Streamflow in the Winnipeg River basin, Canada: Trends, extremes and climate linkages Scott St. George * GSC Northern Canada, Geological Survey of Canada, Ottawa, Ont., Canada Laboratory of Tree-Ring Research and Department of Geosciences, University of Arizona, Tucson, AZ 85721, USA Received 7 January 2006; received in revised form 11 July 2006; accepted 25 July 2006 Summary This study uses a network of long-term discharge gauges to examine how river flow in the Winnipeg River basin, Canada has behaved during the last one hundred years. The Win- nipeg River influences the production of over 4600 MW of hydroelectricity, and is the most important component of the hydrological system used to generate power in Manitoba. Extreme low annual flows are caused by severe reductions in runoff from spring snowmelt, and follow dry weather during the previous summer and autumn over much of the basin. These conditions are associated with enhanced meridional flow across western Canada, and geopotential height anomalies during the previous autumn and winter that are very similar to the positive phase of the Pacific/North American (PNA) pattern. The winter PNA index appears to be an important control on streamflow in the Winnipeg River at both interannual and decadal time-scales, but may be modulated by conditions in the North Atlantic sector. Mean annual flows have increased by 58% since 1924, primarily because of large increases in winter discharge. Because similar trends are observed for both regulated and unregulated rivers, these increases are not artefacts caused by direct anthropogenic interference in the hydrological system. Increasing summer and autumn precipitation is the most probable cause of the changes in streamflow. The observed trends toward higher flows, combined with recent model projections, suggest that the potential threats to water supply faced by the Canadian Prairie provinces over the next few decades will not include decreasing streamflow in the Winnipeg River basin. ª 2006 Elsevier B.V. All rights reserved. KEYWORDS Canada; Climate change; Lake of the Woods region; Nelson River basin; Rivers; Streamflow Introduction A substantial body of evidence suggests that river flow in Canada declined significantly during the 20th century. In a comprehensive study of 243 streamflow records from rivers across Canada, Zhang et al. (2001) found that annual mean 0022-1694/$ - see front matter ª 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.jhydrol.2006.07.014 * Address: Laboratory of Tree-Ring Research and Department of Geosciences, University of Arizona, Tucson, AZ 85721, USA. Tel.: +1 520 235 4599; fax: +1 520 621 8229. E-mail address: [email protected]. Journal of Hydrology (2007) 332, 396– 411 available at www.sciencedirect.com journal homepage: www.elsevier.com/locate/jhydrol
Transcript
Journal of Hydrology (2007) 332, 396–411
ava i lab le at www.sc iencedi rec t . com
journal homepage: www.elsevier .com/ locate / jhydro l
Streamflow in the Winnipeg River basin,Canada: Trends, extremes and climate linkages
Scott St. George *
GSC Northern Canada, Geological Survey of Canada, Ottawa, Ont., CanadaLaboratory of Tree-Ring Research and Department of Geosciences, University of Arizona, Tucson, AZ 85721, USA
Received 7 January 2006; received in revised form 11 July 2006; accepted 25 July 2006
Summary This study uses a network of long-term discharge gauges to examine how river flowin the Winnipeg River basin, Canada has behaved during the last one hundred years. The Win-nipeg River influences the production of over 4600 MW of hydroelectricity, and is the mostimportant component of the hydrological system used to generate power in Manitoba. Extremelow annual flows are caused by severe reductions in runoff from spring snowmelt, and followdry weather during the previous summer and autumn over much of the basin. These conditionsare associated with enhanced meridional flow across western Canada, and geopotential heightanomalies during the previous autumn and winter that are very similar to the positive phase ofthe Pacific/North American (PNA) pattern. The winter PNA index appears to be an importantcontrol on streamflow in the Winnipeg River at both interannual and decadal time-scales,but may be modulated by conditions in the North Atlantic sector. Mean annual flows haveincreased by 58% since 1924, primarily because of large increases in winter discharge. Becausesimilar trends are observed for both regulated and unregulated rivers, these increases are notartefacts caused by direct anthropogenic interference in the hydrological system. Increasingsummer and autumn precipitation is the most probable cause of the changes in streamflow.The observed trends toward higher flows, combined with recent model projections, suggestthat the potential threats to water supply faced by the Canadian Prairie provinces over the nextfew decades will not include decreasing streamflow in the Winnipeg River basin.ª 2006 Elsevier B.V. All rights reserved.
KEYWORDSCanada;Climate change;Lake of the Woodsregion;Nelson River basin;Rivers;Streamflow
0d
G5
022-1694/$ - see front matter ª 2006 Elsevier B.V. All rights reservedoi:10.1016/j.jhydrol.2006.07.014
* Address: Laboratory of Tree-Ring Research and Department ofeosciences, University of Arizona, Tucson, AZ 85721, USA. Tel.: +120 235 4599; fax: +1 520 621 8229.E-mail address: [email protected].
Introduction
A substantial body of evidence suggests that river flow inCanada declined significantly during the 20th century. In acomprehensive study of 243 streamflow records from riversacross Canada, Zhang et al. (2001) found that annual mean
Streamflow in the Winnipeg River basin, Canada: Trends, extremes and climate linkages 397
streamflow generally decreased during the past 30–50years, with the largest decreases occurring in the southernpart of the country. At the regional scale, trends towarddeclining flows have also been reported for rivers in theCanadian Rockies (Rood et al., 2005), the Hudson Bay basin(Dery and Wood, 2005) and the Canadian Prairies (Westma-cott and Burn, 1997; Yulianti and Burn, 1998).
Decreasing river flows are a particular concern becauseCanada is the world’s foremost producer of hydroelectricpower, accounting for roughly 13% of global output (NaturalResources Canada, 2005). Several Canadian provinces relyheavily on hydroelectric energy, and generate substantialrevenue from the sale of electricity to other provincesand the United States. In Manitoba, hydropower providesnearly 95% of total electricity production (Manitoba Hydro,2005). Although most of its watershed is in Ontario, the Win-nipeg River (Fig. 1) is the most important component of thehydrological system used to generate power in Manitoba.
Legend
Flow gauging statio
Communities D
Kilometres0 20 40 60 8
Climate station
95Wº
Lake
Winnipeg
MINNES
MANITOBA
Lake of
the Woods
Kenora
Fort France
Winnipeg
Winnipeg
River
E n g l i s h
R
i v e r
Rainy River
Red
Riv
er
WRS
WRB
RRMR
WHI
Figure 1 Map of the Winnipeg River basin showing the locations ofmap provided by the Lake of the Woods Control Board.
The river supports six hydroelectric generating stations inthe province, which collectively produce 585 MW of elec-tricity (roughly 14% of total provincial production). Further-more, the river provides nearly half (45%) of the flow intoLake Winnipeg and the Nelson River (despite comprisingonly 12% of the Nelson River watershed), and has an impor-tant influence on the production of nearly 4000 MW atdownstream stations on the Nelson River. Widespreaddrought affecting most of the Winnipeg River basin was ci-ted as the main factor responsible for reducing energy pro-duction and sales during the 2003–2004 fiscal year(Manitoba Hydro, 2004). Conversely, a return to wetter con-ditions in 2004 and 2005 was credited with greatly improvedhydroelectric conditions in Manitoba (Manitoba Hydro,2005).
This study uses a network of long-term discharge gaugesto assess how river flow in the Winnipeg River basin has be-haved during the last one hundred years. Whereas most
ns
ams
0 100
50 Nº
52 Nº
4º8N
93Wº
OTA
ONTARIO
Lake Superior
s
Lac Seul
RainyLake
Atikokan
Dryden
Lake St.JosephDiversion
Albany River
RF NAM
ENG
BAS
TUR
the climate station and stream gauges used in this analysis. Base
398 S. St. George
prior research in Canada has used streamflow records asindicators for the purpose of climate change detection(Burn, 1994; Westmacott and Burn, 1997; Yulianti and Burn,1998; Zhang et al., 2001), this study examines the hydrolog-ical record to provide information that may be relevant tothe management of water resources within this importantwatershed. Its main goals are to (1) describe the meanhydrology and hydroclimate of the Winnipeg River basin;(2) evaluate if river flow increased or decreased duringthe period of record, and investigate the potential causesof such changes; (3) determine if extreme high or low flowsare associated with particular configurations of the synopticclimatology; and (4) describe the relative importance ofdecadal-scale variability in streamflow.
Study area
Most of the Winnipeg River basin is located at the easternedge of the Canadian Shield in northwestern Ontario, andcan be described as a rugged wilderness area with low, roll-ing terrain and many large lakes. The Winnipeg River, whichalso drains a small portion of northern Minnesota, flowswestward into Manitoba and empties into the south basinof Lake Winnipeg. The basin has an area of approximately150,000 km2, and supports more than 100 major lakes(including Lake of the Woods) that occupy more than
WRS
Year (AD)
1900 1920 1940 1960 1980 2000
0
500
1000
1500
2000
WRB
0
250
500
750
1000
RRM
0
250
500
750
RRF
0
200
400
600
NAM
Year (AD)
1900 1920 1940 1960 1980 20000
50
100
150
200
250
Mea
nan
nual
flow
(m/s
)3
Figure 2 Mean annual flows for the nine long-term gauge recordsgrey bars mark every two decades.
11,400 km2 (Lake of the Woods Control Board, 2002). Thebasin is largely forested, and serves as a transition zone be-tween the mixed forests of the northern United States andthe boreal forest of northern Canada.
The WRB can be divided into two major sub-basins. TheEnglish River basin comprises the northeastern half of theWRB, and drains an area of roughly 55,000 km2. The EnglishRiver system also includes contributions from Lac Seul and(through a diversion) Lake St. Joseph. The Lake of theWoods/Rainy River system drains 70,000 km2, and includesmost of the area between Lake Superior and the Red Rivervalley. Several protected areas are located in the Lake ofthe Woods/Rainy River basin, including the Boundary WatersCanoe Area and Voyageurs National Park in Minnesota, andQuetico Provincial Park in Ontario.
Water flow in the Winnipeg River basin has been con-trolled since the 1880s, when small dams were constructedto aid navigation on Lake of the Woods (Lake of the WoodsControl Board, 2002). The first hydroelectric dam in the ba-sin was built on the eastern outlet of Lake of the Woods in1892, which was followed by a dam on the lake’s westernoutlet in 1925 (Lake of the Woods Control Board, 2002).The Pinewa Dam, located on the Winnipeg River, went intooperation in 1906 with an initial generating capacity of eightmegawatts. By 1954, seven additional dams had been builtin Manitoba and Ontario, bringing total power production
ENG
Year (AD)1900 1920 1940 1960 1980 2000
0
25
50
75
100
125
BAS
0
20
40
60
80
TUR
0
20
40
60
80
WHI
Year (AD)
1900 1920 1940 1960 1980 20000
10
20
30
40
Mea
nan
nual
flow
(m/s
)3
used in this study. Gauge codes are presented in Table 1. The
Streamflow in the Winnipeg River basin, Canada: Trends, extremes and climate linkages 399
on the river to more than 650 MW. Dams on the WinnipegRiver are run-of-river systems, and have limited storagecapacity. The only major diversion within the basin trans-fers water from Lake St. Joseph, which is part of the AlbanyRiver basin, into Lac Seul. This channel, built in 1935, adds80 m3/s of discharge on average into the Winnipeg River sys-tem (Lake of the Woods Control Board, 2002).
Data
Daily and monthly streamflow data were obtained from theWater Survey of Canada’s HYDAT data archive (Water Sur-vey of Canada, 2005, http://www.wsc.ec.gc.ca/hydat/H2O/). Gauges within the Winnipeg River watershed wereselected for analysis if they had more than 50 years of re-cord, and were still operating at the end of 2003. Nine gaugerecords satisfied these criteria, with the longest record(Winnipeg River below Lake of the Woods) beginning in1892 (Fig. 2; Table 1). The gauge record for the WinnipegRiver at Slave Falls was augmented by data for 2004 ob-tained from Manitoba Hydro (B. Girling, personal communi-cation, 2005). As it is the gauging station closest to theendpoint of the basin, the Slave Falls record provides thebest indication of flow within the entire watershed, and willbe used as a focus for much of the subsequent analysis. Theintegrity of these nine gauge records is excellent, with veryfew (�0.7%) values missing from the monthly series. Missingmonthly values were replaced by estimates obtained usingmultiple regression models developed from other gaugesin the set. As no gauges are currently operating on the Eng-lish River near its confluence with the Winnipeg, a proxy re-cord of English River flow was created by taking thedifference in monthly discharge on the Winnipeg River be-
Table 1 Streamflow gauges used in this study
Station No.a Name Code Latitude L
05PF063 Winnipeg River at SlaveFalls
WRS 50�13 03000N 9
05PE020 Winnipeg River belowLake of the Woods
WRB 49�47 00000N 9
05PC018 Rainy River at ManitouRapids
RRM 48�38 00400N 9
05PC019 Rainy River at FortFrances
RRF 48�36 03000N 9
05PA006 Namakan River at outletof Lac La Croix
NAM 48�23 00000N 9
05QA002 English River atUmfreville
ENG 49�52 03000N 9
05PA012 Basswood River nearWinton
BAS 48�04 05500N 9
05PB014 Turtle River near MineCentre
TUR 48�51 00000N 9
05PH003 Whitemouth River WHI 49�56 02000N 9a Water Survey of Canada station number.b Median annual flows are calculated for the period 1929–2003, wit
1957–2003 period.
tween the Slave Falls gauge and the gauge directly belowLake of the Woods. The four downstream gauges (and theEnglish River proxy) measure flows that are influenced bya series of control structures, while the other five are lo-cated within unregulated watersheds.
Precipitation and temperature data for selected stationswithin the Winnipeg River region was downloaded from theAdjusted Historical Canadian Climate Data (EnvironmentCanada, 2005, http://www.cccma.bc.ec.gc.ca/hccd). Thisdata set consists of rehabilitated and homogenized climaterecords that have been corrected for inhomogeneities andmissing data (Vincent and Gullett, 1999; Mekis and Hogg,1999). This study also uses a subset of the gridded precipi-tation dataset produced by New et al. (1999, 2000) thatspans the 1901–1998 period. Major advantages offered bythis dataset include a lack of missing values, a higher spatialresolution than previous global precipitation compilations,and a method of construction that was intended to minimizeinterpolation errors due to topographic variability or ab-sence of data. The subset used in this study (45�–60�N,80�–120�W) covers all of northwestern Ontario and theCanadian Prairie Provinces, and parts of adjacent Americanstates.
Synoptic climatology maps were produced using the Na-tional Centers for Environmental Prediction/National Cen-ter for Atmospheric Research Reanalysis (Kistler et al.,2001). The Reanalysis provides more than 50 years of globalatmospheric fields over a 2.5� by 2.5� gridded network(equivalent to a horizontal resolution of 210 km). Geopoten-tial height and air temperature are considered ‘‘type A’’variables in the Reanalysis, which are influenced stronglyby observations and are therefore reliable products (Kalnayet al., 1996).
ongitude Drainagearea(km2)
Medianannualflow(m3/s)b
Span Regulated/Unregulated
5�34 01500W 125,000 868.6 1907–2004 R
4�30 05000W 70,400 436.0 1892–2003 R
3�54 04700W 50,200 376.7 1928–2003 R
3�24 00000W 38,600 281.8 1905–2003 R
2�10 04000W 13,400 112.0 1921–2003 U
1�27 03000W 6230 56.3 1921–2003 U
1�39 01000W 4510 38.3 1924–2003 U
2�43 03000W 4870 37.5 1914–2003 U
5�57 02000W 3750 11.8 1942–2003 U
h the exception of the Whitemouth River record, which is for the
Medians and percentiles were calculated for streamflowover the period of record for each gauging station. Meanprecipitation fields were averaged over the 1901–1998 per-iod, with anomaly maps produced by subtracting the gridpoint mean from each observation. Composite precipitationanomaly maps were created by averaging the anomaly fieldsfor several years, with seasonal maps showing aggregateanomalies for different three-month periods.
The procedure described by Yue et al. (2002) was usedto determine if streamflow and precipitation time seriescontained significant linear trends. Despite the earlierbeginning for some gauges, trend analysis was restrictedto the 1924–2003 period to allow comparisons betweengauges using a common interval (with the exception ofthe gauge on the Rainy River at Manitou Rapids, which be-gins in 1929). The streamflow record for the WhitemouthRiver was excluded from trend analysis because year-roundmeasurements are not available from this gauge until1956–1957. For each monthly and annual (October–Sep-tember) streamflow series, the slope of its trend wasdetermined using the robust estimator described by Hirschet al. (1982), and the series was detrended. The first-orderautocorrelation was then removed, and the prewhitenedseries was re-combined with the estimated trend. TheMann-Kendal test was performed on the prewhitened,‘trend-added’ series to determine if the trend was statisti-cally significant.
Hydrological storage, either at the surface (e.g., wet-lands, lakes and reservoirs) or as groundwater, delays thepassage of water through the hydrological system, and cancause streamflow series to have large autocorrelation val-ues, sometimes at lags for several years afterward (Salas,1992). Strong autocorrelation (also described as persistenceor memory) leads to current observations being influencedby prior conditions, and can obscure relationships betweenclimatic forcings and hydrological ‘outputs’. The degree ofyear-to-year persistence was determined by examining theautocorrelation and partial autocorrelation functions foreach monthly and annual (mean October to September) flowseries. Subsequently, persistence within each streamflowseries was removed using either an autoregressive (AR) orautoregressive moving average (ARMA) model. These ‘pre-whitened’ series are not influenced by the condition ofstreamflow in prior years, and should allow more reliablecomparisons between hydrological variability and climaticconditions. Prewhitened hydrological series were used formost of the following analyses.
Monthly streamflow series were correlated against fieldsof monthly precipitation (with precipitation leading stream-flow by 0–11 months), and mapped to display spatial pat-terns. Identified lag times between precipitation withinthe basin and the hydrological response were used to directthe comparisons between streamflow and synoptic climateconditions. Two complimentary approaches were used torelate streamflow variability to the state of atmospheric cir-culation, as represented by monthly fields of geopotentialheights at the 500 mb level. Monthly flow series were corre-lated against the northern North American sector of thegeopotential height field (40�–90�N, 0�–180�W). Anomaly
height maps were produced through compositing years ofextreme (high and low) flows in the Winnipeg River basin,and by examining the height fields for each year includedin the composite.
Patterns of longer-term variability were investigated byfiltering the annual streamflow records to emphasize vari-ance at decadal and greater timescales. Low-pass filteredversions of streamflow series were produced using a 9-point Gaussian filter designed to preserve 50% of theamplitude response at frequencies of 10 years. Nearlyidentical results were obtained using either a binomial orHamming filter.
Results and discussion
Mean hydrological and climatic characteristics
Due to the basin’s location in the interior of North America,its climate is strongly continental, with the difference be-tween maximum and minimum monthly temperatures at Ke-nora, Ontario being more than 36 �C (Fig. 3). FromNovember to March, nearly all (89%) precipitation falls assnow. Snowstorms do occur in October and April and, veryrarely, in September and May. The basin receives most ofits precipitation between May and September, with Julybeing the wettest month.
Compared with regions to its west, the Winnipeg River re-gion has a relatively moist climate. Precipitation increasesfrom west to east across the basin, with mean annual totalsranging from 500 mm/yr in southeastern Manitoba to725 mm/yr directly west of Lake Superior. Wetter condi-tions within the basin are caused mainly by increased sum-mer and autumn precipitation over northwestern Ontarioand northern Minnesota. This part of the basin receives be-tween 240 and 280 mm of precipitation in June, July, andAugust, roughly 40–100 mm more than falls in the southernCanadian Prairies. Total precipitation in autumn is less, butthe east–west moisture gradient is of similar magnitude.
Peak flow at the downstream end of the Winnipeg Riverusually occurs during June and July, with mean daily flowsat Slave Falls on the order of 875 m3/s (Fig. 3). Flow usuallystarts to decline by the beginning of August and reachesminimum values around 650 m3/s in mid-October. Dischargeincreases throughout autumn and is relatively stable fromJanuary to March. May is characterised by an upsurge in flow(typically �120 m3/s) associated with passage of the springsnowmelt pulse through the Winnipeg River system. The lev-els of Lake of the Woods and Lac Seul are regulated, whichundoubtedly has an effect on the timing of peak flows in theWinnipeg River downstream and may reduce the seasonalvariation in flows at Slave Falls. Unfortunately, becauseLake of the Woods was dammed several decades beforethe initiation of discharge measurements on the WinnipegRiver, it is not possible to estimate the magnitude of the ef-fects of regulation using hydrological records. Hydrographsfor smaller, unregulated basins in the upstream reachesshow earlier peak flows (mid-April to mid-May), suggestingthat the seasonal rhythm of flow for these rivers is morestrongly influenced by inputs from local snowmelt. At theseupstream gauges, the steady decline in flow after the snow-melt peak is often interrupted by a brief return to higher
Oct 1 Nov 1 Dec 1 Jan 1 Feb 1 Mar 1 Apr 1 May 1 Jun 1 Jul 1 Aug 1 Sep 1
0
500
1000
1500
2000
2500
20
40
60
80
100
120
-10
0
10
20
30
-20
Prec
ipita
tion
(mm
)M
ean
daily
flow
(m/s
)3
Mean
temperature
(C
)º
Oct 1 Nov 1 Dec 1 Jan 1 Feb 1 Mar 1 Apr 1 May 1 Jun 1 Jul 1 Aug 1 Sep 1
90 pctth
10 pctth
Median
Figure 3 Average hydroclimatic conditions within the Winnipeg River basin. The upper chart shows median and percentiles of dailystreamflow for the Winnipeg River at Slave Falls, calculated over the 1907–2004 interval. The lower chart shows the climograph forKenora, Ontario. Vertical bars shaded grey and white indicate the amount of precipitation that falls as rain or snow respectively.
Streamflow in the Winnipeg River basin, Canada: Trends, extremes and climate linkages 401
flows in June and July that reflects the impact of convectiverainstorms on local runoff.
Trend analysis
Trend analysis indicates that mean annual flows of the Win-nipeg River have increased substantially since 1924 (Fig. 4).At the Slave Falls gauge, annual flows have increased by4.8 m3/s/yr; relative to mean flows at the beginning ofthe record, this cumulative change corresponds to an in-crease of 58%. Results for the monthly flow series suggestthat the shift to higher annual flows has been driven by largeincreases in discharge during the winter. Statistically signif-icant (p = 0.05) increases are observed for flows fromNovember to April, ranging between +60% and +110%. Thereare no significant changes in discharge during the summermonths.
The tendency toward higher winter flows is repeatedthroughout the watershed at gauges upstream from SlaveFalls. All significant trends are positive, with nearly all (37out of 40) occurring between November and April (Table2). For the Turtle River, significant positive trends in monthlydischarge from November to March led to a 46% increase inannual flow. Increasing winter and spring flows were also ob-served on the Basswood River and the English River at Umfre-ville, indicating that these seasonal changes occurred over awide geographic area. However, in the case of the Basswoodand English rivers, these seasonal increases were not largeenough to cause a significant rise in annual discharge. The
trend analysis results also suggest that the English River basinhas experienced larger increases in winter flows than haverivers in the Lake of the Woods-Rainy River watershed. Theproxy flow record for the English River has greater propor-tional increases in discharge than any other gauge in the sys-tem, with changes in monthly flow of up to 160% (March).
Annual and winter discharge also increased on the BigFork, Sturgeon and Roseau rivers in northern Minnesotaafter 1960, indicating that trends in the WRB are not arte-facts caused by changes in techniques used by Canadianagencies to measure discharge. Although some increasesare observed on WRB rivers that are affected by regulationand diversion, the presence of similar trends at upstreamgauges demonstrates that direct anthropogenic interfer-ence in the hydrological system is not the primary causeof these changes. However, the possibility that some frac-tion of the trends in flow at the downstream gauges is dueto regulation effects cannot be excluded.
Rising winter discharge across the basin has coincidedwith increasing annual and seasonal precipitation. Totalannual (October–September) precipitation at Kenora in-creased by approximately 108 mm (15.5%) since 1924. Thischange largely reflects increases in summer (May–July)and autumn (August–October) precipitation, as no signifi-cant trends were observed for winter (November to January)precipitation, and precipitation during spring (February toApril) decreased by 25%. No significant trends were identi-fied in annual, monthly or seasonalised temperature recordsfrom Kenora.
1900
1920
1940
1960
1980
2000
0
500
1000
1500
2000
2500
3000
Janu
ary
1900
1920
1940
1960
1980
2000
0
500
1000
1500
2000
2500
3000
Feb
ruar
y
1900
1920
1940
1960
1980
2000
0
500
1000
1500
2000
2500
3000
Mar
ch
1900
1920
1940
1960
1980
2000
0
500
1000
1500
2000
2500
3000
Apr
il
1900
1920
1940
1960
1980
2000
0
500
1000
1500
2000
2500
3000
May
1900
1920
1940
1960
1980
2000
0
500
1000
1500
2000
2500
3000
June
1900
1920
1940
1960
1980
2000
0
500
1000
1500
2000
2500
3000
July
1900
1920
1940
1960
1980
2000
0
500
1000
1500
2000
2500
3000
Aug
ust
1900
1920
1940
1960
1980
2000
0
500
1000
1500
2000
2500
3000
Sept
embe
r
1900
1920
1940
1960
1980
2000
0
500
1000
1500
2000
2500
3000
Oct
ober
1900
1920
1940
1960
1980
2000
0
500
1000
1500
2000
2500
3000
Nov
embe
r
1900
1920
1940
1960
1980
2000
0
500
1000
1500
2000
2500
3000
Dec
embe
r
+568
m3 /s
+110
%+5
60m
3 /s+1
03%
+536
m3 /s
+101
%+3
60m
3 /s+6
1%
+424
m3 /s
+74
%+5
36m
3 /s+9
9%
Figure
4Lineartrendsin
monthly
stream
flow
fortheWinnipegRiverat
Slav
eFa
llsfrom
1924
to20
03.Statistica
llysign
ifica
ntincreasesin
flow
arehighligh
tedwithaheav
yline.
Totalan
dpercentch
ange
sare
exp
ressedrelative
tothebeginningofthetrendline.Annual
disch
arge
isexp
ressedin
m3/s.
402 S. St. George
Table 2 Trend analysis results for gauges in the Winnipeg River basin, by month
Code Parametera Annual Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
a Values are not shown for months without trends that are significant at the 0.05 level.b Slopes are expressed in units of m3/s/yr.c Percent change is relative to the beginning of the trend line in 1924.d The period of analysis for the Rainy River at Manitou Rapids gauge begins in 1929.
Streamflow in the Winnipeg River basin, Canada: Trends, extremes and climate linkages 403
The changes observed at Kenora are part of a general in-crease in precipitation across much of northwestern Ontario(Fig. 5). Summer precipitation has risen from between 20%to 60% in the eastern sector of the English River basin, thenorth end of Lake of the Woods, and in the Whitemouth ba-sin of southeastern Manitoba. Autumn precipitation has in-creased over a much more extensive area, with changes of+40 to +60% occurring over most of the north-central andnorth-eastern part of the Winnipeg River basin. This areacoincides with the upper portion of the English River basin,which may explain why that basin has shown the largest in-creases in wintertime flows. Over this period, autumn pre-cipitation has increased by similar proportions over muchof the northern Prairie Provinces.
In a recent study, Dery and Wood (2005) demonstratedthat annual flow into the western Hudson Bay basin (includ-ing the Nelson River) declined by 13% from 1964 to 2000.Repeating the analysis for the Winnipeg River using onlydata between 1964 and 2000 produces a declining (butnon-significant) trend in annual flows. Although the Winni-peg River provides only a portion of the total flow into the
western Hudson Bay basin, it seems plausible that some ofthe apparent decrease in flows reported by Dery and Woodis due to anomalously high flows during the 1960s and early1970s, rather than being caused by more recent flows beingabnormally low. However, this suggestion should be evalu-ated using longer flow records from other rivers within thewestern Hudson Bay catchment. In any case, these compar-isons provide a reminder that the results of any trend testsdepend strongly on the interval used for analysis. For a spe-cific application (e.g., evaluating the impact of changingflow levels on a hydroelectric generating facility), it wouldprobably be more useful to tie the analysis to specific inter-vals used for planning (such as expressing changes relativeto the period used originally to determine the range of ex-pected flows).
These results indicate that the Winnipeg River behavedvery differently than most rivers in Canada during the20th century, with flows increasing rather than followingthe national trend towards lower flows. Moreover, these in-creases are 4–5· greater than changes reported in otheranalyses of Canadian streamflow (Dery and Wood, 2005;
20
0
0
20
40
40
404040
40
60
60
80
40
100
20
40
120
40
40
60
60
-20
40
6060
40
60
-40
60
40
80
20
80
(a) Change in May-Jun-Jul precipitation
40
40
20
30
30
304040
40
40
50
50
40
60
50
70
80
20
-50
60
30
50
40
0
(b) Change in Aug-Sep-Oct precipitation
6080
60
60ºN
57ºN
54ºN
51ºN
48ºN
45ºN120ºW
112ºW104ºW 96ºW
88ºW
80ºW
60ºN
57ºN
54ºN
51ºN
48ºN
45ºN120ºW
112ºW104ºW 96ºW
88ºW
80ºW
Figure 5 Patterns of significant trends in (a) summer (May to July) and (b) autumn (August to October) precipitation from 1924 to1998. Contours represent percent changes in seasonal total precipitation relative to the beginning of the trend line at each grid point.
404 S. St. George
Rood et al., 2005). Increasing flows have also been reportedfor rivers in the coterminous United States (McCabe and Wo-lock, 2002), which suggest that the Winnipeg River may bemore closely related to factors that influence streamflowin the northern US than it is to processes affecting other riv-ers in western Canada.
Autocorrelation structure
For the entire basin, roughly 11% of the total variability inannual streamflow is contributed by flow conditions duringthe prior year. This persistence largely reflects the influenceof cold season conditions; monthly flows from November to
Streamflow in the Winnipeg River basin, Canada: Trends, extremes and climate linkages 405
April have significant autocorrelation at lags from one tothree years (Fig. 6). The fraction of variance contributedby autocorrelation varies between 8% (November) and 42%(February) for this interval. In contrast, there is nosignificant autocorrelation for discharge during the warmseason (May to October). Autocorrelation is generally lessimportant for gauges situated on lower-order streams. Forexample, the Turtle River does not show significant autocor-relation in its annual flow record, and only weak autocorre-lation (2–8% of the variance) for flows in March, May andAugust. Similarly, the gauge records from the Basswood Riv-er and the English River at Umfreville do not have significantpersistence at the annual timescale, and have only scat-tered significant (and weaker) autocorrelation for individualmonths.
Autocorrelation in seasonal and annual flow series is highat gauges that are downstream of the largest lakes in thebasin, and low at gauges located on upstream tributaries.This pattern implies that a large portion of the autocorrela-tion observed at the downstream gauges is caused primarilyby the influence of hydrological storage, and is not pro-duced by persistence in the climate system. Furthermore,the autocorrelation function for Slave Falls indicates thatstrong positive or negative flow anomalies can affect waterlevels (in winter) for up to three years afterward.
The variance contributed by autocorrelation was re-moved from the monthly and annualized flow series usinga combination of AR1 and ARMA(1,1) models (Box and Jen-kins, 1976). These prewhitened series were used in subse-quent analysis to remove the possible confounding effectsof hydrological persistence from streamflow-precipitationcorrelations and on the ranking of years based on flowmagnitude.
Correlations between streamflow and precipitation
Correlation maps (not shown) suggest that variations in theflow of the Winnipeg River lag regional precipitation by sev-eral months, and that lags vary between zero and six monthsthroughout the year. Summer flows (June, July and August)at Slave Falls are significantly correlated with precipitationin April, May and June. August is the only month during theyear with significant correlation between flow and precipi-
Jan Feb Mar Apr May
Cor
rela
tion
coef
fici
ent
-0.4
-0.2
0.0
0.2
0.4
0.6
Figure 6 Partial autocorrelation coefficients at one, two and thrmark the 95% confidence limits. Coefficients that exceed the confid
tation in the current month. Streamflow between Octoberand March is most strongly related to precipitation betweenAugust and October, with significant relationships observedusing lags up to 6 months. Cold-season flows are moststrongly correlated with September precipitation for anextensive area over northwetern Ontario and northern Min-nesota. Spring flows (March, April and May) are also signifi-cantly correlated with precipitation during November.Precipitation during May and September are more fre-quently correlated with streamflow than is precipitation inany other months.
Extreme years
Years with extremely high or low streamflow are listed inTable 3, with rankings provided for both the observed andprewhitened flow series for the Winnipeg River at SlaveFalls. The observed flows describe the amount of wateractually measured at the Slave Falls gauge, while the pre-whitened flows reflect the mean discharge that would havebeen observed if hydrological conditions in prior years didnot have any influence. Prewhitening the flow data en-hances the importance of years when extreme flows oc-curred subsequent to opposite conditions in the prioryear. For example, 2003 is the fifth lowest flow in the ob-served record, but is ranked as the lowest flow year in theprewhitened series because above-average flows occurredin 2002. Conversely, prewhitening diminishes the impor-tance of years were anomalously high or low flows followone or more years with similar conditions, such as the early1940s.
The daily hydrological records for the three lowest-flowyears (in the prewhitened dataset) show that these eventsare characterized by severe reductions in runoff receivedfrom spring snowmelt. In each case, discharge begins to de-crease in April and continues to fall until the end of July, in-stead of increasing throughout May and June, as is typical(Fig. 7). In 2003, flows continued to decline until the middleof August. Flows during these three years were also rela-tively low at the beginning of the hydrological year in Octo-ber, and remained at or below the 10th percentile of dailyflow. Discharge measurements during the spring and sum-mer of 1977 are the lowest ever recorded on the Winnipeg
Month
Jun Jul Aug Sep Oct Nov Dec
ee-year lags for monthly flows at Slave Falls. The dashed linesence limits are shaded black.
Table 3 Years with extreme high and low annual (Octoberto September) streamflow at the Winnipeg River at SlaveFalls gauge, showing results for both prewhitened (A) andobserved (B) streamflow
River, including a report of 51 m3/s at Slave Falls on June12. The daily series for 1931 is notable for its near absenceof seasonal variability, with the exception of a brief intervalof higher flows in April. Extreme low flow was observed formost of the 1910/1911 winter, until increasing discharge inthe spring and summer gradually returned flows to close toaverage by the end of summer.
Extremely low flows are typically preceded by wide-spread reductions in the amount of precipitation receivedby the Winnipeg River basin in year prior1 (Fig. 8). Dry con-ditions become established over northwestern Ontario insummer, and by autumn expand to affect southern Mani-toba and northern Minnesota as well. The largest negativeprecipitation anomalies are centered over the Rainy Riverbasin. Winter snow is also reduced, with most of the basinreceiving about 20 mm (�25%) less than average. Springand summer conditions appears to be less critical to thedevelopment of low flows, as mean precipitation anomaliesover the basin are close to average during spring, andslightly above-average in summer. The lowest flows inthe smaller, unregulated watersheds (the Basswood, Eng-lish, Namakan, and Turtle river gauges) also follow largereductions in precipitation during the prior autumn (not
1 2003 is not included in the low flow composites because the Newet al. (1999, 2000) dataset ends in 1998.
shown), but dry conditions are typically restricted to thoseindividual sub-basins, rather than extending over the entireWRB.
Years with exceptionally high flows are associated withnearly the opposite pattern of precipitation anomalies(Fig. 9). Over the entire Winnipeg River watershed, precip-itation during the previous autumn is between 40 and 60 mm(20–30%) above the mean. The unparalleled high flows in1974 followed an unusually wet autumn that saw northwest-ern Ontario receive between 50 and 125 mm (25–70%) moreprecipitation than average. In general, smaller positiveanomalies occur over most of the basin in winter. Althoughprecipitation is close to average during spring, high flowyears are also linked to a second ‘pulse’ of above-averageprecipitation in May, June and July.
Maps of correlation between annual streamflow at SlaveFalls and seasonal precipitation (not shown) have patternssimilar to the high- and low-flow composites, with themap for prior autumn precipitation describing the largestpositive correlations over northwestern Ontario.
Low flow years are associated with enhanced meridio-nal flow during the prior autumn and winter months. Cir-culation patterns during autumn are characterised by astrengthened ridge over British Columbia and a deepenedtrough over eastern North America (Fig. 10). This patternintensifies through the winter, with the centre of the ridgemoving eastward to occupy a position above the northernRocky Mountains. The ridge’s eastward shift is particularlyevident prior to the occurrence of low flows in 1977 and2003. These height anomalies are very similar to thoseassociated with the positive phase of the Pacific/NorthAmerican Pattern (Wallace and Gutzler, 1981), except thatanomalously low heights extend north over Atlantic Can-ada and the North Atlantic. These conditions also coincidewith an enhanced Aleutian low, and a slight weakening ofthe Icelandic low. As noted above, precipitation from Feb-ruary to July is not unusually low, but anomalously warmtemperatures suggest that increased evaporation duringspring and summer acts to reduce runoff and streamflowtoward the end of the water-year. Spring and summertemperatures across the basin (not shown) are roughly2–3 �C warmer in low flow years as compared to high flowyears.
During winter, northwestern Ontario is affected by lowpressure systems that originate from cyclogenesis regionsin Colorado and northern Alberta (Steward et al., 1995).The presence of an enhanced ridge over western Canadamay inhibit cyclogenesis or redirect systems along thesestorm tracks and suppress vertical movement in the atmo-sphere, thereby reducing the amount of precipitationdelivered to northwestern Ontario by winter snowstorms.This scenario is consistent with Angel and Isard (1998)’sobservations that winter cyclones in the Great Lakes regionare less frequent when the PNA index is positive (i.e.,when circulation above western North America is moremeridional).
At the other extreme, high flow years coincide with in-creased zonal circulation, driven by a reduced Rocky Moun-tain ridge and a muted trough in the east. Rogers andColeman (2003) identified a similar association between zo-nal conditions and high stream levels for the Tennessee andOhio rivers. This configuration produces more direct
Oct 1 Nov 1 Dec 1 Jan 1 Feb 1 Mar 1 Apr 1 May 1 Jun 1 Jul 1 Aug 1 Sep 1
0
200
400
600
800
1000
0
200
400
600
800
1000M
ean
daily
flow
(m/s
)3
0
200
400
600
800
1000
0
200
400
600
800
1000
Date
Oct 1 Nov 1 Dec 1 Jan 1 Feb 1 Mar 1 Apr 1 May 1 Jun 1 Jul 1 Aug 1 Sep 1
0
200
400
600
800
1000
(a) 2003
(b) 1977
(c) 1988
(d) 1931
(e) 1911
median
10 percentileth
Figure 7 Daily hydrographs showing discharge in the Winnipeg River at Slave Falls for those years with the five lowest mean annualflows: (a) 2003; (b) 1977; (c) 1988; (d) 1931; and (e) 1911. The upper and lower bound of the gray region represents the median and10th percentile of flow for each Julian day.
Streamflow in the Winnipeg River basin, Canada: Trends, extremes and climate linkages 407
movement of air eastward across western Canada, and mayallow moisture-laden air masses from the Gulf of Mexico topenetrate more easily into northwestern Ontario.
Decadal variability
Between 15% and 25% of the total variance of flow recordsin the basin is contributed by variability at decadal time-scales and greater. At the Slave Falls gauge, the proportionof decadal-scale variance exhibited by seasonal flow seriesranges between 13.6% and 15.6%, indicating that longer-term changes in flow are of equal importance throughoutthe year. The decadally-filtered version of the Slave Fallsrecord (Fig. 11) shows three major periods of persistentlyanomalous flow in the Winnipeg River basin. With theexception of 4 years, Winnipeg River flow was below the1908–2004 mean for the entire period between 1908 and1941. This interval includes the lowest 10-year averageflows in the record centred around 1930. At the opposite
extreme, river flows were at their highest in the 1960sand early 1970s, during which time average dischargewas 200–250 m3/s above average. This interval matchesthe period of highest flows observed in the Little Fork River,Minnesota (Mauget, 2004), which is one of the WinnipegRiver’s American tributaries. The climate record at Kenoraindicates that high river flow during the 1960s coincidedwith above-average precipitation in all seasons, with thelargest increases occurring in spring and autumn. Themean geopotential height field for 1962–1975 shows that,on average, winter circulation during this period was char-acterized by a weakening of the ridge over western Can-ada, and of the Aleutian and Icelandic lows. Thisconfiguration suggests that (1) the processes that controlinterannual changes in streamflow also control longer-termshifts in the hydrological regime and (2) the 1960s wet‘pulse’ was caused by a persistent shift to a winter circu-lation that favoured increased precipitation and river flowsin northwestern Ontario.
Figure 8 Composite maps of seasonal precipitation anomalies for the 5 lowest-flow years (1977, 1988, 1931, 1911 and 1987) in theWinnipeg River basin. Precipitation anomalies are expressed in millimetres.
408 S. St. George
Although long-term discharge has been above the meansince 1991, recent flows have not been as high as those ob-served during the 1960s and 70s. This change has coincidedwith wetter conditions driven solely by increases in summerand autumn rainfall, as there have been no apparentchanges in precipitation during winter and spring at Kenoraduring the last two decades.
Potential connections to the Pacific/NorthAmerican pattern
As noted, extreme low flows in the Winnipeg River areassociated with a winter circulation pattern than resem-bles the Pacific/North American (PNA) pattern. Many ofthe lowest flows in the Winnipeg River (1961, 1977,1981, 1988, 1998 and 2003) followed years when the win-ter (DJF) PNA index was strongly positive (above 0.8).
The winter PNA is also significantly (p = 0.01) correlatedwith annual (r = �0.36) and spring (r = �0.46) flow at SlaveFalls. These results suggest that the state of the winter-time PNA may act as an important control on streamflowin the Winnipeg River basin. The index was negative duringmost of the 1960s and early 1970s, which indicates thatthis mode may also have contributed to persistently highflows during this period. However, the winter PNA indexwas also positive prior to high flows in 1970 and 2001.The geopotential height fields for these two years showsthe typical PNA pattern over western North America, butalso includes regions of anomalous high pressure centeredover northern Quebec, the Labrador Sea and southernGreenland. These patterns suggest that the apparent rela-tionship between the PNA and streamflow in the WinnipegRive basin may be modulated by conditions in the NorthAtlantic sector.
-40
-20
0 20
204020
20
6060
40 40
80
00
20
40
60
0
40
80
20
20
40
040
20
0
0
0
0
20
20
20
40
40
60
20
20
100
40120
20
20
4060
0
0
40
November-December-January
February-March-April May-June-July
prior August-September-October
0
-2020
0
0
020
-20
0
-20
-40
020
40
60
80
-20
0
2040
20
-40
-20
0
60 No
57 No
54 No
51 No
48 No
45 No
120 Wo112 Wo 104 Wo
96 Wo 88 Wo 80 Wo
60 No
57 No
54 No
51 No
48 No
45 No
120 Wo112 Wo 104 Wo
96 Wo 88 Wo 80 Wo
60 No
57 No
54 No
51 No
48 No
45 No
120 Wo112 Wo 104 Wo
96 Wo 88 Wo 80 Wo
60 No
57 No
54 No
51 No
48 No
45 No
120 Wo112 Wo 104 Wo
96 Wo 88 Wo 80 Wo
Figure 9 As Fig. 8, but showing composite precipitation anomalies for the 5 highest-flow years (1974, 1927, 1950, 1969 and 1978).
(a) Composite 500 mb geopotential height map, August to October (low flow yearsminus high flow years)
(c) Composite 500 mb geopotential height map, November to January (low flow yearsminus high flow years)
-0.2
0.2
-0.2
0.2
-0.6 -0.4 -0.2 0 0.2 0.4 0.6
(b) Correlation between the August to October (prior year) 500 mb geopotential heightfield and annual flow in theWinnipeg River at Slave Falls.
(d) Correlation between the November to January 500 mb geopotential height fieldand annual flow in the Winnipeg River at Slave Falls.
-0.6 -0.4 -0.2 0 0.2 0.4 0.6
-0.2
0.2
0.2-0.4
00
-0.1
-20
-20
-80 -60 -40 -20 0 20 40 60 80
40
-20
20
20
40
-40
0
20
-20
0
20
60
80-60
-80
-100
-80 -60 -40 -20 0 20 40 60 80
Figure 10 Comparisons between Winnipeg River streamflow and geopotential heights at the 500 mb level. In (a) and (c), anomalies(in metres) are calculated by subtracting the height anomalies for the five highest flow years from the five lowest flow years.Gridpoints where jrj is greater than 0.26 have correlations that are significant at the 0.05 level.
Streamflow in the Winnipeg River basin, Canada: Trends, extremes and climate linkages 409
Figure 11 Mean annual (October–September) streamflow in the Winnipeg River at Slave Falls. The series has been prewhitened,and transformed to anomalies by subtracting the 1908–2004 mean. The heavy line illustrates variability at timescales above 10years.
410 S. St. George
Conclusion
Long-term gauge records indicate that streamflow in theWinnipeg River basin increased significantly during last 80years, with winter streamflow going up by 60–110% overthe entire basin. Changes in annual and winter streamfloware observed in records from both regulated and unregu-lated portions of the watershed, which point to an underly-ing cause related to climate. Comparisons with long-termmeteorological records suggest that the basin hydrologyhas served to amplify coincident, but smaller, increases inprecipitation during summer and autumn.
Extreme low flows in the Winnipeg River are the productof a series of unusual hydroclimatic conditions within thewatershed, principally widespread reductions in precipita-tion during the previous summer and autumn, warm temper-atures in spring and summer, and very little runoff deliveredby spring snowmelt. These conditions are typically associ-ated with the circulation similar to the positive mode ofthe Pacific North American pattern, which may act to inhibitthe formation of cyclones in western North America, and re-duce the amount of precipitation received by northwesternOntario. A strong, positive PNA-like pattern during autumnand winter appears to be required in order for extremelow flows to occur, but positive PNA anomalies do not al-ways lead to low flows. The apparent relationship betweenstreamflow in the Winnipeg River basin and the PNA may beinfluenced by other competing or complimentary factors,and these interactions should be investigated in futurework.
Reports of declining flow for many rivers in the adjacentCanadian prairies (Westmacott and Burn, 1997; Yulianti andBurn, 1998; Rood et al., 2005; Dery and Wood, 2005) haveled to serious questions about the future reliability of watersupplies, especially because surface water is the primarysource of irrigation for local agriculture (Gan, 2000). Theseconcerns have been exacerbated by suggestions that a dou-bling of atmospheric CO2 could increase the severity andfrequency of droughts in the region (Burn, 1994). The resultsof this study show that hydrological trends in the Winnipeg
River basin during the 20th century are different from thoseobserved on other Canadian rivers, and imply that projec-tions made for the rivers in the Canadian prairies may notbe valid for this watershed. A recent review of results froman ensemble of climate models projects that runoff in theWinnipeg River region and northern and central Manitobawill increase 20–30% by the middle of the 21st century(Milly et al., 2005). Because of these projections, and theobserved trends toward higher flows, it seems likely thatthe potential threats to water supply faced by the CanadianPrairie provinces over the next few decades will not includedecreasing streamflow in the Winnipeg River basin.
Acknowledgements
Financial support for this research was provided by ManitobaHydro, the Manitoba Geological Survey, the Prairie Adapta-tion Research Collaborative and the Natural Sciences andEngineering Research Council of Canada. Dave Meko, KurtKipfmueller and Toby Ault graciously provided Matlabscripts used for parts of the analysis. Precipitation mapswere created using Rich Pawlowicz’s m_map toolbox. GregBrooks, Donald Burn and an anonymous referee reviewed anearlier version of this manuscript.
References
Angel, J.R., Isard, S.A., 1998. The frequency and intensity of GreatLake cyclones. Journal of Climate 11, 61–71.
Box, G.E.P., Jenkins, G.M., 1976. Time Series Analysis: Forecastingand Control. Prentice Hall, Englewood Cliffs, p. 575.
Burn, D.H., 1994. Hydrologic effects of climate change in westernCanada. Journal of Hydrology 160, 53–70.
Dery, S.J., Wood, E.F., 2005. Decreasing river discharge in northernCanada. Geophysical Research Letters 32, L10401. doi:10.1029/2005GL022845.
Gan, T.W., 2000. Reducing vulnerability of water resources ofCanadian prairies to potential droughts and possible climatewarming. Water Resources Management 14, 111–135.
Streamflow in the Winnipeg River basin, Canada: Trends, extremes and climate linkages 411
Hirsch, R.M., Slack, J.R., Smith, R.A., 1982. Techniques of trendanalysis for monthly water quality data. Water ResourcesResearch 19, 107–121.
Kalnay, E., Kanamitsu, M., Kistler, R., Collins, W., Deaven, D.,Gandin, L., Iredell, M., Saha, S., White, G., Woollen, J., Zhu, Y.,Chelliah, M., Ebisuzaki, W., Higgins, W., Janowiak, J., Mo, K.C.,Ropelewski, C., Wang, J., Leetmaa, A., Reynolds, R., Jenne, R.,Joseph, D., 1996. The NCEP/NCAR 40-year reanalysis project.Bulletin of the American Meteorological Society 77, 437–471.
Kistler, R., Kalnay, E., Collins, W., Saha, S., White, G., Woollen, J.,Chelliah, M., Ebisuzaki, W., Kanamitsu, M., Kousky, V., van denDool, H., Jenne, R., Fiorino, M., 2001. The NCEP-NCAR 50-yearreanalysis: Monthly means CD-ROM and documentation. Bulletinof the American Meteorological Society 82, 247–267.
Lake of the Woods Control Board, 2002. Managing the waterresources of the Winnipeg River drainage basin: Ottawa, Canada.20 pp.
Manitoba Hydro, 2004: 2003/04 Annual report and 2004/05 FirstQuarter Financial Results; Manitoba Hydro, News Release,August 11, 2003. Available from: <http://www.hydro.mb.ca/news/releases/news_04_08_11.shtml>.
Manitoba Hydro 2005. 2004/05 Annual Report. Available from:<http://www.hydro.mb.ca/about_us/ar_2004/ar_2004_report.shtml>.
Mauget, S.A., 2004. Low frequency streamflow regimes over thecentral United States: 1939–1998. Climatic Change 63, 121–144.
McCabe, G.J., Wolock, D.M., 2002. A step increase in streamflow inthe coterminus United States. Geophysical Research Letters 24,2185–2188.
Mekis, E., Hogg, W.D., 1999. Rehabilitation and analysis of Canadiandaily precipitation time series. Atmosphere-Ocean 37, 53–85.
Milly, P.C.D., Dunne, K.A., Vecchia, A.V., 2005. Global patterns oftrends in streamflow and water availability in a changingclimate. Nature 438, 347–350.
Natural Resources Canada, 2005. The Canadian Renewable EnergyNetwork. Available from: <http://www.canren.gc.ca/>.
New, M., Hulme, M., Jones, P., 1999. Representing twentieth-century space-time climate variability. Part I: Development of a
1961–90 mean monthly terrestrial climatology. Journal ofClimate 12, 829–856.
New, M., Hulme, M., Jones, P., 2000. Representing twentieth-century space-time climate variability. Part II: Development of1901–96 monthly grids of terrestrial surface climate. Journal ofClimate 13, 2217–2238.
Rogers, J.C., Coleman, J.S.M., 2003. Interactions between theAtlantic Multidecadal Oscillation, El Nino/La Nina, and the PNAin winter Mississippi Valley stream flow. Geophysical ResearchLetters 30, 1518–1521.
Rood, S.B., Samuelson, G.M., Weber, J.K., Wywrot, K.A., 2005.Twentieth-century decline in streamflows from the hydrolog-ical apex of North America. Journal of Hydrology 306, 215–233.
Salas, J.D., 1992. Analysis and modeling of hydrologic time series.In: Maidment, D.R. (Ed.), Handbook of Hydrology. McGraw-Hill,New York, pp. 19.1–19.72.
Vincent, L.A., Gullett, D.W., 1999. Canadian historical and homo-geneous temperature datasets for climate change analyses.International Journal of Climatology 19, 1375–1388.
Wallace, J.M., Gutzler, D.S., 1981. Teleconnections in the geopo-tential height field during the Northern Hemisphere winter.Monthly Weather Review 109, 784–812.
Westmacott, J.R., Burn, D.H., 1997. Climate change effects on thehydrologic regime within the Churchill-Nelson River Basin.Journal of Hydrology 202, 263–279.
Yue, S., Pilon, P., Phinney, B., Cavadias, G., 2002. The influence ofautocorrelation on the ability to detect trend in hydrologicalseries. Hydrological Processes 16, 1807–1829.
Yulianti, J., Burn, D.H., 1998. Investigating links between climaticwarming and low streamflow in the Prairies region of Canada.Canadian Water Resources Journal 23, 45–60.
Zhang, X., Harvey, K.D., Hogg, W.D., Yuzyk, T.R., 2001. Trendsin Canadian streamflow. Water Resources Research 37, 987–998.
