| 包维斌,李成云,张维东,张景庄,郭万里.基于BP、RF模型的黄土区地质灾害易发性评价
——以正宁县山河镇为例[J].甘肃地质,2025,(1):34-41 |
| 基于BP、RF模型的黄土区地质灾害易发性评价
——以正宁县山河镇为例 |
| Evaluation ｏf Geological Disaster Susceptibility ｉn Loess Region Based ｏn BP and RF Models ——Taking Shanhezheng Town of Zhengning County as an example |
| |
| DOI： |
| 中文关键词: 地质灾害 易发性分区 人工神经网络 随机森林 黄土区 |
| 英文关键词:geological disaster, susceptibility mapping, artificial neural network, random forest |
| 基金项目:庆阳市正宁县山河镇1 ∶ 10 000地质灾害精细调查项目资助（甘财资环〔２０２２〕９５号） |
|
| 摘要点击次数: 65 |
| 全文下载次数: 46 |
| 中文摘要: |
| 黄土区土壤侵蚀严重，滑坡、崩塌等地质灾害易发频发，因此开展地质灾害易发性空间分布规律预测研究尤为重要。以正宁县山河镇为例，选取坡度、与侵蚀塬边距离、高差、坡向、坡面曲率和土地利用等6个孕灾条件因子，利用人工神经网络（BP）、随机森林（RF）开展易发性预测评价。结果表明：地质灾害易发性孕灾因子中坡度、与侵蚀塬边距离、高差是所占权重较大的因子，坡向、坡面曲率和土地利用等所占权重较小；中、高易发区主要分布于溯源侵蚀的沟谷、四郎河和贾峪川河左岸的人类活动集中区，低易发区主要分布于溯源侵蚀的塬边，非易发区主要分布于地形平坦的台塬地带。BP、RF模型的AUC值分别为0. 799、0. 857，RF模型预测效果优于BP模型，可为黄土区地质灾害易发性预测和防灾减灾提供指导意义。 |
| 英文摘要: |
| The Soil erosion is severe in loess areas, and geological disasters such as landslides and collapses are prone to occur frequently. Therefore, it is particularly important to carry out research on predicting the spatial distribution pattern of geological disaster susceptibility. Taking Shanhe Town in Zhengning County as an example, six disaster prone factors were selected, including slope gradient, distance from eroded plateau edge, height difference, slope orientation, slope curvature, and land use. Artificial neural network（BP） and random forest （RF） were used to conduct susceptibility prediction and evaluation. The results indicate that slope, distance from the eroded edge, and elevation difference are the factors with higher weights in the susceptibility factors of geological disasters, while slope orientation, slope curvature, and land use have lower weights; The areas with medium and high susceptibility are mainly distributed in the valleys affected by source erosion, as well as in the human activity concentration areas on the left bank of the Silang River and Jiayuchuan River. The areas with low susceptibility are mainly distributed on the edge of the source erosion plateau, while the areas with low susceptibility are mainly distributed on flat plateau terrain. The AUC values of the BP and RF models are 0. 799 and 0. 857, respectively. The RF model has better predictive performance than the BP model, which can provide guidance for predicting the susceptibility of geological disasters and disaster prevention and reduction in loess areas. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
