Abstract

Soil temperature (Ts) exerts critical controls on hydrologic and biogeochemical processes, but the magnitude and nature of Ts variability in a landscape setting are rarely documented. Fiber-optic distributed temperature sensing (DTS) systems potentially measure Ts at high density across a large extent. A fiber-optic cable 771 m long was installed at a depth of 10 cm in contrasting landscape units (LUs) defined by vegetative cover at Upper Sheep Creek in the Reynolds Creek Experimental Watershed (RCEW) and Critical Zone Observatory in Idaho. The purpose was to evaluate the applicability of DTS in remote settings and to characterize Ts variability in complex terrain. Measurement accuracy was similar to other field instruments (±0.4°C), and Ts changes of approximately 0.05°C at a monitoring spatial scale of 1 m were resolved with occasional calibration and an ambient temperature range of 50°C. Differences in solar inputs among LUs were strongly modified by surface conditions. During spatially continuous snow cover, Ts was practically homogeneous across LUs. In the absence of snow cover, daily average Ts was highly variable among LUs due to variations in vegetative cover, with a standard deviation (SD) greater than 5°C, and relatively uniform (SD < 1.5°C) within LUs. Mean annual soil temperature differences among LUs of 5.2°C was greater than those of 4.4°C associated with a 910-m elevation difference within the RCEW. In this environment, effective Ts simulation requires representation of relatively small-scale (<20 m) LUs due to the deterministic spatial variability of Ts.

Soil is a primary interface between terrestrial physical and biological processes. In addition to the critical role of soil in the hydrologic cycle, it mediates most of the biogeochemical interactions between vegetation, groundwater, and the atmosphere (Wilding and Lin, 2006). It has been understood for many years that soil temperature (Ts) influences a host of soil processes, as evidenced by the prominent role it plays in Soil Taxonomy (Soil Survey Staff, 1999). These processes regulate soil carbon dynamics (Davidson and Janssens, 2006; Dornbush and Raich, 2006), soil formation (Jenny, 1980; Chorover et al., 2011; Rasmussen et al., 2011), soil fertility (Bardgett and Wardle, 2010), plant growth (Leguizamón et al., 2005; Mellander et al., 2006; Jonas et al., 2008), and streamflow generation (Seyfried et al., 2009; Buttle, 2011). As a result, variations of Ts across the landscape impact important ecosystem services and the global carbon cycle (Stockmann et al., 2013). It is therefore critical that we develop a quantitative understanding of Ts variability, especially in regions with complex terrain, where slope, aspect, elevation, and vegetative cover may all contribute to the overall Ts variability at relatively small scales.

The spatial variability of Ts may be represented as either deterministic or stochastic (Philip, 1980; Rao and Wagenet, 1985; Seyfried, 2003). With deterministic variability, Ts varies spatially in a predictable manner, usually represented as a theoretically or empirically determined function of more easily measured parameters such as slope and aspect. Numerous methods for this have been developed (Minasny et al., 2013). Stochastic variability is random “noise” and includes, or is indistinguishable from, measurement variability (assuming no bias), small-scale variability, and other “unexplained” variability (e.g., Wilson et al., 2004) and may be described by statistical parameters such as the mean and variance or a probability distribution function. Spatially distributed models used to simulate ecohydrological processes discretize the landscape into units or cells that are assumed to be effectively homogeneous with respect to the parameter of interest (i.e., Ts is represented by a single value). The within-cell variability and accuracy may determine how well processes driven by Ts are simulated or, alternatively, how well models that simulate Ts can be tested or verified.

Air temperature (Ta) is an important determinant of Ts and is often used to estimate the deterministic component of Ts variability (Persaud and Chang, 1983; Soil Survey Staff, 1999; Brown et al., 2000; Bond-Lamberty et al., 2005). In mountainous terrain, this is consistent with the use of the adiabatic lapse rate to discretize the landscape into elevation bands (Ahl et al., 2008; Kang et al., 2000). However, Ts is sensitive to small-scale (relative to Ta) topographic variations (Wundram et al., 2010; Ebel, 2012), soil properties (Hillel, 1998), vegetative cover (Balisky and Burton, 1993; Ebel, 2012; Kang et al., 2000; Lutz et al., 2012), and snow cover (Groffman et al., 2001; Hardy et al., 2001; Seyfried et al., 2001; Link et al., 2004; Schaetzl et al., 2005). Elevation bands (or Ta) do not represent these sources of deterministic variability, potentially increasing the amount of unexplained or stochastic variability. If relatively small-scale landscape features have a substantial effect on Ts, then Ta alone cannot provide a robust description of Ts variation, and higher spatial resolution may be required.

Advances in remote automated measurement technology during the past 25 yr have resulted in a great increase in soil temperature data collection. These data are typically discrete points that provide a limited number of “benchmark” sites that are rarely designed to address questions related to either deterministic or stochastic Ts variability, and hence the degree of Ts spatiotemporal variability in complex terrain remains largely undocumented (Briggs et al., 2012). Fiber-optic distributed temperature sensing (DTS) has recently emerged as a measurement technique that may prove useful for addressing these issues. The method “measures” temperature at roughly 1-m intervals at predetermined time intervals along fiber-optic cable up to 10 km long, thus providing transect-type data with as many as 10,000 measurements to address questions related to temporal and spatial variability. Although DTS has been used since the 1980s, it is only since 2006 that it has been applied to hydrologic systems (Tyler et al., 2009). Hydrologic applications to date have emphasized stream monitoring (Selker et al., 2006a, 2006b; Westhoff et al., 2007; Moffett et al., 2008; Briggs et al., 2012). Other related applications include measurement of temperatures at the snow–soil interface (Tyler et al., 2008) and estimation of soil water content (Ciocca et al., 2012).

One of the keys to advancing hydrologic science is the development of new measurement systems capable of sensing fine-scale spatial and temporal variations of environmental variables (e.g., Kirchner, 2006). As with any new application of technology, numerous practical issues have emerged that have a significant bearing on the ultimate utility of the technology. For example, the rate of calibration drift under variable and wide-ranging (e.g., field) conditions, measurement noise, and instrument power requirements are all instrument issues that have not been fully investigated (Tyler et al., 2009). Evaluation of instrumental performance under a wide range of conditions is a critical step to advancing novel measurement techniques that enable observations of highly dynamic environmental systems.

We are interested in describing processes that are partially controlled by Ts in landscapes dominated by complex terrain and strong elevation gradients where a seasonal snow cover is common and vegetative cover highly variable. In this context, the combined effects of topography, snow cover, and vegetation on the amount and nature of Ts variability are of fundamental importance. We used DTS to measure Ts in different landscape units (LUs) in the Upper Sheep Creek watershed with the following objectives: (i) to evaluate the applicability of DTS to remote field environments; (ii) to describe the spatial and temporal variability of Ts in complex terrain characterized by contrasting LUs; and (iii) to investigate the implications of these findings for modeling Ts and related processes in complex terrain. We anticipate that the results will lead to improved approaches to modeling processes affected by Ts and further application of DTS for the measurement of soil-related processes.

Methods

Site Description

This research was conducted during water year 2010 in the Upper Sheep Creek watershed, which is located within the larger Reynolds Creek Experimental Watershed and Critical Zone Observatory in southwestern Idaho (Fig. 1). The watershed is 0.26 km2 in extent, with an elevation range of 1840 to 2036 m asl, all underlain by basalt. Mean annual precipitation is approximately 426 mm, about 60% of which is snow. Streamflow is intermittent, with an average annual yield of 44 mm.

Snowfall at Upper Sheep Creek is redistributed by wind, resulting in scour and accumulation (drift) zones. Snow drifts commonly exceed a depth of 3 m, while scour zones rarely accumulate more than 10 to 15 cm of snow. Drift and scour zones occur in highly consistent patterns from year to year, resulting in three distinctive, topographically determined snow–soil–vegetation complexes or LUs (Flerchinger et al., 1998; Prasad et al., 2001; Chauvin et al., 2011). The LUs have been defined by the dominant vegetation and common rangeland cover types (Flerchinger and Seyfried, 2014). The low sagebrush (Artemisia arbuscular Nutt.) LU is has shallow (20–40 cm), rocky soils with loam and clay loam textures and is mostly found on ridges or steep, west-facing slopes. The shrubs are short in stature (<30 cm) and, with associated grass, form a sparse cover. Snow accumulation rarely exceeds about 30 cm. The mountain big sagebrush [Artemisia tridentata Nutt. ssp. vaseyana (Rydb.) Beetle] LU is found on deeper (about 1 m) soils derived from loess on north-facing slopes. Soils are typically rock free for >80 cm. Mountain big sagebrush and snowberry (Symphoricarpos oreophilus A. Gray) are the dominant shrubs and form a nearly closed canopy 1 to 1.2 m tall. The shrub understory is a dense cover of grasses and forbs, and maximum snow cover is generally 0.6 to 1 m deep. The aspen (Populus tremuloides Michx.) LU includes some willow (Salix spp.), and the trees grow in dense thickets on north-facing slopes. Soils are deep (~2 m) loams and silt loams derived from loess with low rock contents. The aspen LU is associated with snow drifts that are commonly 3 m deep.

In 2007, a prescribed fire followed by cutting of the aspen and willow in 2008 resulted in a change in vegetative cover. However, snow distribution patterns and underlying soil and topography were unaffected by the fire, and vegetative cover differences among the LUs emerged shortly after the fire. Using previously described methods (Clark et al., 2001; Flerchinger and Seyfried, 2014), we measured the exposed soil surface (no litter or canopy cover), leaf area index, and the percentage of ground covered by plant litter in 1-m2 plots, each with 100 measured points in a grid. There were 16 plots in the low sagebrush LU and eight plots in the mountain big sagebrush LU. In 2010, there were large differences in exposed soil percentage and litter cover, with the low sagebrush LU having greater exposed soil and less litter cover (Table 1). Leaf area index (LAI) was also lower in the low sagebrush LU. The aspen LU was monitored with different methods (Flerchinger and Seyfried, 2014) and had an LAI of 2.7 and, by ocular estimate, virtually no exposed soil.

Instrumentation

Fiber-Optic Distributed Temperature Sensing

The theory of operation for DTS (Dakin et al., 1985; Kurashima et al., 1990) and description of hydrologic applications (Selker et al., 2006a, 2006b) are well documented. Briefly, a pulsed laser generated in the DTS instrument is transmitted through a fiber-optic cable. A very small portion of the pulsed light is returned as Raman backscatter to the DTS instrument, which measures the timing and intensity of the backscatter. Raman backscatter occurs at two wavelengths, Stokes, which occurs at a longer wavelength than the incident light, and anti-Stokes, which is at a shorter wavelength than the incident light. The amplitude of the anti-Stokes backscatter is dependent on the fiber temperature at the point of backscatter, while the amplitude of the Stokes backscatter is essentially temperature independent. The DTS instrument measures both the Stokes and anti-Stokes backscatter and computes the ratio of the two, which is then related to the fiber temperature at the point (length increment) of backscatter.

Measurement precision depends on the amount of light returned to the sensor and is therefore primarily dependent on the measurement duration (Selker et al., 2006a; Ciocca et al., 2012). Temperature resolution approaching 0.01°C may be obtained with very long measurement times (Selker et al., 2006a). For any given measurement duration, precision generally declines with distance from the DTS instrument because the signal dissipates (in both directions) along the fiber. Measurement accuracy is dependent on the quality of calibration, which has two primary components, an offset calibration relating the backscatter ratio to a known temperature, and a loss calibration that accounts for signal loss effects along the length of the cable.

A Sensornet Oryx DTS (Sensornet Ltd.) with a dual 50-μm multi-mode fiber (AFL Telecommunications DNS-1566) was used. The manufacturers report a temperature resolution of 0.1°C at 2000 m from the pulser given a 1-min integration time. The measurements have a nominal spatial resolution of 1 m, although this should be investigated where spatial locations of measurements are critical (Tyler et al., 2009; Rose et al., 2013).

Fiber-Optic Cable Installation.

The fiber-optic cable was installed in a duplexed, single-ended configuration (Hausner et al., 2011) with two calibration coils, one adjacent to the DTS instrument and one at the end of the cable where the two fibers in the cable were fused. The DTS instrument was housed in a small weir house located at the watershed outlet and powered with two 60-W solar panels. We installed 771 m of cable from the instrument to the ridge top, where the second calibration coil was located. The cable was buried 10 cm deep in the following five segments (Fig. 2): (i) the west-facing segment (WF) in the low sagebrush LU with a slope of 14°, (ii) the north-facing segment (NF), in the mountain big sagebrush LU with a 17° slope, (iii) the north-facing segment covered by a seasonal snow drift segment (DNF) in the aspen LU with a 23° slope, (iv) a segment on the ridge covered by a seasonal drift (DR), and (v) a wind-scoured portion of the ridge (R). The DR and R segments were in the low sagebrush LU on gentle, north-facing slopes.

We did not analyze data from unburied cable segments but include a description of the total cable length because all data collected along the cable are part of the same measurement. Starting at Meter 0, where the cable leaves the DTS instrument, the first 58 m were in the weir house (Fig. 2), 25 m of which were rolled in a roughly 60-cm-diameter coil used for calibration and accuracy testing. After leaving the weir house, the cable was laid on the ground surface for 284 m, where the WF segment started and the cable was buried 10 cm deep. The WF segment was 132 m long (Meters 284–416) and traveled about 65 m upslope and then returned to the channel (Fig. 2). At Meter 416, the cable crossed the channel (unburied) and entered relatively flat ground on the opposite side of the channel and was buried at the 10-cm depth. After 24 m, the cable was wrapped in an exposed coil 16 m long. The cable was then buried and, after crossing an access trail, entered the NF slope at Meter 460. The NF segment was 217 m long (Meters 460–677). For the next 54 m, the cable was buried in the zone occupied by a seasonal snow drift. Within this segment, there is a distinct slope and vegetation break on the ridge at Meter 718 that separates the DNF segment (aspen LU) from the DR segment (low sagebrush LU). The cable then passed 8 m on the ridge and entered an exposed calibration coil from Meter 739. At the end of the coil, Meter 771, the two fibers in the cable were welded together, effectively creating a single 1542-m-long fiber.

Calibration and Monitoring.

Upon installation, we performed a slope calibration to account for signal loss along the cable using the calibration coils in the weir house and on the ridge, with the weir house coil in an ice bath and the ridge coil in water at the ambient air temperature. Thereafter, several offset calibrations were performed in an ice bath using the weir house coil (Table 1). Ice bath temperature was measured with a FLUKE (Hart Scientific, 1996) temperature meter (Model 1502a) and platinum resistance thermometer (Model 5613), which is a secondary standard with accuracy of ±0.006°C at 0°C. Each measurement was taken from the weir house coil, which was submerged in the calibration bath (usually an ice bath). Measurement duration was usually 1 min (Table 1). Note that the sampling interval (1 m in this case), may be very different from the measurement spatial resolution of the instrument (Tyler et al., 2009; Selker et al., 2014). Samples taken within the instrument spatial resolution are correlated. Consistent with other observations (e.g., Tyler et al., 2009; Selker et al., 2014), we found that a 2-m buffer between the water bath and outside yielded no visible correlation. That is, the measurements inside the water bath were effectively independent of those outside. This resulted in 20 m of cable in the calibration bath, effectively yielding 20 replicate samples. We then used every third sample for seven independent measurements to provide an estimate of the measurement variability within the calibration bath.

Measurement protocols evolved during the deployment in response to power issues. Initially, we made one 5-min reading each hour. As the season progressed, this was reduced to 2 min and then 1 min to prevent excessive power drain. During each visit, we evaluated the amount of calibration drift that occurred between visits by measuring the DTS temperature in the ice water bath prior to (pre-calibration) and immediately following (post-calibration) instrument calibration. We used three different readings (each with seven measurements) to determine the mean and standard deviation of the measured coil temperature. On 17 April, two PT100 temperature sensors (Campbell Scientific) were installed that were co-located with the calibration coil to monitor calibration drift.

All data reported were made as single-ended measurements. Temperatures measured in the first 771 m (light traveling from the DTS to the ridge) were always in the same fiber, referred to as Fiber 1, while temperatures measured in the second 771 m (light traveling from the ridge to the DTS) are referred to as Fiber 2. Data collection was abruptly terminated on 22 July when cattle found a gap in the fence around Upper Sheep Creek and severed the exposed cable in numerous locations.

Complementary Data Collection

Three sets of preexisting soil water and temperature profiles, one in each of the LUs, were used to provide field validation of the DTS and a detailed description of Ts dynamics. The location of each profile relative to the cable segments is shown in Fig. 2. At each location, a pit was excavated to basalt bedrock, which ranged in depth from 45 cm at the WF site to 220 cm at the DNF site. Two co-located depth profiles of sensors measuring soil water content using time domain reflectometry (TDR) and Ts by thermocouple, separated by about 2 m, were installed in each pit. Snow depth was measured with an ultrasonic depth gauge (Judd Communications) at the WF and NF sites. Continuous snow depth monitoring at the snow drift was not practical, but snow depth was measured by hand coring on two dates as part of the normal snow survey at Upper Sheep Creek. Hydrometeorological data were collected hourly at the J10 climate station (Fig. 2), with precipitation measured with dual weighing gauges (Hanson et al., 2004; Chauvin et al., 2011).

Results and Discussion

Instrument Performance

Calibration Accuracy

We made 10 offset calibrations during the study period, eight in an ice bath (Table 2). Ice bath temperatures, as determined with the FLUKE, were within 0.01°C of 0°C, or effectively 0.00°C. Post-calibration DTS measurements in the ice were between −0.07°C and 0.15°C, with an average deviation (absolute value) from 0°C of 0.09°C (Table 2). The standard deviation (SD) was dependent on the measurement duration: between 0.05 and 0.07°C for a 1-min duration and between 0.09 and 0.1°C for a 30-s duration, which are both in good agreement with previous research (Selker et al., 2006a; Rose et al., 2013). Thus, we could generally expect that, immediately following calibration, measurements made near the calibration coil and near 0°C to be within 0.15°C (and usually much closer) of the true value. However, the calibration and thus measurement accuracy may drift both with time and with distance along the cable, and measurement precision deteriorates with distance from the DTS instrument.

Temporal Calibration Drift

The difference between the pre-calibration temperature on a given date and the post-calibration temperature on the previous date is a measure of the net calibration drift during that time interval. By this measure, calibration drift ranged from 0.40 to 0.04°C, averaging 0.223°C or, expressed in terms of degrees of drift per day, 0.013° d−1. The larger changes represent statistically significant (α = 0.05) differences. We could detect no trend in the rate of drift, which varied dramatically during the study. There does, however, appear to be a dependency of the calibration on the DTS instrument temperature (temperature in the weir house was not controlled) at the time of calibration, which has been noted by others (Tyler et al., 2009). For example, when the instrument temperature (Table 2) dropped 19.55°C between the post-calibration measurement on 1 December and the pre-calibration measurement on 9 December, there was an apparent temperature change of −0.347°C in the calibration coil, while during the next visit (on 6 January), the instrument temperature increased 13.6°C and the apparent calibration coil ice bath temperature increased 0.372°C. Overall, there was a strong linear correlation (R2 = 0.85) between the change in the DTS-measured ice bath temperature and the change in the instrument temperature, indicating a 0.02°C increase in apparent coil temperature per degree of instrument temperature change.

We also evaluated temporal drift using PT100 data that were collected hourly starting 17 Apr. 2010. Due to differing thermal mass and temperature variations along the coil, which was in air between calibrations, the coil and PT100 temperatures changed at slightly different rates during the day. However, when we compared daily average DTS-measured coil temperature to that of the PT100 for all days during which the PT100 was installed, we found a very strong, linear relationship (R2 > 0.99), indicating that the DTS calibration was quite stable with time (Fig. 3). This high degree of calibration stability existed in spite of the temperature dependence described above because those effects are small (at most 0.4°C) relative to the total 20°C temperature change in Fig. 3 and because the broader seasonal warming trend was accounted for by periodic calibration. Based on the data we have collected, it appears that we achieved accuracies of better than ±0.4°C, which is comparable to other field instrumentation (Campbell Scientific Inc., 2012) and acceptable to assess spatiotemporal variations of Ts in relation to complex terrain.

Resolution and Reproducibility

Measurement reproducibility and resolution of the field data are illustrated qualitatively with data from four 1-min readings made at 0241, 0341, 0541, and 0641 h on 10 Feb. 2010 (Fig. 4). Independent Ts measurements at the three soil profiles described above were stable and near 0°C, the ground was entirely covered with snow except on the exposed ridge, there was no incoming shortwave radiation (at night), and there was little air temperature change. Data for the first 771 m are temperatures measured from the DTS instrument in the weir house to the ridge coil using Fiber 1, those in the next 771 m are from the ridge coil to the weir house using Fiber 2 in the same cable. Temperatures in the weir house (Meters 0–58, Fiber 1; Meters 713–771, Fiber 2) cooled about 0.5°C during the period due to air temperature change; otherwise, the four readings were practically indistinguishable along the length of the cable. Temperatures in Fiber 1 after Meter 60 were generally within about 0.5°C of 0 except in or near the exposed ridge coil, where they were less than −2°C. An expanded view of the four readings from Meters 300 to 400 (Fig. 4 inset) reveals slight differences in the four readings, all within about 0.06°C of each other. Very small temperature differences, about 0.05°C, are consistently resolved. Thus, the small temperature variations along the cable, which may appear to be measurement noise, reflect small but actual temperature changes. Note also the apparent temperature jump at the weld and subsequent upward displacement of temperatures the entire length of Fiber 2. Such discontinuities at a weld are common.

Cable Length Effects

To better illustrate the gradual signal change that occurs with distance along the cable, the Fiber 2 data are superimposed on the Fiber 1 data in Fig. 5, so that the temperature measured at each point in the two fibers is matched in terms of distance from the DTS, in this case for the 0241 h reading on 10 February shown in Fig. 4. Recall that the two fibers are side-by-side in the cable so that they are exposed to the same environment. It is clear from visual inspection that the small spatial temperature variations along the cable were recorded by both Fibers 1 and 2, further supporting the measurement reproducibility and resolution limits described above. The absolute temperature values measured along the two fibers are very different, however. There is a large (0.6°C) apparent temperature jump at the weld, followed by a gradual divergence from the weld to the DTS instrument, as shown in both Fig. 4 and 5. The Fiber 1 calibration coil temperature was about 1.35°C less than the Fiber 2 calibration coil temperature. This difference was consistent throughout the study.

The effect of a weld on the backscatter ratio and hence temperature calculation is difficult to predict but was stable during the study period. The change in temperature calibration with distance along the cable is generally assumed to be linear. We confirmed linearity by plotting the apparent temperature difference between Fibers 1 and 2 (ΔT) along the cable length (Fig. 5). Following the apparent temperature increase at the weld, which defines the beginning of Fiber 2, there is a strongly linear (R2 = 0.907) increase in the absolute value of ΔT along the length of the cable, resulting in an apparent temperature increase measured in Fiber 2. The small-scale ΔT variations around the linear trend are due to a combination of measurement noise, which increases with distance from the DTS, and temperature differences that are introduced by the segment-matching procedure. That is, the actual position of Fiber 2 is displaced relative to Fiber 1 by a length <1 m. This effect is small where temperature features are small, and balance out across the length of the fiber, but do add to the apparent measurement noise in the ΔT line. The ΔT dip in Fig. 5 near the weir house (approximately Meter 70) is an example of the effect.

We used the slope of the ΔT line to correct for the apparent temperature increase from the calibration coil along the cable, resulting in the “corrected” line in Fig. 5. Independent 10-cm Ts measurement at the three field sites indicate that the corrected line is closer to the actual field conditions than the uncorrected (Fiber 1) values. This is further supported by TDR data (not shown) that indicated that the soil was frozen at all three sites at this time. (The TDR-measured soil dielectric permittivity changes drastically when moist soil is frozen [e.g., Seyfried and Murdock, 1996]). The subsequent Ts data presented here are from Fiber 1 with a post-processed slope correction for cable length that incorporates the slope of the ΔT vs. distance line shown in Fig. 5.

Soil Temperature Dynamics

Here we first present Ta and precipitation for the measurement period to provide context for the belowground data and then present traditional, one-dimensional Ts profile and snow depth data from sites located near the WF, NF, and DNF DTS segments. Those data illustrate the contrasting Ts dynamics at three points representing the LUs in the watershed and provide a standard basis for comparing DTS-measured Ts. The DTS data are then presented to show how Ts varies across the landscape, both in terms of the deterministic variability associated with LUs and smaller scale, unexplained variability.

Weather

Between 1 Oct. 2009 and 30 July 2010, we measured 593 mm of precipitation, making it a relatively wet year. Daily precipitation totals, in millimeters of water, are partitioned into daily totals of snow and rain in Fig. 6, and it is clear that a large majority of the precipitation was snow, comprising a total of about 490 mm of water equivalent or 82.6% of the measured precipitation. Ignoring the unusual October snow, which melted rapidly, very little snow fell in the watershed prior to DTS data collection initiation on 24 November. Subsequent snowfalls in December added a small amount, but it wasn’t until mid-January that repeated snowfall events built a typical, spatially continuous snow pack. From 18 to 31 January, 119 mm of snow water equivalent fell, which is equivalent to about 70- or 80-cm snow depth if evenly distributed. Numerous events, including notable late season snows on around 1 April, 27 April, and 22 May contributed to the seasonal snowfall total.

Maximum, minimum, and average daily Ta for the DTS measurement period are shown in Fig. 7. There was a pronounced cold period the first 2 wk of December. During the next 4 mo, Ta fluctuated between 10 and −10°C, with relatively cold periods near the end of December, 7 January, 20 February, 9 March, and 9 April. After mid-April, Ta steadily increased to a maxima of approximately 30°C.

Profile Soil Temperature

West-Facing Site.

There was virtually no snow accumulation in November or early December at the WF site (Fig. 8A). Soil temperature dropped sharply in early December, freezing to a depth of 10 cm on 3 December and 30 cm on 6 December in response to the first pronounced cold period. The minimum 10-cm Ts was −5.8°C. In addition, a consistent, upward Ts gradient was established. Soil temperatures rose around 10 December but remained below 0°C owing to the snow cover, which was too thin to effectively insulate the soil but sufficient to provide a temperature buffer at the melting point. This was followed by a second decrease at the end of December in response to the second cold period. Note that the snow depth data can be very noisy in this sloping, shrub-covered, windy environment despite internal noise correction algorithms because the sonic signal used to determine distance to the surface is displaced during wind events, moving the effective measurement location up- or downslope. This problem is especially evident with shallow snow cover. The snowy period starting on 18 January resulted in a snow accumulation of 25 to 30 cm, which was maintained until 27 March. During that time, Ts at all depths gradually approached 0°C and was unaffected by either seasonal or diurnal temperature fluctuations. Following the initial melt-out (here we define melt-out as the effective absence of snow cover as indicated by snow depth and Ts), snow events at the end of March and on 27 April and 22 May caused sharp but temporary Ts declines and dampening of diurnal Ts fluctuations. After the last snowfall, a downward Ts gradient was established and Ts consistently increased with large diurnal fluctuations.

North-Facing Site.

The general pattern of Ts and snow accumulation and melt (Fig. 8B) were similar to those at the WF site, but the timing and amounts were quite different. At the outset, the NF Ts was somewhat cooler and early snow accumulation somewhat greater. The Ts response during the initial cold period was somewhat less than at the WF site, with a minimum Ts at 10 cm of −2.8°C and the 30-cm depth barely crossing the freezing point. The second cold period in December had little effect on Ts and only at the 10-cm depth, and snow depth increased to about 20 cm. Snow depth after the mid-January snowy period was almost twice that at the WF site. As at WF, once the snow depth exceeded about 15 cm, Ts at all depths trended slowly toward 0°C with no diurnal temperature fluctuations and with an upward gradient. The first melt-out was on 22 April, a month after melt-out at the WF site. There was a slight dip in Ts, most evident at 90 cm, in response to the convective flux of 0°C snowmelt water into the soil. After snowmelt, Ts increased steadily except in response to snowfalls around 27 April and 22 May.

Drifted North-Facing Site.

We have no continuous snow depth data for the DNF site, but on snow surveys conducted on 12 February and 24 March, the measured snow depths were 200 and >300 cm, respectively (Fig. 8C). We estimated the accumulation and melt of snow at the DNF site based on timing of storms, measured values, and melt-out data. The Ts at DNF was similar to that at NF during the early and mid-December freezing events, indicating similar snow depths at that time. Shortly thereafter, there was a large accumulation of snow, and Ts at all depths gradually trended toward 0°C with no diurnal fluctuations and a consistent upward gradient. The Ts dip between 17 and 25 May resulted from the large convective flux of cold meltwater. On 25 May, Ts increased rapidly, starting near the surface, signaling the end of snow cover at the DNF site. The melt-out date at DNF was 1 mo after the NF site melt-out and 2 mo after the WF site melt-out. The Ts increases following melt-out were more rapid at DNF than at the other sites.

Distributed Temperature Sensing Data

The DTS data are summarized in Fig. 9 where the x axis is time, from the beginning of data collection until cable destruction by cattle, the y axis is distance along the cable from the DTS instrument in the weir house to the joining weld at the end of Fiber 1, and colors depict DTS-measured temperature (corrected), ranging from −10°C (dark blue) to 45°C (black). The white areas are missing data, mostly due to operator error. The cable segments corresponding to functional landscape units of interest are identified to the right. The corrugated pattern evident in late spring and summer throughout the watershed is the result of diurnal temperature fluctuations. The DTS data are described in terms of the following four time periods that correspond to distinctive weather and snow conditions as described above: (i) early winter with shallow snow cover; (ii) midwinter with deeper snow cover; (iii) spring melt; and (iv) summer. For each time period, DTS transect data representing a diel temperature sequence using data collected near 0000, 0600, 1200, and 1800 h for a single day are presented to further illustrate temperature variations both within and between the different LUs (Fig. 10).

Early Winter.

In early winter (24 Nov. 2009–18 Jan. 2010), Ts was mostly near 0°C, with two dark blue, vertical bands reflecting the relatively cold weather in early and late December (Fig. 9). Diurnal temperature fluctuations, which are clearly evident in the weir house (WH) and ridge coil (RC) segments, are considerably damped in the WF segment and barely discernable in the DNF and DR segments, with the NF segment intermediate in this respect. The Ts fluctuations do not exceed 0°C, suggesting the presence of a shallow snow cover. Horizontal, lighter blue “stripes” show that some points along the transect were consistently warm relative to others. In DNF and DR, just before the ridge coil, Ts is slightly warmer, indicating incipient drift formation. These trends are further illustrated with a representative DTS diel sequence profile obtained on 6 December (Fig. 10A), when Ta ranged from −11.6 to −15.9°C. Relatively large diurnal temperature fluctuations are evident in the weir house coil (−9 to −13°C) and exposed ridge coil (−13 to −16°C). The Ts data for the WF segment were about 5°C warmer than the weir house coil and had diurnal temperature fluctuations of a similar magnitude. The NF segment varied about 2°C and was 2 to 4°C warmer than the WF segment, with virtually no diurnal temperature fluctuation. Within the DR segment, temperatures tended to warm to 0°C, also consistent with incipient drift formation. Temperatures plummeted where the cable left the drift zone and passed across the ridge to the exposed coil.

Midwinter.

The beginning of the midwinter period (18 January–23 March) is marked by substantial depositional snow events and an overall homogenization of Ts, which is expressed as a nearly uniform blue block during that time in Fig. 9. By mid-January, snow accumulations were such that no diurnal fluctuations are discernable and Ts is very close to 0°C throughout the measurement domain. Diurnal temperature fluctuations reflecting atmospheric conditions are apparent only in the weir house and at the exposed ridge coil. These tended to persist with time, as indicated by relatively light or dark horizontal “streaks” in Fig. 9. This period is illustrated with data from 10 February described above (Fig. 10B). Viewed in the context of Fig. 9, it is clear that some of the very small temperature changes along the cable represent consistent spatial temperature patterns.

Spring Melt.

The spring melt period (23 March–16 June) was characterized by spatially discontinuous and temporally intermittent snow cover. The first soil exposure occurred between 9 and 23 March when the DTS did not collect data. Weak diurnal fluctuations are visible on 23 March for most of the WF segment. For the remainder of the melt period, WF Ts tended to exhibit distinctive diurnal fluctuations with a gradually warming trend. An obvious exception resulted from the snowfall at the end of April, when Ts returned to near 0°C, expressed as a blue vertical “band” (see also Fig. 8). Daily maximum Ts values of >20°C were common by the end of the period in the WF segment. Part of the NF segment was snow free around 20 April, and the entire segment was snow free when the 28 to 29 April snowfall occurred. Maximum daily Ts values of 17°C were common at the end of the period. The DNF began to be snow free in early May, and melt-out was complete on 16 June. The DR segment melted much earlier and was considerably warmer than the DNF segment. Figure 10C shows conditions along the cable on 28 March, when the WF segment was partly snow free. Air temperature ranged from 9.2 to 1.7°C that day. Weir house coil temperatures exhibited a similar range. Most of the WF segment exhibited a wide temperature range, from about 9°C to nearly 0°C at some points. The coolest temperatures were at 0600 h, while the warmest were at 1600 h. The NF and DNF segments remained very close to 0°C, with no diurnal fluctuation indicating persistent snow cover. The exposed ridge coil also showed considerable diurnal temperature change.

Summer.

In the summer period (16 June–22 July), there was no snow along the cable transect and diurnal Ts fluctuations were evident throughout as Ts generally increased. The WF segment was considerably warmer than the NF and DNF segments, at times exceeding 40°C, while the NF segment rarely exceeded 28°C. The difference between the NF and DNF segments was reduced to a negligible amount as the DNF segment warmed rapidly after the snow melted. The DR segment was much warmer than the DNF segment and similar to the WF segment. These trends are seen in more detail on 16 July (Fig. 10D), which had daily maximum and minimum Ta values of 30.4 and 19.4°C, respectively. The PT100 data, which are available for this date, were in close agreement with the DTS data. Minimum Ts was at 0700 h for all segments except the ridge coil, which probably received early morning sunlight. The NF and DNF Ts values were about the same at that time (about 15°C), while the WF and DR Ts values were similar at about 20°C. Temperatures diverged considerably by 1300 h, so that the WF segment averaged around 33°C with a high degree of variability and the NF segment averaged about 17°C. The 1900 h Ts values are even warmer for the WF segment, averaging around 40°C (10°C warmer than the max Ta), while the 1900 h NF and drift Ts values are about the same as at 1300 h. The timing difference is due to the western orientation of the WF segment, which received more late afternoon sunlight.

Watershed-Scale Soil Temperature Variability

The profile data demonstrate that Ts dynamics during the measurement period were very different at the three profile locations. From 1 yr of data it’s not clear if those differences are controlled by landscape position, and hence potentially predictable and described as deterministic, or if they vary randomly from year to year or from point to point. It turns out that, in this case, they are temporally persistent. The 9-yr mean annual soil temperatures for the three sites are: WF = 9.6°C, NF = 5.7°C, and DNF = 4.6°C, all significantly different (α = 0.05). Note that Ta is essentially equal across all sites so that differences are due to local, relatively small-scale processes. The question of spatial variability remains because those data give no indication of the extent to which the three sites are representative of the LUs in which they are located. It is possible that those differences are due to random, small-scale, or unknown effects. A unique aspect of DTS is the capability of measuring Ts at high spatiotemporal resolution across the landscape.

Here we describe the overall Ts variability and how it is partitioned among the different LUs with time. Discussion is mostly limited to the three longest cable segments that correspond to the most extensive LUs in the watershed: the WF, NF, and DNF segments. Only those segments have sufficient length (replicate measurements) to effectively describe within-segment variability. As with the calibrations, we used data from every third meter to assure that each measured value was independent. The overall watershed Ts is represented by the aggregate of the three segments. The daily average Ts and standard deviation for each segment and the overall watershed are shown in Fig. 11. Daily average rather than hourly data are used because the hourly data exhibit large diurnal temperature fluctuations that obscure differences among segments.

In general, there was a remarkable degree of similarity between the average Ts dynamics of the cable segments (Fig. 11) and the corresponding representative profiles (Fig. 8). As with the profile data, the average Ts for the WF segment was lower than that of the other segments during the two December freezing periods, about −6°C during the first and −2°C during the second. During the winter snow-covered period, Ts was essentially identical for all segments, again similar to the profile data. Differences among the segments became apparent during melt-out, first on the WF segment, followed by the NF and DNF segments. Snowmelt timing was also similar for the WF and NF segments and profiles, while the DNF profile became snow free on 20 May, about 10 d after the DTS segment Ts rose on 10 May. This discrepancy can be attributed to the long melt-out phase in the DNF segment, which results in a wide range of melt-out dates (Fig. 9). During the summer period, the WF segment diverged from the others, warming more rapidly to temperatures as much as 10°C greater than those in the other two segments. Ultimately, Ts at the three profiles was very similar to that for the corresponding DTS segment.

Within-segment variability, as measured by the SD (Fig. 11B), approached 1°C for the WF and NF segments during the freezing events of December. This was followed by an approximately 2-mo period with a high degree of Ts uniformity, with both within-segment and overall SDs approaching 0°C for all segments. Within-segment variability spiked during periods of melt-out for all segments. Thus, the WF segment had two peaks (late March and mid-April) separated by a brief snow-covered period. Within-segment variability peaked for the NF and DNF segments as they melted out. During the snow-free summer months, within-segment variability rose gradually to SD values of between 1 and 1.5°C.

In general, peaks of within-segment variability correspond to periods of shallow (<15 cm) snow cover due to the high sensitivity of Ts to snow depth in that depth range. This effect was less dramatic in the winter, when variable snow depth resulted in variable insulation, than it was in spring. During spring melt-out, virtually any snow cover effectively limits Ts to near 0°C during warm weather due to the reflection of solar radiation and infiltration of 0°C meltwater into the soil. In the absence of those two processes, the soil warmed rapidly. Thus, even under conditions of highly uniform snow cover and solar radiation, Ts may vary dramatically across relatively short distances during melt-out.

Peaks in the overall variability (Fig. 11B) were due to a combination of within-segment and between-segment variability. The large peak (SD = 5.2°C) in overall variability centered around 19 April resulted from both high within-segment variability as the WF segment melted out and between-segment variability as the WF segment warmed to >10°C while the other segments remained at 0°C. The second major peak in overall variability, of similar magnitude, occurred near the end of July and was mostly attributable to between-segment variability as the WF segment diverged from the other two. During the summer, the within-segment 95% confidence interval for the three segments during summer was always <0.25°C, and there was a significant (α = 0.05) difference between the NF and WF segments, confirming the differences visible in Fig. 9.

Diurnal Variations

The trends of average daily Ts described above do not include the diurnal fluctuations, which became pronounced during the summer period (Fig. 9) when the overall variability was driven by between-segment differences. Figure 12 displays hourly segment average and standard deviations from a 10-d period in July to better elucidate summer time Ts variability. The WF Ts values (Fig. 12A) were much greater than those from either NF or DNF, which were similar, and the diurnal Ts fluctuations in the WF segment, which ranged from about 15 to 40°C, were much greater than those of the other two segments, which ranged from about 12 to 22°C. The average daily maximum Ta during the period was 24.3°C, which the WF segment exceeded by >10°C each day but the NF and DNF segments did not achieve. Similarly, the average daily minimum Ta of 14.8°C was less than the WF daily minima for 9 of 10 d, while daily minima for the DNF and NF segments was usually <14.8°C. Differences between the WF and NF segments were about 5°C at sunrise and increased to about 17°C in the late afternoon.

Within-segment variability was similar among segments and relatively low (SD ~ 1°C) in the early morning hours for all segments (Fig. 12B). During the day, variability increased for all segments and was most pronounced in the WF segment. This appears to be the result of differential heating due to relatively small topographic or plant cover variations that affect the amount and timing of incoming solar radiation. Overall variability was highest during the late afternoon (SD ~ 9), when differences among segments was most pronounced. During the night, temperatures throughout the watershed dropped rapidly, with warmer Ts values dropping more than cooler values, so that overall Ts variability returned to a minimum around sunrise (Fig. 12).

Impact of Landscape Units

These data demonstrate the strong control of topographically and vegetationally defined LUs on Ts. To put the observed differences in a larger context, we consider historical data collected at two sites in the RCEW, Flats and Reynolds Mountain (for data description, see Seyfried et al., 2001). The Flats is at relatively low elevation (1186 m asl), while Reynolds Mountain is at relatively high elevation (2097 m asl), both on nearly level terrain with similar density shrub cover so that Ts differences can be mostly attributed to the nearly 1000-m elevation difference. The mean annual soil temperature at Flats of 11.5°C is 4.4°C greater than that at Reynolds Mountain. Thus, the 5.2°C Ts range associated with LUs within 300 m at the same elevation is greater than that associated with a 911-m elevation difference. An elevation band or Ta basis for Ts simulation would miss most of the variability within the watershed and produce misleading results.

We assume that the first-order driver for energy balance variation in complex terrain is incoming solar radiation, which is controlled by slope, aspect, latitude, and season. In general, the difference in solar radiation between slopes of contrasting aspect is greatest in the winter months and least during the summer months, which are nearly aspect neutral (Tian et al., 2001). This is the opposite of the Ts trends we observed. For Ts, solar inputs are modulated by surface conditions. In winter, when the solar radiation contrast is greatest, Ts was virtually homogeneous in Upper Sheep Creek due to the high albedo and insulating properties of snow. The main topographic effect of solar radiation on Ts was indirect in that it affected the rate of snowmelt. After melt out, Ts is modulated by the live and dead vegetative cover. The vegetation on the shallow soils of the WF segment received less effective precipitation (due to snow redistribution), had lower LAI, less litter cover, and more exposed soil (Table 1) than the NF or DNF segments. For this reason, WF Ts values exceeded Ta, while the shaded Ts segment stayed below Ta. Also note that the R segment, which is in the low sagebrush LU, is on gentle north-facing slopes and has Ts maxima and diurnal fluctuations of similar magnitude to the WF segment (Fig. 9 and 10). The differences in diurnal Ts fluctuations between the WF and NF segments in summer are qualitatively similar to those documented by Balisky and Burton (1993) in experiments testing the impact of vegetative cover on Ts.

Although there are a limited number of studies of Ts in complex terrain, four recent studies illustrate how variable cover and topography produce strongly contrasting Ts environments. Working in northeastern Arizona, Burnett et al. (2008) also measured a mean annual soil temperature difference on contrasting topographically controlled LUs of up to 5.6°C. In their case, snow cover played a small role and vegetative cover was sparse, so that the difference between LUs was more directly controlled by solar radiation. They therefore found that the greatest differences (about 8°C) between north- and south-facing LUs were in winter. Gutiérrez-Jurado et al. (2013), working in central New Mexico, measured very similar annual Ts patterns on LUs on contrasting north- and south-facing aspects. They also found that Ts was nearly identical on both slopes in summer and that the south-facing LU was about 8°C warmer in the winter for soils that were not under a plant canopy. The Ts under tree canopy on the north-facing LU was consistently about 4°C cooler than the south-facing LU, demonstrating the importance of vegetative cover. Finally, Ebel et al. (2012) found that summer Ts in the Colorado Rocky Mountains was about 5°C cooler on a north-facing LU than an adjacent south-facing LU under native forest conditions, but following a fire and vegetation removal from both LUs, the summer Ts was the same on both LUs. In this case, it appears that the much denser forest cover of the north-facing LU moderated the Ts and, after vegetation removal by fire, the small difference in solar radiation during summer resulted in nearly equal Ts values for the two LUs. In related research. Ebel et al. (2012) also found that snowmelt on south-facing slopes was much earlier and resulted in early-season soil warming relative to north-facing slopes.

Spatial Variability

As point-scale critical zone process models are extended to spatial domains, one of the fundamental issues to be addressed is how the target landscape should be discretized. We have shown that topographically and vegetationally defined LUs have substantially different Ts regimes in the same climate. The DTS data further demonstrate the potential improvement in Ts precision when the landscape is discretized by LU. Using the overall variability data as a surrogate for variability within the watershed as a whole, it is clear that, at a scale of 0.26 km2, Ts is highly variable except during snow-covered periods. The overall watershed SD was as high as 9°C so that a ±1 SD Ts estimation range encompasses virtually the full range of biological response. Such imprecise estimates have little utility for either testing Ts models or estimating soil biological activity. On the other hand, most Ts variability was deterministic in the sense that it is associated with specific identified LUs associated with topography and vegetative cover. Variability within LUs, represented by the corresponding within-segment variability, was much lower than for the watershed as a whole, usually with a SD of <1.5°C. Therefore, discretization at the level of the LUs described or the grid resolution required to delineate them can result in reasonably precise estimation of Ts in this terrain. On the other hand, elevation band, Ta, or small watershed delineations cannot be precise because they include multiple LUs.

It is possible that a further increase in spatial resolution may further improve Ts precision. The within-segment Ts variability can be attributed to a combination of instrument precision, cable placement (i.e., depth) variability, and variations in surface conditions due to microtopography and vegetation. The high precision of Ts data during snow cover indicate that instrument precision is not a large contributor to the overall variability. Some amount of cable placement variability is inevitable when installing cable in hand-trenched slots in uneven, sometimes rocky, terrain. Variations in cable depth will cause differences in temporal variability at a point and spatial variability in a segment and be most pronounced when the vertical Ts gradient is greatest. These effects are also probably relatively small because large changes in within-segment variability occur when there is little change in the vertical Ts gradient. This reasoning indicates that small-scale (<10 m) surface variations are probably the primary cause of within-segment variability during summer. At this point, the level of vertical and horizontal spatial resolution required to describe Ts melt-out dynamics is not known and is treated as stochastic.

Summary

Selker (2008) introduced DTS as a means of “taking the temperature of ecological systems.” We applied that vision to the upland portion of a remote headwater catchment with the objectives of evaluating the applicability of DTS to the measurement of Ts in remote field environments, describing the spatial and temporal variability of Ts in complex terrain influenced by a variable snow cover, and assessing the implications of that variability for modeling Ts and Ts–dependent processes. We found that DTS can provide useful and accurate data as a field instrument at a remote site. More specifically, we found that: (i) measurement accuracy on the order of other field temperature sensors (±0.4°C) can be obtained from periodic calibration of a DTS instrument exposed to seasonal temperature variations of 50°C; (ii) cable calibration, which is more difficult (in our environment) can be effectively post-processed if necessary, (iii) instrument resolution of temperature change is on the order of 0.05°C; and (iv) the data collected are well suited to describing the spatial variability of Ts across a landscape. The measurement precision we obtained would have been improved by using longer measurement times, but it’s important to consider the rate of Ts change. In the summer, Ts may change at rates approaching 2°C h−1 or 0.033°C min−1, which argues against measurement times much longer than 1 min.

We also encountered some difficulties with the instrument that may limit its application. The biggest problem is that the measurement of Ts requires digging and refilling a minimum-impact trench. In rugged terrain with no vehicle access, this hard work can be very time consuming, especially if there is interest in a multiple-kilometer study. A second problem is that cable on the soil surface is subject to destruction by animals, so that additional protection may be required. This issue has been noted in other environments (Lutz et al., 2012).

In general, Ts in the 0.26-km2 watershed was highly variable, with LU-averaged differences >16°C at times, rendering an overall average of questionable value. The use of DTS provided a unique view into the nature of Ts variability. We found that Ts variability at Upper Sheep Creek is deterministic in the sense that there are strong spatial and temporal controls on that variability. In the snow-covered months, Ts throughout the watershed was nearly uniform and close to 0°C. Otherwise, much of the Ts variability can be attributed to independently defined LUs that are determined by co-varying snow, soil, and vegetation patterns. Thus, we found that Ts variability within LUs was generally small (SD < 1) except during melt out, while the overall variability was much greater (SD = 2–5).

The measured Ts differences among LUs were large relative to other factors controlling the spatial variability of Ts on the landscape. The mean annual soil temperature range among LUs in the watershed is of similar magnitude to that associated with an elevation difference of 910 m in the RCEW. This high degree of Ts variability related to LUs has been observed in other locations, although not the specifics of how Ts values differ due to the surface conditions.

These findings argue against the use of Ta, elevation bands, or small watersheds as a basis for discretizing the landscape because those groupings inevitably lump contrasting LUs, resulting in a degree of variability that is high relative to the effects on Ts on biological processes and high relative to the overall landscape Ts variability. On the other hand, Ts within easily definable LUs is relatively homogeneous. It remains to be seen how well the observed differences can be predicted based on the processes involved.

We thank Pat Kormos and Julie Finzel for their tireless work installing cable and Steve Van Vactor for his work managing and manipulating the DTS data.

This is an open access article distributed under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).