the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 27 Oct 2025
Talik detection beneath water tracks using three-dimensional Ground-penetrating radar
Abstract. Permafrost is degrading, raising concerns about its impact on global climate. Taliks, year-round unfrozen layers of ground in permafrost environments, affect Arctic hydrological and carbon dynamics and contribute to permafrost degradation. Despite their importance, they are seldom studied because of the difficulty in identifying them. Water tracks (zones overlying permafrost concentrating water flow) have an important role in controlling catchment hydrology, yet their contribution to talik formation remains poorly understood. The objective of this study is to detect and characterize suprapermafrost taliks in three-dimensions at high spatial resolution (dm-scale), with a specific focus on those located beneath water tracks. Fieldwork was conducted in a discontinuous permafrost area near Eight Mile Lake, Alaska (USA), during winter using a Ground-penetrating radar with a 600 MHz antenna. Ground-penetrating radar proved to be a reliable tool for imaging talik depth and extent. While talik tops were clearly detected with a root mean square error of 17 cm, their bottoms were less identifiable due to limited signal penetration. Both isolated and lateral suprapermafrost taliks were observed at the site. Importantly, taliks were more present, shallower and thicker in water tracks than in their adjacent areas. Additionally, thicker snow cover and topographic depressions were significantly associated with shallower taliks (respective significant correlations of -0.65 and 0.53). Water tracks thus appear to be hotspots for talik formation and expansion, with important implications for winter subsurface hydrological connectivity. These findings highlight the need for increased attention to the processes occurring during the winter in the water tracks environments.
- Preprint
(4535 KB) - Metadata XML
- BibTeX
- EndNote
Status: closed
- RC1: 'Comment on egusphere-2025-4667', Anonymous Referee #1, 19 Dec 2025
-
RC2: 'Comment on egusphere-2025-4667', Anonymous Referee #2, 15 Jan 2026
The study presents the acquisition, processing, and interpretation of GPR data to identify talik characteristics. The manuscript is well written and provides valuable reporting of how taliks are distributed in an Arctic environment. I think the research is relevant for the community and appropriate for publication in The Cryosphere. Still, I do think there are several issues that should be addressed before publication.
Major concerns
- I think the interpretation of the processed dataset can sometimes give a false impression to the reader. For example, the uncertainty in the interpretation varies, and in some cases it should be directly mentioned in the caption (or perhaps shown in the figure using a dashed line). The current, limited discussion in the text is not sufficient, in my view (see minor comments).
- Some language used to discuss the GPR data and its processing is approximate and sometimes not in line with common terminology used for GPR processing. Please consider improving imprecise wording, such as using a capital G only for Ground, and using “propagation time” (“two way travel time” or “travel time”). Please also consider adding information such as antenna separation, zero time correction, and related points.
- In Figures 5 to 8, the time axis is not a depth axis, yet it is compared to core data that are depth resolved. The authors adjusted the core depth to overlay it on the transect, but that adjustment is location dependent. Please consider transforming the GPR data to depth, or mention the non-linearity along the transect and show depth-based values for the picked interfaces (in additional figure or supplementary).
Minor comments
Abstract, line 21: “talik top.” Consider clarifying that this corresponds to the bottom of the seasonally freezing and thawing layer (which can fluctuate depending on winter conditions). Referring only to “talik top” can be misleading for the reader.
Line 83: The data represent 2D transects that are later used, after processing, to generate 3D products. The mention of “3D” should be clarified here.
Line 95: This sentence is very general at this point. Consider being more explicit about what causes signal reflections. For example, the sequence of snow ground interface (if present), organic mineral boundary, and the top of permafrost can represent a complex set of possible reflectors. These reflections also lead to reduced energy transmission to deeper layers.
Line 106: “Innovative” sounds too strong here, or explain what is innovative in the GPR processing/application. Still, I do believe the study is of interest because it aims to understand talik distribution.
Line 175: One antenna only? What was the antenna spacing? This information is also relevant for improving the discussion at Line 200 on time zero correction. More technically accurate language could be used. Also consider adding a few words on the expected vertical resolution, and the ability to identify layers at that antenna frequency. Finally, consider discussing additional acquisition and processing limitations.
Line 229: “Teros 12.” Please clarify what it measures and how it is related to dielectric properties.
Figure 5: Time is on the y axis, and depth is used for the cores. Please explain why depth is not shown on the y axis after travel time to depth conversion. I encourage the authors to provide an additional figure to demonstrate the impact of the conversion.
Figure 5c: Please explain the uncertainty in identifying the boundary between frozen active layer and permafrost, versus a change in lithology. Checking polarity reversal or not could help, and/or could be discussed. The interpretation is up to the authors and is part of the process. I would just appreciate to see more nuance regarding what is relatively certain versus what remains quite hypothetical.
Figure 7: Is the top of the permafrost interface hypothetical or quite certain ? Please mention it in the caption or directly on the figure.
Figure 8: Please explain why you believe that outside the main water track, a talik is still present. The reflector is weak there. In Line 379: “Although taliks were difficult to detect outside the water track, they likely extend across the entire grid, considering the deep permafrost detected at the soil core (> 180 cm).” I wonder how strongly the core information influences the decision to pick a reflector there. The image suggests lower attenuation in frozen soil, which feels counterintuitive, although I acknowledge that processing choices could contribute to this behavior on the sides of the talik channel in Figure 8. It is also known that deep taliks can be bordered by very shallow permafrost in some areas. Again, the processing and interpretation choices are up to the authors. I would just appreciate a more nuanced presentation, so the reader can more easily distinguish where the authors consider uncertainty to be high versus where results are more robust.
Citation: https://doi.org/10.5194/egusphere-2025-4667-RC2
Status: closed
-
RC1: 'Comment on egusphere-2025-4667', Anonymous Referee #1, 19 Dec 2025
This manuscript presents ground-penetrating radar (GPR) data collected at a permafrost field site in Alaska. As clearly stated by the authors (lines 17-18), the objective of this study is to detect and characterize unfrozen subsurface layers (taliks) in three-dimensions with a spatial resolution at the decimeter scale. Although the manuscript is well written overall and the application topic of this case study is clearly within the scope of TC, I have some major concerns regarding the geophysical methodology as presented and used in this study. Because of major methodological deficits (including GPR data acquisition, processing, and interpretation), the presented results are, from my point of view, more than questionable and the conclusions are not supported by the presented data. Thus, the manuscript is not acceptable in its present form. In the following, I summarize my major points of criticism in three points. In addition, I provide numerous further comments and examples that should help the authors to realize the limits of their data base and to process and interpret their GPR data following good practice. All these comments should be considered by the authors before resubmitting this manuscript to TC or any other scientific journal.
Major comments
(1) Experimental setup: With the aim of 3D subsurface imaging with a spatial resolution at the decimeter scale (see lines 17-18) the authors claim that they have collected seven 3D GPR data sets at their field site (e.g., line 180). However, considering the spatial sampling of the presented data sets (crossline spacings of 1 m to 2 m; Table A. 1)and the used 600 MHz antennas (resulting in dominant wavelengths of around 0.25 m), the employed crossline sampling is by far to coarse for 3D GPR imaging and the acquired data cannot be regarded as 3D data sets or measurements. It should be noted that the spatial sampling density required for recording proper 3D GPR data (i.e., to avoid spatial aliasing of dipping events including tails of diffraction hyperbolas), are well studied and documented in the geophysical literature (e.g., Grasmueck et al. 2005, Heincke et al. 2005). Thus, the presented data sets do not represent a proper data base to reach the stated objectives of this study.
(2) GPR data processing: Many questions arise around the core GPR processing flow presented in lines 196-205. Many details necessary to follow the general processing strategy and to evaluate individual processing steps are missing. Finally, I have also some major concerns regarding the entire flow; i.e., if the finally processed data allow for an interpretation as presented later under results. I have some doubts. My major concerns are regarding the applied background removal filter (likely removing also horizontal reflection events from the data) and the missing migration step (needed to generate a proper image with correct positions and dips of subsurface structures). Below under “further comments”, I comment these and other processing steps in more detail.
(3) GPR data interpretation: As described in lines 206-212, the authors use horizon tracking for the interpretation of their processed GPR data. In their workflow, they manually identify, select, and pick the target reflection events. Although such an approach is rather common, I have some major problems in understanding and following the author’s interpretation strategy because they do not provide and discuss any criteria used to select a specific reflection event, to follow the selected event, and stop the event tracking. For all picked reflection events shown in Figures 5 to 8 and B. 1 to B. 4, I cannot see a clear and comprehensible strategy for selecting and tracking the highlighted events. Thus, these interpretations (and all derived results and conclusions) are far from being convincing and seem to be biased towards the conceptual subsurface model sought by the authors. Some examples are discussed in more detail below.
Further comments
Line 75: I do not understand why the letter “G” in “Ground-penetrating radar” is always capitalized. This should be corrected.
Lines 83-4 and 99-100: Such statements are not correct because the authors have not collected closely spaced data allowing for 3D GPR imaging (see point 1 under major comments). Consider and realize that your crossline spacing is approximately equal to 5-10 wavelengths and should be in the order of a quarter of a wavelength.
Line 174: It’s good practice to report all fundamental recording parameters. Missing parameters include antenna offset (might be critical, for example, for zero-time correction, time-to-depth conversion, and accurately fitting diffraction hyperbolas), sampling interval, record length, and number of vertical stacks.
Line 176: The unit of frequency is Hertz.
Lines 193-195: I have some doubts that accuracy and the spatial sampling density of the positioning data are sufficient for proper 2D/3D imaging with GPR data collected using 600 MHz antennas (where we expect an accuracy in the order of several cm). The authors use a differential GPS with a sampling frequency of 1 Hz for positioning (lines 175-176). Considering the given walking speed of 1.5 to 2 km/h (line 178) this results in GPS data points with an inline spacing of approximately 0.5 m. Now, the authors exclude low-quality GPS readings and perform moving-average filtering of the remaining data using a kernel size of 10 data points (which corresponds at least to a spatial extension of 5 m). Considering a snowy environment with a typical rough permafrost soil surface and the difficulty in pulling such a sledge (Figure 3) at a constant speed in such an environment, I doubt that the resulting filtered trace positions provide the accuracy needed for detailed 2D or even 3D imaging. This point must be discussed in detail, and the authors have to provide some estimates of the achieved positioning accuracy.
Lines 199-200: The authors should be precise in their formulations; i.e., I don’t see the meaning in the statement “remove the signal before the soil surface”. A zero-time correction is needed because time- zero is unknown. In addition, the authors refer to the “find positive peak method”. I can only guess what’s behind this method and how it is used for zero-time correction. The authors must provide a clear description including a reference (if available) of the methods used.
Lines 200-201: The authors apply a band-pass filter from “120 to 1500 MHz to reduce noise”. I assume that the specified frequency range represents the passband (not precisely stated) and, thus, that they aim at reducing high (>1500 MHz) and low frequency (<120 MHz) noise in their data. The authors must state and briefly discuss how they established these parameters of the used band-pass filter. Is this the bandwidth of the signal they are interested in?
Line 201: The authors state that they use “dewow filtering to remove low-frequency noise”. Because this filter is applied after band-pass filtering (120 to 1500 MHz passband), this suggests that now low-frequencies are characterized by frequencies >100 MHz. This must be clarified. In addition, different approaches of dewow filtering are known from the literature. Because there might be some artifacts introduced by this filtering, the authors have to provide more details including a reference for the approach they implemented.
Lines 201-202: As already indicated above, I have some major concerns regarding background removal filtering. The authors aim at removing horizontal artefacts but do not discuss the problem that also horizontal reflection events might be suppressed or even removed by such a filter. Similar to dewow filtering, different approaches for background removal are known. However, the authors provide no details regarding the strategy they use and, thus, there is no chance to evaluate this processing step. For critical processing steps, good practice includes to illustrate the filter effect by a data example and to discuss the process of parameter testing and selection. I highly recommend to include this for the background removal filter used here.
Lines 202-203: For amplitude scaling, the authors state that they use the “energy decay” method. Again, I can only guess what’s behind this method because different implementations are commonly used. This must be clarified.
Line 203: In addition, to the “energy decay method” the authors apply a constant gain of 16 dB. Later on (e.g., in Figure 5), all amplitudes are normalized between -1 and +1. Although the corresponding normalization step is not further discussed or specified, I don’t see the motivation to apply a constant gain when the amplitudes are normalized afterwards. This must be clarified.
Lines 204-205: The authors state that they “temporarily” apply edge detection based on the Sobel filtering “to facilitate the identification of specific interfaces”. The reader cannot follow how this step supports the interpretation. Thus, more details regarding the implementation and usage of this step are needed and should be supported and illustrated by a data example.
Line 205: As already indicated above, migration represents an essential processing step, especially, when we are aiming at accurate structural imaging. There is a huge amount of literature (including basic geophysical textbooks) that demonstrats the need and benefit of migration, especially, for data collected across complex structural environments. Without migration, I don’t see how to meet the objectives as formulated by the authors.
Lines 229-230: The mentioned “Teros12” sensor and the corresponding data acquisition must be clarified. The authors must provide some information regarding the measurement principle, the resolution capabilities, how they installed it at different depth levels at different field locations, and regarding the used depth sampling interval. Such details are crucial when developing dielectric permittivity and velocity models, respectively, from these data and comparing them to other methods.
Lines 233-244: Here, the authors use “hyperbola fitting” to estimate the dielectric permittivity of different soil layers. Again, they provide no details regarding the used methods and no estimates of uncertainty of the derived permittivity values. A basic hyperbola fitting procedure typically assumes a negligible antenna offset, no lateral velocity variations, a point reflector (much smaller than one wavelength), and that the point reflector is located directly located underneath the profile line. The authors must discuss these assumptions (with respect to their fitting method) and estimate how these assumptions influence the derived permittivity values. In addition, for a layered subsurface, the velocity values estimated using this hyperbola fitting method represent rms velocities and not layer or interval velocities as indicated here (for the frozen and unfrozen soil layers, lines 237 and 241-243). This must be considered and clarified.
Line 262: It’s not clear if the reported value of 25 m represents the radius or diameter of a disc, or the width of a rectangular window. Furthermore, the authors should clarify how this parameter has been selected.
Line 267: The authors use their coring data to distinguish snow, frozen active layer, talik, and permafrost. However, nothing is said regarding sediment composition (e.g., silt or clay content) and how this varies across their field site. A brief discussion of this is needed.
Figure 5 (presentation of GPR data): (1) Here and in the entire manuscript, the authors should follow common GPR terminology; for example, traveltime instead of propagation time. (2) Here (and in all following Figures), the authors present GPR data in terms of normalized amplitudes. However, they do not discuss the applied normalization procedure (e.g., min-max or rms based, trace-based or profile-based). Especially when interpreting reflection amplitudes along horizons (as done by the authors later on; e.g., in lines 314-315), such a normalization might be critical because relative amplitude variations along reflected events might be distorted (in particular after applying a background removal filter). Thus, the authors have to discuss their normalization procedure and its effects on amplitudes in detail.
Figure 5 (interpretation of GPR data): The picked horizons shown in this Figure represent typical examples of GPR data interpretations that I cannot follow and understand. For example, in Figure 5a the dark blue lines (representing top of permafrost) seem to follow basically tails of diffraction hyperbolas that would not be present in properly migrated data. Then, fragments of the same interface are interpreted in Figure 5c. However, I see no reason (except at the coring location) why these reflection elements have been selected (and not, for example, similar deeper features) and picked for lateral distances of a few to several meters (and not further where similar events are visible). Then, I ask why this interface has not been interpreted in Figure 5b where I see numerous similar features as in Figure 5a and 5c. Simialr observations can be made for the other interpreted interfaces and all profile lines presented in this study. Thus, I doubt in the overall validity and reliability of all these interpretations, especially, in view of the overall goal of the authors to characterize this system with a spatial resolution at the decimeter scale.
Line 314: The authors must be precise. What is a “higher reflection”? Higher in amplitudes? Or in depth?
Lines 320-321: With such a vague interpretation (see corresponding points above), such statistical analyses pretending cm accuracy make no sense.
Figure 6b: It’s not clear why the authors select this specific timeslice at 14.3 ns and why they select always different times for all following amplitude slices. This must be clarified. Furthermore, in all of these amplitude timeslices (shown in Figures 6 to 8 and B. 1 to B. 4), isolated anomalies detected only in a single line illustrate the inadequate spatial sampling in the crossline direction, and represent clear indications for spatial aliasing. Thus, and considering my comments regarding data density and spatial aliasing above, these sets of parallel 2D profiles cannot be interpreted in three dimensions.
Figure 6c and 6d: These interpolated horizon maps (as well as all further maps shown in Figures 7 to 8 and B. 1 to B. 4) also clearly illustrate the problem of insufficient spatial sampling. Many “bulls-eye” anomalies represent a clear indication for spatial aliasing. Thus, these maps cannot be regarded as a reliable results and do clearly illustrate the insufficient GPR data base to meet the overall goal of this study (to characterize this permafrost system with a spatial resolution at the decimeter scale).
References
Grasmueck et al. (2005): Full-resolution 3-D GPR imaging. Geophysics, 70, K12–K19.
Heincke et al. (2005): Acquisition and processing strategies for 3D georadar surveying a region characterized by rugged topography. Geophysics, 70, K53-K61.
Citation: https://doi.org/10.5194/egusphere-2025-4667-RC1 -
RC2: 'Comment on egusphere-2025-4667', Anonymous Referee #2, 15 Jan 2026
The study presents the acquisition, processing, and interpretation of GPR data to identify talik characteristics. The manuscript is well written and provides valuable reporting of how taliks are distributed in an Arctic environment. I think the research is relevant for the community and appropriate for publication in The Cryosphere. Still, I do think there are several issues that should be addressed before publication.
Major concerns
- I think the interpretation of the processed dataset can sometimes give a false impression to the reader. For example, the uncertainty in the interpretation varies, and in some cases it should be directly mentioned in the caption (or perhaps shown in the figure using a dashed line). The current, limited discussion in the text is not sufficient, in my view (see minor comments).
- Some language used to discuss the GPR data and its processing is approximate and sometimes not in line with common terminology used for GPR processing. Please consider improving imprecise wording, such as using a capital G only for Ground, and using “propagation time” (“two way travel time” or “travel time”). Please also consider adding information such as antenna separation, zero time correction, and related points.
- In Figures 5 to 8, the time axis is not a depth axis, yet it is compared to core data that are depth resolved. The authors adjusted the core depth to overlay it on the transect, but that adjustment is location dependent. Please consider transforming the GPR data to depth, or mention the non-linearity along the transect and show depth-based values for the picked interfaces (in additional figure or supplementary).
Minor comments
Abstract, line 21: “talik top.” Consider clarifying that this corresponds to the bottom of the seasonally freezing and thawing layer (which can fluctuate depending on winter conditions). Referring only to “talik top” can be misleading for the reader.
Line 83: The data represent 2D transects that are later used, after processing, to generate 3D products. The mention of “3D” should be clarified here.
Line 95: This sentence is very general at this point. Consider being more explicit about what causes signal reflections. For example, the sequence of snow ground interface (if present), organic mineral boundary, and the top of permafrost can represent a complex set of possible reflectors. These reflections also lead to reduced energy transmission to deeper layers.
Line 106: “Innovative” sounds too strong here, or explain what is innovative in the GPR processing/application. Still, I do believe the study is of interest because it aims to understand talik distribution.
Line 175: One antenna only? What was the antenna spacing? This information is also relevant for improving the discussion at Line 200 on time zero correction. More technically accurate language could be used. Also consider adding a few words on the expected vertical resolution, and the ability to identify layers at that antenna frequency. Finally, consider discussing additional acquisition and processing limitations.
Line 229: “Teros 12.” Please clarify what it measures and how it is related to dielectric properties.
Figure 5: Time is on the y axis, and depth is used for the cores. Please explain why depth is not shown on the y axis after travel time to depth conversion. I encourage the authors to provide an additional figure to demonstrate the impact of the conversion.
Figure 5c: Please explain the uncertainty in identifying the boundary between frozen active layer and permafrost, versus a change in lithology. Checking polarity reversal or not could help, and/or could be discussed. The interpretation is up to the authors and is part of the process. I would just appreciate to see more nuance regarding what is relatively certain versus what remains quite hypothetical.
Figure 7: Is the top of the permafrost interface hypothetical or quite certain ? Please mention it in the caption or directly on the figure.
Figure 8: Please explain why you believe that outside the main water track, a talik is still present. The reflector is weak there. In Line 379: “Although taliks were difficult to detect outside the water track, they likely extend across the entire grid, considering the deep permafrost detected at the soil core (> 180 cm).” I wonder how strongly the core information influences the decision to pick a reflector there. The image suggests lower attenuation in frozen soil, which feels counterintuitive, although I acknowledge that processing choices could contribute to this behavior on the sides of the talik channel in Figure 8. It is also known that deep taliks can be bordered by very shallow permafrost in some areas. Again, the processing and interpretation choices are up to the authors. I would just appreciate a more nuanced presentation, so the reader can more easily distinguish where the authors consider uncertainty to be high versus where results are more robust.
Citation: https://doi.org/10.5194/egusphere-2025-4667-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 245 | 125 | 28 | 398 | 15 | 20 |
- HTML: 245
- PDF: 125
- XML: 28
- Total: 398
- BibTeX: 15
- EndNote: 20
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
This manuscript presents ground-penetrating radar (GPR) data collected at a permafrost field site in Alaska. As clearly stated by the authors (lines 17-18), the objective of this study is to detect and characterize unfrozen subsurface layers (taliks) in three-dimensions with a spatial resolution at the decimeter scale. Although the manuscript is well written overall and the application topic of this case study is clearly within the scope of TC, I have some major concerns regarding the geophysical methodology as presented and used in this study. Because of major methodological deficits (including GPR data acquisition, processing, and interpretation), the presented results are, from my point of view, more than questionable and the conclusions are not supported by the presented data. Thus, the manuscript is not acceptable in its present form. In the following, I summarize my major points of criticism in three points. In addition, I provide numerous further comments and examples that should help the authors to realize the limits of their data base and to process and interpret their GPR data following good practice. All these comments should be considered by the authors before resubmitting this manuscript to TC or any other scientific journal.
Major comments
(1) Experimental setup: With the aim of 3D subsurface imaging with a spatial resolution at the decimeter scale (see lines 17-18) the authors claim that they have collected seven 3D GPR data sets at their field site (e.g., line 180). However, considering the spatial sampling of the presented data sets (crossline spacings of 1 m to 2 m; Table A. 1)and the used 600 MHz antennas (resulting in dominant wavelengths of around 0.25 m), the employed crossline sampling is by far to coarse for 3D GPR imaging and the acquired data cannot be regarded as 3D data sets or measurements. It should be noted that the spatial sampling density required for recording proper 3D GPR data (i.e., to avoid spatial aliasing of dipping events including tails of diffraction hyperbolas), are well studied and documented in the geophysical literature (e.g., Grasmueck et al. 2005, Heincke et al. 2005). Thus, the presented data sets do not represent a proper data base to reach the stated objectives of this study.
(2) GPR data processing: Many questions arise around the core GPR processing flow presented in lines 196-205. Many details necessary to follow the general processing strategy and to evaluate individual processing steps are missing. Finally, I have also some major concerns regarding the entire flow; i.e., if the finally processed data allow for an interpretation as presented later under results. I have some doubts. My major concerns are regarding the applied background removal filter (likely removing also horizontal reflection events from the data) and the missing migration step (needed to generate a proper image with correct positions and dips of subsurface structures). Below under “further comments”, I comment these and other processing steps in more detail.
(3) GPR data interpretation: As described in lines 206-212, the authors use horizon tracking for the interpretation of their processed GPR data. In their workflow, they manually identify, select, and pick the target reflection events. Although such an approach is rather common, I have some major problems in understanding and following the author’s interpretation strategy because they do not provide and discuss any criteria used to select a specific reflection event, to follow the selected event, and stop the event tracking. For all picked reflection events shown in Figures 5 to 8 and B. 1 to B. 4, I cannot see a clear and comprehensible strategy for selecting and tracking the highlighted events. Thus, these interpretations (and all derived results and conclusions) are far from being convincing and seem to be biased towards the conceptual subsurface model sought by the authors. Some examples are discussed in more detail below.
Further comments
Line 75: I do not understand why the letter “G” in “Ground-penetrating radar” is always capitalized. This should be corrected.
Lines 83-4 and 99-100: Such statements are not correct because the authors have not collected closely spaced data allowing for 3D GPR imaging (see point 1 under major comments). Consider and realize that your crossline spacing is approximately equal to 5-10 wavelengths and should be in the order of a quarter of a wavelength.
Line 174: It’s good practice to report all fundamental recording parameters. Missing parameters include antenna offset (might be critical, for example, for zero-time correction, time-to-depth conversion, and accurately fitting diffraction hyperbolas), sampling interval, record length, and number of vertical stacks.
Line 176: The unit of frequency is Hertz.
Lines 193-195: I have some doubts that accuracy and the spatial sampling density of the positioning data are sufficient for proper 2D/3D imaging with GPR data collected using 600 MHz antennas (where we expect an accuracy in the order of several cm). The authors use a differential GPS with a sampling frequency of 1 Hz for positioning (lines 175-176). Considering the given walking speed of 1.5 to 2 km/h (line 178) this results in GPS data points with an inline spacing of approximately 0.5 m. Now, the authors exclude low-quality GPS readings and perform moving-average filtering of the remaining data using a kernel size of 10 data points (which corresponds at least to a spatial extension of 5 m). Considering a snowy environment with a typical rough permafrost soil surface and the difficulty in pulling such a sledge (Figure 3) at a constant speed in such an environment, I doubt that the resulting filtered trace positions provide the accuracy needed for detailed 2D or even 3D imaging. This point must be discussed in detail, and the authors have to provide some estimates of the achieved positioning accuracy.
Lines 199-200: The authors should be precise in their formulations; i.e., I don’t see the meaning in the statement “remove the signal before the soil surface”. A zero-time correction is needed because time- zero is unknown. In addition, the authors refer to the “find positive peak method”. I can only guess what’s behind this method and how it is used for zero-time correction. The authors must provide a clear description including a reference (if available) of the methods used.
Lines 200-201: The authors apply a band-pass filter from “120 to 1500 MHz to reduce noise”. I assume that the specified frequency range represents the passband (not precisely stated) and, thus, that they aim at reducing high (>1500 MHz) and low frequency (<120 MHz) noise in their data. The authors must state and briefly discuss how they established these parameters of the used band-pass filter. Is this the bandwidth of the signal they are interested in?
Line 201: The authors state that they use “dewow filtering to remove low-frequency noise”. Because this filter is applied after band-pass filtering (120 to 1500 MHz passband), this suggests that now low-frequencies are characterized by frequencies >100 MHz. This must be clarified. In addition, different approaches of dewow filtering are known from the literature. Because there might be some artifacts introduced by this filtering, the authors have to provide more details including a reference for the approach they implemented.
Lines 201-202: As already indicated above, I have some major concerns regarding background removal filtering. The authors aim at removing horizontal artefacts but do not discuss the problem that also horizontal reflection events might be suppressed or even removed by such a filter. Similar to dewow filtering, different approaches for background removal are known. However, the authors provide no details regarding the strategy they use and, thus, there is no chance to evaluate this processing step. For critical processing steps, good practice includes to illustrate the filter effect by a data example and to discuss the process of parameter testing and selection. I highly recommend to include this for the background removal filter used here.
Lines 202-203: For amplitude scaling, the authors state that they use the “energy decay” method. Again, I can only guess what’s behind this method because different implementations are commonly used. This must be clarified.
Line 203: In addition, to the “energy decay method” the authors apply a constant gain of 16 dB. Later on (e.g., in Figure 5), all amplitudes are normalized between -1 and +1. Although the corresponding normalization step is not further discussed or specified, I don’t see the motivation to apply a constant gain when the amplitudes are normalized afterwards. This must be clarified.
Lines 204-205: The authors state that they “temporarily” apply edge detection based on the Sobel filtering “to facilitate the identification of specific interfaces”. The reader cannot follow how this step supports the interpretation. Thus, more details regarding the implementation and usage of this step are needed and should be supported and illustrated by a data example.
Line 205: As already indicated above, migration represents an essential processing step, especially, when we are aiming at accurate structural imaging. There is a huge amount of literature (including basic geophysical textbooks) that demonstrats the need and benefit of migration, especially, for data collected across complex structural environments. Without migration, I don’t see how to meet the objectives as formulated by the authors.
Lines 229-230: The mentioned “Teros12” sensor and the corresponding data acquisition must be clarified. The authors must provide some information regarding the measurement principle, the resolution capabilities, how they installed it at different depth levels at different field locations, and regarding the used depth sampling interval. Such details are crucial when developing dielectric permittivity and velocity models, respectively, from these data and comparing them to other methods.
Lines 233-244: Here, the authors use “hyperbola fitting” to estimate the dielectric permittivity of different soil layers. Again, they provide no details regarding the used methods and no estimates of uncertainty of the derived permittivity values. A basic hyperbola fitting procedure typically assumes a negligible antenna offset, no lateral velocity variations, a point reflector (much smaller than one wavelength), and that the point reflector is located directly located underneath the profile line. The authors must discuss these assumptions (with respect to their fitting method) and estimate how these assumptions influence the derived permittivity values. In addition, for a layered subsurface, the velocity values estimated using this hyperbola fitting method represent rms velocities and not layer or interval velocities as indicated here (for the frozen and unfrozen soil layers, lines 237 and 241-243). This must be considered and clarified.
Line 262: It’s not clear if the reported value of 25 m represents the radius or diameter of a disc, or the width of a rectangular window. Furthermore, the authors should clarify how this parameter has been selected.
Line 267: The authors use their coring data to distinguish snow, frozen active layer, talik, and permafrost. However, nothing is said regarding sediment composition (e.g., silt or clay content) and how this varies across their field site. A brief discussion of this is needed.
Figure 5 (presentation of GPR data): (1) Here and in the entire manuscript, the authors should follow common GPR terminology; for example, traveltime instead of propagation time. (2) Here (and in all following Figures), the authors present GPR data in terms of normalized amplitudes. However, they do not discuss the applied normalization procedure (e.g., min-max or rms based, trace-based or profile-based). Especially when interpreting reflection amplitudes along horizons (as done by the authors later on; e.g., in lines 314-315), such a normalization might be critical because relative amplitude variations along reflected events might be distorted (in particular after applying a background removal filter). Thus, the authors have to discuss their normalization procedure and its effects on amplitudes in detail.
Figure 5 (interpretation of GPR data): The picked horizons shown in this Figure represent typical examples of GPR data interpretations that I cannot follow and understand. For example, in Figure 5a the dark blue lines (representing top of permafrost) seem to follow basically tails of diffraction hyperbolas that would not be present in properly migrated data. Then, fragments of the same interface are interpreted in Figure 5c. However, I see no reason (except at the coring location) why these reflection elements have been selected (and not, for example, similar deeper features) and picked for lateral distances of a few to several meters (and not further where similar events are visible). Then, I ask why this interface has not been interpreted in Figure 5b where I see numerous similar features as in Figure 5a and 5c. Simialr observations can be made for the other interpreted interfaces and all profile lines presented in this study. Thus, I doubt in the overall validity and reliability of all these interpretations, especially, in view of the overall goal of the authors to characterize this system with a spatial resolution at the decimeter scale.
Line 314: The authors must be precise. What is a “higher reflection”? Higher in amplitudes? Or in depth?
Lines 320-321: With such a vague interpretation (see corresponding points above), such statistical analyses pretending cm accuracy make no sense.
Figure 6b: It’s not clear why the authors select this specific timeslice at 14.3 ns and why they select always different times for all following amplitude slices. This must be clarified. Furthermore, in all of these amplitude timeslices (shown in Figures 6 to 8 and B. 1 to B. 4), isolated anomalies detected only in a single line illustrate the inadequate spatial sampling in the crossline direction, and represent clear indications for spatial aliasing. Thus, and considering my comments regarding data density and spatial aliasing above, these sets of parallel 2D profiles cannot be interpreted in three dimensions.
Figure 6c and 6d: These interpolated horizon maps (as well as all further maps shown in Figures 7 to 8 and B. 1 to B. 4) also clearly illustrate the problem of insufficient spatial sampling. Many “bulls-eye” anomalies represent a clear indication for spatial aliasing. Thus, these maps cannot be regarded as a reliable results and do clearly illustrate the insufficient GPR data base to meet the overall goal of this study (to characterize this permafrost system with a spatial resolution at the decimeter scale).
References
Grasmueck et al. (2005): Full-resolution 3-D GPR imaging. Geophysics, 70, K12–K19.
Heincke et al. (2005): Acquisition and processing strategies for 3D georadar surveying a region characterized by rugged topography. Geophysics, 70, K53-K61.