Given projected increases in stream temperatures attributable to global change, improved understanding of relationships between stream temperatures and hyporheic exchange would be useful. We conducted two conservative tracer injection experiments in a glacial meltwater stream, to evaluate the effects of hyporheic thermal gradients on exchange processes, including preferential flow paths (PFPs). The experiments were conducted on the same day, the first (a stream injection) during a cool, morning period and the second (dual stream and hyporheic injections) during a warm, afternoon period. In the morning, the hyporheic zone was thermally uniform at 4°C, whereas by the afternoon the upper 10 cm had warmed to 6–12°C and exhibited greater temperature heterogeneity. Solute transport modeling showed that hyporheic cross-sectional areas (As) at two downstream sites were two and seven times lower during the warm experiment. Exchange metrics indicated that the hyporheic zone had less influence on downstream solute transport during the warm, afternoon experiment. Calculated hyporheic depths were less than 5 cm, contrasting with tracer detection at 10 and 25 cm depths. The hyporheic tracer arrival at one downstream site was rapid, comparable to the in-stream tracer arrival, providing evidence for PFPs. We thus propose a conceptual view of the hyporheic zone in this reach as being dominated by discrete PFPs weaving through hydraulically isolated areas. One explanation for the simultaneous increase in temperature heterogeneity and As decrease in a warmer hyporheic zone may be a flow path preferentiality feedback mechanism resulting from a combination of temperature-related viscosity decreases and streambed heterogeneity.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 Temperature and hyporheic exchange both play important roles in stream ecosystem structure and function. Temperature influences dissolved oxygen concentrations, stream metabolism, biogeochemical cycling, the suitability of habitat for aquatic organisms, and the timing of important life cycle events such as spawning [Caissie, 2006; Kaushal et al., 2010]. Hyporheic exchange and flow path characteristics such as residence time and length influence the biogeochemical cycling of nutrients, dissolved organic carbon, and inorganic solutes [Chapra and Runkel, 1999; Gooseff et al., 2002; Holmes et al., 1996; Kaplan and Newbold, 2000]. Several studies have shown that warming trends in streams and rivers throughout the United States and Europe are attributable to increases in effluent discharges, impoundments, and runoff from impervious surfaces [Kaushal et al., 2010; Webb, 1996; Webb and Nobilis, 1995, 2007]. Other studies project further warming under the influence of climate change [Eaton and Scheller, 1996; Gooseff et al., 2005; Mohseni et al., 1999]. Hyporheic exchange may play a role in ameliorating the ecosystem impacts of warming trends by buffering stream temperature and creating thermal refugia within the channel [Arrigoni et al., 2008; Cozzetto et al., 2006; Story et al., 2003].
 In this context, a better understanding of the influence of temperature on hyporheic exchange would be useful. The heterogeneity of flow paths within the hyporheic zone may influence interactions between stream temperature regimes and hyporheic exchange. For many situations it is useful to represent the hyporheic zone as a zone of uniform hydraulic characteristics which remains static and in which solutes are well mixed or are transported in a diffuse manner spreading throughout the zone [Harvey and Wagner, 2000; Kaplan and Newbold, 2000]. However, recent studies of hyporheic zone processes have begun to investigate the heterogeneity in structure and transport caused by the opening of preferential flow paths (PFPs). Along PFPs, water will flow and solutes will be transported at local rates faster than the average rates throughout the hyporheic zone. PFPs may form, for example, along fractures, result from macropores created by roots or burrowing organisms, or form in areas with higher permeability than the surrounding matrix [Burkholder et al., 2008; Menichino et al., 2012]. Geophysical techniques are emerging as powerful tools for the study of hyporheic zone hydraulic characteristics. Using electrical resistivity imaging, Menichino et al.  recently observed that macropores created PFPs in river meander bends not only shortening solute travel times through the hyporheic zone but also increasing the degree of solute tailing.
 The potential ecological significance of PFPs in hyporheic zones was demonstrated by Wagner and Bretschko , who showed that there are measurable differences in the velocity of water in these flow paths which in turn can influence the delivery of nutrients and oxygen to a patchy distribution of invertebrates. The ecological significance of hyporheic zone PFPs in creating cool-water refugia for fish and other organisms has been shown by Burkholder et al. . They estimated that 76% of the flow in the hyporheic zone of a gravel bar in a large gravel-bed river was along PFPs. Poole et al.  found that shorter hyporheic flow paths (tens of meters or less) affect the daily range in stream temperatures without influencing the mean daily temperature while longer flow paths (hundreds of meters) have the potential to affect the mean. Shorter flow paths tend to be more numerous in a given hyporheic volume, however, and thus have a more spatially extensive influence on temperature while longer flow paths tend to emerge at distinct locations making their temperature effect more localized [Poole et al., 2008].
 Hydraulic characteristics of hyporheic PFPs, in some situations, may be less stable than the characteristics for diffuse flow paths found for the bulk of a hyporheic zone. In flume studies, for example, Chen et al.  observed that deposition of colloidal particles in the near-streambed subsurface altered hyporheic flow paths, increasing spatial variability in pore water flow. Although the colloid deposition resulted in an overall decrease in permeability, the development of hyporheic PFPs was also observed, leading to more rapid hyporheic exchange in some locations. In alluvial floodplains, PFPs may become activated with changes in river stage [Heeren et al., 2011]. Similarly, dynamic changes in temperature could influence flow in hyporheic PFPs due to the temperature sensitivity of water density and dynamic viscosity, both of which decrease for temperature increases above 4°C. Using numerical simulations, Cardenas and Wilson  found that temperature-induced changes in water viscosity influenced fluid flow and temperature patterns within the hyporheic zone. Temperature influences may be significant for the many stream ecosystems in which temperatures vary substantially on daily time scales (e.g., deserts and alpine ecosystems), seasonal timescales, or in response to disturbances (e.g., wildfire) [Brown and Krygier, 1970; Caissie, 2006; Constantz, 1998; Constantz et al., 1994; Cozzetto et al., 2006; Dunham et al., 2007; Hitt, 2003; Johnson and Jones, 2000; Kaushal et al., 2010; Nelson and Palmer, 2007; Webb et al., 2008].
 The purpose of the study presented here was to explore how temperature-driven changes in density and dynamic viscosity might influence the interactions between temperature, hyporheic exchange and associated PFPs. Glacial meltwater streams of the McMurdo Dry Valleys, Antarctica, provide an excellent setting for studying interactions between temperature variations and hyporheic exchange because they are not influenced by a large-scale groundwater system nor by riparian vegetation. Our approach involved conducting two conservative tracer experiments on the same day in a dry valley stream. One experiment, a chloride injection into the stream, was conducted under cool conditions in the morning. The second experiment was conducted under warm conditions in the afternoon and involved simultaneous injection of chloride into the stream and bromide into the hyporheic zone. When the stream temperatures were cool the hyporheic zone was cold and thermally uniform with depth, but later on when the stream temperatures were warmer the upper layer of the hyporheic zone was also warmer with greater thermal gradients. We found that the hyporheic zone cross-sectional area and the overall influence of hyporheic exchange on the downstream transport of solutes were higher under the cool stream temperatures. The tracer data and modeling are consistent with a conceptual view of the saturated alluvium as containing discrete PFPs weaving their way through hydraulically isolated areas. We propose that viscosity decreases due to increases in stream and hyporheic temperatures could create a flow path preferentiality feedback mechanism that could explain the differences observed in the hyporheic storage cross-sectional areas.
2. Study Site
 The McMurdo Dry Valleys comprise the largest ice-free area in Antarctica (Figure 1). The region lies among the Transantarctic Mountains near McMurdo Sound. The valleys are dominated by barren soils and exposed bedrock, glaciers, permanently ice-covered lakes, and well-established stream channels. The region is an extreme cold desert with mean annual air temperatures between −15 and −30°C [Doran et al., 2002] and precipitation that averages 3–50 cm of water equivalent each year as snow [Fountain et al., 2010]. Soils remain frozen year round except for a <1 m active layer that overlies continually frozen permafrost [Cartwright and Harris, 1981]. There is little movement of groundwater in this frozen ground, and the valleys do not have a large-scale groundwater system [Cartwright and Harris, 1981]. Dry valley streams do however have lateral inflows. These inflows though are from hyporheic water upstream of the reach that has been transported downstream and is re-entering the stream.
 During the warmest months of summer, December and January, air temperatures may rise to −1 and 0°C or warmer [Clow et al., 1988] causing glacial meltwater generation [Fountain et al., 2010] and sustaining stream flow. The stream channels cut through the unconsolidated alluvium and the stream banks are unstable due to the absence of riparian vegetation (supporting information Figures S1 and S2). Many of the streams in Taylor Valley are monitored as part the McMurdo Dry Valleys Long-Term Ecological Research project (MCMLTER). Flows are highly variable on daily and interannual bases [Conovitz et al., 1998; House et al., 1995]. Daily flow changes occur due to increased melt taking place when the sun shines directly on the vertical faces of the glaciers. Daily stream temperature changes average 6–9°C, and maximum temperatures typically range from 8 to 15°C [Cozzetto et al., 2006]. The hyporheic zone surrounding the stream is dynamic as well, expanding gradually throughout the summer as the active layer thaws [Conovitz et al., 2006] until late in the summer when the active layer starts to freeze again. Hyporheic zone temperatures also experience daily cycles, with temperatures at depths of 5 cm changing by almost 8°C in a 24 h period [Cozzetto et al., 2006]. While some streams contain perennial microbial mats that persist in the thalweg through the summer, the invertebrate fauna in these mats is quite sparse and comprised primarily of tardigrades and nematodes, with an absence of burrowing organisms [Treonis et al., 1999].
 The experiments reported here were conducted in Von Guerard Stream in the Lake Fryxell basin in Taylor Valley (Figure 1). Hyporheic sediments in Von Guerard Stream are at least 97% sand and gravel by weight and porosity ranges from 0.29 to 0.36 and averages 0.33 [Conovitz, 2000]. The experimental reach is 142 m long, and was the site of two previous tracer experiments [Cozzetto et al., 2006]. Heat transport simulations for the experiments showed that hyporheic exchange was a significant term (6–21% of nonradiative heat losses) in the heat budget of the stream reach contributing to the buffering of stream temperatures. The overall gradient of the reach was 0.030 m/m, and its sinuosity index was 1.1. The reach does not contain microbial mats and the stream was a single thread during the experiments. The reach was instrumented with Campbell Scientific 107-L temperature probes and thermocouples at 11 stations. Reach widths ranged from 2.4 to 5.5 m. Active layer depths were measured at several points along the stream channel cross section at every downstream thermocouple location as described by Conovitz et al.  and ranged from 26.5 to 38.5 cm with a mean of 34.1 cm.
3.1. Tracer Experiment Overview and Temperature Measurements
 Two conservative tracer injection experiments were conducted in the reach on 16 January 2006. The first surface water only injection experiment took place in the morning between 7:30 and 9:30 when stream temperatures were cool. The second simultaneous surface water and subsurface injection experiment took place in the afternoon between 14:00 and 16:00 when stream temperatures were warmer. The times were selected so that the flows during the experiments would be as similar as possible, occurring after the previous day's peak flow had receded and before the current day's peak flow had arrived. The surface water injections during the morning and afternoon provided information on surface water travel times and average hyporheic characteristics. Concomitantly examining where and when the subsurface bromide injectate emerged in the stream in the afternoon provided information on hyporheic flow paths.
 During the experiments, stream water temperatures were measured at 11 locations: 8.5, 21, 32, 53, 62, 74, 98, 108, 118, 137, and 141 m downstream of the NaCl injection site (Figure 2). Associated streambed temperatures were also recorded at each of the 11 locations at depths of 5, 10, and 20 cm below the streambed surface. Stream temperatures were measured with Campbell 107-L probes while hyporheic temperatures were measured with thermocouple strings (TCs) made from Omega, 0.051 cm diameter, type K chromel-alumel wire installed using 1 in. hollow polyvinyl chloride (PVC) pipes. Temperatures from the 107-L probes and TCs were recorded every minute and were accurate to within ±0.3°C. The TC temperature data were then slightly adjusted after the experiment based on a calibration against a 107-L probe (see supporting information section S1). Mean stream and hyporheic temperatures for the entire reach were calculated by averaging the minute by minute temperatures recorded at each of the 11 T/TC locations and then calculating the weighted average based the proportion of the stream reach represented by each T/TC given the distance between the 11 locations (see supporting information section S2).
3.2. Density, Viscosity, and Hydraulic Conductivity Metrics
 Density (ρ) and dynamic viscosity (µ) are both characteristics of a fluid that change with temperature with μ changing to a much greater extent for a given temperature change. Both ρ and µ influence hydraulic conductivity (K), as described by the following equation:
where k is the intrinsic permeability of the subsurface matrix (i.e., sediments) [Charbeneau, 2000] and g is the gravitational constant. By assuming that k is constant during each experiment and throughout the reach and the top 20 cm of the hyporheic zone, the method of Constantz et al.  can be used to calculate a ratio of warm to cool period hydraulic conductivities at the stream/streambed interface using:
 Ratios of warm to cool experiment dynamic viscosities and hydraulic conductivities associated with averaged temperatures over the reach were calculated for the stream/streambed interface (0 cm) and for hyporheic depths of 5, 10, and 20 cm [Dingman, 2002]. Temperature data at each of the 11 locations and four depths (0, 5, 10, and 20 cm) were also used to calculate relative differences in dynamic viscosities and hydraulic conductivities between the warmest and coolest locations in the top 20 cm of the hyporheic zone.
3.3. Tracer Experiments
 During both experiments, NaCl was injected for 2 h at a constant rate into the surface water in the stream center at a site with rapid mixing. Previous tracer experiments done in the Dry Valleys have shown that chloride behaves conservatively [McKnight et al., 2004; Runkel, 1998]. Estimates of the lateral mixing lengths for the two injections [Chapra, 1997] and experience from past experiments indicated that chloride should have been fully mixed into the stream by the sampling location 5 m downstream of the injection site.
 Injection rates averaged 2.82 mL/s with a coefficient of variation of 0.008 during the cool experiment and 2.78 mL/s with a coefficient of variation of 0.030 during the warm experiment. During the afternoon experiment, NaBr was simultaneously injected into the streambed by dumping 1 L of a 30,000 mg NaBr/L solution into a well at 14:00. The well was constructed of 2.5 in. PVC pipe following the procedure outlined by Gooseff et al. . It was screened at a depth 8–13 cm below the surface and was located 1.5 m upstream from the sodium chloride injection site. Hyporheic water flowing through the well flushed out the bromide. Bromide was used for the subsurface injection because its low background levels in the stream and hyporheic zone, typically less than 0.1 mg/L, made elevated concentrations easy to detect. Soil-water sorption experiments done elsewhere indicated that bromide was conservative [Levy and Chambers, 1987].The statistical analysis to determine background concentrations of chloride and bromide are described in the supporting information section S3.
 Stream and hyporheic water samples were collected at a site 3 m upstream of the injection and at sites located 5, 137, and 142 m downstream of the injection (Figure 2, see further explanation in the supporting information section S4).Two wells made from 0.43 cm inner diameter stiff, polyethylene tubing were installed at each sampling location in the thalweg. One well screened a depth of 8–12 cm and the other a depth of 23 to 27 cm and are thus referred to as the shallow/10 cm wells and the deep/25 cm wells, respectively. For the dual injection, afternoon experiment, additional shallow and deep well sites were installed in the thalweg at 1, 2.5, 7.5, and 51 m to attempt to capture the movement of bromide through the system.
 Background stream and hyporheic samples were collected once every 15–30 min. Other stream samples were collected every 5 min for the first hour of the injection, every 15 min during the second hour of the injection, and every 5 min again during the hour after the injection was stopped. Well samples were collected using peristaltic pumps every 15–30 min during the injections and for 1 h after the injections stopped. Well water was cleared before samples were collected. Samples were filtered through 0.45 μm Pall Gelman filter capsules, collected in high-density polyethylene (HDPE) plastic bottles, and kept chilled until analysis. Analysis for chloride was done on a Dionex ion chromatograph by the University of Colorado's Laboratory for Environmental and Geological Studies.
3.4. Connectivity Calculations
 The method of Triska et al.  was used to calculate the hydrologic connectivity of the 137 and 142 m well locations to the stream. The water sampled from the wells is a combination of chloride tracer-labeled stream water and hyporheic water at background chloride levels. Combining the water and chloride mass balance equations, allows the ratio of stream water to well water to be determined:
 Where, V represents volume of water, Cl represents chloride concentration (mg/L), w represents well, s represents stream, and h represents hyporheic water. Equation (3) assumes that stream, hyporheic, and well waters all have the same density. The proportion of surface water in the well can then be computed as follows:
 Connectivity as defined by this ratio is the proportion of well water derived from the stream. Note that it represents the connectivity of the stream to the location of the hyporheic zone being sampled by the well for the conditions of the tracer experiment (i.e., time scale of injection, flow conditions, etc.). We consider this metric useful because we sought to replicate conditions as closely as possible between the two tracer experiments (with the exception of the thermal regime).
 Separate connectivities for each well and experiment were determined resulting in a total of eight connectivity values. The hyporheic chloride concentrations used were the background values associated with the well samples collected at the start of each injection, i.e., at 7:30 or 14:00, before chloride would have had a chance to reach the wells. The stream concentrations used were plateau values. The well chloride concentrations used were typically the plateau concentration in the well for the experiment under consideration. In the one instance in which a well plateau was not reached (142 m shallow well, cool experiment), the maximum well concentration was used instead. In the one instance (142 m deep well, warm experiment) in which the stream and well plateau concentrations were within 1% of one another, which was in the range of analytical error, we considered the two plateau values to be essentially the same and set the connectivity value equal to 100%.
3.5. Solute Transport Modeling
 The One-Dimensional Transport with Inflow and Storage model with parameter estimation, OTIS-P [Runkel, 1998] was used to simulate stream chloride concentration data from the tracer experiments.
 Four parameters, the transient storage area (As), dispersion (D), the storage exchange coefficient (α), and the main channel cross-sectional area (A) were optimized to simulate chloride concentrations at the 137 and 142 m stream sites. Although the two sites were close to one another, we modeled both for comparison. Q is the average in-stream volumetric flow rate, C is the main channel solute concentration, CS is the storage zone solute concentration, CL is the lateral inflow solute concentration, qL is the lateral inflow rate, and x and t are direction and time along the stream. Because during the experiments, the reach contained a minimum of surficial dead water areas, we assumed that the transient storage in the model represented the hyporheic zone. The OTIS-P model output also includes residual sum of squares (RSSs), t-ratios, and 95% confidence intervals. RSS values and t-ratios are both goodness of fit measures to assess how well the model tracer curve simulates the data, with lower RSS values and higher t-ratios indicating greater confidence in the results. The 95% confidence intervals are provided for each output parameter (i.e., A, As, α, and D). When comparing the parameters for two experiments, if their 95% confidence intervals do not overlap, this suggests that the values are different at a statistically significant level.
3.5.1. Flows in the Reach
 In order to model hyporheic exchange in a reach using the OTIS-P model, it is necessary to estimate streamflow at the upstream (in our case 5 m) end of the reach. To determine these flows for the two experiments, we used a mass balance approach based on known injectate concentrations and injection pump flow rates. Flows were variable, ranging from 4.8 to 8.0 L/s during the cool experiment and 3.6 to 6.1 L/s during the warm one. Alternatively, the tracer was not as well mixed at 5 m as the narrowing of the stream channel and our estimates of lateral mixing lengths indicated. Because of this flow variation, we modeled the experiments under low, medium, and high flow scenarios in which the flow was held at a steady state in order to assess the sensitivity of the hyporheic parameters to flow. The low and high flow scenarios represent the two ends of the flow range. The medium flow scenarios represent the flows calculated based on average in-stream chloride concentrations. Modeling the experiment under several flow scenarios allowed us to determine the sensitivity of the parameters to upstream flow. For the cool experiment, the three flow scenarios used were 4.8, 6.0, and 8.0 L/s, and for the warm experiment, they were 3.6, 4.6, and 6.1 L/s.
 In addition, the chloride transport modeling required that a value be entered representing either the average inflow of water to the stream (gaining reach) or the average outflow of water from the stream (losing reach). For the various flow scenarios, the dilution of chloride concentrations between the 5 and 142 m sites indicates that the reach as a whole was gaining water. To obtain an average inflow rate, we first used background well concentrations at the −3 m site upstream of both the surface water and subsurface injections, available throughout both experiments and background well concentrations at the 5, 137, and 142 m sites, available before the start of the first experiment (i.e., before 7:30) to estimate inflow chloride concentrations in the lateral inflow. We then used these estimates in a mass balance making use of the tracer dilution to calculate downstream flows at 137 and 142 m. The background well concentrations used to determine the lateral inflow concentrations represent values at particular locations within the hyporheic zone, and given natural variability, we do not interpret these data as an exact measure of average chloride concentrations in the inflow. The initial estimates of lateral inflow concentrations and calculated downstream flow values were thus entered into the OTIS-P model, and the model was then used to test various inflow values associated with the range of background chloride concentrations observed in the wells so as to select the inflow value that produced the closest match between simulated and observed downstream (137 and 142 m) concentrations.
 The inflow volumes thus determined, in effect, took into account any water losses due to evaporation and were added to the associated upstream discharges to calculate flows at the downstream end of the reach. Flows at 137 and 142 m appeared to stabilize not showing the same degree of variability as at the upstream location. Downstream flows were within 5% of one another for each site/experiment, no matter what the upstream flow scenario used to calculate them. At 137 m, flows averaged 8.2 and 7.3 L/s for the cool and warm experiments, respectively. At 142 m, they averaged 9.1 and 7.3 L/s for the cool and warm experiments, respectively. Although the reach as a whole was gaining water during both experiments, the 137–142 m subreach was gaining water during the cool experiment and was neutral or losing water during the second experiment. Table 4 provides details of model parameter values.
3.5.2. Chloride Concentrations Used in the Modeling
 The OTIS-P model of hyporheic exchange requires that a value be entered for the stream chloride concentration at the upstream (in our case 5 m) end of the reach, before, during, and after the injection. Chloride concentrations in-stream samples collected at the −3 and 5 m sites before the start of the morning experiment were used to determine a precool experiment in-stream concentration of 7.5 mg Cl/L. For the chloride concentrations during the two injections, we modeled both experiments using single plateau chloride concentration values corresponding to the associated steady flow scenario being modeled. In addition, for the morning injection, we also evaluated a second scenario that involved a double chloride plateau during the experiment (see supporting information section S5).
 Between the two experiments, the stream chloride concentrations at 5 m did not return to the same background levels observed at the 5 m site before the injection took place nor did they return to the same levels at the site 3 m upstream of the injection as observed between the two experiments. The approach for addressing these effects is described in supporting information section S6. At the end of the second, warm injection experiment, the in-stream 5 m chloride concentration did return to background levels and was set at a value of 8.0 mg Cl/L.
3.5.3. Calculation of Hyporheic Metrics
 The exchange metrics which were calculated include: advective velocity (u), the, the storage exchange flux (qs) [Harvey et al., 1996], the ratio of storage to main channel areas (As/A), the fraction of median travel time due to transient storage (Fmed200) [Runkel, 2002] and the storage zone depth (ds) [Harvey and Wagner, 2000]. Associated equations are provided below. The depth of the storage zone was calculated based on the value for As and assuming a 0.33 porosity (n) for the streambed sediments [Conovitz, 2000; Harvey and Wagner, 2000] and an average reach-wide stream width (w) of 3.8 m for both experiments. The equation for calculating storage zone depths also assumes that all storage occurred in the hyporheic zone, not in side-pool “dead zones” and further that it occurred directly beneath the channel and not in the stream banks. The fraction of median travel time due to storage was calculated at a standardized reach length (Lstd) of 200 m. In addition to using the RSS, t-ratio, and 95% confidence interval output from the OTIS-P modeling described above, the reliability of parameter estimates was evaluated by calculating the Damkohler number (DaI) [Wagner and Harvey, 1997], a dimensionless number that compares time scales of transport processes with those of storage processes, such that as DaI approaches 1, the reliability of parameter estimates increase. This number is calculated based on the actual reach length (L = 142 m). The associated equation is provided below.
4.1. Stream and Hyporheic Temperatures, Densities, Dynamic Viscosities, and Hydraulic Conductivities
 Midway through the cool experiment (8:30), temperatures at all depths in the hyporheic zone were within 2°C of the stream temperature (Figure 3a). Extreme values in the hyporheic zone are summarized in Table 1. Temperatures were at their lowest, 1.8°C, at a 20 cm depth at 74 m and at their highest, 6.0°C, at a 5 cm depth at 137 m, corresponding to a 4.2°C temperature range in the monitored top 20 cm of the hyporheic zone. In contrast, during the warm afternoon, experiment, the temperatures were much warmer in the upper layer of the hyporheic zone compared to those at the deeper sites (Figure 3b). The lowest hyporheic temperature at 15:00 midway through the warm experiment, 2.4°C, also occurred at 20 cm depth at the 74 m site. The highest temperature was 12.8°C at a depth of 5 cm at 118 m, which was about 2°C warmer than the overlying stream water. Thus the temperature range during the warm experiment in the top 20 cm of the hyporheic zone was 10.4°C, which is about 2.5 times greater than that during the cool experiment.
Table 1. Comparisons of Water Physical Properties (Density, Dynamic Viscosity, Hydraulic Conductivity) Corresponding to Locations of Hyporheic Temperature Extremes Midway Through Each Experimenta
Cool Experiment at 8:30
Warm Experiment at 15:00
Location—site (m), depth (cm)
Dynamic viscosity (kg/m s × 10−3)
Maximum/Minimum Ratio (% Difference)
Maximum/Minimum Ratio (% Difference)
Minimum and maximum of the temperatures measured at the stream-streambed interface and at three hyporheic depths (5, 10, and 20 cm) along the experimental reach midway through each experiment (8:30 and 15:00). Corresponding densities and dynamic viscosities. Comparisons of dynamic viscosities and hydraulic conductivities between the locations of hyporheic temperature extremes for a single experiment. Ratio refers to value associated with maximum temperature: value associated with minimum temperature; % difference: ((value associated with maximum temperature—value associated with minimum temperature)/value associated with minimum temperature) × 100.
 Compared to the morning period, warmer, more variable, average temperatures were observed throughout the stream-hyporheic system during the afternoon period. In Table 2, temperatures as averaged over the whole reach for the entire 2 h period of the injection are presented for the stream and for three hyporheic depths for both experiments. As an example, average temperatures at 5 cm depth during the warm experiment were 4.6°C higher than during the cool experiment.
Table 2. Comparisons of Average Water Physical Properties (Density, Dynamic Viscosity, Hydraulic Conductivity) Along the Stream Reach for the Cool and Warm Tracer Experimentsa
Stream (0 cm)
5 cm Depth
10 cm Depth
20 cm Depth
Dynamic viscosity (kg/m s × 10−3)
Warm/Cool Ratio (% Difference)
Stream (0 cm)
5 cm Depth
10 cm Depth
20 cm Depth
Temperatures as averaged over the entire stream reach and the 2 h injection time period and the corresponding densities and dynamic viscosities for the stream-streambed interface (0 cm) and for three depths in the hyporheic zone for each experiment. Comparisons of dynamic viscosities and hydraulic conductivities between the two experiments. Ratio: = warm experiment value: cool experiment value; % difference = ((warm − cool)/cool) × 100.
 These temperature differences corresponded to differences in density, dynamic viscosity, and hydraulic conductivity between the warm and cool experiments, which are also presented in Table 2. The ratios of the values during the warm experiment to those of the cool experiment are presented as well. Although density differences are negligible, the dynamic viscosities at the stream/streambed interface were about 18% lower during the warm experiment and hydraulic conductivities were about 22% higher. At hyporheic depths of 5 and 10 cm, viscosities were 13% and 10% lower and conductivities were about 15% and 10% greater, respectively, during the warm experiment. In contrast, at 20 cm, the changes in these values were only 3%. In summary, the temperature regime during the morning experiment was cooler and essentially uniform while that during the afternoon experiment was warmer and heterogeneous with lower viscosities and higher hydraulic conductivities.
4.2. Hyporheic Solute Concentration Dynamics and Connectivity
4.2.1. Background Chloride and Bromide Values
 Chloride cutoff values for the 5 m shallow and deep wells, below which concentrations would be considered to be at background levels, were determined to be 13.2 and 15.2 mg Cl/L, respectively. Statistical analyses (see supporting information section S3) indicated that the remaining background chloride data should be divided into three groups: (1) stream morning, (2) shallow and deep well morning, and (3) combined stream and shallow and deep well afternoon. Cutoff values for these groups were: 7.7, 7.5, and 8.1 mg Cl/L, respectively. Statistical analyses also indicated that the stream, shallow well, and deep well background bromide values for both the cool and warm experiments could all be grouped together. The cutoff value for this group was 0.08 mg Br/L.
4.2.2. Chloride Surface Water Injections
 Chloride concentrations during the cool and warm experiments at the two stream sites located at 137 and 142 m below the injection are compared in Figure 4. During both experiments a plateau concentration was reached before the tracer injection ended. However, during the warm experiment, stream chloride plateau concentrations were reached more quickly than during the cool experiment at both locations. The chloride concentrations reached a plateau between 35 and 45 min faster during the warm experiment (Table 3). Also during the warm experiment, the chloride concentrations on the rising and falling limbs at the 142 m site rose and fell faster than those at the 137 m in contrast to the cool experiment, when the opposite was the case (Figure 5).
Table 3. Time to Plateau and Chloride Concentrations at Plateau for the Cool and Warm Tracer Experiments at Two Downstream Locations
 For the shallow wells at 10 cm depth at those two sites, the differences between the two temperature regimes were also pronounced (Figure 6). At the 137 m site during the cool experiment, the chloride concentration gradually plateaued reaching a concentration that was just over half of the concentration in the stream (connectivity = 51%). During the warm experiment, however, the chloride concentrations in this well reached values close to those in the stream (connectivity = 86%) with a shorter lag to plateau (Table 3). Concentrations also decreased coincidentally with the decrease in stream chloride concentrations when the tracer injection ended. Similarly, for the 10 cm well at the 142 m site plateau chloride concentrations were reached more quickly during the warm experiment than during the cool experiment. In addition, the plateau concentrations were closer to the concentrations in the stream in this well during the warm experiment (connectivity = 75%) than during the cool experiment (connectivity = 70%). In summary, for these two shallow wells there was evidence of more rapid arrival of the tracer and greater stream-well connectivity during the warm experiment.
 The two deep 25 cm wells at these sites exhibited quite different responses from the shallow wells in terms of the time course of the tracer concentrations (Figure 6 and Table 3). The 25 cm well at the 137 m site appeared to be highly connected to the stream despite its depth (connectivities = 94% and 100% for morning and afternoon experiments, respectively), with the tracer concentrations almost exactly mimicking those in the stream water during both experiments. In contrast, at the 25 cm well at the 142 m site for both experiments the tracer did not arrive until the injection had almost ended and the chloride concentration increased only slightly above background levels (connectivities = 27% and 24% for the morning and afternoon experiments, respectively).
4.2.3. Bromide Subsurface Injection
 Bromide first appeared in the stream above background levels at the 142 m site in the 14:10 sample at a value of 0.51 mg Br/L (Figure 7). This rise in bromide coincided with the first rise in chloride above background levels at the 142 m site and occurred before the first appearance of bromide and chloride above background levels at the 137 m site. Bromide and chloride then both returned to background levels before rising again in the 14:20 sample. Bromide first appeared above background levels at 137 m in the 14:20 sample at a value of 1.0 mg Br/L, which was the same time that chloride from the surface water injection first arrived at 137 m. Bromide did not appear above background levels at 5 m until 14:40. However, when it did first appear at 5 m, it was at the extremely high concentration of 201 mg Br/L.
 In general, the shape of the stream bromide concentration curves at the 5, 137, and 142 m sites was one of a sharp pulse with a prolonged tail. We can examine the timing of the bromide peaks to determine if and how the peaks were transmitted downstream. Of the three sites downstream of the bromide injection (5, 137, and 142 m), the first site to peak was the 142 m site at 14:20 at a value of 1.4 mg Br/L. The main 137 m site bromide peak of 1.6 mg Br/L took place at 14:25. It is not clear whether or not this peak was transmitted to the 142 m site. There is no corresponding spike in concentration at 142 m. Given the close, only 5 m, proximity of the sites and the 5 min sampling regime, it seems unlikely that the sampling missed the peak entirely.
 The final major bromide peak to emerge in the stream was that of 201 mg Br/L at 5 m at 14:40. This peak then arrived at the 137 m site at 14:55 when bromide concentrations increased slightly to 0.42 mg Br/L. However, again, this peak does not appear to have been transmitted to the 142 m site. The extremely high bromide value at 5 m suggests that the bromide may have been discharging near that point and that it had not been fully mixed with the stream water.
 During the subsurface injection, the additional wells installed in the thalweg did not consistently capture bromide above background levels until the 51 m site, which suggests that the bromide was not well mixed throughout the stream system at the upstream locations (Figure 8). However, chloride data for the additional wells installed indicate that at the 7.5 and 51 m sites, the deep wells had higher maximum/plateau concentrations than their shallow well counterparts. For the wells at 5 and 137 m, in particular during the morning experiment, this was also the case (Figure 8).
4.3. Solute Transport Modeling
 Modeled hyporheic parameter values and the associated residual sum of squares, t-ratios, and 95% confidence intervals are provided in Table 4. Because our conclusions do not change depending on the flow scenario, we present the medium flow scenario results here and include the low and high flow scenario results in the supporting information (Tables S1–S6 and Figure S3). Further, because our conclusions do not change based on whether we modeled the cool experiment with a single or double plateau, we present the single plateau results here because they did not make any assumptions about changes in 5 m concentrations and include the double plateau results in the supporting information (Tables S2 and S5 and Figures S3 and S4). The double plateau results, however, had lower residual sum of squares (RSSs). The 142 m results had goodness of fit measures (RSSs and Damkohler numbers) indicating that they were more reliable than the 137 m results (Table 4). We thus present the 142 m results first and then include the 137 m results for comparison.
 As expected, given the fairly similar flows between the two experiments, the stream cross-sectional areas and dispersion coefficients were fairly similar across both experiments and sites (Table 4). The 142 m hyporheic exchange coefficient (α) for the cool experiment was 2.61e-04 s−1. The 142 m warm experiment α of 1.99e-04 s−1 was 52% lower. However, the 95% confidence intervals overlapped substantially, suggesting that it is not possible to distinguish clearly between the α values. For the 137 m site, the cool experiment α was 2.01e-4 s−1. The warm experiment α was about half that value. However, again the 95% confidence intervals for the two experiments overlapped substantially.
 The 142 m storage zone cross-sectional area (As) for the cool experiment was 0.027 m2. The 142 m warm experiment As of 0.013 m2 was approximately half that value. The 95% confidence intervals do not overlap being 0.017–0.036 for the cool experiment and 0.010–0.015 for the warm experiment, suggesting that the difference in hyporheic cross-sectional areas was significant. For the 137 m site, the cool experiment As was 0.056 m2. The 137 m warm experiment As was 0.008 m2, which is seven times lower. The 95% confidence intervals, however, did overlap somewhat at the lower end of the range for the cool experiment and the upper end of the range for the warm experiment. However, as noted above, the RSSs and DaIs indicate that the results for 137 m were less reliable than those for the 142 m site, and they correspond to a particularly large 95% confidence interval range for the 137 m cool experiment As.
 Several metrics for hyporheic exchange indicate a lower influence of hyporheic exchange processes during the warm experiment compared to the cool experiment. For the 142 m site, for the cool experiment, the ratio of hyporheic to main channel cross-sectional areas (As/A) was 0.47. The fraction of median travel time due to transient storage as evaluated at a standardized reach length of 200 m (Fmed200) was 10.3%, and the estimated storage zone depth (ds) was 2.1 cm based on the modeling results and a typical stream width measurement of 3.8 m. For the warm experiment, the 142m As/A ratio was 0.24, the Fmed200 was 5.8%, and the calculated ds was 1.0 cm, which are all approximately two times lower than during the cool experiment. For the 137 m site, for the cool experiment, As/A was 1.05, Fmed200 was 13.4%, and ds was 4.5 cm. For the warm experiment, the 137 m As/A ratio was 0.12, Fmed200 was 2.1%, and ds was 0.6 cm, which were nine, six, and seven times lower, respectively, than their cool experiment counterparts. Thus the overall 137 m changes between the warm and cool experiment were in the same direction as for the 142 m site, but were larger. However, again, the 137 m RSSs, DaIs, and t-ratios all indicate the 137 m results are less reliable than those for 142 m.
 It is significant to note that although the estimated storage zone depths for both experiments were quite shallow, the tracer was actually detected at depths well below the estimated modeling values. This observation is consistent with the conceptualization that unlike the concentrations of chloride in the stream water, which integrate processes occurring in the whole stream due to mixing, the chloride concentrations at a selected hyporheic sampling location reflect the hydraulic connections of that individual location in the hyporheic zone to the stream.
Table 4. Estimated Model Parameters and Storage Metrics for the Cool and Warm Tracer Experiments
Fraction of median travel time due to transient storage—evaluated at Lstd = 200 m
Hyporheic zone depth
5.1. Evidence for Preferential Flow Paths
 The results for stream chloride and bromide concentrations during the warm experiment provide the most direct evidence for preferential flow paths (PFPs) in the experimental reach. In contrast to the cool experiment, during the warm experiment, the stream chloride rising and falling limbs at 142 m occurred before those at 137 m suggesting the presence of some rapid PFPs emerging between 137 and 142 m linking the 142 m site more closely to the upstream injection location (Figure 5). Similarly, the emergence of both bromide and chloride above background levels at 14:10 at the 142 m location but not until 14:20 at the 137 m location also suggests the emergence of PFPs between the 137 and 142 m sites. In addition, the bromide peak at 142 m occurred earlier than the bromide peak at 137 m (Figure 7). These results thus suggest a subsurface travel time on par with or faster than that of the surface water.
 The speed of the longer flow paths is surprising, and it may have been that hyporheic water was not moving parallel to the stream thalweg but was instead cutting across the channel bend in the reach and taking a shorter, more straight line route toward a downstream discharge point, in the sense of a plan view of the stream (Figure 2). Other authors have found evidence of such flow paths [Lautz and Siegel, 2006; Peterson and Sickbert, 2006; Takahashi et al., 2008; Wroblicky et al., 1998]. In the Von Guerard Stream reach, the straight line distance across the bend between the NaBr injection point at 1.5 m above the NaCl injection and the 142 m sampling point was 128 m, which would have shortened the distance only 15.5 m. Yet even if the flow path was shortened by 15.5 m, it would seem that some particularly favorable PFPs must have been involved for subsurface flow to be conveyed downstream in travel times that were on par with or faster than the surface water.
 Given the absence of plants and larger organisms in the harsh Dry Valleys' landscape, perhaps PFPs form in areas with particularly high hydraulic conductivity, or perhaps the rapid downstream movement of subsurface water could be explained by formation of soil pipes, a phenomenon commonly observed in hillslope hydrology. Soil pipes are chains of connected subsurface macropores oriented parallel to the soil surface [Faeh, 1997; Uchida et al., 2001], which can form preferential flow paths in diverse ecosystems, including forests, grasslands, and semiarid lands [Uchida et al., 2001]. Furthermore, evidence of pipeflow has been found at the interface between permeable thawed soils and less permeable frozen soils [Carey and Woo, 2000; Koch et al., 2013]. Thus, pipe-like features may form at the boundary between the frozen and unfrozen streambed sediments in dry valley streams. Considering the high density of the streambed injectate (about 1030 kg/m3 or approximately the density of seawater), the injectate may have sunk to a deeper portion of the hyporheic zone near the injection site, encountering pipe-like features and revealing some deeper hyporheic flow paths.
5.2. A Temperature-Induced Flow Path Preferentiality Feedback Mechanism
 An important overall finding from these two consecutive tracer experiments is that the physical properties of the stream-hyporheic zone system, in particular the hyporheic zone cross-sectional area (As), are dynamic on the time scale of hours. These changes occurred over two periods with distinct temperature regimes—cool, uniform and warm, heterogeneous. Although flows also differed somewhat between the two time periods, our modeling indicated that As was not sensitive to potential flow scenarios consistent with the upstream tracer concentrations, which overlapped between the two experiments. Thus, we interpret the results as showing that the increase in hyporheic temperatures in the upper layers of the hyporheic zone during the afternoon resulted in a large decrease in the influence of hyporheic exchange on Fmed200, the fraction of median travel time due to transient storage.
 The substantial change in the stream-hyporheic system suggests that temperature-induced changes in viscosity caused changes in the preferentiality of certain flow paths in the upper layer of the hyporheic zone. The decrease in dynamic viscosity under the warm conditions may increase the connectivity of flow along some paths that have intermediate connectivity under cooler regimes, enhancing their preferentiality during periods of maximum stream temperatures. Evidence for such an increase is seen in the two shallow (10 cm) wells at 137 and 142 m, where the tracer arrived sooner and reached concentrations closer to those in the stream during the warm experiment compared to the cool experiment. Although these two wells represent the transport of tracer for only two points in the upper hyporheic zone, the results may be representative of changes occurring elsewhere in the hyporheic zone. The preliminary dominance of certain flow paths may stem from stream surface and substrate heterogeneity creating initial differences in hydraulic conductivity that become amplified due to viscosity changes.
 It is also noteworthy that these two wells where the tracer arrived sooner during the warmer experiment than during the cool one were much deeper (10 cm) than the calculated storage zone depth (<5 cm), which is based on an assumption that all the saturated alluvium underneath the streambed surface exchanges water with the stream. This result indicates that this assumption is incorrect. Even during the warm experiment when the water had a minimum dynamic viscosity, a large portion of the saturated alluvium was not well connected to the stream. The results from the deep wells (25 cm) during both experiments provide further evidence for deep highly connected hyporheic flow paths, e.g., the well at 137 m, intermingled with areas that are isolated with limited connectivity to the stream (e.g., the well at 142 m).
 These findings suggest that the degree of connectivity within the hyporheic zone can vary due to the dynamic responses of PFPs to changes in water properties. One way to visualize this response may be as changes in the flow moving through a hyporheic “Swiss cheese” with dynamic PFPs distributed among isolated areas of the alluvium that, although saturated, are stagnant (Figure 9). Depending on the water temperature, some PFPs may reach the stream again while others may lead to dead ends. The viscosity-dependent flow of water through these PFPs may be the underlying process generating the response of the larger scale stream-hyporheic system to diel temperature variation.
 Further, temperatures in certain areas of the streambed could increase through the advection of heat along PFPs, in particular during the afternoon when stream temperatures reach their peak. Higher hyporheic temperatures occurring along more dominant PFPs could enhance temperature heterogeneity in the hyporheic zone. A feedback loop could occur in which the warming of PFPs enhances flow rates, resulting in more warm water flowing along the upper PFPs and so on. This increase in the flow path preferentiality could occur at the expense of more diffuse matrix flow in cooler areas of the hyporheic zone, which could become less preferential. For example, in the top 20 cm of the hyporheic zone, at 15:00, during the warm experiment, the hydraulic conductivity of the warmest location, based solely on temperature considerations, was 37% greater than that of the coolest location (Table 1). Overall, such a temperature-induced flow path preferentiality feedback mechanism may explain several differences in the results for the warm experiment compared to the cool experiment: (1) the increase in temperature heterogeneity during the warm experiment, (2) the faster arrival of the tracer at the wells during the warm experiment, (3) the greater similarity in solute concentrations at plateau at the well and the stream during the warm experiment, and (4) the smaller storage zone cross-sectional area (As) during the warm experiment.
5.3. Implications for Stream Ecosystems
 Hyporheic metrics including the ratio of storage to main channel cross-sectional areas and storage zone depths were at least twice as low for the warm experiment, than for the cool, more thermally uniform experiment for both sites. In addition, Fmed200, was also at least 1.8 times lower during the warmer, more thermally heterogeneous experiment than the cooler, more thermally uniform one. This metric that takes into account the simultaneous effects of As, α, A, and the advective velocity on solute transport downstream. Thus, according to most of the metrics calculated, hyporheic storage and exchange had less influence on the downstream transport of solutes during the warm, more thermally heterogeneous experiment than during the cool, more thermally uniform one.
 The finding that the storage exchange flux was lower under the warmer temperature regime for the 137 and 142 m sites contrasts with the findings of Cardenas and Wilson . Using numerical simulations, they showed that temperature-induced viscosity variations in a coarse sand/fine to medium gravel streambed substrate resulted in greater hyporheic fluxes when temperatures were warmer. However, in their simulations, Cardenas and Wilson assumed an idealized streambed that was homogeneous and isotropic. As they pointed out, natural systems are typically heterogeneous and anisotropic, and our results indicate that the combination of viscosity effects and PFPs can result in a different outcome.
 A conceptual model in which PFPs weave through isolated areas of the hyporheic zone also has biogeochemical implications. For example, the more isolated areas may be more apt to become anoxic, and a greater degree of anoxic/oxic interfaces along subsurface flow paths could create “hot spots” of denitrification [McClain et al., 2003]. Then, if an isolated area becomes reconnected to the stream at a longer time scale, this could create a “hot moment” in which reaction rates in either the hyporheic zone or the stream are higher than those of the surrounding time intervals because of the introduction of missing biogeochemical “reactants” [McClain et al., 2003]. The changes in hyporheic metrics noted above between the cool, morning experiment and the warm, afternoon experiment could influence how biogeochemical processes occurring in the hyporheic zone are coupled to processes occurring on the streambed and thus where/when hot spots and hot moments develop. For example, a decreased role for hyporheic exchange during the afternoon may allow some hyporheic areas to become more isolated leading to the development of hot spots within the stream subsurface. An increased role for hyporheic exchange during cooler thermal regimes may result in mornings being “hot moment” time periods in the stream within a daily cycle.
 Here we demonstrate that warmer, thermally heterogeneous conditions in a glacial meltwater stream and its bed drive significantly different hyporheic exchange dynamics compared to a cooler, more thermally uniform condition, as evidenced by spatiotemporal solute dynamics and solute transport modeling. This demonstrated influence of temperature on hyporheic exchange is likely to occur in many streams that experience substantial temporal temperature variability whether it be daily and/or seasonal. Previous studies have examined the effect of thermally induced fluid viscosity variations on hyporheic fluxes through idealized numerical simulations [e.g., Cardenas and Wilson, 2007] and several field investigations have combined empirical measurements with data analysis/modeling to examine the influence of hyporheic exchange on stream temperatures, but did not attempt to pinpoint the influence of hyporheic PFPs on stream thermal heterogeneity and hyporheic exchange processes [e.g., Arrigoni et al., 2008; Burkholder et al., 2008; Cozzetto et al., 2006; Moore et al., 2005; Story et al., 2003]. Our study thus advances understanding of the many controls on hyporheic exchange in coupled stream-groundwater systems. Additionally, we propose a conceptual view of the hyporheic zone of our particular study reach as being dominated by PFPs weaving through hydraulically isolated areas, which contrasts with the more traditional view of the hyporheic zone as being dominated by more uniform and diffuse flow paths dispersed throughout the alluvium.
 We thank the many people who helped out with both tracer experiments, especially Josh Koch and Shannon Horn with the University of Colorado and Kirk Miller and Ray Woodruff with the USGS Wyoming Water Science Center. Valuable logistical support was provided by Raytheon Polar Services and PHI Helicopters, Inc. We thank John Drexler of the University of Colorado's Laboratory for Environmental and Geological Studies for analytical assistance and Rob Runkel for help with the OTIS-P modeling. The authors also appreciate the input provided by Hari Rajaram. This work was supported by the National Science Foundation's McMurdo Dry Valley's Long-Term Ecological Research Program (ANT-1115245).