Insights into initiation of typhoon-induced deep-seated landslides in Southeast coastal China

Zhuang, Yu; Xing, Aiguo; Sun, Qiang; Jiang, Yuehua

doi:10.1007/s11069-023-06138-z

Cited by 1 publication

(1 citation statement)

References 73 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Abancó et al (Abancó et al, 2021) investigated the roles of topography, typhoon rainfall, and soil moisture during typhoon-induced landslides, while Dou et al (Dou et al, 2023a) summarized the mechanisms and early warning research status of vegetation-covered landslides triggered by typhoon rains. Zhuang et al (Zhuang et al, 2023) performed numerical simulations of transient changes in slopes during the entire Typhoon Lekima period, and Wu (Wu, 2023) developed a spatiotemporal cube model to explore the long-term evolution of landslides caused by Typhoon Morakot in 2009, determining the spatiotemporal distribution of landslide recovery post-extreme rainfall events.…”

Section: Introductionmentioning

confidence: 99%

Machine learning-based hazard assessment of landslides within the rainband of typhoon Lekima

Feng,

DING,

et al. 2024

Preprint

View full text Add to dashboard Cite

Amid escalating global climate change, the frequency of extreme weather events, such as powerful typhoons, has been on the rise. Notably, Typhoon Lekima in 2019 severely impacted Zhejiang Province, triggering a series of severe landslides. This study employs six different machine learning approaches to propose a landslide hazard assessment zoning model that takes into consideration the typhoon track and rainbands. For this purpose, Zhejiang Province was selected as a case study area based on the path and internal rainbands of Typhoon Lekima. The cumulative rainfall 48 hours before and after the typhoon's landfall was recorded, and various factors such as topography, geology, vegetation, and human engineering activities were integrated. Landslide hazard was evaluated using six machine learning techniques: Random Forests (RF), Support Vector Machines (SVM), K-Nearest Neighbors (KNN), Logistic Regression (LR), Artificial Neural Networks (ANN), and Gradient Boosting Decision Tree (GBDT). The evaluation results indicate that all models effectively identified landslide-prone areas, with exceptional performance from the GBDT and RF models, achieving AUC values of 0.95 and 0.94, respectively. Furthermore, the distribution of landslide hazards was closely linked to typhoon intensity and trajectory. Landslide risks were high in areas experiencing category 12 typhoons, centrally symmetric along the typhoon path and decreasing towards the borders of the inner rainbands; areas affected by category 10–11 typhoons exhibited asymmetric hazards concentrated on the coastal side, whereas regions under category 9 typhoons showed lower risk levels. Lastly, the local interpretability of models through the LIME (Local Interpretable Model-agnostic Explanations) algorithm enhanced understanding of model predictive behaviors, offering increased accuracy and interpretability in landslide hazard assessment compared to traditional methods. This research provides a scientific basis for developing targeted landslide disaster prevention and control strategies and improving emergency management of such disasters.

show abstract

Section: Introductionmentioning

confidence: 99%