Abstract

Ring-width data from 138 sites in the Canadian Prairie Provinces and adjacent regions are used to estimate summer drought severity during the past several hundred years. The network was divided into five regional groups based on geography, tree species, and length of record: the eastern Rockies, northern Saskatchewan, central Manitoba, southern Manitoba, and northwestern Ontario. Regional tree-ring records are primarily related to summer moisture and drought conditions, and are less responsive to droughts caused by deficits in winter precipitation. These summer-sensitive data are not linearly related to major modes of climate variability, including ENSO and the Pacific decadal oscillation (PDO), which primarily affect the climate of western Canada during winter. Extended drought records inferred from tree rings indicate that drought on the Canadian Prairies has exhibited considerable spatial heterogeneity over the last several centuries. For northern Saskatchewan and northwestern Ontario, intervals of persistently low tree growth during the twentieth century were just as long as or longer than low-growth intervals in the eighteenth or nineteenth centuries. Longer records from southern Alberta suggest that the most intense dry spell in that area during the last 500 yr occurred during the 1720s. At the eastern side of the prairies, the longest dry event is centered around 1700 and may coincide with low lake stands in Manitoba, Minnesota, and North Dakota. Although the Canadian Prairies were dry at times during the 1500s, there is no regional analog to the sixteenth-century “megadroughts” that affected much of the western United States and northern Mexico.

1. Introduction

Most prior studies using tree-ring data to understand past drought in the Canadian Prairie Provinces have been based on either single chronologies (composite time series that represent mean growth within the stand of trees) from individual locations (e.g., Sauchyn and Beaudoin 1998; Sauchyn et al. 2003) or multiple chronologies from a relatively small area [100–200-km transects; e.g., Case and MacDonald (1995); St. George and Nielsen (2002); Girardin and Tardif (2005)]. This work has demonstrated that data from Prairie trees can be an effective proxy for meteorological or hydrological drought, and can be used to develop annually resolved drought records that extend back several hundred years (Watson and Luckman 2006, and references therein).

Despite these encouraging results, making inferences about the drought history of the entire Prairies (or even individual regions) from a limited number of records remains a challenge because tree-ring data often contain multiple signals (of varying strengths) created by different environmental forcings. Synoptic-scale climate is usually a major influence on ring width, but tree growth can also be influenced by disturbances (affecting both single trees and the entire stand) or the local microclimate (Cook 1985). Because it is not possible to separate these influences a priori, even ring-width chronologies developed from climate-sensitive trees can include nonclimatic signals. The problem of whether or not local proxy records can be taken as representative for the broader region is complicated further by the inherent spatial heterogeneity of drought; at a given time, severe drought conditions may be present over a small area, but it does not necessarily follow that the entire Prairies must also be affected by drought. Although many ring-width chronologies from this region have been incorporated into the North American Drought Atlas (Cook and Krusic 2004), attempts to extract regional signals from networks of moisture-sensitive trees in Canada have either focused primarily on drought in regions adjacent to (and, in some cases, overlapping with) the Canadian Prairie Provinces (Watson and Luckman 2004; Watson and Luckman 2005; Girardin et al. 2006b), or have targeted other environmental variables (e.g., area burned; Girardin 2007).

In this paper, we use 138 ring-width chronologies from the Canadian Prairie Provinces and adjacent regions (Fig. 1) to estimate changes in regional drought severity since approximately 1500 a.d. Many of these data are relatively new; almost all records have been developed since the mid-1990s, and nearly half were produced after the publication of the most recent version of the North American Drought Atlas (Cook and Krusic 2004). As noted by Cook and Krusic (2004), the number of moisture-sensitive ring-width chronologies in Canada has, until recently, been quite modest, and these new collections fill substantial gaps in the previous network. Using an extensive (greater than 2 000 000 km2), relatively dense, network of chronologies allows us to focus our analysis on signals recorded by many or all records over large areas. These common signals are more likely to be related to synoptic-scale influences, including regional drought conditions, and should minimize contributions from local, nonclimatic noise. We describe the spatial and temporal variabilities of ring-width chronologies from the Canadian prairies, and examine the seasonality of moisture signals contained within regional tree-ring series. We also investigate potential connections between tree growth and a suite of major climate modes and discuss how the “climate window” (Fritts 1976) recorded by Prairie trees affects their ability to describe the influence of remote climate forcings. Finally, we examine what tree-ring data tell us about the intensity and persistence of drought during the past several hundred years as compared to drought observed during the period covered by instrumental records.

Fig. 1.

Network of ring-width sites from the Canadian Prairies and adjacent regions. Symbols correspond to regional groups: eastern Rocky Mountains, circles; northern SK, upward triangles; central MB, diamonds; southern MB, squares; and northwestern ON, downward triangles. Black circles represent communities mentioned in the text.

Fig. 1.

Network of ring-width sites from the Canadian Prairies and adjacent regions. Symbols correspond to regional groups: eastern Rocky Mountains, circles; northern SK, upward triangles; central MB, diamonds; southern MB, squares; and northwestern ON, downward triangles. Black circles represent communities mentioned in the text.

2. The Canadian Prairies

Most of the area in this study (approximately 47°–58°N and 90°–118°W) is part of the Prairie ecozone (Ecological Stratification Working Group 1996), a predominantly mixed-grass prairie with a semiarid climate and few trees. The prairie ecozone also coincides with the area commonly described as the Palliser Triangle, named after a nineteenth-century British expedition that mapped the territory between Lake Superior and the Pacific coast and assessed its potential for agriculture, mining, and settlement (Spry 1995). Most ring-width data used in this study are derived from trees growing along the margins of the prairie ecozone, including sites from the Montane Cordillera, Boreal Plains, and Boreal Shield ecozone (Ecological Stratification Working Group 1996). For simplicity, we refer to the broader region (shown in Fig. 1) as the Canadian Prairies or Prairies. This region also encompasses the drainage basin for the Nelson–Churchill River system, the third largest watershed in North America.

3. Tree-ring data

We assembled a set of regional ring-width data (Fig. 1; appendix A) from the International Tree-Ring Databank (information online at http://www.ncdc.noaa.gov/paleo/treering.html) and previously published material (Sauchyn and Beaudoin 1998; St. George and Nielsen 2002; Watson and Luckman 2002; Case and MacDonald 2003; Girardin and Tardif 2005; Beriault and Sauchyn 2006; Girardin et al. 2006b; Pederson et al. 2006; St. George et al. 2008). In many cases, these data were developed from trees under ecological conditions known to enhance the importance of moisture to tree growth [e.g., the forest–prairie border, sandy soils, or steep slopes; Fritts (1976); Meko et al. (1995)]. Prior studies have demonstrated that ring-width data from most of these sites exhibit some type of moisture signal, either related to seasonal or annual precipitation (Case and MacDonald 1995; Case 2000; Watson and Luckman 2001; Watson and Luckman 2002; St. George and Nielsen 2002; Sauchyn et al. 2003; St. George et al. 2008), summer drought (Sauchyn and Skinner 2001; Girardin and Tardif 2005; Girardin et al. 2006b; Pederson et al. 2006), area burned (Girardin et al. 2006a; Girardin 2007), or streamflow (Case and MacDonald 2003; Beriault and Sauchyn 2006; Watson and Luckman 2006). The tree-ring network can be divided into five regional groups based on geography, tree species, and length of record: (i) the eastern Rockies {primarily Douglas fir [Pseudotsuga menziesii subsp. glauca (Murray)] and limber pine [Pinus flexilus (James)] from lower-elevation sites in Alberta and Montana}, (ii) northern Saskatchewan {boreal forest trees, mainly white spruce [Picea glauca (Moench) Voss] and jack pine [Pinus banksiana Lamb.]}, (iii) central Manitoba [eastern white cedar (Thuja occidentalis L.) and a mixture of other species], (iv) southern Manitoba {bur oak [Quercus macrocarpa (Michx.)] from riparian forests}, and (v) northwestern Ontario [mainly red pine (Pinus resinosa Sol. ex Aiton) and white pine (Pinus strobus L.)]. The total network includes measurements of nearly 5000 samples taken from almost 3000 trees. Because of a number of factors (including the scarcity of trees in grassland ecosystems, frequent fires in the boreal forest, wet conditions that encourage heartrot, and intensive logging), the region contains relatively few long-lived trees. Although trees in the foothills of the Rockies can be quite old (the oldest tree, from the Teton River valley in Montana, has 846 rings), the median age of trees in the network is only 135 yr.

4. Instrumental climate data

Precipitation data for the Canadian Prairie region were extracted from the Climatic Research Unit (CRU) TS 2.1 gridded (0.5°) dataset of monthly climate observations (Mitchell and Jones 2005) and combined into annual (October–September) and seasonal (December–February, DJF; March–April, MAM; June–August, JJA and September–November, SON) totals. We also used gridded Palmer drought severity index (PDSI) records from the dataset developed by Dai et al. (2004), but we truncated these records at 1900 because of the scarcity of climate stations on the Canadian Prairies prior to the twentieth century. Some analyses were repeated using the Cook et al. (2004a) gridded set of instrumental PDSI data for North America but we did not note any major differences in our results. All of these gridded products are produced from individual station records, and because of the lack of high quality, long-term climate stations in many parts of northern Canada, it is necessary to be cautious when interpreting potential climate signals in the northern Prairies.

Climate indices for several major modes of climate variability, including the Atlantic multidecadal oscillation (AMO), Pacific decadal oscillation (PDO), the cold tongue index (CTI), the Pacific–North America teleconnection pattern (PNA), and the northern annular mode (NAM), were downloaded from the Joint Institute for the Study of the Atmosphere and Ocean Web site (information online at http://jisao.washington.edu) and the National Oceanic and Atmospheric Administration (information online at http://www.cdc.noaa.gov), and averaged to produce either annual (AMO only) or cold season (October–March) means.

5. Analytical procedures

Ring-width measurements were detrended and combined to produce “standard” and “residual” (prewhitened) chronologies (Cook 1985; Cook et al. 1990) for each site. We attempted to emphasize signals that are common to many or all chronologies within each region (which are likely related to synoptic-scale influences like regional climate), and diminish signals that are present at only a few sites (which are probably caused by local factors like microclimatic differences or ecological disturbances). We used either empirical orthogonal function (EOF) analysis (North et al. 1982; Peixoto and Oort 1992) or simple averaging to identify the dominant mode of variability present in each region (i.e., the leading EOF or the regional mean) and construct summary series that represent regional tree growth (“regional tree-ring records”). A more complete description of the methods used to produce the ring-width chronologies and regional tree-ring records, including a discussion of standardization techniques, signal strength statistics, and our EOF approach, is provided later (see appendix C).

Potential climate signals in regional tree growth were evaluated by comparing individual ring-width chronologies and the regional tree-ring records with (i) gridded climate products and (ii) indices of major modes of climate variability that might plausibly influence the Canadian prairies. All significance tests for correlation were adjusted for the loss of degrees of freedom associated with autocorrelated time series (Dawdy and Matalas 1964). Cross-spectral analysis was conducted using Welch’s averaged periodogram method (MathWorks 2007; Trauth 2006). Significance levels for coherence (squared) were computed by a Monte Carlo procedure: 10 000 series with the same spectral properties as the regional tree-ring records were generated by exact simulation (Percival and Constantine 2006), and were subjected to cross-spectral analysis to establish empirical confidence levels.

6. Results

a. Ring-width chronologies

Most ring-width chronologies (60%) describe at least 85% of their stand-wide signal (Wigley et al. 1984; appendixes B and c) back until 1850 a.d., but there is a rapid dropoff thereafter. Only 30 chronologies retain their signal to 1700 a.d., only 15 to 1600 a.d., and only 7 (all from the eastern Rockies) contain a robust signal prior to 1500 a.d. Only the leading eigenvectors (EOF1) from each region were retained for subsequent analysis; effectively, these eigenvectors (Fig. 2) describe the common regional signal that allows tree-ring patterns to be matched between sites several hundred kilometers apart. Because the set of tree-ring sites from central Manitoba includes a small number of chronologies (10) that are weakly correlated with each other, we were not able to define a meaningful summary series to represent tree growth in this region.

Fig. 2.

Eigenvector loadings for EOF analysis of regional ring-width groups: (a) eastern Rocky Mountains EOF1 (standard chronologies), (b) northern SK EOF1 (residual chronologies), (c) central MB EOF1 (standard chronologies), (d) southern MB EOF1 (standard chronologies), and (e) northwestern ON EOF1 (standard chronologies). The values in parentheses indicate the amount of variance described by each pattern relative to the total in each regional network. The period of analysis is 1900–2004. The area of the circle is proportional to the eigenvector loadings for each chronology.

Fig. 2.

Eigenvector loadings for EOF analysis of regional ring-width groups: (a) eastern Rocky Mountains EOF1 (standard chronologies), (b) northern SK EOF1 (residual chronologies), (c) central MB EOF1 (standard chronologies), (d) southern MB EOF1 (standard chronologies), and (e) northwestern ON EOF1 (standard chronologies). The values in parentheses indicate the amount of variance described by each pattern relative to the total in each regional network. The period of analysis is 1900–2004. The area of the circle is proportional to the eigenvector loadings for each chronology.

b. Climate signals in prairie tree rings

In general, regional tree-ring records from the Prairies are correlated with summer (JJA) precipitation over a broad area surrounding the sites used to define each regional pattern (Fig. 3). With the exception of the data from northern Saskatchewan, the regional tree-ring records developed from standard ring-width chronologies were more highly correlated with regional precipitation (and PDSI) than those developed from residual chronologies. Several regional series (eastern Rockies, southern Manitoba, and northwestern Ontario) are correlated with precipitation records from grid points more than 500 km away from the tree-ring sites. The leading pattern for northern Saskatchewan shows the weakest connection to summer precipitation and, instead, has higher and more extensive correlations with spring (primarily May) precipitation in central Saskatchewan (see Fig. S4 in the supplementary material for this paper). The tree-ring records for southern Manitoba and the eastern Rockies are also correlated with spring precipitation, but for the most part, these correlations are weaker and extend over a smaller area than those obtained using summer precipitation. None of the regional tree-ring series are significantly correlated with precipitation during the prior autumn or winter (see Figs. S1–S4 in the supplementary material). The regional series also correlate with annual precipitation, but this correspondence likely reflects the importance of summer rainfall on the Prairies rather than a real physical relationship between ring width and precipitation over the entire year (see the section 7).

Fig. 3.

Correlations between major patterns of regional tree-ring width and gridded summer (JJA) precipitation, calculated for the 1901–2004 interval. Only correlations that are significant at the 0.05 level are shown. The white circles mark the tree-ring chronologies used to define each regional pattern.

Fig. 3.

Correlations between major patterns of regional tree-ring width and gridded summer (JJA) precipitation, calculated for the 1901–2004 interval. Only correlations that are significant at the 0.05 level are shown. The white circles mark the tree-ring chronologies used to define each regional pattern.

Regional tree-ring records are also correlated with July PDSI over large parts of the Prairies (Fig. 4). Tree growth in southern Manitoba and the eastern Rockies appears to record the most widespread drought signals, with both series showing significant correlations over the entire southern prairies and adjacent American states. In general, the regional series are mostly highly correlated with PDSI records from nearby grid points, which likely reflects the direct connection between tree growth and local drought conditions. The series for northern Saskatchewan and southern Manitoba are most highly correlated with drought records one or two grid points away; these offsets are probably artifacts created by interpolating the gridded climate data from a sparse network of stations.

Fig. 4.

Correlations between regional ring-width indices and gridded July PDSI. Only correlations that are significant at the 0.05 level are shown. Red squares mark grid points with the highest absolute correlation between regional ring width and PDSI. The period of analysis is 1900–2002.

Fig. 4.

Correlations between regional ring-width indices and gridded July PDSI. Only correlations that are significant at the 0.05 level are shown. Red squares mark grid points with the highest absolute correlation between regional ring width and PDSI. The period of analysis is 1900–2002.

We identified those grid points whose PDSI records have the highest absolute correlation with the regional tree-ring records (marked by red squares in Fig. 4), and compared the PDSI and ring-width time series (Fig. 5). The PDSI grid points are named after nearby communities: Lethbridge, Alberta; The Pas, Manitoba; Estevan, Saskatchewan; and Dryden, Ontario. The regional series tracks PDSI most closely for the eastern Rockies–Lethbridge (r = 0.62, p < 0.0001) and southern Manitoba–Estevan (r = 0.52, p < 0.0001). Tracking is weakest in northern Saskatchewan (r = 0.26, p < 0.02), possibly because trees in this region are less limited by moisture and less sensitive to drought, but it may also reflect problems with instrumental PDSI records related to the paucity of long-term climate stations in this area (New et al. 2000).

Fig. 5.

Time series for instrumental PDSI (solid lines) and regional ring width (dashed lines) with the interseries correlations (“r”): (a) eastern Rockies PC1, (b) northern Saskatchewan PC1, (c) southern Manitoba mean, and (d) northwestern Ontario PC1.

Fig. 5.

Time series for instrumental PDSI (solid lines) and regional ring width (dashed lines) with the interseries correlations (“r”): (a) eastern Rockies PC1, (b) northern Saskatchewan PC1, (c) southern Manitoba mean, and (d) northwestern Ontario PC1.

The relationship between PDSI and tree growth appears to be frequency dependant. Both PDSI and tree-ring series have predominantly low-frequency spectra (Table 1) and two of the four tree-ring–PDSI pairs (eastern Rockies–Lethbridge and southern Manitoba–Estevan) show greater coherence at low frequencies (Fig. 6). For the eastern Rockies, coherence is significant (0.05 level) at wavelengths longer than about 5 yr. For the southern Manitoba–Estevan pair, coherence is higher (but not statistically significant) above 10 yr and significant (p = 0.05) around 3 yr. Tree-ring and PDSI records from northwestern Ontario show a strong peak in coherence around 2–3 yr, and a second, less significant, peak around 10 yr. Cross-spectral results for northern Saskatchewan confirm that these trees contain a relatively weak PDSI signal: peaks in coherence are found near 3 and 10 yr but neither reaches significance at the 0.05 level.

Table 1.

Fraction of variance at low frequencies (≥5 yr) in regional ring-width and PDSI series.

Fraction of variance at low frequencies (≥5 yr) in regional ring-width and PDSI series.
Fraction of variance at low frequencies (≥5 yr) in regional ring-width and PDSI series.
Fig. 6.

Coherence between regional ring-width and PDSI records. The 90% and 95% confidence levels are represented by the dashed lines and were determined by a Monte Carlo approach that generated 10 000 random series with the same spectral properties as the original time series: (a) eastern Rockies PC1, (b) northern Saskatchewan PC1, (c) southern Manitoba mean, and (d) northwestern Ontario PC1.

Fig. 6.

Coherence between regional ring-width and PDSI records. The 90% and 95% confidence levels are represented by the dashed lines and were determined by a Monte Carlo approach that generated 10 000 random series with the same spectral properties as the original time series: (a) eastern Rockies PC1, (b) northern Saskatchewan PC1, (c) southern Manitoba mean, and (d) northwestern Ontario PC1.

For the most part, individual chronologies display the same sensitivity to summer precipitation as the regional tree-ring series: more than half of all chronologies (82) show significant and spatially extensive correlations with summer precipitation. A smaller number of chronologies (25) are significantly correlated with spring precipitation, and six records, almost all from the eastern Rockies, show an apparent signal related to the amount of snow in winter. Some individual chronologies are more highly correlated with July PDSI than the corresponding regional tree-ring series. In northern Saskatchewan, seven chronologies (out of 16) are more highly correlated with July PDSI at The Pas than the regional tree-ring series. In northwestern Ontario, 10 chronologies (out of 55) are more highly correlated with PSDI than the regional tree-ring series. This tendency is not present in southern Manitoba, where the highest correlation between an individual chronology and PDSI is equivalent to the correlation between PDSI and the regional tree-ring series. It is also absent in the western Prairies; none of the ring-width chronologies from the eastern Rockies are as highly correlated with the Lethbridge PDSI record as the regional tree-ring series.

c. Associations with major climate modes

Mapping the correlation between individual ring-width chronologies and major climate indices shows that ring width across the region is not associated with either the CTI (a measure of ENSO) or the PNA pattern (Fig. 7). The other three indices have a greater number of significant correlations, but for the most part they do not create the coherent spatial fingerprint that would be expected if these modes had a real influence on prairie trees. For example, the PDO is significantly correlated with just five chronologies in the eastern Rockies, and the sign of the correlation is not consistent among these records. Because trees in the eastern Rockies exhibit a strong regional signal and should respond similarly to a synoptic-scale climate forcing, it seems likely that these correlations with the PDO are spurious. Several chronologies from northwestern Ontario are significantly correlated with the AMO, the PDO, and the NAM, but again, these correlations are not present at all sites in the region. The apparent correspondence between the AMO and tree growth in northwestern Ontario is difficult to reconcile with prior analysis of instrumental records, which has shown that the AMO does not affect drought in this part of Canada (Shabbar and Skinner 2004). Finally, none of the regional tree-ring records are significantly correlated with any of the five climate indices (not shown).

Fig. 7.

Correlations between ring-width chronologies on the Canadian Prairies and major climate modes. Colored circles show correlations significant at the 0.05 level, after the adjustment for temporal autocorrelation. The area of the circle is proportional to the correlation coefficient for each chronology: (a) AMO, (b) PDO, (c) CTI, (d) PNAP, and (e) NAM.

Fig. 7.

Correlations between ring-width chronologies on the Canadian Prairies and major climate modes. Colored circles show correlations significant at the 0.05 level, after the adjustment for temporal autocorrelation. The area of the circle is proportional to the correlation coefficient for each chronology: (a) AMO, (b) PDO, (c) CTI, (d) PNAP, and (e) NAM.

d. Tree-ring estimates of past drought

The above results demonstrate that ring-width chronologies from the Canadian Prairies contain regionally coherent signals (with the exception of chronologies from central Manitoba), and that these signals are associated with summer moisture (June–August precipitation) or drought (July PDSI) conditions. The connection between summer drought and tree rings is strongest in the eastern Rockies and southern Manitoba, but a significant drought signal is also present in northern Saskatchewan and northwestern Ontario. We attempted to model the relationship between the four target PDSI series and the suite of regional tree-ring records with traditional, regression-based approaches (Fritts 1976; Bradley 1999), but found that most models explained a modest amount of variance in the PDSI series (generally between 20% and 30%, with a maximum explained variance of 40%). In several cases, the PDSI series was most effectively modeled as a simple linear transformation of its corresponding regional tree-ring records (from Fig. 5).

In light of the weak statistical relationships between ring width and PDSI, we chose to interpret the regional tree-ring record as qualitative measures of past drought severity (Fig. 8). Each regional series was scaled so that, during the period covered by instrumental drought records (1900–2004), its mean was equal to zero and its standard deviation was equal to one. This approach is effectively a simplified version of the composite-plus-scale technique (Jones and Mann 2004; Esper et al. 2005). The regional series were also smoothed with a 5-yr Gaussian filter to emphasize longer-term variability that, at least in some regions, is more closely tied to similar behavior in the instrumental PDSI records. The terminology used to describe the inferred drought records is modeled after Biondi et al. (2002); duration is the number of consecutive years the series remains above or below the instrumental period mean, and intensity is the average departure from the mean for a given duration.

Fig. 8.

The tree-ring record of drought in the Canadian Prairies. Each series is scaled to have a zero mean and equal variance during the instrumental period (1900–2004). The heavy line illustrates variability on time scales >5 yr.

Fig. 8.

The tree-ring record of drought in the Canadian Prairies. Each series is scaled to have a zero mean and equal variance during the instrumental period (1900–2004). The heavy line illustrates variability on time scales >5 yr.

1) Eastern Rockies

The tree-ring record from southern Alberta and northern Montana suggests that at least one drought during the last 500 yr was more intense than droughts observed in the instrumental record. The first attempts to derive drought history using tree rings from this region were based on a limited number of ring-width chronologies and suggested that the 1790s was the driest decade in southern Alberta in the last several hundred years (Case and MacDonald 1995; Sauchyn and Beaudoin 1998; Wolfe et al. 2001). Our much larger dataset indicates that tree growth was exceptionally low in the 1790s at only a few sites (most prominently, the Towers Ridge site near Calgary, Alberta). The regional series suggests that the most intense drought in southern Alberta since 1500 a.d. occurred in the early 1720s. The most extreme low-growth year in the eastern Rockies record (1720) also falls within this interval. The dry spells with the greatest duration occurred in the early twentieth century (1918–42) and the middle of the nineteenth century (1842–77). The earlier drought coincides with the Palliser expedition of 1858–59, which reported sparse rains, dry river beds, and arid country throughout southern Alberta and Saskatchewan (Spry 1995; Rannie 2006). The tree-ring evidence suggests that southern Alberta had already been dry for a decade and a half at the time of Palliser’s expedition, and would remain so until the late 1870s. This interval also encompasses the “Civil War Drought” (1856–65) identified in the central United States (Herweijer et al. 2006), but it does not seem that droughts in both regions can be attributed to the same climate forcing. Herweijer et al. (2006) argued that the prolonged drought in the U.S. Great Plains was driven by the La Niña–like tropical Pacific, but because this configuration generally brings more moisture to western Canada (Shabbar et al. 1997; Bonsal and Lawford 1999), this mechanism cannot account for dry summer weather in the eastern Rockies at the same time. Finally, the first two decades of the twentieth century appear to have been one of most prolonged wet periods in the entire record, lending support to the argument that first major pulse of Euro-Canadian settlement in the western prairies occurred during unusually favorable climate conditions (Sauchyn and Beaudoin 1998; Watson and Luckman 2006). This period also coincided with the early twentieth-century pluvial observed in the western United States (Woodhouse et al. 2006; Pederson et al. 2006), and unusually high flows in the Colorado River (Meko et al. 2007). The early 1800s is the only other wet period of comparable duration; this pluvial included exceptionally high tree growth at sites across the region in 1839.

2) Northern Saskatchewan

Tree-ring data from northern Saskatchewan identify persistent dry conditions in the late 1880s, early 1900s, and circa 1940. The last period is coincident with the lowest recorded flow in the Churchill River (Beriault and Sauchyn 2006), which drains a large portion of northern Manitoba and Saskatchewan. Low growth across northern Saskatchewan during 1869 coincided with poor growth in the eastern Rockies, southern Manitoba, and the western half of the northwestern Ontario network; several prior studies have noted that drought was widespread across most of western Canada in this year (Watson and Luckman 2005; Cook and Krusic 2004). As in southern Alberta, this area appears to have experienced persistent wet conditions at the beginning of the twentieth century.

3) Southern Manitoba

The most prominent feature of the inferred drought record for southern Manitoba and southeastern Saskatchewan is a prolonged absence of wet years around 1700. Between 1688 and 1715, growth was above the instrumental period mean for only 5 yr out of 28. An earlier analysis of the southern Manitoba tree-ring record using a different standardization technique (regional curve standardization) also indicated that this region was dry for several decades circa 1700 (St. George and Nielsen 2002). Radiocarbon dates of drowned trees on the margin of Lake Winnipeg suggest that the lake was lower by roughly 1 m around 1650 a.d. (Nielsen 1998). Given the limitations of radiocarbon dating within the last 200–300 yr (Stuiver and Reimer 1993), it is possible that lowered lake levels coincided with prolonged drought upstream in southern Manitoba. A composite flood history for the Red River in southern Manitoba was developed from gauge records, historical observations, and anatomical signatures in riparian trees and records major floods in 1747, 1762, 1826, 1852, 1862, 1950, and 1997 (St. George and Nielsen 2003). Except for 1747, these floods occurred during periods of persistent above-average wet conditions inferred from the regional tree-ring record.

4) Northwestern Ontario

The regional tree-ring series for northwestern Ontario indicates that the most extreme and prolonged drought conditions occurred during the twentieth century. The duration of recent dryness (most prominently from 1906 to 1915 and 1974 to 1999) is unmatched in the earlier part of the record. The extreme low growth exhibited by regional trees in 1910 is well beyond any of the other negative growth anomalies that have occurred since 1783. This observation led us to conclude that summer drought in northwestern Ontario in the twentieth century was more extreme than drought in the nineteenth and late eighteenth centuries. Twentieth-century wetness appears to be similarly exceptional. Between 1951 and 1972, regional tree growth was above the mean for 20 yr out of 23, including the record’s peak growth year of 1965.

7. Discussion

Our results demonstrate that ring-width data from the Canadian Prairies primarily describe signals related to moisture and drought conditions during summer. In some cases, regional tree-ring records are correlated with summer precipitation and drought records several hundred kilometers distant. Fritts (1976) introduced the concept of the “climatic window,” whereby certain climatic “inputs” pass through the physiological system of a tree and are converted into a certain ring-width “output.” Within this framework, the tree can be thought of as a filter that emphasizes some climatic signals and minimizes others. This idea can be extended to include the behavior of the climate system as well: aspects of the climate system that are highly variable are more likely to create stronger signals in tree-ring series than those that are relatively static. The seasonality (or climate window) of Prairie tree-ring records has important implications for their interpretation as drought proxies.

First, the sensitivity of Prairie tree rings to summer moisture serves to limit their connection to major climate modes. The impact of ENSO and the PDO on precipitation over southern Canada is strongest during winter and is dampened out by the following spring (Shabbar et al. 1997; Mantua and Hare 2002). Several authors have noted that summer drought on the Prairies is statistically associated with wintertime ENSO conditions 1–2 yr prior (Bonsal et al. 1993; Bonsal and Lawford 1999; Shabbar and Skinner 2004), but, although Prairie trees are primarily sensitive to summer drought, there is no apparent association between regional tree-ring records and either ENSO or the PDO. We infer that (i) Prairie trees are largely insensitive to the stronger winter precipitation signals created by either ENSO or the PDO and (ii) any summertime ENSO signal is too weak to be detected using tree-ring records.

Second, the absence of a winter moisture signal at nearly all sites is the most likely reason why tree rings do not exhibit a pronounced negative anomaly associated with the recent 2000–01 drought, which has been described as the most intense drought in the western prairies during the last 100 yr (Sauchyn et al. 2003; Liu et al. 2004). Although the summer was also dry, the 2000–01 drought was caused primarily by a deficit in winter precipitation (Liu et al. 2004), which, because of its timing, had a limited impact on regional tree growth. Trees at sites in the eastern Rockies region formed narrow rings in 2001 but overall, tree-ring records from this area were not as low as during other twentieth-century droughts, such as 1936–37 or 1961.

Finally, the importance of summer rainfall to the Canadian Prairies creates an association between regional tree growth and total annual precipitation. Several prior studies have used ring-width chronologies from the Prairies to reconstruct annual precipitation (e.g., Case and MacDonald 1995; Sauchyn and Beaudoin 1998; St. George and Nielsen 2002; Watson and Luckman 2005). Our results suggest that, at the regional scale, the correspondence between prairie tree rings and annual precipitation may reflect a statistical association driven by a true causal relationship with summer precipitation. Although tree-ring records are, for some regions, significantly correlated with spring precipitation (especially in northern Saskatchewan), they are not significantly related to precipitation during the prior autumn or winter. If these data contained a true annual signal integrating precipitation over the entire year, they should correlate with precipitation during all seasons, not just summer. Instead, the association between all four regional tree-ring series and the annual precipitation is likely an artifact created by the prominence of summer rainfall in this part of Canada. The precipitation regime of the Prairies is summer dominant (Borchert 1950; Bonsal et al. 1999); that is, summer is the wettest season, and summer rainfall is more variable and more highly correlated with annual totals than precipitation during the other seasons. As a consequence, any proxy record that is strongly influenced by summer rainfall will also track annual precipitation, even if that record is not sensitive to precipitation anomalies during the other three seasons. Because summer rainfall and the total annual precipitation are so similar, regional ring-width records can be used to infer the past variability of either variable; however, it should be considered that such inferences depend primarily on signals created by summer conditions.

In some regions, it may be possible to develop better drought reconstructions (i.e., explain more variance) by using a subset of ring-width chronologies rather than all ring-width chronologies within a region. Dendroclimatic studies often use screening procedures to identify chronologies that are correlated with a specific climate variable and use those chronologies (only) as the basis for subsequent analysis (e.g., Cook et al. 1999; Meko et al. 2001; Woodhouse et al. 2006). This approach is best suited for tree-ring networks that contain chronologies that are either uncorrelated with climate or are sensitive to a different aspect of climate than the rest of the set. This alterative may be viable in northern Saskatchewan and northwestern Ontario, as both networks include several chronologies that are more similar to PDSI than their regional tree-ring series. If regional tree-ring series were built using only those chronologies that are highly correlated with PDSI, it may be possible to obtain a more “pure” drought signal and produce a statistically superior regression model. The main weakness of this method is the attendant decrease in sample size: the resulting reconstruction would be derived from fewer trees and would be potentially more vulnerable to the effect of nonclimatic influences. Our results suggest that drought estimates for the eastern Rockies or southern Manitoba would not be improved by some type of screening procedure. For both areas, the regional tree-ring series appears to be a better proxy of PDSI than any of its component ring-width chronologies. However, screening may help isolate ring-width variability that reflects the influence of winter precipitation. A small number of chronologies in the eastern Rockies (particularly Pyramid Lake, Lake Annette, Maligne Canyon, Teton River Valley Douglas fir, and Bears Paw Mountains Douglas fir) are significantly correlated with winter precipitation, but this behavior is present at too few sites to be resolved by the EOF analysis. The apparent sensitivity of these sites to winter precipitation may be due to random chance, but might also reflect a real but subtle signal that is poorly characterized by the current ring-width network. More records in the eastern Rockies are needed to determine if this association describes a real relationship and to identify the site characteristics that make some trees at the lower-forest border sensitive to winter conditions.

Tree-ring evidence suggests that the drought on the Canadian Prairies has exhibited considerable spatial heterogeneity over the last 500 yr, and it is rarely appropriate to describe the entire region as being either wet or dry. Although the two longest inferred drought records (the eastern Rockies and southern Manitoba) are significantly correlated (r = 0.2, p = 0.00001), there are several examples when one region was in drought and the other was not. The 1720s drought, which was extremely intense in the west, does not appear to have affected the eastern prairies, which was itself just recovering from a prolonged drought circa 1700 that went unrecorded in the west. Most parts of the prairies were dry at some time during the 1850s and 1860s, but this mid-nineteenth-century drought lasted the longest in southern Alberta and northern Montana. The lack of regional coherence in the tree-ring records is likely a product of the nature of summer precipitation on the Prairies. Although cyclonic and frontal activity plays a role in summer precipitation, rainfall is predominately convective and has high spatial variability (Raddatz 2005). Consequently, the spatial variability in the Prairie climate creates similarly variable (and localized) signals in the regional tree-ring network. The examples of contrasting drought conditions in the eastern and western sectors imply that the mechanisms that control persistent summer drought in each region are largely independent, at least those that operate at longer time scales, and corroborate Rannie (2006)’s warning against making general statements about drought on the entire Prairies based on proxy evidence from only one region.

We make a pair of observations on the potential connections between inferred drought in the Prairies and climate events at the continental or global scale. First, although the Canadian Prairies were dry at times during the 1500s, there is no regional analog to the sixteenth-century “megadroughts” that affected much of the western United States and northern Mexico (Stahle et al. 2000). Stahle et al. (2000) hypothesized that this widespread drought may have been caused by persistent cool water in the eastern equatorial Pacific (La Niña conditions). The Canadian prairies might be expected to be wet under those conditions, as modern La Niñas are associated with stronger westerlies moving across the eastern North Pacific and enhanced moisture delivery to western Canada during winter (Shabbar et al. 1997; Bonsal and Lawford 1999). Indeed, the eastern Rockies were wet throughout the 1590s, and although the region was dry for several years in the mid- to late 1500s (between 1555 and 1585), this episode does not appear as a prominent drought in this record. The southern Manitoba record also does not show any notable dry intervals during the sixteenth century, with the exception of 1595, the year of the lowest growth in that record. Second, the Prairies were not consistently wet or dry during the putative “Little Ice Age” [1570–1900 a.d.; Matthews and Briffa (2005)]. Tree-ring records exhibit considerable spatial and temporal variabilities during this period and include evidence of both wet and dry intervals. These data suggest that it is not appropriate to conceptualize the Little Ice Age on the prairies as a uniform, prolonged wet (or dry) anomaly extending over the entire region.

Does the instrumental period describe the complete range of drought variability on the Canadian Prairies during the last few hundred years? The answer to that question varies by region, and depends on how far back in time the tree-ring records allow us to look. For the cooler, wetter parts of the Prairies (northern Saskatchewan and northwestern Ontario), drought histories inferred from tree rings are only available for the last 150–200 yr. These records show that, during the twentieth century, intervals of persistently low tree growth were just as long as or longer than low-growth intervals in the eighteenth or nineteenth centuries. From this, we infer that instrumental climate records provide a reasonable approximation of the maximum duration of summer drought in these regions since circa 1800 a.d. The longer perspective afforded by tree-ring records in southern Manitoba and the eastern Rockies presents a different picture. The mid-nineteenth-century drought in southern Alberta and northern Montana had a greater duration than did the drought during the early twentieth century. The 1720s drought was not as long lasting, but was the most intense dry spell in the western Prairies in the last 500 yr. On the eastern side of the Prairies, the most notable dry event was centered around 1700 a.d. and persisted for almost three decades. Because of the limitations of radiocarbon dating, we cannot make unequivocal comparisons between the 1700 dry period in the eastern Prairies and the dry conditions inferred from geological evidence and most (nonvarved) limnological records in the region. Nonetheless, we do note that it is certainly possible that prolonged drought in southern Manitoba coincided with low stands in small lakes from Minnesota and North Dakota (St. George and Nielsen 2002), and low water levels in Lake Winnipeg (Nielsen 1998). In at least these two regions then, tree-ring evidence indicates that droughts that occurred within the last 300 yr were more intense and more persistent than those observed during the period covered by instrumental records.

8. Concluding remarks

The regional drought histories developed in this study may differ from earlier tree-ring reconstructions in some of the details, but the overall message remains largely the same. Tree-ring evidence indicates that parts of the Canadian Prairies have, relatively recently, been affected by summer droughts that were more intense and more long lasting that those observed in instrumental records. If similar droughts occurred in the future, it is not clear what impacts they would have on activities that rely heavily on inputs from summer moisture, such as agriculture, hydropower, or consumptive supplies. It may be prudent to examine the resilience of these systems to alternate worst-case scenarios based on these regional drought records inferred from tree rings. This exercise may be particularly important for the dryland areas of the western Prairie Provinces, which in the last few decades have experienced rapid population growth and agricultural expansion and concomitant increases in regional water demand (Schindler and Donahue 2006).

What aspects of the broader climate system have influenced prairie drought patterns during the last several centuries? The answer to this question remains uncertain, particularly because drought records derived from tree rings show relatively modest regional synchrony during the last several centuries. This apparent spatial heterogeneity suggests that persistent summer drought is principally related to regional- or local-scale forcings, rather than being driven by a single, synoptic-scale control that entrains drought conditions over the entire Canadian Prairies. Furthermore, although several studies have linked preinstrumental droughts in the United States to persistent La Niña conditions (Stahle et al. 2000; Herweijer et al. 2006), prairie droughts cannot be attributed to the same cause because western Canada occupies the opposite node of the precipitation dipole associated with ENSO.

It is important to recognize that conventional tree-ring records from the Canadian Prairies are somewhat limited in their application as proxies for past hydroclimatic conditions. Despite the addition of several dozen new chronologies, the regional tree-ring series are still a relatively poor basis for developing quantitative estimates of precipitation or drought severity, at least in comparison to tree-ring data from other parts of North America. Adding more ring-width chronologies might produce better results, but it might be more productive to supplement the existing ring-width network with records of other tree-ring parameters, such as earlywood–latewood width (Meko and Baisan 2001; Watson and Luckman 2002), wood density (Rigling et al. 2001), or isotopic composition (Leavitt et al. 2002; Masson-Delmotte et al. 2005). Multiproxy tree-ring studies in the southern French Alps (Gagen et al. 2004, 2006) indicate that drought reconstructions based on multiple tree-ring parameters may be more accurate than estimates derived from any single parameter. Because drought records inferred from ring-width data from the Canadian Prairies are relatively uncertain, adding new information from other tree-ring parameters could improve the quality of dendroclimatic reconstructions substantially.

There remains a critical need to develop more long chronologies on the Canadian prairies, either from old living trees (most promising in the eastern Rockies) or by extending living-tree records with logs recovered from historical buildings or geological settings. The current network of tree-ring sites is adequate to describe broad, regional patterns in ring width during the last 100–500 yr (varying by region), but there are not enough long chronologies to describe large-scale variability in any region prior to circa 1500 a.d. In regions like northern Saskatchewan or northwestern Ontario, the limit of tree-ring records is much more recent, with few to no trees older than 200 yr. Even with the relative abundance of old-growth trees in the foothills of the Rocky Mountains, estimates of drought variability prior to the sixteenth century are derived from a much smaller set of tree-ring records and should be viewed with a degree of extra caution. This issue becomes only more important as we attempt to go further back in time, and as a result, it is not yet possible to use tree-ring records to adequately resolve the impacts of the “Medieval Climate Anomaly” (roughly 900–1300 a.d.) evident at many sites in the western United States (Cook et al. 2004b; Meko et al. 2007) and on the Canadian Prairies.

Acknowledgments

This article is a contribution to Natural Resources Canada’s Enhancing Resilience in a Changing Climate Program. This research was funded by Manitoba Hydro, the Manitoba Geological Survey, the Prairie Adaptation Research Collaborative, Natural Resources Canada, and the Natural Sciences and Engineering Research Council of Canada. We thank Julio Betancourt, Michael Evans, and Stephen Wolfe for helpful comments on an earlier version of this manuscript.

REFERENCES

REFERENCES
Beriault
,
A.
, and
D. J.
Sauchyn
,
2006
:
Tree-ring reconstructions of streamflow in the Churchill River basin, northern Saskatchewan.
Can. Water Resour. J.
,
31
,
249
262
.
Biondi
,
F.
,
T. J.
Kozubowski
, and
A. K.
Panorska
,
2002
:
Stochastic modeling of regime shifts.
Climate Res.
,
23
,
23
30
.
Bonsal
,
B. R.
, and
R. G.
Lawford
,
1999
:
Teleconnections between El Niño and La Niña events and summer extended dry spells on the Canadian prairies.
Int. J. Climatol.
,
19
,
1445
1458
.
Bonsal
,
B. R.
,
A. K.
Chakravarti
, and
R. G.
Lawford
,
1993
:
Teleconnections between North Pacific SST anomalies and growing season extended dry spells on the Canadian prairies.
Int. J. Climatol.
,
13
,
865
878
.
Bonsal
,
B. R.
,
X.
Zhang
, and
W. D.
Hogg
,
1999
:
Canadian prairie growing season precipitation variability and associated atmospheric circulation.
Climate Res.
,
11
,
191
208
.
Borchert
,
J. R.
,
1950
:
The climate of the central North American grassland.
Ann. Assoc. Amer. Geogr.
,
40
,
1
39
.
Bradley
,
R. S.
,
1999
:
Paleoclimatology: Reconstructing Climates of the Quaternary.
Academic Press, 613 pp
.
Case
,
R. A.
,
2000
:
Dendrochronological investigations of precipitation and streamflow for the Canadian prairies.
Ph.D. thesis, University of California, Los Angeles, 249 pp
.
Case
,
R. A.
, and
G. M.
MacDonald
,
1995
:
A dendroclimatic reconstruction of annual precipitation on the western Canadian prairies since A.D. 1505 from Pinus flexilis James.
Quat. Res.
,
44
,
267
275
.
Case
,
R. A.
, and
G. M.
MacDonald
,
2003
:
Tree ring reconstructions of streamflow for three Canadian prairie rivers.
J. Amer. Water Resour. Assoc.
,
39
,
703
716
.
Cook
,
E. R.
,
1985
:
A time series analysis approach to tree-ring standardization.
Ph.D. dissertation, The University of Arizona, 171 pp
.
Cook
,
E. R.
, and
P. J.
Krusic
,
cited
.
2004
:
The North American Drought Atlas.
Lamont-Doherty Earth Observatory and the National Science Foundation. [Available online at http://iridl.ldeo.columbia.edu/SOURCES/.LDEO/.TRL/.NADA2004/.pdsi-atlas.html.]
.
Cook
,
E. R.
,
S.
Shiyatov
, and
V.
Mazepa
,
1990
:
Estimation of the mean chronology.
Methods of Dendrochronology: Applications in the Environmental Sciences, E. R. Cook and L. A. Kairiukstis, Eds., Kluwer, 123–132
.
Cook
,
E. R.
,
D. M.
Meko
,
D. W.
Stahle
, and
M. K.
Cleaveland
,
1999
:
Drought reconstructions for the continental United States.
J. Climate
,
12
,
1145
1163
.
Cook
,
E. R.
, and
Coauthors
,
cited
.
2004a
:
North American summer PDSI reconstructions.
IGBP PAGES/World Data Center for Paleoclimatology Data Contribution Series 2004-045, NOAA/NGDC Paleoclimatology Program, Boulder, CO. [Available online at http://www.ncdc.noaa.qov/paleo/pdsi.html.]
.
Cook
,
E. R.
,
C.
Woodhouse
,
C. M.
Eakin
,
D. M.
Meko
, and
D. W.
Stahle
,
2004b
:
Long-term aridity changes in the western United States.
Science
,
306
,
1015
1018
.
Dai
,
A.
,
K. E.
Trenberth
, and
T.
Qian
,
2004
:
A global dataset of Palmer drought severity index for 1870–2002: Relationship with soil moisture and effects of surface warming.
J. Hydrometeor.
,
5
,
1117
1130
.
Dawdy
,
D. R.
, and
N. C.
Matalas
,
1964
:
Statistical and probability analysis of hydrologic data.
Part III: Analysis of variance, covariance and time series. Handbook of Applied Hydrology: A Compendium of Water-Resources Technology, V. T. Chow, Ed., McGraw-Hill, 8.68–8.90
.
Ecological Stratification Working Group
,
1996
:
A national ecological framework for Canada.
Agriculture and Agri-Food Canada and Environment Canada, Ottawa, ON, Canada, 125 pp
.
Esper
,
J.
,
D. C.
Frank
,
R. J. S.
Wilson
, and
K. R.
Briffa
,
2005
:
Effect of scaling and regression on reconstructed temperature amplitude for the past millennium.
Geophys. Res. Lett.
,
32
,
L07711
.
doi:10.1029/2004GL021236
.
Fritts
,
H. C.
,
1976
:
Tree Rings and Climate.
Academic Press, 567 pp
.
Gagen
,
M.
,
D.
McCarroll
, and
J-L.
Edouard
,
2004
:
Latewood width, maximum density and stable carbon isotope ratios of pine as climate indicators in a dry subalpine environment, French Alps.
Arct. Antarct. Alp. Res.
,
36
,
166
171
.
Gagen
,
M.
,
D.
McCarroll
, and
J-L.
Edouard
,
2006
:
Combining tree ring width, density and stable carbon isotope proxies to enhance the climate signal in tree-rings: An example from the southern French Alps.
Climatic Change
,
78
,
26
379
.
Girardin
,
M-P.
,
2007
:
Interannual to decadal changes in area burned in Canada from 1781 to 1982 and the relationship to Northern Hemispheric land temperatures.
Global Biol. Biogeogr.
,
10
,
557
566
.
Girardin
,
M-P.
, and
J.
Tardif
,
2005
:
Sensitivity of tree growth to the atmospheric vertical profile in the Boreal Plains of Manitoba, Canada.
Can. J. For. Res.
,
35
,
48
64
.
Girardin
,
M-P.
,
J. C.
Tardif
, and
M. D.
Flannigan
,
2006a
:
Temporal variability in area burned for the province of Ontario, Canada, during the past 200 years inferred from tree rings.
J. Geophys. Res.
,
111
,
D17108
.
doi:10.1029/2005JD006815
.
Girardin
,
M-P.
,
J. C.
Tardif
,
M. D.
Flannigan
, and
Y.
Bergeron
,
2006b
:
Synoptic-scale atmospheric circulation and boreal Canada summer drought variability of the past three centuries.
J. Climate
,
19
,
1922
1947
.
Herweijer
,
C.
,
R.
Seager
, and
E. R.
Cook
,
2006
:
North American droughts of the mid to late nineteenth century: A history, simulation and implication for mediaeval drought.
Holocene
,
16
,
159
171
.
Jones
,
P. D.
, and
M. E.
Mann
,
2004
:
Climate over past millennia.
Rev. Geophys.
,
42
,
RG2002
.
doi:10.1029/2003RG000143
.
Kim
,
K. W.
, and
G. R.
North
,
1993
:
EOF analysis of surface temperature field in a stochastic noise model.
J. Climate
,
6
,
1681
1690
.
Leavitt
,
S. W.
,
W. E.
Wright
, and
A.
Long
,
2002
:
Spatial expression of ENSO, drought, and summer monsoon in seasonal δ13C of ponderosa pine tree rings in southern Arizona and New Mexico.
J. Geophys. Res.
,
107
,
4349
.
doi:10.1029/2001JD001312
.
Liu
,
J.
,
R. E.
Steward
, and
K. K.
Szeto
,
2004
:
Moisture transport and other hydrometeorological features associated with the severe 2000/01 drought over the western and central Canadian prairies.
J. Climate
,
17
,
305
319
.
Mantua
,
N. J.
, and
S. R.
Hare
,
2002
:
The Pacific Decadal Oscillation.
J. Oceanogr.
,
58
,
35
44
.
Masson-Delmotte
,
V.
, and
Coauthors
,
2005
:
Changes in European precipitation seasonality and in drought frequencies revealed by a four-century-long tree-ring isotopic record from Brittany, western France.
Climate Dyn.
,
24
,
57
69
.
MathWorks
,
cited
.
2007
:
MATLAB Signal Processing Toolbox User’s Guide 6.
The MathWorks, Natick, MA. [Available online at http://www.mathworks.com/access/helpdesk/help/pdf_doc/signal/signal_tb.pdf.]
.
Matthews
,
J. A.
, and
K. R.
Briffa
,
2005
:
The ‘Little Ice Age’: Re-evaluation of an evolving concept.
Geogr. Ann.
,
87
,
17
36
.
Meko
,
D. M.
, and
C. A.
Baisan
,
2001
:
Pilot study of latewood-width of conifers as an indicator of variability of summer rainfall in the North American monsoon region.
Int. J. Climatol.
,
21
,
697
708
.
Meko
,
D. M.
,
C. W.
Stockton
, and
W. R.
Boggess
,
1995
:
The tree-ring record of severe sustained drought.
Water Resour. Bull.
,
31
,
789
801
.
Meko
,
D. M.
,
M. D.
Therrell
,
C. H.
Baisan
, and
M. K.
Hughes
,
2001
:
Sacramento River flow reconstructed to A.D. 869 from tree rings.
J. Amer. Water Resour. Assoc.
,
37
,
1029
1039
.
Meko
,
D. M.
,
C. M.
Woodhouse
,
C. A.
Baisan
,
T.
Knight
,
J. J.
Lukas
,
M. K.
Hughes
, and
M. W.
Salzer
,
2007
:
Medieval drought in the upper Colorado River basin.
Geophys. Res. Lett.
,
34
,
L10705
.
doi:10.1029/2007GL029988
.
Mitchell
,
T. D.
, and
P. D.
Jones
,
2005
:
An improved method of constructing a database of monthly climate observations and associated high-resolution grids.
Int. J. Climatol.
,
25
,
693
712
.
New
,
M.
,
M.
Hulme
, and
P.
Jones
,
2000
:
Representing twentieth-century space–time climate variability. Part II: Development of 1901–96 monthly grids of terrestrial surface climate.
J. Climate
,
13
,
2217
2238
.
Nielsen
,
E.
,
1998
:
Lake Winnipeg coastal submergence over the last three centuries.
J. Paleolimnol.
,
19
,
335
342
.
North
,
G. R.
,
T. L.
Bell
,
R. F.
Cahalan
, and
F. J.
Moeng
,
1982
:
Sampling errors in the estimation of empirical orthogonal functions.
Mon. Wea. Rev.
,
110
,
699
706
.
Osborn
,
T. J.
,
K. R.
Briffa
, and
P. D.
Jones
,
1997
:
Adjusting variance for sample-size in tree-ring chronologies and other regional mean time series.
Dendrochronologia
,
15
,
89
99
.
Overland
,
J. E.
, and
R. W.
Preisendorfer
,
1982
:
A significance test for principal components applied to a cyclone climatology.
Mon. Wea. Rev.
,
110
,
1
4
.
Pederson
,
G. T.
,
S. T.
Gray
,
D. B.
Fagre
, and
L. J.
Graumlich
,
2006
:
Long-duration drought variability and impacts on ecosystem services: A case study from Glacier National Park, Montana.
Earth Interactions
,
10
.
[Available online at http://EarthInteractions.org.]
.
Peixoto
,
J. P.
, and
A. H.
Oort
,
1992
:
Physics of Climate.
Springer-Verlag, 520 pp
.
Percival
,
D. B.
, and
W. L. B.
Constantine
,
2006
:
Exact simulation of Gaussian time series from nonparametric spectral estimates with application to bootstrapping.
Stat. Comput.
,
16
,
25
35
.
Raddatz
,
R. L.
,
2005
:
Moisture recycling on the Canadian prairies for summer droughts and pluvials from 1997 to 2003.
Agric. For. Meteor.
,
131
,
13
26
.
Rannie
,
W. F.
,
2006
:
A comparison of 1858–59 and 2000–01 drought patterns on the Canadian prairies.
Can. Water Resour. J.
,
31
,
263
274
.
Rigling
,
A.
,
P. O.
Waldner
,
T.
Forster
,
O. U.
Bräker
, and
A.
Poutuu
,
2001
:
Ecological interpretation of tree-ring width and intraannual density fluctuations in Pinus sylvestris on dry sites in the central Alps and Siberia.
Can. J. For. Res.
,
31
,
18
31
.
Sauchyn
,
D. J.
, and
A. B.
Beaudoin
,
1998
:
Recent environmental change in the southwestern Canadian prairies.
Can. Geogr.
,
42
,
337
353
.
Sauchyn
,
D. J.
, and
W. R.
Skinner
,
2001
:
A proxy PDSI record for the southwestern Canadian plains.
Can. Water Resour. J.
,
26
,
253
272
.
Sauchyn
,
D. J.
,
J.
Stroich
, and
A.
Beriault
,
2003
:
A paleoclimatic context for the drought of 1999–2001 in the northern Great Plains.
Geogr. J.
,
169
,
158
167
.
Schindler
,
D. W.
, and
W. F.
Donahue
,
2006
:
An impending water crisis in Canada’s western Prairie Provinces.
Proc. Natl. Acad. Sci.
,
193
,
7210
7216
.
doi:10.1073/pnas.0601568103
.
Shabbar
,
A.
, and
W.
Skinner
,
2004
:
Summer drought patterns in Canada and the relationship to global sea surface temperatures.
J. Climate
,
17
,
2866
2880
.
Shabbar
,
A.
,
B.
Bonsal
, and
M.
Khandekar
,
1997
:
Canadian precipitation patterns associated with the Southern Oscillation.
J. Climate
,
10
,
3016
3027
.
Spry
,
I. M.
,
1995
:
The Palliser Expedition: The Dramatic Story of Western Canadian Exploration, 1857–1960.
Fifth House Ltd., 315 pp
.
Stahle
,
D. W.
,
E. R.
Cook
,
M. K.
Cleaveland
,
M. D.
Therrell
,
D. M.
Meko
,
H. D.
Grissino-Mayer
,
E.
Watson
, and
B. H.
Luckman
,
2000
:
Tree-ring data document 16th century megadrought over North America.
Eos, Trans. Amer. Geophys. Union
,
81
,
121
.
St. George
,
S.
, and
E.
Nielsen
,
2002
:
Hydroclimatic change in southern Manitoba since AD 1409 inferred from tree rings.
Quat. Res.
,
58
,
103
111
.
St. George
,
S.
, and
E.
Nielsen
,
2003
:
Palaeoflood records for the Red River, Manitoba, Canada derived from anatomical tree-ring signatures.
Holocene
,
13
,
547
555
.
St. George
,
S.
,
D. M.
Meko
, and
M. E.
Evans
,
2008
:
Regional tree growth and inferred summer climate in the Winnipeg River basin, Canada since AD 1783.
Quat. Res.
,
70
,
158
172
.
doi:10.1016/j.yqres.2008.04.009
.
Stuiver
,
M.
, and
P. J.
Reimer
,
1993
:
Extended 14C data base and revised CALIB rev 3.0.3 14C age calibration program.
Radiocarbon
,
35
,
215
230
.
Trauth
,
M. H.
,
2006
:
MATLAB Recipes for Earth Sciences.
Berlin, 237 pp
.
Watson
,
E.
, and
B. H.
Luckman
,
2001
:
Dendroclimatic reconstruction of precipitation for sites in the southern Canadian Rockies.
Holocene
,
11
,
203
213
.
Watson
,
E.
, and
B. H.
Luckman
,
2002
:
The dendroclimatic signal in Douglas-fir and ponderosa pine tree-ring chronologies from the southern Canadian Cordillera.
Can. J. For. Res.
,
32
,
1858
1874
.
Watson
,
E.
, and
B. H.
Luckman
,
2004
:
Tree-ring based reconstruction of precipitation for the southern Canadian Cordillera.
Climatic Change
,
65
,
209
241
.
Watson
,
E.
, and
B. H.
Luckman
,
2005
:
Spatial patterns of preinstrumental moisture variability in the southern Canadian Cordillera.
J. Climate
,
18
,
2847
2863
.
Watson
,
E.
, and
B. H.
Luckman
,
2006
:
Long hydroclimate records from tree-rings in western Canada: Potential, problems and prospects.
Can. Water Resour. J.
,
31
,
205
228
.
Wigley
,
T. M. L.
,
P. D.
Jones
, and
K. R.
Briffa
,
1984
:
On the average value of correlated time series, with applications in dendroclimatology and hydrometeorology.
J. Climate Appl. Meteor.
,
23
,
201
213
.
Wolfe
,
S. A.
,
D. J.
Huntley
,
P. P.
David
,
J.
Ollerhead
,
D. J.
Sauchyn
, and
G. M.
MacDonald
,
2001
:
Late 18th century drought-induced sand dune activity, Great Sand Hills, Saskatchewan.
Can. J. Earth Sci.
,
38
,
105
117
.
Woodhouse
,
C. A.
,
S. T.
Gray
, and
D. M.
Meko
,
2006
:
Updated streamflow reconstructions for the upper Colorado River basin.
Water Resour. Res.
,
42
,
W05415
.
doi:10.1029/2005WR004455
.

APPENDIX A

Ring-Width Records from the Canadian Prairies

Table A1. Ring-width records from the Canadian Prairies used in this study. Tree species codes may be found online (http://web.utk.edu/~grissino/species.htm). Here, NA indicates information not available.

Table A1. Ring-width records from the Canadian Prairies used in this study. Tree species codes may be found online (http://web.utk.edu/~grissino/species.htm). Here, NA indicates information not available.
Table A1. Ring-width records from the Canadian Prairies used in this study. Tree species codes may be found online (http://web.utk.edu/~grissino/species.htm). Here, NA indicates information not available.

APPENDIX B

Ring-Width Chronology Statistics

Table B1. Signal strength results for prairie ring-width standard chronologies, as calculated after Wigley et al. (1984). The expressed population signal (EPS) measures the ability of each record to represent the ideal population signal, and the between-tree correlation (R_bar) is the mean correlation between all ring-width records within a site. SSS > 0.85 is the earliest year that the chronology is able to estimate at least 85% of its original signal, derived from all trees within the stand. The median series length provides the median number of annual rings contained by the tree-ring samples from an individual site.

Table B1. Signal strength results for prairie ring-width standard chronologies, as calculated after Wigley et al. (1984). The expressed population signal (EPS) measures the ability of each record to represent the ideal population signal, and the between-tree correlation (R_bar) is the mean correlation between all ring-width records within a site. SSS > 0.85 is the earliest year that the chronology is able to estimate at least 85% of its original signal, derived from all trees within the stand. The median series length provides the median number of annual rings contained by the tree-ring samples from an individual site.
Table B1. Signal strength results for prairie ring-width standard chronologies, as calculated after Wigley et al. (1984). The expressed population signal (EPS) measures the ability of each record to represent the ideal population signal, and the between-tree correlation (R_bar) is the mean correlation between all ring-width records within a site. SSS > 0.85 is the earliest year that the chronology is able to estimate at least 85% of its original signal, derived from all trees within the stand. The median series length provides the median number of annual rings contained by the tree-ring samples from an individual site.

APPENDIX C

Chronology Development and Calculating Regional Tree-Ring Series

Chronology development

Age- or size-related trends in ring-width data were removed by dividing each series by a spline with a 50% response wavelength equal to 70% of the length of each ring-width series (Cook et al. 1990). Because tests conducted using a much stiffer spline (a response wavelength equal to 200% of the series length) produced very similar results, we concluded that our analyses were largely insensitive to the choice of standardization approach. The same spline parameters were also used to remove any trend in variance. Detrended ring-width series were then combined to create the site chronology, with variance stabilization (Osborn et al. 1997) applied to adjust for changes in the number of trees over time.

The strength of the population signal in each chronology was assessed using the mean between-tree correlation and the EPS statistic (Wigley et al. 1984; appendix B). Nearly all sites have EPS values above 0.85 (used commonly as a minimum threshold for chronology quality). Although most sites with very high between-tree correlations (0.6 and above) are located in the eastern Rockies, there are no other apparent differences in the quality of the within-stand signal retained by ring-width chronologies in each of the five regions. Chronologies become increasingly noisy back in time, as fewer and fewer trees are available to contribute to the estimate of the population mean. Eventually, the chronology is derived from so few trees that the signal becomes obscured by an unacceptable amount of noise. The subsample signal strength (SSS; Wigley et al. 1984) is a measure of the chronology’s ability to retain its original signal (estimated from all trees within the stand) back through time. We truncated each chronology at the point where the SSS fell below 0.85 and excluded years prior to this cutoff from the final chronology.

Regional ring-width signals

The application of the EOF technique used correlation matrices derived from the regional ring-width networks, with no rotation applied (using covariance matrices or varimax rotation produced similar leading EOFs). Higher-order EOFs either could not be resolved from random white noise using the rule N significance test (Overland and Preisendorfer 1982) or exhibited the same seasonal moisture sensitivity as the leading EOF.

EOF analysis was conducted on ring-width chronologies from each of the five regions separately for the period 1900–2004. In some cases, the principal component time series (PC) associated with EOF1 is almost identical to a simple average of all chronologies in the region. Over the period 1900–2004, the first eigenvectors describe between 30% and 60% of the regional networks’ total variance (Fig. 2).

This initial analysis suggested that robust spatiotemporal patterns could be resolved from each of the five regional tree-ring networks. However, as with individual chronologies, the pool of available records across each region becomes smaller back in time as shorter chronologies become unavailable. Therefore, it is necessary to test whether a progressively shrinking network of chronologies can still provide an accurate estimate of the patterns described by the full network. We conducted a series of nested EOF analyses using a series of overlapping time intervals, each with the same end year but with different beginning years. In each run, chronologies were allowed to enter as inputs if they spanned more than 85% of the interval. Chronologies that did not span the entire interval were padded by replacing missing values with the global mean (Kim and North 1993). The quality of the signal recovered from the reduced network was assessed by comparing the PCs associated with the leading EOFs against those derived from the full network. If the reduced network produced PCs that were highly correlated (r > 0.9) with the PCs from the full network, we concluded that the reduced network was adequate to describe the behavior of that pattern back in time. The analysis was terminated once the PCs from the reduced network began to diverge from the original PCs. Essentially, this approach seeks to maximize the length of the regional PCs while retaining enough sites to define the spatiotemporal pattern accurately.

For the eastern Rockies, the leading pattern (EOF1) defined from the full network of 38 chronologies can be resolved using only the 14 longest chronologies. Because the pattern begins to degrade if fewer chronologies are used, we conclude that EOF1 can be resolved adequately back to 1528. Only seven chronologies are required to resolve EOF1 in northern Saskatchewan back to 1839. The leading eigenvector in the northwestern Ontario network can be described using only 12 chronologies and is stable back to 1783. Unlike the other regions, the first PC from central Manitoba changes radically as chronologies become unavailable back in time; extending the period of analysis by only 25 yr (beginning in 1875 instead of 1900) produces a PC that is only weakly correlated (r = 0.39) with the leading PC derived from the full network. This behavior is likely caused by having too few chronologies (10) in this region to define the initial pattern properly, weak correlations between the chronologies, and by having six chronologies (from different tree species) at the same geographic location.

The nested EOF approach is not appropriate for the set of chronologies from southern Manitoba. This set is made up of several short chronologies (most are less than 150 yr long), and a single long chronology (“Winnipeg composite”) developed from trees taken from historical buildings and alluvial deposits. Although this chronology extends up to 1999, the twentieth-century portion of its record is derived from very few trees. Because this chronology is weakest in the most recent portion of its record, it is not reasonable to judge the quality of its signal based on its ability to mirror regional ring-width variability during the twentieth century. As an alternative, we combined all ring-width data for southern Manitoba and constructed a single regional tree-ring record [this approach is similar to the strategy used by St. George and Nielsen (2002)]. Applying the EPS criterion to the regional series suggests that it remains a reasonable approximation of the unknown population signal back to 1523 a.d.

Footnotes

a

Supplemental information related to this paper is available at the Journals Online Web site: http://dx.doi.org/10.1175/2008JCLI2441.s1.

b

Earth Sciences Sector Contribution Number 20070491.

l

Retired.

Corresponding author address: Dr. Scott St. George, Geological Survey of Canada, 601 Booth St., Ottawa, ON K1A 0E8, Canada. Email: sstgeorg@nrcan.gc.ca

Supplemental Material