Quantification of localized groundwater inflow into streams using ground-based infrared thermography



[1] Due to temperature differences of groundwater and streamwater, localized groundwater inflows into small streams can directly be detected with ground-based thermographic systems in summer or winter. Infrared radiation temperatures of surface water were used to determine mixing length and to calculate the relative fraction of groundwater inflow to downstream discharge. These fractions were comparable to groundwater inflow fractions derived from electrical conductivity, kinetic water temperatures and discharge measurements. This approach advances the immediate detection and quantification of localized groundwater inflow for hydrology, geology and ecology.

1. Introduction

[2] Significant groundwater (GW) contributions to the stream reach can be observed during the dry season [Krause et al., 2007]. In addition to diffuse GW contributions from an aquifer, localized GW contributions from preferred flow pathways through highly permeable sediments [Winter et al., 1998] or fractures in bedrock [Fetter, 2001] are also common. The localization and quantification of GW inflows to the stream will yield important insights into aquifer type, flow pathways, the composition of baseflow regarding the spatial distribution of catchment storage, water quality issues and fish and microinvertebrates in-stream habitat [Hayashi and Rosenberry, 2002].

[3] Few established methods allow the identification of localized GW inflows. These include intensive measurement campaigns of runoff or chemical water composition along stream reaches with high spatial resolution [Krause et al., 2007; Lowry et al., 2007; Schmidt et al., 2006, 2007], application of GW tracers such as 222Rn [Kluge et al., 2007], or emergent aquatic vegetation [Rosenberry et al., 2000]. A well-known tracer for GW-stream water (SW) interaction is temperature. Anderson [2005] described applications of heat as a tracer used since the 1960s to analyze GW recharge and discharge and to constrain GW-SW models. All temperature-based methods take advantage of the fact that in seasonal climates GW has a smaller intra-annual temperature amplitude than surface water. Thus a significant temperature difference between SW and GW can be expected in summer and winter. Several studies detected GW inflows along stream reaches using continuous distributed temperature measurements [Lowry et al., 2007; Selker et al., 2006a, 2006b; Westhoff et al., 2007]. Conant [2004] and Schmidt et al. [2006, 2007] used spatially distributed snap-shots of stream and GW temperatures to detect and estimate GW contributions. Constantz [2008] provided an overview of current temperature based applications for the investigation of SW-GW interactions.

[4] Infrared thermographic systems sensing the spectrum between 8 and 14 μm can detect the intensity of the thermal infrared spectral longwave emission of water bodies. Infrared temperatures (IRT) of surface water have thus been used to detect GW inflows at different spatial scales: remotely sensed water surface IRT were observed by satellite based scanner systems [Becker, 2006; Handcock et al., 2006; Kay et al., 2005], airborne [Banks et al., 1996; Johnson et al., 2008; Loheide and Gorelick, 2006; Rayne and Henderson, 2004; Sams and Veloski, 2003; Torgersen et al., 2001] and ground-based thermographic systems [Cardenas et al., 2008; Chen et al., 2009; Deitchman and Loheide, 2009]. Satellite based systems are only capable to detect GW inflow in large rivers [Handcock et al., 2006]. Airborne observations were used to detect thermal gradients induced by inflowing GW at shorelines of lakes [Banks et al., 1996; Portnoy et al., 1998] and ocean [Johnson et al., 2008]. IRT heterogeneities were recorded using airborne systems with spatial resolutions between 0.2 m and 0.4 m [Loheide and Gorelick, 2006; Rayne and Henderson, 2004; Torgersen et al., 2001]. Emissivity corrected IRT agreed with in-situ water temperatures within 0.5 K between 5 and 27 °C stream temperature [Torgersen et al., 2001]. However, the accuracy of IRT observations is influenced by longwave radiation reflectance, thermal boundary layer effects, and thermal stratification of water [Kay et al., 2005; Rayne and Henderson, 2004; Torgersen et al., 2001].

[5] Unlike airborne and satellite systems, ground-based or hand-held IRT systems have not been used widely to quantify and detect GW contributions to the stream. Cardenas et al. [2008] used IRT to map SW surface temperature differences during baseflow and flood conditions. Chen et al. [2009] used high-resolution thermal images to verify vertical seepage flux measurements by comparison with water temperatures. Deitchman and Loheide [2009] reported for the first time the use of IRT for fine scale detection of water table positions along a stream bank seepage face. Comparing measured saturated conductivities of stream bank sediments with surface infrared temperatures, they found an empirical relation to quantify GW discharge at a spatial resolution of centimetres.

[6] Most applications of remotely sensed IRT successfully detected GW inflows in lakes or streams. Simulations of the stream energy budget [Loheide and Gorelick, 2006; Rayne and Henderson, 2004], applications of the one-dimensional steady-state heat-diffusion–advection equation [Schmidt et al., 2006, 2007] or numerical heat transport models [Schmidt et al., 2007] allowed to quantify GW inflows. However, remotely applicable methods to directly quantify GW inflow in streams are rare [e.g., Deitchman and Loheide, 2009], especially for localized GW contributions. The aim of our study is to localize and quantify GW inflow in small streams, using non-invasive and remotely applicable IRT snap-shot sampling campaigns along headwater streams during baseflow periods in summer and winter with flows less than 10 l/s.

2. Methods

[7] IRT images cannot only be used to detect locations of GW inflows in creeks, but also to determine the stream length for complete mixing of SW and GW. This additional information is fundamental to our approach of quantifying GW inflow using IRT. The three end-members necessary to quantify GW inflow can be directly located and measured with IRT: stream temperature upstream of the GW inflow, GW temperature at the localized inflow and stream temperature below complete mixing downstream of the GW inflow. Complete mixing length depends on stream geomorphology, flow velocity, discharge and physical attributes as viscosity, density, etc. Therefore a method to detect the mixing length is a pre-requisite to significantly reduce the error in quantifying GW inflow (see auxiliary material).

[8] GW inflow can only be quantified if several conditions are met: (1) the IRT change between upstream and downstream of the GW inflow must be significantly larger than the accuracy of the IRT system, (2) the temperature of the water column is constant with depth, (3) GW inflow IRT can be observed in the stream or at the streambed and (4) GW inflow and stream flow are constant during the time of observation, thus cooling or warming influences of air temperature in the local area of observation are below quantification limits. Under these conditions, the relative contribution of GW inflow to downstream discharge can be determined from the IRT of inflowing GW, upstream and downstream SW by using the mass balance of flux-weighted concentrations or temperatures:

equation image

and, following Selker et al. [2006b],

equation image

with discharge Q and T for water temperature or an equivalent solute concentration for the end-members groundwater GW, downstream Down and upstream Up of the inflow location.

[9] Uncorrected IRT can only be used, if the emissivity of all end-members is equivalent. Here, emissivity is considered as an empirical factor (apparent emissivity), which determines the difference between the kinetic temperature of an object and emitted thermal radiation. Robinson and Davies [1972] found the emissivity of freshwater to range from 0.97 to 0.98 between 7.7 μm and 14.3 μm. The relation between IRT of the end-members and kinetic water temperatures of our observations is linear with a R2 of 0.98. The unexplained variance is due to uncertainty in the emissivity of the target and not accounting for other factors such as air humidity and observation distance. Since only relative IRT differences of water surfaces are used to calculate the end-member fractions, a correction for emissivity, humidity and observation distance is not necessary. Much more important is to exclude reflected radiation temperatures of surrounding obstacles or direct solar radiation by radially moving the IRT system around the area of interest. The reflectance of obstacles or direct solar radiation will change according to the observation angle. Since a body radiates in all directions, “real” IRT of the observed object stays nearly constant even if the angle and the point of view is changed [Ishiyama et al., 1996]. To improve our results, close-up topview pictures of all end-members were taken after the detection.

[10] Additional independent methods like electrical conductivity (EC) and in-situ temperature measurements (T) to calculate GW inflow as fractions of downstream discharge were compared to the results of the IRT based end-member calculations. Moreover discharge was measured with the salt-dilution method upstream and downstream of the GW inflow. Up- and downstream Q measurements were also used to determine the absolute amount of inflowing GW. For the four methods (IRT, T, EC, Q) the overall error (relative and absolute) based on the Gaussian error propagation of the errors or uncertainties of the end-member measurements was determined.

[11] The hand-held thermographic system VARIOCam hr INFRATEC, which is based on a focal plane array (FPA) uncooled microbolometer with a spatial resolution of 640 × 480 pixels, was used to observe IRT of surfaces with a relative accuracy of 0.08 K. Observed infrared spectra of the thermographic system ranged between 7.5 and 13 μm. Observation height was approximately 1.2 m. Measurement uncertainty of IRT is estimated using the standard deviation of IRT in the observed areas since the high number of selected pixel values (more than 200 pixels) reduces the measurement uncertainty. Discharge measurements were conducted with a Sommer MST2 system consisting of two conductivity probes and a data-logger for discharge calculation resulting in a total uncertainty between 5 and 10% [Moore, 2005]. T and EC were measured with a WTW LF92 with a relative accuracy for EC of 0.5% and an absolute accuracy for T of 0.2°C.

[12] Localized GW inflows were detected in summer and winter during several field campaigns during dry weather conditions in two different streams, one in a mountainous headwater catchment underlain by magmatic bedrock and one in an agricultural headwater catchment covered with loess in Southwest Germany. To avoid thermal disturbances by longwave emission of the surrounding, the campaigns were conducted in the morning. Discharges in the stream during the baseflow periods varied between 1 l/s and 5 l/s.

3. Results and Discussion

[13] Several GW inflow locations were detected in the mountainous headwater stream. Some locations at stream banks were discrete areas with significantly different stream bank and SW temperatures compared to the surrounding stream bank and SW temperatures, but without significant changes in downstream SW temperatures (example in Figure 1a). One significant localized GW inflow into the stream (Figure 1b) and one in a small detention pond (Figure 1c) were detected. Old hidden tile drains to the stream were located in the agricultural catchment (Figure 1d). These GW inflows caused detectable changes in SW temperatures downstream of the GW inflows. The mixing lengths for complete mixing of GW and SW varied between 3 and 10 m, which can be expressed in terms of channel widths between 7.5 and 25, respectively. These results are consistent with those of Day [1977].

Figure 1.

(a) Small GW inflow without a detectable change in downstream SW temperature; (b) GW inflow during the low-flow period in summer. Sticks and leaves are disturbing the homogeneous temperature fields (black circle); (c) GW inflow into a small detention pond in winter; (d) GW inflow of an old covered tile-drain in the agricultural catchment. Hatched areas are land surfaces, blue open arrows show stream flow directions and white arrows show locations of localized groundwater entries.

[14] At all sites, T, EC and IRT were determined for the three end-members. In-situ upstream stream temperatures ranged from 15.5 to 20.8°C during summer and from 2.5 to 10.8°C during winter. Groundwater temperatures ranged from 13.9 to 16.8°C in summer and 5.3 to 11.1°C in winter. Seasonal variations of kinetic water temperatures of SW and GW were comparable in both catchments. Changes in stream temperatures caused by mixing of GW and SW ranged from 0.3 to 3.2°C in summer and winter. EC did not vary strongly between different seasons, but varied among sites. EC of SW ranged from 135 to 931 μS/cm and for GW from 93 to 1070 μS/cm. GW inflow induced changes of EC in the stream between 7 and 206 μS/cm. Measured discharges in the mountainous headwater catchment varied between 1.7 l/s in summer and 3.2 l/s in winter. The localized GW contributions during the different field surveys varied between 0.3 and 0.6 l/s. In the agricultural headwater catchment discharge was between 1.8 l/s in summer and 4.2 l/s in winter, localized GW contribution by the drain pipes between 1.1 and 2.2 l/s.

[15] Applying equation (2), we calculated the fractions of GW inflow to downstream discharge using the four methods. Figure 2a shows that GW inflow fractions for each observation vary among all methods. Standard deviations among calculated GW inflow fractions of all four methods ranged between 0.047 and 0.111 per site and observation.

Figure 2.

(a) GW inflow fraction of different sites and seasons (W = winter and S = summer), based on discharge, EC, T and IRT observations. The black bars show absolute errors for all calculated fractions; (b) relative error of groundwater inflow fractions calculated by T, EC and IRT compared to estimated GW inflow fractions. The two lines show the uncertainty range of GW inflow fractions calculated from discharge measurements with a relative error of 5 and 10%.

[16] In Figure 2b relative errors of GW inflow fractions of EC, T and IRT are compared to the function of relative errors of estimated GW inflow fractions assuming Q measurement errors between 5% and 10%. EC shows the smallest errors, which vary independently from the GW inflow fraction between 3% and 20%. Errors of IRT vary from 6% for large GW inflow fractions to 55% for smaller ones. The main reason for the larger IRT errors is the inadequate determination of homogeneous temperature fields, which can be influenced by the presence of sticks, stones and leaves in the streambed (see for example Figure 1b). Errors become smaller when smaller IRT fields are considered. With an estimated error of 0.1 K (we used a sensor with an error of 0.2 K) for T, the relative errors of GW inflow fractions are similar to those of IRT. For decreasing GW inflow fractions, the relative error of discharge-based results increases. If GW inflow becomes too small, the determination of inflow fractions is limited either by the measurement error of discharge or the ability to determine a downstream end-member of IRT, EC or T. For smaller fractions, end-member based methods can better quantify inflow as discharge measurements alone.

[17] The accuracy of IRT based surface water temperature estimations is often influenced by longwave radiation reflectance, thermal boundary layer effects and thermal stratification of water [Rayne and Henderson, 2004; Torgersen et al., 2001]. In this study, these effects can be neglected since longwave radiation reflectance and thermal boundary layer effects were avoided by radially moving around the observed location until reflectance and longwave emission effects were eliminated. We also considered the results of Ishiyama et al. [1996] during our measurement campaigns, who found that good IRT based surface temperature observations are possible up to an observation angle of 70°. Thermal stratification did not play a role at our sites though stream depths were within centimetres and well mixed by turbulent flow.

[18] The results highlight several implications for the investigation of GW-SW interactions using heat as a tracer. The quantification of GW inflow with observed SW and GW temperatures is frequently done using relative complex heat transport or heat budget models [e.g., Loheide and Gorelick, 2006; Schmidt et al., 2006, 2007; Westhoff et al., 2007]. The presented approach to directly detect end-members of SW and in particular GW to calculate GW inflow fractions provides an innovative possibility to determine GW discharge from a certain area or sub-catchment directly in field. It offers the possibility to collect data instantly and remotely during the detection of GW sources. Becker [2006] recommended the use of IRT to investigate GW-SW interactions at different scales. In addition to the proposed use of IRT for the estimation of small scale diffuse GW inflow [Deitchman and Loheide, 2009] and larger scale GW-SW processes [e.g., Banks et al., 1996; Handcock et al., 2006; Loheide and Gorelick, 2006; Torgersen et al., 2001] the proposed method may help to understand GW contributions to surface flow at the scale of localized GW-contributions in highly variable material. Even if the stated boundary conditions limit the applicability to small streams, this method might enhance the understanding of GW-SW interaction of first order streams which accounts, e.g., in the U.S. for nearly 48% of the total stream network length [Poff et al., 2006]. Especially during low flow periods, this method may improve our ability of sustainable water management in water scarce regions or the ecological management of in-stream habitats. Additionally, it can support management practices regarding the export of fertilizers or pesticides in agricultural catchments, where spatial extent and contribution of tile drains is often unknown.

4. Conclusions

[19] This study describes ground-based IRT as a non-invasive and remotely applicable method to detect and quantify localized groundwater inflow in small streams. The method offers the possibility to determine the exact location of groundwater inflow and also the reach length of complete mixing of surface and groundwater, which is a requirement for the quantification of end-members independent of the method chosen for quantification. During summer and winter the method was capable to detect and quantify localized GW inflow sources during baseflow periods, with contributions to stream discharge between 15% and 75%. Calculated groundwater inflow fractions of discharge measurements, electrical conductivity and kinetic water temperatures were comparable to those derived from water surface infrared radiation temperatures alone. Ground-based IRT is a valuable and easily applicable tool in stream ecology and process hydrology for the detection and quantification of localized groundwater inflow in small streams.