Globally vs. Locally Trained Machine Learning Models for Landslide Detection: A Case Study of a Glacial Landscape

Round 1

Reviewer 1 Report

In this paper the authors compare global vs local models for landslide detection. The topic and the objective is quite important and relevant for many countries with an high hydrogeolic risk.

The authors provide a wide overivew of the priovious works related to this subject and of the software tools used for the analysis. In my opinion there are too many technical (software) details that soften the key theoretical aspects and messages. I would suggest to create an appendix with most of the software details.

Reading the paper there is an important aspect which is not clear to me, the methods proposed (Therani model / Prakash model/ etc) require a pre-event and a post-event images within 3 months of the landslide event, for a large scale analysis, where the event is not know a priori for each area, how the images are selected? How this limitation can be overcome?

pag 9 , line 318: "we used random classified tiles": who classified the tiles? which dataset is  used?

To evaluate the performaces of Locally-trained models 2 different models with 4 possible different input data combinations are tested. It is not clear how the setting 2 (S1 + S2) can perform better than setting 1 (S1 + S2 + DEM), in the worst case I would expect the setting 1 to perform as well as the setting 2: can be due to an overfitting problem? how many more parameters are needed in setting 1? 

I agree with the authors that S1 data has to be the main source for a wide scale analysis but I think that other layers of infromation, derived by S1 data, has to be considered such as coherence and displacement maps.

In Fig 8, the SAR amplitude images show several linear artifacts, where do they come from? 

Author Response

 

We thank the editors and reviewers for useful comments and the possibility to revise our manuscript. Below we outline how we have addressed the comments.

Author Response File: Author Response.docx

Reviewer 2 Report

The work investigates which image types and machine learning (ML) models are most useful for landslide detection in a Norwegian setting. It is well-written, the objectives are clear, and the results are interesting. Following are some minor comments:

1. Lines 489-494: The authors explained why the Tehrani model did not detect landslides, highlighting the problems, especially for landslides
on north-facing slopes. To improve the overall reading for readerships, I suggest adding more information about the landslide database (e.g. not only the type of landslides, Fig. 1). In this way, a new Table can be helpful to show but also the "orientation" of the 40 historic landslides recorded in the national database.

2. Line 535: The Authors stated, "We believe in the case of S1-only, that the false negatives are due to the landslide expression in this location being different from other areas". This sentence should be clarified. What is the "landslide expression"?.

3. Lines 627-629: I am not sure that for developing an operational landslide detection system, a SAR-only-based approach using a deep-learning model is recommended for rapid detection as part of an emergency response. As recently reported by Mondini et al. (2021), despite undisputed progress over the last 26 years, challenges remain to be faced for the effective use of SAR imagery for landslide detection and mapping. This study concluded that SAR imagery and related pre-processing and classification techniques are well suited to investigate
single-event landslides or populations of landslides. The findings of this paper can be integrated into the discussions section.

Mondini, A. C., Guzzetti, F., Chang, K. T., Monserrat, O., Martha, T. R., & Manconi, A. (2021). Landslide failures detection and mapping using Synthetic Aperture Radar: Past, present and future. Earth-Science Reviews, 216, 103574.

Author Response

 

We thank the editors and reviewers for useful comments and the possibility to revise our manuscript. Below we outline how we have addressed the comments.

Author Response File: Author Response.docx

Reviewer 3 Report

Please see the attachment

Comments for author File: Comments.docx

Author Response

 

We thank the editors and reviewers for useful comments and the possibility to revise our manuscript. Below we outline how we have addressed the comments.

Author Response File: Author Response.docx