A platform for research: civil engineering, architecture and urbanism
A platform for research: civil engineering, architecture and urbanism
Effectiveness of Newmark-based sampling strategy for coseismic landslide susceptibility mapping using deep learning, support vector machine, and logistic regression
Abstract Non-landslide samples play a crucial role in landslide susceptibility mapping (LSM), although unsuitable sampling methods may degrade the performance of the prediction model. The primary objectives of this study are to explore the influence of the traditional buffer-controlled sampling method on model performance and to propose a Newmark-based sampling approach for coseismic landslides. The Jiuzhaigou meizoseismal region of China is selected as the region of study. Six sample datasets are constructed for three machine learning models, namely a deep neural network (DNN), logistic regression (LR), and a support vector machine (SVM). The samples cover two scenarios: scenario-BZ, a set of samples created using different buffer distances, and scenario-LD, a set of non-landslide samples created by the Newmark-based method. Intriguingly, the results indicate that the area under the curve (AUC) is positively correlated with the buffer distance in scenario-BZ (DNN: 0.894–0.979, SVM: 0.894–0.981, LR: 0.797–0.889), but gentle valleys in the buffer zone are assigned over-conservative susceptibility values while the probability of landslides in steep mountains outside the buffer zone is underestimated. In contrast, all models assign more reasonable susceptibility values in scenario-LD (AUC values of 0.969, 0.969, and 0.931 for the DNN, LR, and SVM models, respectively). These results suggest that the landslide susceptibility obtained by the traditional buffer-controlled method may be inaccurate, despite the prediction model achieving excellent performance. The proposed approach can therefore provide insights into coseismic landslide susceptibility in other earthquake regions.
Effectiveness of Newmark-based sampling strategy for coseismic landslide susceptibility mapping using deep learning, support vector machine, and logistic regression
Abstract Non-landslide samples play a crucial role in landslide susceptibility mapping (LSM), although unsuitable sampling methods may degrade the performance of the prediction model. The primary objectives of this study are to explore the influence of the traditional buffer-controlled sampling method on model performance and to propose a Newmark-based sampling approach for coseismic landslides. The Jiuzhaigou meizoseismal region of China is selected as the region of study. Six sample datasets are constructed for three machine learning models, namely a deep neural network (DNN), logistic regression (LR), and a support vector machine (SVM). The samples cover two scenarios: scenario-BZ, a set of samples created using different buffer distances, and scenario-LD, a set of non-landslide samples created by the Newmark-based method. Intriguingly, the results indicate that the area under the curve (AUC) is positively correlated with the buffer distance in scenario-BZ (DNN: 0.894–0.979, SVM: 0.894–0.981, LR: 0.797–0.889), but gentle valleys in the buffer zone are assigned over-conservative susceptibility values while the probability of landslides in steep mountains outside the buffer zone is underestimated. In contrast, all models assign more reasonable susceptibility values in scenario-LD (AUC values of 0.969, 0.969, and 0.931 for the DNN, LR, and SVM models, respectively). These results suggest that the landslide susceptibility obtained by the traditional buffer-controlled method may be inaccurate, despite the prediction model achieving excellent performance. The proposed approach can therefore provide insights into coseismic landslide susceptibility in other earthquake regions.
Effectiveness of Newmark-based sampling strategy for coseismic landslide susceptibility mapping using deep learning, support vector machine, and logistic regression
Xi, Chuanjie (author) / Han, Mei (author) / Hu, Xiewen (author) / Liu, Bo (author) / He, Kun (author) / Luo, Gang (author) / Cao, Xichao (author)
2022
Article (Journal)
Electronic Resource
English
BKL:
56.00$jBauwesen: Allgemeines
/
38.58
Geomechanik
/
38.58$jGeomechanik
/
56.20
Ingenieurgeologie, Bodenmechanik
/
56.00
Bauwesen: Allgemeines
/
56.20$jIngenieurgeologie$jBodenmechanik
RVK:
ELIB18
| British Library Online Contents | 2014
Presenting logistic regression-based landslide susceptibility results
| British Library Online Contents | 2018
Regression models for estimating coseismic landslide displacement
| Elsevier | 2007