CASSIOPE TETRAGONA

Cassiope tetragona (L.) D. Don (Ericaceae) is a long-lived, evergreen, dwarf-shrub with a circumarctic distribution (Hultén, 1968), and is an important component of low shrub-heath, dwarf shrub-heath, and mixed heath communities (Bliss and Matveyeva, 1992). The species initially grows monopodially with upright stems (∼10 cm). Later, the plant may develop into a loose prostrate mat of stems that radiates outward from a central root mass and can grow to cover an area of ∼1 m² (Crawford, 1989; Havström et al., 1993). The plant may form clones when trailing stems produce adventitious shoots that later disintegrate, resulting in physically separate, but genetically identical individuals (Havström, 1995). Along each stem, two alternating sets of opposite leaves are produced, forming four distinct rows (Johnstone and Henry, 1997). Johnstone and Henry (1997) found the pattern in the positioning of leaf node scars in adjacent leaf rows was analogous to the wave-like pattern in leaf-lengths first described by Warming (1908), and later used by Callaghan et al. (1989) and Havström et al. (1993, 1995) to identify and date individual annual growth increments and to develop short chronologies (26 yr). As leaf scars remain visible along the length of a shoot for long time periods, Johnstone and Henry (1997) used the pattern to identify, measure, and date annual growth increments and to generate longer chronologies (26–36 yr).

STUDY AREA

Alexandra Fiord (78°53′N, 75°55′W) lies on the east coast of central Ellesmere Island (Figs. 1A, 1B). The lowland is a post-glacial outwash plain bound by upland plateaus (500–700 m a.s.l.), Twin Glacier, and the coastline of Alexandra Fiord. Closed to semi-closed vegetation occurs over 90% of the Alexandra Fiord lowland (Muc et al., 1989). Heath communities dominated by C. tetragona and Dryas integrifolia cover >50% of the area, occupying habitats such as fellfields, well-drained heath tundra, and snowbeds (Muc et al., 1989). The mean growing season (June–August) air temperature in the lowland is 3.2°C (mean annual temperature, −15.2°C), with summer precipitation of 10–60 mm (annual precipitation, 100–200 mm) (Labine, 1994; C. Labine, unpublished data). Further information on the climate of Alexandra Fiord is available in Labine (1994).

INSTRUMENTAL CLIMATE DATA

We used instrumental climate data from the Eureka HAWS located on the west coast of Ellesmere Island, 250 km northwest of Alexandra Fiord (Fig. 1B) (Meteorological Society of Canada, 1948–1998). Climate data have been recorded continuously at the station since May 1947. While the Eureka HAWS and Alexandra Fiord are not grouped into the same climate region (Maxwell, 1981), climate variables (e.g., air temperature, incoming shortwave, net radiation) recorded during the growing season correlated more strongly between Alexandra Fiord and the Eureka HAWS than with the HAWS at Alert, Ellesmere Island (Labine, 1994). The Eureka HAWS has the warmest summer climate conditions of any of the HAWS, making it a good counterpart to the thermal oasis of Alexandra Fiord (Maxwell, 1981; Edlund and Alt, 1989; Labine, 1994).

CHRONOLOGY DEVELOPMENT

In July 1998, we collected C. tetragona plants at two sites at Alexandra Fiord (hereafter referred to as Lowland [30 m. a.s.l.] and Bench [150 m. a.s.l.]) (Fig. 1C). Twelve to fifteen plants were collected from a visually homogenous area at each site and chosen for the longest stems possible. The plants were air-dried for a week and then stored.

In the laboratory, we selected five to nine stems per plant for measurement. We chose stems for their length and, when possible, for live green leaf buds and leaves at the tip of the shoot. Stems were straightened by placing them in a lukewarm bath for 15 min, and then they were inserted into 1-cm-diameter glass tubes to dry. We removed two adjacent rows of leaves along each stem by hand, being careful to leave in place any flower buds or flower peduncles.

Under a dissecting microscope (10× to 30× magnification), we measured the internode distances between leaf scars using a manually operated caliper system (designed by J. Svoboda, University of Toronto). The internode lengths were measured from the base to the tip of the shoot. We identified the annual growth increments (AGIs) by the wave-like patterns in internode lengths, with the terminus of annual growth delimited by the shortest internode length at the end of each wave-series (Johnstone and Henry, 1997). Annual production of leaves, flower buds, and flower peduncles were calculated by counting the total number of each variable found within the AGI.

Measurement series were visually cross-dated, a pattern matching technique which is used to assign calendar year dates to each AGI (Stokes and Smiley, 1968). We verified the visual cross dating by statistically comparing measured segments with the full chronology (COFECHA; Dendrochronology Program Library) (Holmes et al., 1986). Two to nine stems per plant were cross-dated, and 12 to 15 plants were used to create site chronologies. We also constructed chronologies for the annual production of leaves, flower buds, and flower peduncles, based on the number of variables per AGI.

We standardized all C. tetragona measured series using dendrochronological methods (Fritts, 1976; Cook and Briffa, 1990). Inspection of the individual series revealed a juvenile growth trend in which shorter AGIs characterized the first years of shoot elongation. In addition, because C. tetragona plants at Alexandra Fiord do not reproduce until ∼10–15 years of age, we hypothesized that reproduction may also be influenced by a juvenile growth trend. To minimize the juvenile effect on growth and reproduction, all stem series were standardized and all flower bud and flower peduncle chronologies were truncated to begin at the first year of bud formation or flower production. The sample depth for all Lowland and Bench chronologies ranged from a minimum of 3 to 17 stems for the period of 1895–1915, to a maximum sample depth of 64 (Bench) and 76 (Lowland) stems later in the chronologies.

Based on the signal-to-noise ratio (SNR) and a priori information regarding the plant's growth and reproductive characteristics, chronologies for each site were standardized using flexible cubic splines (ARSTAN; Dendrochronology Program Library) (Wigley et al., 1984; Cook, 1985; Cook and Briffa, 1990). In our study, a smoothing spline was needed to maximize the inter-stem correlation without losing the climate signal entirely. Relatively small changes within the plant's architecture (e.g., inter-shoot shading, resource partitioning, apical dominance) (Johnstone and Henry 1997) could have as great an effect on plant growth as climate over the lifetime of the plant. Thus, intra-plant effects had to be filtered as much as possible. We selected 20-year cubic splines for the growth chronologies, and 60- and 70-year cubic splines for the flower peduncle and flower bud chronologies, respectively. Following standardization, the annual indices from the individual stems were averaged using a biweight robust mean to produce master chronologies for each site (Cook and Briffa, 1990). In order to maintain what remained of the low frequency signal in the site chronologies, we retained the standardized chronologies for analysis.

CORRELATIONS AND RESPONSE FUNCTIONS

We investigated climate-growth and climate-reproduction relationships through correlation and response function analysis (PRECONK, v. 5.11; Fritts, 1999). In PRECONK, a response function analysis is run by regressing principal components of temperature and previous growth (reproduction) on the annual growth (reproduction) indices. We calculated correlation and response function coefficients between the Alexandra Fiord growth and reproduction chronologies and average monthly air temperature from the Eureka HAWS for the common period of 1948–1996. Correlation and response function coefficients were calculated for a 17-month period (previous May to current September) and 2 yr of previous growth (reproduction) were included in the models to account for lag effects.

TRANSFER FUNCTIONS

Following investigation of the correlation and response function coefficients, transfer function models predicting average summer air temperature from C. tetragona growth and reproduction indices were calibrated using stepwise multiple linear regression analysis (MGLH procedure; SYSTAT, v. 5.0, 1990–1992). We calibrated preliminary models predicting temperature for individual months of the growing season (May through September), and for different combinations of months (e.g., June–September) (Case and MacDonald, 1995). We included growth and reproduction indices from the two sites as potential predictors in the transfer function analyses. Both forward (t + 1, t + 2) and backward (t − 1) lagged indices were added to capture persistence effects in the growth-reproduction-climate relationships (Jacoby and D'Arrigo, 1989). Chronologies for the annual production of flower peduncles were not included in the transfer function analysis because of their shorter length.

To demonstrate the temporal stability of the empirically derived equation linking C. tetragona growth and reproduction to temperature, we used the traditional data-splitting method of Fritts (1976). We split the full data set into an “early” period (1948–1980) and a “late” period (1964–1996) and then calibrated transfer functions for each period using stepwise regression. The calibration models developed for the “early” and “late” periods were then verified over the remaining one-third of the data set, 1981–1994 and 1948–1963, respectively. Verification statistics included Pearson's product-moment correlation coefficient (r), reduction of error statistic (RE), the coefficient of efficiency (CE), and the nonparametric sign test of first differences (Fritts, 1976; Briffa et al., 1988).

CHRONOLOGY CHARACTERISTICS

The Lowland growth chronologies are the longest C. tetragona chronologies (1879–1996) yet developed for the Arctic (Table 1a). The growth chronologies from both sites were characterized by low SNR, mean sensitivity, standard deviation and first-order autocorrelation values (Table 1b). In contrast, the reproductive chronologies showed high mean sensitivity and standard deviation values, but low SNR and first-order autocorrelation (Table 1b).

GROWTH/REPRODUCTION–CLIMATE RELATIONSHIP

While many of the correlation and response function coefficients were not significantly different from zero (P > 0.05), the majority of the coefficients were positively associated with air temperature during the growing season months (May–September) (Figs. 2A, 2B). The Lowland growth variables were positively correlated (P < 0.05) with June average air temperature. At the Bench site, growth variables were significantly correlated with June (AGI) and September (AGI, leaves) average air temperature. There was also a significant response function relationship between annual production of leaves and August temperature. Growing season temperature in June, July, and September was positively correlated with flower peduncle production at both sites. At the Lowland site, flower bud production was positively related to July and September temperature, while at the Bench site it was positively associated with May, June, and July temperature.

There were significant correlation and response function relationships between the Lowland and Bench variables and average air temperature during the winter months (October–April) as well (Figs. 2A, 2B). At the Bench site, the response function coefficients for growth were positively associated with previous October (AGI) and January (leaves) temperatures, but negatively associated with February (AGI, leaves) temperatures. In addition, there were negative response function relationships between annual production of flower buds and previous November (Lowland) and December (Bench) temperatures.

The transfer function model predicting July–September average air temperature provided the best results for the period of 1948–1994 (adjusted R² = 0.45) (Tables 2, 3). Early (1948–1980) and late (1964–1996) calibration models accounted for 39% and 46% of the variance (adjusted R²) in the temperature record, respectively (Table 2). For the late (1981–1994) and early (1948–1963) verification periods, Pearson' product-moment correlation coefficients (r = 0.59; r = 0.50) were significant, indicating that the model is able to estimate values not contained in the calibration period. The positive CE (0.10; 0.09) and RE (0.12; 0.22) values indicate that the model has some predictive skill. Although the level of significance cannot be tested, positive RE and CE values are generally accepted as an indication of successful reconstruction of the climatic variable (Briffa et al., 1988). The sign tests for the verification periods were not significantly different from zero. Given that the early and late models passed three out of the four verification tests, the transfer function estimates of July–September average air temperature were considered time stable (Fritts, 1976; Fritts et al., 1979). Therefore, the full model calibrated over the 1948–1994 period was also considered time stable and was used to reconstruct summer air temperature. The full model met the assumptions of parametric statistics (Neter et al., 1996).

AVERAGE AIR TEMPERATURE RECONSTRUCTION

The 100-year-long reconstruction (1895–1994) of July–September average air temperature is the longest C. tetragona–based proxy constructed for the Arctic (Fig. 3). The record was necessarily shortened by the shortest predictor chronology (Bench; number of flower buds, 1894–1996), and by lags and leads included in the full model. Overall, the reconstruction followed the observed data from the Eureka HAWS (1948–1994). However, the model underestimated the instrumentally recorded data in particularly warm or cold years (e.g., 1954, 1961, 1979, 1983), and occasionally smoothed over short periods of moderate inter-annual variability (e.g., 1971–1975). The first 15–20 yr of the air temperature model are characterized by high inter-annual variability (Fig. 3). Despite the early variability, the reconstruction revealed a sharp decrease in temperature at the turn of the century followed by an increase from ∼1905–1910 to the early 1960s (Figs. 3, 4A). A cooling trend characterized the 1960s–1970s of the reconstruction until a shift occurred in the early 1980s when temperatures rose once again (Figs. 3, 4A).

CHRONOLOGY CHARACTERISTICS

The mean and standard deviation values of the growth variables from both sites were similar to those reported in previous studies of Cassiope tetragona at High Arctic sites (Nams and Freedman, 1987; Callaghan et al., 1989; Havström et al., 1993; 1995; Johnstone and Henry, 1997). In contrast, the mean number of flower peduncles produced per year at the Lowland and Bench sites was less (Havström et al., 1995; Johnstone and Henry, 1997). We hypothesize that potentially variable environmental conditions at Alexandra Fiord in the early 20th century may have suppressed flower bud development and flower production in mature plants and delayed it in juvenile ones. It is also possible that fragile flower buds and peduncles fell off the oldest stems. However, manually removed flower buds and peduncles normally leave visible and identifiable scars along the stem. The mean annual flower bud production values reported in our study are the first observations recorded for a period of greater than 1 yr (Nams, 1982).

The low SNR values for the growth and reproductive chronologies at both sites indicate a high amount of noise present in the chronologies. However, no other standardization method yielded higher SNR values. We hypothesize that the high level of intra- (not shown) and inter-plant variability is the result of the plant's architecture and associated microenvironmental conditions, within-plant resource partitioning (e.g., water, nutrients, photosynthates), and poorly understood community dynamics within arctic ecosystems (Rayback and Henry, 2005).

In general, high mean sensitivity and low first-order autocorrelation values for tree-ring series are indicative of high inter-annual variability and sensitivity to yearly climate conditions (Fritts and Shatz, 1975). Previous studies have shown that C. tetragona increases flower bud and peduncle production under short-term, favorable ambient growing season temperatures and under experimentally enhanced temperatures (Nams and Freedman, 1987; Johnstone, 1995; Johnstone and Henry, 1997). In contrast, C. tetragona exhibits a rather conservative growth response to short-term variations in growing season climate. The lower sensitivity may serve to stabilize the annual vegetative productivity of this species in the resource-poor arctic environment (Sørensen, 1941; Shaver and Kummerow, 1992). The mean sensitivity values were similar to those reported by Johnstone and Henry (1997). Low autocorrelation values indicate that C. tetragona growth is relatively uninfluenced by low-frequency climatic trends. However, standardization of the growth chronologies with flexible cubic splines may have removed more of the low-frequency climatic fluctuations than intended (Cook and Briffa, 1990).

GROWTH/REPRODUCTION-CLIMATE RELATIONSHIP

The positive influence of early growing season temperatures on the growth and the initiation of photosynthesis in C. tetragona and other arctic plants is well documented (Callaghan et al., 1989; Havström et al., 1995; Johnstone and Henry, 1997). In addition, warmer, early growing season temperatures initiate the elongation of flower peduncles and the subsequent formation of the flowers early in the summer (Nams, 1982; Johnstone, 1995). Flower buds are generally visible and swollen in the leaf axils by late June to mid-July (Nams, 1982; Johnstone, 1995). Temperature patterns at the end of the growing season may facilitate the translocation of nutrients in the plants for winter storage, as well as cue critical winter hardening that protects the plant tissues from injury due to intracellular freezing and desiccation (Chapin et al., 1980; Fritter and Hay, 2002). Warmer late summer temperatures also facilitate successful reproduction by allowing the further development of flower buds prior to over-wintering (Nams and Freedman, 1987; Johnstone, 1995). Cassiope tetragona pre-forms flower buds 1 yr prior to actual flowering (Sørensen, 1941). Lastly, data on the influence of winter temperature on arctic plants is scarce (Körner and Larcher, 1988). In this study, there were few significant correlation and response function coefficients for the winter months, and those coefficients that were significant may be the product of the arbitrary assignment of winter temperature into monthly increments that are not necessarily physiologically or ecologically based.

SUMMER TEMPERATURE RECONSTRUCTION

Interpretation of the first 15–20 yr of the summer temperature reconstruction is problematic. The high variability revealed in the early portion of the reconstruction may be a statistical artifact, the result of uncommon variance or noise increasing with decreasing sample size early in the chronology (Fritts, 1976; Cook and Briffa, 1990). However, our reconstruction revealed a decrease in temperature at the turn of the century followed by an increase around 1905–1910. The temperature decline during this same period is also recognizable in the Arctic-wide, average summer air temperature series (Fig. 4B) (Overpeck et al., 1997), and as a decrease in lake varve thickness at Lake C2 and Lake C3, Ellesmere Island (Fig. 4C) (Lamoureux and Bradley, 1996; Lasca, 1997). Unfortunately, the short length of our model (100 yr) makes it impossible to determine how the decrease in temperature fits into the overall trend of increasing warmth in the early 20th century that is evident in other arctic proxies (e.g., Overpeck et al., 1997). Furthermore, we acknowledge that the use of flexible smoothing splines does reduce our ability to reconstruct longer-term climate trends.

Proxy evidence from Arctic Canada indicates that within the context of the late Holocene, summer temperatures during the last 100 yr were exceptionally high (Koerner and Fisher, 1990), and the period of 1931–1960 was warmer than 80% of the last 3000 yr (Lamoureux and Bradley, 1996). Beginning in ∼1905–1910, our reconstruction showed an increase in summer temperature which continued into the early 1960s, reaching its peak between ∼1945 and 1965. The Arctic-wide, average summer air temperature time-series (Fig. 4B) (Overpeck et al., 1997), and lake varve records from Lake C2, Lake C3, and Lake Tuborg, Ellesmere Island, also support warmer temperatures in the Canadian Arctic from ∼1920 to 1965 (Figs. 4C, 4D) (Lamoureux and Bradley, 1996; Lasca, 1997; Smith et al., 2004) . In addition, ice core melt records from Ellesmere, Devon, and Axel Heiberg Islands provide evidence of a large increase in the amount of melting during the last 100 yr, particularly during the mid-1920s to 1930s and the 1950s to 1960s (Fig. 4E) (Koerner, 1977; Fisher and Koerner, 1983; Koerner and Fisher, 1990). Since 1920, anomalously high solar radiation and lower volcanic aerosol loading, along with exponentially increasing atmospheric trace gases may have influenced the warming trend across the Arctic (Overpeck et al., 1997, and references therein; Mann et al., 1998; Free and Robock, 1999; Lean and Rind, 1999).

Beginning in the mid-1960s and lasting throughout the 1970s, our reconstruction showed a decrease in summer temperature. A similar cooling trend characterizes the ice core melt records from the Devon Island Ice Cap (Fig. 4E) (Fisher and Koerner, 1994), and the sedimentary records from Lake C3 and Lake Tuborg, Ellesmere Island (Figs. 4C, 4D) (Lasca, 1997; Smith et al., 2004). The cooling trend is also pronounced in the Arctic-wide, average summer air temperature reconstruction (Fig. 4B) (Overpeck et al., 1997). Our model is further supported by instrumental climate records, which are marked by a decline in temperature after 1962–1963 (Meteorological Service of Canada, 1948–1998). Multiple factors may have influenced the mid-century cooling in the High Arctic, including a change in high-latitude atmospheric circulation after 1962 (Bradley and England, 1978; Kahl et al., 1993), an increase in arctic tropospheric aerosols caused by increased volcanism, and a leveling off of solar irradiance by the mid-20th century (Overpeck et al., 1997, and references therein).

Finally, our reconstruction revealed an increase in summer temperature at Alexandra Fiord since the early 1980s. Tree-ring records from the Low Arctic of North America show that trees began to recover from the mid-century cooling by the early 1980s (D'Arrigo and Jacoby, 1993). The upward trend in temperature is also evident in the Arctic-wide, average air temperature time-series (Fig. 4B) (Overpeck et al., 1997). The close correspondence between our model and the Arctic-wide time-series indicates that the Cassiope-based reconstruction contains a large-scale temperature signal. The temperature increase was also detected in the HAWS instrumental record; the 1990s is the warmest decade on record (Meteorological Service of Canada, 1948–1998). The observed warming of the last decade is likely caused by a combination of anthropogenic and natural forcing factors, changes in the predominant mode of atmospheric circulation, as well as climate feedback mechanisms (Mann et al., 1998; Thompson and Wallace, 1998; Serreze et al., 2000; Moritz et al., 2002).

Further work developing C. tetragona chronologies from sites located across broad latitudinal and longitudinal gradients in the Arctic will facilitate local-scale assessment of recent climate change, as well as highlight patterns of regional differences. In addition, longer C. tetragona chronologies that extend back to the end of the Little Ice Age will allow us to place the reconstructions within the longer-term perspective of other proxy data. By increasing the overall number of proxy data sets for Arctic Canada, we will improve our understanding of the spatial and temporal variability of northern climate and the arctic climate system.