Rainfall-Induced Shallow Landslide Susceptibility for Risk Management of Underground Services in a Mediterranean Metropolitan City

comment

revision

Line 24:  Finally, an early-stage

Done

Line 39: - probably it's about Landslide hazard.

Done

Line 208: - is better as a separate figure with numbering.

Done

Line 216:  Figure 4 - it is necessary to cite a source for the lithological map, unless it is the work of the authors.

Done

Lines 221-224: long and unclear sentence

It has been modified.

Line 304:  The title of Table 2 should begin with a capital letter.

Done

Lines 314-315: Should be supplemented with how the hazard (susceptibility) was calculated - formula? Are the weights of the individual factors summed? Why is the sum 70.3 (Table 3) and how are the five classes of hazard determined? It should be clarified.

There were typing errors in the table that have been corrected. A clearer description of the methodology and detail regarding the pairwise comparison matrix, consistency ratio and consensus indicator have been added.

Line 335: Figure 7 is missing a labeled “D”

Done

Lines 357: In Table 5, it is good to indicate “Number of shallow landslides” in the second column or to add [number]

Done

Table 5 - It is strange to give points for the absence of slope deposits (despite the fact that few points are given). This is probably because there are 26 landslides in them. I suggest to specify the term, e.g. „No presence“ or „Without slope deposits“. Or you can also give 1 for the presence of slope deposits

Done

Line 372: The achieved susceptibility map (Figure 8) – It is necessary to supplement and clarify in 2.2. Research Methods how is this map made?

A clearer description in the methods section has been added.

Line 511:  The title of Table 12 should begin with a capital letter.

Done

Please check the English again. There are minor errors and unclear sentences.

Corrections have been done. The paper will be revised by the proper journal service.

comment

revision

Section 1 Introduction:

1. Gap, novelty claim exaggerated. Rephrase to “limited” or “understood,” narrow by geography/context, and add 3–5 targeted citations.

2. Move city specifics, hazard history, and image/title details to the Scope of Study or Methods. Keep only 1–2 sentences of context in the introduction.

3. Objectives are implicit, not explicit. Conclude the introduction with a short paragraph with clear objectives and specific contributions of the paper (what is new compared to previous work).

4. No research questions or hypotheses. Add 2–3 research questions that link shallow landslides triggered by rainfall to network risk screening.

5. Replace “first-born” with “newly initiated/first-time.”

6. Acronyms are introduced without definition. Define all on first use.

7. There are unsupported general claims. Add sources for high-level statements (damage, frequency, rainfall thresholds, utility outages). Give preference to more recent studies focused on urban conditions.

8. Textbook context is too long. Condensate generic definitions of landslide risk to 2–3 sentences, remove redundant items.

9. Risk vs. susceptibility not distinguished. Add a transitional sentence.

10. Figures should be moved from introduction to study area. Also, title should be shorter and more precise.

1.Done

2.Done. Please see the lines 113-118 and 93-96.

3.Done. Please see the lines 96-98:

“The method has been modified from previous researches [19, 47], improving the quantitative component, and then applied to the risk assessment of water and gas distribution pipelines.”

4.The aim of the paper has been clearly stated. Please see the lines 99-104.

“The aim of the present work is to improve the previous researches in shallow landslides susceptibility assessment, increasing the quantitative component, and applying the result to the shallow landslides risk assessment of the water and gas distribution pipelines in the Municipality. The results represent a preliminary evaluation of the more critical stretches and a possible DSS (decision support system) to plan and adopt a risk mitigation strategy and consequent measures”

5.Done. Please see line 50.

6.Done. Please see line 55 and line 88.

7.Add citations in addition to 1-4

8.Done

9.It has been clarified.

10.Done. Figures and Tables have been moved to Materials and Methods.

Section 2.1 Geographical and Geological outline:

1. Correct the list of nine municipalities (currently repeating “Middle East” and “Middle West”) and align it with the official names of the city of Genoa.

2. The history of population and port is irrelevant. Either (a) link population density and coastal infill directly to network exposure/vulnerability (to inform later risk overlays), or (b) remove. Add source for the 560k/800k claim if retained.

3. Extreme rainfall values ​​are listed but not used. Only keep metrics that feed the methods/results (e.g. intensity and duration context for shallow landslides). Add citations for records from October 1970, November 2011, October 2021, and explain how these extremes justify the choice of factors or thresholds.

4. When introducing lithology, immediately link to the selected causal factors and expected failure styles (e.g., clay-rich flysch leads to shallow landslides).

5. Quantitative terrain context is missing. Add one sentence with basic morphometric data that will be used later (given an area of ​​≈240 km², add slope percentiles/elevation range that you are actually modeling). This sets up the sensitivity mapping in Section 2 and the validation in Results.

6. Since the valleys (Bissaño/Polchevera) structure both landslide paths and utility corridors, close the paragraph with how drainage axes affect the accumulation of flow or the concentration of running water that will be used later.

7. Open section 2.1 with a single sentence: “This subsection summarizes the geographic, climatic, and geological controls that are the basis for the selection of factors for susceptibility mapping and subsequent overlap with gas/water networks.” Then, each paragraph should end with a reference to the Methods or Results.

8. Summarize the generic meteorological description in a single sentence. Keep only what stimulates the shallow landslide hazard and supports the selected precipitation in the Methods.

1.Done. Corrected and removed the contents

2.Done. (a) linked population density and coastal infill directly to network exposure/vulnerability (to inform later risk overlays). Please see lines 109-112.

Added source for the 560k/800k (Lines 126-127)

3.Rainfall values have been included as they are relevant for the related ground effects and for the increase in short time duration, as stated in the Discussion section. References and justifications have been added.

4.It has been explained. Please see lines 202-207.

5.Figures have been added. Please see lines 207-212.

6.It has been added. Please see line 212.

7.A short explanation has been added. Please see lines 107-109.

8.The meteorological description has been shortened.

Section 2.2 Materials and Methods:

1. The method label is unclear. In the opening sentence, state the exact workflow (inventory-driven weights for class intervals + AHP factor weights → weighted sum → 5-class sensitivity) and list substeps to reduce ambiguity.

2. The AHP setup is not specified enough. Report the pairwise comparison matrix, the consistency ratio (CR), and how the factor weights were derived. Otherwise, the AHP is not transparent enough.

3. The training and validation are only random. Replace (or supplement) the 80/20 random part with spatial k-short/block cross-validation and report success/prediction curves or AUC to properly assess generalizability. Keep the event basis, but stratify by basin or lithological unit.

4. The choice of 7 factors is good, but the rationale needs a stronger bridge to 2.1 and the results. For each factor (slope, aspect, lithology, land use, profile and tangential curvature, slope deposits), add a one-line “why it matters here” linked to the Genoa controls (Section 2.1) and to the observed class responses in Table 5.

5. The mismatch between data origin and resolution is not addressed (5 m vs. 1 m DEM). Indicate the CRS, resampling/interpolation method, and where the 1 m ALS-DEM (used for SVF/terraces) does or does not yield morphometry derived from the 5 m DTM, to avoid scale artifacts.

6. Factor detection is described for anthropogenic terraces, but the thickness criterion is not clear. Define how “coverage > 1 m” is estimated (mapping rule, replacement or field verification) and how terrace polygons intersect with factor class rasters. Otherwise, reproducibility suffers.

7. Weight table mixes scales. Numbers seem inconsistent

In Table 3, some factor weights are decimals (0.08–0.09), while three are integers (28.0, 13.0, 29.0) – likely formatting/unit errors. Normalize all factor weights to the same scale and explain the normalization.

8. Model evaluation is described only as “calibration”. Define evaluation metrics and include a brief sensitivity analysis of factor weights.

9. Inventory provenance is strong, but temporal misalignment with predictors is not specified. Since the events span decades, clarify whether the land use (2018–2019) and modern DEMs are intended to predict current sensitivity (fine) and note that the inventory was used solely for statistical weighting and validation, not to reproduce past land use states.

10. The risk step relies only on pipe diameter as a vulnerability

Explain diameter as a proxy (exposed value) and confirm missing drivers (material, age, burial depth, grade transition, joints). If data are not available, present this as an early stage review and add a sentence about limitations indicating future refinement.

11. The units for pipe diameter seem unrealistic (most likely a unit error). The table lists water as 10–900 cm and gas as 15–800 cm. These words are ranges in mm. Correct the units throughout the text (text + Table 4 + captions) and align the 5 vulnerability subcategories accordingly.

12. The construction of the risk matrix is ​​not sufficiently explained. Show the numerical scales of hazard (H) and vulnerability (V), the mapping of the H×V product to the 5 classes (thresholds), and why the final classification is “conservative”. Include the matrix table next to Figure 8 for clarity.

1.A clearer description has been added. Please see lines 230-239.

2.Done, the pairwise comparison matrix has been added in the Appendix A. Please see lines 658-663.

3.AUC and ROC reports have been added

4.Done. Please see lines 325-344.

5.The 1 m ALS DTM has been used for terraces identification, but the final result was down sampled to 5m accordingly to the seven factors. Please see lines 309-318.

6.Description of the methodology may be found in the cited references as terraces identification is not the topic of the paper but has been included for a more precise calculation of deposits and then to get to a more reliable final result. Thickness of deposits derive from the Geomorphological map (Genoa Municipality).

7.All numbers have been correctly reported as decimals. Please see Table 3.

8.The model has been described as requested, adding the pairwise matrix, the consistency ratio and consensus index and AUC and ROC computation. Please see lines 349-364.

9.Land use in the studied area has not significatively changed in the last 2-3 decades. Anyway, the statement has been added. Please see lines 367-370.

10.Done 2018–2019

11.It was a typing error: units have been corrected. Please see lines ,372, 372 and 380.

12.The product between the 5 hazard classes (from 1 to 5) and 5 vulnerability classes (from 1 to 5) is described at lines 328-329 and in figure 8 of the original manuscript. Please see lines 382-392.

Section 3. Results:

1. Calibration-only validation inflates reliability. Replace the single success tally (92% in classes 4–5; 73% in class 5) with spatial cross-validation and report AUC, success & prediction rates. Keep Table 6 but add a CV table/curve so generalization is credible.

2. Susceptibility thresholds lack quantitative justification. Explain how class breaks for Figure 8 were derived (equal interval, quantiles, percentiles by event density). Add a small panel or a supplementary table with area (%) per class to contextualize the map.

3. There is narrative mapping without numbers. Where you describe hotspots, add summary stats. A compact table tying hotspots to factors will tighten the story.

4. Risk relies on "hazard from susceptibility + diameter = vulnerability," but math is opaque. Show the H×V risk matrix with numeric scales and thresholds that map to classes 1–5 (and 1f–5f). Place it next to Figure 8/9 or in a short table so readers can reproduce the classification.

5. Justify the 5° break with a citation/sensitivity check, or test 3–7° and state robustness. Clarify that 'f' classes are separate from shallow-slide risk and how they're combined (or not) for overall risk.

6. Running-water concentration is shown, not integrated. In Figure 10, clarify whether flow accumulation is only illustrative or actually feeds the risk score. if integrated, specify the weight or rule. Consider a two-scenario comparison.

7. Network exposure statistics lack units and denominators.

When stating "12% water, 11% gas in classes 4–5," also report total length (km) per network, and km in high-risk classes. Add 95% CIs if derived from CV folds.

8. Ensure all Section-3 text/figures reflect corrected mm (not cm) diameter bins from Section 2.2/Table 4, or the risk by diameter will be misleading. Cross-check captions for Figures 9–11.

9. Figure 9 caption/process unclear about segmenting networks. State how linework was segmented (by homogeneous H×V class? fixed length?). Add one sentence on topology cleaning and snapping to DEM to avoid sliver segments.

10. Overreliance on a single vulnerability proxy (diameter). Reframe results as an early-stage screen and add a limitations sub-paragraph pointing to future inclusion of material, age, burial depth, joints, slope crossings.

11. State explicitly that the inventory informed factor binning and was also used in calibration. Mitigate with spatial CV and by holding out distinct sub-basins for validation.

12. Keep Section-3 phrasing consistent: use "rapid, rainfall-triggered shallow slope failures" for the mapped hazard. Reserve "flash floods/runoff concentration" for channelized processes shown in Figure 10.

13. Keep Figure 9–11 captions descriptive (what/where), and move method rules (e.g., slope<5° handling) into Section 2.2 to reduce redundancy.

14. Method and Result are aligned in structure, but fix figure labels and diameter units.

15. Statistics is partly there (class weights, MK test, calibration %), but key methodological stats are missing (AHP CR; cross-validated performance metrics). I’d add CR to Methods and spatial CV + AUC/prediction rates to Results for credibility.

16. Figures that are formally adequate are F1, F2 (add metadata), F4 (add map source/CRS), F6 (add years/source). Too generic and need concreteness are figures F8, F11, F12 (and F9 needs clearer vulnerability wording).

1.ROC and AUC assessment have been added. Please see lines 448-455.

2.Done. Please see lines 404-413.

3.As mentioned in the introduction, we received five reviews and therefore had to balance the requests of the many reviewers. At present, the manuscript has 14 figures, eight tables and an additional table in Appendix A.

4.Risk definition for shallow landslides and wall collapse/fall has been moved to section 2.2 and the risk matrix in figure 8 now includes both.

5.The f-classes have been clarified. Please see lines 484-488.

6.It has been clarified that running-water accumulation is not integrate in the computation.

7.Unit and reference values has been added.

8.Values are correct; there was a type mismatch in original section 2.

9.Done. Please see lines 471-478.

10.Done. Please see lines 490-493.

11.The role of the inventory has been specified in section 2.2.  Please see lines 228-239.

12.Done.

13.Done.

14.Done.

15.Done in section 2.2 as requested in the relative section comment.

16.Done. Please see the figures.

Section 5 Discussion:

1. Discussion largely repeats Methods/Results. Start with the meaning of the results, not how the model was built. Keep each method summary to one sentence.

2. “Calibration = validation” weakens the lack of credibility. Replace the calibration sum of 92%/73% as the leading claim of confidence by spatial cross-validation and report the AUC/prediction rate. Keep the calibration number as supporting context.

3. Risk logic (H×V) remains implicit. Add a small risk matrix in the Discussion (or refer to a new figure/table): numerical hazard classes (from susceptibility) × numerical vulnerability classes (from pipe diameter), risk classes 1–5 (and 1f–5f), with thresholds and a sentence about conservatism.

4. Vulnerability by diameter alone is insufficiently justified. Explicitly formulate this as an early stage of screening and list missing drivers (material, age, depth of burial, joints, slope crossings). Indicate how this would change the risk class assignment in future work.

5. Urban “f” classes (rockfalls/wall collapse) pinned rather than integrated (Moderate). Clarify whether “f” classes are parallel or combined with shallow landslide risk and justify the 5° slope threshold. Add a sentence about how readers should interpret overlapping hazards.

6. Flow accumulation appears as an afterthought. If you are going to integrate running water concentration into the risk, indicate where it enters (set of factors vs. modifiers after overlapping) and the weight/threshold. If illustrative only, state this explicitly.

7. The inventory of events versus time predictors is not clarified. Cite 2–3 decades of events and 1-hour extreme precipitation (180 mm, 2011). Add one sentence about temporal representativeness (modern DEM/land use predicts present-day sensitivity) and the implication that some “calmer” areas may not have been tested by extremes.

8. Generic conclusions require site-specific insight. Briefly explain why Polchevera/Bissagno and the western sources concentrate high sensitivity (topography, lithology, storm climatology) and relate this to property corridors (valley routes, slope crossings).

1.Done. Corrected and removed the contents, please see lines 518-526.

2.Done. Please see lines 526-529.

3.It has been added in the results section and here could look to be redundant.

4.Limitations in vulnerability are already cited at (original) lines 463-464;

5. The two hazards’sources are not overlapped, as stated in the (new) lines 475-480 in section 2.2.

6.It is not integrated in the method; sentence added. Please see lines 566-567.

7.Land-use in the area has not been significatively changed in the last 2-3 decades. the other request is stated at (old) lines 474-477.

8.Done. Please see lines 588-591.

Section 4 Conclusion.

1. Section numbering should be corrected to 5, and discussion to 4.

2. Reads as a summary, not a synthesis. Start with 2–3 key messages (what we learned about shallow slip susceptibility in Genoa and network risk), not how the model was built. Limit the summary of the method to one short sentence.

3. Relies on the calibration formulation, not on validation evidence. Replace “high percentage after calibration confirms reliability” with a one-sentence summary of spatial cross-validation (AUC/prediction rate) and keep the calibration statistics as an additional detail.

4. The 80/20 random split is called “rigorous” (overestimated). Acknowledge the 80/20 random split as a calibration only. Point out the spatial CV in Results/Discussion for generalizability. One sentence is sufficient in the Conclusion.

5. Numbers that are important to stakeholders are missing. Add key exposure numbers (e.g., % and km of high-risk gas/water networks, classes 4–5) extracted from Figure 11 so that asset managers can see the scale at a glance.

6. Contributions are implicit, not explicit. Convert the change in the set of factors into a clear list of contributions and link each to evidence in the results.

7. The climate trend result is unused. Briefly summarize Table 8 in one sentence (e.g., “Only FIO 1-h shows a significant upward trend, p=0.016; the others are not significant”) and state what this means for projected storms and early-stage screening.

8. The risk logic (H×V) is not re-anchored. Add one explanatory sentence on how sensitivity (hazard) and pipe diameter (vulnerability substitution) combine into 5 risk classes and note that there are “f” classes for rockfalls/wall collapses in urban areas, so that readers leave with a full definition of the risk.

9. The constraints are understated. Add a concise constraint line.

10. The applicability to utilities is weak. Turn classes 4–5 into operational guidelines, so that the Conclusion contains practice, not just a method.

11. Generalization is not covered by the scope.

→ Add one sentence on what is site-specific (Genoa lithology, relief, precipitation and climate) versus transferable (workflow, set of factors, overlapping risks) and what would be needed to transfer to other cities.

12. Missing cross-references to images/sections.

1.Done. Please see lines 517 and 604

2.The first phrase is the key message #1: the risk assessment. #2 is the modification of a previously developed methodology, adapting it and trying to reduce the bias in weighting the factors through the use of statistics. #3 is the reduction of the factors. #4 is the importance of man-made terraces presence that has been precisely identified with LidAR DTM analysis. Please see lines 605-608

3.Done. Please see lines 613

4.Done. Corrected and removed the contents, please see the first paragraph of section 5 Conclusion.

5.Done. Please see lines 629-631.

6.The link has been added in previous sections.

7.Table 8 belongs to the Discussion section and its discussion, together with climate trend is commented there. Please see Table 8.

8.Risk logic has been already clarified in the previous sections as requested. Please see lines 382-392. The wall collapse assessment has been added. Please see lines 635-638.

9.Limitations and constraints have been underlined in previous sections.

10.The possible interventions have been stated in the Discussion section. Please see lines 569-573.

In the conclusion section the importance of the method as a DSS has been underlined. Please see lines 609-625.

11.Done. Please see lines 609-625.

12.Done

comment

revision

The abstract is not consistent with what we would expect in a published journal paper. It suffers from several structural issues. Although the abstract identifies the problem (i.e., rain-induced shallow landslides and the link to infrastructure risk), it is too vague and generic. The opening statement is highly generic and does not add much value—it would be better to address, specifically, the research gap. The abstract is too long and dense, with poorly written sentences. Additionally, it is unclear why this is a novel or important study. Finally, there is no mention of key results of the analysis and thus, it reads like a summary rather than an incisive abstract.

Some modification to the abstract has been done according to the requests of another reviewer.

The introduction section has been modified, shortening it and declaring explicitly the aim of the paper, as requested even by another reviewer.

Overall, the paper is grammatically incoherent, with awkward phrasing and run-on sentences. Terminology is sometimes vague or inconsistent. For example, phrases such as “rapid runoff phenomena” and “modest landslide phenomena” are not helpful or technically descriptive.

The paper has been deeply revised according to the other reviewers’ requests and then rephrased and corrected.

Similarly, the manuscript is too long and poorly structured. Key and essential concepts are muddled by long, unconstructive, repetitive, and disorganized paragraphs—simply put, the paper does not read well. Figures and tables are not implicitly integrated into the argument: for example, see the text relating to Figures 9–12. Also, along these lines, the literature review is incomplete. The assertion that “no previous research” exists on risk assessment of urban underground services is somewhat overstated, and frankly, incorrect. Literature does exist especially with respect to urban resilience planning and critical infrastructure protection. A more thorough and complete literature review would address this.

The paper has been deeply revised according to the other reviewers’ requests and then rephrased and corrected. References have been added and the statement “no previous research” has been corrected.

The methodology is generally sound, but some improvements are needed. Overall, the process is insufficiently explained. For example: What expert panel was involved? How were subjective weights decided? Were consistency checks applied?  Additionally, the slope deposits factor is not sufficiently explained and the details given are not sufficient to enable reproducibility. Finally, there is no analysis of uncertainty in the model which is a major gap in this risk model. The model for combining hazard and vulnerability lacks justification. Pipe diameter is a crude proxy for vulnerability. The model for combining hazard and vulnerability lacks justification. Pipe diameter is a crude proxy for vulnerability.

The whole method has been clarified, adding the pairwise comparison matrix of the AHP method, the consistency ratio and the consensus indicator, together with further details that allow the replicability. Besides, ROC and AUC computation have been added for the validation procedure.

The vulnerability assessment through the pipes’ diameter limitation has been furtherly underlined.

The manuscript is overly long and poorly structured. 

comment

revision

The introduction section is overly fragmented, with too many paragraphs. The description of the innovation points is insufficient, failing to clearly highlight the original contributions of the study.

The Introduction has been shortened and re-structured.

Section 2.1, "Geographical and Geological Outline," is excessively lengthy and should be streamlined.

The section has been shortened and figures have been moved.

There is a lack of detail in the research methods. The core steps of the Analytical Hierarchy Process (AHP) include "constructing the hierarchical structure, expert pairwise comparisons, and consistency checks," but the paper only mentions the determination of factor weights through AHP. It is recommended to supplement the complete AHP methodology, especially the consistency check results, to demonstrate the rationality of the weight allocation.

Methodology has been addressed more deeply and the pairwise matrix has been included in the Appendix A. Consistency ratio and consensus indicator have been computed and added.

The results section lacks clear logical structure and should be divided into multiple subsections.

Results have been more clearly stated and integrated with ROC/AUC computation. Further, clarification for the wall collapse assessment have been added and clarified, recalling the relative clarification in section 2.2.

The model calibration metrics are too simplistic, relying only on percentages to indicate reliability. It is recommended to include ROC curve analysis and calculate the AUC value to enhance the persuasiveness of the model validation.

ROC and AUC have been added.

The conclusion section is overly lengthy and fails to precisely summarize the core contributions of the paper. Additionally, references should not be cited in the conclusion section.

Refences have been cut off, but Conclusion length has not been reduced integrating requests from other reviewers.

Line 117, Figure 2, and Line 406, Figure 9, do not have labels for A and B.

Labels have been added.

Line 319, Table 3, shows the weight distribution of landslide-triggering factors, where the weights for "Slope Gradient," "Land Use," and "Slope Deposits" significantly differ from other factors such as lithology and slope aspect. The reasons for this discrepancy should be explained.

There were mistakes in the values that have been corrected.

There is a repeated numbering issue for the figures, with two instances of Figure 8.

The correction has been done.

The tables are not formatted using a three-line table style.

Done

comment

revision

The abstract would benefit from including one or two quantitative results (e.g., “12% of water pipelines fall into high-risk zones”) to highlight the practical implications of the study.

The gap in literature regarding urban underground utilities could be emphasized more strongly. Please contrast your work with existing pipeline studies in rural/mountain contexts to highlight novelty.

Data in the abstract have been added and modification in the Introduction has been done.

The description of how weights of AHP and seven causal factors were derived (Tables 2 & 3) needs more transparency. Please clarify how consistency of AHP judgments was tested (e.g., CI/CR ratios).

Methodology has been addressed more deeply and the pairwise matrix has been included in the Appendix A. Consistency ratio and consensus indicator have been computed and added.

Figures 7–9 require more detailed descriptions in the text. For example, Figure 7 (causal factors) should explicitly explain how each factor influences susceptibility. In Figure 9 (risk map), consider highlighting specific neighborhoods or infrastructure stretches at highest risk.

A more detailed description has been added. The research focus on water and gas shallow landslides risk, then no other infrastructure has been cited.

The validation process could be improved by reporting additional statistical indices (e.g., ROC curves, AUC values).

ROC and AUC have been added.

Please add at least one case or simulation (e.g., slope bio-engineering, terracing reinforcement) to illustrate applicability of Nature-Based Solutions.

Done

In Tables 7 & 8, only one station shows Mann-Kendall test statistical significance. Please discuss the limitations of spatial representativeness (e.g., why western Genoa is more sensitive) and how this might bias risk mapping.

Done

To enhance reproducibility, it is recommended to make the landslide inventory dataset (485 events), pipeline network layers, and source codes available.

Landslides dataset is public, while the networks layers may not be distributed for safety reasons.

Consider integrating ensemble machine learning and deep learning methods (e.g., Random Forest, XGBoost, or CNN) alongside AHP to compare predictive performance. Relevant studies include DOI:10.1007/s12665-022-10723-z and DOI:10.1080/19475705.2024.2383309

It could be done in further research.