the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 08 Sep 2025
Projecting changes in rainfall-induced landslide susceptibility across China under climate change
Abstract. Landslides pose a significant threat to human lives and property. Evaluating dynamic changes in landslide susceptibility under climate change can provide decision-making support for future disaster prevention. Using historical landslide inventories (2008–2023) and a random forest algorithm, this study develops an annual-scale landslide susceptibility model to assess spatiotemporal patterns of landslide susceptibility across China under different simulated scenarios. The results show that model achieves excellent performance (AUC = 0.97), with annual precipitation being the most influential factor (26 % contribution). Compared to the baseline (1950–2014), China's landslide susceptibility is projected to increase significantly under future climate conditions. By the late 21st century (2076–2100), the national mean annual precipitation is expected to rise by 59–111 mm, corresponding to a 4.3–10.6 % expansion in median to very high susceptibility zones across SSP scenarios. Spatially, the most significant susceptibility increases are anticipated in the Northwest Loess Plateau region (Loess) near the Taihang Mountains and the northern part of the Southwest Karst Mountain region (SW), where SSP5-8.5 amplify risks toward the century’s end. These findings underscore the necessity of proactive risk management in these identified hotspots to mitigate escalating landslide threats.
- Preprint
(1720 KB) - Metadata XML
-
Supplement
(1355 KB) - BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'Comment on egusphere-2025-3834', Francisco Dourado, 20 Oct 2025
-
RC2: 'Comment on egusphere-2025-3834', Anonymous Referee #2, 25 Nov 2025
This paper presents a study on the landslide susceptibility evolution influenced by future climate, in China, at the national scale. This subject is of great interest, and the methodology is clearly defined and described. The plan of the paper permits to have an easy lecture, and the paper is well written.
However, several points must be deeply described and considered :
1) A lot of information is missing, necessary for the understanding and for defining the limits of the approach, especially the limits of the resolution of the different data.
In particular, the influencing factors needs to be more detailed, as well as their maps (to put in supplement), their categories, description, accuracy, how they are obtained...
Landslide inventory must be more detailed and described; some statistical analyses could be provided (typology, size, ..) ; the triggering conditions have to be discussed: as I understand, they are not all triggered by rainfall, but also by earthquake. From a methodological point of view, it is necessary to consider only the ones triggered by rainfall. Moreover, all landslides have been considered here (landslide, collapse, debris flow…), but their predisposing parameters and triggering conditioning are not the same. A separate analysis might be more robust. Finally, it might be useful to provide some statistical analyses on the inventory, related to some predisposing factors (for instance are the landslides localised close to road ? )
The choice of some factors, as well as the data used for characterising these factors, is not justified. For instance, the seismic hazard map might be not adapted to this study as it is focused on landslide triggered by rainfall, and not earthquake. So, I don’t have any explanations on the usefulness of this data, nor the mechanical link of this data to landslide occurrence.
2) In the introduction, the literature analysis related to landslide analysis under climate change is weak, with lack of references. This might be improved.
3) The use of data coming from climate change model is another point: some details must be provided on these models. Is the low resolution of these data adapted for the scientific question of this paper? Is mean annual precipitation adapted to explain the landslide susceptibility? Monthly precipitation might be a proxy of antecedent condition/ soil saturation, but I don’t understand why annual precipitation has an influence on landslide susceptibility.
4) I don’t understand why the influence of key factors are analysed on future susceptibility map, and not on current one. I suggest analysing them on historical landslide susceptibility, and to present this before analysing future precipitation and future susceptibility.
The importance of some parameters is surprising; indeed, the lithological map at 1:3 750 000 is a tricky issue because of the low resolution. That might explain why these data have low relative importance within predisposing factors in RF analysis. Indeed, it is questionable that lithology is not within the most important feature. This point has to be largely discussed; if no more accurate data exist, one solution might be to compare the susceptibility map to other ones at regional scale, or to compare lithology at national and regional scale and discuss the differences.
Moreover, it might also be surprising that geomorphological parameters (e.g. slope, curvature…) are not so important.
At the contrary, NDVI, Elevation and Mean annual precipitation are the most important factors ; this result is also quite surprising and must be discussed.
NDVI reflects the vegetation density, but the effect of vegetation on landslide stability is not easy as several features can explain it : as explained in the paper the vegetation areas may imply more infiltration compared to urban areas ; but it is also important to consider the land use with different typology of vegetation; for instance some trees permit to stabilise the soil due to reinforcement of the roots ; some species also capture the rainfall , leading to the reduction of infiltration within the soil. All these aspects might be analysed and discussed.
As said before, I am not convinced that mean annual precipitation is adapted to explain the landslide susceptibility, as it is not linked with intense rainfall, as suggested in the paper (line 321). At the least, monthly rainfall value before the occurrence of a landslide from the inventory could be a proxy of antecedent rainfall or saturation.
I am not convinced by the discussion concerning landslide susceptibility distribution related to the 475-year RP PGA data (between lines 327- 333) ; It is important to only consider landslide triggered by rainfall within the inventory.
The results also indicate that the elevation is also a key parameter; this result is explained in considering that this parameter is a proxy of lithology. It reinforces the necessity to have an accurate analysis of the lithology information itself.
As a conclusion, I suggest the authors to provide these major corrections before re-submitting the manuscript.
Citation: https://doi.org/10.5194/egusphere-2025-3834-RC2
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 1,737 | 182 | 18 | 1,937 | 82 | 29 | 27 |
- HTML: 1,737
- PDF: 182
- XML: 18
- Total: 1,937
- Supplement: 82
- BibTeX: 29
- EndNote: 27
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
The proposal of the manuscript is interesting; however, it presents several points that warrant consideration:
The rationale for utilizing such contrasting and extreme IPCC precipitation projection scenarios—specifically SSP1-2.6 (very low) versus SSP5-8.5 (very high)—is unclear.
Similarly, the justification for employing such long-term time horizons (e.g., 2051–2075; 2076–2100) is questionable. This period is significantly distant from our current reality, limiting the applicability of meaningful interventions, and is subject to substantial uncertainty.
Given China's vast territorial extent, analyzing the entire country practically precludes the use of a spatial resolution (level of detail) that is adequate for, and compatible with, the scale of most mass-movement events (i.e., landslides, which are often < 30 m) that typically occur globally.
The inclusion of certain 'Influencing Factors' (e.g., NDVI, Distance to River, and Distance to Road) seems inappropriate for an analysis focused on mass movements.
It is suggested that the author reflect upon these points and revise the manuscript accordingly before resubmission.