PS-InSAR-Based Validated Landslide Susceptibility Mapping along Karakorum Highway, Pakistan

Round 1

Reviewer 1 Report

“Variuos landslide-conditioning variable” (line 87-88): it’s too general. Please, provide some additional information to give an idea of the kind of the considered variables (e.g. their nature and the number).

What was the origin of the described geological Units in Chap. 2.1 (line 134-142)? Simply describe the relationships between those Units to understand the geological setting; did they form on a paleo-continental margin? What is the nature of the faults? Such information is relevant because, as you highlight (ex. line 147-148), the geology of the area is a key to understand the landslide susceptibility of the area itself. The steep slopes of the mountains should be controlled by the nature of the rock substratum and by overprinting fault framework.

Most of the geological Units named in Fig. 2 are not described in Chap. 2.1: make a coherent description in the text and map.

Figure 1: Indicate the city location (as in Fig. 2). Please, show the Study area limit with a different colour than the Karakorum Highway path (as in Fig. 2).

Figure 2: Add a cross section through the area to show the structural setting crossed the KKH path (the arrangement of the geological Units, their mutual relationships and the fault framework). The geological setting must be clearly described because all such features contribute to the landslide susceptibility of the area and of the KKH.

Figure 5: Maps d-e-f appears to be totally green and thus we cannot appreciate the variability of the parameters. Rescale the colours to highlight the parameters changes across the area.

Figures 5-6: Increase the character size of the scale and legends to make easier the reading. Provide images with higher resolution.

Table 1: I guess that (at least) some of the "topography" parameters should be more properly indicated as "geomorphology". e.g. curvature, roughness, plan curvature, slope.

Figure 7: I suggest indicating the name of the parameters below the histogram instead of the Xn. It should simplify the reading of the histogram. The figure is large enough to allow it.

“Fig. 1. Rock plots of XGBoots and RF models” (line 282): It’s Figure 10. The figure is too large.

Did you calculate vertical/horizontal velocity (line 300-301)?

What are the evidences for the more accurate predictability you refer to at line 304-305?

Figure 11: Specify the range value of the velocity in the legend. What kind of velocity is showed? it is not clear (e.g. vertical velocity?).

Figure 12: Specify the range value of the landslide susceptibility index in the legend of Fig. a; The legend of Fig. b-c is too small to be read.

Figures 14-15: The areas within the circles appear to be significant for your considerations. if it is true, indicate them with a number and check that they are clearly described and recall in the text. Elevation is not a triggering factor (Table 1), rather, it can promote landslides/make an area prone to landslides. Most of such considerations should be anticipated as a comment of the Fig. 7 to explain it.

General considerations: please define an appropriate size for the figures (es. Figure 7 and Fig. 10 can be smaller). Make the legend text size more homogeneous in all the figures.

The description of the causative factors of landslides is poorly developed in the section “Landslide causative factors”, and they are simply shown through the figures; they are poorly discussed also in the “Result” section. Such sections must be improved.

The factors that control the landslide events are treated only marginally. The description of the causative factors of the landslides are listed but not described, it is not explained how they were chosen; moreover, their interpretation is not clear: which are actually considered triggering factors, and which are not, which concern the geology and / or the geomorphology of the area.

Please integrate such information.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The authors used state-of-the-art machine learning techniques such as RF and gradient boosting to generate maps of landslide susceptibility. The resulting maps were refined with data from the permanent scattterer method. The manuscript contains the described methods satisfying the criteria of novelty, and the chosen region and subject matter responds to the criteria of relevance of the problems. The goals set have been achieved, and the intermediate tasks have been solved in full. In general, the manuscript can be published after adjustments have been made to correct typos and inaccuracies, as well as to clarify a number of points, e.g:
Line 74 - artificial neural networks
Line 96 - decrypt ADInSAR and PS-InSAR
Line 82 - the authors write that Catboost achieves better results, but uses XGBoost. Why?
Line 144 - "the maximum annual mean temperature extends from 16 C to 25 C, and the minimum annual mean temperature rises from -3 C to -21 C" - is this the mean or not? Specify. Add a degree sign.
Line 147 - "The region's geological formations and soils are weak, which also play an essential role" - indicate the prevailing geological composition and soil types.
Picture 1 - the borders of the study area are poorly visible on the satellite image, change the color or add a buffer glow.
Figure 3, line 175 - please specify which Google Earth images were used - resolution, date, imaging equipment.
Line 200 - re-calculated.
Table 1 - how exactly NDVI was counted, for what period/date?
Line 210 - jenks natural breakpoint
Chapter 2.5 - why was the RF method used rather than SVM, which works better on small sets of samples? RF with so many trees and such a small input dataset is overfitting the model.

Author Response

Please see the attachment

Author Response File: Author Response.pdf