Geophysical Research Letters

Detecting unfrozen sediments below thermokarst lakes with surface nuclear magnetic resonance


Corresponding author: A. D. Parsekian, Department of Geophysics, Stanford University, Stanford, CA, USA. (


[1] A talik is a layer or body of unfrozen ground that occurs in permafrost due to an anomaly in thermal, hydrological, or hydrochemical conditions. Information about talik geometry is important for understanding regional surface water and groundwater interactions as well as sublacustrine methane production in thermokarst lakes. Due to the direct measurement of unfrozen water content, surface nuclear magnetic resonance (NMR) is a promising geophysical method for noninvasively estimating talik dimensions. We made surface NMR measurements on thermokarst lakes and terrestrial permafrost near Fairbanks, Alaska, and confirmed our results using limited direct measurements. At an 8 m deep lake, we observed thaw bulb at least 22 m below the surface; at a 1.4 m deep lake, we detected a talik extending between 5 and 6 m below the surface. Our study demonstrates the value that surface NMR may have in the cryosphere for studies of thermokarst lake hydrology and their related role in the carbon cycle.

1 Introduction

[2] Recent studies have demonstrated that thermokarst lakes, formed due to thawing of ice-rich permafrost followed by subsidence, can emit considerable volumes of methane to the atmosphere [Walter et al., 2007]. The methane originates from microbial decomposition of organic matter in unfrozen sediments beneath the lake. Formerly part of the permafrost, these sediments thawed due the thermal influence of the thermokarst lake and occupy a region referred to as lake talik. Modeling and field studies suggest that the zone along the thaw front, i.e., at the boundary of the talik, is the most active in methane production [Kessler et al., 2012]. Determining the geometry of lake taliks is thus important for evaluating the potential for sublake methane production.

[3] Taliks below thermokarst lakes and large rivers are also important in the context of the hydrology of permafrost regions, in particular, for interaction between surface waters and subpermafrost groundwater in the discontinuous permafrost zone [Yoshikawa and Hinzman, 2003]. Depending on the duration of talik development [West and Plug, 2008] and permafrost conditions, some lakes have a “closed” talik that is bounded on all sides by permafrost, whereas other lakes may have “open” taliks that may be connected to deep regional groundwater, ultimately leading to lake drainage in some cases [Yoshikawa and Hinzman, 2003].

[4] Thawed sediments may extend many tens of meters below a lake [e.g., Brewer et al., 1993]. As a result, the use of direct measurements to detect and determine the geometry of taliks is challenging, with only a few borehole investigations undertaken for this purpose [e.g., Brewer et al., 1993]. In an effort to increase and expedite data acquisition in regions that are remote and logistically challenging for borehole studies, noninvasive geophysical methods are increasingly being considered. Past attempts at geophysical investigation of sublacustrine sediments using ground-penetrating radar have been successful; however, this method is limited by investigation depth to the top several meters of sublake sediment [e.g., Schwamborn et al. 2002, Arcone et al., 2006]. Airborne frequency-domain electromagnetic investigations have proven to be reliable at imaging the electrical resistivity up to depths of approximately 100 m over terrestrial permafrost and thermokarst lakes [Minsley et al., 2012]. However, the limitation of any electrical resistivity measurement is that they may be prone to ambiguous interpretation: resistivity is linked to the parameter of interest (i.e., liquid water content), but it is also sensitive to pore water chemistry, temperature, and lithology. Seismic methods have been successfully used for determining talik configuration for one thermokarst lake in Siberia [Schwamborn et al., 2002]; however, this method has not seen widespread adoption.

[5] To overcome challenges faced in using other geophysical methods, we turned to surface nuclear magnetic resonance (NMR) [e.g., Legchenko and Shushakov, 1998]—the only noninvasive geophysical method that is directly sensitive to liquid water (ice has a negligible signal). Surface NMR measures changes in the bulk nuclear magnetization associated with the nuclear spins of hydrogen nuclei in water. The measurement can be summarized as follows: prior to the measurement, the nuclear spins in water in the subsurface are at equilibrium and are aligned with the Earth's static background magnetic field. The equilibrium state is perturbed by an oscillating electromagnetic (EM) pulse generated in a wire loop at the surface that tips the spins away from their initial orientation. The perturbing pulse is tuned to the Larmor frequency (determined by the magnitude of the background magnetic field and the gyromagnetic ratio specific to hydrogen nuclei); therefore, the measurement is directly sensitive to water. When the EM pulse is switched off, the spins “relax” to their initial state. As the spins relax, the signal amplitude is measured as a function of time in the loop at the surface. The magnitude of the initial amplitude of the relaxation signal is proportional to the water content, whereas the decay time (T2*) is related to pore-scale properties (e.g., pore size) [Kleinberg and Horsfield, 1990]. The power of the EM pulse (pulse moment, defined as the product of pulse current and duration) is increased in subsequent measurements to probe deeper into the subsurface. The maximum depth of investigation is approximately equal to the surface loop diameter. Vertical resolution degrades with depth to a maximum of several meters; the horizontal resolution is approximately equal to the diameter of the loop [Hertrich, 2008].

[6] This study assesses the feasibility of using surface NMR to detect a lake talik and determine its geometry. Laboratory NMR measurements have been used to determine the water content in permafrost samples, yielding 4% and 10% in sandstone permafrost and mudstone permafrost, respectively [Kleinberg and Griffin, 2005] and 1% to 10% for silt samples (temperature dependent) [Yoshikawa & Overduin 2005]. We presume that the water content in a lake talik will be much greater than 10%, thus providing a good target for the surface NMR measurement. The only other surface NMR study in permafrost that we are aware of was aimed at the characterization of pingos (massive subsurface ice bodies) in the Fairbanks, Alaska region [Yoshikawa et al., 2006]. The absence of useful surface NMR results in this prior study was most likely due to a lack of contrast in liquid water content in the ice and a high sensitivity of their instrument to EM noise. To the best of our knowledge, our investigation of taliks below lakes and our comparative measurements on terrestrial permafrost provide, for the first time, essential demonstration of surface NMR capability to detect and quantify liquid water in unfrozen zones of permafrost environments.

2 Study sites

[7] Modeling suggests that older, deeper thermokarst lakes tend to have taliks that extend to greater depths [e.g., West and Plug, 2008]. We therefore selected, for surface NMR study, two thermokarst lakes of different depths, assuming that these lakes are of different age and have different talik geometries. Ace Lake (64.832°, −147.934°) is an approximately 7.5 ha lake with water depth of 8 m, and Caribou Lake (64.879°, −147.765°) is an approximately 1.3 ha lake with a water depth of 1.4 m. A boring drilled at the western shore of Ace Lake indicates a permafrost thickness of approximately 49.5 m [Pewe and Bell, 1974]. The nearest borehole to Caribou Lake is about 1300 m to the NE, with a permafrost thickness of approximately 30 m [Pewe and Bell, 1975)]. In addition, we carried out surface NMR measurements at a terrestrial permafrost site, the Bonanza Creek Long Term Ecological Research forest (“Bonanza Creek”, 64.713°, −148.291°). Borehole BZ2 at Bonanza Creek, approximately 100 m from our measurement site, does not fully penetrate the permafrost and indicates a minimum permafrost thickness of 24 m (V. Romanovsky, personal communication). For this feasibility study, we limited the test sites to the vicinity of Fairbanks, Alaska, USA due to the relative ease of access as well as the presence of ancillary data. All measurements were made in March 2012, when lake surfaces were frozen and most accessible.

3 Methods and results

3.1 Supporting measurements and inversion constraints

[8] Lake ice and water column thicknesses were determined by drilling holes in the ice at Ace Lake (four holes) and Caribou Lake (five holes). Average ice thickness at Ace Lake was 0.63 ± 0.06 m, whereas at Caribou Lake, it was 0.56 ± 0.01 m. Water column thickness under these holes (i.e., base of ice to lake bottom) at Ace Lake ranged from 4.2 to 6.3 m, whereas at Caribou Lake, the water column thickness ranged from 0.2 to 0.8 m. Mechanical probing in the center of Caribou Lake with a steel talik probe resulted in a maximum sediment penetration of 5.9 m below the water surface. A 3 m long sediment core was extracted from Ace Lake with a vibra-corer and gravimetrically analyzed at 0.1 m intervals, revealing the water content of sediments within 3 m of the lake bottom ranged from 19% to 48% by volume (M. Wooller, personal communication). Constraints were applied in the inversion (details in the following sections) based on measured ice thickness and water depth, plus the liquid water content for those layers (0 and 1, respectively). At Bonanza Creek, a depth constraint was used that restricted the known frozen zone (based on Borehole BZ2) to be at least 24 m deep with water content of ≤4% that would be expected given the measurements of Yoshikawa and Overduin [2005] for Fairbanks silt. Water content in the talik was constrained to two standard deviations of the measured value from Ace Lake, and talik thickness remained unconstrained.

[9] A proton precession magnetometer measured the total magnetic field at each site to determine the local Larmor frequency. The corresponding variations in Larmor frequency of ±3 Hz were adjusted during surface NMR data acquisition. We completed ground-based time-domain electromagnetic (TEM) soundings at each site using 10 and 40 m loops to characterize the subsurface resistivity. This information was used in the inversion of the surface NMR data; however, it is also a second line of evidence to support NMR interpretations. Details of the TEM method can be found in Fitterman and Stewart [1986]. The resistivity below Ace Lake was between 35 and 144 Ω m. The near-surface material at Caribou Lake had a resistivity of less than 250 Ω m; however, at a depth of 7.7 m, the resistivity increased to more than 10,000 Ω m. At Bonanza Creek, the resistivity structure alternated between approximately 40 Ω m and more than 4000 Ω m. The TEM inversion results are presented in Figure 1; uncertainty for TEM measurements is only presented for vital interfaces to retain the clarity of the figure.

Figure 1.

Inversion results from surface NMR soundings at Ace Lake (a), Caribou Lake (b), and Bonanza Creek (c). The black line for each sounding is the best-fit model; the gray lines represent models that fit the data nearly as well. Gray dashed line shows the maximum sensitivity depth. Colored boxes indicate inversion constraints for layer thickness and water content obtained from direct measurements. TEM resistivity soundings are plotted as colorbars with select error bars.

3.2 Surface NMR methods and inversion results

[10] To overcome the background EM noise in the Fairbanks urban area, we (a) used a multichannel surface NMR instrument with a noise reference loop for digital noise mitigation, (b) used a figure eight–shaped loop that has intrinsic noise cancellation properties [e.g., Trushkin et al., 1994], and (c) stacked 16 to 60 recordings for 24 to 36 pulse moments to maximize signal to noise ratio, which was more than 3.3 at maximum power for all sites. At Ace Lake and Caribou Lake, 50 m circular figure eight loops were used due to the close proximity to EM noise sources (power lines within 200 m at Ace Lake and within 600 m at Caribou Lake). At Bonanza Creek, a 95 m square loop was used; the straight geometry was easier to set out in the presence of shrubs and the larger loop size was possible due to lower background noise. The orientation of the long axis of the figure eight loops was parallel to the direction of strongest EM noise source (i.e., power lines), whereas noise compensation loops were placed in the direction of any other power lines.

[11] Data processing was done within an open-source surface NMR processing package [Müller-Petke et al., 2012]. Raw data were imported, band-pass filtered, and digital noise compensation was applied. The applied QT inversion scheme fits all pulse moments, signal amplitudes, and T2* values simultaneously (see Müller-Petke and Yaramanci [2010]). Four-layer blocky inversions (an optimization approach for sharp boundaries) were used for the lake data sets because of the expected sharp transitions between stratigraphic layers with different water contents: (1) ice, (2) water, (3) saturated sediments, and (4) permafrost. We estimated the maximum depth of sensitivity for each sounding based on the approach of Christiansen and Auken [2012]. The quality of the fit is assessed with a chi-squared (χ2) statistic that is normalized to the noise in the data (Figure 1). Values closer to 1 are better with a χ2 value of 1 indicating that the model perfectly describes the data whereas χ2 values less than 1 suggest that the model overfits the data (i.e., fitting noise). Because we are using a blocky inversion, it is not possible to overfit the data [e.g., Aster et al., 2005] and therefore we consider all small χ2 values (i.e., ~1) to be good fits. Uncertainty is assessed by displaying models that fit the data approximately equally as well as the best fit (Figure 1); due to the use of constraints, only unconstrained intervals change with the additional model realizations.

[12] At Ace Lake, the data support a model with water content of 32 ± 6% by volume (as observed in the extracted sample) across the depth range of 5.5 to 16 m below the lake surface. We constrained the water content to be 32 ± 6% by volume and allowed the thawed depth to vary. Below 16 m, the data indicate that the water content is about the same or slightly lower with a minimum estimate of 14% by volume (Figure 1a). At Ace Lake, there was a trade-off in acquisition of the surface NMR data between fine vertical sampling and maximum depth of investigation (i.e., using many smaller pulse moments and omitting large pulse moments). As a result, the instrument sensitivity is low at depths greater than 22 m. At Caribou Lake, the best-fit water content for the talik is 39%, but models that fit the data nearly as well suggest that the water content may be as low as 25% (Figure 1b). At depths greater than 4 m below the lake surface, the water content is estimated to be 4%. The resulting best-fit model at Bonanza Creek indicates that the permafrost extends to 33 m, at which point a 14 m thick layer with water content up to 37% is present (Figure 1c). Several models that fit the data approximately equally well indicate a layer of elevated water content, although the thickness and absolute water content has an uncertainty as displayed (Figure 1c). The deepest interval is found to have low water content similar to the top layer.

3.3 Surface NMR sounding curves

[13] The raw surface NMR sounding data were analyzed to determine the initial signal amplitude and decay time of the NMR signal as a function of pulse. Although both of these parameters are used in the inversion, we focus here on the initial amplitude (inset, Figure 2a) because that is a direct indicator of water content, the parameter of primary interest in this study. The surface NMR sounding curves (Figure 2) demonstrate clear differences in subsurface conditions at each location. Both lake sites are characterized by high initial amplitudes at low pulse moments due to the water column at the surface. The influence of lake depth, however, is clear: the 8 m deep Ace Lake (Figure 2a) has higher amplitudes throughout the sounding due to the larger volume of lake water present compared to the 1.4 m deep Caribou Lake (Figure 2b). Both lake site soundings are characterized by a transition to lower amplitudes at larger pulse moments, which generally correspond to greater depths. The terrestrial permafrost site at Bonanza Creek displays a different response with low amplitudes in the very small and high pulse moments (<200 nV for 2 < q < 6 As) whereas the intermediate pulse moments have higher amplitude (Figure 2c). Synthetic models for each data set based on the best-fit inversion results are presented for comparison in Figure 2. The EM noise varies between sites with higher noise at the lake sites near Fairbanks (Figure 2). The uncertainty in the signal amplitude (related to the noise level in the raw data) is greatest at Ace Lake and Caribou Lake and least at Bonanza Creek.

Figure 2.

Surface NMR amplitude values derived from the field data (red circles) and best-fit models (blue lines). Ace Lake (a), Caribou Lake (b), and Bonanza Creek (c). Error bars are shown for the summed squares of the residual (SSE) of the relaxation signal and the 95% confidence interval of the fit. The inset in Figure 2a illustrates an example of the observed NMR relaxation.

4 Discussion

4.1 Sensitivity to unfrozen water content below thermokarst lakes

[14] Surface NMR has been previously deployed on frozen lakes for calibration purposes, using a blocky inversion scheme to capture the sharp ice-water boundary and water-lake-bottom boundary [Müller-Petke et al., 2011]; the sounding curves from that study display a pattern similar to pulse moments of less than 1 As at Ace Lake. Although the authors of that study did not focus on estimating water content of the sublake material, their inversion did resolve the expected low water content of the sublacustrine rock material. Our results indicate that a blocky inversion scheme is optimal for thermokarst lake investigations because of the expected sharp transitions between units with high contrasts in unfrozen water content. These abrupt transitions are better replicated by a robust blocky model rather than a smooth one: between lake ice and lake water, lake water and sublake sediments, and talik sediments and permafrost.

[15] Inversion results from Ace Lake show that the data are fit well by models that include water content approximately equal to site-specific, gravimetrically determined values (Figure 1a). It is useful to review the amplitude data as further confirmation of the sensitivity to sublacustrine water content. We reiterate that because an EM pulse at the Larmor frequency will only excite protons in water molecules (ice has a negligible signal), any signal that is measured must be related to the presence of liquid water. Focusing on the signal amplitude values for large pulse moments corresponding to depths below the lake bottom (Figure 2a), it is clear that the observed NMR signal indicates the detection of unfrozen water. The TEM sounding (Figure 1) corroborates this interpretation because the resistivity values are very similar to the 10 to 100 Ω m (up to several hundred ohm meters depending on lithology) expected for a thermokarst lake talik [e.g., Minsley et al., 2012]. The current size and depth of Ace Lake suggests considerable thaw subsidence and a relatively long period of talik expansion in the ice-rich permafrost of this area. Based on the NMR results, the lake size and depth, and permafrost thickness, we conclude that Ace Lake has either a closed talik that extends deeper than the maximum depth of our measurement (22 m) or an open talik, potentially connecting the lake to subpermafrost water. Our observation agrees with model results for talik depths by West and Plug [2008] for similar ice-rich permafrost and lakes (2–22 m deep, ~350 m diameter) on the Seward Peninsula.

[16] The surface NMR inversion data from Caribou Lake (Figure 1b) best describes a system in which a talik extends to a depth of 4 to 6 m below the shallow lake, and is underlain by permafrost. The amplitude data (Figure 2b) also confirm that these measurements are sensitive to liquid water below the lake water column; however, the large pulse moments have very low amplitude, as might be expected for the low water content associated with permafrost. Forcing low water content in the inversion results in elevated χ2 values, indicating that talik water content of more than 25% is necessary to explain the data. The TEM survey (Figure 1) resolved similar low-resistivity values consistent with lake water and saturated, unfrozen sediments to a depth of less than 8 m. The shallow lake depth suggests minimal subsidence and therefore a potentially small volume of unfrozen sediments [West and Plug, 2008]. Based on the surface NMR data with corroborating evidence from the TEM data and mechanical probing, as well as nearby permafrost thickness data of 30 m [Pewe and Bell, 1975], our interpretation is that Caribou Lake has a shallow, closed talik with permafrost at approximately 5 to 6 m below the lake surface.

4.2 Surface NMR observations of terrestrial permafrost

[17] Measurements at the terrestrial permafrost site, Bonanza Creek, provided an opportunity to compare the surface NMR results to borehole data. At Bonanza Creek, the surface NMR results reveal a layer of elevated water content at approximately 40 m depth bounded above and below by low-water content layers. The NMR estimate of a 24 to 33 m thick frozen layer is compatible with the nearby borehole, which indicates that the permafrost is at least 24 m deep. The TEM measurements indicating a layer of lower resistivity present between two layers of very high resistivity corroborate the surface NMR observations (Figure 1c). It is likely that the total thickness of the permafrost at this location is approximately 34 m, the zone of elevated water content corresponds to a regional groundwater aquifer and the deep material with low water content may be bedrock, or some other material with NMR signals that are too small to detect.

4.3 Modeling of surface NMR response to different types of thawed zones in permafrost

[18] To demonstrate the potential for other surface NMR applications to permafrost research, we present forward models showing the distinctly different surface NMR signal responses that would be acquired in five typical permafrost scenarios (Figure 3a)—permafrost with no thaw features, an open talik beneath a lake, a layer of thin permafrost underlain by a deep groundwater aquifer, a closed talik beneath a lake, and an isolated talik enclosed in permafrost. Features consistent with all five scenarios are widespread in permafrost regions but are difficult to measure and characterize in the field [e.g., Yoshikawa and Hinzman, 2003; Minsley et al., 2012]. Using these scenarios, we show the synthetic sounding curves that would be obtained when using a 100 m loop under conditions of low EM noise (Figure 3b). The synthetic models demonstrate the considerable difference between measurements made on a thermokarst lake or other types of permafrost environments. A clear contrast in the sounding curves is visible between thermokarst lake environments where liquid water is present at the surface and in the lake talik versus a vertically continuous permafrost section (Figure 3b). In addition, the permafrost, deep groundwater, and isolated talik scenarios have considerably smaller amplitude values due to the smaller volume of water present near the surface NMR loop compared with the thermokarst lake scenario. The difference in the character between the isolated talik and deep groundwater scenarios indicates that distinguishing between these subsurface conditions would also be possible.

Figure 3.

Five potential scenarios in permafrost settings (a). Numbers and colors of vertical lines correspond to synthetic surface NMR data (b). Liquid water contents are 4% and 33% for permafrost and talik, respectively.

[19] For these modeled scenarios, the advantages of surface NMR over other geophysical methods are the unambiguous, direct sensitivity to liquid water, and depth of penetration potentially up to 150 m. Although surface NMR is not able to cover large areas as quickly as airborne geophysics due to acquisition times of more than 1 h, surface NMR measurements would be highly valuable in combined measurement schemes in which they could provide constraining data, calibration information, or control points for larger scale (e.g., airborne) surveys. As seen in this study, there is value in having other forms of data available (e.g., TEM and boreholes) to assist with the inversion of the surface NMR data.

5 Conclusions

[20] We demonstrate the ability of surface NMR to detect unfrozen sediments below thermokarst lakes of different depths, estimate the depth of taliks and detect unfrozen sediments below the permafrost. Surface NMR is one of the preferable geophysical investigation methods for detecting unfrozen sediments in permafrost regions because of the unambiguous, direct sensitivity to liquid water. Our NMR inversion approach utilizing sharp transitions between layers is well suited to the sharp boundaries between water, saturated sediments below the water column, and permafrost. Using synthetic forward models, we have demonstrated that surface NMR is highly suitable for thermokarst and permafrost studies, particularly for detecting various types of unfrozen zones in permafrost.


[21] We thank A. Behroozmand for TEM data interpretation, M. Wooller for providing vibracore data, and V. Romanovsky for providing permafrost borehole data. This investigation was funded in part by CUAHSI, NASA Carbon Cycle Sciences NNX08AJ37G, and the U.S. Geological Survey - Alaska Science Center, and the Hydrology Program of the U.S. National Science Foundation (grant 0911234). Thanks to all who helped in the field (B. Cable, C. Arp, A. Kholodov, and A. Gusmeroli). Thanks to B. Minsley for comments that improved the manuscript. We used a Vista Clara GMR for surface NMR measurements, a Scintrex magnetometer, and a WalkTEM (Aarhus University) for electrical resistivity measurements. Any use of trade, product, or firm names is for descriptive purposes only and does not imply endorsement by the U.S. Government.