Ice Monitoring in Swiss Lakes from Optical Satellites and Webcams Using Machine Learning
, Rajanie Prabha 2, Tianyu Wu 1, Emmanuel Baltsavias 1, Laura Leal-Taixé 2 and Konrad Schindler 1Round 1
Reviewer 1 Report
This paper presents a methodology to detect freeze up and spring melt of lakes in alpine regions using MODIS, VIIRS and webcam data, and using machine learning maximize the accuracy of the detection of these transitional events on the lakes. MODIS data were obtained for four Swiss alpine lakes and three lakes for VIIRS over 2 full winter periods, and webcam data were examined for two of the lakes with a total of 3 cameras over two winters. Manual examination and machine learning were used detect ice/no ice conditions and the procedures were explained in great detail. My background is limited in this area, but my assessment is that the analysis was complete and the results were very encouraging.
Overall the paper is well written and provided the necessary references to frame this work. Figures and tables were sufficient to convey an understanding of the procedures and results. While there was overlap between the satellite analysis and webcam analysis, these two themes of the paper were not as integrated as I had initially expected. MODIS and VIIRS have their advantages and disadvantages for high temporal resolution ice detection, but a major plus for MODIS is the nearly two decade time series of MODIS data. This was mentioned as future work, but not taken on by the paper. The webcam analysis was very thorough and the authors clearly learned a great deal from analyzing the images. No doubt, the authors could provide a wish list of webcam siting guidelines for optimizing ice detection on alpine lakes. I would suggest adding that. As new webcams come online and existing webcams are replaced, these insights from the authors (if implemented) could increase the accuracy of ice detection from webcams in the future.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The authors present the very great advances in their study of lake ice cover (extent) on-set and off-set according to self-developed algorithms of satellite and on-site webcam imaginary processing. These all is very fine. The only thing I wonder why authors do not search for correspondence with air temperature monitoring and may be simple sum of negative grad – day ice on-set and off-set and formation model. But the work should definitely be published! Thanks!Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
The manuscript entitled “Lake Ice Monitoring from Earth and Space with Machine Learning” evaluates the use of machine learning algorithms to classify lake ice features in both MODIS/VIIRS satellite imagery and web camera imagery. The establishment of a lake ice database for images from web cameras is very useful and could serve as a valuable training dataset for future experiments. The results of the manuscript are promising with the machine learning models appearing stable even when lakes are removed, and different winters are tested.
The methodology appears good and the authors have thoroughly discussed limitations of the different datasets (both satellite imagery and web cameras). One concern with the study that is unavoidable is the lakes that were used. While it is nice that these lakes have accompanying web cam imagery, each lake is only represented by a few pixels in each MODIS image.
The authors have presented the machine learning method as a possibility for development of a future operational system. For this to be the case, it would be valuable to include comparison not necessarily to ground truth data but to other operational products. I have expanded on this in the comments below. Additionally, the literature review should be changed to better represent previous studies. The authors should be wary of informal language within the article (for example instead of cam > camera) and must ensure their in-text citations match the required format.
Title: The title of the paper needs to be modified to be more specific. While the manuscript does indeed focus on monitoring lake ice using machine learning, some reference to the study site may be helpful for readers. For example, “Examining the Application of Machine Learning for Monitoring Lake Ice for Lakes in Switzerland: Towards an Operational Lake Ice Monitoring System”, however, this title is only a quick suggestion and to the authors discretion.
Line 11: “On an average”, can be changed to “On average”
Line 14: The abstract could be reworked to include some reference to the larger applications of the proposed methodology as an operational technique.
Line 18: “a” should be changed to either “the main challenge” or “one of the main challenges”.
Line 19: “Lake observables”, is this referring to lake properties? Maybe either list as with the CCI+ variables or better clarify what this statement means.
Line 28: This reference should be replaced with something more recent – research has been published in the last decade that addresses the socio-economic and biological role of lake ice (i.e. see Knoll et al., 2019)
Line 31: “past years” requires a citation, either a reference to previous research or to the NSIDC lake ice database as done below. It would also be good to provide a frame of reference, since which year?
Line 37: “- daily”, should be changed to “is high temporal resolution (daily) with an accuracy…
Line 50: The section “Target lakes and winters” should be moved to the first part of the methods section. Additionally, some information about the meteorological conditions (temperature, precipitation, etc.) of the areas would be beneficial to the reader.
Figure 1: This figure requires some changes. A scale bar should be provided in the two inset maps and one for the larger overview. It would also be nice to have labels on the different lakes as well as an indication of the date the MODIS image and web camera shots were acquired.
Line 71: “Some researchers…” this requires a citation – which researchers? It is unclear.
Line 76: “Our Contributions”, this section is interesting however, it is more suited for the conclusion. While it highlights the objectives, this was clearly done between lines 43 and 49.
Line 87-102: This section is an interesting overview of previous research on lake ice observation but could be written more concisely. Presently the paragraph simply presents a list of papers and the most critical part is the last two sentences. The section could be improved by briefly summarizing some of the main points from these key articles, focusing on fewer and more significant papers, or by presenting some numbers on how lake ice cover is expected to change.
Line 99: RADARSAT-2 is a SAR satellite; it is better to simply state SAR.
Line 111-112: Make sure references have the proper notation.
Line 116: How accurate were these methods? This is needed to provide context for the reader.
Line 117-124: While still only the first version is available and limited to 250 lakes, authors should mention the newly released CCI Lakes product which uses MODIS data to provide daily ice cover values.
Line 126: The in-text citation is not correct, and the language should be changed, “Previously” could be used as a substitute.
Line 141: “an” should be “and”.
Optical Data Imagery: What specific MODIS product was used? If multiple products were used this should also be specified.
Line 174: The term ‘freezing-frequency’ could be better defined. Does this mean there are years when the lake does not fully freeze? Would it be possible to provide some context over the last decade? Is it fully frozen every other year? Or partially frozen each year?
Line 177: Is there data to support this? How do we know the effect has been small?
Line 193: There is a double ‘are’.
Line 224: “because of” can be changed to “for” to make the sentence concise.
Line 266: ‘(left)’ can be removed.
Line 281: The spectral range of the bands should be provided.
Line 308-310: “Another neat…” this sentence should be revised. “Spatial convolution was applied independently to each channel, followed by a…”
Line 376: ‘predict also’, the ‘also’ is not needed.
Table 8 + 9: Having the names of the lakes in the tables would be good so that it can understood without the caption.
Qualitative Results: While it is a nice comparison to show this data relative to the ground truth data, additional comparison to existing operational products (i.e. MODIS snow and ice) would be good. These threshold methods have shown to be effective and comparing the percentage of ice cover between the two methods would be a good way to demonstrate the increased operational capacity of the machine learning algorithm. Additionally, this would allow quantitative analysis to be conducted on transition dates and determine how both methods perform.
In Figure 12 there are clearly some VIIRS images that have estimated days with no ice cover mid-winter. This also occurs when there should be only water estimated for both MODIS and VIIRS. The explanation appears to be related to thin ice vs water – some explicit examples should be highlighted in the text.
Figure 13: Provide the location of the images shown in this figure.
Line 480: ‘prior art’, this refers to prior results?
Figure 16: Provide the location of the images shown in this figure.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 4 Report
I was particularly disappointed in the English writing of this paper. To me, this paper is still in draft format and needs extensive editing and reorganization. I am interested in the topic presented here and the ability to build a vast database of lake ice on/off dates, this particular manuscript needs further work. More specific comments about the manuscript are given below.
- Introduction: this section seems to meander a lot. A large section on pages 2 and 3 should be moved to the methods section of the paper. The section "1.1 Related Work" shouldn't be a subsection, it should be the introduction. In section 1.1, the first paragraph makes a series of vague statements. Remove these statements or make them meaningful.
- Methods, Ground Truth: ground truth data are very important, and in this study rather subjective. Please go into further detail about how the process. Did you have multiple people look at each image or only one person per image?
- Results, Figure 11: include a legend for the color gradient
- Results, Figure 12: This is a very confusing way to depict the data. I recommend using dates on the x-axis and considering using different colors.
- Results, Figure 13: I am not sure why the legend includes the categories of ice, snow, and clutter.
- Results, Figure 16: no legend included.
- Results: It's unclear to me how you handle the case of persistent snow on the ice. Please describe it further.
- Conclusions, line 605: You make the claim that your demonstrated approach "generalises well across winters and lakes". Given you only have two lakes in the same region and two back-to-back winters, I do not think you should make this statement.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 3 Report
The authors have done a good job integrating and addressing comments throughout the manuscript. However, there are still some areas that need to be improved. No lines were provided in the latest version of the document but additional comments below work to highlight specific areas.
The major fix that needs to be made to the manuscript are the short-bolded sentences. It is unclear whether these are intended to be subheadings, but that appears to be the purpose and should therefore appropriately numbered. Otherwise, these sentences should be removed and connecting sentences should be added to make the purpose of each paragraph clear.
Apologies for the unclear comment regarding the abstract, not a citation, but the abstract should at least mention the larger implications of this research, i.e. the use as an operational system.
“Moreover, winter lake ice is depleting at a record pace…” is there a citation for this claim?
The introduction must be modified. While the changes the authors have made are good, it does not follow a logical structure. The biggest suggestion would be for everything to be written more concisely and for descriptions of past research be shortened to one sentence or two at most. For example, the second paragraph on the second page could be significantly cut (the most important sentences are the first and last, the rest could be rewritten and added in pieces to the first paragraph) and added to the end of the first paragraph, which would give the section more flow. Additionally, the objectives of the study should be reorganized, currently, they are at the bottom of the first and second paragraphs. It would be more appropriate to be listed before the definitions of ice phenology and should be stated in a single paragraph that highlights both the objectives and the need for this research. The bold headings are also confusing. It is unclear if these are subheadings or short sentences, either way, they do not suit an introduction.
Table 2 heading: meteo should be fully expressed as meteorological station.
Figure 3: This figure is out of place; it should be directly below Figure 2 or be combined with Figure 2.
Table 6: This table should be placed above Figure 6 to provide the appropriate context.
Figure 10: This should be below the first methodology paragraph.
In the last paragraph before section 3.2, the last sentence states “This is also the reason why some days were estimated as non-frozen during mid-winter (see VIIRS timeline, February).” Does this mean that the clouds are causing days to appear non-frozen in February? Wouldn’t they be classified as frozen still as seen in April/October? This is confusing.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.
