the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Refined glacial lake extraction in high Asia region by Deep Neural Network and Superpixel-based Conditional Random Field
Yungang Cao
Xueqin Bai
Meng Pan
Ruodan Lei
Puying Du
Abstract. Remote sensing extraction of glacial lakes is an effective way of monitoring water body distribution and outburst disasters. At present, the lack of glacial lake datasets and the edge recognition problem of semantic segmentation networks lead to poor accuracy and inaccurate outlines of glacial lakes. Therefore, this study constructed a high-resolution dataset containing seven types of glacial lakes and proposed a refined glacial lake extraction method, which combines the LinkNet50 Network for rough extraction and Simple Linear Iterative Clustering (SLIC)-Dense Conditional Random Field (DenseCRF) for optimization. The results show that: 1) With Google Earth images of 0.52 m resolution in the study area, the Recall, Precision, F1 Score, and IoU of glacial lake extraction based on the proposed method are 96.52 %, 92.49 %, 94.46 %, and 90.69 %, respectively. 2) With the Google Earth images of 2.11 m resolution in the Qomolangma National Nature Reserve, 2300 glacial lakes with a total area of about 65.17 km2 were detected by the proposed method. The area of the minimum glacial lake that can be extracted is about 160 m2 (6×6 pixels). This method has advantages in small glacial lake extraction and refined outline detection, which can be applied to extracting glacial lakes in the high Asia region with high-resolution images.
- Preprint
(20891 KB) - Metadata XML
- BibTeX
- EndNote
Yungang Cao et al.
Status: final response (author comments only)
-
RC1: 'Comment on tc-2022-267', Anonymous Referee #1, 18 Apr 2023
The comment was uploaded in the form of a supplement: https://tc.copernicus.org/preprints/tc-2022-267/tc-2022-267-RC1-supplement.pdf
- AC1: 'Reply on RC1', Yungang Cao, 10 Aug 2023
-
RC2: 'Comment on tc-2022-267', Connor Shiggins, 14 Jul 2023
The manuscript presented by Cao and co-authors put forward a deep learning approach to delineate glacial lake outlines from Google Earth images.
The manuscript is relatively well-written, however there are several parts which require much more detail to provide more contextual information for the reader.
Major comments:
- The introduction as a whole lacks clarity into why glacial lakes are important in the global context and the technical detail of different delineation methods are very complex and are difficult to follow for those new to the subject matter. See minor comments to try and help mitigate these problems.
- On the whole, the first paragraph about the importance of glacial lakes is very limited and only makes statements with no explanation at all (i.e. why do glacial lakes have a strong relationship with climate change?).
- Outburst floods are not necessarily caused by ‘melting glaciers’ so I would strongly suggest this is changed.
- I really struggled to follow the second paragraph in the introduction (L47 to L58). It currently reads as if there is an assumption that the reader has prior knowledge about these techniques. More context is required to explain these technical approaches. The minor comments may help to resolve this problem.
- Why does a new paragraph begin on L47 for the automated approaches, when the authors present the manual and semi-automated approaches in the previous where they were originally mentioned?
- I would consider having 2.0 as ‘Study area’ as a stand-alone section and moving ‘2.2. Data’ into the Methods section and re-naming ‘3.0 Methods and Data’. It currently seems strange to have the data within the study area section.
- In the data section, can the ‘levels’ of Google Earth images be explained? I don’t tend to use these types of data when using openly accessible optical imagery and I’m sure other readers will not have used them either. What are they? Are they like bands on Landsat collections? Please clarify.
- Can ENVI 5.3 please be clarified to what it is and how it assisted the authors in the labelling process. Again, I don’t use this software and I’m sure others haven’t, so it would be good to note what it is and how it assisted labelling as this part is vital to the training of the dataset.
- It is important to note the spatial and temporal resolution (could be done in L145) of the Google Earth images so the reader understands explicitly. How often are images acquired? Is it sparse, or is it continuous? What times of the year is the analysis being undertaken (i.e. July to September)? What years have available data (i.e. 2010, 2015)?
- Is L248 to 260 (first results paragraph) not just methods? The reader is being told what version of Python is being used and the specific package?
- All maps need more detail, north arrow, labels, scale bar. Also, figure captions are very limited (i.e. why are there different colour circles in Fig 5 and Fig 6?).
- It took me a long time to try and understand what figure 8(b) is actually showing and I’m still not convinced I do. What is this heat map actually showing? There is no label on the colour bar gradient?
- The discussion is quite limited in terms of the importance of the study and why this method is worthwhile.
- I’d suggest there needs to be some expansion on why we care about this approach. For example, the paper clearly states small lakes can be identified, so does this mean they are the smallest lakes ever to be identified? If so, how much total lake volume/area have we been missing from global inventories? Could a comparison be made to these global inventories?
- On a glaciological note, I think it would be worth having a figure or table with the seven lake types (as shown in Figure 2) and how each model performs for each category? It would help the discussion, particularly where it is noted on L395.
- Could I ask the authors to go and check abbreviations and ensure they have had their full title spelt out prior being abbreviated.
- Consider use of ‘besides’ throughout manuscript.
Minor comments:
- In the title, should there be an ‘a’ between ‘in’ and ‘high’?
- L10: Would use the term to the ‘glacial lake outburst flood’ – depends on where glaciers/lakes are situated in proximity to towns/infrastructure to be considered ‘disasters’
- L16: What is IoU – be careful with abbreviations
- Remove two examples of ‘about’ on L18 and L20. Be specific.
- L30: Why? Why are glacial lakes becoming larger as a result of increased warming? Statement needs to be explained.
- L31: Besides does not correctly link to previous sentence
- L32: Why? Why are small lakes ‘more active’ and ‘sensitive to climate change’? The last couple of ‘why’ comments are important as this is the glaciological rationale for studying these lakes. From the introduction so far, I’m questioning why glacial lakes are important as only statements have been made with no explanation.
- L34: I appreciate the need for real-time monitoring of glacial lakes and their risk downstream particularly. However, the introduction so far has not cited any examples of glacial lake outburst floods and the subsequent impacts?
- L35: I think for the first instance of using a different word to extraction for the reader. I understand what you mean, but it could raise questions of what is being extracted? Area, volume? Would suggest rephrasing to ‘For the glacial lake outline delineation’ to simplify the sentence.
- Citation needed after ‘results’ on L37.
- Vigour? Just state that it is time expensive and is therefore not suitable for global lake identification
- Citation needed after machine learning on L39.
- L49: Remove canny
- L51 to L52: This sentence makes no sense. What lake extraction is being referred to, is it from a study? What threshold? How is it automatically defined? Why did it not successfully identify lakes? Change sentence as it is not currently followable
- L53: Again, what threshold? If it is going to be referred too, it needs to be stated at the top of the paragraph so the reader understands what this threshold is - i.e. is it a band threshold?
- L55: why is it complicated?
- L56: Why is it limited to Landsat? Is it limited to all collections? Similar to entire introduction so far, it’s too vague of a statement and needs to be more explicit
- L63: Is there any numbers to show it was ‘better’? Vague currently
- L64: Sentence makes no sense – is this a follow-on point from the previous?
- L65: What is ‘skip connection structure’?
- L66: What are ‘low level features’?
- L66: NDWI needs stating before abbreviating
- L67: What is a ‘space attention mechanism’?
- L70: State spatial resolution of Google Earth images
- L75: ‘et al’???
- L76: end-to-end? Remember readers may not have any prior knowledge of machine learning
- L92: Remove ‘undertaken in’
- L97: So? Why does the presence of rivers matter?
- L105 to L119: remove all examples of ‘etc’
- L110: refer to Fig 2 after ‘seven types’ so the reader knows what type of lakes are being categorised
- L118: Why not in high Asia?
- L124: First mention of ‘levels’ of Google Earth images as noted in major comments. Please address.
- L125: Sub-meters? Be specific and use numbers
- L128: Inputted?
- L130: Why are images divided into 256X256? Is it to lessen the computational expense on storage?
- L133: Why did you assume 20% of the training data was suitable for validation? Does this coverage provide enough assurances that the model can produce realistic results? Was it for time or computational reasons?
- L140: Be specific – which Landsat collections?
- L141: Be specific – which Sentinel collections (1,2,3)?
- L149: Remove phrases like ‘by the way’
- 159: Why is it ‘unsatisfactory’ in snow and ice areas? Surely this is important and needs explaining to failures in the model
- L248: Again, with levels. Move spatial resolution up the manuscript
- L263: Rough? What is meant by rough?
- L263: What difficulties? Snow cover?
- Table 5: Why are some numbers bolded? Is it highest scores? Explain in caption
- L319: Why is it difficult to obtain?
- Figure 7: Cannot see bar chart inset
- L355: Is this sentence stating that the training data did not contain enough variety of glacial lakes, and therefore brown lakes were unable to be identified by the model? If so, please rephrase
- L358: What are evaluation indicators?
- L359: What do you consider complex?
- L362: Quote the percentage of reduced misses
- L367: Differs by how much?
- L368: Sentinel 1,2,3?
- L377: misjudge
- L379: Large-scale instead of big?
- L384: First mention of temporal aspect in ‘winter’ – see prior comments about temporal resolution of data
Citation: https://doi.org/10.5194/tc-2022-267-RC2 - AC2: 'Reply on RC2', Yungang Cao, 10 Aug 2023
- The introduction as a whole lacks clarity into why glacial lakes are important in the global context and the technical detail of different delineation methods are very complex and are difficult to follow for those new to the subject matter. See minor comments to try and help mitigate these problems.
Yungang Cao et al.
Yungang Cao et al.
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
274 | 189 | 19 | 482 | 9 | 9 |
- HTML: 274
- PDF: 189
- XML: 19
- Total: 482
- BibTeX: 9
- EndNote: 9
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1