| 不同环境因子联接和预测模型的滑坡易发性建模不确定性 |
| |
| 引用本文： | 李文彬,范宣梅,黄发明,武雪玲,殷坤龙,常志璐.不同环境因子联接和预测模型的滑坡易发性建模不确定性[J].地球科学,2021,46(10):3777-3795. |
| |
| 作者姓名： | 李文彬 范宣梅 黄发明 武雪玲 殷坤龙 常志璐 |
| |
| 作者单位： | 1.南昌大学建筑工程学院, 江西南昌 330031 |
| |
| 基金项目： | 国家自然科学基金项目41807285国家自然科学基金项目41762020国家自然科学基金项目51879127国家自然科学基金项目51769014江西省自然科学基金项目20192BAB216034江西省自然科学基金项目20192ACB2102江西省自然科学基金项目20192ACB20020中国博士后面上基金项目2019M652287中国博士后面上基金项目2020T130274江西省博士后基金项目2019KY08 |
| |
| 摘 要： | 拟深入探讨滑坡与其环境因子间的非线性联接计算以及不同数据驱动模型等因素，对滑坡易发性预测建模不确定性的影响规律.以江西省瑞金市为例共获取370处滑坡和10种环境因子，通过概率统计（probability statistics，PS）、频率比（frequency ratio，FR）、信息量（information value，Ⅳ）、熵指数（index of entropy，IOE）和证据权（weight of evidence，WOE）等5种联接方法分别耦合逻辑回归（logistic regression，LR）、BP神经网络（BP neural networks，BPNN）、支持向量机（support vector machines，SVM）和随机森林（random forest，RF）模型共构建出20种耦合模型，同时构建无联接方法直接将原始数据作为输入变量的4种单独LR、BPNN、SVM和RF模型，预测出总计24种工况下的滑坡易发性；最后分别使用ROC曲线、均值、标准差和差异显著性等指标分析上述24种工况下易发性结果的不确定性.结果表明:（1）基于WOE的耦合模型预测滑坡易发性的平均精度最高且不确定性较低，基于PS的耦合模型预测精度最低且不确定性最高，基于FR、Ⅳ和IOE的耦合模型介于两者之间；（2）单独数据驱动模型易发性预测精度略低于耦合模型，且未能计算出环境因子各子区间对滑坡发育的影响规律，但其建模效率高于耦合模型；（3）RF模型预测精度最高且不确定性较低，其次分别为SVM、BPNN和LR模型.总之WOE是更优秀的联接法且RF模型预测性能最优，WOE-RF模型预测的滑坡易发性不确定性较低且更符合实际滑坡概率分布特征.
|
| 关 键 词： | 滑坡易发性预测 不确定性分析 联接方法 数据驱动 证据权 随机森林 工程地质学 |
| 收稿时间： | 2020-11-28 |
Uncertainties of Landslide Susceptibility Modeling under Different Environmental Factor Connections and Prediction Models |
| |
| Li Wenbin,Fan Xuanmei,Huang Faming,Wu Xueling,Yin Kunlong,Chang Zhilu.Uncertainties of Landslide Susceptibility Modeling under Different Environmental Factor Connections and Prediction Models[J].Earth Science-Journal of China University of Geosciences,2021,46(10):3777-3795. |
| |
| Authors: | Li Wenbin Fan Xuanmei Huang Faming Wu Xueling Yin Kunlong Chang Zhilu |
| |
| Abstract: | This study aims to explore the influences of some modeling factors including the non-linear correlation calculation between landslides and environmental factors and the different data-based models on the uncertainty law of landslide susceptibility prediction (LSP) modeling. The Ruijin City of Jiangxi Province in China with investigated 370 landslides and 10 environmental factors is used as study case. Accordingly, a total of 20 types of different coupling modeling conditions are proposed for LSP with five different connection methods(probability statistics (PS), frequency ratio (FR), information value (Ⅳ), index of entropy (IOE) and weight of evidence (WOE)) and four different data-based models including logistic regression (LR), back propagation neural networks (BPNN), support vector machines (SVM) and random forest (RF). Meanwhile, four single LR, BPNN, SVM and RF models with the original data as input variables are also proposed, as a whole, a total of 24 types of modeling conditions for LSP are obtained based on the above 20 types of coupling conditions and 4 types of single models. Finally, the uncertainty characteristics in the LSP modeling are assessed using the area under the receiver operation curve (ROC), mean value, standard deviation and significance test, respectively. Results show follows. (1) WOE-based models have the highest LSP accuracy and low uncertainty while PS-based models have the lowest LSP accuracy and the highest uncertainty, and the FR, Ⅳ and IOE-based models are in between. (2) The single data-based models have slightly lower LSP accuracies than those of the coupling models on the whole and cannot calculate the influence law of each sub-interval of environmental factors on landslide evolution, however, the single data-based models have higher modeling efficiency than those of the coupling models. (3) Among all the data-based models, RF model has the highest LSP accuracy and relatively low uncertainty, followed by the SVM, BPNN and LR models, respectively. It is concluded that the WOE is a very excellent correlation method and the RF model predicts the optimal LSP performance, the LSP results of WOE-RF model have the lowest uncertainties and the predicted landslide susceptibility indexes are more consistent with the actual landslides distribution characteristics. |
| |
| Keywords: | |
| 本文献已被 万方数据 等数据库收录！ |
| 点击此处可从《地球科学》浏览原始摘要信息 |
|
点击此处可从《地球科学》下载全文 |
|
