Measuring the spatiotemporal variability of snow depth in subarctic environments using unmanned aircraft systems (UAS) – Part 1: Measurements, processing, and accuracy assessment
Abstract. Snow conditions in the northern hemisphere are rapidly changing, and information on snow depth is critical for decision-making and other societal needs. Unmanned aircraft systems (UASs) can offer data resolutions of a few centimeters at a catchment-scale, and thus provide a low-cost solution to bridge the gap between sparse manual probing and low-resolution satellite data. In this study, we present a series of snow depth measurements using different UAS platforms throughout the winter in the Finnish subarctic site Pallas, which has a heterogeneous landscape. We discuss the different platforms, the methods utilized, difficulties working in the harsh northern environment, and the results and their accuracy compared to in situ measurements. Generally, all UASs produced spatially representative estimates of snow depth in open areas after reliable georeferencing by using the Structure from Motion (SfM) photogrammetry technique. However, significant differences were observed in the accuracies produced by the different UASs compared to manual snow depth measurements, with overall RMSEs varying between 13.0 to 25.2 cm, depending on the UAS. Additionally, a reduction in accuracy was observed when moving from an open mire area to forest covered areas. We demonstrate the potential of low-cost UASs to efficiently map snow surface conditions, and we give some recommendations on UAS platform selection and operation in a harsh subarctic environment with variable canopy cover.
Anssi Rauhala et al.
Status: final response (author comments only)
RC1: 'Comment on tc-2022-239', Anonymous Referee #1, 13 Feb 2023
- AC1: 'Reply on RC1', Anssi Rauhala, 18 Apr 2023
RC2: 'Comment on tc-2022-239', Anonymous Referee #2, 01 Mar 2023
- AC2: 'Reply on RC2', Anssi Rauhala, 18 Apr 2023
Anssi Rauhala et al.
Unmanned aircraft system (UAS) snow depth mapping at the Pallas Atmosphere-Ecosystem Supersite https://doi.org/10.23729/43d37797-e8cf-4190-80f1-ff567ec62836
Anssi Rauhala et al.
Viewed (geographical distribution)
The authors present an overview of the methods employed to measure snow depth from UAVs using Structure from Motion (SfM) techniques over a sub-arctic environment. Notably, they present comparisons between several different UAV platforms in terms of both general platform limitations and with respect to measured vs. observed snow depths. Such comparisons are also partitioned by date and land cover across three separate plots.
In my opinion, the paper is well-written and presents results clearly and concisely through the effective use of tables and figures. The methods are well presented to allow repeatability and the various platforms are well described and compared. I think the work's major novelty lies in the inter-comparison of several platforms, and different baselines, as well as a large range of dates, land cover, and illumination conditions. Bringing all of these components together is challenging, but I think it is done well and presented in a way that will allow others to design and implement effective SfM snow depth mapping projects in similar regions.
With that said, I believe there are a few areas that could be improved. This refers to a need for a heightened emphasis on recommendations especially pertaining to the use of multiple baselines. A clear section or figure aggregating the recommendations from throughout the work would greatly benefit the paper. Without the focus on best practices, this work essentially presents a workflow that has already been established to some degree in the literature. More specific suggestions along these lines are presented below.
Provide more emphasis on recommendations. This could be in the form of a table or a new figure within the discussion section. Such recommendations that are mentioned in the text include: the best platform for accuracy/ease of use (RTK vs. GCPs), general recommendations on GCP use, environmental operating suggestions (cold temps and wind), appropriate baselines, and operations in low light conditions. Currently, this information is largely spread throughout the text.
The use of different baselines in this work and pros/cons should be highlighted more. I believe more focus should be placed on the tradeoffs between using the ALS and UAV-based baselines as well as recommending best practices. There is novelty and value to adding an increased focus on these particular findings. For example, based on the results, it seems to me that (if possible) establishing a LiDAR-based baseline might be a good standard practice before doing SfM work.
The lighting conditions aspect was stated as one of the primary objectives of this study, though I feel it wasn’t sufficiently focused on/addressed – It seemed to be shown that low light didn’t notably affect the retrievals, but this is in contrast to previous literature. The discussion on this topic seemed to be a bit of an afterthought. Other points may be considered to be emphasized instead (like the choice of baselines or the performance across multiple platforms).
L15: revise - “…and the UAV snow depth results compared to in situ measurements.”
L24 - L31: Try to separate the sentences on societal impacts & ecological/environmental implications. Reads as though it could be plant/animal communities OR human communities. Also, generally, can add a bit more detail here on the importance of snow before moving on to monitoring approaches. Especially, how snow relates to nature/the environment.
L45 - L52: Either provide a bit more of an overview on the range of what was done in these studies or point the reader to the latter discussion section where you detail these studies more.
L59: ‘submitted to the same journal’ can be removed
L59 - L60: consider rewording, the data itself shouldn’t have implications; are you referring to the insights it provides, regarding accumulation and melt patterns?
L159: The tree masking is an interesting approach. I suggest adding an additional sentence or two describing the ‘Maximum Likelihood Supervised Classification’. Additionally, assuming there is a related citation/paper to this approach, that should be included here.
Figure 2 (and throughout): The use of the term DEMs of Difference (DoD) is used throughout. I suggest adding the acronym in the figure
L174: Related to the previous point, make sure that the difference (or not) in meaning between DoD and snow depth is clear (and why DoD is used instead of snow depth). Are these terms interchangeable?
L215: Provide citation/source for “Levene’s test”
L225: ‘struggles’ -> ‘performs poorly’ (or similar)
L240: are these biases high or low? It may also be valuable to add a table with information of the bias/errors of all the baselines used (maybe in supplementary material)
Figure 1: If you are going to present the point snow depth observation from the ultrasonic sensor (in Figure 7), make sure to indicate its location in Figure 1
L294: ‘No clear correlation’ -> can you describe this a bit more? It is concerning if the observed and UAV-derived depths are not correlated across all depth products. That draws into question the ability of these approaches to accurately capture the snow variability across the basin, and only suggests it can capture general differences through time. Please add some more interpretation here.
Figure 9: In the results, can you add a comment on the clear high (Dec 12) and low biases (Apr 3)?
L401 – 404: This is a good opportunity to delve into why you think the lowlight conditions did not affect the products notably (and why it did in Revuelto et al 2021). The stated objective of providing an assessment of how low light conditions affect SfM snow depth mapping is left with a somewhat ambiguous conclusion. Make sure this is clear.
L470: Reword – I think a general statement mentioning how weather constraints and unpredictability are unavoidable limitations when working with UAVs + providing your example.
Conclusion: One recommendation - specifically recommend a platform (i..e, the Phantom 4 RTK) for providing the best SfM products (w/ relevant error statistics)