Assessing the Available Landslide Susceptibility Map and Inventory for the Municipality of Rio de Janeiro, Brazil: Potentials and Challenges for Data-Driven Applications

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors The paper builds a landslide-susceptibility map by using a GAM-based model. My major comments are as follows.

1. What is the specific reason to use GAM over other linear/tree-based/NN-based models? Restricting the GAM to only slope and elevation ignores other critical covariates, which might limit the performance.
2. How does the model consider the spatial-temporal feature?
3. The study omits spatial cross-validation, so spatial autocorrelation likely inflates the reported accuracy.
4. Evaluating the map by the share of events in each susceptibility class is biased by class area and should be replaced with threshold-independent metrics.

Author Response

Comment 1:  The paper builds a landslide-susceptibility map by using a GAM-based model. My major comments are as follows.
What is the specific reason to use GAM over other linear/tree-based/NN-based models? Restricting the GAM to only slope and elevation ignores other critical covariates, which might limit the performance.

Response 1: Thank you for this insightful comment. We acknowledge that alternative approaches—such as tree-based methods or neural networks—can provide valuable predictive capabilities. Nevertheless, such approaches often require heavier computing power. Our choice of Generalized Additive Models (GAMs) was guided by their simplicity, interpretability, and methodological transparency, particularly suitable for a preliminary and exploratory assessment. On top of that, we hoped to test such a simple model to verify if it would be useful in many scenarios where there is not enough computing power nor detailed geological databases, thus providing many municipalities in the Global South with tools that can be useful in determining areas that require immediate attention concerning landslides. It is important to note that this work represents an initial but scientifically sound investigation, designed to establish a baseline understanding of the available data and mapping limitations. The analysis’ focus was to test a model that balances explanatory power with ease of communication. As clarified in the revised manuscript (Section 3.2.4), GAMs make it possible to interpret the individual effects of each predictor—such as slope and elevation—and to present results in a format that is accessible to public decision-makers and non-specialist audiences.

Comment 2: How does the model consider the spatial-temporal feature?

Response 2: We appreciate this observation. At this stage, the model does not yet incorporate spatio-temporal components. This limitation has now been clearly stated in the manuscript (section 5.4). Nevertheless, it is feasible to imply this possibility may be incorporated in the future. Again, we emphasize the major objective of this instance of our research was to test the feasibility of using a simple model that provides satisfactory landslide susceptibility maps under limited data and computer power. Future developments in the broader RioSLIDE project will explore spatial-temporal modeling as referenced in the cited literature (e.g., Lombardo et al. 2020; Moreno et al. 2024, Steger et al. 2024).

Comment 3: How does the model consider the spatial-temporal feature? The study omits spatial cross-validation, so spatial autocorrelation likely inflates the reported accuracy.

Response 3: We agree with this important point and have accordingly revised the manuscript to include spatial cross-validation, using a k-means partitioning strategy as implemented in the sperrorest package in R (Brenning, A. 2012). Spatial cross-validation guarantees training and test sets are spatially separated, thus avoiding an artificial accuracy due to overfitting. Also, it works well when data are not regularly distributed, which is a major feature of landslide occurrences. This approach allows us to account for spatial autocorrelation and assess the model’s ability to generalize across spatially distinct areas. In addition to random and spatial cross-validation applied to the original dataset (2011–2016), we have also tested the model on a second, independent inventory (2017–2018) to evaluate its predictive performance on previously unseen data. The revised results (Section 5.2) present the AUC values obtained under each validation scheme, providing a more robust and spatially aware assessment of model performance. These additions are now highlighted in red in the revised manuscript. 

Comment 4: Evaluating the map by the share of events in each susceptibility class is biased by class area and should be replaced with threshold-independent metrics.

Response 4: Thank you for this relevant observation. We have now included a discussion of this limitation in Sections 5.2 and 5.4, and expanded it with a new paragraph (also in red) that presents the use of AUC (Area Under the ROC Curve) as a threshold-independent metric (section 4.2). AUC was calculated using both random and spatial cross-validation procedures, thus ensuring the model’s viability. This addresses the concern of area bias and provides a more statistically relevant evaluation.

Finally, we would like to thank you for taking the time to review our article. We are hopeful that the adjustments we made are deemed acceptable and the paper is fit for publication. Best regards.

Reviewer 2 Report

Comments and Suggestions for Authors

Dear Authors

Scientific comments

The article is good and will be a valuable tool for the scientific community in supporting decision-making. However, the introduction should also consider:

It should be noted that mapping in urban areas faces the constant reorganization of the urban environment, whether through the emergence of new buildings or the reorganization of existing space, particularly in areas undergoing building renovation.

In geotechnical mapping, in particular, urban reorganization allows for the observation of a vast collection of data, and the main challenge lies in the minimum temporal and spatial exposure of the natural terrain.

The use of Geographic Information Systems (GIS) in the development of geotechnical mapping has taken significant steps forward, through the use of geographic databases, new analysis tools, and, more recently, the use of free software and platforms.

Describe the need for thematic geology maps (soils and/or rock masses – rock type, fracturing – dip, dip direction – degree of alteration, among others) and, fundamentally, a thematic slope map. (e.g. … indicate that most of the terrain is (line 313) characterized by low to moderate slopes (less than 25°) and low elevations (generally below line 314 200 m). However, the training sample distributions (Panels B and E) reveal that sliding (line 315) occurs more concentrated in areas with intermediate slopes (15–35°) … this sentence could be more clarifying with this map.

On the other hand, this slope chart will be related and more noticeable, e.g. Fig.6, Panels (A) and (D) present the frequency distribution of slope and elevation, and the others that follow. They convey the same idea, but interpretation will be better within the IAEG concepts.

Note that slope failure in residual or lateritic soils (as in Brazil), resulting from the alteration of granite or other iron-bearing rocks, is often linked to the suction developed in unsaturated soils and which disappears when water infiltrates the soil. Suction and the angle of friction in other areas are important effects that resist the stress imposed by the weight of the soil on the failure surface to be developed.

Geotechnical charts with these mechanical parameters and the slope become important tools in deciding and inferring the factor of safety.

The presentation and understanding of Figures and Tables should precede these, and they should all be cited, e.g. …As shown in Figure 3 …, … (line 113) and line 357 (the text of Figure 10.a does not exist?)

If possible, translate Figure 5. Example of an inspection report. Form filled out by field professionals at the occurrence site.

Are Figures 6 and 9 cited in the text?

The bibliography is adequate; however, it should consider classical authors in soil mechanics and not write general ideas with multiple citations (e.g., line 78 …crucial factors in analyzing landslide occurrences [18–21].

Editorial comments

Check: line 247, 253 Figure??

Check: line 327, 329 Figure7.A and 7.B

Check: line 364 (Figure Figure)

 

Author Response

The article is good and will be a valuable tool for the scientific community in supporting decision-making. However, the introduction should also consider:

Comment 1: It should be noted that mapping in urban areas faces the constant reorganization of the urban environment, whether through the emergence of new buildings or the reorganization of existing space, particularly in areas undergoing building renovation.

Response 1: Thank you for your comment. We recognize that urban environments - particularly informal settlements - are highly dynamic, making it challenging to model their evolution with both high spatial and temporal resolution. This inherent difficulty was a key motivation for adopting a data-driven approach, as discussed in the introduction (with addenda in red).

Comment 2: In geotechnical mapping, in particular, urban reorganization allows for the observation of a vast collection of data, and the main challenge lies in the minimum temporal and spatial exposure of the natural terrain.

Response 2: Thank you for your insightful comment. We agree that the temporary exposure of natural terrain during urban reorganization provides a unique opportunity to observe key geo-environmental interactions, even if spatially and temporally limited. More broadly, this perspective reinforces our rationale for adopting a data-driven approach, as it allows these transient but valuable observations to be formalized and incorporated into predictive models. 

On a practical note, we also emphasize that Geo-Rio is advancing automated landslide detection, which will enable future inventories to be updated in real time. Our choice of a simple yet effective model—based primarily on altimetric and landslide data with minimal additional inputs—was deliberate, as many municipalities in the Global South face limited data availability and computational resources. This ensures accessibility while still offering a meaningful tool for susceptibility mapping.

Comment 3: The use of Geographic Information Systems (GIS) in the development of geotechnical mapping has taken significant steps forward, through the use of geographic databases, new analysis tools, and, more recently, the use of free software and platforms.

Response 3: We appreciate your insightful comment. Indeed, our methodology is fully GIS-based, encompassing all stages from data acquisition to spatial analysis in R. Moving forward, we aim to enhance our framework by incorporating a temporal dimension. This will involve developing dynamic susceptibility maps integrated with nowcasting systems to improve real-time assessment.

Comment 4: Describe the need for thematic geology maps (soils and/or rock masses – rock type, fracturing – dip, dip direction – degree of alteration, among others) and, fundamentally, a thematic slope map. (e.g. … indicate that most of the terrain is (line 313) characterized by low to moderate slopes (less than 25°) and low elevations (generally below line 314 200 m). However, the training sample distributions (Panels B and E) reveal that sliding (line 315) occurs more concentrated in areas with intermediate slopes (15–35°) … this sentence could be more clarifying with this map.

Response 4: Thank you for pointing that out. Some of those maps already exist for Rio de Janeiro, and will be used in further developments of this research (section 5.4). At this point, however, we sought to be as simple as possible. From a geomorphological perspective, starting with slope and elevation—the most fundamental morphometric controls—allows us to isolate their direct relationship with observed landslides, while keeping the model transparent.

Comment 5: On the other hand, this slope chart will be related and more noticeable, e.g. Fig.6, Panels (A) and (D) present the frequency distribution of slope and elevation, and the others that follow. They convey the same idea, but interpretation will be better within the IAEG concepts.

Response 5: We thank the reviewer for this observation. Panels (A) and (D) in Figure 6 show the frequency distribution of slope and elevation for all pixels within the study area, whereas Panels (B) and (E) show the distribution of these variables within the balanced training sample used to fit the model, distinguishing between landslide and non-landslide points. Although both address slope and elevation, they represent different populations: (A) and (D) depict the full-area morphometric characteristics, while (B) and (E) describe the modelling dataset. This distinction is now explicitly stated in the revised manuscript text and figure caption (section 4.1). We would be grateful if the reviewer could clarify the intended application of the IAEG slope classification concepts in this context, and, if possible, provide specific references or examples of similar implementations to guide our revision.

Comment 6: Note that slope failure in residual or lateritic soils (as in Brazil), resulting from the alteration of granite or other iron-bearing rocks, is often linked to the suction developed in unsaturated soils and which disappears when water infiltrates the soil. Suction and the angle of friction in other areas are important effects that resist the stress imposed by the weight of the soil on the failure surface to be developed.

Response 6: The requested addition has been incorporated (lines 127-132).

“In tropical residual soils, like those in the mountainous and urbanized areas of Rio de Janeiro, slope instability has been associated with rainfall infiltration and its effects on the hydro-mechanical behavior of soils. Studies from the region [31-34] describe how infiltration can lead to a reduction in matric suction, which lowers the apparent cohesion and may contribute to shallow slope failures under intense or prolonged rainfall.”

Comment 7: Geotechnical charts with these mechanical parameters and the slope become important tools in deciding and inferring the factor of safety.

Response 7: Whilst we acknowledge the importance of geotechnical charts, we emphasize that we sought simplicity for this initial assessment of our data-driven model. Geotechnical charts that relate mechanical parameters such as cohesion, friction angle, and unit weight to slope angle are indeed valuable for estimating the factor of safety at the slope scale, where detailed field and laboratory measurements are available. However, their direct application becomes challenging in regional or large-scale susceptibility mapping, where such parameters are rarely sampled systematically. Physically based models depend on—and can benefit from—these parameters, but they are difficult to extrapolate reliably for large-area assessments. In contrast, geomorphological and thematic variables can often be measured and extrapolated consistently across the study area, making them more suitable for broader-scale modelling when high-resolution geotechnical data are absent.

Comment 8: The presentation and understanding of Figures and Tables should precede these, and they should all be cited, e.g. …As shown in Figure 3 …, … (line 113) and line 357 (the text of Figure 10.a does not exist?)

Response 8: Thank you. We have tried to correct all miscitations. Nevertheless, image placement is done mostly by LaTeX, so we don’t have much control - at this point - of how figures will be distributed. We have published articles with MDPI before, however, and know their editing services fix all such issues. 

Comment 9: If possible, translate Figure 5. Example of an inspection report. Form filled out by field professionals at the occurrence site.

Response 9: Thank you for your comment. We hoped to use the original inspection report, so readers have a glimpse of how it looks like. In order to provide a meaningful translation, the image was adapted to incorporate translations in red, and below the original text.

Comment 10: Are Figures 6 and 9 cited in the text?

Response 10: Thank you. We have now cited them throughout the text.

Comment 11: The bibliography is adequate; however, it should consider classical authors in soil mechanics and not write general ideas with multiple citations (e.g., line 78 …crucial factors in analyzing landslide occurrences [18–21].

Response 11: Thank you for providing such comments. We have revised the manuscript by incorporating (1) a wider survey of landslide susceptibility literature (throughout section 2.3), including mitigation, and (2)  refined vocabulary to clearly address the difference between risk and susceptibility.

 

Comment 12: Editorial comments

Check: line 247, 253 Figure?? 

Check: line 327, 329 Figure7.A and 7.B

Check: line 364 (Figure Figure)

Response 12: Thank you. We have now revised those minor mistakes and they were corrected.

Finally, we would like to thank you for taking the time to review our article. We are hopeful that the adjustments we made are deemed acceptable and the paper is fit for publication. Best regards.

Reviewer 3 Report

Comments and Suggestions for Authors

At the start you say many landslides a quantity would be helpful. Although mitigation I believe is beyond the scope methods of slope stabilization could have been mentioned. The understanding of the wider field seems weak. Are these methods applied in other urban areas when considering landslides. What does the literature say? Some sort of structured literature review would strengthen the paper. You use the term risk I assume this is likely hood and fatality rate but it needs to be defined. Slope treatment and vegetation could have been discussed further.

Author Response

Comment: At the start you say many landslides a quantity would be helpful. Although mitigation I believe is beyond the scope methods of slope stabilization could have been mentioned. The understanding of the wider field seems weak. Are these methods applied in other urban areas when considering landslides. What does the literature say? Some sort of structured literature review would strengthen the paper. You use the term risk I assume this is likely hood and fatality rate but it needs to be defined. Slope treatment and vegetation could have been discussed further.

Response: Thank you for providing such comments. We have revised the manuscript by incorporating (1) a wider survey of landslide susceptibility literature  (throughout section 2.3), including mitigation, and (2)  refined vocabulary to clearly address the difference between risk and susceptibility. Finally, we would like to thank you for taking the time to review our article. We are hopeful that the adjustments we made are deemed acceptable and the paper is fit for publication. Best regards.

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Thanks for the revision. No further questions.