Coastal Sea Ice Concentration Derived from Marine Radar Images: A Case Study from Utqiaġvik, Alaska

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Comments to the Author:

 

The article introduces a method for extracting sea ice concentration from marine radar images by applying the Canny edge detection algorithm and compares it with existing satellite datasets (MODIS-AMSR2 and NSIDC's CDR). The study has a certain degree of innovation. However, before the paper can be accepted, there is still much to be improved, and the text is a bit rough. Here are some of my comments.

 

(1) This paper uses the Canny edge detection algorithm to extract sea ice, but this algorithm can produce false detections in conditions of haze or precipitation, and the paper does not provide an in-depth analysis and solution to this issue. It is recommended that the author further explore how to optimize the algorithm to reduce the false detection rate or propose improved methods for such complex meteorological conditions.

 

(2) The article describes the comparative analysis of radar images with MODIS-AMSR2 and CDR products, but does not provide sufficient explanation for the differences between the results. For example, why radar images perform better under certain conditions or worse under others. It is recommended to further explore these issues in the discussion.

 

(3) The algorithm used 14 images and 10 ice analysts. The participation of multiple analysts does help to reduce human identification errors, but 14 images might be a bit few, and it would be better to increase the number. Moreover, the selection of images should be more random and diverse to reduce errors and improve accuracy. Also, in Section 2. Data, I did not find any explanation or introduction for these 14 images, please add it.

 

(4) Line 101 on page 3 is an incomplete sentence, please carefully review the entire text to avoid such errors.

 

(5) 2.1. (Near-Real-Time) Climate Data Record Sea Ice Concentration If it is public data, the download link and access time need to be provided.

 

（6）Section 2.2 is confusing; did the author use the method from reference [18] for the merging? If so, the merging method should be introduced in the Methods section. The sea ice extraction methods of MODIS and AMSR2 should also be properly introduced.

(7) On line 140 of page four, how were the upper and lower threshold values selected? Please provide a detailed explanation.

(8) On line 132 of page three, how was the masking process carried out?

(9) On line 142 of page four, how was the masking specifically used to remove the radar image boundary?

(10) On line 164 of page four, why were areas with low radar reflectivity under the shadow of ridges or covered by smooth ice within the edges marked as sea ice? Also, how could there be ridges in the ocean?

(11) Section 3.3 does not clearly state how the parameter optimization was conducted.

(12) Figure 2 should be placed in Section 3.5.

(13) Section 3.4 is also strange; this is not a method and could be merged with Section 3.3.

(14) From Figure 2, it can be seen that the radar area is smaller than one grid of the CDR. Is it reasonable to use CDR data for comparative study?

(15) Lines 186-188 on page five are very confusing. Additionally, the method of calculating the correlation between the CDR, MODIS-AMSR2, and CSIRS datasets could be provided in the Methods section.

(16) On page six, line 19, (Figure 3 (b)). stands alone as a sentence, which is strange.

(17) The method of calculating the inter-analyst error is not given on line 198 of page six.

(18) What is the core innovation of the paper?

(19) What does 'results not shown' mean on page seven, line 226?

(20) The core conclusions of the paper are not quantitatively given in Section 5. The discussion is also not deep or sufficient.

Comments on the Quality of English Language

none.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Marine surveillance radars on board ships or onshore can be used to monitor the ice cover and provide data to be analysed as such or supporting and validating other results. Data can be  arranged by a radar recorder capturing  the radar image time series. The benefits are good  time  and spatial resolution, typically  few seconds and few meters, and the connection to SAR images that ’look the same’ thanks to similar frequency bands. On the other hand, the visibility range of  ice features, 10-20 km, limits the usefulness of the data especially for coastal radars. The radar data is especially suitable for kinematic analyses while the extraction of structural features encounter the same problems a SAR in the classification of roughness features, detection of floe edges and telling smooth level ice apart from open water. A specific radar problem, increasing with distance,  is shadowing behind elevated features (ridges) due to  low grazing angle.

 

The paper presents radar based SIC retrieval in terms of a edge detection algorithm to ice/water demarcation in radar images with visibility range ~10 km. The algorithm parameters are optimised using ice analysts’ estimates and the the results are also compared with visible MODIS-Terra images. The daily averages of radar SIC are then compared  with SIC from CDR climatological dataset  (partially overlaps radar visibility area in one grid cell) and from MODIS-AMSR2 products (partially covers radar visibility area with ~100 cells) over a 18-month period.

 

The algorithm (Canny edge detector, please correct all small c’s) is a standard tool and applied as such so methodologically there are no novelties. The methods are properly presented. The procedure is termed often floe edge detection although the goal is to detect edges of open water areas and not to resolve floe geometry. In the closed pack the emerging structures can be both floe edges and ridges.

 

A major concern is that SAR sea ice research is dismissed, especially as the Canny detector is frequenctly applied to similar problems in SAR context (and also for other satellite data types). Methodologically the manuscript adds to this body of research and it should be expounded how in connets to the state of art and advances or expands it.  It is stated that SAR images are difficult to interpret although SAR texture is usually very similar to that found in radar images. One of the main promises radar is that results obtained in from high spatiotemporal resolution data can be extrapolated to enveloping SAR images. Selected SAR-radar comparisons would considerably increase the impact of the approach. At least the extensive  edge detection and SIC work for SAR should be summarised with selected references and with emphasis on the Canny applications. After all, the SAR is the future prospect of further research on the topic at the  radar site. If there are plans to seize this in the future this is not a reason not to have  proper literature review here.

 

The final goal of the manuscript is therefore a bit unclear. Different data are compared, but abstract states that the algorithm is validated  by satellite products while in the main presentation it appears to be vice versa. As the match between data types is only partial this raises the question how meaningful the comparisons are.  It would be better not the present conclusions on systematic biases of different platforms. The radar algorithm appears to give proper results, but how the observed deviations of  satellite data  from radar can hardly be used (eg. to error estimation as suggested) as they depend on the varying partial overlap and land contamination and these cannot be quantified from this dataset only.  

 

Especially concerns the fast ice zone masked out from the AMSR. In Figure 5 the  2022 April-May SIC oscillates between 0 and 1 for AMSR and 0 and 0.4 for radar? Is this just widening of the coastal lead, fast ice remaining at place? If so, why this choice in the figure? Would it be possible to separate the fast ice and moving pack in the graphs, perhaps using temporal filtering?  It would be also informative to see the 14 images, perhaps indicating where the anlysts’ and algorithm disagree. Figure 4  ab) could be better, larger and with the same red edge in both subimages.

 

The authors should consider concentrating to the on the radar based methods and specifically on the edge detection problem with added exposition on different conditions and seasons and proper referencing. The reader also wonders is there really no ground truth data, for example aerial imagery or LIDAR, obtained during the whole history of the site as any such could be used to validate the SIC algorithm  with archival  data.

 

Some detailed comments.

 

Section 1.

·       Reduce the number encyclopedic Arctic references to few choice summaries or leave out altogether.  Use more recent summary references for satellite data.

·       The Utqiagvik radar timeline and research work could be better referenced and in a topically organised way.  The same applies to ice radar research pertaining to other sites and on board research vessels, with the same division of research topics. The scope of is not so large that it could not be summarised quite comprehensively.

·       Characterise the Utqiagvik ice conditions generally, esp. the fast ice zone and breakup.

·       ’Latest version of CSIRS’:  does this mean physical radar, processing unit, or software?

·       SAR data was easily accssible after ERS-1 was launched 1991.

·       Line 80: recently

 

Section 2.3

·       Are the 4-min images  snapshots or averages/medians over the period?

·       Please give specifications of the Utqiagvik digitisation, the process of image rasterisation and the maximum resolutions for both scan data and for images. In the stored images, why is the pixel size not constant.

 

Sections 2.1-2

·       How does the reported errors in satellite data relate to the 18 month time history?  Are there any quantifications of land or fast ice edge induced errors that could be used to interpret it?

 

Section 3.3.

·       How did you instruct the analysts as concerns the smaller dark patches inside floe matrix, or shadowing. Was some scale indicated? It would be instructive to see the 14 cases with dates and some comparisons between algorithm and human.  There could  more discussion and details, after this was the only real validation of the method that was then applied to ~200 000 images.

 

Section 3.5

·       Explain ’point-by-point comparison’  and ’coarse grained’

 

Section 4.1

·       Explain fig. 3a more clearly.

·       In fig.4 is the algorithms really missing the large open water area in the NW quadrant? And the poor correlation is due to the algorithm? If the satellite/radar comparison is pursued as research topic and not just an illustration, more discussion on the discrepancies is needed. Now it is unclear is the source in the radar data or in the satellite data. For MODIS-AMSR2 this could be analysed in terms of the 1-km data and presented as skill score 4-field.

 

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

Please check the attachment.

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

Comments:

- In the methods section, the authors are encouraged to include a subsection describing the equations used to assess the introduced product.

- Adding a figure (should be figure 1)that shows the study location and the radar's position & range, along with a description of the location and the relevance of retrieving sea ice, would be helpful for readers. This would also provide insights into the applicability of the introduced method in other locations.
- Given the differences in spatial resolution between the benchmark product and the introduced products, what upscaling methods were used to compare ice concentration?

- The discussion should be a separate section. In its current version, the paper merges the discussion and conclusion, resulting in a lack of a clear discussion. The authors are encouraged to compare their results with previous studies that address the retrieval of sea ice concentration from radar data.

- With the risk that MODIS sensors might stop sending data (already one satellite with MODIS sensor is down), are there any alternative optical products that could assist in retrieving sea ice data? Could GOES-R or VIIRS be alternatives?

- The authors are also encouraged to discuss the sources of uncertainty in their retrievals, as this topic needs to be addressed in the context of sea ice retrievals from radar data.

- Could the authors comment on the reliability of the method if applied in estuaries or large rivers? any previous applications?

Author Response

Please see the attachement

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The paper has made some corrections in response to the reviewers' comments. However, there are still some issues. It can be published after appropriate modifications.

 

1. The article describes the comparative analysis of radar images with MODIS-AMSR2 and CDR products, but does not provide sufficient explanation for the differences in the results. For example, why radar images perform better under certain conditions or worse under others has not been reasonably explained.

 

2. The 14 images used by the algorithm, as an important data source, should be introduced in the 2 Materials and Methods section.

 

3. What is the land mask product provided by https://www.fisheries.noaa.gov/inport/item/60548? Is it matched with the radar resolution? How is it operated specifically?

 

4. Due to the differences between the old and new versions, it is difficult for reviewers to locate the specific parts being addressed in the paper. Therefore, it is suggested to put the specific modifications directly in the response to the reviewers' comments. Or provide clear and specific page numbers and line numbers for the modified content.

 

Comments on the Quality of English Language

no

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors I have no further comments and suggest acceptance.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

Authors have addressed all comments. The quality of the paper was greatly improved.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf