the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 09 Sep 2025
The Dual-Edged Role of Vegetation in Evaluating Landslide Susceptibility: Evidence from Watershed-Scale and Site-Specific Analyses
Abstract. Vegetation is widely recognized for its beneficial role in landslide mitigation. However, shallow landslides frequently occur even in densely vegetated regions, suggesting that the influence of vegetation on gravity-driven erosion hazards remains incompletely understood. This study investigates the interactive effects of vegetation and key environmental factors—including rainfall, lithology, wind speed, and slope gradient—on landslide susceptibility in an area with substantial vegetation cover (≥65.5 %). At the watershed scale, we employed structural equation modeling and geographic detectors to assess the primary drivers of landslide susceptibility under high vegetation conditions. At the point scale, we calculated the stability coefficient of a representative landslide, accounting for both vegetation self-weight and artificial waste sediment. Our findings reveal that the combination of vegetation, rainfall, and wind speed significantly increases landslide susceptibility, as evidenced by a 21.3 % rise in high and very high susceptibility zones and a 42.7 % reduction in low and very low susceptibility zones. Interactions among multiple factors exerted a stronger influence than individual factors, with the most pronounced interaction observed between slope gradient and rainfall (Geodetector q = 0.81), followed by rainfall and lithology (q = 0.79). Under saturated conditions, the stabilizing effect of root systems was outweighed by the self-weight of tree vegetation, leading to a marked decrease in slope stability compared to scenarios without additional loading. These results offer new insights into the complex role of vegetation in landslide control and highlight the importance of considering interactive environmental effects at multiple spatial scales.
- Preprint
(2255 KB) - Metadata XML
-
Supplement
(689 KB) - BibTeX
- EndNote
Status: open (until 21 Oct 2025)
- CC1: 'Comment on egusphere-2025-3004', Jun Zhang, 15 Sep 2025 reply
-
RC1: 'Comment on egusphere-2025-3004', Anonymous Referee #1, 06 Oct 2025
reply
Review of the manuscript “The Dual-Edged Role of Vegetation in Evaluating Landslide Susceptibility: Evidence from Watershed-Scale and Site-Specific Analyses”
I would like to thank the authors for submitting this manuscript, which addresses a topic of undeniable scientific and practical relevance. Understanding the role of vegetation in slope stability and landslide susceptibility is an important and timely research question.
However, in its current form, the manuscript presents several major issues that, in my opinion, make it unsuitable for publication without substantial and fundamental revisions. The work requires significant restructuring and methodological clarification before it can be considered for acceptance.
General structure and integration between scales
A central weakness of the paper lies in the lack of clarity regarding how the local-scale analyses are integrated with the regional-scale susceptibility assessment.
As currently presented, the local-scale analyses appear disconnected and scientifically irrelevant, adding little to the main argument. If the authors cannot clearly demonstrate the conceptual and methodological linkage between the two scales, I would recommend removing the local-scale component entirely.
Methodological adequacy – Use of AHP
The use of the Analytic Hierarchy Process (AHP) as the core method for landslide susceptibility mapping raises serious concerns.
AHP is highly subjective and largely outdated, having been replaced in the literature by more objective and data-driven approaches (e.g., statistical models, machine learning algorithms, ensemble frameworks).
The authors do not provide any convincing justification for this methodological choice. In its current form, this decision undermines the robustness and reproducibility of the results.
Landslide inventory and model validation
The description of the landslide inventory is inconsistent and insufficient. The manuscript states that the inventory was downloaded from an online repository and then integrated with 227 manually identified landslides; yet later it is mentioned that the total number of landslides is 227. This discrepancy must be clarified.
The authors should:
(i) provide a detailed and transparent description of the inventory, including sources, validation, and completeness; and
(ii) clearly explain the selection criteria for non-landslide points, as this strongly affects the ROC/AUC results.
Without this information, the reported validation accuracy appears potentially overestimated and unreliable.
Reference to established best practices
The authors should explicitly compare their approach with the guidelines proposed by Reichenbach et al. (2018), who outlined key criteria for producing reliable landslide susceptibility maps.
Currently, the manuscript neither demonstrates adherence to these well-established standards nor engages with them critically.
Analysis of variables and use of geodetector
The attempt to explore how different variables (and their combinations) influence susceptibility is potentially interesting.
However, the application of the GeoDetector method appears methodologically flawed: the authors use the susceptibility map (derived from AHP) as the dependent variable, rather than the landslide inventory itself.
Since susceptibility is already a model — and a highly subjective one — this approach introduces a strong bias, making any subsequent inferences about controlling factors or vegetation effects questionable.
Such analyses should be based on observed landslide occurrences, not on the output of another model.
Structure and clarity of the manuscript
The manuscript is redundant and lacks structural clarity. I recommend:
(i) removing repetitive sentences;
(ii) improving logical flow and conciseness throughout the text.
In particular, several statements about the “ambiguous role” of vegetation are speculative and not sufficiently substantiated by quantitative evidence.
Vegetation
The paper focuses on vegetation but does not include an accurate description of the types of vegetation present in the study area. This is a major issue in my opinion.
Specific (but not minor) comments
l29: “landslide frequently occur” → How frequently? More than in other soil-cover conditions? I suggest replacing frequently with may.
l59: What is meant by “good vegetation”?
l60: change to: “mitigation also depends on…”
l73: What do you mean by “susceptibility to external disturbances”? I believe susceptibility refers specifically to landslides in this paper.
l75: You probably mean apparent cohesion, since cohesion itself is not reduced by rainfall.
l76: “downslope forces generated by gravitational water distribution” — please clarify.
l89: What are the “geoscience factor weights”? In any case, multiple techniques for landslide susceptibility exist, and the references at the end of the sentence are insufficient.
l93: “few studies” → please cite them.
l110: “frequent” again — misleading: it suggests landslides mostly occur in vegetated areas.
l113: GeoDetector and structural equation modeling require references.
l127: Forest is 65%. Which type of forest? And what about the remaining 35%?
l127: “terrain slopes from southwest to northeast” — what do you mean? This sounds strange.
l137: It is unclear why you introduce this major landslide and how it connects to the large-scale analysis. This point is crucial.
l174: SoilGrids (Hengl et al., 2017) describes soils, not lithology. How did you derive lithology from those data?
l178: Why “maximum”? Please clarify.
l189: “Landslide hazard point data were obtained from the GeoCloud platform...” — this sentence is poor. The GeoCloud data are never mentioned again. No description is provided for the method or imagery used to prepare the inventory. Were these recent landslides? Is it a geomorphological inventory? What imagery and what dates were used? Without this, the inventory’s quality cannot be assessed — a major issue.
l196: “Landslide susceptibility analysis was conducted by overlaying landslide sites with…” → You only overlaid the landslide points?
l238: “An equal number of non-landslide points were created” — how?
l253: The GeoDetector method should be applied with a large number of landslides; are 227 sufficient?
l258: Why did you use landslide susceptibility as the dependent variable? It is already a model. Why not use the landslide inventory?
l259: what’s “geological hazard risk” — a scientific paper should use terms accurately.
l324: “84 landslides covering 53.77% of the total area” — you said landslides are points: how can they cover an area? In pixels?
l336: This section pops up quite suddenly and without any clear justification, in my opinion.
l400: “faults occur where the structural stability of the slopes is poor” — this is scientifically incorrect; the causal direction is reversed.
l462: “public factors” — what do you mean by this?
l500: The statements throughout this section are not supported by data analysis, in my opinion.
Overall Assessment
The study addresses a relevant topic, but the current manuscript suffers from serious methodological, conceptual, and structural weaknesses.
I recommend a major revision involving:
-a clearer definition of objectives and scale integration,
-a more rigorous and modern susceptibility modeling approach,
-a detailed and transparent description of the landslide inventory and validation procedures, and
-a substantial reorganization of the manuscript to improve clarity and coherence.
Only after these fundamental issues are properly addressed can the paper be reconsidered for publication.
Citation: https://doi.org/10.5194/egusphere-2025-3004-RC1
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 1,411 | 20 | 7 | 1,438 | 14 | 13 | 10 |
- HTML: 1,411
- PDF: 20
- XML: 7
- Total: 1,438
- Supplement: 14
- BibTeX: 13
- EndNote: 10
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
This work explores the "double-edged sword" role of vegetation in the landslide susceptibility, conducting multi-scale research at both the watershed scale and typical points, and is of certain innovation and academic value. Thus, to some extent, it is very meaningful. But some of content should be improved, the suggestions are show below. So, I give it moderate revision.
Introduction
The introduction provides sufficient background information regarding landslide hazards to introduce readers to the study. The gap in the literature, that more research is needed in regions with high vegetation, and the limitations of prior studies has been clearly explained. The motivation for the study and its objectives are stated clearly. But, also need some revisions just as follows:
1 The following studies cited are older than 10-15 years: Regmi et al., 2010; Yilmaz, 2009, Fell et al., 2008; Hürlimann et al., 2008; Sezer et al., 2010; Hu & Bentler, 1999; Goren et al., 2010; Manzella et al., 2008. It is preferable to cite recent articles in a manuscript, and only in exceptional circumstances should references going back more than 10-15 years be cited. It is preferable to cite recent articles because older references may be irrelevant given more recent advancements in the field of study. Exceptions to this rule should be reserved for seminal works directly relevant to the topic of research. Citing recent articles also helps journal editors see that there is a potential audience for your topic of research;
2 There is some repetition in the introduction, which can be frustrating for your readers;
3 The last paragraph is so long, please split into two part.
4 Line 67, “reposted” should be “reported”.
Methods and materials
The methods present all the necessary information to be reproduced by other researchers, and the reasons for choosing specific methods have been included, where relevant. The methods are presented in a logical order. The procedure for statistical analyses of the data collected has been outlined under Methods. But followings should be paid more attention,
1 Figures 1,2 and 4 were referenced in the text after the appearance of the figure. Please amend;
2 Clarify why the specific study area (Jinkouhe District) was chosen—how do its characteristics contribute to the relevance of this research?
3 Provide more detailed descriptions of the modeling processes, especially SEM, including assumptions made during factor selection.
4 please redraw the Fig. 6.
Results
The order of the results corresponds to that of the methods. The tables and figures present the data clearly and are referenced in the manuscript. Table and figure details corresponded with those in the text.
1 Lines 330-332, It would be useful to give the areas (north, southwest, etc.);
2 Line 335, please left a space between Fig.7.landslide…………
3 Lines 352-353, the title 3.3 can be revised as “Slope stability calculation considering artificial waste sediment and vegetation self-weight”.
Discussions
This section provides a thorough discussion of the results-based part and the differences and improvements compared to previous studies, and also offers an outlook on future work.
However, it is necessary to more clearly point out the unique aspects of this research (for example, "integrating macroscopic susceptibility with microscopic mechanics"). The outlook for future research can be more specific, for example: how to utilize interpretable machine learning and multi-source data fusion, rather than just making general statements. More specific as below:
1 Line 364, the title should be changed to “Analysis of landslide driving factors and their interaction pathways”. This part, the authors mainly emphasize the factors and the interactions.
2 Line 453, the title should be precise. This part mainly compared the landslide susceptibility under different factors combination, so maybe this title will be more suitable: “differences and explanations of landslide susceptibility results under different factor combinations”.