CHRONOLOGY DEVELOPMENT

To develop the dendrohydrological reconstruction of precipitation, increment cores were collected during the summer of 2006. At each site, 25–40 trees were cored using a Hagloff increment borer (internal diameter ∼4.3 mm). A minimum of two cores were obtained at breast height from opposite sides of the tree.

Preparation and analysis of the tree cores followed standard dendrochronological methods (Stokes and Smiley, 1968; Fritts, 1976). Cores were visually cross-dated and marker rings were recorded following the list method of Yamaguchi (1991). Cross-dated samples were measured using a Velmex UniSlide traversing table and digital encoder with an accuracy of 0.001 mm. Visual cross-dating was checked using the software program COFECHA (Holmes, 1983). Site chronologies were standardized using conservative detrending methods and the software program ARSTAN (Cook and Holmes, 1986). To measure the strength of the common signal between samples from individual sites, the series intercorrelation was determined by averaging the correlation of each individual series with the site master chronology (Holmes, 1983). The mean sensitivity for each site chronology was also determined as the relative difference in ring width from one year to the next. Mean sensitivity can vary from 0 (no difference) to 2 (repeating pattern of alternating narrow and wide rings) (Fritts, 1976). The agreement between site chronologies was examined using Principal Components Analysis (PCA) during the period of common overlap (1874–2005) for the 11 longest chronologies.

In an effort to maximize the length of the reconstruction and produce a statistically robust regional chronology, master chronologies were developed using two different methodologies. (1) Yellowknife master 1: Site chronologies were constructed for each of the 12 sampling sites using the computer program ARSTAN. Following the methods of Meko (2002), a sample-size-weighted mean chronology (SSWMC) was developed. In this method, the resultant master chronology is a weighted mean of all the available chronologies in that year with the weight being proportional to the number of trees in each particular chronology. Complete details regarding the computation of the SSWMC can be found in Meko (2002). (2) Yellowknife master 2: Using the site chronologies developed with ARSTAN, PCA (Varimax rotation) was utilized to develop a master chronology; however, the shortest chronology (PREE) was excluded to maximize the length of the PCA-derived master chronology. While this resulted in a much shorter master chronology than method 1 (SSWMC), it afforded the opportunity to compare the two regional series to assess the similarity and robustness of the chronology development procedures by calculating a Pearson correlation between the two master chronologies.

CHRONOLOGY CHARACTERISTICS

Summary statistics for the 12 jack pine chronologies are presented in Table 1. The longest chronology spans the period 1679–2005 (Fox Lake), while the shortest extends from 1936 to 2005 (Prelude Lake East). The average time span covered by all of the chronologies is approximately 1825–2005. Series intercorrelation for all sites is high (Table 1), suggesting a common growth signal within respective sites. Correlations between all 12 chronologies over the common period of overlap (1936–2005) are high (mean r  =  0.715) (Table 2), indicating that the common signal also persists across the study region. Similarly, PCA of the site chronologies indicates the weights of individual chronologies are relatively evenly distributed across the sampling sites (Table 2). The first PC explains a significant amount of variance in the data set (50.9%). The analyses both support a common regional signal in tree growth that is likely climatically controlled.

Table 2

Correlations between Jack pine chronologies for the Yellowknife region, Northwest Territories. The common period used in the analysis spans 1936–2005 (n  =  70). All correlations are significant at p ≤ 0.05. *Principal Components Analysis (Varimax rotation) of the 11 longest site chronologies for the common period of overlap (1874–2005); values are the coefficients of tree growth for PC1 for each of the site chronologies. The variance explained by PC1 is 50.9%.

Annual growth at the 12 jack pine sites is low as indicated by the small annual growth increment (Table 1). Mean ring width among the sites varies from 0.380 mm at APIT to 0.750 mm at Prelude Lake East (PREE). PREE has the youngest trees which consist of a greater proportion of juvenile growth than the series from most of the other sites. Regardless, PREE still correlates relatively well with nearly all of the other 11 site chronologies (mean r  =  0.695) (Table 2).

Ring widths exhibit considerable variability from one year to the next as indicated by the mean sensitivity for the sites which varies from 0.335 to 0.435. Based on data housed at the International Tree-ring Databank (ITRDB), mean sensitivities reported here are higher than those for white spruce growing near treeline at comparable latitudes in Yukon Territory and Alaska which are typically around 0.200. Year-to-year variability in ring width is also supported by the first-order autocorrelation for the 12 sites, which ranges from 0.180 to 0.504 for the standard chronologies. These values are also lower than those typical of white spruce populations in northwest North America (typically >0.750) (based on data from the ITRDB). These lower autocorrelation values suggest that year-to-year variability in the Yellowknife jack pine chronologies is greater and the long-term, low frequency variability is lower than reported for northern white spruce populations. While autocorrelation of the jack pine chronologies is low in comparison to other northern tree species, it is still considerably higher than the autocorrelation in the Yellowknife instrumental precipitation record. For example, first-order autocorrelation for June precipitation (1943–2005) is −0.04. Therefore, autocorrelation of the individual site chronologies was adjusted to match that in the instrumental record (sensu Meko et al., 2001).

Considering the strong correlations between the 12 site chronologies, the similar responses to climate across the region (see below and Table 3), and the relatively even distribution of site chronologies on PC1, individual series from each of the 12 sites were combined using the SSWMC procedure to construct a single regional jack pine chronology for Yellowknife spanning from 1680 to 2005. The strong correlation between the PCA-derived and sample-size-weighted mean chronologies (r  =  0.757; p ≤ 0.05) (Fig. 3a) suggests the SSWMC (Fig. 3b) is a robust and valid record of tree growth for the region, and subsequently was used to develop the dendrohydrological reconstruction of June precipitation described below.

Figure 3

(a) Two jack pine master chronologies developed for Yellowknife using the Sample-size-weighted mean chronology (SSWMC) and Principal Components (1874–2005). The correlation between the series during the period of overlap is r  =  0.730 (p ≤ 0.05). (b) Jack pine standard chronology for Yellowknife region. Thick black line is the SSWMC chronology based on the 12 site chronologies, while the thin gray lines are the individual site chronologies highlighting the variability between sites. (c) Running Rbar and (d) expressed population signal (EPS) and sample depth. The Rbar and EPS values were computed using a 30-yr moving window advanced 1 year. The gray bars on the Rbar plot are two standard errors. The shaded gray area in (d) is the number of individual tree-ring series in the regional chronology through time.

Table 3

Pearson correlation coefficients for significant relations between the Yellowknife regional Jack pine chronology (SSWMC) and monthly (a) precipitation totals and (b) mean temperatures recorded at the Yellowknife Airport. Period of correlation analysis with temperature and precipitation is 1943–2005. (c) Palmer Drought Severity Index (PDSI) values for three grid points near Yellowknife. Instrumental PDSI data from Cook et al. (2004); period of correlation analysis between tree-ring chronologies and PDSI data is 1900–1990. Only significant correlations (p < 0.05) are shown.

The strength of the common signal in the regional chronology was assessed using the running series of average correlation (Rbar) (Fig. 3c) and the expressed population signal (EPS) statistics (Wigley et al., 1984) (Fig. 3d). The Rbar statistic was truncated at 1819, prior to which it was highly variable. From 1819 onwards the Rbar statistic is generally consistent at ∼0.3. The EPS statistic is a measure of how well a composite chronology reflects a hypothetically perfect chronology, with values above 0.85 generally acceptable (Briffa and Jones, 1990). The running EPS statistic is above 0.85 from 1819 onwards. Prior to 1819, EPS falls below 0.85 as sample depth declines to 10 series or less. These statistics indicate that 1819 can be considered an acceptable cutoff for the Yellowknife precipitation reconstruction. However, we present the full reconstruction spanning the period 1680–2005 because (1) the correlation is high between the SSWMC and the PCA-derived chronology during the period of overlap, and (2) it shares compelling similarities with other dendrohydrological reconstructions from Canada and the United States as discussed below.

TREE-RING/CLIMATE RELATIONSHIPS

Precipitation and temperature records from Yellowknife Airport (1943–2005) were used to evaluate the climate signal contained in the 12 site chronologies (Table 3). The adjusted precipitation and homogenized temperature records were utilized since both have been corrected for inhomogeneities and missing data (Mekis and Hogg, 1999; Vincent and Gullett, 1999). Adjusted climate data for Yellowknife Airport are available from the Adjusted Historical Canadian Climate Data (AHCCD) web site ( http://www.cccma.ec.gc.ca/hccd/). Pearson correlations were computed between each of the 12 site chronologies and monthly precipitation and temperature records for the 20-month period from the previous May to December of the current growing season. Seasonal (e.g., May–July precipitation) and annualized precipitation (12 month periods; e.g., _previousJuly–June, _previousAugust–July) indices were also developed and correlations were computed for these indices and the 12 site chronologies and the SSWMC (Table 3a). Ring width and June precipitation consistently showed the strongest correlations. Growing or summer season precipitation indices (total May–July, June–July, and June–August precipitation) were also highly correlated with the individual site chronologies and the regional chronology.

Previous studies investigating hydrologic controls on tree growth have found that annualized data incorporating precipitation contributions over more than one growing season are often more strongly correlated and explain a higher percentage of variance in tree-ring records (e.g., Case and MacDonald, 1995; Sauchyn and Beaudoin, 1998; Watson and Luckman, 2001; St. George and Nielson, 2002). This was tested using the Yellowknife jack pine chronologies and calculating correlations between tree growth and precipitation over two growing seasons (e.g., _previousJuly–June). While the regional chronology did correlate with various annualized periods, it did not result in stronger relations than that of June precipitation or the growing season precipitation indices (Table 3). Furthermore, when individual sites were analyzed, the relations were not consistent across all sites, unlike the June and the combined June–July index.

Correlations between ring width and monthly temperatures were also calculated over the same 20-month period (Table 3b). The months of January and August were the only months that had significant relationships with tree growth over the period of overlap (1943–2005). Seven of the 12 site chronologies had positive and significant (p ≤ 0.05) correlations with January temperatures while only 3 of the 12 had significant correlations with August temperatures (Table 3).

Instrumental and reconstructed PDSI data were obtained from North American Drought Atlas website ( http://iridl.ldeo.columbia.edu/SOURCES/.LDEO/.TRL/.NADA2004/.pdsi-atlas.html). The PDSI are averaged values for the summer months (June–July–August) and generally fall between +6 and −6. Several of the individual site chronologies are significantly correlated with the instrumental PDSI records for the Yellowknife region (Table 3). The SSWMC is also correlated with PDSI for the three grid points near Yellowknife, but the correlations are weaker than for June precipitation.

CLIMATE RECONSTRUCTION

Based on the correlation analysis presented in Table 3, a reconstruction of June precipitation was developed using the SSWMC. Given the lack of autocorrelation in June precipitation in the Yellowknife region, lagged predictors were not included as potential predictors in the regression analysis. Since instrumental climate data for Yellowknife is only available since 1943, a leave-one-out cross validation procedure was used to evaluate the skill of the model (Gordon, 1982). The regression model using only the SSWMC as a predictor explains approximately 42% (R²_adj) of the variance in the instrumental record during the full calibration period.

MODEL VERIFICATION

The reconstruction model for the full period (1943–2005) took the form:

1

The calibration and verification statistics are provided in Table 4. The full model passed all of the independent verification statistics that were used (p ≤ 0.05). A positive value for the reduction of error (RE) statistic indicates that the model possesses reconstructive skill and that meaningful climatic information exists within the reconstruction (Fritts, 1976). The Durbin-Watson statistic of 1.94 suggests there is no issue of autocorrelation in the residuals during the calibration period. Since the full model demonstrated significant abilities at predicting June precipitation during the leave-one-out process, we consider the transfer function to be robust and time stable. Subsequently, the model developed above was used to reconstruct June precipitation back to 1680.

Table 4

Calibration and verification statistics for Yellowknife June precipitation reconstruction model^a

Figure 4 shows that there is good correspondence between the reconstructed values and the instrumental June precipitation data throughout the period of overlap (r  =  0.654). Mean June precipitation is 26.9 mm for the instrumental record and 23.3 mm for the reconstructed data during the period 1971–2000. The instrumental and reconstructed June precipitation records also share similar autocorrelation signatures: −0.04 for the instrumental data and −0.15 for the reconstructed values (1943–2005). The reconstruction is particularly effective during the period 1943–1995 where broader trends in the instrumental data are captured quite well. A series of high June precipitation values (1984, 1988, and 1991) are not reconstructed well by the tree rings. Similar difficulties have been noted in other hydrological reconstructions (e.g., Graumlich et al., 2003), probably in response to decreased sensitivity of tree growth to above-average growing conditions. A return to average June precipitation values after 1991 is captured by the tree rings, but they fail to closely track the year-to-year variability with much skill until 2003 (Fig. 4).

Figure 4

A comparison of the tree-ring reconstructed total June precipitation and the instrumental data from Yellowknife Airport during the 1943-2005 calibration period. Correlation between the predicted and actual June precipitation is r  =  0.65 (p ≤ 0.05).

The reconstruction of June precipitation (presented as anomalies from the 1680–2005 mean) is presented in Figure 5a. The 326-year June precipitation reconstruction for the Yellowknife region extends the precipitation record by more than 260 years. Mean June precipitation for the full reconstruction (1680–2005) is 21.7 mm. Over the reconstructed period, the most prolonged period of dry anomalies was the mid-20th century. During the 1927 to 1976 period, June precipitation anomalies were consistently negative, in particular, from 1958 to 1976. The driest June during the entire record was found to be 1927 (Table 5). Other notable periods of extended negative June precipitation anomalies include 1879–1893, 1842–1865, 1776–1796, and 1698–1735. Periods of positive June precipitation anomalies include 1976–2005, 1894–1926, 1822–1841, and 1756–1775. These prolonged periods of above/below average precipitation coincide with similar dry/wet episodes in dendrohydrological reconstructions from western Canada, the Prairies, and parts of the United States at multi-decadal time scales (Figs. 5a–5e). It also displays coherence with a record of Pacific climate variability suggesting some climatic forcing originates in the Pacific basin (Fig. 5f). The instrumental climate record from Yellowknife also shows coherence with Pacific climate variability (Table 6), suggesting that coherence with the proxy record is not simply an artifact of the tree-ring reconstruction. Five-year averages of June precipitation from Yellowknife Airport are highly correlated with similarly averaged values of the Pacific Decadal Oscillation (PDO) Index for numerous months of the year and annually (Table 6). The highest and most consistent relations with the monthly PDO index values are with precipitation during the months of February, May, and June.

Figure 5

Time series plots of dendrohydrological reconstructions and Pacific climate variability for select regions of North America. All series are presented as z-scores (smoothed with a 15-year moving average), except the Yellowknife June precipitation reconstruction which shows anomalies with respect to 1680–2005 mean. The time series are as follows: (a) Yellowknife June precipitation reconstruction (this study). The vertical dashed line separates the portion of the reconstruction that is statistically robust based on EPS and Rbar statistics (1819–2005). Prior to 1819 the reconstruction is less robust. (b) PDSI Grid point 93 (Cook et al., 2004); (c) Lake Athabasca water level for three time periods (Stockton and Fritts, 1973); (d) PDSI Grid point 53 (Cook et al., 2004); (e) North Saskatchewan riverflow (Case and MacDonald, 2003); and (f) Reconstructed Pacific Decadal Oscillation Index (MacDonald and Case, 2005). The gray shaded columns indicate periods of extended negative June precipitation anomalies in the Yellowknife reconstruction. Numbers in brackets indicate the Pearson's correlation coefficient and significance level for the smoothed time series and the Yellowknife precipitation reconstruction. Significance levels were adjusted for decreased degrees of freedom that resulted from smoothing the time series with a 15-year moving average.

Table 5

Extreme June precipitation values in the reconstructed time series, 1680–2005. (Values in brackets predate 1819 when sample replication is low and should be interpreted with caution.)

Table 6

Pearson correlation coefficients for significant relations between Yellowknife Airport monthly precipitation totals and monthly Pacific Decadal Oscillation Index values ( http://jisao.washington.edu/pdo/ PDO.latest). Both series are 5-year averages and the period of the correlation analysis is 1943–2005. Only significant correlations (p < 0.05) are shown. Significance levels were adjusted for decreased degrees of freedom that resulted from smoothing the time series with a 5-year moving average.

CLIMATE-GROWTH RELATIONS

Correlation analysis between the jack pine site chronologies and climate data from Yellowknife Airport suggests that a significant portion of variability in annual growth is determined by precipitation during the month of June (Table 3). Similar but slightly weaker correlations are noted with total precipitation for the periods May–July, June–July, and June–August, indicating the annual growth increment is influenced by precipitation falling throughout the short growing season and not just the single month of June. It is interesting to note that only a few of the individual site chronologies are significantly correlated with annualized precipitation indices across 12-month periods (e.g., _previousJuly–June, _previousAugust–July) (Table 3). These annualized indices often incorporate climatic inputs during two growing seasons while also including the contribution of non-growing season precipitation which can be important for soil moisture recharge and availability during the growing season (Case and MacDonald, 1995). In Yellowknife, the last day with measureable snow on the ground averages 4 May (earliest 22 April; latest 15 May; 1943–2005), while the date at which mean monthly temperatures generally rise above 10 °C is approximately 3 June ( http://www.climate.weather office.ec.gc.ca/climate_normals/index_e.html) (1943–2005). Ko Heinrichs et al. (2007) determined the onset of jack pine growth near Lac Duparquet, Quebec, commenced in mid-May when mean temperatures fluctuated widely, but were generally above 10 °C. Fluctuating air temperatures and increasing day length can promote bud burst and signal the onset of cambial reactivation and xylem cell production (Partanen et al., 1998). In Yellowknife, similar temperature conditions generally do not occur until early June, several weeks after the snow has melted and moisture on outcrops has been lost to runoff or evaporation.

The potential decoupling of snowmelt and the onset of the growing season has implications for jack pine phenological processes. Zahner (1968) suggested that a lack of adequate moisture is most important during early wood cell growth. The asynchrony between snowmelt and the onset of the growing season could result in moisture stress early in the growing season, inhibiting photosynthetic activity and protein production and ultimately reduce the resources for early wood formation and overall ring width. The correlation analyses appear to support this idea since no winter season precipitation series were significantly correlated with the site chronologies (Table 3). This is in contrast with the Campbell Dolomite Uplands in northwest NWT (Szeicz and MacDonald, 1996) where snowmelt occurs several weeks later and can provide important moisture supplies for early season tree growth. The strong response to precipitation conditions during the single month of June may be a unique feature of precipitation-sensitive trees from northern sites in contrast to those from more southerly latitudes which are able to respond to a longer growing season.

JUNE PRECIPITATION 1819–2005

From 1819 to 2003 there is good agreement between the Yellowknife precipitation reconstruction and multi-decadal variability in records from other regions of North America and the Pacific basin (Fig. 5). The period of negative precipitation anomalies reconstructed for Yellowknife from ∼1840–1865 broadly coincides with dry conditions in other parts of Canada and the United States, including reconstructed PDSI values for the Yellowknife region (Fig. 5b) (Cook et al., 2004). Lake Athabasca water levels from 1866 to 1868 were found to be the lowest during the reconstructed period (1810–1967) (Stockton and Fritts, 1973) (Fig. 5c), and Case and MacDonald (2003) reconstructed low flows on the North Saskatchewan River from 1841 to 1873 (Fig. 5d). Watson and Luckman (2004) identified the period from 1839 to 1859 as a prolonged interval of dry conditions in the southern Canadian Cordillera which is also apparent in the PDSI reconstruction (Grid Point 53; Fig. 5d) from the same region (Cook et al., 2004). Other reconstructions have identified extreme dry or drought conditions centered on the 1860s in the southern Canadian Cordillera (Watson and Luckman, 2001), and the Great Plains and south-central United States (Blasing et al., 1988; Meko, 1992). Decreased June precipitation during the early 20th century in Yellowknife also corresponds to widespread drought during the Dustbowl years throughout western Canada, the Prairies, and numerous regions of the United States (Fig. 5).

JUNE PRECIPITATION 1679–1819

Although sample depth is low in the early part of the reconstruction, the portion from 1679 to 1819 also contains multi-decadal variability that appears to coincide with similar dry/wet periods in other regions of Canada and the United States (Fig. 5). The multi-decadal variability in the Yellowknife reconstruction also appears to vary in step with sea surface conditions in the Pacific Basin related to the PDO (Fig. 5f). For example, the Yellowknife reconstruction suggests very dry June conditions during the period 1698–1735 when only 11 years with positive anomalies occurred. Similarly dry conditions are noted in the Rocky Mountain Foothills (Case and MacDonald, 1995), the southern Canadian Rockies (Watson and Luckman, 2001; Cook et al., 2004), and on the Canadian Prairies (Case and MacDonald, 2003). This period also coincides with a negative state of the PDO as reconstructed by MacDonald and Case (2005) (Fig. 5f). While interpretations of the record prior to 1819 are statistically tenuous because of low sample replication, it does appear the early portion of the reconstruction may provide valuable climatic information for the region.

CLIMATIC DRIVERS OF PRECIPITATION IN THE YELLOWKNIFE REGION

The Yellowknife June precipitation reconstruction shares many similarities with dendrohydrological reconstructions from other sites across western North America, which may provide insights into drivers of summer precipitation in the Yellowknife region. It has been widely illustrated that interactions between coupled ocean-atmosphere systems can lead to large-scale changes in climate and air mass circulation (e.g., Minobe, 1997; Nigam et al., 1999; Barlow et al., 2001). The large inertia of the oceans can lead to persistence of such anomalies for multi-decadal time periods leading to meso-scale reorganization of climate (Ropelewski and Jones, 1987; Bonsal et al., 1993; Zhang et al., 1997). Within the Pacific basin, sea surface temperatures (SSTs) have been identified as a primary driver of climate throughout western North America across multiple time scales (Namais and Cayan, 1981; Trenberth et al., 1988; Case and MacDonald, 1995; Woodhouse and Overpeck, 1998; Cayan et al., 1999; Graumlich et al., 2003; MacDonald and Case, 2005; Cook et al., 2007; MacDonald, 2007). Similarly, numerous studies have described the importance of North Pacific circulation patterns in the generation of precipitation in the Mackenzie Basin (Smirnov and Moore, 1999; Petrone and Rouse, 2000; Spence and Rausch, 2005). Given (1) the similarity in timing between periods of positive/negative precipitation anomalies in the Yellowknife reconstruction with other dendrohydrological records from North America; (2) the strong correlations between the instrumental precipitation data from Yellowknife Airport and monthly PDO index values; and (3) the coherence between the Yellowknife June precipitation reconstruction and the PDO reconstruction from MacDonald and Case (2005), it appears that conditions in the Pacific basin could be an important driver of decadal scale variability in summer precipitation in the North Slave region. The exact mechanisms that control these relations are still not clear, warranting further investigation.