An Improved R-Index Model for Terrain Visibility Analysis for Landslide Monitoring with InSAR

Round 1

Reviewer 1 Report

The manuscript extends the R-index of Notti et al. (2014) to include also far passive layover regions in terrain visibility analysis of SAR images. Although the extension is based on similar hillshade techniques as used for active and near passive layover regions and for shadow regions by Notti et al., the proposed improvement includes a clever idea to look the terrain upside down. As expected, the proposed improved R-index yields better results in the experiments than the previous R-indices.

The manuscript is very clearly written and especially, the geometrical drawings are very informative. The weaknesses include that the R-index seems to use a fixed incidence angle although it increases from near range to far range. The proposed improved R-index gives similar visibility results as obtained by Kropatch and Strobl (1990) (which is used as ground truth in the experiments), so although the manuscript improves the R-index, it does not outperform other methods for visibility analysis. In fact, the contribution of the manuscript would be higher if the proposed improved R-index were compared also to other existing methods given in [21, 25, 26, 31, 32]. It is also well known that descending satellites monitor better W-facing slopes and ascending satellites E-facing slopes, so the conclusions in this regard provide nothing new.

Detailed comments:

Lines 161-164: If the surface is flat (alpha=0), then R-index=sin(theta). So, "smaller" should be "not greater" and "greater" should be "not smaller".

Figure 1c and lines 172-174: The definition of the angle between the ground surface and radar beam seems to be incorrect for the active shadow area. It should be pi - gamma_4 instead of gamma_4, and then R-index is negative and not positive (line 173). This definition agrees also with the legend for R-index in shadow areas in Figs. 5-8 and 10.
 
Lines 199-200: Not positive means that R-index_(m)<=0. 

Figures 5-8 and 10: In all these figures, the legend for R-index is incorrect, because the value R-index = 0 should be included in the layover and shadow area and not in the foreshortening area. 

Table 1: It seems that in all the R-index models considered, the incidence angle theta is kept fixed although in the reality, it varies from near range to far range. This should be taken into account in the terrain visibility analysis and comparisons in Section 3. It would require that the light source employed in the hillshade model were a point source. If this is not possible to realize in ArcGIS or some other software, then this limitation should be at least discussed in Section 4.

Lines 403-414 and Figure 9: The text does not match with Fig. 9 in regard to the W-facing and E-facing slopes. The layover and foreshortening regions for the descending satellites are on the E-facing slopes in Figs. 9b-c and not on the W-facing slopes. The good visibility and shadow regions are on the W-facing slopes in Figs. 9a and 9d and not on the E-facing slopes. The descending satellites are more suitable for monitoring W-facing and ascending satellites for E-facing landslides, and not the other way round.

Lines 478-480: The shadow coefficient was already present in the modified R-index by Notti et al., so the proposed improved R-index detects only the layover regions better than the modified R-index and not the shadow regions.

Lines 485-487: Descending satellites -> W-facing landslides; ascending satellites -> E-facing landslides.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

The manuscript aims to modify the so called R-index, which defines the visibility of the terrain to the satellite SAR line-of-sight, and is particularly important for mountain areas and landslide investigations with InSAR methods.

 

Major comments:

 

1) The authors build upon the original definition of the R-index proposed in [22], later improved in [23,24] and [26,27], and add into the established formula a step to identify the “far passive layover” regions. This is the main aspect of novelty.

On the other hand, the approach used by the authors for the demonstration of the improved index, and also the whole discussion of the influencing factors are very similar to the [26] study, as detailed below.

Only part of the existing similarity is clearly stated throughout the text, but many sections seem not to fully account for experiments already carried out in past publications, e.g. [23,24,26,30]. Therefore, the discussion of the results in section 4 should be thoroughly revised to give credit to existing studies where the same observations were made in the past.

 

2) Layover and shadow masking: lines 80-83, and 186-191.

This approach was further developed and tested not only in [24] but also in [26], by modelling shadow and layover using hill-shading models with azimuth and altitude angles defined according to the LOS orientation. These lines in the manuscript should be revised accordingly.

 

3) Effects of LOS parameters: lines 23-24, 93-94, section 4.1 lines 398-426, and 474-475.

These effects have been already widely discussed by other authors, so are not novel or original contributions of this manuscript. For instance, section 2.2.4 and figs.9-10 in [26] already analyzed the influence of the incidence angle in the resulting topographic distortions. The above lines and section 4.1 should be revised to take past studies into account.

 

4) Lines 410-411, and 484-485: larger incidence angles are not always better options for mountain regions, because it is true that layover decreases with higher incidence angles, but the shadow increases gradually too, so overall the distortions will not be less. Refer to fig.10 in [26], and revise the text accordingly.

 

5) Effect of DEM resolution: lines 24-25, 94-95, section 4.2 lines 435-460, and 475-476.

This effect has been largely investigated in [26], so it is not a novel or original contribution. For instance, section 2.2.3 and figs.6-8 in [26] already analyzed the influence of the spatial resolution on the input DEM product used for the modelling, in the identification of topographic distortions. Therefore, the above lines and section 4.2 should be revised to take past studies into account. Also, contextualize Figure 11 with results already shown in fig.7 in [26].

 

6) Lines 451-456: explain better how do you identify the 30 m resolution DEM as the best trade-off between the performance of the index and the cost of the DEM product. Provide some costs to show such trade-off, indicate which DEMs are freely accessible, and which ones are commercial. Add some examples of costs for DEM products with appropriate references, to support this conclusion.

 

7) Limitations of original R-index: lines 76-83, and 178-186.

These were largely discussed in [30] and [26], and demonstrated in fig.4 in [26]. Therefore, the above lines should be revised to take past studies into account.

 

8) Factor “Fa”: lines 223-228. Please note that according to Figure 3c, the hillshade model includes not only the “far passive layover”, but also the “active layover”. Check the figure and revise the text accordingly.

 

9) Figures 5, 6 and 7: it is very difficult to see the differences in the maps. Try to zoom into one or more zones of the study region, to make the differences more visible. Figure 8 is, instead, easier for the readers to understand.

 

10) Figure 8: in (a) and (b), edit the colour to indicate the true layover and shadow, to better distinguish them from the modelled ones in (c), (d), and (e).

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

The authors have carefully revised the manuscript to address my initial comments, and the flow of the narration has also been largely improved.

There are some minor revisions that they should consider and implement:

 

Line 243: explain the acronym P-NG and add reference.

 

Line 510: is this cost of 504  RMB for a single DEM tile? By square kilometre? Please specify the reference product size for this price.

 

Figure 12: the added plot has different scales in the X and Y axes, so could be very confusing.

 

Regarding existing literature, in my first report I forgot to mention the following paper and tool, that could be also considered by the authors while assessing existing methods for geometric distortions mapping in SAR imagery: https://doi.org/10.1016/j.envsoft.2017.01.027

Author Response

Please see the attachment.

Author Response File: Author Response.docx