| 基于机器学习的滑坡易发性预测建模及其主控因子识别 |
| |
| 引用本文： | 黄发明,胡松雁,闫学涯,李明,王俊宇,李文彬,郭子正,范文彦. 基于机器学习的滑坡易发性预测建模及其主控因子识别[J]. 地质科技通报, 2022, 41(2): 79-90. DOI: 10.19509/j.cnki.dzkq.2021.0087 |
| |
| 作者姓名： | 黄发明 胡松雁 闫学涯 李明 王俊宇 李文彬 郭子正 范文彦 |
| |
| 基金项目： | 国家自然科学基金项目41807285国家重点研发计划项目2019YFC0605001中国博士后基金项目2019M652287中国博士后基金项目2020T130274江西省博士后基金项目2019KY08 |
| |
| 摘 要： | 不同机器学习预测滑坡易发性的建模过程及其不确定性有所差异, 另外如何有效识别滑坡易发性的主控因子意义重大。针对上述问题, 以支持向量机(support vector machine, 简称SVM)和随机森林(random forest, 简称RF)为例探讨了基于机器学习的滑坡易发性预测及其不确定性, 创新地提出了"权重均值法"来综合计算出更准确的滑坡主控因子。首先获取陕西省延长县滑坡编录和10类基础环境因子, 将因子频率比值作为SVM和RF的输入变量; 再将滑坡与随机选择的非滑坡样本划分为训练集和测试集, 用训练好的机器学习预测出滑坡易发性并制图; 最后用受试者工作曲线、均值和标准差等来评估建模不确定性, 并计算滑坡主控因子。结果表明: ①机器学习能有效预测出区域滑坡易发性, RF预测的滑坡易发性精度高于SVM, 而其不确定性低于SVM, 但两者的易发性分布规律整体相似; ②权重均值法计算出延长县滑坡主控因子依次是坡度、高程和岩性。实例分析和文献综述显示RF模型相较于其他机器学习模型属于可靠性较高的易发性模型。
|
| 关 键 词： | 滑坡易发性预测 不确定性分析 主控因子识别 支持向量机 随机森林 |
| 收稿时间： | 2021-04-08 |
Landslide susceptibility prediction and identification of its main environmental factors based on machine learning models |
| |
| Abstract: | The modelling processes and uncertainties of various machine learning models for landslide susceptibility prediction (LSP) are different, and effectively identifying the main conditioning factors of landslide susceptibility is of great significance. Aiming at these problems, this study aims to discuss the LSP processes and the uncertainties of landslide susceptibility based on machine learning models, namely, support vector machine (SVM) and random forest (RF), and then to innovatively propose the "weighted mean method" for calculating more accurate landslide main control factors. First, the landslide inventories and 10 basic environmental factors of Yanchang County in Shaanxi Province are obtained, and the frequency ratios (FRs) of the environmental factors are taken as the input variables of the SVM and RF models.Then, the landslide and randomly selected nonlandslide samples are divided into model training and testing datasets. Furthermore, the trained RF and SVM models are used to predict the landslide susceptibility and draw the landslide susceptibility prediction (LSP) map.Finally, the uncertainties of LSP modelling are evaluated by the receiver operating characteristic (ROC) curve, mean value and standard deviation, and the main landslide control factors are calculated.The results show that ① Machine learning models can effectively predict the susceptibility of regional landslides. The accuracy of RF in LSP is higher, and its uncertainties are lower than those of SVM. As a whole, the landslide susceptibility distribution rules of the two models are similar.②The main control factors of landslide susceptibility in Yanchang County calculated by the weighted mean method are slope, elevation and lithology.③Case studies and literature reviews show that the RF model is a more reliable susceptibility model than other types of machine learning models. |
| |
| Keywords: | |
|
| 点击此处可从《地质科技通报》浏览原始摘要信息 |
|
点击此处可从《地质科技通报》下载免费的PDF全文 |
|
