Characterizing Spatial Structures of Field-Scale Snowpack using Unpiloted Aerial System (UAS) Lidar and SfM Photogrammetry

Cho, Eunsang; Verfaillie, Megan; Jacobs, Jennifer M.; Hunsaker, Adam G.; Sullivan, Franklin B.; Palace, Michael; Wagner, Cameron

doi:https://doi.org/10.5194/egusphere-2024-1530

Preprints

https://doi.org/10.5194/egusphere-2024-1530

Preprints

21 Jun 2024

| 21 Jun 2024

Characterizing Spatial Structures of Field-Scale Snowpack using Unpiloted Aerial System (UAS) Lidar and SfM Photogrammetry

Eunsang Cho, Megan Verfaillie, Jennifer M. Jacobs, Adam G. Hunsaker, Franklin B. Sullivan, Michael Palace, and Cameron Wagner

Abstract. Uncrewed Aerial Systems (UAS) lidar and structure-from-motion (SfM) photogrammetry have emerged as viable methods to map high-resolution snow depths (~1 m). These technologies enable a better understanding of snowpack spatial structure and its evolution over time, advancing hydrologic and ecological applications. In this study, a series of UAS lidar/SfM snow depth maps were collected during the 2020/21 winter season in Durham, New Hampshire, USA with three objectives: (1) quantifying UAS lidar/SfM snow depth retrieval performance using multiple in-situ measurement techniques (magnaprobe and field cameras), (2) conducting a quantitative comparison of lidar and SfM snow depths (< 35 cm) throughout the winter, and (3) better understanding the spatial structure of snow depth and its relationship with terrain features. The UAS surveys were conducted over approximately 0.35 km² including both open fields and a mixed forest. In the field, lidar had a lower error than SfM compared to in-situ observations with a Mean Absolute Error (MAE) of 3.0 cm for lidar and 5.0–14.3 cm for SfM. In the forest, SfM greatly overestimated snow depths compared to lidar (lidar MAE = 2.7–7.3 cm, SfM MAE = 32.0–44.7 cm). Even though snow depth differences between the magnaprobe and field cameras were found, they had only a modest impact on the UAS snow depth validation. Using the concept of temporal stability, we found that the spatial structure of snow depth captured by lidar was generally consistent throughout the period indicating a strong influence from static land characteristics. Considering all areas (forest and fields), the spatial structure of snow depth was primarily influenced by vegetation type (e.g., fields, deciduous, and coniferous forests). Within the field, the spatial structure was primarily correlated with slope and forest canopy shadowing effects.

Received: 23 May 2024 – Discussion started: 21 Jun 2024

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this preprint. The responsibility to include appropriate place names lies with the authors.

Download & links

Preprint (PDF, 2256 KB)

Supplement (159 KB)

Download & links

Eunsang Cho, Megan Verfaillie, Jennifer M. Jacobs, Adam G. Hunsaker, Franklin B. Sullivan, Michael Palace, and Cameron Wagner

Status: final response (author comments only)

RC1:
'Comment on egusphere-2024-1530', Anonymous Referee #1, 09 Jul 2024
Summary
This study introduces a novel dataset of snow depth (HS) observations for a small study site with mixed vegetation and open, non-vegetated areas. HS was measured using both, LiDAR and SfM systems mounted on a UAV at different dates during one snow season. The study further evaluates the HS maps against different manual in-situ reference measurements and compares the LiDAR and SfM products against each other. The dataset is finally analysed using the relative distance concept and by comparing the mean relative distance (MRD) to 5 different spatial features describing soil and terrain characteristics.
Novel datasets of HS distribution using UAV-based remote sensing are strongly encouraged and of great importance for the hydrologic community (especially if the presented data sets will be made publicly available). Only a few other data sets exist that provide repeated, high-resolution HS observations from UAVs. For cold-maritime environments in in the north-eastern US, no comparable data sets exist (Except for the data set presented for the same study site by Jennifer M. Jacobs et al. (2020). The data therefore has the potential to provide new insights into the specific drivers for the spatial distribution of HS in such environments and is scientifically significant. The performed analysis, however, lacks a clear scientific aim and does not give new insights. The findings of the manuscript, that i) SfM provides less reliable observations with larger data gaps underneath the forest canopy and ii) HS distribution being influenced by topography and vegetation are not new to the scientific community. See for instance Harder et al. (2020) that performed a comprehensive comparison of SfM vs LiDAR HS maps derived from UAVs and Mazzotti et al. (2023) that discussed the different drivers of HS variability at high spatial resolutions. The presented method moreover lacks a co-registration of the snow-off and snow-on maps. I see the greatest potential of your manuscript and data to further discuss the relative difference maps and analyse their temporal stability quantitatively. Understanding the consistency of spatial HS patterns is an important aspect of current snow research (Geissler et al., 2023; Pflug & Lundquist, 2020).
Major Comments:
Overall Style:
The manuscript contains erroneous links to figures and references.

The manuscript overall has a good structure but could benefit from a careful proofreading (See also minor comments) and a more scientific writing style.

Some figures are not satisfactory and need to be improved (See minor comments).

Method:
I encourage the authors to perform a co-registration of the snow-off and snow-on surveys. If there are no snow-free areas within the snow-on surveys to perform a co-registration with, maybe a co-registration of the point clouds (or at least the elevation models) for each survey using SfM and LiDAR could help to reduce systematic over-/underestimation of the products. Check https://github.com/jgenvironment/cluster_snow for ideas on how to perform a three-dimensional co-registration of point clouds using vegetation. Did you check whether your snow-off surveys are comparable between both systems?

Subsequently, I would be interested in a transect – plot (of the underlying point clouds) that could give a better understanding of i) where the differences between SfM and LiDAR originate and ii) give an idea of the sub-canopy point density of the two products. Providing the point densities of the different products for the different surveys is essential. (Overall, forest, open).

I suggest removing the different validation strategies (cameras vs. magnaprobe) from your manuscript, as the number of camera locations is not sufficient to give reliable results. Moreover, as you show, the differences are negligible. Your magnaprobe measurements are important to provide the overall accuracies of your data sets but are too small in number to provide reliable insights into potential deficits of your HS maps.

Instead, you could further examine the relative difference maps, discuss their temporal stability quantitatively and relate areas of persistent relative differences with areas of varying relative differences to your topographic and soil features. This is where I see the greatest potential of your work.

Secondly, I would try to work out explicitly where and when SfM is a suited method to measure HS and what strengths and weaknesses of this sensor are. Did you find some features/flight conditions that impacted the SfM more than LiDAR?

Introduction and Discussion:
As noted above, I would focus your work i) on the comparison of SfM and LiDAR and ii) the subsequent analysis of relative differences, including the quantitative assessment of temporal stability. For i), I would explicitly state for which conditions you can recommend using SfM. Obviously, your results show that SfM is not capable of measuring sub-canopy snow (r^2 = 0.01 and MAE being almost three times as high as the mean HS, see Figure 3). I am missing a discussion on potential applications where choosing SfM could be reasonable. Is SfM suited to measure shallow snowpacks? Maybe analyse your difference-maps also with regard to your topographic and soil features! For ii) a more thorough introduction and discussion of literature working with snow distribution pattern is required (e.g. (Geissler et al., 2023; Pflug et al., 2021; Revuelto et al., 2020; Sturm & Wagner, 2010; Vögeli et al., 2016))

Minor Comments
Title: Here and throughout the entire manuscript. I would avoid the word ‘structures’ in this context as snow structures could also refer to the microstructure (e.g. grain size, type or specific surface area) of the snow. Maybe use pattern or distribution instead.
L15: LiDAR not introduced and make sure that you use the same abbreviations throughout your manuscript (lidar vs LiDAR).
L23: This sentence is not clear, you have only a very few measurements from the cameras and the measurements were taken at different locations. Thus, it is not surprising that they differ. I would skip this analysis.
L42: Repetition of L30ff.
L43: I would improve this section with a more thorough literature review. From what Geissler et al. (2023) and Pflug and Lundquist (2020) showed, patterns are rather stable.
L56: …and most sensors cannot measure through the canopy of trees.
L64-65: The sensors can measure the HS, not the interactions between the snowpack and land/soil characteristics.
L70: Not sure what you mean with transition periods.
L76: Again, you mean spatial distribution and not structure of snow depth and probably persistence and not stability.
L88: Erroneous reference. Here and many others. Check entire manuscript.
L90: Actively managed and unmown grassland – sounds contradictory.
L131: You need to specify the point densities of all of your surveys and data sets (Overall vs field vs forest) together with the size of the data gaps.
L133: remove ‘-‘
L147: How and when did you rasterize your products? Before or after the substruction from snow-off data? I am missing a co-registration of your datasets. See major comments.
L167: ‘made’ – rephrase.
L171: Style again: …DEMs are derived based on the points classified as ground within…
L182: © and TM antenna – please check author guidelines.
L194: Make sure to be consistent field-scale vs field scale.
L196: Please clarify what variables are physical.
L201: Formula not referenced.
L211: At what date? The incidence angle changes throughout the season. And what about sub-canopy shadow hours? More details on the underlying method are needed! Check grammar.
L231: Something is wrong with the references.
L234: Define winter season explicitly.
L236: I would rephrase this sentence.
L249: You have not really talked about standard deviations so far – this sentence is not clear? Is it needed at al?
L251: You did not introduce accumulation and ablation periods so far. What are you referring to? Sentence could become clearer after rephrasing.
L255: Could the increased variability of originate from the reduced point density/increased data gaps? -> Would become clearer with the transect plot (See major comments).
L258: Remove ‘snow observing’.
L262-265: Unclear – sentences should be more concise.
L271: r² - check layout!
L277: remove space in snow.
L287: 0 cm?
L291: remove daily
L297 : ‘were some gaps’ – rephrase and be more precise. How do these gaps emerge? Using your formula 1, I assume that areas with no snow (‘patchy snow cover’) would result in a relative difference of -1. I think this needs to be further discussed as this is what makes your data set special and valuable to the community! There are not many data sets that have several revisits during one season and could be used to analyse also patchy snow covers and their evolution.
L299: over the time period – better: for all survey dates.
L303: AOI does not contain areas south of forests? And…
L305: ..no forest is located in the wester field? This is very confusing. Maybe add small numbers that you could refer to Figure 7.
L311: You don’t know what primarily drivers of snow distribution. This would require a more thorough analysis.
L330: It is well known that LiDAR outperforms SfM (Harder et al., 2020).
L337: What features? Can you give examples? This is where it gets interesting!
L340: Do you mean image overlap? You only have one point cloud for each survey.
L360: You only showed the overall relationships!
L375: Underneath rather than beneath?
L386: would help the snow community.
L388-389: Please give some examples for the ‘numerous studies’. Not sure to what section/results you refer this time stability. This needs to be further analysed, see major comments.
L411: remove space.
Tables and Figures:
Table 1: These are the numbers of magnaprobe measurements per survey? Please clarify!
Figure 1: The patterns of the slope are interesting, as they vary on very different scales comparing forests and open sites. Can you confirm this from your knowledge of the sites or could this be due to some problems in the ground/no-ground classification of the point cloud? Here, again, a transect of the point clouds could help to get an idea of the topographic characteristics of your site. For the aspect: can you confirm these ‘stripes’ from your observations? Can you reproduce them with your SfM- snow-off map?
Move coniferous into the middle of the dark green bar in the legend. Did you explain what these outlined areas are in the middle of the Western field and in between the western and eastern field?

Figure 2:
This plot is very unclear and it is very difficult to understand what you want to show. For instance, you can assign the uncertainty ranges to the individual dots. I would completely revise this Figure. Showing the meteorologic forcing together with the HS timeseries is important, but I would reorganize this figure. For instance, a plot containing Meteo forcings and the three (!) time series from your camera locations. A comparison with your available in-situ measurements is already done in Figure 3.
The caption has some formatting problems.
Figure 3:
I would change the x-axis range to the data (e.g. 0 – 40 cm). I assume this plot combines all surveys? Clarify! It is hard to differentiate the colors of the SfM circles. After incorporating my major comments, it might be more suited to use the color of the dots to visualize different, more relevant informations such as the survey date instead of Magneprobe vs camera.
Figure 4:
Is it needed to show all the outliers of your SfM data in this plot? I would be more interested in the Scatterplots and density plots for the more relevant range e.g. <50 cm.
Figure 5:
It seems to me as if you sometimes have less data gaps in the difference map compared to the SfM – how is that possible? (e.g. 2/20/21).
Caption introduces SD. Either introduce the abbreviation for the entire manuscript or not.
Figure 7:
Are these dark-red areas along the edges of your study site potentially due to misclassifications (ground/no-ground) of your point cloud? Or can you explain them otherwise?
Figure 8: Why do you compare combined vs field and not field vs forest? Or all?
References
Geissler, J., Rathmann, L., & Weiler, M. (2023). Combining Daily Sensor Observations and Spatial LiDAR Data for Mapping Snow Water Equivalent in a Sub‐Alpine Forest. Water Resources Research, 59(9), Article e2023WR034460. https://doi.org/10.1029/2023WR034460
Harder, P., Pomeroy, J., & Helgason, W. D. (2020). Improving sub-canopy snow depth mapping with unmanned aerial vehicles: Lidar versus structure-from-motion techniques. The Cryosphere, 14(6), 1919–1935. https://doi.org/10.5194/tc-14-1919-2020
Jennifer M. Jacobs, Adam G. Hunsaker, Franklin B. Sullivan, Michael Palace, & Eunsang Cho. (2020). Shallow snow depth mapping with unmanned aerial systems lidar observations: A case study in Durham, New Hampshire, United States. https://www.researchgate.net/publication/339338229_Shallow_snow_depth_mapping_with_unmanned_aerial_systems_lidar_observations_A_case_study_in_Durham_New_Hampshire_United_States https://doi.org/10.5194/tc-2020-37
Mazzotti, G., Webster, C., Quéno, L., Cluzet, B., & Jonas, T. (2023). Canopy structure, topography and weather are equally important drivers of small-scale snow cover dynamics in sub-alpine forests. Hydrology and Earth System Sciences Discussions, 2023, 1–32. https://doi.org/10.5194/hess-2022-273
Pflug, J. M., Hughes, M., & Lundquist, J. D. (2021). Downscaling Snow Deposition Using Historic Snow Depth Patterns: Diagnosing Limitations From Snowfall Biases, Winter Snow Losses, and Interannual Snow Pattern Repeatability. Water Resources Research, 57(8), Article e2021WR029999. https://doi.org/10.1029/2021WR029999
Pflug, J. M., & Lundquist, J. D. (2020). Inferring Distributed Snow Depth by Leveraging Snow Pattern Repeatability: Investigation Using 47 Lidar Observations in the Tuolumne Watershed, Sierra Nevada, California. Water Resources Research, 56(9). https://doi.org/10.1029/2020WR027243
Revuelto, J., Alonso-González, E., & López-Moreno, J. I. (2020). Generation of daily high-spatial resolution snow depth maps from in-situ measurement and time-lapse photographs. Cuadernos De Investigación Geográfica, 46(1), 59–79. https://doi.org/10.18172/cig.3801
Sturm, M., & Wagner, A. M. (2010). Using repeated patterns in snow distribution modeling: An Arctic example. Water Resources Research, 46(12), Article 2010WR009434. https://doi.org/10.1029/2010WR009434
Vögeli, C., Lehning, M., Wever, N., & Bavay, M. (2016). Scaling Precipitation Input to Spatially Distributed Hydrological Models by Measured Snow Distribution. Frontiers in Earth Science, 4. https://doi.org/10.3389/feart.2016.00108
Citation: https://doi.org/10.5194/egusphere-2024-1530-RC1
- AC1: 'Reply on RC1', Eunsang Cho, 05 Nov 2024
  
  Dear Reviewer, Thank you for the constructive comments which helped us to significantly improve the manuscript. Please see our detailed response to all comments in the attached file.
  
  Citation: https://doi.org/10.5194/egusphere-2024-1530-AC1
RC2:
'Comment on egusphere-2024-1530', Anonymous Referee #2, 28 Aug 2024

The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2024/egusphere-2024-1530/egusphere-2024-1530-RC2-supplement.pdf

Citation: https://doi.org/10.5194/egusphere-2024-1530-RC2
- AC2: 'Reply on RC2', Eunsang Cho, 05 Nov 2024
  
  Dear Reviewer, Thank you for the constructive comments which helped us to significantly improve the manuscript. Please see our detailed response to all comments in the attached file.
  
  Citation: https://doi.org/10.5194/egusphere-2024-1530-AC2
RC3:
'Comment on egusphere-2024-1530', Ross Palomaki, 29 Aug 2024

The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2024/egusphere-2024-1530/egusphere-2024-1530-RC3-supplement.pdf

Citation: https://doi.org/10.5194/egusphere-2024-1530-RC3
- AC3: 'Reply on RC3', Eunsang Cho, 05 Nov 2024
  
  Dear Dr. Palomaki, Thank you for the constructive comments which helped us to significantly improve the manuscript. Please see our detailed response to all comments in the attached file.
  
  Citation: https://doi.org/10.5194/egusphere-2024-1530-AC3

Eunsang Cho, Megan Verfaillie, Jennifer M. Jacobs, Adam G. Hunsaker, Franklin B. Sullivan, Michael Palace, and Cameron Wagner

Supplement

https://doi.org/10.5194/egusphere-2024-1530-supplement

Eunsang Cho, Megan Verfaillie, Jennifer M. Jacobs, Adam G. Hunsaker, Franklin B. Sullivan, Michael Palace, and Cameron Wagner

Viewed

Total article views: 478 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	Supplement	BibTeX	EndNote
312	138	28	478	33	11	12

HTML: 312
PDF: 138
XML: 28
Total: 478
Supplement: 33
BibTeX: 11
EndNote: 12

Views and downloads (calculated since 21 Jun 2024)

Month	HTML	PDF	XML	Total
Jun 2024	118	28	8	154
Jul 2024	65	21	7	93
Aug 2024	59	32	9	100
Sep 2024	16	16	0	32
Oct 2024	6	6	1	13
Nov 2024	45	30	3	78
Dec 2024	3	5	0	8

Cumulative views and downloads (calculated since 21 Jun 2024)

Month	HTML	PDF	XML	Total
Jun 2024	118	28	8	154
Jul 2024	65	21	7	93
Aug 2024	59	32	9	100
Sep 2024	16	16	0	32
Oct 2024	6	6	1	13
Nov 2024	45	30	3	78
Dec 2024	3	5	0	8

Viewed (geographical distribution)

Total article views: 463 (including HTML, PDF, and XML) Thereof 463 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 13 Dec 2024

Short summary

Uncrewed Aerial Systems (UAS) lidar and structure-from-motion (SfM) photogrammetry are effective methods for mapping high-resolution snow depths. However, there are limited studies comparing their performance across different surface features and tracking spatial patterns of snowpack changes over time. Our study found that UAS lidar outperformed SfM photogrammetry. With limited wind effects, the snow spatial structure captured by UAS lidar remained temporally stable throughout the snow season.


Total:	0
HTML:	0
PDF:	0
XML:	0