hydrogeologic controls on streamﬂow sensitivity to climate

HYDROLOGICAL PROCESSESHydrol. Process. 22, 4371–4385 (2008)Published online 16 June 2008 in Wiley InterScience(www.interscience.wiley.com) DOI: 10.1002/hyp.7041

Hydrogeologic controls on streamflow sensitivity to climatevariation

Anne Jefferson,1* Anne Nolin,2 Sarah Lewis2 and Christina Tague3

1 Department of Geography and Earth Sciences, University of North Carolina at Charlotte, 9201 University City Boulevard, Charlotte,NC 28223, USA

2 Department of Geosciences, Oregon State University, 104 Wilkinson Hall, Corvallis OR 97331, USA3 Bren School of Environmental Science and Management, University of California, Santa Barbara, CA 93106, USA

Abstract:

Climate models project warmer temperatures for the north-west USA, which will result in reduced snowpacks and decreasedsummer streamflow. This paper examines how groundwater, snowmelt, and regional climate patterns control discharge atmultiple time scales, using historical records from two watersheds with contrasting geological properties and drainageefficiencies. In the groundwater-dominated watershed, aquifer storage and the associated slow summer recession are responsiblefor sustaining discharge even when the seasonal or annual water balance is negative, while in the runoff-dominated watershedsubsurface storage is exhausted every summer. There is a significant 1 year cross-correlation between precipitation anddischarge in the groundwater-dominated watershed (r D 0Ð52), but climatic factors override geology in controlling the inter-annual variability of streamflow. Warmer winters and earlier snowmelt over the past 60 years have shifted the hydrograph,resulting in summer recessions lasting 17 days longer, August discharges declining 15%, and autumn minimum dischargesdeclining 11%. The slow recession of groundwater-dominated streams makes them more sensitive than runoff-dominatedstreams to changes in snowmelt amount and timing. Copyright 2008 John Wiley & Sons, Ltd.

KEY WORDS groundwater; snowmelt; climate variability and change; low flows; Cascades; Oregon

Received 10 January 2007; Accepted 27 February 2008

INTRODUCTION

The mountains of the Cascade Range provide criticalwater supplies for agriculture, municipalities, and ecosys-tems in the Pacific Northwest region of the USA. Recentanalyses show that this region is sensitive to current andprojected climate warming trends, specifically reducedsnow accumulation and earlier spring melt, leading to adecline in summer streamflow. Temperature trends overthe past half-century indicate warmer winters across thewestern USA (Folland et al., 2001) and particularly in thePacific Northwest (Mote, 2003). These data are corrob-orated by ample proxy evidence of recent temperatureincreases in the Pacific Northwest (Cayan et al., 2001;Regonda et al., 2005; Stewart et al., 2005). Snowpacksin western North America have declined over the past50 years, primarily due to an increase in winter tem-peratures (Mote et al., 2005). In accordance with earlierspring snowmelt, streamflow timing is earlier by 1 to4 weeks, compared with the middle of the 20th century(Stewart et al., 2005). Climate models predict continuedwinter warming of 0Ð2 to 0Ð6 °C per decade in the PacificNorthwest (Christensen et al., 2007; Mote et al., 2003;Washington et al., 2000). By 2050, Cascade snowpacksare projected to be less than half of what they are today

* Correspondence to: Anne Jefferson, Department of Geography andEarth Sciences, University of North Carolina—Charlotte, 9201 Univer-sity City Boulevard, Charlotte, NC 28223, USA.E-mail: [email protected]

(Leung et al., 2004), potentially leading to major watershortages during the low-flow summer season. Regional-scale analyses identify climatic gradients as primary con-trols on changing streamflow regimes, but the poten-tial for other hydrologic factors, notably groundwater, toinfluence this response has received much less attention.

Snowpack is the dominant water storage compartmentin many watersheds, and snowmelt is quickly translatedinto streamflow. In the western USA, where there islittle summer rainfall, the snowmelt peak is often theyear’s highest discharge event, and marks the beginningof a long decline in streamflow through the summermonths. Groundwater systems can mediate the translationof water from snowpack to streamflow by providing anadditional storage compartment and damping variationsin precipitation and recharge. Extensive groundwatersystems are often found where the bedrock is basalt,limestone, or fractured rock.

The object of this work is to develop an understand-ing of how discharge is controlled at the event, sea-sonal, and inter-annual scales by the interactions amonggroundwater storage, snowmelt dynamics, and regionalclimate signals. The over-arching goal is to better pre-dict hydrologic responses to future climate change ingroundwater-dominated watersheds of the western USA.The focus of this research is an analysis of the his-torical discharge, precipitation, and snowpack recordsfrom two adjacent watersheds on the western slope ofthe Oregon Cascades. One watershed has an extensive

Copyright 2008 John Wiley & Sons, Ltd.

4372 A. JEFFERSON ET AL.

groundwater system (McKenzie River at Clear Lake),while the other watershed is runoff-dominated with lit-tle groundwater storage (Smith River). Results from thestudy watersheds should be broadly applicable to geo-logically similar watersheds in the Oregon Cascades andnorthern California.

Most watersheds on the western slope of the Ore-gon Cascades encompass elevations that receive winterprecipitation as a mixture of rain and snow. These water-sheds have complex winter hydrographs that are depen-dent on the distribution of rain and snow during individ-ual events, which in turn is controlled by storm temper-atures and catchment hypsometry. Snow cover typicallyaccumulates at temperatures close to the melting point,and thus is at risk from climate warming because temper-ature affects both the rate of snowmelt and the phase ofprecipitation. With a projected 2 °C winter warming bymid-century, 9200 km2 of currently snow-covered areain the Pacific Northwest would receive winter rainfallinstead (Nolin and Daly, 2006).

Summer and autumn discharges are of interest becausethey occur during periods of low flows, high waterdemand, and critical temperatures and flows for salmonidsand other fish species. Decreases in summer streamflowsrestrict junior water rights users, constrain hydropowergeneration, and make in-stream flow targets difficultto meet. Lowered summer discharge can also intensifystream temperatures warmed by hot summers (Webbet al., 2003), and some species are very temperature-sensitive (Selong et al., 2001), such as the threat-ened bull-trout that inhabits the McKenzie River water-shed. Decreased minimum flows have also been linkedto drought stress and mortality in riparian vegetation(Naiman et al., 1998). Regional climate models predictthat Pacific Northwest summers will become hotter anddrier over the next century (Christensen et al., 2007),exacerbating existing stresses.

Spatial patterns of summer streamflow in the Ore-gon Cascades exhibit significant differences betweenthe geologically-distinct High and Western Cascadeprovinces (Tague and Grant, 2004). A key control onthese differences is the partitioning of water to slow-draining basaltic aquifers in the High Cascades, versusshallow subsurface storm flow through the upper fewmetres of soil and bedrock in the Western Cascades. Saarand Manga (2004) estimated a groundwater recharge rateof 100 cm yr�1 for the High Cascades, while Ingebritsenet al. (1994) estimated a recharge rate of 3 cm yr�1 forthe Western Cascades bedrock aquifers. High Cascadesaquifer volumes are seven times greater than annual dis-charge (Jefferson et al., 2006). Groundwater discharge toHigh Cascades streams also has a moderating effect onwater temperatures, resulting in lower diurnal variabilityand colder summer temperatures (Tague et al., 2007).

Tague et al. (2008) used the hydro-ecological modelRHESSys to examine the influence of geology on stream-flow response to 1Ð5 °C and 2Ð8 °C warming scenar-ios. Their study included the High Cascades watershedused in this project, McKenzie River at Clear Lake.

Their model showed that warmer temperatures resultedin greater reductions in August discharge and annualminimum flows for the High Cascades than the WesternCascade watershed, both in terms of absolute volumesand normalized by drainage area. The Western Cascadestream, however, showed greater relative reductions inthese summer streamflow metrics. Model results illustratethat differences between the responses of the two siteswere primarily due to differences in groundwater flow, asmanifested in drainage efficiency of the watersheds. Spa-tial differences in recharge characteristics and the timingof snow accumulation and melt were shown to be impor-tant, but secondary, in terms of explaining responses atthe two sites. Further, they found that in the 2Ð8 °C warm-ing scenario, August discharge of the McKenzie River atClear Lake was reduced 23% from current mean values.

This study complements the previous model-basedanalysis through in-depth analysis of meteorological andstreamflow records in order to more fully characterizeclimate–snowmelt–groundwater interactions for the Ore-gon Cascade region. To gain insight, analyses of dis-charge dynamics at the event, seasonal, inter-annual, anddecadal time scales are presented utilizing time-seriesanalyses and correlations between parameters. Whiledynamics are explored during all seasons, the focus isparticularly on late summer streamflow because of itsimportance for water resources. Through these analyses,the primary and secondary controls on discharge variabil-ity are identified at each time scale. This will facilitatemodel development that appropriately factors in the influ-ence of geologic setting and in order to better predict theeffects of future climate variability and change.

STUDY WATERSHEDS

Study watersheds were selected on the basis of theirlocation at rain–snow transitional elevations, availabilityof historical discharge and snow records, absence ofstream regulation or diversion, and contrasting geology.The two study watersheds adjoin each other in the upperMcKenzie River watershed on the west side of theOregon Cascades (Figure 1). The US Geological Survey(USGS) gauge for the McKenzie River at Clear Lakeis located at N 44° 220, W 122° 00. The watershedupstream from that point (the Clear Lake watershed)covers 239 km2 and ranges in elevation from 918 to2051 m, with a weighted average elevation of 1215 m.The USGS gauge at Smith River above Smith RiverReservoir is located at N 44° 200, W 122° 020, withan area of 42 km2 and elevations ranging from 803to 1753 m; weighted average elevation is 1216 m. TheSmith River is tributary to the McKenzie River 16 riverkm downstream of Clear Lake and 6 km downstream ofthe Smith River gauge. The McKenzie River provideswater and electricity for Eugene, the second largest cityin Oregon, supports a major recreational economy, andsustains superb salmonid habitat. The McKenzie Riveris a major tributary to the Willamette River, one of thelargest rivers in the USA.

Copyright 2008 John Wiley & Sons, Ltd. Hydrol. Process. 22, 4371–4385 (2008)DOI: 10.1002/hyp

HYDROGEOLOGIC CONTROLS ON STREAMFLOW SENSITIVITY TO CLIMATE VARIATION 4373

Figure 1. Map showing the study watersheds relative to their location in Oregon and to Cascades topography and geology. On the Oregon map, theletter ‘a’ next to a gauge symbol indicates the position of the stream gauge for South Santiam River below Cascadia, ‘b’ indicates Smith River above

Smith River Reservoir, ‘c’ indicates McKenzie River at Clear Lake, and ‘d’ indicates McKenzie River at McKenzie Bridge

The study watersheds lie within the Cascades volcanicarc, which includes two distinct geologic provinces: theHigh Cascades and the Western Cascades. The HighCascades are known for their active composite volca-noes and extensive Quaternary basaltic lavas, while theWestern Cascades are the products of Tertiary arc vol-canism, and have been extensively folded, faulted, andweathered (Priest et al., 1983). High Cascade lava flowshave high permeability, resulting in extensive ground-water systems, low slopes, and low drainage densities(Jefferson et al., 2006), whereas the Western Cascades aredeeply dissected with well-developed drainage networksand shallow subsurface storm flow (Harr, 1977). We usethe mapping of Sherrod and Smith (2000) to define theHigh Cascades as including all Quaternary igneous rocksand glacial till, while the Western Cascades includes allTertiary igneous rocks. The Clear Lake watershed is 66%High Cascades, 25% Western Cascades, and 9% lakesand alluvium. The Smith River watershed is composedentirely of Western Cascades geology.

The study watersheds are largely forested, althoughcover is patchy on the youngest lava flows in theClear Lake watershed. Forests include stands of Douglas

fir (Pseudotsuga menziesii ), noble fir (Abies procera),Pacific silver fir (Abies amabilis), and western hemlock(Tsuga heterophylla). Wet and dry meadows are alsofound throughout the watersheds. Land in the water-sheds is primarily within the Willamette National Forest,although there are small in-holdings of private forestryland in the Clear Lake watershed. Forest harvest hasoccurred throughout the past 50 years in both watersheds,but less than 15% of either watershed was harvested orin an early regenerative stage in any year during theperiod. Both watersheds lie within the McKenzie Riverat McKenzie Bridge watershed, which has been deemedsuitable for studying the effects of climate on waterresources and is part of the Hydro-Climatic Data Network(Slack et al., 1993).

Daily discharge data for the study watersheds areavailable from the USGS streamflow gauges. For theMcKenzie River at Clear Lake, data are available forthe 1913–1915 water years (water years begin Octo-ber 1) and continuously since October 1947 (gauge14 158 500) (Herrett et al., 2005). Summer flow in theMcKenzie River at Clear Lake consists largely of dis-charge from springs along the lakeshore. Additional



wet season discharge includes direct precipitation ontothe 0Ð6 km2 lake, seasonal runoff from Fish Lake, andperennial runoff-dominated flow from Ikenick Creek(Figure 1). Based on occasional discharge measurementsfrom August 2003 to September 2005, Ikenick Creek is avery minor contributor of summer flow to the McKenzieRiver at Clear, ranging from 3% on 6 June 2005 to 0Ð03%on 20 September 2005. Runoff from the creeks is prob-ably much more significant during snowmelt periods. ForSmith River (gauge 14 158 790), USGS daily dischargedata are continuously available since October 1960 (Her-rett et al., 2005). Smith River is runoff-dominated withno lakes or mapped springs.

There are two Natural Resources Conservation Servicesnowpack telemetry (SNOTEL) sites within the ClearLake watershed that collect real-time, automated mea-surements of snow water equivalent (SWE), precipitation,and temperature (http://www.wcc.nrcs.usda.gov/snow/).The Hogg Pass SNOTEL is located at 1451 m and San-tiam Junction SNOTEL is at 1143 m. A third SNOTELsite, Jump Off Joe (1067 m), lies <5 km west of the studywatersheds. This site was included in the analyses to pro-vide additional information about low elevation areas.Approximately 20% of the Clear Lake watershed has anelevation higher than Hogg Pass, 63% is higher than San-tiam Junction, and 83% is higher than Jump Off Joe. ForSmith River, 12%, 64%, and 83% of the watershed arealies above the three SNOTEL stations, respectively.

Annual precipitation ranges from 1000 to 3200 mm,and 70% falls between November and March, basedon the historical data from the SNOTEL stations. Thewatersheds receive a mixture of rain and snow at allelevations. At Hogg Pass, 56% of annual precipitationaccumulates as snow, whereas at Santiam Junction it is37%, and at Jump Off Joe it is 25%. From 400 to 1200 melevation in the Cascades, snowpacks may accumulateand melt several times in the winter season and rain-on-snow events are common (Harr, 1981). Above thiselevation, a seasonal snowpack of 2 to >7 m accumulatesthrough the winter months and melts over the periodMarch through June. In forested areas of the Cascades,up to 60% of falling snow can be intercepted by theforest canopy, but is later released by melt drip or fallingsnow masses. Sublimation accounts for only a minorcomponent of ablation (Storck et al., 2002). Peak snowaccumulation occurs around 1 April at Hogg Pass, andaround 1 March at Santiam Junction and Jump Off Joe.The highest monthly stream discharges at both Clear Lakeand Smith River have historically occurred in May, dueto snowmelt, but individual peak flows may be triggeredby rain or rain-on-snow earlier in the water year.

DATA COMPILATION AND ANALYSIS

Daily time series datasets were compiled for discharge,precipitation, snow water equivalent, and temperaturefrom the available USGS and SNOTEL records. Vary-ing record lengths, as described below, were used for

the analyses. Quality control on SNOTEL records wasperformed by the Oregon Climate Service (Daly, pers.commun., 2005). The Santiam Junction and Jump OffJoe SNOTEL sites precipitation and SWE records extendfrom water year 1979 through 2005, while for Hogg Passprecipitation and SWE records are from water years 1980through 2005. Temperature records for Hogg Pass andJump Off Joe span water years 1985–2003. For SantiamJunction, temperature records span 1983–2003. Reportedminimum and maximum daily temperatures were aver-aged to generate a single daily value for temperature.

To understand the relationship between climatologicaland hydrological variables in space and time, Pearson’scorrelations, and auto- and cross-correlations were per-formed. Annual values of 49 parameters were derivedfrom the discharge, precipitation, snow water equivalent,and temperature time series (Table I). Pearson’s prod-uct moment correlations were calculated for 41 param-eters, and significance was determined through F-testsand a 95% confidence limit. Daily values of discharge,precipitation, SWE, and recharge (rain C snowmelt) for1979–2005 were normalized using daily means, andauto- and cross-correlations were computed followingBox and Jenkins (1976). A similar analysis was con-ducted for annual time-series. Auto- and cross-correlationprocedures generate correlation coefficients for data pairsseparated by variable time lags. Auto-correlation isa measure of correlation between values in a singleseries, while cross-correlation is a measure of correlationbetween values in two different series. Confidence inter-vals for auto-correlation and cross-correlation coefficientswere calculated using the effective number of obser-vations following procedures outlined in Haan (2002)and Kite (1977). Auto- and cross-correlations were con-sidered to be significant if they fell outside the 95%confidence interval around the null hypothesis of no cor-relation.

Monthly values of the Nino 3Ð4 Index of sea sur-face temperature (Trenberth and Stepaniak, 2001) andthe Southern Oscillation Index of surface pressure(Ropelewski and Jones, 1987) over the period 1978–2004were used to explore correlations between El Nino/Southern Oscillation (ENSO) with annual discharge andsnow water equivalent. Discharge and SWE were alsocorrelated with the Pacific Decadal Oscillation (PDO)data set from the University of Washington (Mantuaet al., 1997).

Simple water balances for the Clear Lake and SmithRiver watersheds were constructed for the 2001–2004water years to examine the magnitudes and seasonalvariations in water fluxes and stores. Discharge was cal-culated at each USGS gauge, and precipitation, SWE, andrain plus snowmelt at the areally weighted average eleva-tion (1215 m) were derived by linear interpolation fromvalues at Jump Off Joe and Hogg Pass. As a check on theappropriateness of linear interpolations of seasonal pre-cipitation and rain plus melt, results were compared withobserved values at Santiam Junction. Additionally, water



Table I. Parameters derived from hydrological and climatic data. P indicates parameters for which Pearson’s product momentcorrelations were calculated. R indicates parameters which were used in regression models of autumn minimum discharge. DOWY

stands for day of water year, and DOY stands for day of calendar year

Parameter Units Clear Lake Smith River Hogg Pass Santiam Jct. Jump Off Joe

Water year year P P P P PMean discharge for water year m3 s�1 P PMean discharge for December–August m3 s�1 R RMean August discharge m3 s�1 P PAutumn minimum discharge m3 s�1 PR PRDate of autumn minimum discharge DOYTemporal centroid DOWY P PMean discharge for previous December–August m3 s�1 R RAutumn minimum discharge of previous year DOY R RTotal precipitation in water year mm PR PR PRAccumulated precipitation on 1 April mm P PR PAccumulated SWE mm PR PR PRSWE present on April 1 mm PR PR PRDate of Peak SWE DOWY PR PR PRPeak SWE mm P P PLast date with SWE DOWY P P PTemporal centroid of rain plus melt DOWY PR PR PAverage winter temperature (November–March) °C P P PAverage spring temperature (April–June) °C P P PAverage summer temperature (July–October) °C P P P

year 2004 precipitation was compared with spatially-distributed PRISM model outputs (Daly et al., 1997,2002). Basin-averaged evapotranspiration (ET) was cal-culated in RHESSys, a physically-based hydro-ecologicalmodel previously calibrated for the Clear Lake watershed(Tague et al., 2008). ET estimates in RHESSys accountfor canopy transpiration, evaporation of canopy intercep-tion, litter/soil evaporation and snow sublimation losses.ET losses are calculated using Penman, for evaporativelosses, and Penman–Monteith, for transpiration losses.Additional details on the RHESSys modeling approachare available in Tague and Band (2004). Watershed-averaged lake evaporation was estimated by multiply-ing the lake-covered area in the Clear Lake water-shed (<2 km2) by the evaporation rate of Crater Lake(1200 mm yr�1; Redmond, 1990), located at 1883 m ele-vation in the Oregon Cascades. The resulting rate was<10 mm yr�1 for the Clear Lake watershed, and theSmith River watershed has no lakes, so lake evaporationwas neglected in water budget calculations.

Predictive models of September–November minimumdischarge using available climatic and hydrologic mea-surements were developed using stepwise regression inSAS 9Ð1 (Table I). Separate models were developed forthe Clear Lake and Smith River watersheds. The mod-els were based on data from the 1979–2005 water years,and stepwise selection of parameters proceeded until allvariables in the model were significant at the 0Ð05 leveland no other variables met the 0Ð05 significance levelfor entry into the model. Colinearity was minimized bymanual elimination of strongly correlated variables.

Discharge records were investigated for secular trendsin August discharge, minimum discharge, and tempo-ral centroid. Less than 50 year record lengths for ClearLake and Smith River watersheds may lead to ambiguous

conclusions about the presence or absence of trends. Tosupport these records, discharge data from two water-sheds with longer historical records were also examined.The McKenzie River at McKenzie Bridge (USGS gauge14 159 000) drains 901 km2, including the Clear Lake andSmith River watersheds. It has a weighted average ele-vation of 1286 m, and a gauge elevation of 432 m. It is56% High Cascade geology, and has a daily dischargerecord that extends from August 1910 to October 1995.A hydroelectric project has been in operation upstream ofthe gauge since 1963, but the reservoirs are not operatedfor flood control or low-flow supplementation. The SouthSantiam River below Cascadia (USGS gauge 14 185 000)drains 451 km2, immediately to the west of the studywatersheds. It has a weighted average elevation of 875 m,and a gauge elevation of 236 m. The watershed is 5%High Cascade geology, and it has daily discharge datafrom September 1935 to the present. Both watershedsare part of the Hydro-Climatic Data Network, a set ofdischarge records that have been deemed suitable forstudying the effects of climate on water resources (Slacket al., 1993).

EVENT AND SEASONAL SCALE VARIABILITY

The groundwater-dominated Clear Lake watershedexhibits a muted hydrograph response relative to theadjacent Western Cascade, runoff-dominated Smith Riverwatershed (Figure 2). Base flows are higher, and peakflows are less frequent, smaller, and somewhat delayedrelative to Smith River. Peak flows are generated by rain-on-snow, snowmelt, and rain events, with rain-on-snowevents typically having sharp, isolated peaks. For exam-ple, on 7–9 January 2002 (event A in Figure 2), 53 mm



Figure 2. Water years 2001–2002 Santiam Junction precipitation and unit discharge hydrographs for the McKenzie River at the outlet of ClearLake and Smith River above Smith River Reservoir. Event A corresponds to 7–9 January 2002; event B corresponds to 1–23 April 2002; event C

corresponds to 21 October 2000; and event D corresponds to 15–16 May 2001. Events are discussed in the text

of precipitation fell at Santiam Junction yielding an addi-tional 44 mm of melt. The McKenzie River at Clear Lakehad a peak flow twice the pre-event flow, while SmithRiver peak flow was six times greater than before theevent. Snowmelt events are generally sustained, build-ing gradually to their peak flows, often with intermediatepeaks. For example, between 1 and 23 April 2002 (eventB), 188 mm of precipitation fell at Santiam Junction and564 mm of snow melted. The peak flow of the McKen-zie River at Clear Lake was four times greater than thepre-event discharge, while at Smith River it was eighttimes greater. Flows twice as large as the pre-event dis-charge were sustained for 14 days in the McKenzie Riverat Clear Lake and 6 days in the Smith River. The mag-nitude of response to rain events depends largely onthe antecedent wetness in the watershed. For example,¾60 mm rain events in 21 October 2000 and 15–16 May2001 (events C and D) produced very different hydro-graphs.

Annual unit discharge of the McKenzie River at ClearLake and the Smith River are comparable, but theSeptember unit discharge at Clear Lake is over six timeshigher than at Smith River, as illustrated in Figure 2.The ratio of May to September monthly discharge atClear Lake is only 3Ð3, while for the Smith River itis 39Ð8. Tague and Grant (2004) provide an extendeddiscussion of the low flow characteristics of High andWestern Cascade streams and attribute the slow summerrecession of High Cascade streams to their groundwaterinputs.

Clear Lake and Smith River have very different auto-correlation structures (Figure 3a). The auto-correlationcoefficient for Clear Lake is statistically significant atthe 95% confidence level for 301 of the first 366 dailylags, whereas at Smith River statistically significant auto-correlations are observed for only 36 of the first 366 days,and recharge time series derived from rain and snowmeltat Hogg Pass has statistically significant correlations for

only 22 days. These results suggest that both runoff-dominated and groundwater-fed stream systems havedamped discharge variation relative to the input variationas estimated from a single SNOTEL site, reflecting thedistribution of response times of shallow subsurfaceand groundwater flow and in the channel network. Theprevalence of longer response times associated withgroundwater systems of the Clear Lake watershed resultin substantial smoothing of the hydrograph and sustainedhigh auto-correlation values, relative to the faster shallowsubsurface flow, runoff-dominated Smith River.

The maximum cross-correlation between discharge ofClear Lake and Smith River occurs at a one-day lag(Figure 3b), as do the peak cross-correlations of ClearLake discharge and recharge at Santiam Junction andJump Off Joe. These results suggest that the ClearLake watershed experiences a minimal lag in dischargerelative to the recharge time series or a runoff-dominatedstream. This result is in stark contrast to the results ofManga (1999), who found a lag of 47–137 days forfour snowmelt-dominated spring-fed streams on the eastside of the Oregon Cascades compared to neighbouringrunoff-dominated streams. Tague and Grant (2004) reportlags averaging 30 days in their analysis of Western versusHigh Cascade streams. However, their analysis did notnormalize the datasets to remove seasonal effects. Laterwork by Tague et al. (2008) show that discharge fromClear Lake includes both spring and shallow subsurfaceflow contributions. Thus, short lag times reflect the ‘fast’shallow subsurface component of the Clear Lake system.

WATER BUDGET

Water budgets, constructed as described earlier, revealdivergent seasonal patterns of water storage and fluxfor the runoff- and groundwater-dominated watersheds.Over the 2001–2004 period, precipitation equalled 99%



Figure 3. Auto- and cross-correlations of time-series data. Auto-correlation time series for (a) Clear Lake discharge, (b) Smith River dis-charge, and (c) Hogg Pass recharge (rain plus snowmelt). Correlations areconsidered to be significant if they fell outside the 95% confidence inter-val (dashed lines) around the null hypothesis of no correlation. (d) Cross-correlation time series for Clear Lake discharge versus Smith Riverdischarge, Santiam Junction recharge, and Jump Off Joe recharge. Allcross-correlations shown on the graph are outside the 95% confidenceintervals of 0Ð0551 for Smith River, 0Ð0303 for Santiam Junction, and

0Ð0292 for Jump Off Joe

of discharge plus ET in the Clear Lake watershed and80% of discharge plus ET in the Smith River watershed(Figure 4). In water year 2004, linear interpolation esti-mated precipitation as 2Ð14 m, and PRISM precipitationfor the Clear Lake watershed is 2Ð11 m. In the SmithRiver watershed, PRISM precipitation was 2Ð49 m for2004, 18% higher than in the Clear Lake watershed. Thisresult suggests that the linear interpolation method cor-rectly estimated precipitation in the Clear Lake watershedand underestimated precipitation by ¾20% in the SmithRiver watershed. This difference accounts for the closure

of the annual water budget in the Clear Lake water-shed and the discrepancy in the Smith River watershed.The interpolations based on linear regression from HoggPass and Jump Off Joe over-predict the Santiam Junctionrecord. This was expected because Santiam Junction liesin the rain shadow of the Western Cascades and oftenbecomes snow-free earlier than Jump Off Joe.

The year was divided into three seasons based on thestatus of the snowpack. November through March consti-tutes the snow accumulation season; April through Juneis the snow ablation season; and July through Octoberis the snow-free season. On average for 2001–2004,in the snow accumulation season, 35% of precipitationwas stored as seasonal snowpack (450 mm), which wasmelted during the snow ablation season. Thus, despitethe accumulation season having 350% of the precipita-tion of the ablation season, available water amounts werenearly equal in the two periods. Between 2001 and 2005,snowpack storage had varying importance for runoff,depending on the size of the snowpack, timing of themelt, and the amount of spring rain. In 2002, snowmeltwas 137% of April through June Clear Lake runoff, butin 2005, it was only 29% owing to a warm dry wintercombined with a wet spring.

For the Clear Lake watershed, during the 2001–2004period, 71% of available water during the snow accu-mulation season was accounted for as runoff, and5% was lost to ET (Figure 4). The remaining 24%(¾200 mm), probably replenished the soil moisture sup-ply and groundwater storage. During the snow ablationseasons in the Clear Lake watershed, snowmelt and rainsupplied enough water to account for runoff and ET andprovided ¾125 mm to groundwater storage. In the snow-free season in the Clear Lake watershed, runoff and ETwere approximately equal and accounted for 260% ofprecipitation. Groundwater storage diminished to sustainstreamflow and supply water to vegetation.

For the Smith River watershed, available water dur-ing all seasons was increased by 20% over the ClearLake watershed, as suggested by the PRISM outputs(Figure 4). During the 2001–2004 snow accumulationseasons, all available water was used for runoff and ET,with no groundwater recharge occurring. In the snowablation seasons, ¾160 mm of water was available torecharge groundwater and soil moisture, after runoff andET were accounted for. In the snow-free seasons, runoffaccounted for ¾20% of precipitation and ET required130%. The ET flux was six times as great as the runoffflux, suggesting that recharge during the snow ablationseason was largely to the plant-available soil zone ratherthan to bedrock groundwater.

INTER-ANNUAL VARIABILITY

Generally, discharge, precipitation and SWE vary to-gether throughout the period of record and among thethree SNOTEL sites (Table II). Mean annual dischargesat Clear Lake and Smith River are strongly correlated



Figure 4. Water budget for Clear Lake (CL) and Smith River (SR) watersheds for the average of water years 2001–2004. The rain plus transientsnow category includes all liquid precipitation and snow that falls and ablates within the same season. Seasonal snow falls and is stored during the

accumulation season, and it melts and contributes to runoff during the ablation season. Inputs for SR are 120% of inputs for CL

with annual precipitation at all three sites (r > 0Ð88), butonly at Hogg Pass is there strong correlation betweenaccumulated snow water equivalent (SWE) and mean dis-charge (r D 0Ð74 for Clear Lake). The two watershedsshow similar levels of variability in mean annual dis-charge, with the coefficient of variation (CV) equal to0Ð25 for Clear Lake and 0Ð27 for Smith River. However,variability in August discharge is smaller for Clear Lake(CV D 0Ð32) than for Smith River (CV D 0Ð36). Augustdischarge at Clear Lake is more strongly correlated withprecipitation and SWE parameters than is August dis-charge at Smith River, which is more strongly tied towinter and spring temperatures. This suggests that sum-mer streamflow in the groundwater-fed watershed is morestrongly linked to hydroclimatic conditions than summerstreamflow in the runoff-dominated watershed, wheretemperature-modulated ET demand may have more influ-ence. Winter temperature was most strongly correlatedwith accumulated SWE at Jump Off Joe (r D �0Ð71),which lies within the transient snow zone and is stronglyaffected by temperature-induced phase changes in precip-itation. The amount of peak SWE and the date at which itis reached are only weakly correlated for any individualsite (r < 0Ð55).

There is an inter-annual memory in the Clear Lakewatershed as a result of the High Cascades groundwatersystem. Discharge at Clear Lake is moderately cross-correlated with the previous year’s precipitation (r D0Ð52 for Hogg Pass), and the cross-correlation is higherthan the 1-year discharge auto-correlation (r D 0Ð45).Both of the above correlations for Clear Lake dischargeare higher than those for Smith River discharge with thatof the previous year or the 1-year lagged auto-correlationof precipitation, which are not statistically significant. At2 year lags and beyond all auto- and cross-correlationsare not statistically significant.

The relationships between two major climate indicesthat have been correlated with streamflow in the PacificNorthwest were examined and quantified: El Nino/Southern Oscillation Index and the Pacific Decadal Oscil-lation (Beebee and Manga, 2004). Correlations betweenSWE and the monthly Nino 3Ð4 index for 1977–2000 arehighest for December, and are stronger for the lower ele-vation SNOTEL sites. Hogg Pass SWE had a correlationof �0Ð54 with December Nino 3Ð4, while the correla-tion at Santiam Junction is �0Ð60, and that at Jump OffJoe is �0Ð63. These finding suggest that the temperatureshifts associated with the ENSO cycle are more impor-tant than the precipitation signal in determining snowaccumulation in the transient snow zone. There are nosignificant correlations with annual or August dischargein either watershed. These results indicate that ENSO, asexplained by the Nino 3Ð4 index, is a moderate predic-tor of SWE, but not of discharge. Correlations were alsoexplored using the SOI but it was found that the corre-lations are much lower than for the Nino 3Ð4 index. Itwas found that there are no significant correlations withannual discharge, August discharge or station SWE.

These results contrast with those of Beebee and Manga(2004), who found inverse correlations between annualdischarge, peak runoff and ENSO indices for eightsnowmelt-dominated watersheds in central and easternOregon. However, the current study area has a higherproportion of rain versus snow than the watersheds theystudied, and it is also substantially more humid. Theyalso used the SOI averaged over June–September andNino 3Ð4 averaged over September–November. Individ-ual monthly values of the indices were used here ratherthan seasonal aggregates to better understand the timelags between discharge and the climate indices. Lowercorrelations were found for seasonally averaged values ofboth SOI and Nino 3Ð4. Like Beebee and Manga [2004],no significant correlation with PDO was found.



AUTUMN MINIMUM DISCHARGE

The presence of groundwater also strongly influences thelowest flows of the year.

Annual minimum discharge occurred during Septem-ber through November during 52 of 64 years of theClear Lake record and 41 of 45 years of the Smith riverrecord. In the remaining years, annual minimum dis-charge occurred later in the winter, probably as a resultof snowy or dry autumns that didn’t recharge the streamand prolonged the recession. For consistency of analyses,the minimum daily discharge between 1 September and30 November was determined, henceforth called autumnminimum discharge (Qmin). On a unit area basis, Qmin

is ¾9 times greater in the Clear Lake watershed thanthe Smith River watershed. As discussed above, thegreater Qmin in the groundwater-dominated watershed isthe result of a slower summer recession. Clear Lake Qmin

are also more strongly influenced by inter-annual fluc-tuations in precipitation (Figure 5). The CV for ClearLake Qmin is 0Ð26, while for Smith River it is 0Ð22. It isproposed that subsurface storage is nearly exhausted inthe Smith River watershed every year, whereas Qmin inthe Clear Lake watershed are sensitive to aquifer levels,which in turn are controlled by the current and previ-ous year’s precipitation inputs. Using the previous year’sDecember–August mean discharge at Clear Lake as aproxy for precipitation, the Pearson correlation with ClearLake Qmin is 0Ð5.

Stepwise multiple linear regression was performed onthe series of 1979–2005 Qmin for Clear Lake and SmithRiver, using 18 potential predictor variables for eachwatershed (Table I). Stepwise multiple linear regressionfor Smith River yielded a model using the accumulatedsnow water equivalent at Santiam Junction (SSJ, mm) thatexplains 38Ð1% of the variation in the dataset. No othervariables met the 0Ð05 significance level for entry into themodel. The low association between Smith River Qmin

and hydroclimatic variables supports the contention thatsubsurface storage in the watershed is nearly exhaustedeach year, and thus, insensitive to climatic variability.

0.04

0.03

0.02

0.01

0.001.0 1.5 2.0 2.5 3.0 3.5

Uni

t Dis

char

ge (

m3

s-1 k

m-2

)

Annual precipitation (m)

Clear Laker2 = 0.77

Smith Riverr2 = 0.25

Figure 5. Relationship between annual precipitation at Santiam Junctionand autumn minimum unit discharge for the Clear Lake and Smith River

watersheds

In contrast to the results for Smith River, the bestmodel for Clear Lake Qmin uses four variables to explain93Ð4% of the variation in the dataset. The predictivevariables are mean discharge from December–August(QMean, m3 s�1), the previous year’s autumn minimumdischarge (QPrevMin, m3 s�1), SSJ, and the day of thecalendar year on which autumn minimum dischargeoccurs (A). The regression equation is:

Qmin D 0Ð288 ð Qmean C 0Ð00175 ð SSJ C 0Ð292

ð QPrevMin � 0Ð0197 ð A C 4Ð73 �1�

The high correlation between precipitation and meandischarge suggests that the discharge term represents theyear’s precipitation. The snow term is probably related tothe timing of snowmelt and recharge to the aquifers. Theprevious minimum discharge term represents the inter-annual memory in the groundwater system and the stateof the aquifer in the previous year. The autumn minimumflow date term represents the timing of the first majorautumn rains that bring an end to the summer recession.Thus, Qmin is shown to be a function of subsurfacestorage, the year’s precipitation, timing of snowmelt, andtiming of the fall rains. Even when A is not included, theother three variables explain 90% of the variation in thehistorical record, enabling Qmin to be predicted by theend of August, using

Qmin D 0Ð312 ð Qmean C 0Ð00174 ð SSJ C 0Ð260

ð QPrevMin � 1Ð28. �2�

SECULAR TRENDS

Investigation of secular trends in climatic and hydrographparameters suggests that inter-annual variability maskspotential trends in precipitation and SWE data derivedfrom the <30 year SNOTEL record. The only SNOTEL-derived snow or precipitation parameter to exhibit astatistically significant relationship with time is the dateof last snow cover at Santiam Junction. There is a weak(r D �0Ð39) trend towards earlier loss of snow cover atthis site, with a slope of 7Ð8 days per decade. There are nostatistically significant trends at either the higher or lowerelevation sites. Santiam Junction may be particularlysusceptible to warming-induced earlier snow cover loss,because it lies near the lower limits of seasonal snowcover. Future warming may result in much more transientsnow cover interspersed with winter rain.

Longer discharge records suggest that there are somesecular trends, although the signal is still dominated byinter-annual variability. The hydrograph temporal cen-troid is the day of the water year when half of the annualdischarge has occurred, and has been used as an indica-tor of climate change throughout the mountainous west-ern USA (Regonda et al., 2005). It was found that the



Tabl

eII

.C

orre

latio

nco

effic

ient

sfo

rse

lect

edva

riab

les.

Bol

dte

rms

are

stat

istic

ally

sign

ifica

ntat

the

95%

confi

denc

ele

vel.

Mea

n,A

ugus

t,an

dau

tum

nm

in.

are

daily

disc

harg

es.

CT

isth

ete

mpo

ralc

ente

rof

the

hydr

ogra

ph.S

WE

and

Prec

ip.a

recu

mul

ativ

eva

lues

for

the

wat

erye

ar.P

eak

day

isth

eda

yof

peak

SWE

accu

mul

atio

n,an

dPe

akSW

Eis

the

amou

nt.W

inte

rte

mpe

ratu

reis

aver

age

daily

med

ian

tem

pera

ture

for

Nov

embe

rth

roug

hM

arch

,an

dsp

ring

tem

pera

ture

isth

eav

erag

eda

ily

med

ian

tem

pera

ture

for

Apr

ilto

June

Cle

arL

ake

Smith

Riv

er

WY

Mea

nA

ugus

tA

utum

nM

in.

CT

WY

Mea

nA

ugus

tA

utum

nM

in.

CT

Cle

arL

ake

Mea

n1Ð0

0A

ugus

t0Ð8

81Ð0

0A

utum

nM

in.

0Ð93

0Ð92

1Ð00

CT

�0Ð10

0Ð28

0Ð04

1Ð00

Smit

hR

iver

Mea

n0Ð9

40Ð8

60Ð8

4�0

Ð041Ð0

0A

ugus

t0Ð5

00Ð7

60Ð5

90Ð4

30Ð8

61Ð0

0A

utum

nM

in.

0Ð56

0Ð74

0Ð74

0Ð41

0Ð49

0Ð62

1Ð00

CT

�0Ð26

0Ð05

�0Ð14

0Ð89

�0Ð23

0Ð23

0Ð21

1Ð00

Hog

gPa

ssSW

E0Ð7

40Ð7

70Ð6

90Ð2

50Ð8

00Ð4

40Ð4

70Ð0

0Pr

ecip

.0Ð9

10Ð7

50Ð8

1�0

Ð190Ð9

50Ð4

40Ð4

0−

0Ð34

Peak

Day

0Ð14

0Ð25

0Ð17

0Ð15

0Ð21

0Ð39

0Ð11

0Ð21

Peak

SWE

0Ð67

0Ð79

0Ð68

0Ð39

0Ð74

0Ð53

0Ð56

0Ð11

Win

ter

T�0

Ð08�0

Ð33�0

Ð20−

0Ð60

�0Ð18

−0Ð5

1−

0Ð43

−0Ð5

0Sp

ring

T�0

Ð33−

0Ð52

−0Ð4

7−

0Ð45

�0Ð38

−0Ð5

5−

0Ð52

�0Ð34

Sant

iam

Junc

tion

SWE

0Ð57

0Ð76

0Ð68

0Ð48

0Ð61

0Ð58

0Ð62

0Ð27

Prec

ip.

0Ð88

0Ð82

0Ð85

�0Ð01

0Ð95

0Ð55

0Ð51

�0Ð22

Peak

Day

0Ð21

0Ð34

0Ð28

0Ð23

0Ð24

0Ð18

0Ð43

0Ð05

Peak

SWE

0Ð51

0Ð73

0Ð60

0Ð51

0Ð57

0Ð60

0Ð60

0Ð30

Win

ter

T�0

Ð09�0

Ð32�0

Ð23−

0Ð58

�0Ð20

−0Ð4

0�0

Ð36−

0Ð52

Spri

ngT

−0Ð4

6−

0Ð64

−0Ð6

0�0

Ð35−

0Ð47

−0Ð5

6−

0Ð67

�0Ð27

Jum

pO

ffJo

eSW

E0Ð3

60Ð5

50Ð4

40Ð6

40Ð3

80Ð3

60Ð4

80Ð4

9Pr

ecip

.0Ð9

30Ð8

20Ð8

8�0

Ð100Ð9

30Ð5

10Ð5

0�0

Ð26Pe

akD

ay0Ð1

40Ð2

90Ð2

60Ð2

80Ð0

50Ð2

30Ð3

60Ð2

6Pe

akSW

E0Ð3

40Ð5

40Ð4

00Ð6

90Ð3

90Ð4

30Ð5

00Ð5

3W

inte

rT

�0Ð11

�0Ð34

�0Ð26

−0Ð5

5�0

Ð14�0

Ð32−

0Ð42

−0Ð5

3Sp

ring

T�0

Ð15�0

Ð30�0

Ð26�0

Ð36�0

Ð23�0

Ð35�0

Ð31�0

Ð34



Tabl

eII

.(C

onti

nued

)

Hog

gP

ass

San

tiam

Junc

tion

Jum

pO

ffJo

e

SW

EP

reci

p.P

eak

Day

Pea

kS

WE

Win

ter

TS

prin

gT

SW

EP

reci

p.P

eak

Day

Pea

kS

WE

Win

ter

TS

prin

gT

SW

EP

reci

p.P

eak

Day

Pea

kS

WE

Win

ter

TS

prin

gT

Hog

gPa

ssSW

E1Ð0

0Pr

ecip

.0Ð7

41Ð0

0P

eak

Day

�0Ð06

0Ð13

1Ð00

Pea

kS

WE

0Ð92

0Ð66

0Ð05

1Ð00

Win

ter

T�0

Ð34�0

Ð14−

0Ð43

−0Ð6

11Ð0

0Sp

ring

T�0

Ð40�0

Ð35−

0Ð62

−0Ð4

40Ð4

61Ð0

0

San

tiam

Junc

tion

SW

E0Ð8

10Ð5

30Ð0

20Ð9

0−

0Ð59

−0Ð4

21Ð0

0P

reci

p.0Ð7

70Ð9

50Ð1

30Ð7

5�0

Ð27−

0Ð39

0Ð68

1Ð00

Pea

kD

ay0Ð3

30Ð1

80Ð0

70Ð5

5−

0Ð44

0Ð05

0Ð45

0Ð24

1Ð00

Pea

kS

WE

0Ð75

0Ð48

0Ð05

0Ð91

−0Ð7

4�0

Ð350Ð9

40Ð6

40Ð5

51Ð0

0W

inte

rT

�0Ð34

�0Ð17

�0Ð35

−0Ð6

00Ð9

50Ð5

5−

0Ð60

�0Ð33

−0Ð4

1−

0Ð73

1Ð00

Spri

ngT

−0Ð3

8−

0Ð41

−0Ð6

4−

0Ð42

0Ð31

0Ð94

−0Ð3

8−

0Ð45

�0Ð06

�0Ð26

0Ð38

1Ð00

Jum

pO

ffJo

eS

WE

0Ð67

0Ð33

�0Ð05

0Ð75

−0Ð5

9�0

Ð370Ð8

80Ð4

80Ð2

90Ð8

0−

0Ð59

�0Ð28

1Ð00

Prec

ip.

0Ð69

0Ð95

0Ð14

0Ð64

�0Ð16

�0Ð36

0Ð57

0Ð97

0Ð15

0Ð51

�0Ð21

−0Ð4

50Ð3

91Ð0

0P

eak

Day

0Ð06

0Ð05

0Ð24

0Ð25

�0Ð30

�0Ð15

0Ð35

0Ð06

0Ð43

0Ð32

�0Ð21

�0Ð27

0Ð29

0Ð02

1Ð00

Pea

kS

WE

0Ð68

0Ð33

�0Ð07

0Ð77

−0Ð6

7−

0Ð39

0Ð86

0Ð47

0Ð29

0Ð84

−0Ð6

4�0

Ð230Ð9

60Ð3

60Ð2

71Ð0

0W

inte

rT

�0Ð27

�0Ð08

−0Ð4

2−

0Ð50

0Ð81

0Ð40

−0Ð6

5�0

Ð23−

0Ð48

−0Ð6

90Ð8

00Ð3

1−

0Ð71

�0Ð13

−0Ð4

4−

0Ð71

1Ð00

Spri

ngT

�0Ð23

�0Ð22

−0Ð6

1�0

Ð240Ð4

10Ð9

3�0

Ð25�0

Ð250Ð1

0�0

Ð190Ð5

40Ð8

5�0

Ð26�0

Ð21�0

Ð01�0

Ð290Ð4

41Ð0

0 temporal centroid is significantly correlated with temper-ature and SWE-derived variables, but not precipitation.Thus, early dates of the temporal centroid are indica-tive of warm years with winter rain or early snowmelt,regardless of the total precipitation or discharge. Theregression between temporal centroid and water year forthe Clear Lake discharge record is statistically significant(P D 0Ð06), revealing that the temporal centroid movedearlier by 17 days between 1948 and 2007. This is con-sistent with the findings of Stewart et al. (2005), whoreported temporal centroids moving earlier by 1–4 weeksbetween 1948 and 2002 in snowmelt-dominated basins ofwestern North America. The regression between temporalcentroid and water year is not statistically significant forSmith River (P D 0Ð85), possibly reflecting the shorterrecord length in that watershed.

Warmer winters and earlier spring snowmelt also seemto be affecting low flows from groundwater-dominat-ed watersheds disproportionately relative to runoff-dominated watersheds. In the Clear Lake watershed,August discharges, normalized by mean water year dis-charges, (QAug/Qwy) have experienced statistically sig-nificant declines, as have Qmin/Qmean ratios (Figure 6).QAug/Qwy declined 15% from 1948 to 2005, andQmin/Qmean has declined 11% in the same period. In con-trast, QAug/Qwy and Qmin/Qmean ratios lack statisticallysignificant trends in the Smith River watershed. Longerrecord watersheds mirror the results of the Clear Lakeand Smith River watersheds: the groundwater-dominatedMcKenzie Bridge watershed shows significant declines,whereas the runoff-dominated South Santiam does not.These parallel results suggest that the trends are not afunction of the period of record, but reflect the adjustmentof flow regimes to changing climatic conditions.

In groundwater-dominated watersheds, summer andautumn low flows are decreasing as snowmelt occursearlier, the slow, summer hydrograph recession beginssooner, and it occurs over a longer period of time. Asthe start date to the summer recession becomes earlier,discharge for a specified interval of the recession period(e.g. August discharge) decreases because the intervalis later on the recession limb. In the same manner,with a longer recession, an ultimately lower endpoint(i.e. autumn minimum discharge) is reached. In runoff-dominated watersheds, the fast recession leaves summerand autumn low flows insensitive to climatic variabilitybecause subsurface storage is nearly exhausted early eachsummer.

DISCUSSION

The different behaviours of groundwater- and runoff-dominated watersheds on event, seasonal, and inter-annual scales are clearly revealed by the analysis ofhistorical records in the Clear Lake and Smith Riverwatersheds. These contrasting behaviours allow one toidentify the primary and secondary controls on watershedresponse at each time scale (Table III). These different



behaviours have implications for how the watershedshave responded to warming winters in the past 45 yearsand how they might respond to future climate warming.

At the event scale, discharge is controlled primarilyby the magnitude and intensity of the rain or snowmeltand secondarily by the geology of the watershed. Therecharge signal is spatially and temporally varying,because of topographic effects on both precipitationand temperature. Particularly important to determining

Table III. Controls on discharge variability for Cascades water-sheds

Scale Primary Control Secondary Control

Event Precipitation GeologySeasonal Geology Winter–spring temperatureAnnual Precipitation Geology (minor)

discharge is the location of the transition between snowand rain during a precipitation event and the averagetemperature during a melt event. The bedrock hydraulicconductivity determines whether most recharge is quicklyrouted to streams, resulting in a flashy hydrograph, suchas that of Smith River, or whether most recharge reachesa deeper aquifer where the signal becomes damped andlagged, as in the Clear Lake watershed.

At the seasonal scale, variations in discharge are pri-marily controlled by the watershed geology, and secon-darily by winter and spring temperature. For aquifers withseasonal recharge, damping of the hydrograph is a func-tion of aquifer length, storativity, transmissivity and theperiodicity of the recharge (Townley, 1995). The aquifercharacteristics of the Clear Lake watershed produce sig-nificant damping of the discharge time series relative tothe recharge time series. This damping results in the sus-tained release of groundwater during the summer drought,

Figure 6. Relationship between low flow parameters and calendar year: (a) ratio of August mean discharge to water year mean discharge (Qaug /Qwy)and (b) ratio of autumn minimum discharge to the average discharge of the preceding nine months (Qmin/Qmean)



even though the seasonal water balance is negative. Incontrast, the shallow subsurface flow system of the SmithRiver watershed rapidly transmits pressure waves fromrecharge to produce changes in streamflow (Torres et al.,1998). This results in a flashy winter hydrograph andrapid summer recession.

In the Clear Lake watershed, the streamflow season-ality, as measured by the ratio of discharge in the high-est flow month to that in September, is related to thepartitioning of precipitation between rain and snow andthe timing of the snowmelt. Years with higher snow torain ratios have lower winter flows and relatively morerecharge in the spring, resulting in more water to sustainsummer discharge. Years with colder spring tempera-tures have delayed snowmelt, resulting in a later onsetof the summer recession and higher discharge during theautumn minimum flow period. Both of these scenariosresult in less month-to-month variability in discharge.

At the annual scale, mean discharge is controlled by thecurrent year’s precipitation. Although it has been shownthat the previous year’s precipitation has some influenceon mean discharge at Clear Lake, the inter-annual vari-ability in discharge is similar in the two watersheds.This suggests that the groundwater system does not havea significant moderating effect on overall inter-annualflow variability. Drought and high water years showup with approximately the same frequency and severityin both groundwater- and runoff-dominated watersheds.However, during the summer recession, groundwater canhave a more important moderating effect on inter-annualvariability. Conversely, at the tail end of the summerrecession groundwater increases the effects of inter-annual variability. This corresponds with the hypothesisthat subsurface storage is nearly exhausted every year inthe runoff-dominated watershed, resulting in less inter-annual variability.

Based on the patterns of the past several decades,one can begin to forecast what the next few decadeswill likely bring to streams in the Oregon Cascades.The correlations between accumulated SWE and wintertemperature and ENSO indices, combined with the long-term decrease in the snow-covered season at SantiamJunction, suggest that there is very strong elevationsensitivity in SWE parameters. In particular, it seemsthat areas at the transition between transient and seasonalsnowpacks are most sensitive to warmer winters. Ifwinters continue to warm, as suggested by regionalclimate models (Christensen et al., 2007), it is projectedthat the elevation of the transient snow zone will shiftupward and substantial portions of the previously snow-dominated Cascades will experience a higher frequencyof winter rain events. For these watersheds that inthe future have more transient and less seasonal snow,winter discharge will increase and the snowmelt peakwill become less important. Thus, watersheds in theOregon Cascades most sensitive to warming will be thosewith much of their area between 1100 and 1300 m.For watersheds already dominated by transient snow(<1100 m), the proportion of winter rain will increase,

but the hydrograph will likely be less affected than forhigher elevation watersheds. For the highest elevationwatersheds (>1500 m), seasonal snow cover will persistbut the snowmelt peak will occur earlier in the spring.In the Willamette River watershed, 14% of the CascadeRange lies at an elevation range between 1100 and1300 m, and 91% lies below 1500 m. Approximately37% of the area of the Clear Lake watershed lies withinthe 1100–1300 m range, as does 43% of the Smith Riverwatershed.

The historical record suggests that the annual hydro-graph has already shifted and that Qmin in groundwater-dominated watersheds is already declining as snowmeltoccurs earlier (Figure 7). Since the late 1940s the peakdischarge month for Clear Lake has shifted from Mayto April, as a function of earlier snowmelt. This leadsto longer summer recession periods, and ultimatelylower Qmin in groundwater-dominated watersheds. Iffuture warming proceeds as predicted by regional cli-mate models, and simulated by Tague et al. (2008) forthe Clear Lake watershed, the summer recession willfurther lengthen and minimum flows will decline evenmore. It is expected that in runoff-dominated watersheds,longer recession periods will exhaust subsurface storagesooner, but that the effect on Qmin will be smaller thanin groundwater-dominated watersheds.

CONCLUSIONS

Analyses of the historical datasets for two watersheds inthe Oregon Cascades clearly highlight the importance ofgroundwater systems in sustaining summer streamflow.Discharge and evapotranspiration in the groundwater-dominated watershed account for 260% of incomingprecipitation between July and October, so groundwaterstorage and the associated slow recession are respon-sible for sustaining discharge even when the seasonalor annual water balance is negative. While ground-water significantly influences the shape of the annualhydrograph, climatic factors control the inter-annual vari-ability of streamflow. Even though Cascades aquifers

Figure 7. Monthly percentages of annual discharge for the Clear Lakewatershed in two historical periods (1948–1952, 2001–2005) and ahydrograph based on 1Ð5° warming modelled by Tague et al. (2008).

October to December are repeated to illustrate the low flow period



store multiple years worth of water (Jefferson et al.,2006), and there is a one year memory associated withthe groundwater system, these effects are not enoughto dampen inter-annual variability in streamflow ingroundwater-dominated streams, as evidenced by com-parable coefficients of variation between the two studywatersheds.

Warmer winters and earlier snowmelt over the past¾60 years have shifted the hydrograph for theMcKenzie River at Clear Lake, resulting in an ear-lier temporal centroid (17 days), longer summer reces-sions and lowered August discharge (15%) and autumnminimum discharge (11%). Projections of future cli-mate suggest that this trend will continue over thenext 50–100 years. The slow summer recession ofgroundwater-dominated streams makes their low flowsmore sensitive to changes in snowmelt amount and timingthan those of runoff-dominated streams. Groundwater-dominated watersheds that are transitioning between tran-sient and seasonal snow regimes will be the most affectedby projected climatic change, particularly in terms of latesummer and fall streamflow. In the Oregon Cascades,the region of maximum sensitivity seems to be between1100 and 1300 m, based on historical trends and pro-jected changes over the next 50 years.

These results suggest that water resource managersin the mountainous western USA must identify ground-water-dominated watersheds and those that are perchedat the transient/seasonal snow transition, and that theyshould be prepared for significant changes to the annualhydrograph in those areas. Downstream water users andhydroelectric dam operators will have to adjust theirusage schedules or face shortfalls. Aquatic species, suchas salmonids, will be subject to diminishing flows andpossibly increasing water temperatures.

ACKNOWLEDGEMENTS

This material is based upon work supported under aNational Science Foundation Graduate Research Fellow-ship and grants from the Eugene Water and ElectricBoard and the Institute for Water and Watersheds at Ore-gon State University. We thank Gordon Grant for helpwith project design and useful discussions, and we thankMeredith Payne for assistance with data compilation andpreliminary analysis. The manuscript benefitted from thethoughtful comments of two anonymous reviewers.

REFERENCES

Beebee R, Manga M. 2004. Variation in the relationship betweensnowmelt runoff in Oregon and ENSO and PDO. Journal of theAmerican Water Resources Association 40: 1011–1024.

Box GEP, Jenkins G. 1976. Time Series Analysis: Forecasting andControl . Holden-Day: Boca Raton, FL.

Cayan DR, Kammerdiener SA, Dettinger MD, Caprio JM, Peterson DH.2001. Changes in the onset of spring in the western United States.Bulletin of the American Meteorological Society 82: 399–415.

Christensen JH, Hewitson B, Busuioc A, Chen A, Gao X, Held I,Jones R, Kolli RK, Kwon W-T, Laprise R, Magana Rueda V,

Mearns CG, Menendez J, Raisanen J, Rinke A, Sarr A, Whetton P.2007. Regional climate projections. In Climate Change 2007: ThePhysical Science Basis. Contribution of Working Group I to the FourthAssessment Report of the Intergovernmental Panel on Climate Change,Solomon S, Qin D, Manning M, Chen Z, Marquis M, Averyt KB,Tignor M, Miller HL (eds). Cambridge University Press: Cambridge;847–940.

Daly C, Gibson WP, Taylor GH, Johnson GL, Pasteris P. 2002. Aknowledge-based approach to the statistical mapping of climate.Climate Research 22: 99–113.

Daly C, Taylor G, Gibson W. 1997. The PRISM approach to mappingprecipitation and temperature In 10th Conference on AppliedClimatology. American Meteorological Society: Reno, NV; 10–12.

Folland CK, Karl TR, et al. 2001. Observed climate variability andchange. In Climate Change 2001: The Scientific Basis. Contributionof Working Group I to the Third Assessment Report of theIntergovernmental Panel on Climate Change, Houghton JT, Ding Y,Griggs DJ, Noguer M, van der Linden PJ, Dai X, Maskell K,Johnson CA (eds). Cambridge University Press: Cambridge; 99–181.

Haan CT. 2002. Statistical Methods in Hydrology . Iowa State Press:Ames, Iowa.

Harr RD. 1977. Water flux in soil and subsoil on a steep forested slope.Journal of Hydrology 33: 37–58.

Harr RD. 1981. Some characteristics and consequences of snowmeltduring rainfall in western Oregon. Journal of Hydrology 53: 277–304.

Herrett TA, Hess GW, Stewart MA, Ruppert GP, Courts M-L. 2005.Water Resources Data for Oregon, Water Year 2005. US GeologicalSurvey: Washington, DC.

Ingebritsen SE, Mariner RH, Sherrod DR. 1994. Hydrothermal systemsof the Cascade Range, north-central Oregon. In U.S. GeologicalSurvey . US Geological Survey: Washington, DC.

Jefferson A, Grant GE, Rose TP. 2006. The influence of volcanichistory on groundwater patterns on the west slope of theOregon High Cascades. Water Resources Research 42: W12411–,DOI:12410Ð11029/12005WR004812.

Kite GW. 1977. Frequency and Risk Analysis in Hydrology . WaterResources Publications: Fort Collins, CO.

Leung LR, Qian Y, Bian X, Washington W, Han J, Roads JO. 2004.Mid-century ensemble regional climate change scenarios for thewestern United States. Climatic Change 62: 75–113.

Manga M. 1999. On the timescales characterizing groundwater dischargeat springs. Journal of Hydrology 219: 56–69.

Mantua NJ, Hare SR, Zhang Y, Wallace JM, Francis RC. 1997. A Pacificinterdecadal climate oscillation with impacts on salmon production.Bulletin of the American Meteorological Society 78: 1069–1079.

Mote PW. 2003. Trends in snow water equivalent in the Pacific Northwestand their climatic causes. Geophysical Research Letters 30: 1601,DOI:1610Ð1029/2003GL017258.

Mote PW, Hamlet AF, Clark M, Lettenmaier DP. 2005. Decliningmountain snowpack in western North America. Bulletin of theAmerican Meteorological Society 86: 39–49.

Mote PW, Parson E, Hamlet AF, Keeton WS, Lettenmaier DP, Man-tua NJ, Miles EL, Peterson D, Peterson DL, Slaughter R, Snover AK.2003. Preparing for climatic change: the water, salmon, and forests ofthe Pacific Northwest. Climatic Change 61: 45–88.

Naiman RJ, Fetherston KL, McKay SJ, Chen J. 1998. Riparian forests.In River Ecology and Management: Lessons from the Pacific Northwest,Naiman RJ, Bilby RE (eds). Springer: New York; 289–323.

Nolin AW, Daly C. 2006. Mapping “at-risk” snow in the PacificNorthwest, U.S.A. Journal of Hydrometeorology 7: 1164–1171.

Priest GR, Woller NM, Black GL, Evans SH. 1983. Overview ofthe geology of the central Oregon Cascade Range. In Geologyand Geothermal Resources of the Central Oregon Cascade Range,Priest GR, Vogt BF (eds). Oregon Department of Geology and MineralIndustries: Salem, OR; 3–28.

Redmond KT. 1990. Crater Lake climate and lake level variability. InCrater Lake: An Ecosystem Study, Drake ET, Larson GL, Dymond J,Collier R (eds). American Association for the Advancement ofScience: San Francisco, CA; 127–141.

Regonda SK, Rajagopalan B, Clark M, Pitlick J. 2005. Seasonal cycleshifts in hydroclimatology over the western United States. Journal ofClimate 18: 372–384.

Ropelewski CF, Jones PD. 1987. An extension of the Tahiti-DarwinSouthern Oscillation Index. Monthly Weather Review 115: 2161–2165.

Saar MO, Manga M. 2004. Depth dependence of permeability in theOregon Cascades inferred from hydrogeologic, thermal, seismic, andmagmatic modeling constraints. Journal of Geophysical Research 109:B04204, 04210Ð01029/02003JB002855.



Selong JH, McMahon TE, Zale AV, Barrows FT. 2001. Effects oftemperature on growth and survival of bull trout, with applicationof an improved method for determining thermal tolerance in fishes.Transactions of the American Fisheries Society 130: 1026–1037.

Sherrod DR, Smith JG. 2000. Geologic Map of Upper Eocene toHolocene Volcanic and Related Rocks of the Cascade Range, Oregon.US Geological Survey: Washington, DC.

Slack JR, Lumb AM, Landwehr JM. 1993. Hydro-Climatic Data Network(HCDN): Streamflow Data Set, 1874–1988 . US Geological Survey:Washington DC.

Stewart IT, Cayan DR, Dettinger MD. 2005. Changes toward earlierstreamflow timing across western North America. Journal of Climate18: 1136–1155.

Storck P, Lettenmaier DP, Bolton SM. 2002. Measurement of snowinterception and canopy effects on snow accumulation and melt in amountainous maritime climate, Oregon, United States. Water ResourcesResearch 38: 1223, DOI:1210Ð1029/2002WR001281.

Tague C, Farrell M, Grant GE, Lewis SL, Rey S. 2007. Hydrogeologiccontrols on summer stream temperatures in the McKenzie River basin,Oregon. Hydrological Processes 21: 3288–3300.

Tague C, Grant GE. 2004. A geological framework for interpretingthe low flow regimes of Cascade streams, Willamette River Basin,Oregon. Water Resources Research 40: W04303–04310Ð01029/02003WR002629.

Tague C, Grant GE, Farrell M, Choate J, Jefferson A. 2008. Deepgroundwater mediates streamflow response to climate warming in theOregon Cascades. Climatic Change 86: 189–210.

Tague CL, Band LE. 2004. RHESSys: regional hydro-ecologic simula-tion system–an object-oriented approach to spatially distributed model-ing of carbon, water, and nutrient cycling. Earth Interactions 8: 11–42.

Torres R, Dietrich WE, Montgomery DR, Anderson SP, Loague K.1998. Unsaturated zone processes and the hydrologic response ofa steep, unchanneled catchment. Water Resources Research 34:1865–1879.

Townley LR. 1995. The response of aquifers to periodic forcing.Advances in Water Resources 18: 125–146.

Trenberth KE, Stepaniak DP. 2001. Indices of El Nino evolution. Journalof Climate 14: 1697–1701.

Washington WM, Weatherly JW, Meehl GA, Semtener AJ, Bettge TW,Craig AP, Strant WG, Arblaster J, Wayland VB, James R, Zhang Y.2000. Parallel Climate Model (PCM) control and transient simulations.Climate Dynamics 16: 755–774.

Webb BW, Clack PD, Walling DE. 2003. Water-air temperaturerelationships in a Devon river system and the role of flow. HydrologicalProcesses 17: 3069–3084.


hydrogeologic controls on streamﬂow sensitivity to climate

Documents