Evaluating topography-based predictions of shallow lateral groundwater discharge zones for a boreal lake-stream system
Abstract
Groundwater discharge along streams exerts an important influence on biogeochemistry and thermal regimes of aquatic ecosystems. A common approach for predicting locations of shallow lateral groundwater discharge is to use digital elevation models (DEMs) combined with upslope contributing area algorithms. We evaluated a topography-based prediction of subsurface discharge zones along a 1500 m headwater stream reach using temperature and water isotope tracers. We deployed fiber-optic distributed temperature sensing instrumentation to monitor stream temperature at 0.25 m intervals along the reach. We also collected samples of stream water for the analysis of its water isotope composition at 50 m intervals on five occasions representing distinct streamflow conditions before, during, and after a major rain event. The combined tracer evaluation showed that topography-predicted locations of groundwater discharge were generally accurate; however, predicted magnitude of groundwater inflows estimated from upslope contributing area did not always agree with tracer estimates. At the catchment scale, lateral inflows were an important source of streamflow at base flow and peak flow during a major rain event; however, water from a headwater lake was the dominant water source during the event hydrograph recession. Overall, this study highlights potential utility and limitations of predicting locations and contributions of lateral groundwater discharge zones using topography-based approaches in humid boreal regions.
Key Points
- Topography-based predicted locations of discrete lateral subsurface inflow along a boreal stream agreed with tracer estimates
- Topography-based predicted magnitude of lateral subsurface inflow did not always agree with tracer estimates
- Flow from a headwater lake was much greater than lateral subsurface inflow contributions to streamflow following a rain event
1 Introduction
Biogeochemical and thermal regimes of fluvial ecosystems are strongly influenced by location and magnitude of groundwater inflows along streams [Hunt et al., 2006; Pacific et al., 2010; Leach and Moore, 2011]. Heat and nutrients, such as carbon and nitrogen, are transported from hillslopes to streams primarily through subsurface flow paths, which provide a critical connection between terrestrial and aquatic environments [Grabs et al., 2012; Blume and Van Meerveld, 2015; Kurylyk et al., 2015]. The spatial arrangement of these subsurface flow paths and the time that they connect to surface waters influence the heterogeneity of stream ecosystems by creating spatially variable streamflow, nutrient, and stream temperature conditions [McGuire and McDonnell, 2010; Mallard et al., 2014]. Zones of groundwater inflow often create thermal refugia for aquatic organisms and can be hotspots of biological activity [Ebersole et al., 2001; McClain et al., 2003; Dugdale et al., 2013]. Therefore, identification of lateral groundwater inflow locations along streams is needed to effectively manage and protect aquatic ecosystems from terrestrial disturbances such as forest harvesting, urbanization, and agricultural activities [Osborne and Kovacic, 1993; Gorsevski et al., 2008; Laudon et al., 2016].
Various approaches have been used to identify and predict locations of groundwater inflow zones. Examples include use of wells and piezometers, hydrologic tracers, thermal imaging, and groundwater flow modelling [Krause et al., 2007; Schuetz and Weiler, 2011; Atkinson et al., 2015]. A common approach for predicting locations of shallow lateral groundwater inflow along streams is the use of digital elevation models (DEMs) and upslope contributing area algorithms [Jencso et al., 2009; Grabs et al., 2010]. This approach is attractive since high-resolution DEMs are readily available for many regions and the implementation is relatively quick and requires limited data compared to alternate approaches for determining groundwater discharge locations, such as using groundwater flow models or field measurements [Batelaan et al., 2003; Kalbus et al., 2006]. The combined DEM and upslope contributing area algorithm approach relies on the assumption that subsurface water flow paths, often the dominant process by which water travels from land to surface water in forested catchments, are related to surface topography. This assumption has been shown to be reasonable for some catchments [Anderson and Burt, 1978; Rodhe and Seibert, 1999; Jencso et al., 2009]; however, others have highlighted that this assumption may be inappropriate in certain environments and under specific conditions [Grayson and Western, 2001; Devito et al., 2005a]. For example, topography-based approaches have been found to not work well in catchments with extensive bedrock fractures [Genereux et al., 1993], where bedrock or till topography differs from surface topography [Hutchinson and Moore, 2000; Tromp-van Meerveld and McDonnell, 2006], or in arid to subhumid environments where evapotranspiration dominates the water balance [Carey and Woo, 2001; Devito et al., 2005b]. Consequently, there is a need to further evaluate topography-based predicted locations and relative contributions of groundwater inflows to streams if these approaches are to be used for management purposes, such as protecting surface water resources from forest harvest impacts [Laudon et al., 2016; Tiwari et al., 2016].
Hydrologic tracers provide another method to determine groundwater input locations to streams. A variety of tracers exist and each provides unique insights on hydrologic processes [Abbott et al., 2016]. For example, distributed temperature sensing (DTS) using fiber optic cables allows for high-resolution temperature measurements in space and time [Selker et al., 2006b; Tyler et al., 2009] and has been used over the last decade in hydrology research to examine and identify locations of groundwater inflow to surface waters [Briggs et al., 2012; Matheswaran et al., 2014]. Thermal tracing exploits temperature differences between groundwater and surface water: the former typically having a more stable annual temperature pattern [Anderson, 2005]. Much of the DTS research has been focused on identifying discrete groundwater upwelling along streambeds of lowland rivers [Briggs et al., 2012; Krause et al., 2012; Matheswaran et al., 2014; Hare et al., 2015; Sebok et al., 2015]. In addition, some studies have used DTS to identify groundwater inputs along headwater streams [Selker et al., 2006a; Westhoff et al., 2011]. The majority of these studies have used DTS observations for post hoc identification of groundwater inflows. To our knowledge, no studies have used DTS methods to evaluate a priori predictions of lateral groundwater inflow locations.
Using heat as a tracer is attractive for hydrology research because high-resolution measurements are possible in both space and time; however, temperature is highly reactive which limits its utility as a tracer in certain conditions and environments [Anderson, 2005]. In addition, distinct thermal signatures of different water sources are needed to be able to infer hydrologic processes from temperature observations. In comparison to heat, stable water isotopes, such as
O and
H, have low reactivity and can potentially provide more robust information on water flow path and source [Abbott et al., 2016]. However, conventional sampling methods for stable water isotopes limit the number of measurements that can be made in space and time, particularly in comparison to DTS methods for heat tracing.
The overall objective of this study was to employ temperature and stable isotope tracers to evaluate predictions of shallow lateral groundwater inflow location and magnitude made using a topography-based approach along a boreal stream. Our specific research questions were: (1) Can maps of upslope contributing area successfully predict locations of discrete and diffuse shallow lateral groundwater inflows to a stream? and (2) Can upslope contributing area along a stream channel be used to predict the relative contribution of shallow lateral groundwater inflow to stream discharge? By addressing these questions, forest and water resources managers will be better informed about the advantages and limitations of hydromapping tools (i.e., maps based on upslope contributing area analyses and similar products). These hydromapping tools have been proposed by researchers as useful for managers to help minimize human impacts on surface water quality in humid environments where shallow groundwater systems may be sensitive to land disturbance such as forest harvesting and soil rutting caused by machine operation [Murphy et al., 2008; Laudon et al., 2016].
The study was conducted along a 1500 m stream reach located in the boreal zone of northern Sweden. The stream reach has no tributaries but drains a small headwater lake. Because our study focused on a headwater lake and outlet stream system, we also had the opportunity to address a related research objective examining the relative roles of groundwater, lake water, and event water (rain) contributions to stream discharge before, during, and after a major rain event. This research objective complements the two specific research questions stated above by providing context for the importance of shallow lateral groundwater contributions to streamflow in catchments with headwater lakes.
2 Study Site
The study was conducted within the Krycklan Catchment Study located approximately 50 km northwest of the city of Umeå in northern Sweden [Laudon et al., 2013]. At the location of an outlet hydrometric station, the Krycklan catchment is 6790 ha and ranges in elevation from 127 to 373 m above sea level. Bedrock is composed of Svecofennian gneissic bedrock with metasediments and metagraywacke covered by a layer of quaternary deposits of glacial till that varies in thickness up to tens of meters [Laudon et al., 2013]. The climate of Krycklan is defined as cold temperate humid with persistent snow cover during the winter season. The 30 year (1981–2010) mean annual air temperature and precipitation are 1.8°C and 614 mm, respectively [Laudon et al., 2013]. About 35–50% of total annual precipitation falls as snow and the mean period of snow cover is 167 days per year (1981–2015) [Laudon and Ottosson-Löfvenius, 2016].
The study focused on a 1500 m stream reach bounded by two hydrometric stations (C5 upstream and C6 downstream; Figure 1). The catchment areas drained at C5 and C6 are 65 and 110 ha, respectively. The stream reach has no tributaries although saturation excess overland flow is sometimes visible in select areas along the riparian zone during wet conditions. C5 is located about 100 m downstream of Stortjärnen Lake, which has a surface area of approximately 4 ha and is primarily fed by a mire system. The headwater lake provided advantages for using hydrologic tracers, since the lake water had thermal and isotopic signatures that were distinct from those of the hillslope groundwater during the study period. The catchment area draining the reach between C5 and C6 is primarily covered with forest dominated by Scots pine (Pinus sylvestris) and Norway spruce (Picea abies) with some birch (Betula pubescens). Well-developed iron podzols dominate the forest soils. The organic content increases with proximity to the stream and the riparian zones are dominated by peat. The channel bankfull width of the 1500 m study reach ranges between 0.3 and 2 m and the mean channel slope is about 3%. Streambed substrate consists of dense decomposed organic matter, sand, and cobble/boulder sections.
Map of the study reach showing locations of the lake, flumes, inflow isotope sample locations, subsurface temperature measurements, and logarithm of the D8 upslope contributing area. Distances downstream of C5 are provided for reference and indicate the five major inflow zones predicted by the upslope contributing area mapping (350, 550, 750, 975, and 1300 m). Note that the black lines do not represent locations of surface water or tributaries to the main channel, but topography-predicted networks of subsurface flow paths. Location of the Krycklan Study Catchment within Sweden is shown in the inset.
3 Methods
3.1 Field Observations
Hourly discharge was estimated at the lake outlet (C5) and downstream (C6) hydrometric sites using stage measurements at H-type flumes and rating curves developed using manual streamflow measurements made over a range of flow conditions at both stations. Hourly precipitation was measured using a tipping-bucket gauge at the Svartberget meteorological station located about 1 km from the C6 station [Laudon et al., 2013].
A 1700 m fiber optic cable was installed in the C5-C6 stream reach. The cable used was a 6 channel tight buffered fiber optic cable with aramid yarn protection and polyurethane coating (Silixa Ltd, Elstree, UK). The cable was installed along the streambed in the middle of the channel and fixed in place using rocks and weights made from stainless steel. The cable was connected to a Silixa XT-DTS (Silixa Ltd, Elstree, UK) with a sampling resolution of 0.25 m and temperature resolution of 0.01°C. The instrument was set in double ended measurement mode and temperature scans logged at 6 min intervals. Minimal fiber damage and splicing to the cable resulted in a clear measurement signal with no abrupt offsets. Calibration was conducted by placing 15 m of coiled cable in an ice bath of 0°C located in an electric cooler and initially comparing temperature measurements with a Greisinger GMH 3750 handheld thermometer (accuracy ±0.03°C) and continuous measurements with a Class A PT100 thermistor (accuracy ±0.1°C). Colocated submersible temperature loggers (Tidbit v2 Temp, Onset Computer Corporation, accurate to ±0.2°C) and 5 m coiled lengths of cable distributed along the stream at 10 locations were used to check for measurement drift and inconsistencies. Temperature loggers were installed at sites with sufficient water depth so that they were not exposed during low flows. The loggers were shielded with slotted white PVC pipe and were calibrated at 0 and 20°C before and after field deployment. Measurements were logged at 15 min intervals. Mean hourly water temperature comparisons between the DTS and submersible temperature loggers were within ±0.2°C. We also conducted manual spot temperature measurements to check for lateral variations in stream temperature to ensure the assumption of complete mixing was met. Surveys suggested lateral temperature variations were within ±0.1°C, which corresponds with the accuracy of the manual temperature sensor used. The DTS device was housed in an insulated flume hut to help keep the unit temperature stable. The DTS system logged from 3 September 2015 to 5 October 2015. Hourly air temperature measurements at the lake outlet (C5) and downstream (C6) hydrometric sites, as well as the Svartberget meteorological station were used to diagnose when parts of the DTS cable were out of the water.
Subsurface water temperature measurements were made at four near-stream locations to provide an estimate of subsurface inflow temperature (Figure 1). Measurements were made using TruTrack water level sensors equipped with thermistors located at the bottom of the instrument (WT-HR; Christchurch, New Zealand; accuracy ± 0.3°C). TruTracks were placed in groundwater wells that were installed to depths of 0.4–0.8 m below the ground surface and temperature measurements were stored every 60 min. Wells were located within 3 m of the stream channel and three of the four groundwater wells were installed at sites associated with relatively large upslope contributing areas. Wells and temperature sensors were installed at all the predicted major discrete subsurface inflow locations; however, due to instrument failure we were limited to measurements made at the four locations shown in Figure 1.
Five sets of manual sampling for
O measurements were made during different hydrologic conditions: two during base flow, two during peak flow at the downstream C6 hydrometric station, and one during peak flow at the lake outlet (C5) hydrometric station. Water samples were collected every 50 m along the stream reach (28 samples per sampling set) and at any obvious seepage locations where water from the hillslope was observed entering the stream at the channel bank (1–3 samples per sampling set). Sampling for each set took under 1.5 h to complete. Samples were collected in dark glass bottles and refrigerated prior to analysis. Water samples were analyzed using a Picarro cavity ringdown laser spectrometer L2130-i with a vaporizer module A0211 (Picarro, Inc., Santa Clara, USA). Each sample was analyzed four times and averaged after correcting for drift and memory effects [van Geldern and Barth, 2012]. Isotopic signatures of the samples were calibrated using laboratory standards calibrated against three International Atomic Energy Agency (IAEA) official standards: the Vienna Standard Mean Ocean Water (VSMOW), the Greenland Ice Sheet Precipitation (GISP), and the Standard Light Antarctic Precipitation (SLAP). Accepted standard deviation of the control samples was 0.15% for
O.
3.2 Analysis
3.2.1 Lateral Groundwater Inflow Location Prediction and Evaluation
Shallow lateral groundwater inflow locations along the stream were predicted using LiDAR derived DEMs and standard upslope contributing area and flow accumulation algorithms (Figure 1). The DEM used was the Swedish National elevation model generated by the Swedish Mapping, Cadastral, and Land Registration Authority. The base DEM has a cell resolution of 2 m and was generated from a point cloud with a density of 0.5–1 points per m2 with a horizontal and vertical error of 0.4 and 0.1 m, respectively. We resampled the 2 m DEM, using bilinear interpolation in SAGA GIS [Conrad et al., 2015], to generate 5 and 10 m resolution DEMs in order to compare the influence of DEM resolution on predicted inflow locations.
Sinks in the DEM were solved by breaching using GoSpatial [Lindsay, 2015]. The hydrologically corrected DEM was used to model stream networks using the deterministic-8 (D8) algorithm [O'Callaghan and Mark, 1984] in SAGA GIS. We compared six upslope contributing area and flow accumulation algorithms for predicting locations of lateral groundwater inflow (deterministic 8 [O'Callaghan and Mark, 1984]; deterministic infinity [Tarboton, 1997]; multiple flow direction [Freeman, 1991]; multiple flow direction based on maximum downslope gradient [Qin et al., 2011]; triangular multiple flow direction [Seibert and McGlynn, 2007]; and random eight-node [Fairfield and Leymarie, 1991]).
Temperature and
O longitudinal profiles were examined during contrasting hydrologic conditions (i.e., base flow, peak flow at C6 (downstream) and C5 (lake outlet)) and compared to the predicted upslope contributing area along the stream. Step changes in temperature and
O downstream profiles were assumed to be associated with input of thermally and isotopically distinct groundwater [Selker et al., 2006a]. We also examined the standard deviation of diurnal temperature which has been used to diagnose locations of groundwater inflow using DTS [Lowry et al., 2007; Matheswaran et al., 2014]. Throughout this manuscript, we use the terms prediction to refer to groundwater inflow locations obtained from upslope contributing area methods and identification to refer to those obtained from tracer approaches.
3.2.2 Relative Contribution of Groundwater Inflow
We used two approaches to evaluate predicted relative contribution of lateral groundwater inflows to the stream based on upslope contributing area. First, we calculated theoretical longitudinal stream temperature profiles using upslope contributing area and assumptions on stream channel mixing and compared those to observed profiles measured with the DTS sensor. Second, we focused on predicted major discrete inflow zones and compared predicted relative contributions based on upslope contributing area with mixing equation estimates using temperature and stable isotope measurements.
Theoretical longitudinal temperature profiles were generated using upslope contributing area along the stream channel and assuming that all downstream temperature changes were due solely to mixing of discrete and diffuse lateral groundwater inflows. This assumption is reasonable for this system, since the temperature of lake water entering the stream may be near equilibrium with the energy exchanges at the water surface [Garner et al., 2014]. Therefore, deviations between this theoretical profile and observed temperature may be attributed to errors in using upslope contributing area as a proxy for actual lateral inflows, not accounting for energy exchanges at the stream surface, and errors in groundwater inflow temperature.
(1)
is stream temperature (°C) at channel grid cell i, Tinf is lateral groundwater inflow temperature (°C) determined from the subsurface temperature measurements, and Qi is stream discharge at i and is computed as:
(2)
is the upslope contributing area along the stream channel at i,
is the catchment area at the C5 (lake outlet) hydrometric station subtracted from the catchment area at the C6 (downstream) hydrometric station, giving the catchment area of the C5-C6 stream reach, and
and
are discharge at the C5 (lake outlet) and C6 (downstream), respectively. Uncertainty bounds for the longitudinal temperature profile were accounted for by incorporating errors in discharge from C5 and C6 (±15% based on replicated manual measurements) and uncertainty in Tinf (±0.6°C based on the maximum daily range of measured groundwater temperatures).
(3)
(4)
O samples were also taken.
(5)
is the standard deviation of the ratio of Qinf to Qus,
is the variance of upstream temperature,
is the variance of the downstream temperature,
is the variance of the inflow temperature, and
is the variance of the DTS sensor (0.1°C based on the calibration procedure).We estimated Tus and Tds from the DTS data using the fitted line approach proposed by Selker et al. [2006a]. This approach involved fitting linear regression models to temperature data from stream reaches above and below the discrete groundwater inflow zones where surface water and groundwater appeared well-mixed by evidence of a relatively stable longitudinal temperature signal. The upstream and downstream temperatures were then estimated by projecting the slope of these fitted lines to the centre of the mixing reach. Temperature data for 10–50 m above and below the mixing reach were used to fit the regression models. The length depended on the variability of the temperature signal and which length provided the best fit. The fit of the regression models were used to estimate
and
. The variance of the groundwater well temperature measurements were used to estimate
.
O measurements. The
O measurements were used in the mixing equations (equation 4) to estimate relative groundwater inflow and uncertainty was accounted for using the approach of Genereux [1998]. Upslope contributing area, as determined from the DEM, was used to predict the relative contribution of groundwater (R) at major discrete inflow locations (i) using the following equation:
(6)
is the upslope contributing area for lateral inflow at location i.3.2.3 Changes in Streamflow Contributions Over Time
(7)
, Qlake, Qgw, and Qppt are discharge (L/s) at the catchment outlet (C6 station), discharge at the lake outlet (C5 station), discharge from water stored as groundwater in the catchment prior to the rain event (preevent water), and discharge from precipitation that fell during the rain event (event water), respectively.In order to partition the streamflow components, we followed the two-step approach presented by St Amour et al. [2005] for a three-component separation involving one tracer. First, we used a standard two-component hydrograph separation to estimate lake outlet (Qlake) and hillslope (
) contributions. Second, we further partitioned the hillslope contributions, again using a standard two-component approach, into preevent groundwater (Qgw) and event precipitation (Qppt). Uncertainty in the separation estimates was determined using the methods described by Genereux [1998] and standard error propagation methods [Bevington and Robinson, 2003] to propagate errors through the two-step approach.
4 Results
4.1 Predicted Locations of Lateral Groundwater Inflow
The six upslope contributing area and flow accumulation algorithms generally predicted similar locations of lateral subsurface inflow along the stream between the lake outlet (C5) and downstream (C6) hydrometric stations (Figures 2, 3, and supporting information Figure S1). In addition, the predicted locations were generally consistent across the 2, 5 and 10 m DEMs. Although predictions were mostly similar overall, there were a couple noticeable exceptions. Flow routing associated with the 975–1300 m downstream location was sensitive to a small flat area upslope of the channel on the right bank near the 975 m location. The impact of this flat area was most pronounced with the D8 method using the 5 m DEM, which routed the flow and associated upslope contributing area further downstream to the 1300 m inflow location. Field observations of a visible seepage face associated with the predicted inflow at 975 m along the right bank suggest that the D8 prediction using the 5 m DEM is incorrect for that section of the stream. Considering the similarity between predictions made using the different algorithms and DEM resolutions, we decided to use the 2 m D8 predictions for the remaining analyses. This was done for simplicity and because the D8 method is commonly used and computationally efficient.
Cumulative upslope contributing area along the stream computed using six methods (D8: Deterministic 8 [O'Callaghan and Mark, 1984], Dinf: Deterministic Infinity [Tarboton, 1997], MFD: Multiple Flow Direction [Freeman, 1991], MMD: Multiple Flow Direction based on Maximum Downslope Gradient [Qin et al., 2011], MTF: Triangular Multiple Flow Direction [Seibert and McGlynn, 2007], Rho8: Random eight-node [Fairfield and Leymarie, 1991]) and three DEM resolutions (2, 5, and 10 m).
Maps of upslope contributing area along the stream computed using six methods (D8: Deterministic 8 [O'Callaghan and Mark, 1984], Dinf: Deterministic Infinity [Tarboton, 1997], MFD: Multiple Flow Direction [Freeman, 1991], MMD: Multiple Flow Direction based on Maximum Downslope Gradient [Qin et al., 2011], MTF: Triangular Multiple Flow Direction [Seibert and McGlynn, 2007], Rho8: Random eight-node [Fairfield and Leymarie, 1991]) and three DEM resolutions (2, 5, and 10 m).
The upslope contributing area maps predicted five major discrete zones of lateral subsurface inflow along the stream between the C5 (lake outlet) and C6 (downstream) stations that correspond with large step changes in cumulative upslope contributing area along the stream channel (Figures 1-3, and supporting information Figure S1). Of the predicted five major zones of groundwater inflow along the stream reach, one zone consisted of a single discrete contribution (350 m downstream of the lake outlet (C5) station); whereas, the remaining inflow zones consisted of clusters of two or more discrete inflows (550, 750, 975, and 1300 m distances downstream of the lake outlet (C5) station; Figure 1). Upslope contributing areas computed from the 2 m DEM using the D8 algorithm for the five discrete inflows (350, 550, 750, 975, and 1300 m locations) were 2.2, 6.0, 2.4, 10.0, and 2.4 ha, respectively. The cumulative length along the stream of these inflow zones accounts for less than 5% of the total stream reach length but make up approximately 60% of the total predicted contributing area for the C5-C6 stream reach.
4.2 Hydrometeorology Overview
The early part of the study period was characterized by an extended dry period with low streamflow conditions (<1.5 L/s), followed by 67 mm rain falling between 16 and 20 September (Figure 4, top plot). The downstream C6 site responded to the rain event before C5 (lake outlet) and flow peaked at 58 L/s at C6 on 18 September. Comparatively, the lake outlet C5 site had a delayed and muted response to the rain event and peaked at 18 L/s on 21 September. Precipitation
O was enriched during the onset of the major rain event and became more depleted throughout the event (Figure 4, middle plot). The lake outlet C5
O was relatively enriched and remained stable at around −10.9% throughout the study period, consistent with an evaporative influence on the lake water. The downstream C6
O was more depleted than C5 (lake outlet) during the base flow period and became more enriched during the rain event and approached C5 values during the recession limb of the downstream C6 hydrograph. The three groundwater locations sampled for
O (Figure 1) were depleted compared to those of C6 (downstream) during base flow and became more enriched during the beginning of the rain event and converged around −12.0% at the end of the rain event (Figure 4). Water temperatures measured at C5 (lake outlet) were higher than those at C6 (downstream) by 0.5–5°C during the study period (Figure 4, bottom plot). The period of greatest difference between C5 (lake outlet) and C6 (downstream) temperatures was during base flow. Temperatures at the two sites began to converge with the onset of the rain event. Groundwater temperatures ranged between 6.2 and 9.5°C during the study period (Figure 4, bottom plot). During base flow, temperature at C6 and the groundwater inflow temperatures were generally within 1–2°C.
Summary of precipitation (PPT), streamflow (Q),
O isotope sampling, and stream, air and groundwater temperatures during the study period. Vertical grey bands and associated labels in the top plot show sampling periods for manual oxygen isotope sampling. The “Weighted PPT” shown in the middle plot refers to an incremental mean weighted approach for the rainfall isotope signature [McDonnell et al., 1990].
4.3 Predicted Lateral Groundwater Inflow Location Evaluation
A summary of the five isotope sampling periods and temperature measurements are shown in Figure 5 and an animation of the stream temperature dynamics along the study reach is provided in the supporting information. The longitudinal upslope contributing area along the reach highlights the five major zones of predicted lateral inflow as indicated by pronounced step increases and indicated by the grey bands in the Figure 5. During base flow (Base flow 1 and 2), the most upstream discrete inflow (350 m) was not associated with an obvious step change in isotope signature or stream temperature. In contrast, the remaining four inflow zones were associated with a step decrease in
O signature and stream temperature. For the two isotope measurement, periods during the downstream C6 peak flow (a and b which were conducted about 5 h apart),
O signature decreased downstream; however, step changes were less pronounced than during the base flow periods. Inflow zones at approximately 550 and 1300 m downstream were associated with a clear depletion of the
O signature; whereas, step decreases for the other three inflow zones were less pronounced or nonexistent. In contrast, stream temperature showed a considerable step change at the 350 m inflow zone and minor or no change at the other four predicted inflow zones. However, longitudinal patterns in the diurnal standard deviation of stream temperature were associated with the major predicted inflow zones (Figure 5, bottom plot). Finally,
O composition along the reach was relatively uniform during the lake outlet C5 peak flow period. In addition, stream temperature showed only a modest declining pattern downstream (
C difference between C5 and C6) with no strong step changes. The diurnal standard deviation of stream temperature during the day of the lake outlet C5 peak flow showed a gradual increase downstream with no obvious step changes.
Summary of event sampling periods. (top row) Upslope contributing area along the stream (constant in time), (second row) manual
O isotope samples (C6 peak flow sampling periods a and b are indicated in blue and red, respectively), (third row) daily mean stream temperature, and (bottom row) standard deviation of the daily stream temperature. Since both isotopes sampling for the C6 peak flow were conducted on the same day there is only one associated trace for daily mean stream temperature and diurnal standard deviation of stream temperature. Vertical grey bands correspond with predicted locations of major subsurface lateral inflows (350, 550, 750, 975, and 1300 m). Isolated large spikes in the standard deviation figures are associated with dewatered cable sections.
4.4 Relative Groundwater Inflow Contribution
Longitudinal temperature profiles for three hydrologic settings (base flow, downstream C6 and lake outlet C5 peak flows) comparing theoretical and observed temperatures are shown in Figure 6. For base flow, step changes in the theoretical profile generally occur at the same locations as step changes in the observed temperature profile; however, the observed temperature profile has a lower temperature over the first 500 m than the theoretical profile. In addition, the difference in magnitude between observed and theoretical profiles does not appear to vary systematically. During the downstream C6 peak flow, both theoretical and observed temperature profiles generally agree for the first 500 m after which the observed profile is generally stable around 10.4°C, whereas the theoretical profile exhibits continued step changes and reaches 8°C at the downstream C6 site. Observed and theoretical temperature profiles during the lake outlet C5 peak flow both show a relatively linear pattern and the theoretical uncertainty bounds capture the observed profile.
Longitudinal stream temperature patterns during (a) base flow, (b) downstream C6 peak flow a, and (c) lake outlet C5 peak flow. The black line represents the observed temperature profile measured using the DTS sensor and the red line represents the theoretical downstream temperature change accounting for groundwater water mixing. The error band associated with the theoretical temperature profile accounts for uncertainties in discharge and groundwater temperature. The results for C6 peak flow b are not shown as they were similar to those for C6 peak flow a.
Figure 7 shows the percent contribution of the five major inflow locations relative to streamflow immediately upstream of the inflow, predicted using discharge and upslope contributing area,
O isotopes, and three temperature approaches. For the Base flow 1 period, the upslope contributing area predictions and tracer estimates agree for the 350, 975, and 1300 m inflow zones. For the inflow zones at 550 m and 750 m, the tracer approaches suggest relative contributions of inflow of between 20–45% and 5–20%, respectively, which are greater than predicted by the upslope contributing area approach (5% and 3%, respectively). For the Base flow 2 period, there was relatively strong agreement between approaches for the 350, 750, and 1300 m locations. The isotope method suggested a greater relative contribution of inflow (
) than the upslope contributing area (
) and temperature approaches (
) for the 550 m location. The upslope contributing area (
) and isotope methods (
) were similar for the 975 m location and were greater than the three temperature approaches (
). During the downstream C6 peak flow periods, the upslope contributing area and isotope approaches tended to predict similar relative contributions, although there was considerable uncertainty in the isotope estimates due, in part, to large variability in the estimated groundwater inflow isotope signature (–12.23 to −11.28%). In contrast, the utility of the temperature approach was diminished during the downstream C6 peak flow period and predictions were at or near 0% for most of the inflow locations, with the exception of the 350 m zone where upslope contributing area, isotope, and temperature methods tended to provide similar predictions (
). For the lake outlet C5 peak flow period, all five approaches suggested only minimal relative contribution (<10%) from each of the five inflow locations.
Predicted relative inflow contribution of five inflow locations compared to streamflow immediately upstream of the inflow using the upslope contributing area approach and
O isotope and stream temperature (T
) tracers for the five sampling times. Error bars represent ± one standard deviation.
4.5 Changes in Streamflow Contributions Over Time
Hydrograph separation using
O isotopes highlight the changing contributions of lake water, preevent water, and event water to discharge measured at C6 before, during and after the rain event (Table 1). During base flow conditions, discharge at the downstream C6 site was estimated to comprise about one quarter lake water and three quarters preevent groundwater. During the rain event, event water accounted for approximately 30–51% of the discharge at C6 and the contribution of lake water diminished to 0.1%. However, there is considerable uncertainty in the hydrograph separation analysis during the C6 peak flow period due to minimal differences in isotope signature between water sources, as well as uncertainty in the isotope signature of hillslope water (the sum of preevent and event water). The relative contributions of preevent water and event water during the lake outlet C5 peak flow period are also subject to considerable uncertainty; however, the hydrograph analysis suggests that lake water was the dominant water source during this period (
).
| Event | Discharge (L/s) | Contribution to C6 Discharge (%) | |||
|---|---|---|---|---|---|
| Lake Outlet C5 | Downstream C6 | Lake Water | Preevent Water | Event Water | |
| Base flow 1 | 1.0 | 1.3 | 21 ± 6 | 79 ± 6 | 0 |
| Base flow 2 | 1.0 | 2.9 | 24 ± 7 | 76 ± 9 |
 |
| Downstream C6 peak flow a | 2.2 | 21.7 | 16 ± 21 | 54 ± 33 | 30 ± 33 |
| Downstream C6 peak flow b | 5.1 | 58.4 |
 |
49 ± 86 | 51 ± 86 |
| Lake outlet C5 peak flow | 17.5 | 20.0 | 75 ± 3 | 12 ± 63 | 13 ± 63 |
- a Event refers to the sampling periods shown in Figure 4. Discharge (L/s) is the hourly value observed at the lake outlet C5 and downstream C6 hydrometric stations during the sampling period. Estimated contribution (% and ± one standard deviation) of lake, preevent, and event water to C6 discharge based on isotope hydrograph separation.
5 Discussion
5.1 Predictions of Lateral Groundwater Inflow Location
A topography-based approach for predicting locations of lateral inflow worked reasonably well for our study reach based on results from the dual tracer evaluation. During base flow conditions, four of the five predicted major discrete lateral inflow locations were associated with abrupt downstream changes in longitudinal temperature and
O patterns. One predicted location, around 350 m downstream of C5, was not associated with a major step change in tracer during base flow conditions, although it was associated with a step change in temperature during the rain event.
Patterns of
O during base flow suggest evidence of diffuse inflows along the stream, indicated by a gradual longitudinal depletion downstream (e.g., Figure 5: the first 200 m downstream of C5). This is also corroborated by some of the observed temperature patterns. In particular, the decreasing downstream temperature pattern for the first 300 m was consistent with the predicted diffuse lateral inflows suggested by the upslope contributing area map, at least during base flow and C6 peak flow periods (Figures 6a and 6b). There are no abrupt step changes in temperature and
O patterns beyond those corresponding with the upslope contributing area predictions. This suggests that the topography-based predictions were successful at capturing all major discrete subsurface inflows along the stream reach.
A number of studies have questioned the role of topography as a first-order control on runoff processes, particularly for low-relief headwater catchments [Buttle et al., 2004; Devito et al., 2005a; Tetzlaff et al., 2015; Klaus et al., 2015]. Results from our study site suggest that topography can be used to successfully predict the location of lateral subsurface water inflows. Other research from humid Scandinavian forests dominated by till soils also suggest that topography is a first-order control on subsurface water distribution in this landscape [Rodhe and Seibert, 1999; Ågren et al., 2014]. In contrast to other regions [Genereux et al., 1993; Hutchinson and Moore, 2000; Buttle et al., 2004; Devito et al., 2005a], the ability to predict lateral groundwater inflow at our site from topography may be due to the relatively humid conditions and the rapid decrease in hydraulic conductivity with depth of these soils [Nyberg et al., 2001; Bishop et al., 2011; Ameli et al., 2016]. These conditions likely promote a shallow water table with subsurface flows following surface topographic gradients [Hutchinson and Moore, 2000; Ågren et al., 2014]. Although we report results from a single 1500 m stream reach, our findings are consistent with other research from humid boreal zones that have found relations between topography and subsurface water distribution, such as locations of wet areas and high plant species richness being associated with large upslope contributing areas [Rodhe and Seibert, 1999; Kuglerová et al., 2014a; Ågren et al., 2014].
5.2 Lateral Inflow Contributions to Streamflow
The upslope contributing area map successfully identified locations of lateral inflow; however, there was less success in predicting relative contributions of these lateral inflows to streamflow. Contributions of lateral inflows associated with both small and large upslope contributing areas did not always match estimates of inflows based on the tracers. For example, the magnitude of diffuse inflows was underestimated by the upslope contributing area method along sections of the stream reach during base flow conditions (Figure 6a). In contrast, predicted contributions for the five major discrete inflows during base flow generally agreed with tracer-based estimates of inflows although there were considerable uncertainties (Figure 7). For those locations with disagreements, the predicted contributions using upslope contributing area were generally less than those estimated from the tracers. During the downstream C6 peak flow, nearly all the tracer estimated contributions were lower than the upslope contributing area predictions (Figure 7). During the lake outlet C5 peak flow period, predicted and tracer-based estimates were similar but suggested that the relative contributions of the major inflows were minimal.
There may be a number of reasons why predicted contributions of lateral inflow did not always agree with tracer estimates. These include complexities in runoff processes not accounted for by the upslope contributing area maps, changes in actual upslope contributing area with time, limitations of the tracers, and incorrect representation of tracer end-members. Discrepancies between predicted and estimated lateral inflow contributions could be due to the inability of the upslope contributing area approach to account for potential nonlinear runoff response behavior. The upslope contributing area approach assumes a linear increase in lateral inflow with streamflow; however, a number of studies have highlighted nonlinear and threshold response behavior due to influences such as variability in soil depths and hillslope water storage, changes in hydraulic conductivity with depth, shifts in dominant runoff processes (e.g., saturated matrix flow to saturation excess overland flow), and preferential flow processes [Buttle et al., 2004; Tromp-van Meerveld and McDonnell, 2006; Zehe et al., 2007; Spence, 2010; Gannon et al., 2014; Nippgen et al., 2015]. It is also possible that spatial variability in subsurface permeability and topography of a confining layer contributed to disagreements in predicted inflow contributions [Hutchinson and Moore, 2000; Ameli et al., 2016]. In addition, upslope contributing areas will be sensitive to DEM resolution and may have also contributed to errors in the predicted lateral inflow contributions [Thompson and Moore, 1996; Sørensen and Seibert, 2007; Ågren et al., 2014]; however, our comparison of different DEM resolutions and flow accumulation algorithms suggests that this a minor source of uncertainty for our site.
Another limitation of the topography-based maps for predicting lateral subsurface inflow is that they assume upslope contributing areas are static in time; therefore, predicted lateral inflow at any point along the stream is proportional to the cumulative upslope area [Beven and Freer, 2001]. The assumption of static contributing areas avoids the need to set up a hydrologic model to account for dynamic changes in upslope contributing area, making this approach more attractive to practitioners. However, as previously highlighted by others, the relative contribution of inflows to streamflow should be expected to vary in time, particularly during periods of catchment wetting and drying [Woods and Rowe, 1996; Beven and Freer, 2001; Troch et al., 2002]. Indeed, many of the discrepancies we found between estimates of inflow contributions made from upslope contributing area and tracers, particularly during the rain event, can be explained by not accounting for dynamic changes in upslope contributing area. During base flow, the catchment may be in an approximate steady state condition; therefore, the predictions of inflow contributions tended to agree with tracer estimates. During the rain event, actual inflow contributions may depend more on local hillslope geometry, not the size of upslope contributing area [Troch et al., 2002]. This would also result in diffuse inflows being more influential at this time, consistent with some of our results.
Limitations of the tracer approaches confounded our ability to fully evaluate the topography-based groundwater inflow predictions. The utility of temperature as a tracer diminished both in the downstream direction and as the response to the rain event progressed, due to decreasing temperature differences between stream, subsurface inflow, lake, and rain event water (Figure 7). A few days following the rain event, as streamflow receded, the general step pattern in longitudinal temperature seen during base flow re-emerged (data not shown). This was presumably due to a return to conditions where hillslope inflows were more important than lake outlet contributions, discrete inflows contributed more than diffuse inflows, and larger differences existed between stream and inflow water temperatures.
It is likely that subsurface isotope and temperature observations made at seepage zones and using the groundwater well sensors, respectively, do not adequately represent the spatiotemporal signature of lateral inflow water. For example, not all five major inflow zones predicted by the upslope contributing area algorithms were sampled for isotope and temperature signatures. In addition, near-stream soil water temperatures and
O are known to vary with soil depth [Laudon et al., 2004; Peralta-Tapia et al., 2015; Leach and Moore, 2015]. Considering that lateral flow may have been generated at depths shallower than the depths of groundwater temperature sensors, the lack of soil temperature observations from near the surface could explain the disagreement in theoretical and observed longitudinal stream temperature profiles (Figure 6b). A reasonable explanation for the difference in longitudinal temperature patterns in Figure 6b from about 350 m onward is that inflow temperatures may have been closer to 10.4°C (the approximate temperature of the 350–1445 m portion of the stream during this period) than the inflow estimate of 7.6°C calculated from the well observations. However, applying an inflow temperature of 10.4°C to the theoretical calculations results in an overprediction of stream temperature along the stream reach, particularly for the first 500 m (analysis not shown). This highlights that spatial variability of inflow temperatures for this site are likely greater than the observed 0.5–1.0°C daily spatial range typically observed from our limited groundwater temperature network. Additional measurement locations, as well as sampling from different depths in the riparian soil, would provide a more robust tracer description of these lateral inflows and reduce uncertainty in the inflow estimates.
The lake-stream network provided the opportunity to examine the importance of lateral inflows relative to headwater lake contributions to downstream discharge. The shifting contributions of lake and hillslope water sources highlight an important role of headwater lakes on regulating and influencing downstream discharge dynamics, which may have a critical influence on stream ecosystems and how water quality observations made downstream of lakes are interpreted [Vadeboncoeur, 1994; Arp et al., 2006; Larson et al., 2007; Jones, 2010; Kalinin et al., 2016; Pépino et al., 2016]. In contrast, streams draining headwater catchments without a lake would be entirely sourced from hillslope contributions and this may suggest that headwater catchments without lakes are more sensitive to riparian and upland disturbances [Mellina et al., 2002]. This speculation requires further study and would depend on disturbance type and the water quality impacts concerned.
5.3 Topography-Based Lateral Inflow Predictions for Forest Management
There has been increased interest in using high-resolution DEMs coupled with upslope contributing area or hydrologic modeling to inform forest management plans in order to minimize impacts on aquatic ecosystems. For example, maps based on upslope contributing area have been proposed for delineating wet areas connected to surface waters and designing riparian buffer zones and other harvesting strategies in order to minimize ecological and biogeochemical impacts of forest harvesting on surface water [Gorsevski et al., 2008; White et al., 2012; Kuglerová et al., 2014b; Ågren et al., 2015; Laudon et al., 2016]. In low-relief topography typical of many boreal forest regions, visually identifying source areas of lateral inflow in the field can be challenging during both summer (supporting information Figure S2) and winter when snow obscures any potential visual indicators, such as vegetation type [Kuglerová et al., 2016]. In addition, soil maps are often not of sufficient spatial resolution to be useful for many forest operation plans aimed at minimizing impacts to surface waters [Bailey et al., 2014]. In contrast, upslope contributing area maps and similar products produced from high-resolution DEMs require minimal computation and can be done using numerous software products; therefore, facilitating their use by forest and water resources managers for designing best management strategies.
Our results suggest that for humid boreal forests upslope contributing area maps may be suitable for identifying locations of shallow groundwater inflows to streams. Having the ability to identify groundwater inflow zones from readily available topographic data can be useful for certain forest management applications, such as minimizing harvesting and driving on wet riparian soils, increased forest retention around groundwater source areas, locating forestry roads and stream crossings, and designing riparian buffer structure based on predicted soil wetness [Laudon et al., 2016; Tiwari et al., 2016]. Our study showed that locations of lateral inflows can be identified using upslope contributing area; however, the ability to predict the relative contribution of these inflows to streamflow was less successful. This is an important limitation to these hydromapping tools. Forestry management related to protecting surface water resources often requires binary decision making (e.g., to harvest an area or not; to drive heavy equipment through an area or not). Our findings highlight that it may be difficult to decide, based on upslope contributing area magnitude alone, whether a specific inflow zone is a critical influence on downstream water quantity and quality. At our study site, it appears that the inflows associated with the largest upslope contributing areas were indeed important contributors to streamflow as individual inflows were estimated to contribute up to 75% of streamflow immediately downstream (Figure 7). However, the importance of inflows associated with low and moderate upslope contributing areas was less clear. More work is needed to clarify relations between topography and the role of lateral inflows on stream water quality and to develop better management tools for predicting which source areas are critical contributors to stream ecosystems.
6 Conclusions
It has been debated whether topography is useful for predicting the location and magnitude of lateral groundwater inflows to streams, particularly for streams in low-relief landscapes such as the boreal region. We found that predicted locations of lateral groundwater inflow using upslope contributing area generally matched those locations identified from temperature and isotope tracers, particularly for base flow conditions. This result suggests that for humid boreal regions with basal till soils, surface topography is useful for identifying parts of the landscape that tend to exhibit hydrologic connections to surface waters. Although locating groundwater inflows using upslope contributing area was generally successful, predicting the magnitude of these inflows was less successful. From a forest management perspective, this raises potential issues around classifying which inflow source areas should be actively protected from disturbance to ensure impacts to stream systems are minimized. At the catchment scale, lateral groundwater inflows comprised most of the stream water leaving the catchment during base flow and peak flow conditions; however, water from a headwater lake was the dominant stream water source during the hydrograph recession. Therefore, headwater lakes may have a strong influence on downstream water quality and this influence likely varies substantially through time. Overall, this study highlights the advantages and limitations of topography-based approaches for predicting lateral groundwater inflows and identifying the spatiotemporal importance of lateral inflow contributions to streamflow.
Acknowledgments
Financial support for this project was provided by Kempe Foundation, FORMAS, Future Forests, Vetenskapsrådet, SITES, VINNOVA, Knut och Alice Wallenberg Stiftelse, Skogssällskapet, and Stiftelsen Oscar och Lili Lamms Minne. We are thankful for the support and help from Peder Blomkvist, Ida Taberman, Johannes Tiwari, Ola Olofsson, Stefan Krause, and the team at HydroResearch. We appreciate the valuable feedback and detailed reviews from four anonymous reviewers and Christa Kelleher that helped improve the manuscript. We are particularly thankful for important contributions and thoughts from the Associate Editor Charles Luce. Data used in this study are available from the authors upon request.
