Dynamic landslide susceptibility mapping over recent three decades to uncover variations in landslide causes in subtropical urban mountainous areas

TL;DR

This study tackles the problem of dynamic landslide susceptibility in a subtropical urban mountainous setting by constructing yearly LS tasks (1992–2019) and applying a hybrid framework that combines RF-based prediction for data-rich years with meta-learning for few-shot adaptation in data-scarce years. It couples SHAP-based interpretation to identify evolving landslide inducing factors (LIFs) and employs MT-InSAR to enhance and validate LSMs against ground deformation signals. The results show slope and annual extreme rainfall days (AERD) as dominant drivers, with AERD’s importance modulated by climate change and the Landslip Prevention and Mitigation Programme (LPMitP). The approach achieves higher accuracy and AUROC than traditional methods and demonstrates a practical pathway to track temporal variation in landslide causes, informing targeted risk mitigation in rapidly changing environments. Overall, the dynamic LSA framework offers a robust, interpretable, and deformation-validated tool for proactive landslide management in urban mountainous regions.

Abstract

Landslide susceptibility assessment (LSA) is of paramount importance in mitigating landslide risks. Recently, there has been a surge in the utilization of data-driven methods for predicting landslide susceptibility due to the growing availability of aerial and satellite data. Nonetheless, the rapid oscillations within the landslide-inducing environment (LIE), primarily due to significant changes in external triggers such as rainfall, pose difficulties for contemporary data-driven LSA methodologies to accommodate LIEs over diverse timespans. This study presents dynamic landslide susceptibility mapping that simply employs multiple predictive models for annual LSA. In practice, this will inevitably encounter small sample problems due to the limited number of landslide samples in certain years. Another concern arises owing to the majority of the existing LSA approaches train black-box models to fit distinct datasets, yet often failing in generalization and providing comprehensive explanations concerning the interactions between input features and predictions. Accordingly, we proposed to meta-learn representations with fast adaptation ability using a few samples and gradient updates; and apply SHAP for each model interpretation and landslide feature permutation. Additionally, we applied MT-InSAR for LSA result enhancement and validation. The chosen study area is Lantau Island, Hong Kong, where we conducted a comprehensive dynamic LSA spanning from 1992 to 2019. The model interpretation results demonstrate that the primary factors responsible for triggering landslides in Lantau Island are terrain slope and extreme rainfall. The results also indicate that the variation in landslide causes can be primarily attributed to extreme rainfall events, which result from global climate change, and the implementation of the Landslip Prevention and Mitigation Programme (LPMitP) by the Hong Kong government.

Figures (18)

• Figure 1: Location of study area and distribution of landslides from 1992 to 2019.
• Figure 2: Landslide thematic maps. (a)Land use, (b) lithology, (c) DEM, (d) slope, (e) curvature, (f) aspect, (g) NDVI (The value is amplified 10,000 times), (h) SPI, (i) TWI, (j) catchment & drainage, (k) road lines & faults, (l) precipitation.
• Figure 3: (continued)
• Figure 4: The overflow of the proposed method.
• Figure 5: The proposed meta-learning strategy. $\theta_i'$ is the updated network parameters of $\mathcal{T}_i$ in the inner loop computation. $\theta'$ is the network parameters of the intermediate model in the outer loop computation. $\mathcal{F}$ is the function for meta-objective.