1. 基于不同机器学习的震后滑坡易发性建模研究.
- Author
-
周天游, 刘 畅, 薛 鹏, 杨 豹, and 舒建冬
- Abstract
The modeling process and uncertainty of different machine learning models to predict landslide susceptibility are different. High-precision assessment of post-seismic landslide susceptibility is the key to effective disaster prevention and mitigation. This paper established the post-seismic landslide database of the Jiuzhaigou earthquake, and 12 causative factors were finally selected for landslide susceptibility modeling by eliminating two factors from 14 initial landslide factors through factor covariance diagnosis. Then selects logical regression(LR), support vector machine(SVM), random forest(RF), and artificial neural network(ANN) machine learning models and 10-fold cross-validation data sampling methods. Finally, the receiver operating characteristic curve and the distribution characteristics of the susceptibility index(mean and standard deviation) are used to discuss the modeling of landslide susceptibility based on machine learning and its uncertainty. The results show that the high susceptibility areas of landslides after the earthquake are mainly developed and distributed along the epicenter and valleys. The AUC values of the four machine learning models all exceed 0.87, which has achieved good applicability for predicting potential post-seismic landslides in the study area. Among them, the mean value and standard deviation of the susceptibility index of SVM are small, indicating that the model has good recognition ability for landslides, and it has the highest prediction accuracy(CVmean=0.925), whereas the other models are ANN(0.920), LR(0.894) and RF(0.877). The results of this paper are of great significance for popularizing and promoting the development of machine-learning models in other regions. [ABSTRACT FROM AUTHOR]
- Published
- 2023
- Full Text
- View/download PDF