Impact of seasonal variability and monitoring mode on the adequacy of fiber-optic distributed temperature sensing at aquifer-river interfaces

Authors

  • Stefan Krause,

    Corresponding author
    1. School of Geography, Earth and Environmental Sciences, University of Birmingham, Birmingham, UK
    • Corresponding author: S. Krause, School of Geography, Earth and Environmental Sciences, University of Birmingham, Edgbaston, Birmingham B15 2TT, UK. (s.krause@bham.ac.uk)

    Search for more papers by this author
  • Theresa Blume

    1. Section 5.4 Hydrology, Helmholtz Centre Potsdam, GFZ German Research Centre for Geosciences, Potsdam, Germany
    Search for more papers by this author

Abstract

[1] Fiber-optic distributed temperature sensing (FO-DTS) has been frequently applied for analyzing thermal patterns, including the identification of groundwater-surface water exchange fluxes across aquifer-river interfaces. However, the impacts of (a) seasonal variability in signal strength (given by the difference between groundwater and surface water temperatures) and (b) monitoring modes on the accuracy of FO-DTS surveys have not yet been determined. This study uses a well-investigated field site as model system for quantifying the accuracy and uncertainty of FO-DTS surveys in dependency of seasonal signal variation and monitoring mode. The analysis of the relationship between seasonal variability in signal strength and diurnal oscillations in end-member temperatures at the study site revealed that winter conditions, with substantially lower diurnal temperature oscillations, provide the highest temporal stability in signal strength. The choice of monitoring mode proved to have significant impact on the accuracy of FO-DTS surveys. The proposed two-way single-ended averaging of FO-DTS surveys had significant advantages compared to single-ended or double-ended surveys, with a higher accuracy in signal detection, in particular for small-scale temperature variations. Since FO-DTS surveys in two-way single-ended averaging mode were better suited for detecting the full complexity of spatial temperature patterns for the investigated aquifer-river interface, we recommend its wider application in similarly complex systems with small-scale thermal patterns.

1. Introduction

1.1. Groundwater-Surface Water Exchange Fluxes at the Aquifer-River Interface

[2] The mixing of groundwater and surface water can have substantial impact on stream and streambed thermal patterns [Malcolm et al., 2002; Hannah et al., 2004, 2009; Krause et al., 2011b], the availability of dissolved oxygen, nutrient cycling and carbon respiration [Pinay et al., 2009; Mulholland et al., 2000, 2008; Krause et al., 2008a, 2008b; Pinay et al., 2009; Zarnetske et al., 2011a, 2011b], transport and transformation of contaminants [Conant et al., 2004; Ellis and Rivett, 2007; Rivett et al., 2011], as well as the ecohydrological functioning of the riverine environment [Brunke and Gonser, 1997; Dole-Olivier et al., 1997; Boulton et al., 1998; Malcolm et al., 2002; Stubbington et al., 2009; Robertson and Wood, 2010; Krause et al., 2011b, 2011c]. A major challenge for quantifying the impact of physical streambed controls on the biogeochemical and ecohydrological functioning is the difficulty to identify temporal dynamics and detailed spatial patterns of groundwater-surface water exchange at stream reach and larger scales [White, 1993; Krause et al., 2011a]. Recent studies have successfully investigated aquifer-river exchange fluxes by using temperature differences between groundwater and surface water as a tracer [Constantz et al., 2003; Constantz, 2008; Conant, 2004; Anderson, 2005; Keery et al., 2007; Schmidt et al., 2007; Hatch et al., 2010].

1.2. Fiber-Optic Distributed Temperature Sensing

[3] In particular, advances in distributed sensor technologies such as fiber-optic distributed temperature sensing (FO-DTS) have supported the monitoring of thermal patterns at aquifer-river interfaces at increasing spatial and temporal resolution [Selker et al., 2006a, 2006b; Tyler et al., 2009]. FO-DTS technology, which is based on the monitoring of thermal patterns along a fiber-optic cable of up to several kilometer length, allows for continuous measurements of temperatures at a spatial resolution between currently 1–2 m and measurement precision of 0.05–0.1°C for sampling intervals of 30 s [Selker et al., 2006a, 2006b; Hausner et al., 2011; Van de Giesen et al., 2012].

[4] In addition to successful studies of spatial patterns and temporal dynamics of exchange fluxes between groundwater and surface water [Lowry et al., 2007; Westhoff et al., 2007; Vogt et al., 2010; Slater et al., 2010; Briggs et al., 2012; Krause et al., 2012a], FO-DTS has experienced a rapid increase in interdisciplinary applications for thermal monitoring in hydrology and hydrogeology, glaciology, soil sciences, civil engineering, and meteorology [Selker et al., 2006a, 2006b; Tyler et al., 2008, 2009; Henderson et al., 2009; Steele-Dunne et al., 2010; Slater et al., 2010; Keller et al., 2011; Suárez et al., 2011; Krause et al., 2012a, 2012b]. Different experimental designs and strategies have been deployed in order to optimize the FO-DTS setup to accommodate the specific monitoring requirements at the aquifer-river interface, including the installation of coiled fiber-optic cables for a 5–10 times increase in spatial sampling resolution along a vertical profile [Vogt et al., 2010; Briggs et al., 2012].

[5] The tracing of groundwater-surface water exchange represents a complex challenge for FO-DTS surveys where uncertainties related to sampling design or monitoring mode are still largely unknown. Spatial patterns in river or streambed temperatures caused by aquifer-river exchange can involve either gradual alteration of temperatures or rather discrete temperature changes [Selker et al., 2006a, 2006b; Westhoff et al., 2007; Lowry et al., 2007; Krause et al., 2012a]. Furthermore, groundwater-surface water mixing can cause temperature changes that propagate downstream [Selker et al., 2006a, 2006b; Westhoff et al., 2007] or localized temperature anomalies of limited spatial extent [Lowry et al., 2007; Krause et al., 2012a]. As shown by Rose et al. [2013], in particular, the precise detection of smaller signals that do not exceed the respective FO-DTS sampling resolution by at least three to four times can be critically limited, causing concern for the accurate interpretation of FO-DTS survey results under such conditions.

[6] Alongside the increasing numbers of interdisciplinary applications of FO-DTS and problem-specific adaptations of the monitoring design, several studies have started to investigate the implications of different calibration techniques and experimental design for the quality and accuracy of FO-DTS monitored temperature patterns [Tyler et al., 2009; Hausner et al., 2011; Van de Giesen et al., 2012]. The recent works by Hausner et al. [2011] and Van de Giesen et al. [2012] provided benchmark analyses of the specific advantages and restrictions as well as calibration demands of single-ended and double-ended FO-DTS monitoring modes. While the experimental setup of single-ended or double-ended monitoring modes has been shown to have an impact on the accuracy of signal detection, with a degradation of precision toward the cable ends or at the center of a fiber-optic cable [Tyler et al., 2009; Hausner et al., 2011; Van de Giesen et al., 2012], the monitoring mode does not affect the sampling resolution of temperature measurements. However, the quantitative impact of FO-DTS monitoring modes on the accuracy of detected temperatures and temperature patterns has yet to be identified. As the majority of previous FO-DTS publications did not report on the applied monitoring modes, it still has to be established to what degree the accuracy of earlier reported FO-DTS survey results has been affected by the chosen monitoring mode.

[7] In order to adopt a sampling design and FO-DTS monitoring mode that are best suited to accurately detect site-specific signal patterns, it is required to identify the FO-DTS sampling mode dependent limitations and uncertainties in signal detection. While many recent studies have adopted double-ended FO-DTS monitoring modes in order to increase the robustness of signal detection, it is unknown how monitoring mode decisions affect the detection of signals of smaller spatial extent or the accuracy of predictions of discrete signal locations and signal changes.

[8] Moreover, the strength of the investigated temperature signal has been shown to have substantial impact on the detection accuracy of thermal patterns in a FO-DTS survey [Rose et al., 2013]. As FO-DTS surveys of aquifer-river exchange fluxes use the temperature difference between groundwater and surface water as a tracer signal, seasonal variability of the signal strength has to be taken into account. With the seasonal differences in groundwater and surface water thermal regimes, also the signal strength (as given by the temperature difference between groundwater and surface water) often varies on a seasonal basis [Lapham, 1989; Bartolino and Niswonger, 1999; Stonestrom and Constantz, 2003], presenting a challenge for accurate FO-DTS surveys. FO-DTS surveys at aquifer-river interfaces have been carried out during summer and winter conditions, utilizing the seasonally warmer as well as colder groundwater than surface water temperatures. The impacts of seasonal differences in detected signal strength as well as short-term (i.e., diurnal) signal variability on the accuracy of FO-DTS surveys have yet to be identified. In particular, for dynamic systems with small-scale variability and discrete signal patterns as well as for the quantitative interpretation of FO-DTS survey results, detailed quantitative understanding of the implications of monitoring modes and seasonal signal differences on the accuracy and uncertainty of FO-DTS surveys is paramount and a prerequisite for adapting the monitoring design to best accommodate site-specific survey demands.

1.3. Aims and Objectives

[9] This study quantifies the impact of (i) the seasonal variability of signal strength and (ii) the FO-DTS sampling design and monitoring mode on the accuracy and limitation of FO-DTS surveys at aquifer-river interfaces. Its specific objectives are therefore to (1) analyze the seasonally variable signal strength represented by the temperature differences between groundwater and surface water and its diurnal variation; (2) quantify the impact of seasonal variation in signal strength on the detection accuracy of groundwater-surface exchange flow patterns by FO-DTS and identify seasonally variable uncertainties and limitations in signal detection; (3) compare the impact of standard single-ended and double-ended monitoring modes on the accuracy of the detection of size, location, and spatial extent of exchange flow patterns at the aquifer-river interface; (4) and present an alternative approach that combines the advantages of single-ended and double-ended monitoring modes for optimizing the detection accuracy of FO-DTS surveys.

[10] This study uses the investigation of exchange fluxes between aquifer and river in a well-investigated field site as a model system for quantifying the accuracy and uncertainty of FO-DTS application in dependency of seasonal signal variation and experimental design. The findings of this study, however, are transferable to other situations and also to FO-DTS applications in systems other than aquifer-river interfaces.

2. Material and Methods

2.1. Experimental Field Site

[11] This study is focusing on a well-investigated, approximately 300 m long, stream section of the River Tern (United Kingdom; 52°86′N, 2°53′W; Figure 1). The local geology at the field site is characterized by Permo-Triassic Sherwood Sandstone, which represents one of the United Kingdom's major aquifers and is overlain by drift deposits of variable depth. The 5 to 8 m wide river channel is limited by steep, on average 2 m high river banks, and comprises a succession of pool-riffle-pool sequences, representing the highest topographic energy of the investigated stream reach (Figure 1c). Local landuse is characterized by pasture. Long-term (1971–2010) mean annual precipitation at the field site is 583 mm. Mean daily air temperature ranges from 3.7°C (January) to 15.8°C (July), with long-term (1957–2007) mean annual temperatures of 9.3°C (Hannah et al., 2009). Mean river discharge at the Environment Agency operated Ternhill gauging station (52°87′92″N, 2°55′12″W, basin area 92 km2, elevation 62 m above sea level) is 0.9 m3 s−1 with a 95% exceedance (Q95) of 0.4 m3 s−1 and a 10% exceedance (Q10) of 1.39 m3 s−1 (data period 1971–2010, UK National River Flow Archive, http://www.ceh.ac.uk/data/nrfa/data/station.html?54044).

Figure 1.

(a) Location of the River Tern field site in the United Kingdom, (b) experimental infrastructure including the setup of the fiber-optic cable installation at the field site (flow direction from north to south), and (c) longitudinal streambed profile of the investigated stream section with location of the fiber-optic cable entry and exit points.

[12] The research area has been subject to intensive previous investigations which provide experimental infrastructure and background data for this study (Table 1). The U.K. Natural Environment Research Council selected the River Tern as a representative lowland catchment within its Lowland Catchment Research Programme [Wheater and Peach, 2004]. Previous application of heat tracers including FO-DTS identified groundwater in the research area to be generally upwelling, with upwelling patterns predominantly controlled by physical streambed conditions such as the geometry of low conductivity streambed structures [Krause et al., 2011b, 2012a]. The discrete spatial patterns of groundwater upwelling have been found to result in distinct thermal anomalies in streambed temperatures during the 2009 summer baseflow conditions [Krause et al., 2012a].

Table 1. Temporal Resolution and Accuracy of the Monitored Environmental Parameters and FO-DTS Experimental Setup
Environmental VariableObservation IntervalInstrumentationAccuracy
Temperature: SW5 minSolinst LT M5/F15 diver, combined level and, temperature logger±0.05°C
Temperature: GW15 min±0.05°C
Temperature: air1 hKeele University, meteorological station±0.05°C
Precipitation1 hKeele University, meteorological station (18 km distance)±0.2 mm
Discharge (Q)1 hEA gauging station Tern Hill±5%
DTS streambed temperature surveys10 February, 16 March, 11 August 2010FO-DTS (Sensornet Halo)±0.05°C

2.2. Fiber-Optic Distributed Temperature Sensing

[13] FO-DTS makes use of the temperature-dependent backscatter properties of a laser signal that propagates through a fiber-optic cable [Selker et al., 2006a, 2006b; Tyler et al., 2009]. Due to the elastic collisions between photons and the molecules of the fiber core, the majority of backscatter is returned at the original frequency of the laser pulse. Inelastic photon collisions, and the thermal excitation of glass molecules (Raman scatter), cause a shift in the return energy level below (Stokes band) or above (anti-Stokes band) the Rayleigh scatter band. Stokes and anti-Stokes are exponentially dependent on the temperature with the anti-Stokes being more affected. Temperatures for given intervals along the fiber-optic cable are identified by comparing the Stokes/anti-Stokes amplitude ratio in combination with travel-time information of the propagating laser pulse [Selker et al., 2006a, 2006b]. The spatial resolution of FO-DTS monitoring is limited by the instrument's detection capabilities of the laser backscatter and is assumed to be at least two times the sampling resolution [Tyler et al., 2009; Van de Giesen et al., 2012] or even higher [Rose et al., 2013]. The precision of fiber-optic temperature measurements is a function of the total number of photons detected and depends on the integration time, with most systems working at a precision of 0.05°C–0.1°C for measurement intervals of less than 1 min [Selker et al., 2006a, 2006b; Sensornet, 2009; Hausner et al., 2011; Van de Giesen et al., 2012].

[14] FO-DTS monitoring can either be carried out in single-ended or double-ended mode (Figure 2). In single-ended mode the application of the laser pulse and detection of its backscatter are unidirectional and occur from one end of the fiber, while in double-ended mode the direction of laser application and detection are alternating between both fiber ends (Figure 2). Both monitoring modes encompass specific advantages and restrictions as well as calibration demands which are likely to have different impacts on the accuracy of signal detection, with potential smoothing and averaging effects at the cable ends or the center of a fiber-optic cable [Tyler et al., 2009; Hausner et al., 2011; Van de Giesen et al., 2012]. The nature of monitoring modes (single-ended, double-ended) defines the way the FO-DTS system can be best calibrated, and where reference baths are best placed along the fiber-optic cable [Hausner et al., 2011].

Figure 2.

Measurement setup for FO-DTS in different monitoring modes and respective calibration strategies: (a) single-ended monitoring with unidirectional measurements and associated correction of signal drift and offset, (b) double-ended monitoring with bidirectional measurements and associated averaging of bidirectional traces and correction of signal offsets, and (c) two-way single-ended averaging mode with alternating bidirectional single-ended tracing and associated correction of signal drift and offset before averaging the corrected temperature traces. Dashed red and blue lines represent uncorrected traces, and solid lines represent the corrected temperatures.

2.3. Experimental Design and Field Site Infrastructure

[15] At the field site, surface water (at one location), groundwater levels, and temperatures (at 10 observation wells; Figure 1) were monitored for the period from 12 June 2009 to 11 August 2010. The 3 m deep boreholes were installed in the shallow drift deposits overlying the Permo-Triassic Sandstone [Krause et al., 2011b, 2012a]. Meteorological conditions were recorded at a nearby meteorological station (52°59′55.86″N, 2°16′12.90″W). Four groundwater boreholes (GW1, GW2, GW3, and GW7) and the river-stage gauging station (Figure 1) were instrumented with pressure transducers, which in addition to water levels also monitored surface water and groundwater temperatures at 5 to 15 min intervals (Table 1).

[16] The current study used a Sensornet Halo FO-DTS (Elstree, United Kingdom), which analyzed the backscatter properties of a 10 ns light pulse for temperature observations along the fiber-optic cable. With a sampling resolution of 2 m [Sensornet, 2009], the Halo FO-DTS provides a monitoring resolution of >4m [Van de Giesen et al., 2012; Krause et al., 2012a]. In contrast to most previous studies [except Lowry et al., 2007], the fiber-optic cable at the field site was not installed at the surface of the streambed (where it would have recorded water column bottom temperatures) but was buried within the streambed in order to directly monitor temperatures at the aquifer-river interface [Krause et al., 2012a]. As FO-DTS measurements were conducted more than a year after cable installation, it is assumed that effects of possible sediment and flow path disturbance due to the installation procedure as, for instance, reported by Rosenberry et al. [2010] are marginal. The fiber-optic cable deployed in this study was a 1000 m long armored BruSteel dual-fiber cable (Brugg/CH) of which a section 462 m length was buried at ∼5 cm depth in the streambed. The remaining ∼500 m of fiber-optic cable was stored on a reel outside the river (Figure 2). In order to investigate the impact of variable GW-SW temperature differences, FO-DTS surveys were conducted for winter and summer conditions as well as for a transitional period in spring. Reference sections for the FO-DTS calibration were located at both ends of the fiber-optic cable (Figure 2) to carry out duplex single-ended or double-ended calibrations [Hausner et al., 2011; Van de Giesen et al., 2012] (Figure 2) using ice baths in spring and summer and constant temperature baths in winter, realized by surface water of several degree Celsius differences to groundwater temperatures. Control bath temperatures were independently monitored by thermocouples and remained stable (variation <0.1°C) throughout the monitoring intervals. Differential GPS surveys were performed for measuring the exact positioning of the fiber-optic cable as well as locations and elevations of the installed groundwater boreholes.

[17] FO-DTS monitoring in this study was carried out in single-ended and double-ended measurement mode as well as in a combination of both methods (Figure 2). In single-ended monitoring mode, signal drift and offset are corrected for measurements in one direction [Tyler et al., 2009]. In contrast, in double-ended monitoring mode, the drift correction is based on the integration of measurements of differential loss or gain in forward and reverse directions and applied as direction-independent attenuation correction alongside an offset correction [Smolen and van der Spek, 2003; Tyler et al., 2009; Van de Giessen et al., 2011] (Figure 2). In an attempt to combine the advantages of single-ended and double-ended monitoring modes, surveys in two-way single-ended averaging mode were carried out with alternating forward and backward measurements along the same fiber, similar to monitoring in double-ended mode (Figure 2). However, in contrast to standard double-ended measurements, the signal drift was corrected individually for both measurement directions (similar to measurements in single-ended mode), and only after that, the drift-corrected temperature traces were averaged during the data postprocessing (Figure 2). This procedure also enabled to correct longitudinal signal offsets along the fiber-optic cable. In this study a signal shift by one monitoring interval between forward and reverse traces was corrected for in two-way single-ended averaging mode. This shift is a known problem of bidirectional surveys and results from the fact that the length of the deployed fiber-optic cable does not usually represent an exact multiple of the FO-DTS sampling resolution. However, this shift cannot directly be accounted for in double-ended monitoring mode. FO-DTS surveys in all monitoring modes were based on 100 measurements of 30 s, covering a total monitoring period of 50 min per respective survey.

2.4. Data Analysis

[18] For the quantification of the strength of the FO-DTS monitored signal, temperature anomalies along the buried fiber-optic cable were analyzed, following the methodology proposed by Krause et al. [2012a]. The strength of temperature anomalies (AT) at a specific location was quantified by the difference of local temperature measurements from the spatial average temperatures during the respective sampling period (equation (1)).

display math(1)

where xi is the measurement location along the cable; all units are in degree Celsius and meters.

[19] As this study focused on a strongly gaining lowland river, temperature anomalies during summer, when groundwater temperatures were usually lower than surface water temperatures, were expected to be mainly negative, while during winter, when groundwater was usually warmer than surface water, warm anomalies were expected to result from the groundwater inputs.

[20] The variation in strength and patterns of temperature anomalies were investigated for their dependence from (i) seasonal differences between groundwater and surface water temperatures and (ii) alternative FO-DTS monitoring modes.

[21] The temporal variability of FO-DTS monitored temperature anomalies AT(xi) (°C) was quantified by the temporal standard deviation (STDEV; equation (2)).

display math(2)

3. Results

3.1. Hydrometeorological Conditions

[22] Hydrometeorological conditions during the observation period followed clear seasonal trends with higher precipitation and stream discharge (max >2.5 m3 s−1) during the winter months and lower baseflow (min <0.5 m3 s−1) in summer (Figure 3). Precipitation events with intensities of up to 14 mm/h in May and June 2010 caused only minor and short-lived increases in stream discharge that did not exceed 1.0 m3 s−1. Air temperatures varied by more than 30°C during the observation period with Tmax = 26.3°C, Tmin = −7.6°C, and an average of 10.0°C (Figure 3). Diurnal air temperature amplitudes varied substantially over the observation period with maximum day-night temperature differences of up to 15°C in summer and minimum differences of 4.1°C in winter (average diurnal temperature oscillation = 9.4°C).

Figure 3.

(a) Hydroclimatological conditions during the observation period (12 June 2009 to 11 August 2010) with air temperature and (b) precipitation (precip) and stream discharge (Q) observed at the River Tern field site with indication of the timing of FO-DTS surveys on 10 February, 16 March, and 11 August 2010.

3.2. Thermal Variability of GW-SW Difference (Signal Strength)

[23] Temporal dynamics of surface water temperatures generally followed patterns observed for air temperatures, with maxima in June and July and minimum temperatures in January (Figure 4). In contrast, groundwater temperatures peaked in September and October with minimum temperatures in March (Figure 4), indicating a time lag of several weeks in response to surface water and atmospheric conditions (Figures 3 and 4). Surface water temperature varied by more than 20°C with Tmin = 0.07°C and Tmax = 20.3°C, while the seasonal variability of groundwater temperature observed at four shallow groundwater boreholes in the drift deposits during the monitoring period was lower with a range of 6.3°C (Tmin = 7.1°C, Tmax = 13.4°C; Figure 4).

Figure 4.

Seasonal variability of surface water (one location) and groundwater (four locations) temperatures for the observation period (12 June 2009 to 11 August 2010) including two periods of data losses due to logger failure. Timing for FO-DTS surveys is shown in Figure 3.

[24] The contrasting thermal regimes of groundwater and surface water, with high seasonal temperature variation in surface water and low variability in groundwater, resulted in a distinct seasonality of temperature differences between groundwater and surface water (Figure 5a). While surface water temperatures can be up to 9.7°C higher than groundwater temperatures in summer, during winter the more stable groundwater temperatures can be up to 9.3°C higher than the temporally more dynamic surface water (Figure 5a). During spring and autumn, when surface water temperatures were rising or decreasing toward groundwater temperatures, differences between groundwater and surface water could be less than ±2°C for periods of several weeks. During these transition periods, the direction of groundwater-surface water temperature gradients changed on several occasions (Figure 5a).

Figure 5.

(a) Seasonal variability of ΔT differences between surface water and groundwater temperatures, (b) extent of diurnal temperature oscillation (daily maximum-daily minimum temperature) in surface water (SW) and mean groundwater (GW), and (c) seasonal variability in daily minimum (TSW − TGW)min and maximum (TSW − TGW)max temperature difference between mean groundwater and surface water during the 12 June 2009 to 11 August 2010 observation period.

[25] Diurnal temperature amplitudes in the surface water almost matched daily air temperature oscillations with averages of 1.4°C and maximum amplitudes of 4.4°C (Figure 5b). In contrast, groundwater temperatures exhibited no clear diurnal periodicity. Maximum daily changes in groundwater temperatures were below 0.3°C (Figure 5b). While diurnal changes in groundwater were constantly low, surface water amplitudes in summer were three to four times higher than in winter (Figure 5b). Such temporal variation in diurnal surface water oscillations had a recognizable impact on the strength of the signal analyzed by FO-DTS, which is constituted by the difference between groundwater and surface water temperatures. The daily variation in signal strength indicated by the range between daily minimum (TSW − TGW)min and maximum (TSW − TGW)max temperature difference between groundwater and surface water was significantly lower in winter, when the signal strength did vary by less than 0.5°C, than in summer when diurnal variation in signal strength could be up to eight times higher (Figure 5c). Overall, average daily signal strength (given by ΔT, the difference of surface water and groundwater temperatures) and the corresponding intensity of temporal signal variation (Tamplitude, given by the diurnal amplitude of surface water-groundwater temperature differences) were correlated in summer and winter conditions (Figure 6). The directions of correlations, however, differed between the seasons. While increases in positive temperature differences (TSW > TGW) and signal strengths during summer were positively correlated with an increase in diurnal signal variation, daily variability in signal strengths in winter decreased with increasing signal strength (Figure 6).

Figure 6.

Comparison of average daily signal strength ΔT given by the difference of surface water and groundwater temperatures and the corresponding intensity of temporal signal variation given by the diurnal amplitude of surface water-groundwater temperature differences (Tamplitude) during the 12 June 2009 to 11 August 2010 observation period.

3.3. Seasonal FO-DTS Surveys in Different Monitoring Modes

[26] The FO-DTS surveys in single-ended (two directions) and double-ended mode revealed the existence of distinct warm (winter) and cold (summer) spots in streambed temperatures (Figure 7). Locations of warm and cold anomalies within the three sampling dates coincided (Figures 8 and 9), although the spring survey on 16 March 2010 only identified warm spots at the southern downstream section (Figure 7). Warm and cold spots in the upstream and center sections that were identified on 10 February and 11 August 2010 were not detectable on 16 March 2010 (Figure 7). Winter warm and summer cold spots mainly included the most downstream and upstream river sections as well as a smaller section in the center of the stream reach (Figure 7). While cold and warm spots in the downstream section usually coincided at both parallel cable sections along the cross-sectional profile, this was not always the case in the upstream and center section where cold or warm spots were occasionally limited to one side of the channel cross section.

Figure 7.

Average FO-DTS monitored temperatures over 100 measurements in single-ended (both directions, clockwise and anticlockwise) and double-ended monitoring modes, respectively, for surveys on (top) 10 February 2010, (center) 16 March 2010, and (bottom) 11 August 2010. Numbers in brackets mark the monitoring intervals at the cable entry and exit points as well as in the area of the downstream and upstream turning points in the streambed.

Figure 8.

Time-averaged FO-DTS temperatures for spatial sampling intervals (x axis) on 10 February, 16 March, and 11 August 2010 in single-ended monitoring mode in (a) clockwise and (b) anticlockwise sampling direction; standard deviations (STDEV) of local temperature measurements as indicator of temporal signal variability for single-ended monitoring mode in (c) forward and (d) backward sampling direction. Downstream and upstream turning points of the fiber-optic cable are located in the region of 70 and 180 sampling intervals.

Figure 9.

Time-averaged FO-DTS temperatures for spatial sampling intervals (x axis) on 10 February, 16 March, and 11 August 2010 in (a) double-ended monitoring mode and (b) two-way single-ended averaging mode; standard deviations (STDEV) of local temperature measurements as indicator of temporal signal variability for (c) double-ended monitoring mode and (d) two-way single-ended averaging mode.

[27] The spatial extent of the discrete cold and warm spots did not exceed stream sections longer than 25 m (Figure 7). Cold and warm spots were defined by sharp boundaries with intensive temperature differences of up to 4°C between succeeding measurement sections (Figures 8 and 9). Both, summer and winter FO-DTS surveys identified distinctive streambed temperature baselines, which did not increase or decrease with streamflow direction (Figures 8 and 9). Baseline temperatures were not affected by the cold or warm spots and quickly receded back to baseline levels at the end of detected temperature peaks (Figures 8 and 9). Generally, temporal averages of winter baselines varied by a maximum of 0.2°C in space, less than half of the spatial variation of the averaged summer baseline with 0.5°C (Figures 8 and 9).

[28] Temporal variation during 100 measurements at the same location (indicated by STDEV) for the summer FO-DTS surveys exceeded the temporal variability during winter and spring surveys in all sampling modes. STDEV of double-ended FO-DTS surveys on 11 August 2010 were significantly higher (three to four times) than double-ended surveys in winter or spring as well as single-ended surveys or surveys in two-way single-ended averaging mode during all sampling dates (Figure 9). STDEV of FO-DTS surveys in single-ended mode followed spatial trends and increased toward the cable end in survey direction (Figure 8). Similar spatial trends in STDEV were not identified for double-ended or two-way single-ended averaging surveys (Figure 9). The intensity of warm spots in winter exceeded that of the cold spots in summer (Figures 8 and 9). Winter warm spots of more than 5°C above the temperature baseline were detected in single-ended mode on 10 February 2010, while summer cold spots in single-ended monitoring mode on 11 August 2010 did not exceed differences of 3°C to the baseline (Figure 8).

[29] Although the locations of warm and cold spots detected by FO-DTS surveys in different monitoring modes coincided in the winter and summer FO-DTS surveys, their intensity or signal strength varied depending on the specific monitoring mode (Figure 10). In order to quantify the detected strength of the investigated temperature signal independently from the seasonally variable global temperature shifts, temperature anomalies defined by local temperature differences to the spatial average AT(xi) were analyzed in Figure 10. Signal strength was thus measured by the normalized positive (winter) and negative (summer) deviation from the spatial mean. Peak signal strengths for both winter warm anomalies and summer cold anomalies were higher when monitored in single-ended mode than double-ended mode (Figure 10). The difference relative to the spatial average in double-ended monitoring mode was particularly large for peak anomalies of small spatial extent (e.g., at 12, 116, 172, and 182 m, Figure 10), at occasions more than 1°C smaller than in single-ended. In two-way single-ended averaging mode, signal strength as well as their spatial extent matched the ranges of observations in single-ended mode (Figure 10). An analysis of differences in temperature anomalies monitored in double-ended (AT(DE)(xi)) and two-way single-ended averaging (AT(2WSE)(xi)) mode indicated that while peak temperature anomalies monitored in double-ended mode were frequently smaller than in two-way single-ended averaging mode, in particular, values between quickly succeeding peaks were usually higher in double-ended mode (Figure 11).

Figure 10.

Comparison of signal strength for spatial sampling intervals (x axis) during the 10 February, 16 March, and 11 August 2010 monitoring by means of local differences from the spatial average of FO-DTS measured temperature for measurements in single-ended monitoring mode in (a) clockwise and (b) anticlockwise sampling direction and (c) double-ended monitoring mode and (d) two-way single-ended averaging mode.

Figure 11.

Difference ΔT between temporally averaged streambed temperature anomalies detected by FO-DTS in double-ended (AT(DE)(xi)) and two-way single-ended averaging (AT(2WSE)(xi)) mode along the buried fiber-optic cable (spatial sampling intervals on the x axis) for observation dates 10 February 2010, 16 March 2010, and 11 August 2010.

4. Discussion

4.1. Analysis of Temperature Signal Strength

[30] Differences between groundwater and surface water temperatures in the research area (Figure 4) produced a robust signal for FO-DTS-based monitoring of streambed temperature anomalies associated with groundwater upwelling (Figure 5a). In consequence of the temporal variability of groundwater and surface water thermal dynamics and the resulting temperature differences, the signal strength varied seasonally with differences between groundwater and surface water temperatures in winter and summer exceeding ±9°C, while signal strength during the autumn and spring transition periods declined to not more than 2°C. Temperature differences at the field site were in a similar range as previous FO-DTS applications [Selker et al., 2006b; Lowry et al., 2007; Henderson et al., 2009], and in particular, summer and winter conditions in the research area qualified with their distinctively different thermal regimes of groundwater and surface water (Figure 5a) for FO-DTS surveys of aquifer-river exchange.

[31] While the generally high signal strength did not vary substantially between summer cold anomalies and winter warm anomalies (Figure 5a), the substantially higher diurnal oscillations in surface water temperatures during summer (Figure 5b) resulted in subsequently increased daily variability of summer signal strengths. In contrast, in winter, when the range between minimum and maximum groundwater-surface water temperature differences was significantly smaller (Figure 5c) due to the reduced diurnal oscillation of surface water temperatures (Figure 5b), the daily variability of signal strength was substantially lower. In fact, the comparison of average daily signal strength and the diurnal variation in signal strength revealed contrasting trends for summer and winter conditions with diurnal signal variation in summer increasing with signal strength, while it decreased with signal strength in winter (Figure 6).

[32] This suggests that for the investigated conditions, more robust signal conditions can be expected during winter when surface water was colder than groundwater. The timing of FO-DTS monitoring in winter can be expected to have only minor impact on the signal strength, whereas summer surveys can be noticeably affected by the choice of the monitoring period. As the signal strength in the research area could vary by >30% during this time, the timing of the survey can have potential implications. These can be critical when quantitative assumptions are inferred from the detected signal strength [e.g., Roth et al., 2010; Westhoff et al., 2007, 2011a, 2011b] or long-term FO-DTS surveys are run for 24 h and longer [e.g., Selker et al., 2006b; Lowry et al., 2007; Roth et al., 2010].

4.2. Seasonal Consistency of FO-DTS Monitored Temperature Patterns

[33] FO-DTS surveys for winter (with groundwater temperatures above surface water, Figures 4 and 5) and summer conditions (with groundwater temperatures below surface water, Figures 4 and 5) provided evidence of distinct streambed temperature anomalies caused by upwelling groundwater (Figure 7). The strongest streambed anomalies during those surveys (Figure 7) even reached groundwater temperatures at the respective time period (Figure 4). In contrast, FO-DTS surveys in spring did reveal significantly lower streambed temperature anomalies (Figure 7), highlighting that the signal strength constituted by the temperature differences between groundwater and surface water (Figure 5) was critically low, and consequently, FO-DTS surveys during such transition seasons were of limited suitability for detecting spatial patterns of groundwater upwelling at the field site. Similar temporal dynamics in FO-DTS signal detection on shorter (diurnal) time scale have been previously identified for substantially smaller streams, with proportionally higher groundwater contributions to the stream discharge [e.g., Selker et al., 2006b].

[34] As locations of winter-identified warm anomalies and summer-identified cold anomalies spatially coincided, it can be assumed that spatial patterns of groundwater upwelling at the field site were stationary over the observation period. This coincides with findings of Krause et al. [2012a] which identified, in a 2009 FO-DTS survey, low conductivity streambed structures to represent the major controls of groundwater upwelling patterns, in contrast to temporally more dynamic streambed topography and surface features. This hypothesis is furthermore supported by the occurrence of warm or cold anomalies that are restricted to one side of the channel cross section, which can be explained by spatial variability in streambed hydraulic conductivity [Krause et al., 2012a] and does not coincide with any streambed topographic features at these locations (Figure 1c).

[35] The fact that the spring FO-DTS survey only detected a downstream temperature anomaly but not the upstream and center sections identified by winter and summer surveys could theoretically indicate that the upwelling of groundwater in these locations is seasonally reduced. However, this is unlikely given that winter as well as summer FO-DTS surveys detected strong positive or negative temperature anomalies at these locations. Taking into account that the regional variability of groundwater temperatures (Figure 4) during the FO-DTS survey on 16 March 2010 nearly matched the difference between average groundwater and surface water temperatures (Figure 5), the nondetection of these temperature anomalies is interpreted as indication that signal strength was too low for detection.

4.3. Seasonal Variability of FO-DTS Detected Signal Strength

[36] Although spatial patterns of temperature anomalies during winter and summer FO-DTS surveys in the study area coincided, the detected signal strengths in both seasons varied substantially. Warm anomalies on the 10 February 2010 were higher than cold anomalies detected on the 11 August 2010 at the same location (Figure 10), indicating that indeed winter FO-DTS surveys at the field site might be more suitable for identifying groundwater upwelling patterns with higher accuracy than in summer. With temperature anomalies of more than +4°C in winter and more than −2.5°C in summer, FO-DTS detected signal strength in the research area was in a similar range as reported by Lowry et al. [2007] for streambed temperatures during summer conditions or by Selker et al. [2006b] for river column temperatures.

[37] It could be argued that neither the winter nor the summer FO-DTS survey dates coincided with the highest signal strength possible during these seasons (Figure 5a). However, despite the strength of the temperature signal in summer and winter being of comparable magnitude (Figure 5a), summer increases in diurnal oscillation of surface water temperatures (Figure 5b) and differences between minimum and maximum surface water-groundwater temperatures (Figure 5c) evidently caused higher signal instability during the summer FO-DTS survey than in winter, as indicated by the overall increased STDEV for the 11 August 2010 FO-DTS survey (Figures 8 and 9). Theoretically, the smaller variability in spatial fluctuation of streambed temperature baselines during winter compared to summer (Figures 8 and 9) could potentially be caused by the impact of direct radiation [Carrivick et al., 2012; Hannah et al., 2009] and shading effects [Hannah et al., 2008], which would have a stronger impact during summer than winter and therefore cause less “noisy” baseline temperatures in winter. However, roaming surveys of stream temperatures during the surveys indicated spatial temperature variability in the stream of less than 0.1°C, presumably because the rather high flow velocities caused a fast dissipation of any potential shading effects. The observed signal instability of summer baseline temperatures might also be affected by the higher temperature ranges between summer maximum temperatures and calibration ice baths.

[38] As the baseline of streambed temperature was not affected by the locally constrained temperature anomalies, and the upwelling of relatively warmer groundwater in winter or colder groundwater in summer did not result in a downstream cooling or warming effects, it can be assumed that in summer as well as winter, overall groundwater contributions to the stream discharge were too insignificant to change the entire thermal regime of the investigated stream section. Independent from seasonal differences in detected streambed anomalies, the upwelling of groundwater at the study area caused only local effects [Krause et al., 2012a], which differs from previous findings in smaller streams [e.g., Selker et al., 2006b; Westhoff et al., 2007].

[39] In summary, this qualifies the local winter and summer conditions as adequate for FO-DTS surveys of groundwater upwelling induced streambed temperature anomalies, with expected higher monitoring accuracies during the temporally more constant winter conditions.

4.4. Impact of FO-DTS Monitoring Mode on Signal Detection

[40] The monitoring mode of the FO-DTS surveys evidently had substantial impact on the qualitative (spatial extent and location) as well as quantitative (signal strength) accuracy of the survey (Figure 10). The reduction of the intensity of temperature anomalies (Figures 8 and 9) detected in double-ended monitoring mode (in comparison to single-ended observations) may be attributed to a smoothing effect caused by the fact that (1) the attenuation correction in forward and reverse modes are integrated and hence constant in double-ended survey mode [Smolen and van der Spek, 2003; Tyler et al., 2009; Van de Giessen et al., 2011] (Figure 2) and (2) longitudinal signal shift along the cable by one monitoring interval between forward and reverse traces has been corrected in two-way single-ended averaging but was not possible in double-ended mode.

[41] This smoothing did predominantly cause a reduction of peak signal strengths of temperature anomalies with small spatial extent (Figures 10 and 11), which can have a critical impact, in particular, on the quantitative interpretation of FO-DTS survey results or their further use in heat transport models [Roth et al., 2010; Westhoff et al., 2010, 2011a, 2011b]. Where temperature anomalies did not exceed one or two sampling intervals, double-ended surveys resulted not only in an underestimation of peak thermal anomalies compared to the anomalies observed in single-ended mode but also in a smoothing or overestimation of temperature anomalies in the short-termed “gaps” between quickly succeeding peaks (Figures 10 and 11). This means that overall, double-ended monitoring leads to a reduction in variability of detected temperature anomalies if compared to monitoring in single-ended mode, which is of particular relevance for highly variable systems with discrete signal changes and small-scale temperature variability.

[42] If the two single-ended traces, however, were corrected for direction-dependent signal drift before averaging the calculated temperatures as proposed in the two-way single-ended averaging mode (Figure 2), detected peak intensities of temperature anomalies as well as gaps between swiftly succeeding peaks were similar to single-ended observations (Figure 11). The smoothing character of double-ended monitoring is well known and has been reported previously [e.g., Tyler et al., 2009; Van de Giesen et al., 2012]. However, a wide range of previous FO-DTS applications (not just at aquifer-river interfaces) has been carried out in double-ended mode as measurements in opposite directions provide more robust observations than single-ended surveys, allow for the analysis of symmetry, and are expected to provide higher survey quality in case of cable damage or light losses [Tyler et al., 2009; Van de Giesen et al., 2012]. As shown by our results, the proposed two-way single-ended averaging mode has the potential to yield improved signal strengths and may therefore represent an alternative to the application of standard double-ended monitoring mode, in particular, if the spatial extent of the anomalies or temperature signals is expected to be close to the monitoring resolution as in the presented case.

[43] Recently, laboratory experiments have shown that the accuracy of FO-DTS signal detection declined quickly with reduced signal sizes, resulting in an inaccurate detection of signal strength and spatial extent for signals that were not significantly above the sampling resolution of the applied FO-DTS instrument [Rose et al., 2013]. Such inaccuracies are of particular concern when the correct detection of discrete signals, their strength, and spatial alteration is paramount because they form an integral part of the investigated problem [e.g., Lowry et al., 2007; Krause et al., 2012a, 2012b]. As the robustness of monitoring in double-ended mode is higher due to the inherent redundancy provided by measurements in opposite directions, it will provide stronger evidence for the existence of small-scale temperature anomalies. The predictive capacity, however, will be impaired if the strength of the detected signal is underestimated as shown in this study. The proposed two-way single-ended averaging provides a useful alternative, detecting signals at similar strength and spatial resolution as single-ended monitoring combined with the robustness of a survey in double-ended mode. Although this method requires extra data postprocessing, the additional effort is not substantial and is outweighed by the advantages in detection accuracy.

5. Conclusions

[44] The identified seasonal variability in signal strength has critical implications for the design of future FO-DTS surveys which should aim to reduce the amount of diurnal signal variation (e.g., by choosing a period/season with minimal short-term fluctuation or diurnal oscillations) of one or both end-members that are causing the traced thermal anomalies. For the investigated stream reach, with a thermal regime comparable to many rivers of midlatitudes without glacial or tidal influences, winter conditions with substantially lower diurnal temperature oscillations provide the highest stability in signal strength.

[45] The tested two-way single-ended averaging of FO-DTS surveys proved to have significant advantages compared to standard single-ended or double-ended surveys, with higher accuracy in signal detection, in particular for small-scale peak temperature anomalies as well as for discontinuous signal peaks that were not captured by surveys in state of the art double-ended monitoring mode. FO-DTS in two-way single-ended averaging mode was better suited to detect the full complexity of spatial temperature patterns in the investigated highly heterogeneous aquifer-river interface.

[46] The findings of our case study are not limited to the investigated field site or the investigation of groundwater-surface water exchange. The quantification of accuracies and uncertainties of FO-DTS surveys in dependency of seasonal signal variation and monitoring mode is transferable to other situations, and we recommend the testing of the two-way single-ended averaging mode for FO-DTS application in systems other than aquifer-river interfaces as well.

Acknowledgments

[47] The authors would like to acknowledge funding for this work by the Royal Geographical Society, EPSRC, and NERC (the U.K. Natural Environment Research Council, NE/I016120). The authors thank Kevin Voyce (Environment Agency of England and Wales) and Ian Wilshaw (University of Keele) for the provision of hydrometeorological data and Tim Millington (formerly University of Keele) for technical support with data processing routines. The authors thank Michael Mondanos (SILIXA, formerly Sensornet) as well as two anonymous reviewers and Alex van der Spek for very helpful discussions, which helped to improve this manuscript.

Ancillary