the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Processing methodology for the ITS_LIVE Sentinel-1 ice velocity product
- 1Division of Geological and Planetary Science, California Institute of Technology, Pasadena, 91125, USA
- 2Jet Propulsion Laboratory, California Institute of Technology, 5 Pasadena, 91109, USA
- 1Division of Geological and Planetary Science, California Institute of Technology, Pasadena, 91125, USA
- 2Jet Propulsion Laboratory, California Institute of Technology, 5 Pasadena, 91109, USA
Abstract. The NASA MEaSUREs Inter-mission Time Series of Land Ice Velocity and Elevation (ITS_LIVE) project seeks to accelerate understanding of critical glaciers and ice sheet processes by providing researchers with global, low-latency, comprehensive and state-of-the-art records of surface velocities and elevations as observed from space. Here we describe the image-pair ice velocity product and processing methodology for ESA Sentinel-1 radar data. We demonstrate improvements to the core processing algorithm for dense offset tracking, “autoRIFT”, that provides finer resolution and higher accuracy data products with improved computational efficiency when compared to earlier versions. A novel calibration is applied to the data to correct for Sentinel-1A/B subswath- and full swath-dependent geolocation errors caused by systematic issues with the instruments. Sentinel-1’s C-band images are affected by variations in the total electron content of the ionosphere that results in large velocity errors in the azimuth (along-track) direction. To reduce these effects slant-range (line-of-sight or LOS) velocities are used and accompanied by LOS parameters that support map coordinate (x/y) velocity inversion from ascending and descending slant-range offset measurements, as derived from 2 image-pairs. The described product and methods comprise the MEaSUREs ITS_LIVE Sentinel-1 Image-Pair Glacier and Ice Sheet Surface Velocities: Version 2 (https://its-live.jpl.nasa.gov).
Yang Lei et al.
Status: final response (author comments only)
-
RC1: 'Comment on essd-2021-393', Anonymous Referee #1, 16 Mar 2022
This paper presents a detailed methodology of processing Sentinel-1 radar data (TOPS mode) using an updated module “autoRIFT '' of ISCE platform in order to generate ITS_LIVE Sentinel-1 ice velocity products. The paper walks through the different elements of the sequential processing chain and highlights key points that improve the resolution and accuracy of the velocity products. The paper is clear and generally well-written. I have a few comments and suggestions which may be incorporated for more clarity.
MAJOR COMMENTS:
- The products will be openly available. The module of ISCE platform will also be publicly available. These free resources will be used by many folks across the globe for their scientific analysis or processing data over areas other than polar regions. This paper limits the presentation and analysis of ITS_LIVE velocity products to Greenland, but the products will be available for mountain glaciers as well. The title should reflect this aspect; maybe by adding “polar regions'' in the title. Alternatively, more insights based on presentation and analysis over mountainous regions (e.g. European Alps, High-Asia) should be added in the paper. It is obvious that ITS_LIVE products and associated uncertainties are different in regions other than polar ice sheets.
- We have a number of ice velocity products based on Sentinel-1 radar data and it is increasingly challenging which product is the best way to carry out a scientific analysis without processing the raw GRD/SLC data. Boncori et al., 2018 compared ice velocity products from several international research groups, highlighted different strategies on the processing and uncertainty estimation and found significant differences, also recommending a universal approach. This paper provides a new or updated algorithm (which is great) but needs to be compared with similar contemporaneous products (e.g. PROMICE). Otherwise, the scientific users will have to do this exercise or cherry-pick one of the available products. Both will not serve the ongoing efforts of establishing standard method development, ice velocity product generation and documentation. It would be nice to compare Sentinel-1 ITS_LIVE products with previous ITS_LIVE products obtained from optical remote sensing data (e.g. Landsat).
- If this paper serves only a method development, there should be some more test cases (e.g. ice shelves in Antarctica, debris-covered glaciers in Alaska/high-Asia) to present the applicability of the algorithm other than Greenland Ice Sheet.
- So finally these products will not be average over a certain time-period like PROMICE 21-day ice velocity mosaics? Please clarify and highlight, if this is the case, in your paper as this is a unique aspect.
MINOR COMMENTS:
L40: Several satellite derived regional ice velocity products are released annually
L45: As described in Lei et al., 2021a (CHECK ELSEWHERE)
L55: 6 days repeat is not everywhere but limited to polar regions and Europe or some key areas of the world.
L70: Revise “ We do …… Greenland”
L90: Have you ever considered 2m Arctic DEM instead of GIMP DEM? That may be a better choice for transformation between radar and geographic coordinates.
L95: There is no reference velocity for mountain glaciers. What will be the approach for those areas?
L100: What do you mean by “successful match”? Based on cross-correlation or similarity function value?
L130: ..the extent of..
L140-145: Repetitive
L175 or elsewhere: “autoRIFT” should be clearly distinguished – italic?
L180-200: I strongly recommend a graphical representation of an algorithm – clearly distinguishing chip size, overlapping region, search size etc. on images.
L220-230: It was not very much clear why it was done that way. A lot of parameters with equations break the flow of writing here. Simpler writing with rationale might help us better understand.
L290: Is this subjective? Any insights?
L305: 7>>seven
L450: Maybe I miss something, but the slant range displacement component contributes to ground range (East-West (your x) and North-South (your y)) and vertical movement (z) and the azimuth displacement component contributes to x and y only. Your equations don’t consider vertical velocities. It is known that the vertical velocities exist as well due to slope/elevation changes in the flow direction or ablation. Please clarify.
-
RC2: 'Comment on essd-2021-393', Anonymous Referee #2, 21 Mar 2022
This study provides a processing chain for Sentinel-1 TOPS mode data. Authors have used a modified module of autoRIFT to generate ice velocity global products. This paper is successful in demonstrating the processing chain and efforts in overcoming associated errors. This study also demonstrated improvement in terms of accuracies and resolution of velocity products. The paper is methodologically well organized. However, I have a few comments. Authors may include these comments from the reader’s perspective. My sequential comments are;
Abstract: Authors should re-write the abstract clearly indicating quantitative improvements when they say *higher accuracy*, *finer resolution*, *improvements*. In the present abstract, the reader cannot find what level of improvements, how much accuracy, and what resolution authors refer to. In general, this abstract reads very generically as a project report. To attract a wider audience, this abstract should be heavily revised to clearly state main achievements quantitative, limitations of the current product, and comparative analysis with existing data.
Introduction: This section only focuses on operational velocity product generation attempts. Authors should enrich this section by providing literature on glacier velocity generation in general, the use of glacier velocity in glaciological studies, various ways of deriving velocity fields and uncertainties, and the strengths of different methods. Currently, this introduction section does not discuss existing regional attempts using a variety of methods. After the literature review, the authors should provide gaps in current knowledge and what additional knowledge this study provides to the scientific community.
2.1 Product and methodology overview
2.1.1 Input dataset: This section provides only example input datasets for the Greenland case study. I will suggest including other reference input datasets used for other regions. This can be added in the supplementary information as a table or description. Reference velocity for the Greenland ice sheet is mentioned in the study but I am wondering which stable reference velocities are being used for other parts of the globe. Similarly, input DEMS for other regions should be included in the supplementary information. The impact of varying resolution of DEMs in different regions should be discussed. Have you tried Arctic DEM?
2.1.2 ITS_LIVE Sentinel-1 Image-Pair Data Product
Authors should elaborate *offset tracking success* for readers. What criteria do they use for deciding offset tracking and do they calculate it quantitatively?
“Hence, the selection of the appropriate sensor combination is dependent on the actual use case including data availability, quality, study area, etc.” Such statements are made a couple of times without actually providing clear guidance on these criteria. The authors should clearly state how did they estimate such criteria for different regions on the globe. Authors should provide practical challenges in deciding these criteria while selecting input datasets.
The output glacier velocity maps are generated in 120 m spatial resolution. Is this specifically because computational resources or authors have used a specific criterion for choosing this as the optimal resolution.
How have authors calculated the magnitude of errors? The authors have discussed thoroughly geolocation and ionospheric errors introduced in the analysis and how they tried to overcome these errors. However, the final product error magnitude calculations seem to be missing. If I am not wrong, can you re-direct to the existing literature for elaborating this process?
“This sparse/dense combinative searching strategy substantially improve computational efficiency”. I would like to see numbers on efficiency and improvement.
General comments: Authors have demonstrated the processing on Greenland but state that the products are global. I am wondering if they can describe practical implementation challenges in processing global datasets and how uncertainties associated with other regions are dealt with. This is important because the research community will use these datasets for their research in the coming years and they would like to see practical challenges in the Arctic, Antarctic, mountain areas in the Himalayas, Alps. Also, the major limitation of this paper is that the authors have only focused on Greenland. I will suggest including a good distribution of test sites covering different parts of the globe.
Validation: As we know that there are multiple velocity products are being generated both using SAR and optical methods. This study does not provide any guidance on comparing the present product with existing velocity products. This will be very useful for researchers as they would like to know reliable velocity products for their studies. Similarly, new algorithms are being developed for improvement in velocity products. This study does not comment on state-of-the-art methods and the comparison with other algorithms. Eventually, the scientific community will be benefited from such inter-comparison experiments.
Ground validation: Authors have not mentioned about validation of velocity products with in situ measurements. Have they attempted such validation?
---------------------------------------------
-
EC1: 'Comment on essd-2021-393', Kenneth Mankoff, 26 Mar 2022
Dear Authors,
In addition to the two reviewer comments, I have the following observations, suggestions, comments, and questions.
The current version only validates against other Sentinel SAR imagery. I think it would strengthen the data description to take some outputs and compare against one of the many already existing velocity products. See Sect. 3.5 of https://essd.copernicus.org/articles/10/2275/2018/ I suggest validating against one or more of the many other MEaSUREs (or PROMICE) velocity products in Greenland, and other velocity products in Antarctica and in alpine regions. Why should I use this instead of the many other MEasUREs or PROMICE products? How does it compare to them? What are the causes of the disagreements, assuming they exist? Are the errors random or systematic bias?
All errors appear internal to your processing scheme. Are there other errors that impact the final data product? See Sect. 3.6 of above URL.
Given ITS_LIVE claim of global, only 3 validation regions in Greenland is limiting. Antarctica and especially small Mountain Glaciers are fundamentally different creatures - are the results and errors the same in those locations? I think an in-depth discussion of your data product and its quality issues is needed for mountain glaciers and Antarctica. Reviewer 1 suggests that the paper title could be changed to reflect "polar regions", but I disagree. The dataset is global, so this data description paper must cover the data, and cannot focus on just a subset of the data.
A more detailed description of the data product is needed. For example, how many images are there? How many have 6 day resolution, and how many have 60 day resolution? Are there differences between A and B? Which should I use for my (insert common use case scenario here)? Regarding the Greene et al (2020) paper, how does that impact the 6 vs 60 day w.r.t. under/over sampling highs/lows? By "large glacierized regions" are you really "global" or do you exclude small mountain glaciers? What criteria defines your cutoff?
Ken Mankoff
Editor
- AC1: 'Comment on essd-2021-393', Yang Lei, 08 Jun 2022
Yang Lei et al.
Data sets
MEaSUREs ITS_LIVE Sentinel-1 Image-Pair Glacier and Ice Sheet Surface Velocities: Version 2 (Greenland Sample Products) Yang Lei, Alex S. Gardner and Piyush Agram https://doi.org/10.5281/zenodo.5606118
Model code and software
autoRIFT (autonomous Repeat Image Feature Tracking) Alex S. Gardner, Yang Lei and Piyush Agram https://doi.org/10.5281/zenodo.5643820
Yang Lei et al.
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 590 | 296 | 22 | 908 | 17 | 20 |
- HTML: 590
- PDF: 296
- XML: 22
- Total: 908
- BibTeX: 17
- EndNote: 20
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1