SEARCH

SEARCH BY CITATION

Keywords:

  • ENSO;
  • PDO;
  • climate variability;
  • groundwater levels

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Study Area, Data, and Methods
  5. 3. Results and Discussion
  6. 4. Conclusions
  7. Acknowledgments
  8. References
  9. Supporting Information

[1] Thirty one hydrological time series of shallow groundwater levels, precipitation, and moisture-sensitive tree ring chronologies were analyzed and related to two climate indices: Niño 3.4 and PDO. Spearman rank correlation and spectral analyses (multitaper method, continuous wavelet transform, and wavelet coherence) were used to document the influence of El Niño Southern Oscillation (ENSO) and the Pacific Decadal Oscillation (PDO) on shallow (depth < 20 m) groundwater level records from the Canadian Prairies. Modes of variability in the 2–7, 7–10, and 18–22 year bands were detected and reconstructed. Correlations and wavelet coherence between these oscillation modes and the climate indices suggest that variability in the 2–7 and 7–10 year bands is highly influenced by ENSO. The oscillation modes in the 18–22 year band reflect a negative correlation with the PDO index. When either of these teleconnections (ENSO/PDO) is in their respective positive phases, groundwater levels reflect the effect of associated warmer and drier winters experienced over much of interior Canada and the US, affecting important resource inputs to the hydrological cycle and groundwater recharge.

1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Study Area, Data, and Methods
  5. 3. Results and Discussion
  6. 4. Conclusions
  7. Acknowledgments
  8. References
  9. Supporting Information

[2] Global climate variability at interannual to interdecadal scales has been linked to oceanic and atmospheric teleconnection patterns such as the El Niño Southern Oscillation (ENSO) and the Pacific Decadal Oscillation (PDO) [Rubio-Alvarez and McPhee, 2010; Wang et al., 2008; among others]. In North America, these large-scale climate oscillations are major drivers of hydroclimate variability, and therefore the understanding of how these teleconnection patterns affect hydroclimatic variables has been a major research focus for the past two decades [Hurrell and van Loon, 1997; Bonsal et al., 2001; Enfield et al., 2001; McCabe et al., 2004; and others].

[3] Some coupled sea-surface temperature (SST) and atmospheric teleconnection anomaly patterns recur on a regular basis. ENSO and the PDO [Rasmusson and Carpenter, 1982; Shabar et al., 1997; Mantua et al., 1997; Minobe, 1997; Bonsal et al., 2001; Coulibaly and Burn, 2004; MacDonald and Case, 2005; Fleming and Quilty, 2006; Gan et al., 2007; Gobena and Gan, 2006, 2009] are two of the more important coupled SST and atmospheric teleconnection phenomena that affect climate in the Northern Hemisphere [Speer, 2010]. ENSO is a cyclical 2–7 year coupled ocean-atmosphere oscillation in the tropical Pacific Ocean, consisting of a warm (El Niño) phase, during which SSTs in the tropical eastern Pacific are anomalously warm, and a cool (La Niña) phase associated with cool SSTs in the eastern Pacific [Case, 2000; Rasmusson and Carpenter, 1982]. Prairie precipitation is less affected by ENSO than temperature; however, winter precipitation during El Niño (La Niña) events is significantly lower (higher) than normal [Shabbar et al., 1997]. The PDO is a large scale pattern of SST variability characterized by a gradient between anomalously cool SSTs in the east-central North Pacific and anomalously warm SSTs off the central West Coast of North America recurring every 15–25 and 50–70 years [Case, 2000; Mantua et al., 1997; Minobe, 1997]. North Pacific SSTs affect precipitation amounts on the Prairies through links with 500 mb height anomalies in the midtroposphere. During a positive phase of the PDO, this teleconnection occurs more frequently, resulting in long-term persistence of the negative precipitation anomaly [Knox and Lawford, 1990]. The PDO has positive (warm) and negative (cool) phases that appear to persist for two–three decades. Positive PDO (warm) phases occurred between 1925–1946 and 1977–1998. These periods have been associated with warm winter and spring time temperatures in northwest North America, and dry conditions in the west to central North America [Fagre et al., 2003]. Negative (cool) PDO phases occurred between 1900–1924 and 1947–1976, bringing with them cool winter and spring temperatures over western North America, as well as wet conditions in the Pacific Northwest, northern Rockies, and central Prairies.

[4] Climatic fluctuations in precipitation and temperature, and their effect on amounts of evapotranspiration and snow cover, are key determinants in linkages of large-scale climatic patterns to groundwater recharge [Tremblay et al., 2011]. Precipitation and temperature drive infiltration are indirect components of the groundwater reservoir budget, thereby suggesting that the influence of climatic patterns on aquifer reservoirs should be visible through groundwater levels. Subsurface hydrologic response to natural climate variability is of particular interest in semiarid and arid regions, where groundwater resource availability and sustainability are important [Hanson et al., 2004]. Deep confined aquifers in the southern prairies present a response to mechanical loads that could be caused by moisture conditions and snow [Van der Kamp and Maathuis, 1991], however, unconfined shallow aquifers might present a different response to local mechanical loads but perhaps they could also respond to atmospheric pressure that affect large areas and are associated with atmospheric circulation patterns such as PDO. Knowledge of subsurface hydrodynamic responses to climate variability in these regions is limited because of a general lack of field observations over time scales long enough to document infiltration and percolation of infrequent recharge events [Gurdak et al., 2007]. Despite the importance of groundwater resources, few studies have documented natural variability in Canadian groundwater systems. Fleming and Quilty's [2006] study was the first in Canada and one of the first worldwide to directly quantify the effects of organized modes of climatic variability upon groundwater resources, by examining the influence of ENSO on water levels in shallow aquifers in British Columbia. They showed that during La Niña years water levels are above average and below average during El Niño years reflecting variability in winter and spring precipitation that recharges the aquifer systems; a finding consistent with, but also having differences from, those in comparable surface water resources; highlighting that groundwater responses are something distinct from river responses and require direct evaluation. Perez-Valdivia and Sauchyn [2011]found strong correlations between groundwater levels in southwestern Canada and moisture-sensitive tree ring records and were able to reconstruct historical groundwater levels for the past 300 years at two wells in southern and central Alberta, Canada. They analyzed the natural variability in groundwater systems by spectral analyses; main findings of their study suggest the existence of oscillation modes in groundwater levels at 2–8 and 8–16 year frequencies. A most recent study byTremblay et al. [2011]analyzed the variability of groundwater systems for three different regions across Canada and their linkage to climate indices such as the North Atlantic Oscillation (NAO), the Arctic Oscillation (AO), the Pacific-North American pattern (PNA), and El Niño Southern Oscillation (ENSO). The regions represented by that study are Prince Edward Island, southern Manitoba, and Vancouver Island, and only one groundwater time series per region was analyzed. Results of their study suggest that groundwater in different regions has different patterns of variability. They found that groundwater variability in the Prince Edward Island region is mostly influenced by the NAO and AO, groundwater variability in Manitoba is influenced by the PNA, while groundwater variability for Vancouver Island is influenced by NAO, AO, and ENSO.

[5] Groundwater is the source of domestic fresh water for almost nine million Canadians (30% of the population); with 67% or six million living in rural areas. It is estimated that 23%, 43%, and 30% of the population in Alberta, Saskatchewan, and Manitoba, respectively, relies on groundwater [Statistic Canada, 1996]. Global groundwater withdrawals have increased by 10% since 1950, being close to 1000 km3 per year [Shah et al., 2007] and are expected to continue to increase in response to economic and population growth. Increments in groundwater demand can reflect climatic conditions such as droughts which rapidly limit surficial water resources.

[6] Over the historical period of instrumental records, repeated droughts have highlighted the Canadian Prairies' vulnerability to even small deviations from normal moisture conditions. Widespread drought throughout the region has disrupted farming and caused economic hardship several times in the last 100 years, most notably during the 1930s, 1980s, and 1999–2004 [Bonsal et al., 1999]. The “dustbowl” of the 1930s is one of the worst drought periods in memory in western North America, and caused much hardship both environmentally and economically. During 1961 (the worst single year drought on the Prairies, with approximately 50% of normal growing season precipitation), total net farm income dropped by 48% ($300 million) from the previous year [Bonsal et al., 1999]. The drought of 1988 had many impacts on the agricultural sectors of Canada (with emphasis on Manitoba and Saskatchewan), including wind erosion, production, grain quality, inventories, marketing, livestock, incomes, farm management, global production, and prices [Wheaton et al., 1992]. The sparse snow cover and high spring temperatures resulted in little or no spring runoff from prairie watersheds in 1988, such that the mean volume was 60% to 70% of normal volume. The lack of precipitation caused an agricultural output decrease of 12%, resulting in a direct production loss of $1.8 billion in 1981 [Wheaton et al., 1992]. The drought of 2001 resulted in an economic shortfall of $4 billion or more and required supportive measures from various governments. Grain yield data reveal a similar shortfall in total yields during 1961 and 1977, both also being severe drought years [Khandekar, 2004].

[7] With the prediction that droughts are likely to become more frequent under global warming [Kharin and Zwiers, 2000], understanding the natural variability in groundwater levels and its drivers is fundamental in order to manage water resources under extreme climatic conditions (including droughts). Thus this study aims to identify the influence of major teleconnection patterns, the El Niño Southern Oscillation and Pacific Decadal Oscillation, on shallow aquifers through the Canadian Prairies. Despite that a similar study [Tremblay et al., 2011] has partially investigated this question with coincident tools; our study is the first to investigate the influence of these climate drivers on the variability of groundwater levels over such a large area and the number of wells in the Canadian Prairies.

2. Study Area, Data, and Methods

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Study Area, Data, and Methods
  5. 3. Results and Discussion
  6. 4. Conclusions
  7. Acknowledgments
  8. References
  9. Supporting Information

2.1. Study Area

[8] The Canadian Prairie Provinces of Alberta, Saskatchewan, and Manitoba (Figure 1) extend from the eastern slopes of the Rocky Mountains to the Hudson Bay. The climate varies from cold winters to subhumid summers. In Calgary, Regina, and Winnipeg the mean winter temperatures (1971–2000) are −7.5°C, −13.8°C, and −15.6°C, respectively. Mean summer temperatures reach over 15°C, increasing to the east (15.2°C; 17.7°C; 17.9°C), with maximum means of 17.3°C in Calgary, 20.3°C in Regina, and 20.6°C in Winnipeg. On the contrary, mean annual temperatures increase toward the southwest (southern Alberta) where it is 5.0°C to 6.0°C, while Calgary, Regina, and Winnipeg exhibit mean annual temperatures of 4.0°C, 2.8°C and 2.2°C, respectively.

image

Figure 1. Study area map giving the locations of the groundwater observation wells and tree ring sites, and cities with long precipitation records.

Download figure to PowerPoint

[9] The annual precipitation in the Prairies varies from just below 300 mm in southeastern Alberta and southwestern Saskatchewan to over 900 mm in the adjacent Rocky Mountains. Mean annual precipitation for the period 1971–2000 at Calgary, Regina, and Winnipeg was 457, 444, and 574 mm, respectively. Spring and summer register over 60% of the annual precipitation; however, much is lost to evapotranspiration, mostly from the unsaturated soil water zone where it is available for annual plant growth [Perez-Valdivia and Sauchyn, 2011]. Winter precipitation (snow) is an important source of water to recharge surface and groundwater systems. Approximately 25% of the annual precipitation occurs as snow fall during the months of November to March where mean temperatures are below zero. Given the distribution of precipitation and temperature patterns throughout the Prairies, south central areas are generally under moisture stressed conditions and rely on water supplies received from the surrounding Rocky Mountains and foothills.

2.2. Data

2.2.1. Groundwater, Precipitation, and Climate Indices

[10] Twenty one monthly groundwater level time series were obtained from Alberta Environment, Saskatchewan Watershed Authority, and Manitoba Conservation. These records are part of a larger monitoring network that extends throughout the Prairie Provinces. The selection of the 21 wells was based on length and continuity of the records (at least 20 years), percentage of missing data (less than 20%), anthropogenic effects (no pumping), and depth of the aquifer (shallow <20 m depth). Missing values were estimated using correlations with nearby water level records and median values were used when correlation was not possible. The continuous time series were plotted to check that the process of replacing missing data did not create outliers. The selected groundwater wells respond mostly to spring snowmelt and the spring and summer precipitation period in which maximum levels are observed. Table 1 summarizes the groundwater records used in this study.

Table 1. Groundwater Wellsa
Well NameWell IDProv.Depth of Water Table (m)LatitudeLongitudeElevation (m)PeriodAquifer TypeLithology
  • a

    Bedrock aquifers are typically composed of sandstone and in some other cases by fractured shale or coal. Surficial aquifers are composed of sand, gravel, silt, and clay. Intertill aquifers have glacial gravels, sands and silts between layers of till.

AttonAttoSK16.252.816−108.869536.4Feb-66Dec-08SurficialSand
Bangor BBan BSK15.350.899−102.287527.6Nov-70Oct-08IntertillSand
BeauvalBeauSK16.255.117−107.745434.3Nov-74Apr-09IntertillSand
Conq 500C500SK19.251.574−107.174555.2Dec-71Dec-08IntertillSand
Conq 501C501SK8.251.579−107.315572.6Feb-72Apr-09SurficialSand/silt
Conq 502C502SK19.251.570−107.315572.0Jan-72Apr-09IntertillSand/gravel
ForgerForgSK5.949.705−102.863606.5Apr-66Dec-08SurficialSand
MelfortMelfSK10.652.950−104.448451.1Nov-67Apr-09SurficialSand/silt
Winkler #5Ob005MB6.750.875−97.855278.7Jun-69Jun-09SurficialSand/gravel
M0-5Og003MB9.149.768−97.300236.9May-67Apr-09Carbonate-rockLimestone Dolomite
RicetonRiceSK22.450.164−104.317579.1Jul-68Dec-07IntertillSand
Simpson 13-04SI13SK7.251.457−105.193496.6Jul-68Dec-08SurficialSand
StenenStenSK14.651.821−102.410naSep-65Dec-08SuficialGavel/sand
SwansonSwanSK9.251.650−107.066534.9Jan-72Dec-08SurficialSand
VerloVerlSK12.850.373−108.897737.6Feb-65Apr-09SurficialClay/silt
Waterton DamW105AB12.249.320−113.6341176.0May-65Oct-93SurficialSand
Elkwater 2294EW108AB33.549.661−110.2881220.0Sep-85Jul-05SurficialSand/gravel
Barons 615EW117AB19.849.993−113.077964.5Jul-71Dec-07BedrockSandstone
ElnoraW129AB13.751.967−113.552894.3Jul-62Jul-08SurficialClay
Devon #2W159AB7.653.388−113.691693.3Apr-67Jul-05SurficialSand
Duvernay 2489EW270AB20.753.773−111.700580.0Jun-88Jul-07BedrockSandstone

[11] Seven time series of monthly precipitation, from stations relatively close to the groundwater wells (Figure 1, Table 2), were obtained from the Adjusted Historical Canadian Climate Database [Mekis and Hogg, 1999]. These data have been homogenized and adjusted and are suitable for studies of climate variability and change. Monthly climate indices of ENSO (Niño 3.4), and PDO were obtained from the Earth Systems Research Laboratory (National Oceanic and Atmospheric Administration, available at http://www.esrl.noaa.gov/psd/data/climateindices).

Table 2. Weather Stations and Tree Ring Chronologies
NameLatitudeLongitudeElevation (m)Period
Weather Stations
Estevan49.070−103.0005681945–2004
Lethbridge49.700−112.7808941909–1986
Calgary51.117−114.01710711895–2005
Edmonton53.570−113.5206681929–2004
Regina50.433−104.6675741904–2003
Saskatoon52.233−106.6174911906–2004
Winnipeg49.900−97.2332371895–2005
 
Tree Ring Chronologies
Devon Farm (DEV)50.470−101.8204401862–2005
Oldman River (OMR)49.900−114.20014271203–2007
Whirlpool Point (WPP)52.000−116.45013561062–2007
2.2.2. Tree Ring Chronologies

[12] As climate records often predate groundwater monitoring by several decades, these data can potentially be used to estimate changes within an aquifer prior to the establishment of direct measurement [Ferguson and St. George, 2003]. Although tree rings have most often been used to reconstruct precipitation and streamflow and to study the low-frequency variability in these hydroclimate variables [Watson and Luckman, 2001; Axelson et al., 2009], the direct relationship between tree growth and the soil water that recharges groundwater suggests that dendrochronological data may also be useful in historical groundwater studies [Ferguson and St. George, 2003]. Albeit many reconstructions of precipitation and streamflow exist, only a few have successfully investigated the relationship between tree rings and groundwater within the Canadian Prairies [Ferguson and St. George, 2003; Perez-Valdivia and Sauchyn, 2011], recognizing the common response of tree growth and groundwater levels to effective precipitation, and the often lagged response times [Perez-Valdivia and Sauchyn, 2011].

[13] In this study, tree ring indices are used as a surrogate for long-term instrumental records [Hanson et al., 2004] in order to identify relationships at significant periodicities (found common among groundwater, precipitation, and tree ring records), such as the low frequency of the PDO which exceeds the length of all but a few of the instrumental hydroclimatic records from the Canadian Prairies. Three tree ring chronologies, from Pinus flexilis (Limber pine; Oldman River and Whirpool Point) in the foothills of central and southern Alberta, and Quercus macrocarpa(Burr Oak; Davon Fram) in southeastern Saskatchewan were analyzed to identify coherent low-frequency climate signals across the Prairie Provinces (Table 2). These tree ring chronologies contain annual moisture signals spanning more than 150 years in Saskatchewan and more than 900 years in Alberta. The wood and tree ring data were processed in the Tree ring Laboratory at the Prairie Adaptation Research Collaborative, University of Regina, using standard dendrochronological methods [Stokes and Smiley, 1968; Fritts, 1976; Cook, 1985; Cook and Kairiukstis, 1990] and the semiautomated tree ring analysis software WinDendro Density (version 2009a). Tree growth trends were removed using a negative exponential curve or a cubic spline with 67% of the length of the time series. This detrending is considered conservative because it preserves most of the low-frequency signals recorded in the tree rings. Crossdating, which assigns the correct calendar year to individual tree rings, was verified with COFECHA [Holmes, 1983]. The lengths of the chronologies are 946, 805, and 152 years for Whirlpool Point (WPP), Old Man River (OMR), and Devon Farm (DEV), respectively.

2.3. Methods

[14] The methodology used in this study is similar to that proposed by Hanson et al. [2004], Gurdak et al. [2007] and Tremblay et al. [2011]to assess the relationship between groundwater time series and climatic variability. It consists of detrending hydrological time series by fitting a low-order polynomial. This detrending removes part of the low frequency (signals that are longer than half of the period of record) and anthropogenic effects. The groundwater wells are not directly affected by local pumping; their implementation was only to monitor groundwater levels in different aquifers. However, these groundwater records might be influenced by pumping through other wells in the same aquifer. The residual time series, the difference between the observed data and the values calculated with the polynomial, are standardized to a zero mean and unit variance for intercomparison. Spearman's rank (nonparametric) correlation was used to calculate lagged correlations for up to 36 months between monthly groundwater levels, precipitation, temperature, and indices of Niño 3.4 and PDO. The significance of the correlation coefficients was tested under the hypothesis that our time series are distributed as the Students'st distribution with n − 2 degrees of freedom. A lag of up to 36 months was considered to be enough for groundwater time series to reflect the influence of teleconnection patters and precipitation because all the wells in this study are shallow and located in material of high hydraulic conductivity. The multitaper method [MTM; Ghil et al., 2002] of spectral analysis was used to identify dominant oscillation modes in the groundwater (residuals), precipitation, and residual tree ring time series and to extract and reconstruct significant quasiperiodic signals. MTM is nonparametric and uses a series of tapers that reduce the variance of spectral estimates which is an advantage over other spectral window methods [Percival and Walden, 1993]. We applied the SSA-MTM toolkit [Ghil et al., 2002], with robust background estimation, to the residual time series. The bandwidth parameter p was set at two and the numbers of tapers was set at three, which has been suggested for the analysis of climate records consisting of a few hundred observations [Mann and Park, 1993].

[15] Continuous wavelet transform [CWT; Torrence and Compo, 1998; Grinsted et al., 2004] was also used to identify the main oscillatory modes of variability in groundwater levels and the tree ring chronologies. The CWT complements the MTM method by identifying periods of dominant oscillation modes, with the added feature of the time-frequency domain, which depicts when these dominant frequencies occur throughout the time series [Torrence and Compo, 1998; Jevrejeva et al., 2003]. Among all the wavelet families, Morlet wavelet (w0 = 6) was used since it provides a good balance between time and frequency domains and is recommended when the purpose is to extract dominant signals [Grinsted et al., 2004]. The statistical significance was assessed against a red noise background at a 95% confidence level. The CWT function creates a cone of influence (COI) that delimits a region of the wavelet spectrum beyond which edge effects become important and the power could be suppressed [Torrence and Compo, 1998; Jevrejeva et al., 2003; Grinsted et al., 2004; Gobena and Gan, 2009]. The spectral power outside the cone of influence should be interpreted with caution.

[16] Significant oscillation modes detected using MTM and wavelet analyses were reconstructed in the time domain, using information from the multitaper decomposition via the SSA-MTM toolkit [Ghil et al., 2002], and correlated with the respective frequencies of known oscillation modes of atmospheric and sea-surface temperature (SST) forcings, ENSO and PDO indices, thought to be primary drivers of hydroclimatic variability affecting groundwater recharge. Areas in the time-frequency plane where two time series exhibit common power (coherence) or consistent phase behavior indicate a relationship between the signals. Wavelet transform coherence (WTC) analysis greatly facilitates the detection of the quasiperiodic component indicative of a system anomaly [Grinsted et al., 2004]. Following Torrence and Webster [1999] coherence is defined as

  • display math

where S is a smoothing operator. inline image gives a quantity between 0 and 1, and measures the cross correlation between two time series as a function of frequency. The equation above represents the normalized covariance between two time series because the wavelet transform conserves variance [Gobena and Gan, 2009]. The statistical significance level (p < 0.05) of the wavelet coherence is estimated using Monte Carlo methods with a red noise background resulting in significant periodicities of coherence delineated by significance contours. As in the CWT, regions outside of the COI should be interpreted with caution.

3. Results and Discussion

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Study Area, Data, and Methods
  5. 3. Results and Discussion
  6. 4. Conclusions
  7. Acknowledgments
  8. References
  9. Supporting Information

3.1. Correlations

[17] The results of correlation analyses (Table 3) suggest that most of the residual groundwater time series have significant correlations with precipitation at lags of 9 to 24 months. Significant correlations suggest that these shallow unconfined aquifers are recharged primarily by precipitation that infiltrates the soil.

Table 3. Correlation Coefficients for Monthly Standardized Groundwater Levels and Monthly Standardized Climate Indices, Precipitation, and Mean Temperaturea
WellENSOPDOPrecipitationTemperature
  • a

    Bold coefficients are significant at α = 0.01 and the subscript number denotes the lag in months at which maximum negative or positive correlation was reached.

Atto(n=515)−0.2417−0.20170.1012−0.055
Banb(n=456)−0.2015−0.10190.0712−0.027
Beau(n=366)−0.2315−0.19110.1914−0.487
C500(n=455)−0.15260.05360.05260.008
C501(n=415)−0.0814−0.08210.2214−0.258
C502(n=510)−0.0812−0.04190.2612−0.306
Forg(n=513)−0.2015−0.15190.2712−0.296
Ob005(n=475)−0.2514−0.20140.012−0.106
Og003(n=453)−0.089−0.19170.329−0.373
Melf(n=440)−0.203−0.14360.0812−0.037
Rice(n=478)−0.0820−0.26340.080−0.049
Si13(n=486)−0.182−0.13360.1812−0.156
Sten(n=520)−0.2813−0.1790.1924−0.227
Swan(n=444)−0.3612−0.10180.2211−0.244
Verl(n=504)−0.1515−0.2260.1311−0.0916
W105(n=322)−0.0925−0.02190.4113−0.626
W108(n=239)−0.0436−0.32330.2212−0.346
W117(n=445)+0.0434−0.08360.1812−0.286
W129(n=441)−0.037−0.1280.4312−0.506
W159(n=476)−0.166−0.11200.3511−0.476
W270(n=230)−0.4211+0.0870.2211−0.245

[18] Correlations between groundwater levels and mean temperature indicate that groundwater records are significantly correlated (p < 0.05) with temperature over a five month period (lags 3 to 8; Table 3). As expected, temperature correlates negatively with most groundwater time series, suggesting that increased temperatures, primarily during the warmer months of May to October, causes increased amounts of evapotranspiration, resulting in decreased recharge and groundwater levels.

[19] Wells Atto, Banb, C500, Ob005, Melf, and Rice did not show significant correlations at any lag, for either precipitation or temperature. These results indicate that although these six wells are in sand; they may be influenced more by nearby streamflow recharge, or by beds of lower hydraulic conductivity or a degree of confinement, than by precipitation as their primary sources of recharge. Further research should be focused on determining the recharge mechanisms for these aquifers and how local hydrogeologic environments may modulate groundwater responses to ENSO and PDO.

[20] Sixteen of the 21 groundwater records show significant negative correlations with either or both the Niño 3.4 and PDO indices, suggesting that large scale SST and atmospheric teleconnections may be drivers influencing fluctuations in groundwater levels. Most of these correlations occur around lags of 13–20 months; however, correlations with PDO occur at slightly longer lags than for ENSO. The most significant correlations were for wells Sten (−0.28 lag 13) and W108 (−0.32 lag 33) with ENSO and PDO, respectively.

[21] Few wells (Si13, Sten, and W159) with significant correlations to climate indices, exhibit shorter lag times with ENSO and PDO than with local precipitation. These findings suggest that precipitation may not be the main source of groundwater recharge for these wells; rather, relationships with remotely generated or recharged, and ENSO/PDO driven streamflow or subsurface flows may exist. Future analyses are suggested and required to investigate this further.

3.2. Spectral Analyses

3.2.1. Multitaper Method (MTM)

[22] Results of the MTM suggest that there are significant modes of high-frequency, interdecadal, and low-frequency variability in groundwater levels in the Canadian Prairies at frequencies of 2–7 years, 7–10 years, and 18–22 years (Table 4). Similar findings were reported by Perez-Valdivia and Sauchyn [2011] identifying coincident periodicities of high and interdecadal frequencies in their tree ring reconstructions of groundwater levels for two wells in Alberta, Canada. MTM applications to annual precipitation records show similar results, highlighting significant oscillations also found within the 2–7 and 7–12 year frequencies (Table 5). MTM analysis of the moisture-sensitive tree ring chronologies show that they register oscillation modes in the 2–7, 9–15, 22–25, 64, and 85 year frequency ranges (Table 5; Figures 2, 3, and 4). The interannual, interdecadal, and multidecadal variability found within the tree rings provides evidence that both the tree ring chronologies and groundwater wells may be responding to the same climatic drivers affecting available precipitation resources for both tree growth and aquifer recharge.

image

Figure 2. Multitaper (left) and wavelet analysis (right) for (a) the tree ring chronology WPP and (b) the residual time series for well W159. Both records are from the upper North Saskatchewan River Basin.

Download figure to PowerPoint

image

Figure 3. Multitaper (left) and wavelet analysis (right) for (a) the tree ring chronology OMR and (b) the residual time series for well W117. Both records are from the upper South Saskatchewan River Basin.

Download figure to PowerPoint

image

Figure 4. Multitaper (left) and wavelet analysis (right) for (a) the tree ring chronology DEV and (b) the residual time series for well Si13. Both records are from the Assiniboine River Basin.

Download figure to PowerPoint

Table 4. MTM and Wavelets Analyzes Summarized for Each Wella
Well NameWell IDMTMWavelets Periodicities
32–64 Months (2.6–5.3 yr)64–128 Months (5.3–10.6 yr)>128 Months (>10.6 yr)
  • a

    Note there was no spectral analyzes for W129 because the record had more than three consecutive years with missing data. hp: high power detected but not significant at p < 0.05; ooci: significant periodicity but out of the cone of influence.

AttonAtto2.3, 4.5, 9.5, 21.4yeshp
Bangor BBan B2.4, 5.3, 19yesyesyes
BeauvalBeau7.8, 17.2hphp
Conq 500C5002.0, 3.2, 5.0, 18yesooci
Conq 501C5012.9, 7.8, 18.6hphp
Conq 502C5022.9, 7.8, 18.7hphp
ForgerForg3.4, 5.7, 21.4
MelfortMelf4.1, 8.5, 20.8yesyesyes
Winkler #5Ob0056.5, 20.0
M0-5Og00320.9hphphp
RicetonRice2.7, 7.1, 19.8yesooci
Simpson 13-04SI131.7, 2.1, 3.6, 20.5yesyesyes
StenenSten1.8, 2.2, 4.3, 21.9hpyes
SwansonSwan2.1, 3.2, 7.8yesyeshp
VerloVerl4.1, 8.5, 21.2hpyeshp
Waterton DamW105
Elkwater 2294EW10810yeshp
Barons 615EW1173.4, 18.3yesyes
Devon #2 (North)W1592.9, 19
Duvernay 2489EW2709.6yes
Table 5. Multitaper and Wavelet Analyses Summarized for Precipitation and Tree Ring Chronologiesa
Well NameMTM (yrs)Wavelets Periodicities
32–64 Months (2.6–5.3 yr)64–128 Months (5.3–10.6 yr)128–256 Months (10.6–21.3 yr)
  • a

    hp: high power detected but not significant at p < 0.05; ooci: significant periodicity but out of the cone of influence.

Edmonton2.3, 9.0hpyeshp
Calgary3.1, 4.0, 6.1, 12.2yeshpyes
Lethbridge12.2yeshpyes
Saskatoon3.8yesoocihp
Regina2.3, 4.6, 10.7yesyesyes
Estevan4.5yesyesyes
Winnipeg3.0yesyeshp
Tree Ring Chronologies    
Oldman River2.5, 2.9, 7.3, 14, 24, 64yesyesyes
Whirlpool Pool2.7, 4.7, 6.2, 9.2, 85yesyesyes
Devon Farm2.4, 22.7yesyesyes
3.2.2. Continuous Wavelet Transform (CWT)

[23] Wavelet results (Table 4) mirror those of the MTM analysis, suggesting natural variability, at high and interdecadal frequencies in groundwater level time series. Lower frequency activity, greater than 20 years, is not identified in the groundwater levels by wavelet analysis, likely given the short length of the time series. Significant interannual variability coinciding with 2–6 year frequencies, usually related to ENSO [Shabbar and Skinner, 2004; Gobena and Gan, 2006], is evident throughout much of the entire lengths of the groundwater time series, however, it is concentrated most heavily during the late 1960s–early 1970s, mid 1980s, early 1990s, and 2000s (Figures 2, 3, and 4). Interdecadal oscillations (coinciding with frequencies at 7–12 years) are also depicted in the groundwater time series, most notably during the 1980s and 1990s.

[24] Precipitation records yield similar results to the MTM analysis, with significant oscillations within the 2–6 and 7–10 year bands, but also identify the presence of low-frequency oscillations within the 20–30 year band (Table 5). The high and low frequencies detected previously via MTM analysis are also identified by wavelet analysis for the tree ring chronologies (Table 5; Figures 2, 3, and 4), and coincide to periods of occurrence found within both groundwater and precipitation time series, but also provide the advantage of long record length and thus the identification of low-frequency signals, in this case in the bands of 16–32, 32–64, and 64–128 years. In Alberta, the low frequency in the WPP and OMR chronologies (Figures 2 and 3) concurs with the oscillation modes detected in precipitation records at Calgary and Lethbridge, and also reveals a significant periodicity of lower frequency variability around the 64-year band, documented previously byAxelson et al. [2009]. The DEV chronology in southeastern Saskatchewan (Figure 4) shows significant activity at high frequency (2–7 years) and also at periodicities coinciding with the ∼20 year oscillation mode detected in groundwater levels and precipitation at Calgary, Lethbridge, Regina, and Estevan (Table 5, Figure 4). Frequencies within the 20–30 year bands have been associated with ocean-atmosphere teleconnection forcings, such as the PDO [Mantua et al., 1997; Minobe, 1997], where shifts in oscillation patterns, “warm” or “cool” phases, are marked by widespread variations in the Pacific Basin and North American climate. PDO-related temperature and precipitation patterns have been strongly linked to regional hydrologic variable (i.e., snow pack and streamflow) anomalies, especially in western North America [Cayan, 1996; Manuta et al., 1997; Bitz and Battisti, 1999; Nigam et al., 1999]. The commonality in results of both MTM and wavelet spectral analyses, among groundwater, precipitation, and tree ring time series, suggests that all three are responding to a common coherent signal, being driven by the same mechanism or forcings (i.e., large scale atmospheric teleconnections).

3.3. Reconstruction of Oscillation Modes

[25] To further associate plausible drivers of groundwater level variability, modes at significant frequencies were extracted from groundwater time series and correlated with oscillations associated with sea-surface temperature and atmospheric forcings, specifically ENSO and PDO (Table 6). The high-frequency reconstructed mode (2–7 year) has a negative correlation with ENSO (Table 6), but it also seems to be related to the high-frequency component of the PDO, causing much of the interannual variability found within the groundwater level time series. Correlation coefficients range from −0.02 to −0.33, however, stronger correlations are reached at lags longer than 10 months. The high-frequency variability corresponding to ENSO was previously identified in other analyses of prairie hydrologic time series [e.g.,Coulibaly and Burn, 2004; Gan et al., 2007].

Table 6. Correlation Coefficients Between Reconstructed Oscillation Modes and ENSO and PDO Indicesa
Well2–7 yr7–10 yr∼20 yr
ENSOPDOENSOPDOPDO
  • a

    The subscript number denotes the lag at which maximum negative correlation was reached. These correlations have to be interpreted with caution because their significance has not been evaluated considering the reduction of the degrees of freedom through the signal reconstruction.

Atto(n=515)−0.2918−0.1415−0.307−0.100−0.2429
Banb(n=456)−0.3318−0.1918−0.3317
Beau(n=366)−0.110−0.140−0.1719
C500(n=455)−0.1726−0.1923−0.0534
C501(n=415)−0.1312−0.1319−0.0919−0.0536−0.051
C502(n=510)−0.169−0.1612−0.170−0.0400.011
Forg(n=513)−0.1015−0.058−0.429
Ob005(n=475)−0.1729
Og003(n=453)−0.371
Melf(n=440)−0.3230−0.2026−0.282−0.070−0.1228
Rice(n=478)−0.0216−0.0421−0.0420.0035−0.3736
Si13(n=486)−0.176−0.1515−0.340
Sten(n=520)−0.2416−0.2114−0.4114
Swan(n=444)−0.127−0.010−0.125−0.070
Verl(n=504)−0.315−0.140−0.1100.100−0.431
W105(n=322)
W108(n=239)−0.0200.050
W117(n=445)−0.095−0.155−0.3212
W129(n=441)
W159(n=476)−0.5211
W270(n=230)−0.2400.020

[26] The reconstructed modes in the 7–10 year band show stronger correlations with ENSO than for PDO (Table 6). Correlation coefficients range up to −0.30 with the strongest correlations having lags of up to seven months, which suggest that at this periodicity there is a lag response of groundwater levels to ENSO events. Similar results are reported by Fleming and Quilty [2006] in their investigation of ENSO signals in precipitation, streamflow, and shallow groundwater levels in southwestern British Columbia. Even though ENSO events influence the annual hydrometeorological cycle, they do not always appreciably impact the timing of peak streamflows, lagging precipitation by one to two months, with groundwater having further lags by two to four months, or longer [Bonsal and Shabbar, 2008]. Since aquifers are relatively slow responding systems, winter precipitation anomalies associated with ENSO strongly control water levels for the remainder of the year [Bonsal and Shabbar, 2008].

[27] The PDO seems to have a negative influence on the low-frequency variability in groundwater levels (Table 6). From the 16 groundwater time series with significant low-frequency variability, 11 show stronger correlations with the PDO indices, ranging from −0.17 to −0.52 for the reconstructed 18–22 year band. This multidecadal (∼20 year) oscillation mode was also very apparent in the moisture-sensitive tree ring chronologies in southwest Alberta and southeast Saskatchewan, suggesting that due to the limitation of short groundwater time series, the low-frequency variability affecting the Prairie region is more reliably inferred from the tree ring chronologies, which ultimately serve as proxies of hydroclimatic variability.

3.4. Links to Oscillatory Indices

[28] Links between climate indices, groundwater level, and tree ring time series, were further examined by plotting the extracted significant frequencies and their respective climate indices to visually depict when these cycles were in their respective negative or positive phases, and by computing wavelet transform coherence (WTC) between groundwater and tree ring chronology time series, and the climate indices (ENSO and PDO) (Figures 5 and 6). WTC was used to identify both frequency bands and time intervals at which groundwater and tree ring time series and climate indices are covarying [Torrence and Webster, 1999]. Significant coherence is shown in each run of the WTC analysis, where contours enclose statistically significant periods (p≤ 0.05), based on a red-noise process as determined by a Monte Carlo experiment [Jevrejeva et al., 2003]. Another important feature of the WTC analysis is the ability to investigate the phase difference between components of the two time series during periods of significant coherence. The vectors plotted in each of the WTC analyses (Figures 5 and 6) indicate the phase differences between significant spectral oscillations identified for the groundwater level time series and the climate indices at each respective frequency, and at each time and period. Vectors pointing to the right indicate that the two signals are in phase, whereas a left-pointing arrow indicates an antiphase (180 degrees out-of-phase) relationship. Arrows deviating from the horizontal are indicative of lead-lag relationships between the two signals [Gobena and Gan, 2009]. If a link exists between two time series, a consistent or slowly varying phase lag would be expected, and the phenomena would be considered to be phase locked (i.e., phase arrows point only in one direction for a given wavelength; Grinsted et al. [2004]). Because the lead/(lag) relationships can be difficult to interpret (i.e., a lead of 90° can also be interpreted as a lag of 270° or a lag of 90° relative to the antiphase—opposite sign; Grinsted et al. [2004]), for this study, phase angle associations were noted strictly as either being in-phase/antiphase locked, or having a lead/lag relationship.

image

Figure 5. Plots showing the correlations between the reconstructed oscillation modes and the climate indices (left) and wavelet coherence between selected groundwater time series and climate indices (right). (a) Correlation between ENSO index and reconstructed groundwater time series in the 2–7 year band for wells BanB and Verl and wavelet coherence between groundwater levels at Banb and ENSO. (b) Correlation between ENSO index and reconstructed groundwater time series in the 7–10 year band for well Atto and wavelet coherence between groundwater levels at Atto and ENSO. (c) Correlation between PDO index and reconstructed groundwater time series at the ∼20 year band for wells W159 and Rice and wavelet coherence between the DEV tree ring chronology and PDO. The dashed lines in each plot (left) represent the lagged time series by nmonths. Vector directions, pointing to the right, illustrated in the wavelet coherence plots (right), indicate signals are in phase, whereas a left-pointing arrow indicates an antiphase (180 degrees out-of-phase) relationship. Arrows deviating from the horizontal are indicative of lead-lag relationships between the two signals.

Download figure to PowerPoint

image

Figure 6. (a) Wavelet coherence between precipitation at Regina and the DEV chronology. (b) Wavelet coherence between the DEV chronology and the PDO index. Vector directions pointing to the right, illustrated in the wavelet coherence plots (right), indicate signals are in phase, whereas a left-pointing arrow indicates an antiphase (180 degrees out-of-phase) relationship. Arrows deviating from the horizontal are indicative of lead-lag relationships between the two signals.

Download figure to PowerPoint

[29] For simplicity and to limit the length of the paper, only results for reconstructed frequencies of groundwater levels having the strongest correlations with climatic indices are shown (Figure 5). Results discussed describe findings common to reconstructed significant frequencies and WTC analyses for all groundwater wells, suggesting a common coherent response of the groundwater levels to common drivers of variability.

[30] Groundwater level time series show common coherence with high frequencies of the ENSO index at periodicities corresponding to 2–3 years (Figure 5a), most predominantly in the early 1970s and throughout much of the 1980s and early 1990s. Results at the interdecadal time scale (7–10 years; Figure 5b) indicate a very distinct antiphase relationship between the groundwater time series and ENSO index as depicted by the WTC analysis. This antiphase relationship is first evident in the plots of phase relationships, during the early 1990s, and is more evident throughout much of the late 20th and early 21st centuries. During positive ENSO (El Niño), warmer and drier winters affect the hydrological cycle and groundwater recharge, over much of the Canadian Prairies, as is evident in the decreased amounts of groundwater levels during these periods; a finding similar to results of investigations of streamflow and the Niño-3 index in Southern Alberta [Gobena and Gan, 2009].

[31] As wavelet analyses were unable to depict the low-frequency variability (≥20 years) in groundwater level time series, due to their limited record lengths (although evident in the MTM analyses), coherency analyses were carried out between tree ring and precipitation indices and with the PDO indices. Previous correlation analyses and the coherence indicate a strong significant relationship between the DEV chronology and precipitation time series, and ultimately the precipitation time series with the PDO index (Figure 6), validating the assumption that all three variables are responding to similar drivers of variability. Therefore, using the tree ring indices as surrogates for longer term instrumental records to depict variability at low frequencies was deemed valid. As depicted in Figure 5c, both the graphed extracted frequencies of groundwater time series, and the WTC between the DEV chronology and the PDO index, indicate a very significant antiphase relationship, supporting a physical mechanism of signal propagation from the PDO to precipitation indices, which ultimately is reflected in tree growth and groundwater recharge [Grinsted et al., 2004]. Figure 5c shows that when the PDO index switches from its extreme cool phase to warm phase during the mid 1970s, there are decreased groundwater levels. When either of these teleconnections (ENSO/PDO) is in their respective positive phases (Figure 5), groundwater levels display the effect of associated warmer and drier winters experienced over much of interior Canada and the US, affecting inputs to the hydrological cycle and groundwater recharge.

4. Conclusions

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Study Area, Data, and Methods
  5. 3. Results and Discussion
  6. 4. Conclusions
  7. Acknowledgments
  8. References
  9. Supporting Information

[32] Twenty-one groundwater level time series, seven precipitation records, and three moisture-sensitive tree ring chronologies, were investigated with the objective of analyzing the effects of climate variability on groundwater levels in the Canadian Prairies. Correlation analyses suggest that the shallow aquifers show significant positive and negative relationships to precipitation and temperature, respectively, indicating that groundwater levels are significantly influenced by hydroclimatic variability. Negative correlations were also indentified between groundwater levels and ENSO and PDO indices, suggesting that these ocean-atmosphere oscillation patterns may be possible drivers of hydrologic variability affecting groundwater levels and aquifer recharge.

[33] Results of spectral analyses (MTM and CWT) agree that there is significant variability in groundwater levels at frequencies of 2–7 and 7–10 years. Oscillation modes in these two bands were also detected in precipitation records throughout the Prairie Provinces. MTM results also suggest a multidecadal oscillation mode in groundwater levels, with periodicities of 18–22 years. This 18–22 year frequency was also identified in precipitation records, and detected in the moisture-sensitive tree ring chronologies, which record available soil moisture, a hydroclimate variable with more persistence than precipitation.

[34] Significant spectral peaks determined by MTM and wavelet analyses were investigated further via extracting reconstructions of significant frequencies, and detecting periods of common coherence between the groundwater level time series and the coupled ocean-atmosphere teleconnections, ENSO and PDO.

[35] The extracted significant high-frequency oscillation modes (2–7 and 7–10 years) in groundwater levels were found to exhibit a negative correlation to the ENSO index. Strong relationships were also determined to exist between the 18–22 yr oscillation mode in precipitation and tree ring indices and the PDO index. The linkage between groundwater levels and tree growth in their responses to precipitation serves to reinforce that both respond to a common coherent signal (i.e., precipitation) and are limited by moisture availability. Therefore we conclude that the ENSO and PDO indices show significant linkages with high- and low-frequency variability in the groundwater levels. The physical link between the PDO and groundwater levels can be explained in terms of the influence of the PDO on winter precipitation, and snow melt water as the primary source of the shallow groundwater. The results of spectral analyses coincide with the conclusions of other previous studies, which have detected the influence of ENSO and PDO on hydrologic variables in western Canada.

[36] Further investigation is recommended to examine the interactive effects of climate drivers on groundwater levels and how the atmospheric pressure associated with them might affect groundwater levels in shallow unconfined aquifers. Assessing the relationship, if there is any, between atmospheric pressure and groundwater levels in shallow unconfined aquifers might provide important information of how low-frequency atmospheric patterns might affect long-term groundwater levels. The spatial variability of teleconnection patterns represented by the different oscillation modes detected in the tree ring chronologies also needs to be considered. Special attention to the stationary, but not consistently significant signals, detected in the DEV chronology in southeastern Saskatchewan is recommended for the study of long-term hydrologic variability. Finally, moisture-sensitive tree ring chronologies capture similar climate signals to groundwater level records which may prove them useful for reconstructing historical records of groundwater levels in shallow aquifers in southern Saskatchewan as previously demonstrated by the authors for southern Alberta.

Acknowledgments

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Study Area, Data, and Methods
  5. 3. Results and Discussion
  6. 4. Conclusions
  7. Acknowledgments
  8. References
  9. Supporting Information

[37] This research was funded by the Drought Research Initiative, the Inter-American Institute for Global Change Studies (CRN2047), and the Prairie Adaptation Research Collaborative. We thank Garth van der Kamp of Environment Canada for his support and advice.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Study Area, Data, and Methods
  5. 3. Results and Discussion
  6. 4. Conclusions
  7. Acknowledgments
  8. References
  9. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Study Area, Data, and Methods
  5. 3. Results and Discussion
  6. 4. Conclusions
  7. Acknowledgments
  8. References
  9. Supporting Information
FilenameFormatSizeDescription
wrcr13231-sup-0001-t01.txtplain text document2KTab-delimited Table 1.
wrcr13231-sup-0002-t02.txtplain text document1KTab-delimited Table 2.
wrcr13231-sup-0003-t03.txtplain text document1KTab-delimited Table 3.
wrcr13231-sup-0004-t04.txtplain text document1KTab-delimited Table 4.
wrcr13231-sup-0005-t05.txtplain text document1KTab-delimited Table 5.
wrcr13231-sup-0006-t06.txtplain text document1KTab-delimited Table 6.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.