevation mountain hemlock sites during the 20th century is similar to that observed in the declining Alaskan yellow cedar from the LIA through the 20th century. Correlation of mountain hemlock growth to April precipitation at all sites is becoming more positive, suggesting an increased importance of spring snowpack to tree growth. These observations are consistent with the hypothesis that spring snowpack is a critical insulating layer for root protection needed to maintain healthy tree growth in mountain hemlock.
We thank Glacier Bay National Park and Preserve and the National Forest Service (U.S. Forest Service Special Use Permit FS-2700-4) for allowing us to sample the mountain hemlock and yellow-cedar trees for this study and the National Park Service for logistical support. We also thank Marilyn Loveless for the many ecology discussions pertaining to this study and are grateful for the constructive feedback of two anonymous reviewers. This work was supported by the National Science Foundation (ATM-0902799 and AGS-1202218). Part of the Excursion Ridge chronology sampling was made possible by a grant from the National Geographic Society. Lamont-Doherty contribution #7652.