Seasonally varied hillslope and groundwater contributions to streamflow in a glacial till and fractured sedimentary bedrock dominated Rocky Mountain watershed

Whereas a lack of streamflow response to significant forest disturbance (e.g., forestry, wildfire, and insect infestation) has been observed at multiple locations in the Canadian Rocky Mountains, a region with sedimentary bedrock 10 overlain by glacial till, mechanisms governing this lack of change remain unclear. Some inferences can be drawn from conceptualizations of runoff generation (e.g., runoff thresholds and hydrologic connectivity) in physically similar watersheds, although much of the focus has been on rainfall-runoff dynamics. Thus, there is a need to describe runoff generation in this snow-dominated area to interpret how forest disturbance may impact streamflow quantity for downstream users. Stream water and source water (snow, rain, hillslope groundwater, till groundwater, bedrock groundwater and seeps) were sampled in four 15 sub-watersheds (Star West Lower, Star West Upper, Star East Lower and Star East Upper) in Star Creek, SW Alberta. Principal component analysis was used to determine the relative dominance and timing of source water contributions to streamflow over the 2014 and 2015 hydrologic seasons. An initial displacement of old water stored in the hillslope over winter occurred at the onset of snowmelt, before the stream responded significantly. This was followed by a dilution effect as snowmelt saturated the landscape, recharged groundwater, and connected the hillslope to the stream. Fall baseflows were dominated by either hillslope 20 groundwater or bedrock groundwater in Star West. Conversely, in Star East, stream water was similar to hillslope water in August but was unlike the measured sources in September and October. Temperature and chemical signatures of groundwater seeps suggest highly complex subsurface flow pathways. Hydrologic resilience in the Rocky Mountain eastern slopes may be, in part, due to these complex subsurface pathways in combination with the slow release of groundwater from glacial till. 25 https://doi.org/10.5194/hess-2020-105 Preprint. Discussion started: 16 March 2020 c © Author(s) 2020. CC BY 4.0 License.


Introduction
Forest disturbance from wildfire, pine beetle infestation, or forest harvesting removes the forest canopy increasing the total precipitation that reaches the forest floor (Williams et al., 2019;Burles and Boon, 2011;Boon, 2012;Varhola et al., 2010) 30 often altering the dominant flow pathways, increasing streamflow quantity and changing the timing of flows in forested watersheds (Stednick, 1996;Scott, 1993;Winkler et al., 2017). However, large variability has been observed in streamflow responses following disturbance due to differences in disturbance type, vegetation type, precipitation regimes, and soil moisture storage (Brown et al., 2005;Stednick, 1996). Some studies have reported little, if any, change in streamflow following disturbance (Williams et al., 2015;Harder et al., 2015;Goodbrand and Anderson, 2016;Andres et al., 1987) but the 35 mechanisms or features potentially responsible for the lack of flow response remain unclear. Watershed resilience, when a watershed maintains a similar form and function following forest disturbance (Creed et al., 2011), has been used to explain the lack of change in streamflow following disturbance (Harder et al., 2015). Watersheds exhibiting resilience might be associated with a large storage capacity and complex subsurface flow pathways (Harder et al., 2015) where water can be stored for months or years and be subsequently released to the stream gradually thereby buffering the flow response of watersheds from 40 disturbance (Creed et al., 2011).
High bedrock permeability is a watershed feature that has been associated with the slow release of stored water to streams during baseflow (Uchida et al., 2006;Liu et al., 2004;Pfister et al., 2017), thereby potentially influencing the hydrological resilience. Uchida et al. (2006) reported that a watershed with greater bedrock permeability had larger aquifer storage, and the subsequent release of stored water maintained baseflow later in the year. Similarly, Liu et al. (2004) showed that the recession 45 limb of the annual hydrograph in the Colorado front range Rocky Mountains was driven by baseflow released from fractured bedrock. Deep soils and till deposits with large storage capacities have also been shown to sustain baseflows during drier periods (Floriancic et al., 2018;Shanley et al., 2015). Deep sediment deposits in the Poschiavino watershed, in Switzerland, were associated with larger storage capacity and higher winter baseflows compared to watersheds with shallow sediment deposits (Floriancic et al., 2018). Similarly, deep basal till in Sleepers River watershed in Vermont was associated with large 50 storage capacity and low permeability that promoted extended maintenance of baseflow (Shanley et al., 2015).
While these studies illustrate the influence of bedrock, deep soils, or till on baseflows, few studies have explored the combination of these storage zones on streamflow contributions (Burns et al., 1998;Dalke et al., 2012;Spencer et al., 2019). Burns et al. (1998) characterized the difference in deep (bedrock) and shallow (soils and till) flow systems in the Catskill Mountains in New York, a region with both glacial till and permeable sedimentary bedrock. Baseflow was maintained by 55 discharge from perennial springs, which originated from bedrock fractures, rather than contributions from the soil and till flow system (Burns et al., 1998). Similarly, precipitation-runoff relationships in Alberta's Rocky Mountains indicated that runoff generation is complicated by the interaction of two storage zones -soil and till storage and bedrock storage  where coupled flow and tracer approaches would be needed to reduce uncertainty in estimated flow contributions from each storage zone. Others have also shown the importance of groundwater contributions to streamflow in alpine watersheds in 60 https://doi.org/10.5194/hess-2020-105 Preprint. Discussion started: 16 March 2020 c Author(s) 2020. CC BY 4.0 License.
the Rocky Mountains (Hood and Hayashi, 2015;McClymont et al., 2010). However, due to the highly heterogeneous nature of glacial till, more research is needed to differentiate between baseflow contributions from glacial till and bedrock storage sources.
Chemical signatures of source water (e.g., glacial till groundwater, bedrock groundwater, soil water, and precipitation) and stream water can be used to determine which sources are contributing to streamflow during different flow conditions using 65 end-member mixing (Christophersen and Hooper, 1992). Accordingly, the objectives of this study were to: 1) characterize how sources of stream water (precipitation, soil water, hillslope groundwater, till groundwater, bedrock groundwater, and seeps) vary spatially across four sub-watersheds of a Rocky Mountain watershed and temporally from spring snowmelt to fall low flows, and; 2) determine the relative contributions of source water to the stream from spring to fall for each sub-watershed.
Two adjacent geologically similar watersheds were compared to explore the relative variation in stream water sources across 70 key physiographic zones (higher elevation alpine/sub-alpine vs lower elevation upper montane) that were evident within Star Creek watershed. This study should help inform the current conceptualization of runoff generation in northern Rocky Mountain watersheds, thereby providing new understanding of the potential factors influencing watershed resilience to close a critical evidence gap.

Study Site 75
Star Creek watershed (10.4 km 2 ; Figure 1) is located in the eastern slopes of Canada's Rocky Mountains; a region with fractured sedimentary bedrock overlain by glacial till. Area-weighted average precipitation is 950 mm, with 50-60 % in the form of snow. Soils are Eutric Brunisols (Can. Soil Classification, or Eutric Cambisols in Food and Agriculture Organization system) approximately 1 m deep, on average. Unsorted and uncompacted glacial till is 3 m deep, on average, with some clayrich layers distributed unevenly throughout the watershed. Sedimentary geologic formations are primarily composed of shale 80 and sandstone and are highly fractured due to folding and faulting (AGS, 2004).
There are two main sub-watersheds, Star East (3.9 km 2 ; 1537-2628 m above sea level) and Star West (4.6 km 2 ; 1540-2516 m above sea level). Unvegetated talus slopes beneath exposed bedrock form the upper portion of alpine zones in both subwatersheds. Star West has a larger alpine region with till deposits that includes a narrow marshy area proximal to the stream that holds water throughout the summer and drains into the main channel that is primarily bedrock in the upper reaches. The 85 Star East alpine region is smaller and more constricted than in Star West ( Figure 2) and is comprised mostly of a grassy meadow with the stream originating from springs where the water table reaches the soil surface and is incised in colluvium with large boulders. In the lower reaches, streams in both sub-watersheds are composed of a series of step-pools incised in alluvium and colluvium where the lower sub-alpine and upper Montane zones are dominated by sub-alpine fir (Abies lasiocarpa) and Englemann spruce (Picea Englemannii) above lodgepole pine (Pinus contorta) dominated forests in lower 90 reaches (Dixon et al., 2014;Silins et al., 2009). https://doi.org/10.5194/hess-2020-105 Preprint. Discussion started: 16 March 2020 c Author(s) 2020. CC BY 4.0 License.  https://doi.org/10.5194/hess-2020-105 Preprint. Discussion started: 16 March 2020 c Author(s) 2020. CC BY 4.0 License.

Stream water chemistry
Stream water samples were collected from the four streamflow gauging stations (SEL, SEU, SWL, SWU; Figure 1) every two weeks from April to October, in 2014 and 2015, to capture the full range in streamflow chemistry over the hydrologically active period. One litre plastic bottles were triple rinsed prior to sample collection. Samples were analyzed for major cations and anions (Na + , Mg 2+ , Ca 2+ , K + , Cl -, SO4 2-) and silica (Si as SiO2) in the Biogeochemical Analytical Service Laboratory 105 (University of Alberta). An inductively coupled plasma-optical emission spectrometer (Thermo Scientific ICAP 6300) was used to measure Na + , Mg 2+ , Ca 2+ , K + with analytical precision of 1.9 %, 3.0 %, 1.9 %, and 2.4 %, respectively. An ion chromatograph (Dionex DX600 and Dionex ICS 2500) was used to measure Cland SO4 2with analytical precision of 2.4 % and 3.1 %. Flow injection analysis (Lachat QuikChem 8500 FIA automated ion analyzer) was used to measure Si with analytical precision of 3.4 %. 110

Source water chemistry
Stream water sources were a priori hypothesized to be rain, snowmelt, soil water, hillslope groundwater, and groundwater seeps. All source water samples were collected in triple rinsed (with source water) 50 ml plastic vials and analyzed with the same methods as the stream water samples to support application of end-member mixing analysis. Source water collection and sampling methods are detailed below. 115

Rainfall and snowmelt
Rain samples were collected in a clean bucket that was rinsed with deionized water. Buckets were placed in open areas throughout the watershed or in the nearby townsite (Coleman, AB; within approximately 8 km of Star Creek watershed) after a rainstorm began. Locations were chosen opportunistically depending on storm timing and site access. Samples were collected at the end of the day or once there was enough water in the bucket to sample, to prevent changes in chemical composition due 120 to dry deposition of dust or evaporation. Five, four, and three samples were collected throughout the summers of 2013, 2014, and 2015, respectively. The difficulty of capturing large convective storms and the large frequency of storms less than 5 mm (Williams et al., 2019) prevented the collection of more rainfall samples. prior to snow accumulation. Samples were collected directly from snowmelt troughs and snow bridges with clearly visible melt. Snowmelt was sampled, instead of the snowpack, to better reflect the meltwater signature during the snowmelt period (Johannessen and Henriksen, 1978).

Soil water
Suction lysimeters were installed between 30-60 cm depth using a hand auger in two locations near the toe of the hillslope in each sub-watershed in early spring 2014 (2015 for SEU; Figure 1). Suction lysimeters consisted of a 0.5 Bar ceramic cup and 38.1 mm PVC pipe to ensure ample water was collected for chemical analyses. Water from the suction lysimeter was sampled using a hand pump every two weeks between April and October in 2014 and 2015. Suction lysimeters were pumped dry 135 following sampling and pressure was applied. Thus, soil water was composed of water that was able to pass through the ceramic cup over the two-week period until the lysimeter was at equilibrium pressure with the surrounding soil. Shallow depths were targeted with the intention to collect the unsaturated soil water above the saturated zone in the hillslope, which was sampled separately.

Hillslope groundwater 140
Hillslope wells were installed with a shovel or hand auger to depth of refusal or maximum auger depth (1.5 m) near the hydrometric gauging stations at SEL, SEU, SWL and SWU (Figure 1). A site was added at SEU at the end of the summer in 2014, whereas the other sites were established during summer 2013. Wells were installed in three locations at each site: riparian, toe slope, and hillslope positions to determine the full range in hillslope groundwater. Well depths ranged between 0.5 m (riparian wells) and 1.6 m. Wells were purged using a hand pump prior to sampling. Samples were collected 145 approximately every two weeks, as available, between April and October in 2014 and 2015. Samples from the upper hillslope wells were generally only obtained during the snowmelt or high flow period; these wells were often dry during late summer.
Riparian and toe slope wells contained water for all or most of the year, respectively.

Groundwater seeps
At the onset of this research, lack of access to backcountry sites restricted the installation of deep bedrock or till groundwater 150 wells in upper sub-watersheds. Rather, groundwater seeps were used to characterize the possible range in groundwater signatures (both bedrock and till groundwater) within Star Creek. Seeps are defined here as areas of visible water seeping from hillslopes proximal to the stream or from small wetland areas further from the stream that form small tributaries or rivulets that flow into the stream. The east and west forks were surveyed from the confluence with the main stem to the stream origins in the alpine area in July 2013; 25 visible seeps were identified. Samples were collected during three flow conditions: high 155 flow (May/June), recession flow (mid-July), and baseflow (early September prior to fall rains), in both 2014 and 2015. Water temperature was also measured during sample collection to aid in differentiating between deep bedrock groundwater, till groundwater and hillslope groundwater.

Bedrock and till groundwater
Preliminary end-member mixing analysis showed that a water source was missing from those initially collected (above) 160 highlighting the need to characterize deeper groundwater. Due to monetary and access limitations, a single borehole was drilled to 12 m depth (15.2 cm in diameter) in the topographic ridge between SEL and SWL (approx. 500 m upstream from gauging sites) in October 2015 ( Figure 1). Two wells were installed in the borehole, one well in a water-baring formation in the bedrock at 11 m depth, and a second well in the glacial till deposits at 4.5 m depth, to characterize the differences in bedrock and till groundwater chemistry. Both wells had screens that were 1.5 m in length. Sand was used to backfill the borehole around the 165 screened section of the bedrock groundwater well and was capped with bentonite clay. Local material removed during drilling was used to backfill the borehole up to the till layer. The same method of back filling (sand, bentonite clay, local material) was used for the till groundwater well. Bedrock and till wells were sampled every two to four weeks in 2016 and 2017. Water in the till well was purged until dry prior to sampling. Water in the bedrock well was purged for 2-5 minutes prior to sampling because the recharge rate was faster than the pump rate. Water table depth and temperature were measured continuously with 170 pressure transducers (HOBO U20, Onset Computer Corp., Bourne, MA, USA) at 10-minute intervals. High concentrations of Na + , Cl -, and SO4 2in till groundwater ( Figure 3) and large variability between years suggested that the 175 till groundwater well was likely contaminated by the bentonite clay used to backfill and seal between layers (Remenda and van der Kamp, 1997). Slow recharge rates (and therefore, low hydraulic conductivity) of glacial till prevented the removal of three pipe volumes when sampling and the corresponding low hydraulic conductivity resulted in little flushing of bentonite contaminants. Faster recharge rates (and therefore, higher hydraulic conductivity) of the bedrock groundwater would aid in better flushing of bentonite contaminants which would reduce the effects on bedrock groundwater chemistry (Remenda and 180 van der Kamp, 1997). As a result, the till groundwater samples were not included in the analyses herein; however, water table depths and water temperature dynamics could still be used to understand the differences between till and bedrock groundwater responses and their roles in runoff generation.

Data Processing
End-member mixing analysis (EMMA) was used to visualize multi-variate source water and stream chemistry by reducing the 185 dimensionality of the data with principal component analysis (PCA; Christophersen and Hooper, 1992). The key assumptions for EMMA are: 1) the tracers are conservative; 2) the mixing process is linear; 3) source chemistry does not change temporally or spatially over the period or area studied (Inamdar, 2011;Hooper, 2003), and; 4) all sources have been identified and have the potential to contribute to streamflow. In addition, there were multiple subjective decisions required prior to running EMMA, such as choosing tracers/ions and defining sources. Two methods were used to determine if tracers were appropriate 190 to use in the analysis. A matrix of bivariate plots of stream chemistry data (ion concentrations), used most commonly in geographical hydrograph separations, was used to determine if ions were conservative in nature (Hooper, 2003). The tracer variability ratio (TVR), used most commonly in sediment source apportionment studies, was used to determine if the difference in ion concentrations between groups was larger than the variation within a source group (Pulley et al., 2015). The TVR was calculated for each tracer pair and compared between each group as the percent difference between source group median 195 divided by the average coefficient of variation between group pairs (Pulley et al., 2015). TVR should be greater than 2 to be considered appropriate for use in mixing calculations (Pulley and Collins, 2018), although depending on the dataset in question, a greater threshold may be adopted to make the tracer selection more stringent and to help reduce the numbers of tracers included in further data processing.
Box and whisker plots and linear discriminant analysis (LDA) were used to remove the subjectivity of defining sources (Ali 200 et al., 2010;Pulley and Collins, 2018). Box and whisker plots were used as a visual means of discriminating between sources.
LDA was then used to determine if the combined sources exhibited sufficiently robust statistical separation (Pulley and Collins, 2018). Other statistical classification methods, such as hierarchical clustering or k-means clustering, were not appropriate because source categories were known a priori. The data were processed using LDA in the MASS ('lda' function; Venables and Ripley, 2002) and klaR ('stepwise' function; Weihs et al., 2005) packages in R (R Core Team, 2014). The 'stepwise' 205 https://doi.org/10.5194/hess-2020-105 Preprint. Discussion started: 16 March 2020 c Author(s) 2020. CC BY 4.0 License. function was used with the 'backwards' direction in an attempt to maintain the most tracers; 'lda' method, and 'ability to separate' criterion.
After the sources were characterized, the stream water was processed using principal component analysis (PCA; 'prcomp' function in R; R Core Team, 2014) as a method of dimensionality reduction to create a two-dimensional (2D) mixing space (Christophersen and Hooper, 1992). Stream water was standardized (subtracting the mean and dividing by the standard 210 deviation for each sampling point), for each tracer, to create equal variance between chemical components and used to create a correlation matrix. PCA was conducted on the correlation matrix to calculate eigenvectors and eigenvalues. Standardized stream water was then projected into the end member mixing-space by multiplying by eigenvectors. Ideally, two principal components (PCs) explained most of the variation in the data and were used to generate a 2D mixing space, which corresponds to three sources in EMMA (Hooper, 2003). Other studies have used the 'Rule of 1' to determine how many dimensions, and 215 therefore sources, should be used to create the mixing space (Ali et al., 2010;Barthold et al., 2011). For this study, the mixing space was set to two dimensions for ease of visualization but used all appropriate sources as presented by Inamdar et al. (2013) to provide a full description of potential source contributions. Source water was then standardized using stream water means and standard deviations for each ion and projected into the 2D mixing space as defined by the stream water (Christophersen and Hooper, 1992;Hooper, 2003). Stream water sources should create an outer boundary or polygon around all stream water 220 samples if all sources were correctly identified and adequately sampled.

Tracer and source water group selection
Bivariate plots were created and TVR was calculated to determine which tracers were appropriate for use in EMMA. Pearson correlation coefficients were calculated between all stream bivariate plots for stream water at each sub-watershed (Figure 4). 225 These showed that all tracers exhibited acceptable linear trends with at least one other tracer (Pearson's r > 0.5) and were thereby likely conservative in nature. Average TVR for almost all tracers at all sites (exception Si at SW and SE) were below 2, which suggested that the within-group variation exceeded the between-group variation and were considered unacceptable.
As a result, rather than calculating mixing ratios or percent contributions on the basis of an un-mixing routine, trends in stream water distribution were described in relation to source water dynamics and runoff processes, as suggested by Inamdar et al. 230 (2013).
The a priori classification of water sources was rain, snowmelt, soil water, riparian water, toe slope water, upper hillslope water, groundwater seeps and bedrock groundwater; however, box and whisker plots showed that the distribution of rain and snowmelt were similar as well as the distribution of soil water, riparian water, toe slope water, and hillslope water for most sites. redundancy of the tracers and to ensure that samples were well separated; on this basis, 99.7 %, 91 %, 98.6 %, and 99.9 % of samples were well separated in SWL, SWU, SEL, and SEU, respectively. In all sites, all tracers were retained to maximize the ability to distinguish between the source groups. Overall, these results support the conclusion that there was good separation 240 between source water groups as re-categorized for the individual sites. Groundwater seeps were sampled in leu of bedrock or till groundwater wells to characterize the variability in the chemical 245 signature of groundwater throughout Star Creek. The ion concentrations of the groundwater seeps were not chemically distinct because they were generally similar to stream water or hillslope groundwater in the PCA analyses (data not shown). However, examination of groundwater seep temperatures from spring to fall revealed that some seeps had consistently cool temperatures https://doi.org/10.5194/hess-2020-105 Preprint. Discussion started: 16 March 2020 c Author(s) 2020. CC BY 4.0 License. and others had larger fluctuations in temperature. This suggests that some seeps were potentially groundwater-fed and others were fed by shallow subsurface water, respectively. For example, in SEL, the temperature of a groundwater seep ranged 250 between 2.2-3.7 °C throughout the summer, which is indicative of a bedrock groundwater source because the temperature range was muted and was largely not influenced by radiative warming (Taniguchi, 1993). Similarly, in SEU, the temperature of a groundwater seep ranged from 2.5-3.5 °C, also indicating a bedrock groundwater source. Temperatures in the till groundwater well ranged between 2.7-9.7 °C and the bedrock groundwater ranged between 5.1-5.8 °C. Both groundwater seeps mentioned above had low variability like bedrock groundwater but were cooler suggesting a deeper bedrock groundwater 255 source than in the well. Temperatures of the other groundwater seeps were more similar to bedrock groundwater, ranging from 3.6-5.4 °C, and others were more similar to surface water sources, ranging from 3.8-9.4 °C.

Star West Lower
Sources were grouped as precipitation (rain and snow), hillslope groundwater (soil water, riparian water, and toe slope water), 260 and bedrock groundwater and plotted in PCA mixing space ( Figure 5). Minimal variation across all precipitation samples (standard deviation (SD) of 2.4 and 1.1 for PC1 and PC2, respectively) and overlap of snow and rain samples in the mixing space confirmed that it was appropriate to aggregate all samples (snow and rain) taken across all sites (Star Creek, York Creek, and Coleman). Hillslope groundwater exhibited greater variation across samples (SD of 3.8 and 2.0 for PC1 and PC2, respectively), but no clear temporal pattern was observed. Bedrock groundwater chemistry showed slight temporal variation, 270 with more positive values in PC2 in the spring than in the fall (SD of 2.9 and 4.8 for PC1 and PC2, respectively).

Star West Upper
Water sources were similarly grouped as precipitation (rain and snow), hillslope groundwater (soil water, toe slope water, and upper hillslope water), and riparian water (Figure 6a). Bedrock groundwater samples were collected from a lower elevation in the watershed and may not be representative of higher elevation groundwater signatures; therefore, they were excluded as a 275 source at the upper sites. Precipitation clustered tightly in one location except for a couple samples of snow and rain, which increased the SD for precipitation (SD of 2.7 and 2.3 for PC1 and PC2, respectively). All sources showed similar variation as precipitation; hillslope groundwater had a SD of 4.3 and 2.7 for PC1 and PC2, respectively and riparian water had a SD of 3.0 and 2.0 for PC1 and PC2, respectively. A temporal pattern was observed for hillslope water in which hillslope water became less like precipitation from spring to fall (Figure 6b). Temporal variation was also observed across months for riparian water, 280 in which SO4 2concentrations increased from spring to fall (Figure 6b).

Star East Lower
Water sources were grouped as precipitation (snow and rain), soil water, riparian water, groundwater seep and bedrock groundwater (Figure 7a). Precipitation and bedrock sources were the same as those used in SWL. Again, precipitation was 290 tightly clustered in the PCA biplot for SEL (SD of 1.4 and 1.1 for PC1 and PC2, respectively). Hillslope groundwater samples were initially grouped together as a single source but high standard deviations and clustering within the group suggested the separation of riparian water (SD of 1.0 and 1.5 for PC1 and PC2, respectively) and soil water (SD of 4.3 and 1.5 for PC1 and PC2, respectively) as individual sources. Soil water was most different from stream water and varied from spring to fall (increased Ca 2+ and Mg 2+ concentrations). Riparian water was most similar to stream water and did not vary over the season. 295 Bedrock groundwater was clustered in a linear pattern, but no temporal variation was observed. A single groundwater seep that was chemically similar to stream water but temperatures were consistently cool was retained to aid in the explanation of stream water dynamics (Figure 7b).

Star East Upper 305
Water sources were grouped as precipitation (rain and snow), hillslope groundwater (soil water, riparian water, and toe slope water), and groundwater seep (Figure 8a). Precipitation was clustered in a linear pattern (SD of 2.4 and 1.1 for PC1 and PC2, respectively) but did not vary temporally. Large variation was observed for hillslope groundwater (SD of 9.3 and 7.3 for PC1 and PC2, respectively). Toe slope water and riparian water had some chemical dissimilarities but were not different enough from each other, or soil water, to be considered as different groups. No temporal pattern was observed for toe slope water or 310 riparian water chemistry. Soil water varied temporally where it was more similar to riparian water and stream water in the spring and less like all stream water and all other samples (higher concentrations of ions) in the fall. As in SEL, a single groundwater seep that was chemically similar to stream water but temperatures were consistently cool was retained to aid in the explanation of stream water dynamics (Figure 8b).

Stream water characterization
Stream water chemistry for all four sites showed temporal variation throughout the months of open-water flow but little variation between years. As a result, the temporal pattern of stream water was characterized for each site in general for both 2014 and 2015 combined. Further, due to the lack of source water samples during winter months, the temporal pattern of stream water was characterized from April to October, which represents the dynamic hydrologic period from the beginning of 325 snowmelt through to the start of the next year's snow accumulation period.

Star West Lower
PCA analysis indicated that 87 % of the variation in the data was explained by the first two PCs. Stream water samples were contained within an outer boundary created by source water when the temporal variation in source water was considered, but not if mean values of source water were considered ( Figure 5). In April, stream water was most similar to the hillslope 330 groundwater. Stream water transitioned through May to become most similar to precipitation source water in June. In July, stream water was slightly more similar to hillslope groundwater and bedrock groundwater. In August, September and October, stream water chemistry was more variable and was similar to precipitation and hillslope and bedrock groundwater. The temporal pattern associated with stream water variation through the fall was perpendicular to the direction of the bedrock temporal pattern, suggesting that hillslope groundwater, rather than bedrock groundwater, was driving stream water chemistry 335 in the fall.

Star West Upper
PCA analysis indicated that 77 % of the variation in the data was explained by the first two PCs (Figure 6a). Sources formed two ends of a spectrum for stream water mixing rather than a triangle or polygon defining the mixing space. It is possible that a source was missed in the sampling campaign but when the variation in sources was considered, the stream water was bounded 340 by the two points. In April, stream water was most similar to hillslope groundwater and riparian water. Stream water transitioned through May and was more similar to precipitation in June and July. Stream water was similar to hillslope groundwater again through August, September, and October, but did not follow the same pathway as in the early summer months. These temporal differences in stream water between spring and fall are similar to the temporal differences observed in riparian water from April to October rather than hillslope water (Figure 6b). It is likely that there was mixing between 345 riparian water and stream water producing a similar temporal variation in their chemical signature.

Star East Lower
PCA analysis indicated that 86 % of the variation in the data was explained by the first two PCs (Figure 7a). Stream water samples were mostly contained within an outer boundary created by source water when the temporal variation in source water was considered with the exception of September/October stream water, which plotted outside this boundary. In April, stream 350 https://doi.org/10.5194/hess-2020-105 Preprint. Discussion started: 16 March 2020 c Author(s) 2020. CC BY 4.0 License.
water was most similar to the hillslope or bedrock groundwater. Stream water transitioned through May and was most similar to precipitation in June. In July and August, stream water became dissimilar from precipitation and was once again similar to riparian water or bedrock groundwater. In September and October, stream water was less similar to riparian water and plotted outside the boundary created by the identified sources. Since stream water was not contained within the boundary created by the source water, it is likely that a source was missed from the analysis. The temporal variation in the groundwater seep 355 followed the same pattern as the September/October stream water, suggesting the same source water for the groundwater seep and late fall baseflow (Figure 7b). Consistently cool temperatures of the seep (2.2-3.7 °C) suggest a deeper groundwater source.

Star East Upper
PCA analysis indicated that 83 % of the variation in the data was explained by the first two PCs (Figure 8a). Similar to SEL, some of the stream water samples were outside the boundary formed by the stream water sources. In April, stream water was 360 most similar to hillslope water. Stream water was similar to precipitation in May and June. In July and August, stream water was similar to hillslope groundwater but not as much as in April. In September and October, stream water became chemically dissimilar from all identified sources and plotted outside the boundary created by these sources. As in SEL, the temporal variation observed in the stream in September and October was also observed in the groundwater seep (spring to fall; Figure   8b). Again, it is likely that there was another source contributing to streamflow in the fall that was not captured by the field 365 sampling. This was likely a deeper groundwater source due to consistently cool temperatures (2.5-3.5 °C) observed for the groundwater seep.

Temporal and spatial variation in source water signatures
Chemical signatures of source water have been shown to vary seasonally and annually (Rademacher et al., 2005) as well as 370 spatially across sub-watersheds (James and Roulet, 2006). As a result, James and Roulet (2006) suggested that only source water from within individual sub-watersheds of interest should be used in un-mixing calculations. Inamdar et al. (2013) further argued that mixing proportions should not be calculated because multiple assumptions are often violated and can lead to significant errors in un-mixing proportions. Rather, temporal and spatial variation in stream water and source water should be examined and used to describe or to develop a physically-based conceptualization of runoff mechanisms. 375 Two of the key assumptions for EMMA, the chemical composition of sources does not change over 1) the time scale considered or 2) with space (Hooper, 2003;Inamdar, 2011), were violated in this dataset. Source water chemistry varied greatly across the watersheds. For example, when all hillslope samples from each sub-watershed were projected into the mixing space created by stream water at the watershed outlet (SM), large variability was evident between sites (Figure 9). While there was some overlap between some sites (SWL and SEU), SWU was clearly different than the other hillslope samples. As a result, source 380 water from within individual sub-watersheds was used to reduce the uncertainty associated with large spatial variability in https://doi.org/10.5194/hess-2020-105 Preprint. Discussion started: 16 March 2020 c Author(s) 2020. CC BY 4.0 License. source water chemistry. However, the variability within sites was also quite large. The coefficient of variation (CV) of source water was often larger than the CV of the stream water (there should be little to no variation in source water over time ;James and Roulet, 2006;Inamdar, 2011), particularly for K + . The occasions where source water CV were smaller than stream water CV for most ions were for seeps in SEU and SEL, bedrock groundwater in SWL and SEL, and hillslope and riparian water in 385 SWU. Despite the violation of assumptions, interesting and interpretable temporal trends in source water chemistry were observed. 390 For example, temporal variations in riparian water in SWU were observed from spring to fall and followed the same pattern observed in stream water chemistry in May compared to September/October (Figure 6b). It is not clear if the stream chemistry responded to variation in riparian chemistry or if riparian water responded to stream chemistry, but these pools of water were likely mixing to create the same temporal pattern rather than the variation in hillslope water chemistry influencing the stream.
Other studies have also shown the importance of the riparian zone for "buffering" water chemistry and influencing stream 395 water contributions, although often at the event scale (McGlynn and Seibert, 2003;Grabs et al., 2012). Spencer et al. (2019) presented water table dynamics in hillslope groundwater wells and suggested that the upper hillslope and the stream were only connected during the spring freshet. This would suggest that hillslope water chemistry should reflect the dilution from snowmelt and the subsequent drying of the hillslope, which should increase the concentration of water held in the soil from spring to fall. The corresponding temporal pattern in hillslope water chemistry was observed in SWU ( Figure 6b) and for soil 400 water chemistry in SEL and SEU.
Bedrock groundwater was consistent between years, but there is uncertainty around the variability of bedrock groundwater across the watershed. Although most of the stream is situated within the same geologic formation, there may be differences in bedrock groundwater chemistry associated with heterogeneous sedimentary layers or contact time (Freeze and Cherry, 1979).
Groundwater seeps that were sampled along the stream length could be used as a potential indicator of groundwater variability 405 because they may flow through various formations or flow pathways. Temperature signals from seeps suggested some were groundwater-fed (consistent cool temperatures) and others were fed by shallow subsurface water (larger fluctuations in temperature; Taniguchi, 1993). However, ion concentrations of the seeps were not chemically distinct because they were generally similar to stream water or hillslope groundwater. This suggests that there are likely many complex subsurface flow pathways making it difficult to differentiate between subsurface sources. Other tracers such as oxygen and hydrogen isotopes 410 or non-conservative tracers such as nitrogen and dissolved organic carbon may help to better differentiate between seeps, hillslope groundwater, and bedrock groundwater (e.g., Cowie et al., 2017;Ali et al., 2010;Orlova and Branfireun, 2014). More wells in the bedrock and till are also required to characterize the variability in groundwater more robustly across the watershed (Rinderer et al., 2014) and improve the EMMA results.

Temporal variation in stream water contributions 415
Stream water contributions can be generalized for all sub-watersheds in a number of ways. The water that was stored in the hillslope was likely the first to reach the stream in the early spring prior to high flow as snowmelt started to saturate the landscape. The delivery of old water to the stream at the onset of snowmelt is similar to the flushing mechanism observed in the Turkey Lakes Watershed in central Ontario, Canada (Creed and Band, 1998) where high nitrogen concentrations were observed prior to peak streamflow. McGlynn et al. (1999) also observed the displacement of old water to the stream at the 420 onset of snowmelt in Sleepers River Research Watershed in Vermont, USA and suggested this was due to a small volume of snowmelt being added to a large storage of water already in the subsurface. This initial displacement of old water was likely followed by a dilution effect, where large volumes of low concentration snowmelt mixed with soil water and contributed to streamflow. Snowmelt was the major event that produced a water table response in all wells and connected the hillslopes to the stream . The initial snowmelt period was also the only time overland flow was observed at the study 425 site. Other studies have also reported that snowmelt creates a dilution response in the stream (Rademacher et al., 2005;Cowie et al., 2017). Conversely, the opposite has also been observed whereby a previously disconnected source was connected to the stream and caused an increase in solute concentrations (McNamara et al., 2005). Although this was the main period of hydrologic connectivity in Star Creek, we did not observe the increase in stream water chemistry associated with newly connected sources. 430 Stream water contributions were more similar within Star East (SEL and SEU) and Star West (SWL and SWU) sub-watersheds than between alpine/sub-alpine (SEU and SWU) and upper montane (SEL and SWL) sub-watersheds. PCA plots for SEL and SEU showed that stream water was most like precipitation in May and June, whereas this dilution effect occurred in June and July in SWL and May to July in SWU. The delayed response in SWL and SWU is consistent with the watershed storage https://doi.org/10.5194/hess-2020-105 Preprint. Discussion started: 16 March 2020 c Author(s) 2020. CC BY 4.0 License. estimates from baseflow recession analysis  that suggested that the west fork sub-watersheds had a larger 435 storage capacity than the east fork sub-watersheds. Accordingly, more water would be required to fill storage before saturation or hydrologic connectivity occurred.
Differences in the east and west forks were also evident later in the year. In SWL and SWU, stream water was chemically similar to hillslope groundwater in the fall (Figures 5 and 6a). In SEL and SEU, stream water was similar to hillslope groundwater in August but fell outside the boundaries created by the identified sources in September and October (Figures 7a  440   and 8a). A groundwater seep in SEL and SEU followed similar temporal patterns as stream water from spring to fall (Figures   7b and 8b) and may provide insights into the sources of stream water in the fall. Consistently cool temperatures of the groundwater seeps suggest they have a deeper bedrock groundwater source different than the bedrock groundwater well. Star Creek has spatially heterogenous surficial deposits and geology (AGS, 2004) which likely has a large influence on groundwater chemistry throughout the watershed. It is possible, therefore, that additional bedrock groundwater sources were contributing 445 to streamflow in Star Creek.
Topographic transitions and convergent zones have been associated with groundwater contributions to streamflow (Covino and McGlynn, 2007;Hjerdt et al., 2004). While minimal groundwater discharge occurred over the mountain front recharge zone in Humphrey Creek, southwest Montana, USA, considerable groundwater discharge was observed in the valley bottoms (Covino and McGlynn, 2007). Large increases in the concentration of stream water ions that may be associated with strong 450 groundwater upwelling were not evident between April and October or along the length of the stream. However, chemical signatures of groundwater seeps suggest that some bedrock groundwater may not be distinguishable from stream water.
Consequently, these transitions may not be visible in water chemistry along the length of the stream.
Although contamination of the till groundwater well has limited the inferences that could be made from water chemistry, water levels in the bedrock and till groundwater wells do provide some insights into their contributions to streamflow. The till 455 groundwater well showed consistently slower recession curves compared to the bedrock groundwater well ( Figure 10). Similar to the post-glacial landscape in Sleepers River watershed (Shanley et al. 2015), slower drainage from till groundwater may be partly responsible for maintaining streamflow during late summer. Heterogeneous glacial till deposits with different physical characteristics were also linked to the variable release of stored water, and thus the variability in baseflow, in the Scottish Highlands (Blumstock et al., 2015). Glacial till in the Rocky Mountains can be spatially heterogenous and likely has multiple 460 flow pathways within it (Langston et al., 2011). Clay lenses can create perched water tables that have different response times than the rest of the till matrix (Evans et al., 2000) or create complex groundwater flow pathways (Freeze and Witherspoon, 1967).
The complexity in subsurface flow pathways that appears to influence runoff generation in Star Creek watershed may be important in evaluating watershed resilience to disturbance. Primarily vertical percolation or groundwater recharge (Spencer 465 et al., 2019) and delayed or variable release of stored water could mute the impact of disturbance on peak flows. However, excess water associated with forest disturbance would infiltrate into the subsurface, which may subsequently increase baseflows. Star Creek sub-watersheds have a large storage capacity (300-450 mm) in glacial till and fractured bedrock (Spencer https://doi.org/10.5194/hess-2020-105 Preprint. Discussion started: 16 March 2020 c Author(s) 2020. CC BY 4.0 License. et al., 2019). When compared to average annual precipitation (950 mm), it is possible that storage can mitigate the effects of the increased net precipitation reaching the ground following disturbance by temporarily holding and slowly releasing the 470 excess water. Harder et al. (2015) also postulated that complex subsurface flow pathways and large storage capacity may be partly responsible for the resiliency observed in Marmot Creek watershed (Alberta, Canada) following disturbance; however, more research is needed to determine the influence of disturbance on groundwater.

Conclusions
Stream and source water chemistry in four sub-watersheds of Star Creek showed that Star East (SEL and SEU) and Star West (SWL and SWU) sub-watersheds were more similar than alpine/sub-alpine (SEU and SWU) and upper-montane (SEL and SWL) physiographic zones. PCA plots were used to conceptualize runoff generation. In general, old water reached the stream 480 first at the onset of spring melt in all sub-watersheds. This was followed by a dilution effect as the snowmelt saturated the landscape and the hillslope was connected to the stream. Fall baseflows differed between Star East and Star West forks. Star West stream water was once again similar to hillslope water, but Star East was unlike all measured sources. Slower recession rates (and likely lower hydraulic conductivity) in the till groundwater well than in the bedrock groundwater well suggest that water recharged into the till groundwater may be slowly released to the stream, thereby muting the effects of disturbance on 485 peak flows. While this large storage zone may be an important factor in watershed resilience to disturbance that has been observed in front-range Rocky Mountain watersheds in Alberta, the influence of disturbance on groundwater dynamics is not well understood. More research on the variability of bedrock and till groundwater chemistry is needed to clarify the difference between these sources and their contributions to streamflow throughout the year. https://doi.org/10.5194/hess-2020-105 Preprint. Discussion started: 16 March 2020 c Author(s) 2020. CC BY 4.0 License.