- Title
- Uncertainties of landslide susceptibility prediction due to different spatial resolutions and different proportions of training and testing datasets
- Creator
- Huang, Faming; Chen, Jiawu; Tang, Zhipeng; Fan, Xuanmei; Huang, Jinsong; Zhou, Chuangbing; Chang, Zhilu
- Relation
- Chinese Journal of Rock Mechanics and Engineering Vol. 40, Issue 6, p. 1155-1169
- Publisher Link
- http://dx.doi.org/10.13722/j.cnki.jrme.2020.1119
- Publisher
- Zhongguo Kexueyuan Wuhan Yantu Lixue Yanjiuso
- Resource Type
- journal article
- Date
- 2021
- Description
- It is of great significance to explore the influence of factors such as spatial resolution and the proportion of model training and test sets on the uncertainty of landslide susceptibility prediction. Taking the landslide in Ningdu County, Jiangxi Province as an example, the frequency ratio of each environmental factor at different spatial resolutions (15, 30, 60, 90 and 120 m) was calculated first; The ratios of /2, 7/3, 6/4 and 5/5 are divided into different model training and testing data sets, and the input and output variables of the model under 25 combined working conditions are obtained; then they are imported into the support vector machine. The landslide susceptibility prediction is carried out in the support vector machine (SVM) and random forest (RF) models, and finally the prediction accuracy and susceptibility index characteristics are used to analyze the uncertainty of susceptibility modeling under various working conditions. . The results show that: (1) 15m resolution, 9:1 training and test set ratio, and RF model have the highest susceptibility prediction accuracy, and the important environmental factors under each working condition are elevation, slope and terrain relief, etc.; (2) As the resolution decreases or the proportion of training and test sets decreases, the prediction accuracy of the SVM and RF models gradually decreases, the mean value of the susceptibility index increases and the standard deviation decreases; (3) In all working conditions As the resolution and the proportion of training and test sets decrease, the susceptibility prediction accuracy gradually decreases and its uncertainty increases; (4) The RF prediction accuracy is better than SVM under each working condition, and the impact of resolution on the RF model It is significantly larger than the proportion of training and testing sets, and the influence of the two factors in SVM is not much different.
- Subject
- slope engineering; landslide susceptibility; uncertainty analysis; machine learning; environmental factors; spatial resolution
- Identifier
- http://hdl.handle.net/1959.13/1450119
- Identifier
- uon:43831
- Identifier
- ISSN:1000-6915
- Language
- eng
- Reviewed
- Hits: 1537
- Visitors: 1537
- Downloads: 0