• Loading metrics

Dissolved oxygen sensor in an automated hyporheic sampling system reveals biogeochemical dynamics

  • Matthew H. Kaufman ,

    Contributed equally to this work with: Matthew H. Kaufman, Ruby N. Ghosh

    Roles Data curation, Formal analysis, Investigation, Methodology, Project administration, Supervision, Validation, Visualization, Writing – original draft, Writing – review & editing

    Affiliation Pacific Northwest National Laboratory, Richland, Washington, United States of America

  • Ruby N. Ghosh ,

    Contributed equally to this work with: Matthew H. Kaufman, Ruby N. Ghosh

    Roles Conceptualization, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Supervision, Validation, Visualization, Writing – review & editing

    Affiliation Opti O2, LLC, Okemos, Michigan, United States of America

  • Jay Grate,

    Roles Conceptualization, Funding acquisition, Investigation, Methodology, Resources, Writing – review & editing

    Affiliation Pacific Northwest National Laboratory, Richland, Washington, United States of America

  • Dean D. Shooltz,

    Roles Conceptualization, Methodology, Writing – review & editing

    Affiliation Opti O2, LLC, Okemos, Michigan, United States of America

  • Michael J. Freeman,

    Roles Conceptualization, Investigation, Methodology, Writing – review & editing

    Affiliation Opti O2, LLC, Okemos, Michigan, United States of America

  • Terry M. Ball,

    Roles Data curation, Investigation, Visualization, Writing – review & editing

    Affiliation Opti O2, LLC, Okemos, Michigan, United States of America

  • Reza Loloee,

    Roles Conceptualization, Methodology, Writing – review & editing

    Affiliation Opti O2, LLC, Okemos, Michigan, United States of America

  • Charles W. McIntire,

    Roles Formal analysis, Visualization, Writing – review & editing

    Affiliation Opti O2, LLC, Okemos, Michigan, United States of America

  • Jackie Wells,

    Roles Investigation, Methodology, Writing – review & editing

    Affiliation Pacific Northwest National Laboratory, Richland, Washington, United States of America

  • Chris Strickland,

    Roles Conceptualization, Investigation, Methodology, Project administration, Resources, Supervision, Writing – review & editing

    Affiliation Pacific Northwest National Laboratory, Richland, Washington, United States of America

  • Vince Vermeul,

    Roles Conceptualization, Methodology, Writing – review & editing

    Affiliation Pacific Northwest National Laboratory, Richland, Washington, United States of America

  • Kenton A. Rod,

    Roles Conceptualization, Investigation, Methodology, Validation, Writing – review & editing

    Affiliation Pacific Northwest National Laboratory, Richland, Washington, United States of America

  • Rob Mackley,

    Roles Conceptualization, Methodology, Writing – review & editing

    Affiliation Pacific Northwest National Laboratory, Richland, Washington, United States of America

  • Xinming Lin,

    Roles Data curation, Formal analysis, Investigation, Visualization, Writing – review & editing

    Affiliation Pacific Northwest National Laboratory, Richland, Washington, United States of America

  • Huiying Ren,

    Roles Data curation, Formal analysis, Investigation, Software, Supervision, Writing – review & editing

    Affiliation Pacific Northwest National Laboratory, Richland, Washington, United States of America

  • Amy Goldman,

    Roles Conceptualization, Data curation, Resources, Software, Supervision, Writing – review & editing

    Affiliation Pacific Northwest National Laboratory, Richland, Washington, United States of America

  •  [ ... ],
  • James Stegen

    Roles Conceptualization, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Supervision, Visualization, Writing – review & editing

    Affiliation Pacific Northwest National Laboratory, Richland, Washington, United States of America

  • [ view all ]
  • [ view less ]


Many river corridor systems frequently experience rapid variations in river stage height, hydraulic head gradients, and residence times. The integrated hydrology and biogeochemistry of such systems is challenging to study, particularly in their associated hyporheic zones. Here we present an automated system to facilitate 4-dimensional study of dynamic hyporheic zones. It is based on combining real-time in-situ and ex-situ measurements from sensor/sampling locations distributed in 3-dimensions. A novel dissolved oxygen (DO) sensor was integrated into the system during a small scale study. We measured several biogeochemical and hydrologic parameters at three subsurface depths in the riverbed of the Columbia River in Washington State, USA, a dynamic hydropeaked river corridor system. During the study, episodes of significant DO variations (~+/- 4 mg/l) were observed, with minor variation in other parameters (e.g., <~+/-0.15 mg/l NO3). DO concentrations were related to hydraulic head gradients, showing both hysteretic and non-hysteretic relationships with abrupt (hours) transitions between the two types of relationships. The observed relationships provide a number of hypotheses related to the integrated hydrology and biogeochemistry of dynamic hyporheic zones. We suggest that preliminary high-frequency monitoring is advantageous in guiding the design of long term monitoring campaigns. The study also demonstrated the importance of measuring multiple parameters in parallel, where the DO sensor provided the key signal for identifying/detecting transient phenomena.


Hyporheic zones, the riverbed and bank sediments where pore water is primarily sourced from and returns to the open channel, play host to important biogeochemical processes including aerobic respiration, nitrification, denitrification, and many other biotic and abiotic processes [14]. Concentrations of microbially and geochemically relevant solutes such as dissolved oxygen (DO) and nitrate present at any given time and location in hyporheic zones are a function of both reaction rates and residence times [5, 6]. Rivers are also highly dynamic, driving subsequent dynamics in reaction rates [7] as well as flowpaths and residence time distributions [810]. Hydrologic dynamics in rivers occur over a wide range of timescales [1, 11, 12]. Increasing attention has been paid to the impact of short term hydrologic perturbations (hours to days), such as those created by hydropeaking, tidal forces, evapotranspiration, and other similar processes [13, 14]. These highly dynamic processes combine to control the concentrations of biogeochemically-relevant solutes in the hyporheic zone, particularly electron donors such as organic carbon and acceptors such as DO and nitrate [15].

Ultimately, perturbations of biogeochemically-relevant solutes like DO are driven by varying ratios of reaction rate timescale to transport timescale, a ratio described by the Damkohler number [6, 16]. Traditionally, such perturbations have been named “hot moments” if they meet the criteria laid out by McClain, Boyer [7]: “biogeochemical … hot moments are defined as short periods of time that exhibit disproportionately high reaction rates relative to longer intervening time periods”. In highly dynamic and spatially heterogeneous systems, variations in flowpaths and residence times can lead to similar perturbations of biogeochemically-relevant solutes without changes to reaction rates. In this study, the primary solutes of interest are DO and nitrate, which may vary independently of changes in reaction rate [17] while providing indications of spatial and/or temporal variation in hyporheic zone dynamics.

Variable stage elevations can lead to rapid changes in hydraulic head gradients that drive water into and out of the hyporheic zone [18, 19]. Biogeochemical responses to such hydrologic perturbations are often attributed to variable mixing of surface water and groundwater [2022], particularly through the mechanism of variable reaction rates driven by changing microbial access to nutrients, organic matter, and electron acceptors as those mixing ratios change [23, 24]. However, these hydrologic perturbations also have the potential to drastically alter residence time distributions in the hyporheic zone and access to preferential flowpaths [25, 26]. While several studies conclude that changes in residence time and flowpath length drive biogeochemical responses [17, 27, 28], little is known about biogeochemical responses to short term hydrologic perturbations in field locations with low levels of groundwater/surface water mixing. This results in a limited ability to isolate hydraulic head gradient-driven, residence-time based effects from mixing-driven, reaction-rate based effects. We aimed to study hydrologic and biogeochemical dynamics while minimizing the influence of this mixing.

The dynamic and spatially variable nature of hyporheic zones pose significant challenges to studying the integrated hydrology and biogeochemistry of these systems. The challenges are particularly stark in large, high-order rivers where traditional methods (e.g., reactive tracer studies) developed for lower-order systems are extremely difficult (e.g., due to very large tracer volumes needed for large rivers). Our main objectives were to develop an automated system that couples in-situ and ex-situ components for 4-dimensional monitoring of temperature, dissolved oxygen, nitrate, conductivity, pH, and oxidation-reduction potential in a large-scale hyporheic zone, then provide an example of how the system can be used to study dynamic and spatially variable hyporheic zones. Given that the data span only 16 days and are from a small spatial domain, we consider the outcomes of our study to be hypothesis-generating, as opposed to providing strong tests of a priori hypotheses.


Field monitoring system

The field setting for this study was the shore and near-shore environment of the Columbia River at the 300 Area of the Hanford site near Richland, Washington (Fig 1). The Hanford Reach of the Columbia River is contained within the Pasco Basin in southeast Washington State. The river is surrounded by an unconfined aquifer consisting of fluvial Ringold Formation and Hanford Formation flood sediments [29]. The hydraulic conductivity of the riverbed in this area is highly variable, ranging from 2.8x10-5 cm/s to 4.3x10-2 cm/s [30].

Fig 1. Field site location near Richland WA, USA.

Dense array aquifer tube clusters are indicated with orange dots. The specific cluster sampled for this study is indicated with a red triangle. Wells, piezometers, and the RG3 river gauge are indicated with blue squares. Base layer [31].

Subsurface biogeochemistry was monitored from April 7th to April 23rd 2018 using a single cluster of aquifer sampling tubes installed in the “dense array” automated 4-dimensional sampling system, a unique hyporheic zone water sampling system that draws from a large array of riverbed monitoring points (Fig 1). While short, the study period was chosen specifically to coincide with relatively high river stage (this reach of the Columbia River is typically highest in the spring, due to snowmelt) in order to limit the influence of groundwater discharge on hyporheic zone biogeochemistry. A central measurement system located on-shore pulled water samples from an array of tubes buried in a 3-dimensional grid array in the hyporheic zone. The central measurement system automatically cycled through a pre-programmed subset of the aquifer sampling tubes, to gather physical and chemical data from the grid of spatial locations at this field site. This system used ex-situ sensors to measure fluid electrical conductivity (measured as specific conductivity: SpC), pH, nitrate, oxidation-reduction potential (ORP), and DO, and in-situ sensors to measure temperature.

The aquifer sampling tubes were installed in September of 2016 in a grid pattern in both permanently and variably wetted hyporheic zone sediments along the shoreline. The tubes were installed in clusters and were spaced over an approximately 10 m wide (perpendicular to the shoreline) by 90 m long (parallel to the shore) grid that was 10 clusters long and 3 clusters wide. Each cluster consisted of 2 or 3 tubes installed to approximate depths of 0.5m, 1m, and 2m below the sediment-water interface, allowing for four-dimensional sampling of the hyporheic zone. Due to substrate heterogeneities some locations could not be penetrated with the hand installation tools used and as a result not all clusters contained a 2m monitoring point. The end of each tube was screened to form a mini-piezometer, with slots cut into the last 7 to 15 cm of each tube to allow water to be sampled. These slots were covered with Teflon mesh to reduce sediment clogging. The distal ends of the tubes were closed with stainless steel thermistor (Fig 2). After the full installation was complete, each tube was developed. Development occurred in September of 2017 and proceeded according to the following protocol:

  1. Set up vacuum pump and peristaltic pump. Fill bucket with river water. Put empty bucket nearby for outflow water. Verify correct direction for pumping water into the aquifer tube using the peristaltic before starting (“forward” or “reverse”).
  2. Turn on vacuum pump. Attach aquifer tube extension to pump to pull water out of tube and start timer. Record start time for pump and amount of time until water appears.
  3. Allow flask to fill to 500mL and record amount of time from water appearance until 500 mL reached.
  4. Disconnect vaccum pump from aquifer tube. Disconnect vaccum pump from beaker and empty water.
  5. Attach tube to peristaltic pump with a filter between the aquifer tube and the pump connection.
  6. Put inflow tube of peristaltic pump into the bucket of river water and pump into the aquifer tube for one minute.
  7. Disconnect peristaltic from aquifer tube and reconnect vaccum pump to aquifer tube.
  8. Pump water out of tube for two minutes, empty flask, reconnect to aquifer tube, and record start time of vaccum and when water appears (likely immediately).
  9. Allow flask to fill to 500mL and record time.
  10. Disconnect vacuum pump.
  11. Assess if the rate of water output from the tube or the clarity changed.
  12. Repeat steps 5 to 11 until there is no improvement
Fig 2.

(a) shows the tip of an aquifer sampling tube, with the screened section and thermistor cap. (b) shows the central on-shore measurement system with banks of solenoid valves, a series of sensors, and a datalogging system. (c) shows the hand-installation of an aquifer tube. (d) shows detail of the central on-shore measurement system. The green oval in box 5 of (d) is the OptiO2 dissolved oxygen sensor flow cell. In (d), box 1 houses the syringe pumps, boxes 2 and 4 house the solenoids and solenoid controllers, boxes 3 and 5 house the various sensors mounted in their flow-through manifolds, box 6 houses the Campell CR3000, and box 7 houses interface units that allow the CR3000 to communicate with the sensors in box 3. The equipment enclosures in (b) and (d) are in the same relative positions.

Typically, by the time no improvement was achieved, it took <2 minutes to withdraw 500ml of water from each tube. Water quality parameters were also monitored for stabilization. The system was operated regularly in various configurations between the development date and the start of this study.

Each of the sampling tubes was connected to its own solenoid valve, and all valves were controlled by a CR3000 datalogger through four SDM-CD16AC 16-Channel AC/DC Relay Controllers (Campbell Scientific, Logan UT). For any given sampling scheme, solenoid valves corresponding to selected sample locations were opened one at a time and two coupled Kloehn syringe pumps (IMI Precision Engineering, Las Vegas NV) alternately drew sample water from the tube and then discharged it to the monitoring manifold at a rate of 5 mL s-1. The tubing manifold volume was kept as small as possible, but was not measured. Vacuum applied to the sample lines was also not measured, however the pump was approximately 4m above the surface of the river, which requires a minimum vacuum of 392 millibar to raise the water. No bubbles were observed in the sampling lines or the syringe pumps while under vacuum. In this study, each location was sampled for 41 minutes before switching to another aquifer tube. This purges the prior sample and provides time for the current sample to achieve stable readings representative of the sampled location. The 41-minute timing was selected after brief initial observation of the system in operation. With regard to the sampled sediment volume, assuming a porosity of 0.3–0.6, 41 minutes of pumping equates to a sampled sphere with a radius of 17-21cm.

The central measurement platform relied on the transport of samples from the point of origin in the river bed to the central measurement system, while keeping the sample composition unchanged for measurement at the central location. The plastic tubing used for sample transport is made of LLDPE (linear low density polyethylene), with an outside diameter of 9.5mm (3/8 inch) and an inside diameter of 6.35mm (1/4 inch). The potential for oxygen outside the tube to diffuse into the sample within the transfer tubing was a primary design consideration, and thus was measured in the laboratory. A source of nitrogen sparged water was connected to the luminescent dissolved oxygen sensor of an MS5 Sonde (Hydrolab, Loveland CO, USA) via 4.6, 6, 30.5, and 122 m (15, 20, 100, and 400 foot) lengths of the same black LLDPE tubing type being used in the field. In all cases, at a flow rate of 1 mL s-1, the measured DO was zero at the sensor distal from the source of nitrogen sparged water. Tubing lengths in the field were 6–45.7 m with tubing volumes ranging from 190ml to 1447ml respectively, and the flow rate in this field study was five times faster than the laboratory test. Therefore, the sample residence time in the field was at least twenty times less than in the laboratory experiment with the 122 m length.

Our focus was to provide an example use of this system in a ‘hypothesis-generating’ mode. There is an inherent tradeoff between temporal resolution and number of locations monitored due to transit time of pumped water. For hypothesis generation we obtained high time resolution sampling from multiple depths at a single x-y location in the dense array (Fig 1, red triangle). An alternative design would be to sample all locations infrequently to assay spatial variation across longer timescales. By focusing on one vertical transect of 3 aquifer sampling tubes and using 41 minutes of pumping time per solenoid, each depth was sampled every two hours.

The time-resolved DO measurements were acquired and logged with a novel optical DO sensor developed by Opti O2, LLC ( The DO probe, housed in a flow cell, was incorporated into the measurement platform of the dense array system as shown in Fig 3a. DO was recorded every 30 seconds as the solenoid valves cycled between the three selected aquifer sampling tubes (Fig 3b). This high temporal resolution was necessary to cleanly identify when the system had reached a new quasi steady state after a solenoid/aquifer sampling tube switch.

Fig 3.

(a) shows the Opti O2 DO sensor housed in a flow cell incorporated into the dense array manifold. (b) shows a 36-hour subset of the raw data recorded by the sensor, showing the variation in DO signal as the dense array switched between the three sampling depths.

The Opti O2 probe was a self-contained optical spectrometer which output a temperature and pressure corrected DO signal. Molecular oxygen was detected by monitoring the phosphorescence emission from molybdenum chloride optical indicators [32]. Ultraviolet photons pumped the optical indicators to a spin triplet excited state, which is specifically quenched by the ground state of molecular oxygen, 3O2. The optical indicators were immobilized in a polymer matrix, resulting in a sensing film which was placed in direct contact with the liquid to be analyzed [33]. The sensing film had minimal cross sensitivity to organic and inorganic species, and did not suffer from photobleaching [34]. These properties enabled DO measurements with high sampling rates for an extended period from the unconditioned field water samples, with highly variable temperatures and complex chemical constituents without the need for maintenance or field calibration.

In addition to the DO instrument, the pumps delivered sample water to a series of instruments for real time measurements of water chemistry (Fig 2). Measurements were made for nitrate using a S::CAN spectro::lyser UV-Vis 35mm (S::CAN Messtechnik GmbH, Vienna, Austria), SpC with Rosemount 400 (Emerson, Irvine, CA), pH and oxidation-reduction potential (ORP) with Rosemount 3900 (Emerson, Irvine, CA). All data logging and instrument control for these instruments was provided by the CR3000. The instruments were cleaned and maintained biweekly. The S::CAN instruments used factory calibration, while the Rosemount sensors were calibrated according to the manufacturer’s instructions. The Opti O2 device was factory calibrated pre and post field deployment.

Approximately 17 meters downstream of the downstream edge of the dense array (42 m downstream of the sampled aquifer tube cluster), a transect of five 3.81 cm inner diameter stainless steel piezometers were installed. One of the piezometers (RG3) was left open to the river with the screen at the river level. The other four were capped and had a pressure sensor installed at the screen to allow for discrete measurements of hyporheic zone hydraulic head at the different depths. Of the four capped piezometers, two were installed in the permanently inundated channel with one shallow (P1S) screened at 104.24masl and the other deeper (P1D) screened at 101.67masl. The other two were located on the ephemerally inundated bank with one shallow (P2S) screened at 105.68masl and the other deeper (P2D) screened at 103.88masl. The piezometers were each outfitted with an Aqua TROLL sonde (In-Situ, Inc. Ft. Collins CO) for continuous measurement of pressure, temperature, and SpC. The sonde measuring the river was vented and the other four sondes were not vented but were corrected for barometric pressure at the data logger. The sondes were connected to a CR1000 data logger (Campbell Scientific, Logan UT) and data was downloaded twice a month during the field season. A 10-year (2008–2018) hourly spatiotemporal dataset from a network of groundwater wells at the 300 Area was used for background SpC and hydraulic head gradient information [35].

Data processing

Data was recorded at 30 second intervals for the analytes in the dense array, and 15 minute intervals for the wells, piezometers, and river channel data. Solenoids switched from one of the three depths to the next every 41 minutes. Because the tubes were relatively long, it took approximately 30 minutes for water from the new sampling point (after a solenoid switch) to reach the sampling manifold. We expect heterogeneous and dynamic conditions in the hyporheic zone, thus we did not necessarily expect to ever come to a true steady state after a solenoid switch. Rather, we were looking for the time when the rapidly changing data associated with the transition between sampling locations gave way to more slowly varying data that was reflective of the conditions surrounding the new sampling location. To deal with this, the first 31 minutes from the raw time-series measurement for each 41 minute monitoring event was discarded. The measurements from the final 10 minutes were averaged and are presented in Figs 46. During two of these 41 minute periods this was not enough time for the system to achieve quasi-steady state conditions, and these values were removed. This provided one measurement for each analyte for each depth approximately every 2 hours for the duration of the study.

Fig 4.

(Top) Dissolved oxygen concentration and (Second) oxygen percent saturation at each depth. (Third) stage height of the river at RG3 and water table elevations at wells 3–19 and P1D. Left and right Y axis ranges are the same. (Bottom) Cross valley hydraulic head gradient computed between wells 3–19 and P1D (positive values indicate water flowing toward the river). The pairs of solid and dashed vertical lines delineate perturbations A and B, respectively. The horizontal dashed line is the cross valley hydraulic head gradient above which hysteresis was observed in the relationship between DO and the cross valley hydraulic head gradient.

Fig 5.

(Top panel) Specific conductivity at the three different hyporheic zone depths and in the river, and periods of precipitation. The weather station was located approximately 8 km from the study site and precipitation patterns in the study area can be extremely localized, so precipitation events should be regarded as approximate. (Bottom panel) temperatures recorded at the three monitoring depths and in the river channel.

Fig 6.

(Top 3 panels) DO vs. cross valley hydraulic head gradient for the entire study period. Dashed vertical line indicates a gradient of 1x10-4. (Bottom panel) to facilitate interpretation, the DO and cross valley hydraulic head gradient time series are shown for perturbation A. The sequential numbering is the same as in the top 3 panels.

Data analysis

The study site was surrounded by many monitoring wells (Fig 1), which when combined with the water table elevations collected near the sampling points provided numerous combinations for calculating both approximately vertical and approximately cross-valley hydraulic head gradients. Cross-valley hydraulic head gradient was defined such that a positive value represents a situation where hydraulic head in the monitoring well was higher than head in the river piezometer, thereby driving water generally toward the river. Negative cross-valley hydraulic head gradients describe the opposite condition, where water tends to flow away from the river. Cross-valley hydraulic head gradients were computed between 4 monitoring wells to the west (Fig 1) of the river and the various piezometers located in or near the river. One vertical hydraulic head gradient was also calculated using the P1S and P1D piezometers. One down-valley (as opposed to cross-valley) hydraulic head gradient was also calculated between wells 1–2 and 3–19. All hydraulic head gradients strongly covary with each other by Spearman rank correlation [36] (Table 1, S1 Table). A single head gradient, between wells 3–19 and P1D (Table 1), was chosen to represent the system as it was most strongly correlated with all others.

Table 1. High rank correlation between cross-valley hydraulic head gradient at well pair P1D,3–19 and vertical and cross-valley hydraulic head gradients between other local well pairs.

Results and discussion

Across the 16 days of monitoring, the dense array performed well and met the primary objective of the effort: to generate new hypotheses about the hydrology and biogeochemistry of a dynamic, large-scale hyporheic zone. There are, however, limitations to the system and improvements could be made to elevate the utility and quality of the generated data. Below we discuss the scientific implications of the generated data and end with a discussion of advantages and limitations of the dense array as well as ideas for future uses within and beyond this study system.

Overview of DO and hydraulic head gradient dynamics

Over the course of the study, there were multiple periods of short-lived but large-magnitude excursions from the mean DO concentrations at each of three sampling depths (Fig 4). We focus on two 48-hour periods labeled “perturbation A” and “perturbation B”. Large DO excursions were evident during these periods, particularly in the signals from 54cm and 100cm depths. DO readings in all locations during most of this period were above 6 mg/L, which agrees with prior observations of the typically-oxic nature of both the surface water (S1 Fig) and shallow groundwater in this region [37]. Examining other measured parameters, the next most dynamic variable was nitrate concentration (S4 Fig), although here the excursions were less distinct from the normal amount of variation in nitrate over the data set. pH and ORP showed little systematic variation (S2 Fig). Dissolved oxygen was, therefore, the most sensitive indicator of changing conditions.

River stage (referenced to an arbitrary vertical datum), well and piezometer water table elevations (referenced to mean sea level) and cross-valley hydraulic head gradients during the same period are plotted in Fig 4. Note the rapid change in magnitude and direction of the cross-valley hydraulic head gradient during perturbation A. In addition to cross-valley hydraulic head gradients, we computed a down-valley hydraulic head gradient using wells near the river. The magnitude of the down-valley hydraulic head gradient was approximately an order of magnitude lower than the cross-valley hydraulic head gradient, and also showed less variation in time. This is different from the down-valley dominated hydraulic head gradients Voltz, Gooseff [38] observed, however that work took place in a much steeper headwater valley. Larkin and Sharp [39] predict that the magnitude of cross-valley hydraulic head gradients will be large compared to down-valley hydraulic head gradients in wide valleys like the one the Columbia river occupies.

River stage provides the primary control for both the water table elevation in the unconfined aquifer as well as cross-valley hydraulic head gradients. The generally high (though variable) hydraulic conductivity of the unconfined aquifer near the study site leads to tight coupling between river and subsurface hydraulics. This allows the river stage to strongly influence water table elevations both near the river and farther away at monitoring wells installed approximately 450 meters away from the river [40]. Fluctuating cross-valley hydraulic head gradients driven by relatively high frequency stage fluctuations are common both in the general area of the field site [18], and also in other river systems [19]. These water table elevations in turn control the cross-valley hydraulic head gradient.

DO perturbations were not due to variation in mixing or temperature

DO and nitrate varied with the cross-valley hydraulic head gradient (Fig 4, S4 Fig), which could indicate one or more underlying mechanisms such as variation in groundwater/surface water mixing and/or temperature. An influence of mixing would not be surprising as it is common for variable groundwater/surface water mixing to exert a strong influence on subsurface biogeochemistry at this site and others [18, 22, 41]. However, that does not appear to be the primary process at work at this site during the monitoring period. SpC of pore fluid in the sediments was typically in the range of 120–160 μS/cm (Fig 5) which was similar to the SpC in the open channel of ~160 μS/cm (Fig 5). In contrast, the average SpC of groundwater from the unconfined aquifer (monitored at well 3–19) was approximately 430 μS/cm and changed very little over the course of the study. Note that DO was also not strongly correlated with SpC (S3 Fig), further indicating that groundwater/surface water mixing was not an important driver of DO dynamics. Previous work at this site has established large-scale (100s of meters inland) hydrologic exchange between the groundwater aquifer and the river, that the mixing can be traced using SpC, and that SpC of groundwater is markedly higher than SpC in the river [20, 40]. For example, Arntzen, Geist [20] used 658 μS/cm as indicative of pure groundwater, and Johnson, Versteeg [40] found groundwater to be near 400 μS/cm. While there is variation in the SpC of groundwater, it is clearly much higher than SpC of both the river water and the pore water monitored during our study. The SpC dynamics observed during our monitoring period indicate potential for a small amount of groundwater mixing with river water in the hyporheic zone. However, the increases in SpC above river water levels were relatively minor. We therefore infer that the small amount of groundwater that did move through the system did not have a large influence over DO dynamics. The DO dynamics in periods and places with more groundwater contributions may, however, be influenced by that mixing, as implied in previous work in the same field system [15, 22, 42].

While the SpC measurements indicate that the monitoring locations used in this study exist in primarily surface water-dominated conditions, it is possible that groundwater/surface water mixing is occurring further into the banks or bed of the river. This is consistent with prior modeling studies including Shuai, Cardenas [42], who found most nitrification occurred “adjacent to the bank where oxic river water and groundwater interacted, while denitrification occurred deeper into the aquifer and below the riverbed where oxygen was depleted”. As we do not show a strong nitrate pattern over depth or time, nor suboxic or anoxic conditions, it may be that our monitoring locations simply occur closer to the river than the zones of groundwater/surface water interaction or oxygen depletion described by Shuai, Cardenas [42].

Temperature is another driver of biogeochemical transformation rates [43], providing another potential cause of the observed perturbation. However, the temperatures recorded at the monitoring points (Fig 5), while correlated with changes in cross-valley hydraulic head gradient, only varied by about 2°C over the study period. This indicates that temperature variations within the subsurface, and the corresponding reaction rate variations that would follow, were not the primary drivers of biogeochemical perturbations at this study site.

DO perturbations were closely associated with hydraulic head gradients

Having established that neither groundwater/surface water mixing nor subsurface temperature appear to be the primary drivers of the observed DO perturbations, we further analyzed the relationship between DO and cross-valley hydraulic head gradient (Fig 6). During our study, DO responded to fluctuations in cross-valley hydraulic head gradient in two distinct forms of relationship. These occurred on either side of an apparent tipping point, corresponding to a cross-valley hydraulic head gradient of ~1x10-4. At cross-valley hydraulic head gradients that were lower (including negative) than this threshold, the relationship between DO and cross-valley hydraulic head gradient was non-hysteretic. In contrast, at cross-valley hydraulic head gradients that were larger (more positive) than ~1x10-4, hysteretic relationships emerged (Fig 6). While a linear regression model may provide a reasonable fit across the entire hydraulic head gradient range, examining the temporal progression of DO and hydraulic head gradient (above a gradient of ~1x10-4) confirmed hysteresis at all depths.

The non-hysteretic relationship was observed for the bulk of the time series, including perturbation B. While perturbation B did include some time periods of positive cross-valley hydraulic head gradients, they were not sustained, and did not reach the threshold of ~1x10-4. The slope of the relationship varied across the three monitoring depths, but the non-hysteretic relationship between DO and cross-valley hydraulic head gradient did not. Additionally, the DO concentrations at the three depths tended to vary together when the DO/cross-valley hydraulic head gradient relationship was non-hysteretic (Fig 4). Such non-hysteretic DO/cross-valley hydraulic head gradient relationships may be common in our study system, as they have been observed in previous years [20].

Above the apparent tipping point, we hypothesize that the stronger and more prolonged positive gradient may activate previously-clogged flowpaths, or allow water that has been trapped in the banks for longer times (by negative or only weakly/briefly positive cross-valley hydraulic head gradients) to be drawn through the monitoring locations. When the cross-valley hydraulic head gradient threshold of ~1x10-4 was exceeded, hysteretic behavior emerged at all depths in the riverbed indicating a shift in overall system behavior. Specifically, the DO data from perturbation A indicate that this single field site functioned like two different hyporheic zone systems separated in time. From 54 to 200cm the relationship between riverbed DO and cross-valley gradient took the form of hysteretic loops, as evidenced by the sequential time series in Fig 6.

These hysteretic relationships are similar to what Soulsby, Malcolm [44] observed during a series of flood events in the river Dee in Scotland. In the hyporheic zone, Soulsby, Malcolm [44] observed a hysteretic relationship between DO and discharge during one flood at 30cm depth but not 15cm, during another flood at 15cm but not at 30cm, and during a third flood at both depths. While Soulsby, Malcolm [44] found inconsistent appearance of hysteretic behavior, we observed hysteretic loops at all depths despite divergence in the DO concentrations at the three depths: the 54cm probe recorded a strong decrease in DO concentration, while the 100cm probe recorded an equally strong increase, rising to saturation and staying there for approximately 24 hours, and finally the 200cm probe recorded little variation in DO (Fig 6). Hysteretic DO responses to hydrologic perturbations have also been observed in laboratory flume experiments [27]. Qualitative consistency in the emergence of hysteresis across all depths despite quantitative divergence in DO concentration dynamics indicates an overriding and general shift in system behavior above the cross-valley hydraulic head gradient-based tipping point of 1x10-4 in our system.


From the full suite of patterns studied here, we hypothesize that the combination of “old” water flushing from the hyporheic zone via activation of preferential flowpaths leads to DO concentrations that are both dynamic and variable in space. Here “old” water refers to rain-derived or river-derived water that has moved into the banks and/or hyporheic zone and that has been residing there for a significant (but unknown) amount of time. The general hydrologic setting of the field site includes high and spatially variable hydraulic conductivities [40]. These conditions, combined with large and rapid stage fluctuations, imply that changing residence times driven by hydraulic head gradient fluctuations may be the mechanism behind the observed dynamics in DO and potentially nitrate as well. Perturbation A took place during a local minimum in river stage (Fig 4), which generated a relatively strong positive (toward-river) cross-valley hydraulic head gradient that lasted for ~40 hours. Such a period of toward-river cross-valley hydraulic head gradient may allow clogged flowpaths to open, or allow long-residence-time hyporheic water to be transported through the monitoring locations. The long residence time would likely deplete DO due to microbial respiration. However, the DO perturbation across the monitoring depths differed in both magnitude and direction, potentially due to preferential flowpaths. The SpC data alludes to an influence of preferential flow, with SpC at the 100cm monitoring point dropping below river water SpC, sometimes coincident with rain events (Fig 5). Similar declines in SpC at other depths were not observed. Additionally, the cross-valley hydraulic head gradient correlates worse with SpC at 100cm than at the other two depths (S1 Table). Low-SpC rain water may therefore have preferential access to the 100cm depth.

We further hypothesize that high-frequency stage fluctuation were a key driver of threshold behavior in which the system switched from non-hysteretic to hysteretic dynamics on either side of a cross-valley hydraulic head gradient of ~1x10-4. This hypothesis agrees with previous studies, including Cardenas and Wilson [45] who concluded that “the [interfacial exchange zone] develops only when a current threshold is overcome.” Furthermore, it is well known that relatively small changes in hydraulic head gradients can lead to large changes in hyporheic exchange [4649]. Additionally, De Falco, Boano [50] found that exchange flux determined oxygen consumption, consistent with the hypothesis that DO dynamics in our study were not due to changes in respiration rates.

Consistency between our inferences and those of previous studies suggests that the dynamics we observed could exist in other hyporheic zone systems that experience high frequency stage fluctuations. Such systems are common, with high frequency fluctuations driven by a broad range of factors such as evapotranspiration, tides, water withdrawals, and/or dam operations. Nonetheless, to test the transferability of the dynamics observed here will require high temporal resolution datasets including both subsurface DO and hydrology across additional systems, ideally coupled to process-based models. In addition, while we infer a strong influence of residence time and preferential flow, other mechanisms such as those associated with groundwater/surface water mixing are known to be important in this system and others [23, 24, 42]. Truly transferable understanding of hyporheic zone function will necessitate understanding the coupling among these different mechanisms.

Limitations and future directions

While multi-parameter spatial and temporal monitoring in the subsurface is challenging, it is required to understand integrated hydrologic and biogeochemical function of hyporheic zones. This requires instrumentation that can gather a diverse set of measurements from a large number of sampling locations on rapid time scales. The dense array system used in this study provides a new way to interrogate hyporheic zones, but comes with challenges, such as ~ 30 minute stabilization time. This results in a tradeoff between temporal resolution and spatial coverage. Some creative solutions to the system’s limitations include the following: One could actively monitor concentrations and use a mathematical stabilization criterion to allow for on-the-fly optimized stabilization times. Additionally, the quasi steady state periods were identified visually in our relatively small dataset, but could be identified more quantitatively by analyzing rates of change or standard deviations among sequences of datapoints in future studies. One could generate high temporal resolution data in a small number of locations and coarse temporal resolution data across more spatially distributed locations. Secondly, multiple sampling tubes could be purged simultaneously, resulting in almost no lag time between collection of stable measurements when switching from one tube to another. A radically different experimental approach would be deployment of spatially distributed in situ real time measurement devices directly within the riverbed. For DO, the key biogeochemical indicator in our study, this approach requires a sensor with stable calibration for long term deployment such as the optical technology employed in this work. For electrical resistivity tomography (ERT), the thermistors of the dense array itself have been used as electrodes for time-lapse ERT imaging. This real time in situ data would then guide where and when to pull samples for multi-parameter analysis. Systems like the dense array will be important in obtaining the data necessary to understand the integrated hydrologic and biogeochemical processes present within dynamic hyporheic zones.

Supporting information

S1 Fig. Dissolved oxygen in the surface channel at the study site.


S2 Fig. pH and ORP from 54 cm, 100 cm, and 200 cm sampling locations.


S3 Fig. Bivariate plots of dissolved oxygen vs. specific conductivity at the 3 sampling depths.


S4 Fig. Nitrate measured 54 cm, 100 cm, and 200 cm at the sampling location.


S1 Table. A correlation matrix of Spearman ranks between the variables contained in the dataset.



We thank the editorial staff as well as reviewers for their valuable contributions to this manuscript. The project completed environmental reviews as required under federal NEPA law, and obtained all required Washington State and federal permits for this research, including WDNR right of entry 23-B91879 and USACE nationwide permit NWS-2014-763 The dataset for this publication is hosted on ESS-DIVE and can be accessed at the following location:


  1. 1. Boano F, Harvey JW, Marion A, Packman AI, Revelli R, Ridolfi L, et al. Hyporheic flow and transport processes: Mechanisms, models, and biogeochemical implications. Rev Geophys. 2014;52(4):603–79.
  2. 2. Boulton AJ, Findlay S, Marmonier P, Stanley EH, Valett HM. The functional significance of the hyporheic zone in streams and rivers. Annu Rev Ecol Syst. 1998;29:59–81.
  3. 3. Brunke M, Gonser T. The ecological significance of exchange processes between rivers and groundwater. Freshwater Biol. 1997 Feb;37(1):1–33.
  4. 4. Fischer H, Kloep F, Wilzcek S, Pusch MT. A river’s liver—microbial processes within the hyporheic zone of a large lowland river. Biogeochemistry. 2005 Nov;76(2):349–71.
  5. 5. Zarnetske JP, Haggerty R, Wondzell SM, Baker MA. Dynamics of nitrate production and removal as a function of residence time in the hyporheic zone. J Geophys Res-Biogeo. 2011 Mar 10;116.
  6. 6. Zarnetske JP, Haggerty R, Wondzell SM, Bokil VA, González-Pinzón R. Coupled transport and reaction kinetics control the nitrate source-sink function of hyporheic zones. Water Resour Res. 2012 Nov;48(11):15.
  7. 7. McClain ME, Boyer EW, Dent CL, Gergel SE, Grimm NB, Groffman PM, et al. Biogeochemical hot spots and hot moments at the interface of terrestrial and aquatic ecosystems. Ecosystems. 2003 Jun;6(4):301–12.
  8. 8. Gomez‐Velez J, Wilson J, Cardenas M, Harvey J. Flow and residence times of dynamic river bank storage and sinuosity‐driven hyporheic exchange. Water Resour Res. 2017;53(10):8572–95.
  9. 9. Gomez J, Wilson J. Age distributions and dynamically changing hydrologic systems: Exploring topography‐driven flow. Water Resour Res. 2013;49(3):1503–22.
  10. 10. Larsen LG, Harvey JW, Maglio MM. Dynamic hyporheic exchange at intermediate timescales: Testing the relative importance of evapotranspiration and flood pulses. Water Resour Res. 2014 Jan;50(1):318–35.
  11. 11. Kirchner JW, Neal C. Universal fractal scaling in stream chemistry and its implications for solute transport and water quality trend detection. Proceedings of the National Academy of Sciences. 2013;110(30):12213–8. pmid:23842090
  12. 12. Woessner WW. Stream and fluvial plain ground water interactions: Rescaling hydrogeologic thought. Ground Water. 2000 May-Jun;38(3):423–9.
  13. 13. Singh T, Wu L, Gomez‐Velez JD, Lewandowski J, Hannah DM, Krause S. Dynamic Hyporheic Zones: Exploring the Role of Peak Flow Events on Bedform‐Induced Hyporheic Exchange. Water Resour Res. 2019;55(1):218–35.
  14. 14. Wolke P, Teitelbaum Y, Deng C, Lewandowski J, Arnon S. Impact of Bed Form Celerity on Oxygen Dynamics in the Hyporheic Zone. Water. 2020;12(1):62.
  15. 15. Stegen JC, Johnson T, Fredrickson JK, Wilkins MJ, Konopka AE, Nelson WC, et al. Influences of organic carbon speciation on hyporheic corridor biogeochemistry and microbial ecology. Nature Communications. 2018;9:1–11. pmid:29317637
  16. 16. Damköhler G. Einflüsse der Strömung, Diffusion und des Wärmeüberganges auf die Leistung von Reaktionsöfen.: I. Allgemeine Gesichtspunkte für die Übertragung eines chemischen Prozesses aus dem Kleinen ins Große. Zeitschrift für Elektrochemie und angewandte physikalische Chemie. 1936;42(12):846–62.
  17. 17. Briggs MA, Lautz LK, Hare DK. Residence time control on hot moments of net nitrate production and uptake in the hyporheic zone. Hydrol Proc. 2014 May 30;28(11):3741–51.
  18. 18. Fritz BG, Arntzen EV. Effect of rapidly changing river stage on uranium flux through the hyporheic zone. Ground Water. 2007 Nov-Dec;45(6):753–60. pmid:17973753
  19. 19. Watson JA, Cardenas MB, Ferencz SB, Knappett PS, Neilson BT. The effects of floods on the temperature of riparian groundwater. Hydrol Proc. 2018;32(9):1267–81.
  20. 20. Arntzen EV, Geist DR, Dresel PE. Effects of fluctuating river flow on groundwater/surface water mixing in the hyporheic zone of a regulated, large cobble bed river. Riv Res Appl. 2006 Oct;22(8):937–46.
  21. 21. McAllister SM, Barnett JM, Heiss JW, Findlay AJ, MacDonald DJ, Dow CL, et al. Dynamic hydrologic and biogeochemical processes drive microbially enhanced iron and sulfur cycling within the intertidal mixing zone of a beach aquifer. Limnol Oceanogr. 2015;60(1):329–45.
  22. 22. Stegen JC, Fredrickson JK, Wilkins MJ, Konopka AE, Nelson WC, Arntzen EV, et al. Groundwater-surface water mixing shifts ecological assembly processes and stimulates organic carbon turnover. Nature Communications. 2016 Apr 7;7:11237. pmid:27052662
  23. 23. Hester ET, Eastes LA, Widdowson MA. Effect of surface water stage fluctuation on mixing‐dependent hyporheic denitrification in riverbed dunes. Water Resour Res. 2019;55(6):4668–87.
  24. 24. Marzadri A, Tonina D, Bellin A, Valli A. Mixing interfaces, fluxes, residence times and redox conditions of the hyporheic zones induced by dune-like bedforms and ambient groundwater flow. Adv Water Resour. 2016;88:139–51.
  25. 25. Poole GC, Stanford JA, Running SW, Frissell CA. Multiscale geomorphic drivers of groundwater flow paths: subsurface hydrologic dynamics and hyporheic habitat diversity. J N Am Benthol Soc. 2006 Jun;25(2):288–303.
  26. 26. Sawyer AH, Cardenas MB, Bomar A, Mackey M. Impact of dam operations on hyporheic exchange in the riparian zone of a regulated river. Hydrol Proc. 2009 Jul 15;23(15):2129–37.
  27. 27. Kaufman MH, Cardenas MB, Buttles J, Kessler AJ, Cook PLM. Hyporheic hot moments: Dissolved oxygen dynamics in the hyporheic zone in response to surface flow perturbations. Water Resour Res. 2017;53(8):6642–62.
  28. 28. Valett HM, Morrice JA, Dahm CN, Campana ME. Parent lithology, surface-groundwater exchange, and nitrate retention in headwater streams. Limnol Oceanogr. 1996 Mar;41(2):333–45.
  29. 29. Hartman M, Dresel P. Hanford site groundwater monitoring for fiscal year 1997. Pacific Northwest Lab., 1998.
  30. 30. Arntzen E. In-situ testing and water quality sampling in the hyporheic zone of the Columbia River, Hanford Reach, Washington. Portland State University. 2001.
  31. 31. Office TNMU-F-AAPF, cartographer FSA 10:1 NAIP Imagery m_4611946_ne_11_1_20150625_20151102 3.75 x 3.75 minute JPEG2000
  32. 32. Ghosh RN, Baker GL, Ruud C, Nocera DG. Fiber-optic oxygen sensor using molybdenum chloride cluster luminescence. Appl Phys Lett. 1999;75(19):2885–7.
  33. 33. Ghosh RN, Askeland PA, Kramer S, Loloee R. Optical dissolved oxygen sensor utilizing molybdenum chloride cluster phosphorescence. Appl Phys Lett. 2011;98(22):221103.
  34. 34. Ghosh RN, Loloee R, Askeland PA, Weeks CT. Optical sensor and sensing system for oxygen monitoring in fluids using molybdenum cluster phosphorescence. US Patent 10,520,435, issued Dec 31, 2019. 2019 Dec. 31, 2019.
  35. 35. Ren H, Cromwell E, Kravitz B, Chen X. Using deep learning to fill spatio-temporal data gaps in hydrological monitoring networks. Hydrology and Earth System Sciences Discussions. 2019:1–20.
  36. 36. Chatfield C, Collins A. Introduction to multivariate analysis: CRC Press; 1981.
  37. 37. Ahmed B, Cao B, Mishra B, Boyanov MI, Kemner KM, Fredrickson JK, et al. Immobilization of U (VI) from oxic groundwater by Hanford 300 Area sediments and effects of Columbia River water. Water research. 2012;46(13):3989–98. pmid:22683408
  38. 38. Voltz T, Gooseff M, Ward AS, Singha K, Fitzgerald M, Wagener T. Riparian hydraulic gradient and stream‐groundwater exchange dynamics in steep headwater valleys. Journal of Geophysical Research: Earth Surface. 2013;118(2):953–69.
  39. 39. Larkin RG, Sharp JM. On the relationship between river-basin geomorphology, aquifer hydraulics, and ground-water flow direction in alluvial aquifers. Geol Soc Am Bull. 1992;104:1608–20.
  40. 40. Johnson T, Versteeg R, Thomle J, Hammond G, Chen X, Zachara J. Four‐dimensional electrical conductivity monitoring of stage‐driven river water intrusion: Accounting for water table effects using a transient mesh boundary and conditional inversion constraints. Water Resour Res. 2015;51(8):6177–96.
  41. 41. Liu Y, Xu F, Liu C. Coupled hydro-biogeochemical processes controlling Cr reductive immobilization in Columbia River hyporheic zone. Environ Sci Technol. 2017;51(3):1508–17. pmid:27996242
  42. 42. Shuai P, Cardenas MB, Knappett PS, Bennett PC, Neilson BT. Denitrification in the banks of fluctuating rivers: The effects of river stage amplitude, sediment hydraulic conductivity and dispersivity, and ambient groundwater flow. Water Resour Res. 2017;53(9):7951–67.
  43. 43. Zheng LZ, Cardenas MB, Wang LC. Temperature effects on nitrogen cycling and nitrate removal-production efficiency in bed form-induced hyporheic zones. J Geophys Res-Biogeo. 2016 Apr;121(4):1086–103.
  44. 44. Soulsby C, Malcolm I, Tetzlaff D, Youngson A. Seasonal and inter‐annual variability in hyporheic water quality revealed by continuous monitoring in a salmon spawning stream. Riv Res Appl. 2009;25(10):1304–19.
  45. 45. Cardenas MB, Wilson JL. The influence of ambient groundwater discharge on exchange zones induced by current-bedform interactions. J Hydrol. 2006 Nov 30;331(1–2):103–9.
  46. 46. Fox A, Laube G, Schmidt C, Fleckenstein J, Arnon S. The effect of losing and gaining flow conditions on hyporheic exchange in heterogeneous streambeds. Water Resour Res. 2016;52(9):7460–77.
  47. 47. Galloway J, Fox A, Lewandowski J, Arnon S. The effect of unsteady streamflow and stream-groundwater interactions on oxygen consumption in a sandy streambed. Scientific reports. 2019;9(1):1–11. pmid:30626917
  48. 48. Trauth N, Schmidt C, Vieweg M, Oswald SE, Fleckenstein JH. Hydraulic controls of in‐stream gravel bar hyporheic exchange and reactions. Water Resour Res. 2015;51(4):2243–63.
  49. 49. Trauth N, Schmidt C, Maier U, Vieweg M, Fleckenstein JH. Coupled 3-D stream flow and hyporheic flow model under varying stream and ambient groundwater flow conditions in a pool-riffle system. Water Resour Res. 2013 Sep;49(9):5834–50.
  50. 50. De Falco N, Boano F, Bogler A, Bar‐Zeev E, Arnon S. Influence of stream‐subsurface exchange flux and bacterial biofilms on oxygen consumption under nutrient‐rich conditions. Journal of Geophysical Research: Biogeosciences. 2018;123(7):2021–34.