Deep Learning for Exploring Landslides with Remote Sensing and Geo-Environmental Data: Frameworks, Progress, Challenges, and Opportunities

Round 1

Reviewer 1 Report

Thanks for your efforts to illustrate the art of state of DN  through  introducing the frameworks, Progress, Challenges, and Opportunities. It offers an in-depth analysis of how deep learning technologies can be utilized in landslide prediction. It might need some revison:

1.This review is too long for readers to grasp the main ideas and conclusion, could you please shorten the unrelated parts?

2.Why not you can list and introduce the evolution from ANN, CNN, RNN to DNN?

3.You used so many mathematical equations to introduce the principles learning models. For a review article , I not think it is necessary , you may focus on the progress, challenges, and opportunities of deep learning,

4.The article mentions many models, parameters, and metrics related to deep learning. Why not added some detail about model construction, how the dataset is divided, the parameters are selected, the differences between the training set, validation set, and test set, the problem of training model hyperparameter optimization, overestimation. It might to clear for readers to understand how the deep learning model works

5. Regarding the interpretability of the model, the SHAP is a good choice for explaining and interpreting factor weights. Could you provide some examples of interpretable models using in landslide domain? Meanwhile, it is a bit limited just mentioned the SHAP, what about LIME, CAM, Grad-CAM?

6.The article seems to lack a specific and detail discussion on the computational cost resource when using deep learning. Such as GPU, TPU?

7     you mentioned GPT or BERT， but the challenges, cost, problem and future of this technology, what about other AI technology?

 8. The figures are not very clear, I couldn’t see the text in the figure.

9 Considering the depth of the topic, the document could also benefit from including case studies or real-world applications demonstrating the successful application of deep learning models in landslide prediction and management.

Author Response

Author Response File: Author Response.pdf

Reviewer 2 Report

Dear  Authors,

I finally finished the review of the manuscript entitled “Deep Learning for Exploring Landslides with Remote Sensing: Frameworks, Progress, Challenges, and Opportunities” by Qi Zhang  and Teng Wang.

I found the manuscript very interesting and well addressed to the theme proposed by the title. The reading of the manuscript has been very demanding as the manuscript consists of ~ 40 pages. The length of the manuscript is adequate for the type of manuscript proposed by the authors (Review) and, in my opinion, the topic is mainly addressed to experts in the subject.

However, the manuscript is well written and logically organized. The English is good, only few minor changes needed. So I suggest a quick re-reading by a mother tongue.

I have only few suggestion/requests that should further improve the quality of the manuscript.

-The first regards a best definition of “landslides” in the Introduction. According to Varnes (1978) the landslides can be classified according to material or movement type. Which types of landslides can be subject of what the authors discuss in the manuscript?  All kinds of landslides? Or just some type? The dimension of the landslide can play a role in the detection and mapping? I think that the authors should argue this issue in the introduction.

-The second request regards the quality of the figures. Unfortunately, the quality of figures is poor.  Many figures seem blurry images (low resolution) and the written are not clearly readable. So the figures need to be improved.

Once solved these two requests/suggestions, I think that the manuscript is ready to be published. 

The quality of English language is good. Only few minor changes are needed. I suggest a rapid re-reading by a native speaker.

Author Response

Author Response File: Author Response.pdf

Reviewer 3 Report

The manuscript provides a comprehensive overview of the technical framework of machine learning and deep learning, along with their current application status, existing challenges, and opportunities in landslide research. This article holds significant reference value for scholars involved in this field of study, particularly junior researchers.

There are some minor suggestions and comments on this manuscript:

(1) The technical framework introduced in this paper and the literatures summarized not only involve deep learning analysis of remote sensing data but also encompass data analysis from various other sources. Therefore, it is worth considering whether the title matches the paper.

(2) The clarity of the image provided in this paper could be improved, suggesting further enhancement.

(3) When summarizing the applications of deep learning techniques in the four core aspects of landslide research field, it is recommended to conduct additional statistical analyses on the reviewed literatures pertaining to data indicators, accuracy, and efficiency. This will help illustrate both application effects and limitations of deep learning technology through visual representations such as charts.

Author Response

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Thank you very much for answering my question. The paper has a clear logic and comprehensive description of the progress, opportunities and challenges of deep learning in the field of remote sensing landslide, which is suitable for relevant researchers to understand. I read the article again and I have 2 small issues to point out:

1. In the article, you added some tables to compare the performance of deep learning modules, giving corresponding data, which is very good. I also noticed that formula (9) gives the mathematical relationship between F1-Score, Recall, and Precision. I suggest you check the three indicators in Table 5, Table 6 and Table 7 to find the correctness of the data and improve the quality of the references. (Example: Last row in Table 5, reference No. 103)

2. I have a small question about the fourth line in Table 7 (Yi et al. (2022), Reference No. 128). Why is the indicator not a specific value? You wrote AUC (RNN), F1-score,(RNN), Recall (ANN),Precision (RNN).

Author Response

Dear Reviewer,

Thank you for your insightful comments and suggestions. We have carefully considered each point raised and have revised our manuscript. Attached please find our reply and manuscript with revision marks. 

Thank you,
Authors

Author Response File: Author Response.pdf