4.1 Electrode installation and measurement schedule

The SP setup consists of 20 unpolarizable Pb/PbCl₂ SP electrodes of the Petiau-type (Petiau, 2000), arranged in a 4×4 grid of 16 electrodes on the surface with an approximate distance of 7.5 m between rows and columns (Fig. 2). Another two electrodes are located near the lower-left electrode (no. 1), with one electrode (no. 19) used to replicate electrode no. 1 and the other electrode (no. 20) located approximately 20–30 cm deeper than electrode 1 in the soil. Electrodes 1 and 19 are shown in Fig. 3. A similar arrangement is implemented for electrode no. 16, with electrode no. 17 duplicating it and electrode no. 18 buried below them. Electrodes were placed at 30–40 cm below the surface to reduce diurnal temperature effects caused by direct solar radiation and to ensure proper contact with the soil. We refrained from embedding the electrodes in bentonite clay, due to unknown freezing/thawing behavior and because salt leaching and corresponding electrical diffusion/concentration potentials can be expected in saturated flow regimes (e.g., Perrier et al., 1997), as expected for the test site. As discussed by Hu et al. (2020), even in nonfreezing, unsaturated, conditions it seems that the used electrode type will degrade relatively fast. Coupled with the fact that it is still unclear how the unpolarizable electrodes will age in continuous freezing/thawing environments, it is important to carefully monitor electrode performance. In the preceding years of experiments with SP electrodes in permafrost regions, we sometimes encountered loosened membranes at the tips of the electrodes. However, these electrodes were usually not properly embedded into the soil, thereby missing corresponding counter-forces under thermal changes, or experienced some kind of physical force due to improper installation. Proper embedding into the soil ensures proper counter-forces during freezing, which should prevent the membrane from being pushed out by frozen electrode slush. As recommended, e.g., by Corwin (1989), we performed baseline offset measurements of the SP electrodes in saline solution. Out of a pool of available SP electrodes, electrodes were selected (“matched”) to minimize their baseline offset when measured within a saline solution. A random selection of SP electrodes initially yielded baseline measurements of up to −5 mV (with respect to the reference electrode), while matching reduced this effect to within ±1 mV (also visualized in Fig. S6). Each electrode is fitted with a Pt100 temperature sensor, located directly at the electrode membrane.

The electrodes were connected to the multiplexer units of the data logger (dataTaker DT80M Series 4, detailed below in Sect. 4.2) in a way to allow for the measurement of specific electrode dipoles. Multiple layers of multiplexing were involved to measure voltages. First, a hard-wired routing board was used for a first-line multiplexing of the electrode signals to multiple logger input channels – schematic of this routing (multiplexing) board can be found in the Supplement Fig. S5. Second, each logger input channel was then internally connected to the measurement unit of the logger using the additional internal multiplexers built into the logger and its extension units. Third, each logger channel is comprised of up to four individual inputs, allowing for a variety of measurement operations, including the measurement of voltages between various combinations of these four inputs.

Measurements of electrical SPs were first conducted with respect to the reference electrode no. 1 (lower-left corner), resulting in 15 relative potentials (voltages). To increase system reliability and provide means for consistency checking, another 14 SP potentials were measured with respect to electrode 16, which is located at the upper-right corner of the electrode grid (the dipole between electrodes 1 and 16 was not measured twice, resulting in only 14 measurements). As such, 29 dipoles were physically connected to the multiplexer channels, requiring some duplication of the electrode leads (see Fig. S5 for a wiring diagram). Additionally, the potentials between reference electrodes and their duplicates (electrode pairs 1–19 and 16–17), located only a few centimeters apart, were also recorded to analyze electrode drift and localized inhomogeneities, as were potential differences between reference electrodes and their associated electrodes at deeper locations (pairs 1–20 and 16–18).

Electrical voltages and various operational data (battery voltage, solar voltage, internal logger temperatures) were measured every 10 min in the daytime and every 30 min at nighttime, in order to conserve power. To reduce aliasing effects, voltage measurements were repeated and averaged 10 times, with the internal integration time of the data logger set to 50 ms (20 Hz). Once per hour, temperatures and voltages across the temperature sensors were measured. In order to prolong measurement operations for low-power situations, depending on the battery voltage, these measurement intervals were reduced to 1 h measurements for a battery voltage between 12.3 and 12.5 volts and to 6 h for battery voltages below 12.2 volts. Once per day the resistances between horizontally adjacent electrodes, as well as vertical pairs at the beginning of each electrode row (i.e., pairs 1–5, 5–9, 9–13; see Fig. 2), were recorded using a 2.5 µA constant current source for excitation. This measurement is referred to as the contact resistance between the two involved electrodes and incorporates the interface resistance between electrodes and soil, as well as the soil resistance itself.

System time was set to UTC in order to prevent mismatches in time stamps with external data sources, and measurement data was uploaded by email every 6 h. Once per day the system updated its internal real-time clock using the network time protocol (NTP) in a scheduled online session. The detailed measurement schedule can be found in Table S1 in the Supplement, and the programming for the data logger can be found in listing 1 of the Supplement.

Figure 4Schematic overview of essential system components. Components included in the group “Central measurement system” are housed in an aluminum box (see also Fig. 5), while all other components are external and connected by cables. The term “electrode group” refers to one pair of SP electrode and attached temperature sensor. The components termed “lightning protection and low-pass filter” were only installed and tested in 2019.

Download

4.2 Composition of the measurement system

The overall system design, as constructed and deployed in the years 2017–2019, is presented in Fig. 4. Power is provided by two 12 V Exide ES950 gel batteries, connected in parallel to reach a design capacity of 190 Ah. A 30 watt solar panel is used to charge the batteries using a Steca Solsum solar charger. Data are logged by a dataTaker DT80M, Series-4 data logger, to which three dataTaker CEM20 extension modules (internal multiplexers) are connected, resulting in 62 available analog-in channels. Each input channel of the data logger consists of four terminals, allowing, among others, one four-point resistance measurements or up to four voltage measurements between different terminal pairs (i.e., a limited multiplexing capability within each channel). The logger measures voltages and temperatures sequentially, using the internal multiplexers to select the input channel to work on. It was found that the system operates more reliably under cold conditions if power to the logger was increased from 12 to 24 V using a DC–DC converter. A metal rod inserted into the soil a few meters away from the system was used as system ground.

Two cellular network antennas are connected to the data logger, providing improved signal reception capabilities. Apart from one Pt100 temperature sensor used to measure air/snow temperature, directly attached to the box housing the measurement system, each unpolarizable SP electrode is paired to one Pt100 temperature sensor, with both sensors referred to as an “electrode group”. Each electrode group is connected and grounded to the measurement system via an eight-wired shielded ethernet cable resistant to UV radiation. The actual measurement system is housed in an aluminum case that can withstand expected snow covers and is lined with Styropor on the inside for thermal shielding (Fig. 5).

Figure 5Central measurement system. The red rectangle marks the patch panel at which the incoming connections from the electrodes are transferred to standard ethernet cables.

The ethernet cables of each sensor package are introduced into the box via water-tight cable glands and connected to an ethernet panel via standard insulation displacement connectors (IDC). Signal lines from 20 electrode groups are internally routed and multiplexed to the input channels of the logging system using custom-made printed circuit boards (PCBs), interconnected using ethernet cables (see Fig. S1). In response to frequent lightning damage in the preceding years, a lightning protection circuit board based on electrostatic discharge (ESD) protection diodes was tested in the summer of 2019 to protect all SP input channels against voltage surges caused by nearby storms. Signal lines must be routed to multiple logger input channels to conduct the various measurements, requiring the separation of temperature sensor signals from the SP signals using a “splitter” board. Temperature lines are then directly connected to logger input channels, while the SP lines are further multiplexed to allow for resistance measurements between selected dipoles and for the measurement of the actual SP signals. This second signal splitting was required due to technical constraints in measuring high-ohmic resistances. Figure S1 depicts the signal flow within the system, and Figs. S2–S5 show the schematics of the four interconnected PCB boards, ending in the individual data-logger input channels.

As discussed by DesRoches and Butler (2016), the internal filtering capabilities of data loggers are not always sufficient to suppress high-frequency noise superimposed on the, usually weak, SP signals, which prompted them to install discrete low-pass filters based on RC circuits at the input channels of the data logger. DesRoches and Butler (2016) also pointed out that the cutoff frequency of simple RC filters depends on the soil resistance. Our 2019 tests also included a low-pass filter (cutoff frequency 1 Hz) to reduce the amount of aliasing noise. Due to technical restrictions, the cutoff frequency of the low-pass filter could not be adapted to the Nyquist limit of the measurement schedule. As such, some remaining data noise caused by aliasing of high-frequency components is to be expected.

The electrode groups were connected to the ethernet cables using small switching boxes, again using IDC technology, located a few centimeters next to each sensor (Fig. 6b). To prevent any contact issues caused by humidity, the connection boxes were sealed with two-component epoxy resin (rated for use below 0 ^∘C). A planned but only partially implemented next step was to embed these resin-covered boxes in a plastic case filled with silicon to provide additional physical protection (Fig. 6c).

Figure 6Installation of SP electrodes. (a) Pb/PbCl₂ electrode with attached Pt-100 temperature sensor. Note that this specific electrode depicted in the picture was only used for construction testing and was not used for SP measurements due to the dried-out membrane. (b) Connection box in which the electrode/temperature cables are connected to the ethernet cable used for long-distance signal transport;(c) connection box from (b), embedded in silicon within a sturdy PVC box for physical protection.

5.1 Auxiliary measurements

In order to better assess the system operations and expected measurement results, two additional experiments were conducted. One laboratory experiment was designed to investigate electrical potentials generated within electrodes during freezing and thawing under no-flow conditions. The other experiment was a high-frequency noise measurement, conducted on site at the Schilthorn in the summer of 2019 using the largest available SP dipole, aimed at investigating the magnitude and stability of noise components encountered during routine operations of the monitoring system.

Figure 7Laboratory experiment investigating electrode effects during freezing and thawing. (a) Sketch of experimental setup. The container is 30 cm in length, 20 cm in depth, and 5 cm high. Electrodes are shown in gray, with the wood membranes in contact with the surroundings highlighted in red and assigned electrode numbers in white boxes. (b) Thermal image (surface temperatures) during thawing. (c) Electrical voltages and soil temperatures; vertical black line indicates time of thermal image in (b). Voltages were offset-corrected to the beginning of the observed time frame.

Download

5.2 Quality assessment

In light of the harsh measurement environment, system integrity must be monitored with special care. This is important to not only plan maintenance and repair activities but also to flag bad data caused by malfunctions or strong external noise activities (see Fig. 8). Besides the monitoring of regular system activity in the form of data downloads, battery and solar voltages are monitored with respect to expected system behavior (not shown here). In the following we discuss some of the quality-assessment procedures implemented for the presented system, targeted at the measurement of SP data.

Figure 8Noise measurement on dipole E16-E01, conducted in the summer of 2019. Raw data are drawn in black, a 2 h moving average filtered version in red. (a) Time series on scale of raw data. (b) Filtered time series with zoomed-in y-scale.

Download

5.2.1 Contact resistances

Daily contact resistances are used as a rough proxy for electrode health, ensuring that an electrical path exists between the two electrodes involved. Successful measurements are in general dominated by the interface resistance between the electrodes and the soil, with the actual soil resistance assumed to be of minor importance (e.g., Corwin, 1989). As such, contact resistance data and their temporal evolution can, within limits, be used to assess the state of the electrode–soil interface. Very low contact resistance measurements could point towards short circuits in the system, while failing measurements (outside the measurement range of the data logger), caused by very large resistances, point to missing electrode–soil contact, or broken cabling.

Figure 9 shows the temporal evolution of contact resistances over the whole measurement period of 2017 and 2018. No extreme data points that indicate short circuits or lost electrode contacts were observed. However, the actual values of the resistances differ with the electrode pairs, with very high resistances (red colors) occurring for some electrode pairs and relatively low resistances (blue) showing up at early times for the electrode pairs 3–4 and 6–7.

Figure 9Contact resistances (R), measured daily between selected electrode pairs. For electrode numbering, refer to Fig. 2.

Download

It is not clear if and to what extent the actual values of the resistances, in contrast to the patterns and extreme values, can be used to assess electrode and soil characteristics. While the patterns present in Fig. 9 do not necessarily correlate to the temperatures of the involved electrode temperatures, one possible influence could be the freezing of soil between electrodes. This freezing would lead to a significant change in subsurface resistance between the electrodes, thereby also changing the contact resistance, somewhat contradicting the general assumption that the interface between electrode and soil dominates the readings (e.g., as observed for the dipoles 3–4 and 6–7). As such, differences between electrode pair resistances suggest spatially variable subsurface conditions (lithology – grain size, pore size, etc. – but also water/ice/air content), while temporal variations indicate changes in pore water chemistry, temperature, and freezing in the direct vicinity of the electrodes.

Note that the use of a static current source leads to electrode polarization, which prevents useful SP measurements with the involved electrodes for some time after (as already noted by Corwin, 1989). As such, only one measurement of the contact resistance is conducted per day, and, based on laboratory experience, we suggest ignoring any SP data for up to 1 h after these measurements.

Please also note that the measured contact resistances differ from those reported for ERT at the same location (e.g., Mollaret et al., 2019, and references within) for multiple reasons. The construction of SP electrodes is, by design, very different from those used for active methods such as ERT. Active methods require a low internal electrode resistance and a large surface area to properly inject current into the soil, while passive SP electrodes are only used to measure the electrical potential generated by external means, with the additional requirement of low influence of electrode polarization (e.g., Corwin, 1989; Petiau, 2000). As such, a high internal resistance is, within certain bounds, actually favored for potential measurements in order to reduce any leakage currents and associated distortions of the measurements. In this regard, the choice of steel electrodes for ERT measurements is a common compromise between good current injection capabilities and acceptable potential measurements. Another difference is the surface area. The SP electrodes employed in this study yield only a small membrane of 3 cm diameter in contact with the soil, while the ERT electrodes offer a much larger surface area in soil contact.

Figure 10Comparison of two SP dipoles, both of which were measured directly and constructed by means of superposition, and corresponding rolling correlation coefficient. (a  b) Dipole between electrodes 5 and 1. (c, d) Dipole between electrodes 8 and 1. (a, c) Filtered direct (black) and indirect time series (purple) moving-average filter; window length 12.5 h. The time series between both reference electrodes (E16-E01, gray solid line) is shown here for reference and is used in the superposition procedure. Red dots indicate raw data of dipoles, from which the rolling mean was computed. Raw data for dipole E16-E01 and for the superposition are not shown. The labels A and B in (a) depict temporally localized anomalies that will be discussed in detail in Sect. 5.3.3. (b, d) Results of moving correlation; window length: 24.5 h.

Download

5.2.2 Superposition of SP dipoles

While contact resistances are easily used to assess electrode contact to the soil, they do not provide information on noise components in the data. Given the large number of signal and noise sources that contribute to the measured SP signatures, ways to isolate process-related signals from noise components are essential. One way of isolating signatures from noise components is the use of low- or high-pass filtering, given a suitable understanding of the process dynamics. This also implies that the process signal-strength dominates a given frequency band. What follows is the challenge of making sure that noise was suitably removed by the filtering operation.

The SP potentials are measured redundantly with respect to two reference electrodes, nos. 1 and 16. This not only provides resilience towards a failure of one of the reference electrodes but also allows consistency and noise checks by recomputing dipoles via superposition of measurements with respect to the other reference electrode and subsequent comparison of the results to direct measurements (this would be akin to the common procedure of re-referencing SP measurements for mobile dipoles). For example, the dipole E05-E01 can also be computed by superposition: E05-E01_super= E05-E16 + E16-E01. Naturally, noise components will increase in the computed dipole by means of error propagation, leading to different noise levels in direct and indirect measurements. If a signal extraction is now attempted, for example, by applying a low-pass filter to the data, then we can compare the filtered direct and indirect time series. Any deviation from both time series can be attributed to incompletely removed noise components (which are expected to vary between both time series). Note that the different noise levels actually help to assess the quality of noise removal. Only for completely removed or suitably suppressed noise do we expect a good comparison of both signals.

We chose to use a rolling linear correlation to present this concept (Fig. 10). Given a filtered time series and its corresponding indirect one (filter applied before superposition), a linear correlation coefficient (R²) is computed for rolling windows of data, providing time-localized information on the status of noise removal. Dipole E05-E01, presented in Fig. 10a, shows two distinct signals that are attributed to meltwater flow (A/B, to be discussed in detail in Sect. 5.3.3), while dipole E08-E01, presented in Fig. 10c, does not exhibit clear signals from temporally localized events. The rolling linear correlation of dipole E05-E01 with its superposition counterpart is generally close to 1, indicating a strong, clear signal. Only one spike to lower correlation values is observed and can be directly linked to a clear deviation of both signals at this time step (around 5 May). Contrarily, while the correlation of dipole E08-E01 shows periodic improvements, roughly at daily intervals, the correlation between direct and indirect time series remains relatively low, indicating incomplete removal of high-frequency noise components.

Please note that the chosen window length of 12.5 h does slightly distort the SP data and is chosen here just as a conservative choice to produce a filtered time series with a low share of noise components (Fig. 10a). If the procedure presented here is used in an actual process analysis, the rolling-average filter should not be used for low-pass filtering due to its underperforming frequency response. Given a suitable understanding of the signal shape of the processes to be analyzed, a signal extraction could also be attempted in the time domain. The consistency check presented in this section could then still be used for quality assessment.

We interpret low correlation coefficients merely as an incomplete removal of noise components, whose definition will vary with specific application goals. For example, it seems that daily variations manifest differently in dipoles E16-E01 and E08-E01 (Fig. 10c, black and gray lines), leading to a cyclic variation in correlation (Fig. 10d), and could be caused by different dipole lengths and locations and associated different temperature responses, which in itself could be a valuable target of investigation.

Figure 11(a) Temperature curve for sensor at electrode  1. (b) Corresponding voltages (blue), measured over the Pt100 resistance of the temperature sensor before current injection, and logger temperature (red).

Download

5.3 Measurement data

In this section we present selected results of the operating time in 2017 and 2018, with two longer interruptions caused by power issues. These results serve first to show that reliable and consistent data can be measured with the system and used only in secondary role as an attempt at extracting process-related information (which would go beyond the scope of this study). Temporal and spatial consistency of the data are shortly discussed, as well as two captured SP events in May 2018, which are interpreted as being caused by snow meltwater infiltration near selected electrodes.

Figure 12Overview of SP time series and temperature data. (a) All 15 SP time series measured with respect to reference electrode 1. Data are filtered with a rolling-mean filter (window: 48.5 h). (b) Corresponding temperature curves for all electrodes. Temperature readings that showed significant voltage drops over the temperature sensors were discarded. (c) Correlation of SP dipole E05-E01 with electrode temperature difference, ΔT_5-1.

Download

5.3.1 Long-term SP data

Filtered SP data for the full measurement time span, starting in September 2017 and ending in May 2018, are presented in Fig. 12a, with corresponding electrode temperatures shown in Fig. 12b. In order to visualize long-term trends, a rolling-average filter with a window length of 48.5 h was applied to the electrical data, suppressing high-frequency noise and pronounced diurnal cyclic signals (cf. Fig. 10a, c). The spectral content of the unfiltered data is discussed in Sect. 5.3.4.

The evolution of electrical SP signals is characterized by roughly four phases. First, there is a freezing phase (Fig. 12a – A), which shows both temporal and spatial variations in the SP signals. This period is followed by a “quiet” time, showing only minor variations between dipoles over time (Fig. 12a – B). After all electrodes reached temperatures below 0 ^∘C, curves do not diverge much but show much larger temporal variability (Fig. 12a – C). Finally, following the data gap in early 2018 (malfunction of solar panel), curves start to diverge again, while maintaining their larger temporal variability (Fig. 12a – D).

While some electrodes dipped below 0 ^∘C, phase A is still dominated by temperatures around 0 ^∘C, indicating only partially frozen subsoil, in which water flow is still possible. This possibly heterogeneous water flow, coupled to ongoing freezing processes, could lead to correspondingly heterogeneous streaming potentials and chemical potentials, explaining the variations between electrodes. Note also that the temporal evolution of freezing of the soil water can be highly heterogeneous even on the scale of the presented measurement setup, possibly explaining the observed variations and dynamics of electrode temperatures.

Long-term trends of the SP dipoles show a correlation to the temperature differences between the electrodes, with the SP data and temperature curve for dipole E01-E05 presented exemplary in Fig. 12c). Note, however, that the observed variation in the electrical signals by far exceeds the temperature coefficients of the electrodes (ca. 20 µV ^∘C⁻¹; Petiau, 2000), suggesting additional processes at play (i.e., highly localized water pockets/movement, or geochemical changes). Similar correlations of SP signals to temperature differences at the electrodes were also observed in other settings, e.g. in Friedel et al. (2004). Hu et al. (2020) also suggested variable membrane potentials of the electrodes as sources of measured SP signals.

Figure 13Measured voltages between references electrodes 1 and 16 and their respective duplicated electrode, located a few centimeters away. (a) Temperature readings of all four involved electrodes. (b) Voltage readings between duplicate electrode pairs 1, 19 and 16, 17.

Download

5.3.2 Voltages with respect to duplicate electrodes

Both reference electrodes, with numbers 1 and 16, were duplicated in order to investigate electrode drift and small-scale heterogeneities (Fig. 13 shows temperatures of all four electrodes, as well as the SP voltages between the pairs). As long as all temperatures remain above 0 ^∘C, measured SP voltages fluctuate within 6 mV (Fig. 13a, b, marker A). Afterwards, both voltages stabilize at fixed values, which, however, deviate from 0 mV (time span indicated by B). Finally, two temperature events are correlated with transient spikes in the measured voltages (markers C and D).

We interpret the initial phase A as a sign of ongoing hydrological and chemical processes in the soil, with corresponding heterogeneities in chemical and streaming potentials. After freezing (B), a chemical equilibrium ensues, and water flow stops, leading to stable voltage readings. The two events (C, D) associated with temperature increases are partially attributed to internal electrode effects but are also a clear sign of meltwater infiltration down to the electrodes. These events are discussed in the next section.

Figure 14SP time series and associated data over the time window containing three distinct temperature events. (a) Daily snow cover height. (b) Soil moisture content from nearby PERMOS borehole. (c) Temperature curves for electrodes 1 (cyan) and 5 (red). Temperature readings that showed significant voltage drops over the temperature sensors were discarded, leading to the temporal gaps in the curves. (d) SP time series between electrodes 5 and 1. Blue: raw data; black filtered data (6.5 h rolling mean).

Download

5.3.4 Spectral power analysis

In order to get a first-order idea of spectral components of the data, the irregularly spaced raw data were analyzed using the Lomb–Scargle periodogram (VanderPlas et al., 2012; VanderPlas and Ivezic, 2015). Data from the central data block (10 November–28 January 2017) were used, and periods between 1∕6 d (4 h) and 2 d were analyzed (Fig. 15). Data were detrended before analysis, but no further analysis regarding stationarity was done; this would be required if further analysis was attempted.

Figure 15Spectral power analysis of E01-E16, using the Lomb–Scargle periodogram. Data were detrended before analysis.

Download

The data exhibit clear diurnal components around the 24 h period (23, 24, 25 h). The daily cycles can also clearly be observed in Fig. 10a and c. Additional signal components can be found for 4, 6, 8, and 12 h periods. At this point we do not try to attribute any signal sources to those periods and would suggest further advanced time series analysis before any period other than the 24 h signal is analyzed due to the inherent nonstationarity of the time series. However, we would like to note that all recovered periods can be associated with global tidal processes (e.g., Egbert and Booker, 1992; MacAllister et al., 2016). Also, the geomagnetic field is known to be influenced by additional perturbations of the ionosphere, caused among other sources by solar wind conditions and magnetic pulsations (e.g., McPherron, 2005; Chave and Jones, 2012). We can also exclude large influences of the regular ERT measurements conducted at the Schilthorn. While these measurements have a planned measurement interval of 24 h, measurements only take up to an hour, which would lead to corresponding disturbances in the SP signals for only this duration and thus cannot explain the observed continuous signals changes over the day (e.g., visible in Fig. 10a, c).

Elucidating the true nature of any SP signals associated with one or more of these periods is beyond the scope of this study. However, given that the study site remained largely frozen for much of the presented time frame, we can exclude, or at least doubt, some of the explanations given for other studies in respect to our study. Direct temperature-related causes can be excluded due to the absence of clear diurnal temperature signals being recorded at the electrodes. We can also exclude that surface meltwater was generated throughout the whole season, given that we only expect significant meltwater in winter due to strong surface radiation caused by direct sunlight, which is often not present in case of cloud covers and limited on the north-facing slope of the Schilthorn.

Blake and Clarke (1999) and Kulessa et al. (2003) suggested electrochemical and streaming potentials as possible causes for diurnal variation. However, their potential measurements were explained with variations in electrical fluid conductivity and water pressure caused by periodic pressure and meltwater flows in a summertime glacier. We do not expect significant periodic water flow and changes in geochemistry throughout the time of frozen soil at the Schilthorn.

Friedel et al. (2004) discussed magneto-telluric currents, capillary flow in unsaturated media, evaporation, and atmospheric air pressure as additional sources for diurnal or semidiurnal periods. Based on the site characteristics, we see only magneto-telluric signals as possible sources for the signals observed at the Schilthorn, again due to missing flow opportunities in the frozen soil.

6.1 Monitoring of streaming potentials in Alpine permafrost

Our primary goal is the monitoring of subsurface water flow (generated by melt processes, precipitation, or other means) in alpine permafrost regions. Yet, commissioning a monitoring SP system, such as described in this study, is only one step in a whole pipeline of steps required to produce reliable estimates of water flow, outlined in the following. It was shown that assessment of data quality becomes essential for measurements in such a harsh surrounding, and suitable assessment tools have been discussed to separate signals from random noise components. The specific nature of these noise signals will change for different field sites, requiring the establishment of standard operating procedures to guarantee useful SP measurements. Here, it should be noted again that daylight times at the Schilthorn summit shall be considered as a worst case in terms of anthropogenic noise (cf. Fig. 8), due to the operation of heavy cable car machinery and associated large ground currents (yet, nighttime data shows considerably less noise, still allowing for long-term data monitoring if noise levels overwhelm process signals in the daytime). Following the separation of signals and noise, process signals need to be decomposed further into relevant and irrelevant components, separating streaming potentials from all other sources of electrical potentials. One approach would be by isolating certain frequency bands known to contain more or less periodic water flow information or to select transient signals in response to additional environmental data, such as borehole data, precipitation, and other weather measurements (in our case provided by the PERMOS network and the Federal Office of Meteorology and Climatology MeteoSwiss).

On the other hand, if the nature of ”noise” processes involved is known, they often can be modeled, and therefore their contribution can be subtracted from a given measurement signal. This is a common procedure, for example, to eliminate telluric signal components (e.g., Perrier et al., 1997) or to account for systematic changes in SP signals due to topography (the “topographic effect”, e.g. discussed in Scapozza et al., 2008). Hu et al. (2020) even attempted to model not only process signals but also electrode effects over time.

However, it is important to remember that each measurement system is constrained in terms of spatial and temporal resolution due to the placement of electrodes and the implemented measurement intervals. Therefore, a thorough analysis of the expected processes and their corresponding spatial and temporal scales should be conducted with the aim either of detecting and removing these “sub-scale” events or to provide guidance as to when and where not to attempt a data analysis due to the interference of these events. One example of a problematic time span would be the onset of meltwater flow, presented in Sect. 5.3.2, Fig. 13A/B). The different thermal and electrical environments of the electrodes in close vicinity to each other indicate processes on small scales that cannot be resolved by the surface electrode grid. As such, care should be taken when attempting to infer large-scale hydrological flow information in this time frame.

Having finally isolated the SP signals attributed to subsurface water flow, actual flow parameters need to be derived. Analysis can be done either using the Helmholtz–Smoluchowski equation or by considering the excess charges at the EDL and their dynamics under water flow (e.g., Jougnot et al., 2020). In any case, usually the coupling coefficient is determined from complementary laboratory information.

Finally, the streaming potentials need to be analyzed with regard to not only time but also space, given the large surface and subsurface heterogeneities in mountainous areas. Here, methods ranging from simple to advanced could be used, correlating the orientation of specific dipoles to water flow (simple) or applying advanced tomographic algorithms to produce 3D subsurface water flow patterns from measured SP signals (e.g., Jardani et al., 2007; Ahmed et al., 2013, 2020). Any inversion of the data would also enable further analysis with regard to a successful extraction of process signatures. Systematic errors (i.e., remaining noise components) that are not covered by the physical forward modeling (or the data error estimates in the inversion) should lead to systematic patterns in the data misfits of the inversion results, opening up additional possibilities for data assessment.

As such, we believe that the presented system offers huge potential for the monitoring of water flow in permafrost regions and can be replicated with relative ease to allow a wide application of the method. However, future efforts should be focused on integrating SP measurements in the established monitoring networks, such as PERMOS, in order to provide complementary data required for the analysis outlined above.

6.2 Possible future extensions and development directions

At this point the system is capable of monitoring SP signatures and electrode temperatures. While additional environmental data are provided for the Schilthorn site by the PERMOS network station located right next to our system, future extensions should integrate measurements of all relevant data required to analyze SP signatures with regard to subsurface flow in order to attain self-sufficiency. The additional, integrated measurement of snow cover, fluid pressure, salinity and pH, and subsurface electrical conductivity greatly enhances the analysis possibilities and is required if subsurface tomography is to be implemented. Measurements of local electrical permittivity and fluid conductivity could further improve subsurface parameterization for process-based analysis and could be used for calibration of electrical resistivity measurements, which exhibit spatially variable resolution characteristics. Also, contact resistances between electrodes should be measured for the same dipoles as used for the SP readings (at this point, for technical reasons this is not the case), allowing better analysis of long-term changes. Contact resistances should optimally be measured using alternating currents, thereby reducing electrode polarization and increasing accuracy of the readings. With regard to the findings of Sect. 5.1.1, temperature gradients along the electrodes caused by infiltrating meltwater could be reduced by installing the electrodes slightly tilted in the soil, in parallel to the surface, thereby placing the whole electrode in one front of the meltwater, instead of leaving the upper part of the electrode susceptible to localized warming by meltwater. Aside from the need to improve electrical isolation in the system, the measurement of voltage drops over temperature sensors provides a useful quality assessment tool that can help in identifying periodic signals caused by systematic measurement errors.

As discussed previously and seen in the presented data of this study, SP measurements commonly contain much noise, originating in various sources. In this regard the practice of “traveling electrode referencing” (e.g., Perrier and Pant, 2005) could be adopted to further reduce electrode noise effects. However, this method would require substantial adaption to the surroundings and constraints of mountain permafrost (i.e., to limited on-site visits). As such its effectiveness in such an environment would require validation.

On the technical side, one obvious shortcoming of the system is its reliance on solar power to charge the batteries. While this solution is cheap, small, and suitable for rough terrain, the use of fuel cells can greatly enhance available power and its reliability. Adding more available power to the system would also allow for higher measurement frequencies, which would facilitate the capture of faster flow events, improve noise monitoring, and reduce or fully suppress aliasing caused by the large cutoff frequency of the filters currently implemented. The dependence of the low-pass filter on the soil resistance could be reduced by using a high-impedance-input, low-impedance-output operational amplifier buffer, which would also reduce effects of the input resistance of the data logger. For example, Zimmermann et al. (2008) realized an input impedance of approximately 500 GΩ for active electrical potential measurements with such a setup. Finally, at this point we refrained from implementing a bidirectional data exchange with the logging system, mainly due to power deliberations. This turned out to be much more constraining than previously thought, given that we were unable to reprogram the logger system in response to the data received. Future changes to the system will focus on establishing bidirectional communication.

The design and implementation of the presented SP monitoring system is based on our experience with established classical geophysical devices. Yet, the recent emergence of the so-called internet of things (IoT) also lead to a large number of technical advances that can be easily envisioned to be of use for geophysical applications, especially in niche fields such as the one discussed in this study. Prominent examples of novel sensor networks are the Xsense and PermaSense networks (Beutel et al., 2011; Weber et al., 2019). High-quality data acquisition and processing units are being produced in huge quantities, driving down costs for design and implementation. Microcomputer-based environments around the Arduino or ESP32 environments allow fast and cheap development of targeted hardware, while facilitating large consumer-orientated development environments. The use of 3D printing allows researchers to easily design custom casings, and freely available software libraries speed up development of the corresponding firmware and back-end software. In addition, we see the increasing demand for low-power connectivity between IoT devices as the most effective development for geophysical applications, having lead to the creation of connectivity technologies such as LoRa and Sigfox, which offer low-bit-rate connections over very long ranges (km range), while using much less power than established wireless technologies, such as Wi-Fi or cellular networks. As such, we envision a move from centralized measurement setups, based around one expensive logger system, towards a distributed system of individual, connected, low-cost measurement units. SP signatures could then be measured by small systems with only few connected electrodes, increasing resilience towards disruptions by the elements, increasing area coverage without requiring very long cable connections, and finally driving down deployment and maintenance costs. One small downside of such a system is that the distributed nature also implies multiple reference potentials toward which measurements are conducted, requiring additional processing steps (e.g., Minsley et al., 2008).

Measurements of environmental data in alpine environments are commonly prone to be disrupted by various events, such as power failures and lightning damage, and thus result in irregular spaced time series. If precise frequency analysis of the measured data is required, for example, in order to extract or identify relevant signatures concentrated on certain frequency bands, then advanced analysis techniques must be applied to irregular time series, such as the Lomb–Scargle periodogram (VanderPlas et al., 2012; VanderPlas and Ivezic, 2015) or improved multi-taper spectral estimation methods (e.g., Chave, 2019).