We designed a system composed of digital temperature sensors mounted on an array of cascaded printed circuit boards (PCBs) connected to a custom-designed low-power logger. The sequentially addressable digital temperature sensors (TMP117AIDRVR) are low cost, low power, and high accuracy, with a resolution of 0.0078125 ∘C and a factory-assured, National Institute of Standards and Technology-traceable (NIST) accuracy of ± 0.1 ∘C across a temperature range of -20 to 50 ∘C (http://www.ti.com/lit/ds/symlink/tmp117.pdf, last access: 25 February 2022). All sensors on the probe are connected to the data logger's two-wire interface (TWI, also known as I2C), and each sensor is accompanied by a discrete D flip-flop. The arrangement of these flip-flops constitutes a shift register that propagates an address bit along the probe, sequentially enabling each sensor. This approach enables a readout of an arbitrary number of sensors with just six signals (3.3 V supply, ground, TWI clock, TWI data, address, and address clock). The board-to-board connections between probe sections rely on custom-designed press-fit PCB connectors to ensure lasting structural stability and electrical contact under mechanical and thermal stress in the field. Once assembled, each PCB section is 20 cm long and contains two or four temperature sensors to enable 10 or 5 cm spacing, respectively. The upper PCB section is connected to a press-fit wire-to-board adaptor to link the entire probe assembly to the logger. The electrical design of the boards minimizes capacitive loading and crosstalk of the communication signals. In combination with a TCA9803 TWI bus buffer, this allows for sensor arrays over 2 m long, without affecting signal integrity. The entire temperature probe is powered down in between measurements, resulting in a 0.0 µA idle current and a reduced impact of electrical failures along the probe. A measurement of 16 sensors along the probe takes 100 ms and requires up to 220 µA per TMP117AIDRVR.

The logger is a custom-designed embedded system built around a low-power wireless system on a chip (NRF52832 ARM Cortex M4) that enables Bluetooth Low Energy (BLE) connectivity. On-board provisions include a TCA9803 TWI buffer, a load switch, a TMP117AIDRVR temperature sensor, 32 MB of low-power NOR flash memory for storing measurement data, a temperature-compensated real-time clock (RTC, i.e., PCF2129AT) for accurate timekeeping and generating watchdog and measurement interrupts, multiple connectors for existing and future sensor expansion, and an RFM95W LoRa modem for future integration in LoRa wireless sensor networks (Wielandt and Dafflon, 2020). The system operates in the 1.8–3.6 V range, allowing for operation with two AA batteries without requiring further power supply circuitry. The microcontroller and its peripherals are mostly asleep, drawing a system idle current of 7.085 µA. Taking regular BLE advertising and a 15 min sensor-measurement interval for 16 sensors into account, the total system's current consumption averages at 22 µA. Using Energizer L91 batteries (https://data.energizer.com/pdfs/l91.pdf, last access: 25 February 2022) with a 3500 mA h capacity, a total battery lifetime of 18 years can be reached in theory. With each measurement taking up 6 B + 2 B per sensor, the above described probe with 16 sensors would have sufficient memory for 3 years of measurements.

Logger parameters (measurement interval, time, etc.) and on-board stored data are managed using BLE connectivity and a custom companion app for Android devices. The app provides a list of nearby probe identifiers ranked by their Bluetooth signal strength, which usually correlates to the distance from the Android device. The app allows a user to erase logger memory, reset the system, synchronize the on-board clock, set a logging interval, transfer data, and assign GPS coordinates through the phone's GPS. Current data transfer speed is ∼60 kb s-1, which means that 2 weeks of data are downloaded every second (assuming a 15 min sampling interval and 16 connected sensors). The transferred data are converted into a .csv format.

The probe is built by cascading sensor boards to the desired length and inserting the sensor assembly into a tube, which is then further filled with a sealing urethane mixture and connected to a logger and its enclosure. Different types of tubes and connections to the logger can be used, based on the application. The default tube is a 38 in. (∼10 mm) outer diameter (OD) and 14 in. (∼6 mm) inner diameter (ID) white-colored cellulose acetate butyrate (CAB) plastic tube that is flexible, UV resistant, high albedo, and structurally stable in cold and warm temperatures. Alternatives include a 38 in. (∼10 mm) OD and 14 in. (∼6 mm) ID 304 stainless-steel tube or a 12 (∼12 mm) OD and 14 in. (∼6 mm) ID CAB tube. A cable gland is tightened and – if necessary – glued on the top of each tube before inserting the sensor assembly. Then, the tube is filled from the bottom with a urethane blend using a syringe to reduce the chance of air bubbles. The urethane blend (20-2360 from Epoxies Inc., https://www.epoxies.com/_resources/common/bulletins/20-2360R.pdf, last access:25 February 2022) is a thermosetting mixture designed for electrical potting applications over a temperature range of -40 to 125 ∘C and has a measured thermal conductivity of 0.191 W m-1 K-1. Its coefficient of thermal expansion (2.28×10-4) and high tensile strength (400 psi; pounds per square inch; 2.758×106 Pa) limit the risk for the probe to warp or snap under a large range of thermal or physical conditions. A 38 in. (∼10 mm) OD metal spike is added at the bottom of the tube to act as a stopper while the urethane mixture sets, ease ground entry during deployment, and enable some electrical grounding with the use of a grounding wire attached to the bottom of the sensor assembly. A 4 oz (∼120 mL) polypropylene (PP) jar can be mounted to the top of the probe and serves as a UV-resistant and dust- and splash-proof enclosure for the data logger (Fig. 1). Depending on the application, sealant can be applied on the jar seams to achieve long-term waterproofing and submergibility.

The cost of materials for the default DTP system, including the logger, can be as low as USD 95 for a 1.2 m long probe with 16 temperature sensors, assuming a batch size of ∼300 probes. The cost is distributed between the logger components and manufacturing (USD 19), the batteries (USD 2), the CAB tube (USD 4), the logger enclosure (USD 1), the urethane mixture (USD 2), the cable glands (USD 2), and the sensor boards (USD 65) which include the cost of USD 2.50 per TMP117 sensor. A 304 stainless-steel probe implies an additional USD 20, distributed between the stainless-steel tube (USD 10) and the brass tube fitting (USD 10). The above-mentioned cost of the logger and sensor boards is only obtained under optimal factory yields and strongly depends on choices and fluctuations in component and PCB manufacturing prices. In sub-optimal conditions, additional costs can easily add up to ∼ USD 120 per probe. In addition, these price estimates do not consider the cost of the mechanical assembly of the various sensor boards and logger into their final housing. Under ideal conditions (i.e., the assembling of a large batch with adequate equipment), a person can assemble a probe in less than 15 min.

The field deployment of the DTP system can be performed in various ways, depending on the probe housing and application. For plastic probes, a custom-length drill bit with the same diameter of the probe is used to drill a guide hole in which the probe is then inserted. The probe can be inserted completely into the ground, or part of it can be left above ground (Fig. 1). Aboveground installation of the DTP system for snow temperature measurement is done by attaching the probe to a PVC or wood stake using low-temperature-resistant zip ties with a 1 cm plastic tube spacer attached at mid-distance between consecutive temperature sensors.

A procedure was developed to evaluate the accuracy of a sensor marketed with a factory-assured NIST-traceable accuracy of ±0.1 ∘C and possibly to improve its accuracy with an additional calibration procedure. The most common method for calibrating a temperature sensor consists of a single point calibration where a sensor is submerged in an ice bath, made by saturating 2–3 mm particles of shaved or crushed ice in distilled water and allowing the mixture to equilibrate (Mangum, 1995). If carefully prepared, the latent heat of fusion, which is needed for the phase change, stabilizes the bath within a few ten-thousandths (∼0.01) of 0 ∘C (Thomas, 1938). Cable et al. (2016) used this calibration approach to increase the accuracy of thermistors from 0.1 to approximately 0.02 ∘C for subsurface temperature measurements. While temperature-controlled water baths at temperatures above 0 ∘C (Aguilar et al., 2018; Naranjo and Turcotte, 2015) can be used for calibration using a reference thermometer, reaching an accuracy of 0.01 ∘C is challenging.

While the standard ice-bath approach is adapted for calibration of individual sensors or a string of sensors, initial tests performed in this study did not provide satisfactory results when scaling up this approach to submerge an entire 1.2 m long DTP system. Initial tests were conducted by filling a 1.5 m long 25 cm diameter pipe with a mixture of cool distilled water and cold crushed ice. The DTP system was centered in the pipe with a Fluke reference thermometer (Fluke 1524) collocated next to one of the DTP sensors for additional comparison. Results have shown that building a fine mixed water–ice bath at that scale was time-consuming, not always successful because of the difficulty of having a well homogenized mixture in such a large volume, and thus not adequate for calibrating hundreds of DTP systems.

In this study, a novel 0 ∘C point calibration approach was developed to calibrate tens of probes in one single run while achieving accuracy similar to the ice-bath method. Our approach includes cycling through water–ice phase changes in a 1.5 in. diameter tube filled with cold distilled water and with the DTP probe suspended at its center and logging every minute. Multiple probes and tubes are placed into an incubator (Thermo Scientific Precision Incubator) at -5 ∘C over a period of 12–24 h to ensure frozen conditions and then moved into an incubator at 3 ∘C until melting is complete. The average offset of the measured temperature from 0 ∘C occurring at the melting point due to the latent heat of fusion (heat-induced zero curtain) is extracted from the data for each individual sensor.

Numerical experiments were performed to evaluate the impact of various DTP characteristics and environmental factors on DTP measurement accuracy, beyond the sole sensor accuracy. In particular, we investigated how the temperature measurements are potentially impacted by the probe tubing material and diameter, different ground and probe surface heating, the air gap between the probe and soil, sensor positioning error, and variable soil thermal diffusivity.

A finite volume numerical model on an axis-symmetric cylindrical grid was developed to simulate heat transfer through conduction, in and between the probe and the soil. The thermal conductivity and heat capacity were explicitly represented in the model. Temperature in each cell was updated in time by summing the contributing heat flow across each cell boundary, and stability was controlled by heuristically reducing the time step to between 10-4 and 10-1 s. The model spanned across 50 cm and 50 cells vertically and 5 cm and 100 cells radially and was parameterized with the thermal conductivity and heat capacity of the probe and soil. The initial conditions and the moving Dirichlet boundary conditions at the top and bottom were calculated using the analytical solution for diurnal heat transfer in the half plane (Turcotte and Schubert, 2002). Boundary conditions at the outside edges were similarly obtained with the analytical solution using the ghost point method. The internal boundary condition at zero radius was treated as a zero-flux Neumann boundary condition for symmetry (Langtangen and Linge, 2017, p. 251). Validation of the numerical model was carried out by applying a naive T=0 Dirichlet boundary condition at the outside and bottom of a homogeneous domain and a sinusoidal forcing function at the upper surface. The simulated temperatures closely matched the analytical solution. For the numerical experiments, the simulated temperatures inside the probe using the finite volume model were compared to the analytical solution. The differences were evaluated through the percentage mismatch and time delay in amplitude.

Vertically resolved measurements of soil and snow temperature can be used to infer various properties, including soil thermal parameters, snow depth, zero-curtain duration, first bare-ground date, frozen- and thawed-layer thickness, and many empirical indices. In this study, we evaluate the value of the DTP system to autonomously estimate snow depth, soil frozen, and thawed-layer thickness, as well as the possible probe upward displacement relative to soil surface that can occur in frost-susceptible soil using acquired temperature measurements.

The developed sensor accuracy assessment approach was validated by repeating the approach several times with one DTP system and then applying the approach on 846 sensors from 70 probes (Fig. 2). The zero curtain induced by the water phase change is consistently observed around 0 ∘C, with offsets that are always smaller than the ±0.1 ∘C factory-assured accuracy. Repeating the calibration cycle three times with the same probes shows that the offset of each sensor across the calibration cycles varies over a range of 0.015 ∘C. The offsets of 846 sensors indicate a relatively Gaussian distribution of offsets with a mean of +0.02433 ∘C, a standard deviation of 0.02095 ∘C, and a 95th percentile interval between -0.022 and 0.062 ∘C.

The results of the sensor accuracy assessment indicate that the sensor accuracy can be improved using the developed approach from a ±0.1 ∘C factory-assured accuracy to about ±0.015 ∘C. Note that the maximum offset measured on 846 sensors was +0.07 ∘C, which indicates the already high accuracy of the factory calibration for the tested sensors. The only observed caveat for the calibration approach is that the top sensor along the probe does not always show the clear zero curtain needed for precise calibration because it is not consistently covered with ice. This issue results from the need to leave sufficient air space, with some safety margin, between the water surface and the bottom of the logger enclosure, in order to account for ice extension, which is needed to avoid ice pushing on the logger enclosure directly and breaking the probe.

The effect of different probe characteristics and environmental factors on the measured temperature accuracy is quantified using numerical experiments. Accuracy is evaluated through the relative difference and time delay in diurnal amplitude between soil temperatures simulated with and without the numerical representation of the probe characteristics (Table 1 and Fig. 3). The maximum percentage error between the hypothetically measured and the true temperature at all times is also considered, through their absolute difference divided by the maximum amplitude at the same depth. Note that sensor accuracy is not considered in these numerical experiments and thus needs to be added to calculate the maximum total error.

The simulated response of a DTP system in its standard plastic housing (10 mm (∼ 38 in.) OD plastic tubing filled with urethane blend) indicates a maximum difference of -0.11 % between the hypothetically measured and the true amplitude and a time delay ranging between 60 and 95 s for a soil diffusivity between 0.15 and 1 mm2 s-1 and assuming an absence of an air gap between the probe and the soil. The maximum difference in amplitude is as small as 0.01 % with a soil diffusivity of 0.5 mm2 s-1. The temperature maximum error at any time and for the full range of soil diffusivity is ∼0.7 %.

Increasing the diameter of the plastic probe to 15 mm OD, or considering a hypothetical 1 mm air gap, produces a time delay of up to 160 s in a soil with a diffusivity of 0.5 mm2 s-1. The amplitude and measurement error can be as high as -0.03 % and ∼1.2 %, respectively. While the error is almost double the standard case, it is acceptable for many applications. Still, results show that further increasing the probe diameter or the air gap increases inaccuracies significantly. For example, the presence of an extreme hypothetical 5 mm air gap produces a time delay of up to 505 s (Fig. 3).

Different surface heating between the soil and aboveground probe surface, though difficult to assess because of the complexity of the surface energy exchanges, primarily influences the surficial soil (top 5 cm) temperature measurements. A temperature difference between the probe and soil surface equal to half the diurnal variation can create an amplitude difference of ∼6 % at 1 cm depth, decreasing to less than 0.03 % at a depth deeper than 5 cm. Such different heating or cooling responses between the soil and probe surfaces can result from different surface emissivity, insolation, near-surface wind, and water phase changes in the soil. Deployment of probes at locations where environmental factors may strengthen this source of error could benefit from burying the probe and the logger separately and using thin diameter plastic probes.

The use of stainless-steel housing may cause a slightly reduced accuracy in soil temperature measurement compared to the standard plastic 10 mm (∼38 in.) OD tubing. Stainless steel, which facilitates vertical heat transfer along the probe, results in a maximum negative time delay of 235 s and a max error of 0.1 % between the measured and true amplitude. The stainless-steel standard probe setting increases the potential for an overestimation of the in situ amplitude in the top 15 cm and then an underestimation similar to the plastic standard case. Though stainless-steel tubing limits the accuracy in the top part of the soil, overall it can provide a tighter contact with soil because the stainless-steel probe can be driven in a thin guide hole or in some cases directly into the soil. Stainless-steel tubing with no air gap has the potential to provide relatively comparable performance to a plastic probe with an air gap larger than 2 mm. Finally, it can be noted that the use of aluminum instead of stainless steel is inadequate because it strongly decreases the measurement accuracy (Table 1).

The effect of most characteristics and factors mentioned above is minor compared to the error resulting from possible inaccuracies in positioning the sensor at a specific depth, which can occur with all measurement methods. Here, a hypothetical 1 cm downward shift of the probe can lead to an amplitude and measurement error of ∼8 % and ∼12 %, respectively, and a time delay of 1244 s. These errors are 2 times larger than the effect of an air gap of 5 mm between the soil and the probe.

For the case where the DTP system is installed temporarily for capturing a single time or snapshot of the soil temperature for mapping purpose, the amount of time needed to approach temperature equilibrium between soil and sensors depends on environmental factors and desired measurement accuracy (Fig. 4 and Table 2). The DTP system in its standard plastic housing (base case; 10 mm (∼38 in.) OD plastic tubing filled with urethane blend) and the stainless-steel probe require 824 and 1040 s, respectively, to reach 1 % of the initial difference of temperature between the probe and the soil. Results indicate that a 1 mm air gap produces a significant delay in the early time of the equilibration process, although it reaches 1 % of the initial difference after a comparable amount of time, i.e., 1070 s. In the presence of low soil diffusivity, the equilibration time increases to 1748 s, implying that leaving a probe in place for about 30 min is appropriate for many applications. Finally, results indicate again the importance of ensuring a good coupling between the probe and the soil, as seen by the effect of a hypothetical 5 mm gap between the probe and soil, which more than doubles the equilibration time needed for reaching similar accuracy.

Quantifying snow and water distribution in snow-dominated mountainous watersheds is critical for managing downstream water resources and societal services (Viviroli et al., 2007), especially at a time when their functioning is increasingly altered by climate change (Barnett et al., 2005). Climate change and interannual variability in precipitation intensity and surface temperature strongly impact snowpack dynamics and snowmelt timing, streamflow, groundwater recharge, and surface energy balance. A particular challenge to the predictive understanding of watershed dynamics and response to perturbations is monitoring snowpack properties, the timing and magnitude of snowmelt events, and the repartitioning of water into surface and subsurface flow. Such monitoring must be conducted at multiple scales across complex terrains as needed to accurately capture the impact of a large range of gradients in topography, air mass exposure, and vegetation cover on these dynamics (e.g., Lundquist et al., 2019; Strachan et al., 2016). The DTP system has the potential to significantly improve the sampling of these properties and their variability along these gradients.

Two collocated DTP systems in a standard plastic housing, one above and one below the ground surface at a mountainous headwater site in the East River watershed of the upper Colorado River basin (Hubbard et al., 2018; Tran et al., 2019; Wainwright et al., 2022), are used to illustrate the DTP data information content on the timing and amplitude of thermal and hydrological processes in the snow and soil columns (Fig. 5). The snow DTP system provided temperature with 0.1 m resolution between 0.05 and 0.85 m above the ground surface and 0.05 m resolution between 0.85 and 1.15 m. The DTP system installed in the soil, next to the snow DTP system, provided measurements with 0.05 m resolution between 0 and 0.3 m depth and 0.1 m resolution between 0.3 and 0.7 m depth. Sensor accuracy was 0.06 ∘C, as probes were deployed before the development of the described calibration method. The snow-thickness algorithm is applied to the DTP system placed above the ground surface, and the estimated snow depth dynamic is compared to the snow depth pattern observed using a sonic sensor at the Butte SNOTEL site (snow telemetry; https://wcc.sc.egov.usda.gov/nwcc/site?sitenum=380, last access: 25 February 2022) located 2 km away and 350 m higher in elevation. Though similar snow depth trends are expected at both sites, their absolute values cannot be compared due to their difference in elevation and location. Soil moisture measurements at 0.1 and 0.5 m depth about 1 m away from the DTP probes are used to further evaluate the information contained in the DTP data.

The snow depth estimated using the DTP system is consistent with the snow depth observed at the Butte SNOTEL site (Fig. 5). The DTP system captures the main changes in snow depth linked to snow precipitation, snowmelt, and/or snow compaction visible in the SNOTEL dataset. The earlier timing in snowmelt at the DTP site is explained by the fact that the SNOTEL station is at higher elevation than the DTP probe. Overall, the main differences between these methods is the lower temporal and spatial resolution of the DTP-inferred snow depth, which is caused by the use of a 24 h moving window to estimate snow depth, the occurrence of days involving snow precipitation events and/or very little diurnal fluctuations, and the 5–10 cm spacing between temperature sensors along the probe.

The DTP system placed below the ground surface shows that the soil freezing, which is estimated by extracting the 0 ∘C isotherm from the temperature data, starts in mid-October and reaches 0.4 m depth by mid-November. The first significant snowfall at the end of November increases the insulation of the ground, which leads to a slow decrease in the frozen-layer thickness from the bottom (Fig. 5a). The soil thawing accelerates in March after the snowpack becomes thicker and air temperature gets warmer. The thawing of soil occurs relatively quickly, likely because of the presence of a relatively dry soil, as indicated by the absence of a clear zero-curtain effect expected in the presence of a large amount of ice and subsequently latent heat absorbed during phase transition. After mid-March, the entirely thawed soil, still covered with snow, remains at an almost constant temperature for about 3 weeks, with less than 0.01 ∘C change per day (Fig. 5c).

The major snowmelt event occurring at the end of March is captured by the aboveground DTP system via the strong decrease in snow depth as well as by the temperature of 0 ∘C throughout the snowpack. Indeed, once the entire snowpack reaches 0 ∘C, the additional thermal energy entering the snowpack initiates the phase change and water infiltration throughout the snowpack (Fig. 5a) (Dingman, 2014; Reusser and Zehe, 2011). The snowmelt water reaching the ground is close to 0 ∘C, while the ground at this location shows a relatively constant temperature of 0.41, 0.65, and 1.38 ∘C at 10, 20, and 50 cm depth, respectively. The snowmelt infiltrating into the ground creates a slight decrease in soil temperature that is apparent in the soil temperature data (Fig. 5a) and more clearly identified by looking at the change in the 24 h average temperature difference (Fig. 5c). The temporal change in temperature with depth, which has a different shape than at earlier times during the winter – when heat conduction was dominating heat transfer, is related to the water infiltration. This change in soil temperature is consistent in timing with a soil moisture increase at 10 cm depth around mid-April and 10 d later at 50 cm depth. Overall, the high vertical resolution and accuracy of the DTP system and its deployment above and below ground enabled the observation of snowpack dynamics and its impact on the soil heat (and to some extent hydrological) fluxes at a resolution that is not achievable with traditional sensors.

Large uncertainty remains in how northern high-latitude environments will evolve under climate warming and in particular in how thaw and release of permafrost carbon will be offset by increased vegetation carbon uptake (Jorgenson et al., 2010; Parazoo et al., 2018). Arctic annual average air temperatures between 1971–2017 increased by 2.7 ∘C, at 2.4 times the rate of the Northern Hemisphere average (Box et al., 2019). This change in temperature is complemented with changes in other atmospheric properties, including humidity, cloud formation, rainfall, and snowfall precipitation. One particular challenge involves improving the predictive understanding of how permafrost regions transition to unfrozen ground and disentangling the various controls and their individual impact on the carbon cycle (Jorgenson et al., 2010). Overcoming this challenge requires improving our capability to estimate the soil freeze/thaw depth, the impact of spatially variable temporal shifts in insolation and insulation on the subsurface temperature, and the water/heat fluxes.

Here, a DTP system in a standard plastic housing located in a discontinuous permafrost environment along Teller Road (mile 27) near Nome, Alaska (Léger et al., 2019; Uhlemann et al., 2021), is used to further illustrate the value of the DTP system in monitoring temperature, frozen-layer thickness, and thawed-layer depth, as well as to discuss the potential issue of DTP system displacement relative to the ground surface (Fig. 6). The displayed DTP dataset comes from a probe deployed at a location where the permafrost table is deeper than the bottom of the probe located at 1.05 m depth. The deployed probe provided temperature with 0.05 m resolution from 0.05 m above the ground surface to 0.25 m depth and with 0.1 m resolution from 0.25 to 1.05 m depth. The bottom of the frozen layer in the fall and wintertime, as well as the bottom of the thawed layer in the spring to fall season, were estimated by selecting the deepest sensor with soil temperature below and above the 0 ∘C isotherm, respectively. The dataset discussion involves an evaluation of the snow depth and air temperature obtained at a nearby monitoring site (<1 km) using a sonic-based snow sensor and air temperature sensor, respectively (https://ngee-arctic.ornl.gov/data/pages/NGA243.html, last access: 25 February 2022).

Soil freezing, which starts at the end of October before being slowed by a warm event coupled with snow precipitation in late November, reaches a depth deeper than the length of the DTP probe in early February (Fig. 6b). The small amount of snow (<30 cm) on the ground favors soil freezing until snow event intensity and air temperature increase in March and April. Consequently, the ground temperature increases and the temperature of the entire soil column reaches temperatures slightly below 0 ∘C at the end of April. The soil thawing process starts after the first bare-ground day, as indicated by the diurnal daily temperature variation becoming visible at the ground surface. The soil thawing occurs slowly, with a zero-curtain effect indicating the presence of wet conditions. The thawed-layer thickness increases from mid-May to mid-August, at which time the thawing occurs deeper than the DTP probe.

The detection of a persistent and negligible time delay in daily minimum temperature between the aboveground sensor and the underlying nearest sensor indicates the presence of a second sensor above the ground surface and thus an upward displacement of about 2.5 cm (±1.5 cm) of the DTP system relative to the ground surface around 3 June. Then, there is an additional displacement around 20 August, leading to a 7.5 cm (±1.5 cm) total displacement during the thawing season (Fig. 6c). The developed detection method provides reliable detection of probe movement relative to the ground, though it does not enable centimeter-scale resolution. Still, the approach allows us to flag the data for lower accuracy and possibly apply subsequent corrections to the reference depth of temperature data and inferred metrics.

Overall, the high vertical resolution and accuracy of the DTP system enables the monitoring of temperature – and related frozen-/thawed-layer thickness – in the Arctic environment at a resolution that allows us to disentangle the impact of various processes on soil warming and changes in hydro-biogeochemical processes. Even with the difficulties in monitoring extreme environments, the DTP system offers a way to account for various sources of measurement uncertainties and potentially develop the dense datasets needed to improve predictive understanding of Arctic feedback to climate change.