Co-registration and residual correction of digital elevation models: A comparative study
Abstract. Digital elevation models (DEMs) are currently one of the most widely used data sources in glacier thickness change research, due to the high spatial resolution and continuous coverage. However, raw DEM data are often misaligned with each other, due to georeferencing errors, and a co-registration procedure is required before DEM differencing. In this paper, we present a comparative analysis of the two classical co-registration methods proposed by Nuth and Kääb (2011) and Rosenholm and Torlegard (1988) . The former is currently the most commonly used method in glacial studies, while the latter is a seminal work in the photogrammetric field that has not been extensively investigated by the cryosphere community. Furthermore, we also present a new residual correction method using a generalized additive model (GAM) to eliminate the remaining systematic errors in DEM co-registration results. The performance of the two DEM co-registration methods and three residual correction algorithms (the GAM-based method together with two parametric-model-based methods) was evaluated using 23 Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) DEM pairs from the western margin of the Greenland Ice Sheet. The experimental results confirm our theoretical analysis of the two co-registration methods. The method of Rosenholm and Torlegard has a greater ability to remove DEM misalignments (4.6 % on average and 15.3 % maximum) because it models the translation, scale, and rotation-induced biases, while the method of Nuth and Kääb considers translation only. The proposed GAM-based method performs statistically better than the two residual correction methods based on parametric regression models (high-order polynomials and the sum of the sinusoidal functions). A visual inspection reveals that the GAM-based method, as a non-parametric regression technique, can capture complex systematic errors in the DEM co-registration residuals.
Tao Li et al.
Status: final response (author comments only)
RC1: 'Comment on tc-2022-205', Anonymous Referee #1, 19 Dec 2022
- AC1: 'Reply on RC1', Xiang Shen, 04 Mar 2023
RC2: 'Comment on tc-2022-205', Anton Schenk, 11 Jan 2023
- AC2: 'Reply on RC2', Xiang Shen, 04 Mar 2023
Tao Li et al.
Tao Li et al.
Viewed (geographical distribution)
Li et al. compared different methods of correcting 3D-shifts and others biases in digital elevation models (DEM) with the ultimate goal to reduce uncertainties in glacier elevation change estimates. They first compared the widely-used Nuth & Kaab (2011) to the less popular Rosenholm and Torlegard (1988) algorithms for DEM coregistration. On top of a simple 3D shift, the latter algorithm also account for any rotation or scale differences between the DEMs. Further, they proposed an improved correction of the structured-biases between the DEMs that have a proper signature in the two directions of image acquisitions (along and across tracks). They go beyond fit by polynomials and sums of sinusoids by proposing a spline-based non parametric model.
I found nothing wrong with this study. However, I am not super convinced that this article, in its current form, fits well in The Cryosphere and its readership. I see two main reasons for that:
1/ DEM differencing is a popular technique to measure glacier changes. However the scope of the present study is really technical with no direct application to glacier changes. The study sites only marginally include glaciers.
2/ The added value of the proposed method is modest. I am not convinced that a gain of 0.2 m (5%) in standard deviation of the residuals between DEMs only covering a single (and not very challenging) test site is sufficient to convince the glaciological community to rethink the way they coregister DEMs. The added value of the spline-based correction of along track residuals is higher but would need to be confirmed in different settings.
Overall the paper is well written, the work is performed seriously but I missed some novel results that would make a real impact on the glaciological community.
1/ Do the conclusions apply in other settings? Map of elevation differences are constructed from several ASTER DEMs in western Greenland with a strong proportion of stable terrain. Ice-covered terrain is restricted to the eastern part of the images/DEM. It seems that images are almost cloud free. This site and the cloud-free images are appropriate to design and test the different methods but are not representative of real case scenario. In my experience, further challenges for DEM coregistration comes from: vegetation (changing with time), large fraction of glacier areas vs. stable terrain, gaps or unreliable data in the DEMs due to clouds, the rough topography leading to higher noise level in the DEMs. Authors did not explore these difficulties and thus their results are not representative of more complex and more realistic situations.
2/ Do the improvements over stable terrain percolate to ice-covered areas? To convince the readers (glaciologists, the readership of TC) of the added values of the proposed methods, authors would need to demonstrate real improvements over glacier terrain. Such a validation is tricky, I reckon, because glacier elevations are constantly changing. I see two ways for the authors to demonstrate this
(a) apply their methods to DEMs derived from images acquired just a few days apart so that the assumption of no elevation change is almost valid. They would then be in position to coregister and bias correct their DEMs over the stable terrain and then check the improvements on glaciers (where no change should be measured over a few days).
(b) find sites where ASTER DEMs are acquired simultaneously to higher resolution DEMs (for example from the Arctic DEM project) so that a reference elevation change map is available. This second solution is more tricky to identify.
3/ The discussion is rather weak. There is a long part about the “extrapolation error” that is mostly unrelated to the rest of the article.
4/ I was also a bit disappointed to see that the techniques are only apply to ASTER DEMs. This also reduce the scope/relevance of the results.
L30. I find it unbalanced that three out of four references on the use of DEMs for glacier elevation change mapping are from Chinese colleagues. Others more seminal papers on the topic could be cited here.
L31. Leprince et al., is about mapping surface displacements (using Cosi-Corr), I am not sure this reference is appropriate for DEM errors. Can authors double check?
L50. What are these "scenarios"?
L71. Authors need to explain why they need to revisit the Nuth & Kaab's method and why they present in details these flavours of their method. It is not straightforward for the reader what is the aim here. Also because in the end the results are almost identical...
Figure 1. I did not really understand the figure because I did not understand what were representing the different letters/segments. Annotation to be clarified.
Figure 2. the terminology "master" and "slave" are not very the best ones for ethical reasons. “Reference” and “secondary” DEMs are better.
L204. I do not understand why 23 DEM pairs are first mention and then only 2 DEM pairs are presented in detail in Table 2. Rather include an appendix with the dates and ID of all the DEMs so that the study can be reproduced. Also, as you read in my general comment, a more extensive study using a variety of study sites would be more convincing.
L208. The NDBI index is not often used in the glaciological community so need to be explained a bit more.
L215. I do not understand how this 3-sigma rule is applied to check outliers from the classification. Authors need to elaborate more.
Figure 4 does not really bring much. I think it will be pretty obvious to most readers and can be explained in a few sentences in the text.
L239. Example of why the application of the algorithm to a greater diversity of images is needed.
L302. English is not really correct I think. Check
L320. This statement (and the rest of the paragraph) about extrapolation error and elevation error as a function of altitude comes a bit out of nowhere. Why discussing extrapolation when this was not mentioned before.
L328. This is exactly what the revised study should do: include cases where bare terrain is rare and see how the different methods compare.
L349. I found the improvements rather modest. “outperformed” is a bit overselling.