Multi-Temporal Satellite Image Composites in Google Earth Engine for Improved Landslide Visibility: A Case Study of a Glacial Landscape

Round 1

Reviewer 1 Report

I am not an expert in remote sensing -based landslide mapping. I believe that the authors have done a good job on landslide detection with Google Earth Engine. In this round of review, without going into details of the paper, I’d like to give some broad suggestions.

1. The to-be-solved question should be focused, all through the paper. In the abstract, the authors state that "databases of historic landslide events are often spatially biased towards roads or other infrastructure, with few reported in remote areas". This is a potential to-be-solved question. Then, the conclusions of this paper should answer: (1) to what extent the procedure proposed in this paper have reduced the spatial bias, like “a decrease from 100% to 30%” the authors mentioned in the abstract; (2) why the procedure proposed in this paper can reduce the spatial bias.

2. Any of remote sensing -based landslide mapping technologies will be useful in detecting landslides in remote areas, and in turn will help in reducing spatial bias. And, I suppose that there are also many previous studies that had mapped landslides with Google Earth Engine. The authors should, show the advantage of their procedure, compared with other remote sensing -based technologies, and more specifically compared with previous applications of Google Earth Engine. The authors should show the readers that their procedure is an improvement of previous ones, and the improvement gives the decrease of spatial bias.

I suppose that the above two questions might be more or less answered in the main text, but I fail to find the answers in the abstract. Other relevant sections should also be adjusted to clarify the contributions of this paper.

A separate methodology section is needed, so that the difference and improvement of the presented procedure compared with previous ones can be directly shown to the readers.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

Very interesting and highly technological work.
However, it is necessary to specify that the results are pertinent to specific forms of instability, typical of the investigated area.
Furthermore, if the methodology could result important in hazard and risk mapping, the connection between hazard and risk and early warning must be clarified. While the triggering thresholds can be useful for this purpose, the methodology does not examine the fundamental aspect of early warning, that is, the positive variations in the speed of the landslide body.

a little observation:

m.a.s.l.

m is a symbol, not a shortening. So it does'nt need the dot:

m a.s.l.

Author Response

Thank you to reviewer 2. I am glad you find it to be interesting!

R2:  it is necessary to specify that the results are pertinent to specific forms of instability, typical of the investigated area.

EL: I have added more details about the types of landslides and the glacial geomorphology that controls the type of landslides observed in the region into the case study description. I also expanded the section in the discussion that describes how the landscape affects the results, and how the results may vary in other areas. 

R2: if the methodology could result important in hazard and risk mapping, the connection between hazard and risk and early warning must be clarified. While the triggering thresholds can be useful for this purpose, the methodology does not examine the fundamental aspect of early warning, that is, the positive variations in the speed of the landslide body. 

EL: I wonder if you are referring to local landslide early warning instead of regional landslide early warnings? I realise now that I did not specify that I am focussing on regional LEWS. I have added the word regional into the abstract, and changed the sentence in the introduction that said 'evacuate people' to warn people to avoid risky areas'. In Norway at least, they tend to only evacuate people when it is a local LEWS, but they do not evacuate people from entire regions. So I can see that that was unclear. 

I could be too narrow in my focus - but I don't think the speed of the landslide body is relevant in regional landslide early warnings. 

R2: correct m a.s.l.

EL: thanks! done. 

Reviewer 3 Report

The use of satellite data for rapid, large-area slope stability analysis is an emerging issue. The study presented here is an important contribution in this area, supported by labor-intensive field observations. 

Some notes and comments are included in the pdf document - please read and respond.

In the issues raised, it is vital to integrate, or be aware of, the need for verification/analysis that takes into account the deep geological structure that may determine landslide activity. Hence, please add in the text - discussion chapter - for which landslides the presented method is best applied - shallow/deep landslides or maybe other factor determines the usefulness - applicability range of the method.

All best

Comments for author File: Comments.pdf

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 4 Report

I find the content of the paper novel and significant. The quality of the presentation is acceptable but commented on what could be improved. I have also some moderate comments related to the introductory part of the paper. See the attached pdf file for more details. 

Comments for author File: Comments.pdf

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

I have no further concerns regarding the statement of the to-be-solve question.

Author Response

Thank you!

Reviewer 4 Report

The authors have done a good job in responding to most of the comments but I still have three comments remaining (see the attached file). 

Comments for author File: Comments.pdf

Author Response

R4: points 1 & 2 

(1) R4. Your answer put some light into my question but I think a more rigorous text based on that should be incorporated in the paper. The novelty of your study is now clear to me but you should present more rigorously the most important previous studies of multiple landslides (for example those that you cite or you mention in your answer) and explain the bias in each of them separately.

(2) R4: I do not see answer to my question in the quoted abstract. Please, elaborate. Moreover, I suggest to list the five countries that you mean in parentheses.

EL: I have expanded on the 2 paragraphs, explaining in more detail the bias and what is not currently done, and adding the list of countries and a fuller explanation of how the completeness is estimated:

However, for operators of LEWS, collecting landslide data and verifying warnings is a difficult and tedious task [7]. Generally, the registration of events and preparation of landslide inventories are not performed systematically [8]. Presently, single landslide event data is collected mainly from ground observations, including reports from road and rail authorities, and to a lesser degree from the public, or mined from social media and news reports [9,10]. For triggering events (e.g. extreme precipitation or earthquake) with multiple resulting landslides, mapping using aerial or satellite optical images may be undertaken on a case-by-case , depending on motivation and resources [11–13]. Such mapping is most commonly limited to areas that are already known to have been affected by landslides (e.g. [14–16]), or to verify the exact location of landslides that may have been reported without precise coordinates, for example in media reports [17]. Systematic mapping is generally not yet done over large areas, or to identify previously unknown landslides in remote areas.

The resulting landslide databases, however tend to show clusters of observations around linear infrastructure and populated areas, thus not giving a realistic representation of the true spatial distribution of landslides, as can be seen in Figure 1. Additionally, landslide inventories tend to be rather incomplete. For example, in Europe a comparison between landslide density based on landslide records from the Geological Surveys of Europe, and the European landslide susceptibility map (ELSUS 1000 v1), only three countries (Poland, Italy, and Slovakia) have inventories to be over 50% complete [18]. These issues with the spatial bias and missing data limit how the landslide data can be used for predictive models [19].

R4 point 3 - fix the figure coordinate labels

EL: done!

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.

Round 1

Reviewer 1 Report

Some works and comparison have been performed in this work. However, I cannot catch the novelty and significance of this work. Personally I think this manuscript need a restructure and to re-summarize the novelty and significance. It was suggested to re-submit this manuscript after a dramatic major revision. Detailed reasons are listed below, to help improve this manuscript.

1. First of all, the title was too wide and was not appropriate. Actually, this paper is a case study with a result and some comparison. In the conclusion, I cannot see the potential and limitation of Sentinel-2, or they are superficial.

2. The novelty and significance should be summarized and emphasized after a deep thinking.

To be honest, the landslides mapping and comparison in this work were very simple. It seems all the landslides mapping were performed by visual interpretation. In fact, if the coverage of Planet images or drone images are wide enough, the Sentinel-2 images will not be used. This kind of visual interpretation of these images and comparison are clear and well-known.

What the roles of SAR images comparison and NDVI comparison play in this study should be considered more. It is clear SAR images will have poor performance compared to Sentinel-2 images.

If the dNDVI was a big part of this paper, is this paper a methodology analysis or application of Sentinel-2 images landslides mapping? The next question for the authors will be dNDVI have better performance over the visual interpretation as shown in Figure 2? Therefore, it is strange that the dNDVI results become a chapter in the results, which is different from the previous results and comparison.  

The introduction part was too wide. If the topic is Sentinel-2, the literature review should focus on the landslide mapping with Sentinel-2, including different methods, rather than review on SAR, NDVI, Lidar,GIS as its current stage.

Overall, the logic of the whole manuscript need to be considered again.

Reviewer 2 Report

General observation:

The paper is attempting to review the possible potentiality and drawbacks of employing the Sentinel-2 data for landslide detection.

The authors begin by talking about landslide detection in general, its importance, and then discuss the completeness and quality of the inventories that are generated by different experts. Systemic annotation and demarcation of landslides follow different conventions and interpretations by different experts. While some countries record every single detail, others only record some important information like damage events or triggering events. The authors comment that countries with large areas of landslide susceptibility, low population density, and lack of systematic landslide mapping tend to have less complete inventories. Therefore, examples such as Hong Kong are a good instances that have been monitoring and collecting landslide records since the 1980s. Mapping with freely available data such as the Sentinel-1 SAR and Sentinel-2 multispectral data can have huge potential, as said by the authors, and which are also testament to the many literature that available. Moreover, a brief overview of the limitations of the spatial resolution limitations are discussed when compared to higher resolution satellite images for landslide detection.

Comments:

1. As stated by the authors, “The purpose of this study was to take a critical look at the potential and limitations of using Sentinel-2 data for dNDVI-based landslide detection” is not reflected in the main topic of the paper.
2. Their principal conclusions stating that Sentinel 2 “far outperforms” is only within the context of the dNDVI. But research with other techniques (like deep learning) suggest otherwise (https://www.researchgate.net/publication/351297987_Improving_Landslide_Detection_on_SAR_Data_through_Deep_Learning) where Sentinel-1 at times produces results much closer to Sentinel-2 data.
3. Their principal conclusions stating that Sentinel 2 “far outperforms” is only within the context of the dNDVI. But research with other techniques (like deep learning) suggest otherwise (https://www.researchgate.net/publication/351297987_Improving_Landslide_Detection_on_SAR_Data_through_Deep_Learning) where Sentinel-1 at times produces results much closer to Sentinel-2 data.
4. Their principal conclusions stating that Sentinel 2 “far outperforms” is only within the context of the dNDVI. But research with other techniques (like deep learning) suggest otherwise (https://www.researchgate.net/publication/351297987_Improving_Landslide_Detection_on_SAR_Data_through_Deep_Learning) where Sentinel-1 at times produces results much closer to Sentinel-2 data.

Based on my observation and the above comments I suggest resubmitting the manuscript after considering the comments.

Reviewer 3 Report

The authors show the limitations of using satellite images for improving landslides detection and collection. Their study is based on the comparison of different images (Sentinel-1-RGB, Sentinel-2-dNDVI, Planet Scope and drone images) and refers to a case study in western Norway, recently hit by a rainstorm event. Their workflow and the results achieved are clear. The authors provide readily available results, but at the same time I believe the paper could be improved in some aspects. Furthermore, the innovation of the study/results is not clear or not well expressed. For example, the authors cite machine learning techniques for landslides mapping, but do not explain why they prefer to use a more classical approach based on qualitative images comparison.

General comments:

Title. As they mention in the text, their investigation needs more tests (lines 551-552; lines 650-651), to evaluate the effects of vegetation cover and seasonal conditions. I suggest using a more applicative title for their paper, and even less exhaustive (“potential and limitation” are arduous words for a single case study. For example, the authors who cited, i.e. Montini et al., 2019, present the results of a pilot study on 32 worldwide cases and use a less challenging title: Sentinel-1 SAR Amplitude Imagery for Rapid Landslide Detection). A title related to the usefulness of Sentinel-2 data-dNDVI for landslide detection in the test site (Jølster, western Norway) could be more appropriate.

Abstract. The sentences written in the abstract appear as juxtaposed concepts, one separated from the other. For example, you begin introducing the use of satellite images for landslide detection (line 99). Then you introduce the limitation of the automation for the detection of landslides (lines 10-11). So you start with your case study (lines 11-12). What is the link between your study and the limitation you mentioned? Perhaps it is enough to introduce simple links between sentences. In the cited case you can use: “to better evaluate the limits of the landslide detection a case-study based on a heavy rainstorm in Norway in 2019 has been considered…”, or something else. Please, try to make more connections between sentences.

Introduction. Please, supplement the introduction by also considering the modern technique of automatic landslide recognition and mapping, for example based on machine learning, using satellite images. Also in lines 576-579 you mention recent research working on automatic process based on machine learning, but you do not provide any references.

Some repetitions are found in the text. For example, the “Results” section begins with a few lines describing the organization of the results (283-292), and shortly after, in the “Discussion” section (lines 582-585) the same is repeated. Please read and synthetize.

Discussion. The whole beginning of this section appears as a conclusion (lines 574-585). Furthermore, the interpretation proposed by the authors in this section is only discussed with respect to two previous studies (Mondini et al, 2019; Casagli et al. 2016). Are there any other studies to mention and use for a broader discussion?

Specific Comments:

Author affiliations: please include complete address information (city, zip code, state/province, country). NTNU, NGI, HVL, NGU may not be known to everyone.

Figure 1: please define NGU.

Line 240: analysis performed or…better…performed data analysis?

Lines 420-421: the sentence on the S1-RGB images appears somewhat in contrast with the partial detectability sentences introduced in the previous part (lines 378-383). Could you rephrase it, better explaining the limitation in using S1-RGB images?

Figure 4: the images show numbered red squares, the meaning of which is explained in the next figure (fig.5). I suggest to indicate here explanation of the superimposed squares, i.e. the locations (1. Arnes, 2. Slatten ecc.). The caption of fig. 5 can be shortened by simple introducing “see also Fig. 4 for location” or something else.

Line 573. “5. Discussion”, instead of “4. Discussion”

Line 656: “6. Conclusion” instead of “5. Conclusion”