当前位置: X-MOL 学术Catena › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
Application of alternating decision tree with AdaBoost and bagging ensembles for landslide susceptibility mapping
Catena ( IF 5.4 ) Pub Date : 2019-12-11 , DOI: 10.1016/j.catena.2019.104396
Yanli Wu , Yutian Ke , Zhuo Chen , Shouyun Liang , Hongliang Zhao , Haoyuan Hong

Landslides are a common type of natural disaster that brings great threats to the human lives and economic development around the world, especially in the Chinese Loess Plateau. Longxian County (Shaanxi Province, China), a landslide-prone area located in the southwest part of the Loess Plateau, was selected as the study area. The main purpose of this paper is to map landslide susceptibility using Alternating decision tree (ADTree) as well as GIS-based new ensemble techniques involving ADTree with bootstrap aggregation (Bagging) and ADTree with adaptive boosting (AdaBoost). Initially, a landslide inventory map was prepared with 171 determined historical landslides events in the study area, 120 landslides (70%) were randomly selected for training dataset and the remaining 51 landslides (30%) were used for validation dataset. Subsequently, eleven landslide conditioning factors were considered in the landslide susceptibility mapping. Then, an optimization operation on selection of landslide conditioning factors was performed using correlation attribute evaluation method and Spearman’s rank correlation coefficient. Afterwards, landslide susceptibility maps were generated with the three models. Finally, receiver operating characteristic (ROC) curve, area under the ROC curve (AUC) and statistical measures were applied to evaluate and validate the performance of the models. The results show success rates of the ADTree model, the ADTree with Bagging (ADTree-Bagging) model and the ADTree with AdaBoost (ADTree-AdaBoost) model were 0.872, 0.917, and 0.984, respectively, while prediction rates of the three models were 0.696, 0.752 and 0.787, respectively. In sum, the two ensemble models proposed prohibited better performance than the ADTree model did, and the ADTree-AdaBoost model was selected as the best model in the study. Hence, ensemble techniques can provide new and promising methods for spatial prediction and zonation of landslide susceptibility.

更新日期:2019-12-11
down
wechat
bug