Automatic Detection and Segmentation on Gas Plumes from Multibeam Water Column Images

Round 1

Reviewer 1 Report

This manuscript is very difficult to understand due to copious grammatical errors. It needs a thorough copy edit to fix the subject-verb disagreement, extraneous articles, incorrect tenses and misused words. This review did their best to overcome the grammatical issues to provide a review of the content, but it was difficult to understand the manuscript. This review is not well versed in feature detection methodologies. It is unclear how novel the use of a Haar-LBP detector is. Is this a commonly used feature detection algorithm? This leads to the most significant problem that this reviewer has with the manuscript. The comparison between the Haar-LBP feature extraction to Haar and LBP alone does not appear to support the statements in lines 502-508. The results of this method should be compared with other gas plume detection methodologies – the methodologies described in the introduction – to determine if this method is an improvement. Instead, the authors compare the results of a two-part algorithm to the results of using each of the parts separately. In regards to the statements in lines 502-508: How does this method significantly improve gas plume detection? Do other gas plume detection methods use only Haar-like feature detection or LBP feature detection? If other gas detection methods use different detection algorithms the statement that this method improves the efficiency of gas plume detection is not supported. How was the correct number of gas plumes determined? The authors claim this method has a “high correct detection rate” but how the correct values were determined is unclear. What are the correct detection rates of other gas plume detection algorithms? It appears that detection outside of the MSR is due to improved multibeam technology, namely dual swath and frequency modulated beams. Dual swath and frequency modulated beams depress the seafloor sidelobes, enabling target detection outside the MSR. It is unclear how the detection algorithm improves detection outside the MSR compared with other detection methods. If the multibeam itself is depressing the seafloor sidelobes, then other detection algorithms will also be able to detect plumes outside the MSR. The figures need to be larger. It is very difficult to see the gas plumes in the fan-view multibeam images. How are the black and white pixels used in the Haar-like feature selection section determined? Are the black and white pixels differentiated by a threshold filter? If so, it is difficult to see how this method differs from those described in the third paragraph of the introduction. It is also difficult to see how this method will overcome the limitations of threshold filters discussed in the introduction if it uses a threshold filter itself. If a threshold is not used in the Haar-like feature selection, this method needs to be more thoroughly explained. Figure 6 does not appear to correlate with the description in the text of the LBP feature method. Step 2 describes a method where a pixel with value lower than the surrounding pixels is given vale 1 and otherwise 0. However, this does not appear to be how the values of 1 and 0 are given in figure 6. It is difficult to understand how the background differential segmentation differs from the segmentation methods in the references provided in the introduction. QPS (which the authors reference) for example, does differentiation and segmentation filtering and how that differs from this method is unclear. If it is not unique to this method it should be stated that other multibeam target detection methodologies use the same background differential segmentation method. Line 107: A more thorough description of Haar-like feature extraction is needed. Lines 221-223: What are “obvious” grey changes and local texture features? How are they determined in the processing algorithm? Lines 350 – 353: Why is value 0.7591 mentioned? It appears that the Haar-like detector has precision recall higher than 0.7591 but the text says that it does not? This whole paragraph is very confusing. How is detection accuracy determined? Were the MWC images manually analyzed to determine the correct number and extent of gas plumes? Line 377 – 378: I do not agree that the histograms are clearly similar. There are statistical tests to determine the relatedness of distributions (the Kolmogorov-Smirnov test for example) that can quantify whether the distributions are significantly different. Values of “histogram similarity” are given, but it needs to be explained how they are computed.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

This paper describes a ML-based approach to the automated detection of gas plumes in the water column of multibeam bathymetry data. This is an open and active field of research at present, and the paper advances and contributes to the body of knowledge in this field. The concept of using Haar-like features and LBP features is valid.

I am not sure why the number of training samples is large - would this not depend on the number of gas plumes detected, or the N value in equation 1?

I am not entirely convinced by the effectiveness of the algorithm in the sidelobe region. The examples presented contain relatively little background interference, and it is not possible to assess the effectiveness of the algorithm at reducing sidelobe interference. The approach may have limitations in certain environments, but the authors recognise this in the discussion.

English grammar throughout the paper is poor, and the text is repetitive in places. I recommend a thorough editing in revision.

Line 180 lacks a Figure number.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Application of acoustic backscatter from multibeam echosounder water column datasets is currently in its preliminary step of development. Therefore, any research dealing with this kind of data may supplement the existing methodology with interesting solutions. I would like to see this paper published after considering following general and specific comments:

Introduction: A the beginning of the introduction I encourage to describe wide context of your study. You may consider other applications of underwater acoustics and the main utilization of MBES, like detection and characterization of the seabed. You may refer to following recent developments:

• Janowski, L.; Madricardo, F.; Fogarin, S.; Kruss, A.; Molinaroli, E.; Kubowicz-Grajewska, A.; Tegowski, J. Spatial and Temporal Changes of Tidal Inlet Using Object-Based Image Analysis of Multibeam Echosounder Measurements: A Case from the Lagoon of Venice, Italy. Remote Sensing 2020, 12, doi:10.3390/rs12132117.
• Hoffmann, J.J.L.; Schneider von Deimling, J.; Schröder, J.F.; Schmidt, M.; Held, P.; Crutchley, G.J.; Scholten, J.; Gorman, A.R. Complex Eyed Pockmarks and Submarine Groundwater Discharge Revealed by Acoustic Data and Sediment Cores in Eckernförde Bay, SW Baltic Sea. Geochemistry, Geophysics, Geosystems 2020, 21, doi:10.1029/2019gc008825.
• Gaida, T.C.; van Dijk, T.A.G.P.; Snellen, M.; Vermaas, T.; Mesdag, C.; Simons, D.G. Monitoring underwater nourishments using multibeam bathymetric and backscatter time series. Coastal Engineering 2020, 158, doi:10.1016/j.coastaleng.2020.103666.

Methods: Provide details of used processing software.  

Line 36: Since you are changing the topic of introduction, I recommend to start from the new paragraph.

Line 61-63: Reference needed

Lines 57 and 63, 239: References should be numbered in a continuous manner. MDPI recommend e.g. EndNote for automatic management of references.

Line 65: Can you add some explanation of segmentation parameters? Are they related with the thresholds provided in the following sentences? Explain why evaluation of segmentation parameters is so complicated?

Lines 257-258: You need to be more precise, e.g. provide specific MBES that you consider. Not all MBESs allow measuring multi-sector and dual-swath

Lines 293-301; lines 443-448, lines 471-475 - provide more details about measuring platform (name of the research vessel) and measuring system, including e.g. motion reference unit, positioning system. Provide more details of measurement parameters, e.g. pulse length, ping rate, signal type, etc. Provide details about processing software (FMMidwater (?)).

Manufacturer details of all used software and devices will be also needed. 

Lines 319-329 - Since you are describing used methodology, you should move this part to section 2.

Figure 14 - provide frequency units for all plots 

Extensive english editing required (e.g. lines 184-186; 343-344; 375; 461; 512; 

Extensive formatting and style corrections required (e.g. lines 244-245; line 369)

Linies 394-398 - this part is related with methods section

Lines 402-404 - you mentioned dB values. Did you used MBES calibrated in terms of acoustic backscatter? If so, how it was calibrated? Explanation needed.

The range of 2 std in Figure 15a is blurred. 

Lines 480 - on what basis you assumed that the seabed is homogenous in that area? What kind of homogeneity you mentioned (substratum, morphology)?

Lines 502-503; 509-510 - you need to refer to other research for which this method is significant improvement and superior. At the moment there is no reference point.

Lines 504-505 - the description of study sites is very poor. Therefore, it is hard to judge how these marine environments were different from another. Even small map with locations of study site in methods section would be an advantage.

Line 549 - you may consider reference to marcoalgae acoustic detection from MWC in Sptisbergen (Kruss, A.; Tęgowski, J.; Tatarek, A.; Wiktor, J.; Blondel, P. Spatial distribution of macroalgae along the shores of Kongsfjorden (West Spitsbergen) using acoustic imaging. Polish Polar Research 2017, 38, 205-229, doi:10.1515/popore-2017-0009.

You may consider reference to some recent works about multibeam water column data research:

• Lohrberg, A.; Schmale, O.; Ostrovsky, I.; Niemann, H.; Held, P.; Schneider von Deimling, J. Discovery and quantification of a widespread methane ebullition event in a coastal inlet (Baltic Sea) using a novel sonar strategy. Sci Rep 2020, 10, 4393, doi:10.1038/s41598-020-60283-0.
• Idczak, J.; Brodecka-Goluch, A.; Lukawska-Matuszewska, K.; Graca, B.; Gorska, N.; Klusek, Z.; Pezacki, P.D.; Bolalek, J. A geophysical, geochemical and microbiological study of a newly discovered pockmark with active gas seepage and submarine groundwater discharge (MET1-BH, central Gulf of Gdansk, southern Baltic Sea). Sci Total Environ 2020, 742, 140306, doi:10.1016/j.scitotenv.2020.140306.

References - according to Remote Sensing Instructions for Authors: "DOI numbers are not mandatory, but hihgly encouraged".

Author Response

Please see the attachment.

Author Response File: Author Response.pdf