The study employed deep learning techniques based on convolutional neural network architectures (U-Net and SegFormer) to segment landslides across regions with distinct geological characteristics. The primary objective was to assess the effectiveness and limitations of automated landslide mapping in practical scenarios and to evaluate whether deep-learning approaches can reliably replace manual mapping. The results highlight several important aspects, including a comparison of models trained using different architectures and input layers. The discussion further examines the impact of incorporating a lithological layer into the model and presents an analysis of buildings potentially affected by landslide hazards. Despite the interesting approach to landslide segmentation, the manuscript presents structural and organizational weaknesses that at times hinder readability. A revision is therefore recommended to improve the clarity of the methodological framework and the presentation of results.

The main objective of this work is evaluation of deep learning methods for automated landslide mapping during rapid response scenario using the 2023 Emilia Romagna Disaster as a case study. Two deep learning methods are tested with limited training data, different combinations of raster layers and in different geological settings and using expert judgment as opposed to just accuracy metrics. This paper is interesting but currently it is too long and difficult to follow. Relevant information is scattered all around the paper and repeated. The format requires constant scrolling to different parts of the paper to make sense of some of the findings. Major revision is needed, particularly towards improving flow and shortening the paper. Please see specific comments.

Major comments

1. Introduction

A paragraph is needed on evolution of automated mapping systems highlighting pixel- and object-based systems and how deep learning systems have made improvements rather than just diving into CNNs.

The authors have also used SegFormer which is transformer-based. No information has been provided on this in the introduction. A section on how transformers are different than CNNs and its use on rapid response should be cited.

3. Study area: should be shortened. Lots of information repeated in individual sections again.

3.1 Data

This section is long and difficult to follow. I commend the authors for providing detailed descriptions and providing alternative satellite sources that can be used as opposed to what was used in this study. To make this section more readable I suggest removing these and shortening the text.

It seems L4 and L5 were resampled to 2 m. However, how these were combined with 10 m Sentinel-2 for specific tests are not clear. Same goes for the 5 m DEM, L7.  Also, what is the rationale behind combining same optical bands from medium and very-high optical image?

3.4 Models

What tile size was used as input? Was there any hyperparameter tuning done or did the authors decide on the hyperparameters?

3.5 Training

How inventory developed using high-resolution imagery was adapted to generate mask for Sentinel-2? I am sure many small landslides will not be visible in Sentinel-2 data. Did author do any sort of area thresholding of manual inventory to use with Sentinel-2 data?

4.1 Model result and performance

I think it is important to list metric in the paragraph itself. Referring to table is very difficult; it took me a while to find metrics for S2 that have only 0.01 difference.

Fig 8 and 9 is a little bit confusing. You have true map and predicted map and then TP, FP and FN with same colors. I guess just showing the accuracy metrics is fine. Color can be changed as it is difficult to visualize TP and FP.

5 Discussion

The manuscript is too long as it is, I think the building at risk analysis although useful information can be removed to shorten it and keep the focus at landslide detection.

Also, we need authors perspective on how the shortcoming found in this work can be addressed in future.

Minor comments

Abstract

Line 27-29 should be rewritten.

Line 31 what is FAA?

Line 103 Aerial or satellite imagery?

Fig 1 a what is the source for the rainfall totals?

Line 148- 160 Table 1 should be cited here.

Figure 4 d source of slope map?