Time-of-flight monitoring reveals higher sediment redistribution rates related to burrowing animals than previously assumed

Main concerns

I like this study and I like the presentation of this novel method to measure soil redistribution. But…I am not sure I agree with all aspects of the application of this technique and the emphasis in terms of objectives in this study.

The authors had two study sites, with different climates, and presumably different species, and I think more could be made of this. How does sediment redistribution differ between the arid and the Mediterranean site, and what is the relative contribution of animal burrowing vs rainfall (which both presumably differ between the two sites) to this? This is an interesting question that the authors sort of get at, but not quite as explicitly, i.e. really teasing the two contributions apart within the two sites. To me that should almost be the main research question and focus. I.e. how effective are animals vs rainfall at moving sediment short distances (the authors are not following the sediment all the way down the slope after all)? I suggest splitting the paper, with one paper being a short methodological paper, with all the technical details of the method, and accuracy testing (I like how that was approached). And the other focusing on this comparison between arid and Mediterranean, using a clearly phrased research question, such as the one that I suggest. But, I would be careful in how to approach answering this question.

The authors tease apart these two contributions by looking at sediment redistribution within areas affected and not affected by burrowing animals and also during and after rainfall. When I read the section about the burrowed and non-burrowed sites the first time, I assumed some of the hillslopes were burrowed and some were not. But then I realized “affected” and “not affected” is at a much smaller scale, with the unaffected areas being in between and in close proximity to the burrows. How unaffected are the “unaffected” sites at this small scale then really? Surely the burrowing animals walk around in between their burrows, trampling the soil. Essentially, is the sediment redistribution in the non-affected areas representative of sites that do not have burrowing animals at all? If we took away the animals, would the situation in the non-affected sites be representative of how much sediment was redistributed? This has to be discussed a bit or acknowledged if there is doubt.  

I also had concerns with upscaling, both spatially (to the size of the hillslope) and temporally (to a year). From a spatial perspective, I could not find anywhere how large a “scan” was, i.e. what are they upscaling from? Or how uniform the burrow sizes at any given point in time were? Burrows in the landscape are presumably at different stages of creation – most natural systems are dynamic systems with burrows being created and destroyed by rainfall continuously. What about other non-burrowed features in the landscape? Big rocks or trees etc. Are the scans and the non-burrowed areas within them really representative of the landscape? Can one just upscale from one burrow to a whole slope of burrows? I also don’t think it’s necessary to upscale to answer the interesting questions in this study (see paragraph 1 of my comments for what I think this is). That can be done at the smaller scale of individual burrows and non-burrowed areas (with some justification of why the non-burrowed areas are representative and giving careful thought regarding what to use as replication). In terms of the temporal upscaling (from seven months to a year), I am also not convinced. Burrowing animals do not continue burrowing at the same pace. And presumably the authors did not catch the burrowers right in the beginning, when the burrowing first started and was at its fastest. At the most the size of the mound gives an estimate of the minimum volume of soil contributed to the hillslope. How about seasonal effects? Some species dig burrows to nest, but are not fossorial, spending their time thereafter aboveground. I was also missing information on who these burrowers are. What sort of animals are we talking about? I assume it’s not the same species in the arid and the Mediterranean system? Or are there multiple species in both systems? This would influence how realistic upscaling is. Again, I also don’t think they need to upscale temporally to compare the two systems.

To summarise: I suggest two papers: one methods paper and one focusing on one main research question comparing the two systems (see my paragraph 1). Other interesting research questions that can potentially be answered with the data set are listed below.

1. How variable are the burrow sizes in this landscape?
2. Does this differ between the arid and Mediterranean system?
3. How fast do the animals burrow?
4. How variable is this?
5. How does this change over time?
6. How fast do the burrows deteriorate after a rainfall event?

Most of these questions are more ecological in nature (but can certainly be phrased in the context of biogeomorphology). I understand that this was not the intention of the article, and the focus is meant to be far more geomorphological, contributing to hillslope erosion modelling. But I am concerned that they cannot really contribute as much from an erosion modelling/geomorph perspective as they think, given my concerns with upscaling. I do see many interesting biogeo questions that can be answered though…

Some smaller issues

• What is meant with autonomous in this context? Do they mean automated?
• Lines 49 to 50: This is exactly where it would be interesting to tease apart how much of this was a result of burrowing and how much was rainfall? Both the rainfall and the burrowing species presumably differ between the systems. This is an interesting question to phrase the whole project around. At the moment, the fact that they had two different systems is almost an “aside”.
• Consider adding a study species section after the study site section.
• Is Figure 4 necessary?
• Line 403: Exemplarily is not a word.
• I have not commented on the results and discussion sections, as I think these will change if the focus of the study changes and the paper splits, and the upscaling is taken out.
• In general, I thought the article was quite long, with many relatively complicated diagrams to follow. Can this be simplified? I think splitting the paper in two will already help with this.

The study illustrates an interesting application of time-of-flight cameras in geomorphology. Furthermore, it very well highlights the potential of custom build sensor systems with simple components (i.e. Pi) with full automatic/autonomous capabilities. The manuscript is well structured and easy to follow. I agree with the first reviewer that the already captured data entails enough novel information to present in ESurf. However, some issues regarding the methods should be addressed in more detail, which are displayed in detail below:

Chapter 1: What is actually a low-cost ToF? No prices (or at least rough estimates) are mentioned and therefore the statement low-cost is not possible to assess. The authors discuss the drawback of laserscanning, as being a lot more expensive. However, laserscanning also reaches a lot farther compared to the ToF cameras. Therefore, the types of studies that can be performed are not relatable due to the different observation scales. The authors miss mentioning time-lapse photogrammetry as another already applied low-cost (as even track-cameras might be used) topographic monitoring technique that can be applied at different observation distances and thus scales (e.g. James et al., 2014 and Galland et al., 2016 – volcanology, Eltner et al., 2017 – soil erosion, Mallalieu et al., 2017 – glacier, Kromer et al., 2019 and Blanch et al., 2021 – rock falls).

Line 162-165: I find the explanation of the pulsed ToF principle confusing. It should be added that the receiver is opening the first window simultaneously and synchronised with the pulse emission, i.e. the receiver opening the window with the same delta t as the emitted pulse. And then the second window is opened, for the same duration delta t, synchronised with the closing of the first window. Thus, the captured photon number (i.e. measured by electrical charge) in both windows can be related according to equation 1 (the higher g1 the shorter the distance) to solve for the distance, which can also be considered as solving for the phase shift and thus solving the ToF. Maybe, the authors can also shortly mention that in general the ToF cameras rely on the principle of measuring the phase shift and that there are different options to modulate the light source to be able to measure a phase shift, e.g. the camera in this study using pulsed modulation.

Line 172-173: The spatial resolution also depends on the orientation of the camera. The more oblique the perspective, the more the variation.

Line 174-175: The point cloud can be both binary and encoded. The authors are actually describing a cloud stored in a binary format being transformed into an ASCII (?) encoded data format.

Line 176-179: As I understand this, the centre of the camera sensor defines the origin of the local, 3D Cartesian coordinate system?

Equation 2: What is actually wrong with the original Z-value? Are the authors aiming to transform the measurements to a local coordinate system, where the X-Y-axes are parallel to the soil surface (for the subsequent transformation of Z-values to a 2.5D dataset)? If yes, would a simple rigid body transformation not be enough? Furthermore, why is the distance of a distance, i.e. distance(y₁-y_i), calculated? Do you mean solely (y₁-y_i)? Also, if the authors refer to the distance of the origin, thus the radius, I would suggest to use r_xy (sqrt(sqr(x₁-x_i)+sqr(y₁-y_i))) instead of y. This causes confusion, as y is already explained as the y-coordinate. How did the authors calculate the angles and with what accuracies? This seems to be tricky in the field.

Equation 3: Why did the authors choose the scaling of 1 standard deviation and not e.g. 1,5 or 2?

Chapter 3.3: Why did the authors not compare the ToF data in the lab experiment with SfM (sub-mm accuracy at that close range possible) or a triangulation based LiDAR (µm accuracy possible)? Such references allow the assessment of spatially distributed errors or potential spatially correlated errors. If the authors use SfM they could have also done an accuracy assessment outside under the actual observation conditions.

Line 209-211: I understand that the authors average the data from several subsequent scans to reduce the noise, assuming a random (Gaussian distributed) error. However, in regard of the accuracy estimation, I would suggest to display the standard deviation, also spatially distributed, to get a better grasp on the variation of each scan.

Line 211-212: How did the author assure a smooth surface? What was the surface made of? I suggest, instead of using the standard-deviation of the Z-coordinate as error estimate (which will be overestimated if the surface is tilted), to fit a plane into the point cloud and calculate the distance to that surface to get the variation in the distance measurements.

Line 223: Please, also display the standard deviation to assess the random error and potentially display a boxplot to better illustrate the inherent variability in your method as you have 45 measurements allowing for such a display.

Chapter 3.5: The choice of the parameters to derive the entrance or mound seem arbitrary. The motivation and reasoning for the choices as well as the defined thresholds should be explained in more detail.

Line 266-267: What is the spatial resolution of the DSM?

Line 274: Why 16 squares and what was their size?

Line 297-298: What is the standard deviation of the five scans? This could also be used to assess the accuracy of the measurements?

Line 323: Why five scans? Did the authors test that at this number, accuracy does not increase much more after averaging? Or is this due to storage or power consumption?

Chapter 3.7: How did the author ensure that there is no mixture/overlap of different processes, e.g. erosion due to rainfall happening shortly after digging?

Line 337: What machine learning algorithm was used? After checking the cited paper, I understood that an in the other study trained random forest was used again in this study. I would suggest to add this information; thus others can follow the manuscript without needing to check the references. 

Equation 7: What is M? Did you mean Vol?

Equation 7-10: The authors observed the sites solely for 7 months and upscale then to yearly changes. Can this be done so easily. For instance, at the Mediterranean site at least a full year should be observed to capture all the seasons. For the desert site the observation period would need to be even longer.

Chapter 3.8: Did the authors perform any validation of their up-scaled data, e.g. by leaving out some samples for testing?

Figure 4: Please, also state the standard deviation because it looks high according to the scatterplot.

Line 590-592: Why would more sporadic measurements be less reliable? The cumulative signal can be more significant than the more frequent measurements with smaller signal to noise ratio.

References

Blanch, X.; Eltner, A.; Guinau, M.; Abellan, A. Multi-Epoch and Multi-Imagery (MEMI) Photogrammetric Workflow for Enhanced Change Detection Using Time-Lapse Cameras. Remote Sens. 2021, 13, 1460.

Eltner, A., Kaiser, A., Abellan, A., Schindewolf, M., 2017. Time lapse structure from motion photogrammetry for continuous geomorphic monitoring. Earth Surf. Process. Landf. 42 (14), 2240–2253.

Galland, O., Bertelsen, H.S., Guldstrand, F., Girod, L., Johannessen, R.F., Bjugger, F., … Mair, K., 2016. Application of open-source photogrammetric software MicMac for monitoring surface deformation in laboratory models. J. Geophys. Res. Solid Earth 121 (4), 2852–2872.

James, M. R. and Robson, S.: Sequential digital elevation models of active lava flows from ground-based stereo time-lapse imagery, ISPRS J. Photogramm. Rem. Sens., 97, 160–170, doi:10.1016/j.isprsjprs.2014.08.011, 2014b.

Kromer, R.; Walton, G.; Gray, B.; Lato, M.; Group, R. Development and optimization of an automated fixed-location time lapse photogrammetric rock slope monitoring system. Remote Sens. 2019, 11, 1890.

Mallalieu, J., Carrivick, J.L., Quincey, D.J., Smith, M.W., James, W.H.M., 2017. An integrated structure-from-motion and time-lapse technique for quantifying ice-margin dynamics. J. Glaciol. 63 (242), 937–949.