Main controlling factor analysis and susceptibility assessment of landslides in southeastern Tibet based on optimized random forest model
-
摘要:
以藏东南地区为研究对象, 通过优化随机森林模型对滑坡易发性进行深入分析, 探讨影响滑坡发生的主要因素.通过现场调查、遥感数据分析和文献综述, 系统筛选了滑坡与非滑坡样本, 并优化了模型的滑坡-非滑坡样本筛选方法、影响因子的选取、联结方法的应用以及超参数的优化.采用非支配排序算法(NSGA-Ⅱ)优化随机森林模型(多目标优化)与RF-GA模型(单目标优化)进行对比分析, 最优准确率、召回率、精确率、F1四项指标较RF-GA模型分别提高了3.3%、8.7%、3.2%、1.9%.在滑坡易发性分区方面, 高、较高、中易发区面积占比分别提升了2.7%、3.1%、1.2%.通过绘制ROC曲线和计算AUC值, 验证了RF-NSGA-Ⅱ模型的高准确性(AUC=0.877).研究结果显示, 藏东南地区的滑坡易发区主要集中在易贡藏布与帕隆藏布的交汇处以及雅鲁藏布江的大拐弯区域.在影响因子的重要性排序中, 距道路距离、高程和距河流距离排名靠前, 地质构造复杂、断层发育密集以及长期构造活动的影响使得这些区域滑坡发育频繁, 特别是在断层纵横交错、岩石破碎和节理发育明显的高易发区域.
Abstract:Taking the southeastern Tibet Region as the research object, the paper analyzes the landslide susceptibility based on optimized random forest(RF) model, and discusses the main influencing factors of landslides. With field survey, remote sensing data analysis and literature review, the study screens the landslide and non-landslide samples systematically, and optimizes the landslide and non-landslide sample screening methods of models, selection of impact factors, application of connection approach and hyperparameters. The random forest model (multi-objective optimization) is optimized by non-dominated sorting genetic algorithm (NSGA-Ⅱ) and compared with RF-GA mode (single objective optimization). The four indexes of optimal accuracy, recall rate, precision rate and F1 are increased by 3.3%, 8.7%, 3.2% and 1.9%, respectively, compared with the RF-GA model. Besides, the high, relatively high and medium susceptible areas increased by 2.7%, 3.1% and 1.2%, respectively, in terms of landslide susceptibility zoning. The high accuracy of RF-NSGA-Ⅱ model is verified (AUC=0.877) by drawing ROC curve and calculating AUC value. The results show that the landslide susceptible areas in southeastern Tibet are mainly concentrated in the intersection of Yigong Zangbo River and Palong Zangbo River and the big bend area of Yarlung Zangbo River. In the importance ranking of landslide impact factors, distance from road, elevation and distance from river take up the top places, as complex geological structures, densely developed faults and influence of long-term tectonic activities cause frequent landslides in such areas, especially in high susceptible areas with criss-crossing faults, broken rocks and developed joints.
-
-
表 1 滑坡灾害影响因子数据来源表
Table 1. Data sources for landslide hazard impact factors
编号 主控因子 评价因子 因子致灾特征及数据来源 1 高程 2 坡度 3 坡向 4 地形地貌 坡度变率 坡体特征; ASTER GDEM v2.0(12.5 m分辨率)，数字地形高程通过ArcGIS软件生成 5 平面曲率 6 地形起伏度 7 坡度变率 8 坡向变率 9 地层岩性 地层岩性分区 坡体岩土物理指标; 通过1:20万地质图( http://dcc.ngac.org.cn )提取10 水文 距河流距离 坡体水文因子特征及侵蚀坡脚状况; 通过1:25万基础地理( http://dcc.ngac.org.cn )提取11 人类活动 距道路距离 人类工程设施对坡体影响; 通过1:25万基础地理( http://dcc.ngac.org.cn )提取12 气象 多年有效平均降雨量 降雨入渗对滑坡坡体稳定性影响; 通过国家气候中心获取降雨数据, 1 km栅格( http://www.ncc-cma.net/cn )13 地质构造 距主要断层距离 地质构造对坡体发育影响; 基于1:20万地质图(公开版) 14 环境地质 土地利用类型 影响坡体发育环境特征; 基于Landsat 8遥感影响数据，30m分辨率 15 地表切割深度 16 地形粗糙度 17 归一化植被指数 
下载: 导出CSV
表 2 主控因子量化结果表
Table 2. Quantitative results of main control factors
序号 类型 分类 影响因子 分级范围 归一化值 A 坡度 0~87° B 坡向 0~360° C 高程 742.039~7109.46 m D 连续型 地形地貌因子 平面曲率 0~70.19 0~1 E 剖面曲率 0~38.96 F 地形起伏度 0~2294 m G 坡度变率 0~38.95 H 坡向变率 0~70.19 I 连续型 水文因子 距河流距离 0~5600 m 0~1 J 连续型 人类活动因子 距道路距离 0~4800 m 0~1 K 连续型 气象因子 多年平均降雨量 0~1289.88 mm 0~1 L 连续型 地质构造因子 距断层距离 0~40000 m 0~1 M 地表切割深度 0~586.3 N 连续型 环境地质因子 地形粗糙度 1~3.526 0~1 O 归一化植被指数 -0.20~0.99 P 离散型 土地利用类型 1~10类 0~1 Q 地层岩性因子 地层岩性分区 1~4类
下载: 导出CSV
表 3 测试集AUC值变化表
Table 3. AUC value changes of test set
影响因子筛选 未剔除 第1次剔除 第2次剔除 第3次剔除 第4次剔除 第5次剔除 剔除因子 地表切割深度 地形粗糙度 坡向变率 剖面曲率 NDVI 测试集AUC值 0.746 0.754 0.778 0.762 0.768 0.775
下载: 导出CSV
表 4 优化前后超参数对比
Table 4. Comparison of hyperparameters before and after optimization
参数名 RF-NSGA-Ⅱ RF-GA 训练用时 34 h 42 min 48 s 2s 树的数量 50 100 内部节点再分裂最小样本数 50 2 叶子节点所需最小样本数 48 1 树的最大深度 12 10
下载: 导出CSV
表 5 优化前后训练集结果对比
Table 5. Comparison of training set results before and after optimization
训练集 准确率 召回率 精确率 F1 NSGA-Ⅱ-RF 0.827 0.819 0.801 0.815 GA-RF 0.794 0.732 0.769 0.796
下载: 导出CSV
表 6 滑坡易发性评价分区统计结果
Table 6. Statistics of landslide susceptibility assessment zoning
易发性等级 高 较高 中 较低 低 面积占比/% 分区面积 滑坡面积 分区面积 滑坡面积 分区面积 滑坡面积 分区面积 滑坡面积 分区面积 滑坡面积 RF－NSGA－Ⅱ 44.42 14.95 28.79 29.52 15.46 4.63 2.34 17.94 2.34 12.96 RF－GA 41.69 12.25 28.66 26.52 21.67 23.63 8.71 21.94 2.68 15.67
下载: 导出CSV
表 7 典型滑坡群样本分布表
Table 7. Sample distribution of typical landslide groups
分段 河道长/km 滑坡数/处 高易发区面积占比(优化RF)/% 高易发区面积占比(常规RF)/% 滑坡密度/(处/km2) 派镇麦朗村-直白村 16.8 23 5.87 4.96 1.24 巴嘎沟-沟东 22.6 28 7.14 7.64 1.16 索松村-蹦嘎 23.8 27 6.89 6.23 0.87 本宗-增多 34.9 32 8.16 7.89 0.68 丹娘乡-卧龙镇 77.2 48 16.67 15.23 0.63 卧龙-米堆 57.8 44 12.24 10.28 0.57 森波-帮羌纳乡 52.8 36 9.18 8.16 0.48 玉松村-日加贡巴 16.8 15 3.83 2.96 0.41 朗多-白拉村 92.3 45 11.48 10.48 0.26 岗派公路附近 87.9 32 8.16 7.96 0.25 派村-格嘎村 91.7 15 3.83 3.21 0.19 羌纳乡-则隆弄 58.4 12 3.06 2.89 0.14 米堆-101道班 284.5 35 8.93 8.36 0.11
下载: 导出CSV
-
[1] 许强. 对滑坡监测预警相关问题的认识与思考[J]. 工程地质学报, 2020, 28(2): 360-374.
Xu Q. Understanding the landslide monitoring and early warning: Consideration to practical issues[J]. Journal of Engineering Geology, 2020, 28(2): 360-374.
[2] 韩用顺. 藏东南地质灾害易发性评价[J]. 山地学报, 2021, 39(5): 700.
Han Y S. Assessment of geological disaster susceptibility in southeastern Tibet[J]. Mountain Research, 2021, 39(5): 700. (in Chinese)
[3] 冉涛, 徐如阁, 周洪福, 等. 雅砻江流域深切河谷区滑坡类型、成因及分布规律——以子拖西-麻郎错河段为例[J]. 中国地质, 2024, 51(2): 511?524.
Ran T, Xu R G, Zhou H F, et al. Type, formation mechanism and distribution regularity of landslides in the deeply incised valley area of Yalong River Basin: A case study of Zituoxi-Malangcuo River section [J]. Geology in China, 2024, 51(2): 511-524.
[4] 牛犇, 冯春, 丛俊余, 等. 基于CDEM颗粒流的三维高速远程滑坡成灾范围分析[J]. 岩石力学与工程学报, 2023, 42(S2): 4018-4027.
Niu B, Feng C, Cong J Y, et al. Three-dimensional high-speed remote landslide disaster area analysis based on CDEM particle flow [J]. Chinese Journal of Rock Mechanics and Engineering, 2023, 42(S2): 4018-4027.
[5] 袁浩, 郭长宝, 吴瑞安, 等. 西藏易贡高位远程滑坡研究进展与展望[J]. 地质通报, 2023, 42(10): 1757-1773.
Yuan H, Guo C B, Wu R A, et al. Research progress and prospects of the giant Yigong long run-out landslide, Tibetan Plateau, China[J]. Geological Bulletin of China, 2023, 42(10): 1757-1773.
[6] 王哲, 赵超英, 刘晓杰, 等. 西藏易贡滑坡演化光学遥感分析与InSAR形变监测[J]. 武汉大学学报(信息科学版), 2021, 46(10): 1569-1578.
Wang Z, Zhao C Y, Liu X J, et al. Evolution analysis and deformation monitoring of Yigong landslide in Tibet with optical remote sensing and InSAR[J]. Geomatics and Information Science of Wuhan University, 2021, 46(10): 1569-1578.
[7] 韩盼, 余国安, 侯伟鹏. 藏东南帕隆藏布流域冲积扇地貌发育特征及空间分异研究[J]. 地理科学进展, 2024, 43(4): 784-798.
Han P, Yu G A, Hou W P. Geomorphological development characteristics and spatial differentiation of alluvial fans in high mountain canyon areas: A case study in the Parlung Tsangpo Basin, southeast Tibet[J]. Progress in Geography, 2024, 43(4): 784-798.
[8] 杨帆, 马志刚, 文艳, 等. 联合SAR影像相位和幅度信息的多形变量级滑坡监测——以金沙江白格滑坡为例[J]. 自然资源遥感, 2024, 36(2): 257-267.
Yang F, Ma Z G, Wen Y, et al. Monitoring a landslide with a multi-deformation magnitude based on the phase and amplitude information of SAR images: A case study of the Baige landslide in Jinsha River [J]. Remote Sensing for Natural Resources, 2024, 36(2): 257-267.
[9] 付杰, 张青, 仉文岗. 基于多源数据与机器学习的降雨型滑坡灾害危险性动态预警[J]. 地质论评, 2024, 70(S1): 217-218.
Fu J, Zhang Q, Zhang W G. Dynamic warning of rainfall-induced landslide hazard based on multi-source data and machine learning[J]. Geological Review, 2024, 70(S1): 217-218.
[10] 张冬梅, 张莉. 对话姚檀栋: 走近第二次青藏高原综合科学考察[J]. 科学通报, 2019, 64(27): 2765-2769.
Zhang D M, Zhang L. Tandong Yao: The second Tibetan Plateau scientific expedition and research[J]. Chinese Science Bulletin, 2019, 64(27): 2765-2769.
[11] 王悦, 曹颖, 许方党, 等. 考虑非滑坡样本选取和集成机器学习方法的水库滑坡易发性预测[J]. 地球科学, 2024, 49(5): 1619-1635.
Wang Y, Cao Y, Xu F D, et al. Reservoir landslide susceptibility prediction considering non-landslide sampling and ensemble machine learning methods[J]. Earth Science, 2024, 49(5): 1619-1635.
[12] 李正, 冷亮, 孙永鑫, 等. 基于信息量-机器学习耦合模型的水电梯级开发流域滑坡易发性评价[J]. 测绘通报, 2024(S1): 237-241.
Li Z, Leng L, Sun Y X, et al. Landslide susceptibility assessment in the river cascade development basin based on the Ⅳ-LM coupling model[J]. Bulletin of Surveying and Mapping, 2024(S1): 237-241.
[13] 茹曼, 郑燕, 张斌, 等. 基于SVM-RF模型的地质灾害易发性评价——以河南省博爱县青天河景区为例[J]. 地质与资源, 2023, 32(5): 633-641. DOI: 10.13686/j.cnki.dzyzy.2023.05.014.
Ru M, Zheng Y, Zhang B, et al. Assessment of geological disaster susceptibility based on SVM-RF MODEL: A case study of Qingtian River scenic area in Boai County, Henan Province[J]. Geology and Resources, 2023, 32(5): 633-641. DOI: 10.13686/j.cnki.dzyzy.2023.05.014.
[14] 李成林, 刘严松, 赖思翰, 等. 基于逻辑回归和支持向量机耦合模型的滑坡易发性分析[J]. 自然灾害学报, 2024, 33(2): 75-86.
Li C L, Liu Y S, Lai S H, et al. Landslide susceptibility analysis based on the coupling model of logistic regression and support vector machine[J]. Journal of Natural Disasters, 2024, 33(2): 75-86.
[15] 杜鹏, 陈宁生, 伍康林, 等. 基于随机森林模型的藏东南地区滑坡易发性评价及主控因素分析[J]. 成都理工大学学报(自然科学版), 2024, 51(2): 328-344.
Du P, Chen N S, Wu K L, et al. Evaluation of landslide susceptibility in southeast Tibet based on a random forest model[J]. Journal of Chengdu University of Technology (Science & Technology Edition), 2024, 51(2): 328-344.
[16] 林泊锟, 丁勇, 李登华. 一种基于深度学习与图像局部特征提取的边坡异常监测技术[J]. 测绘通报, 2024(4): 23-28.
Lin B K, Ding Y, Li D H. A slope anomaly monitoring technology based on deep learning and image local feature extraction[J]. Bulletin of Surveying and Mapping, 2024(4): 23-28.
[17] 冯旻譞, 毛伊敏, 贾俊, 等. 地质灾害危险性评价中不同机器学习方法优劣对比: 以宁强县大安镇为例[J/OL]. 中国地质, 1-13[2024-06-17].
Feng M X, Mao Y M, Jia J, et al. Comparison of the advantages and disadvantages of different machine learning methods in geohazard risk assessment: Taking Da'an Town, Ningqiang County as an example [J/OL]. China Geology, 1-13[2024-06-17].
[18] 蒋宏伟, 刘健鹏, 王新杰, 等. 随机森林优化的静动态耦合模型在滑坡位移预测中的应用[J]. 常州大学学报(自然科学版), 2024, 36(3): 80-92.
Jiang H W, Liu J P, Wang X J, et al. Application of static dynamic coupling model optimized by random forest in landslide displacement prediction[J]. Journal of Changzhou University (Natural Science Edition), 2024, 36(3): 80-92.
[19] 黄发明, 欧阳慰平, 蒋水华, 等. 考虑机器学习建模中训练/测试集时空划分原则的滑坡易发性预测建模[J]. 地球科学, 2024, 49(5): 1607-1618.
Huang F M, Ouyang W P, Jiang S H, et al. Landslide susceptibility prediction considering spatio-temporal division principle of training/ testing datasets in machine learning models[J]. Earth Science, 2024, 49(5): 1607-1618.
[20] 王本栋, 李四全, 许万忠, 等. 基于3种不同机器学习算法的滑坡易发性评价对比研究[J]. 西北地质, 2024, 57(1): 34-43.
Wang B D, Li S Q, Xu W Z, et al. A comparative study of landslide susceptibility evaluation based on three different machine learning algorithms[J]. Northwestern Geology, 2024, 57(1): 34-43.
[21] 周剑, 汤明高, 裴芳歌, 等. 基于机器学习的库岸滑坡变形短期预测[J]. 山地学报, 2023, 41(6): 891-903.
Zhou J, Tang M G, Pei F G, et al. Short-term deformation of reservoir slope based on machine learning[J]. Mountain Research, 2023, 41(6): 891-903.
[22] 翟文华, 王小东, 吴明堂, 等. 基于频率比模型和随机森林模型耦合的地质灾害易发性评价[J]. 自然灾害学报, 2023, 32(6): 74-82.
Zhai W H, Wang X D, Wu M T, et al. Geological disaster susceptibility evaluation based on coupling of frequency ratio model and random forest model[J]. Journal of Natural Disasters, 2023, 32(6): 74-82.
[23] Merghadi A, Yunus A P, Dou J, et al. Machine learning methods for landslide susceptibility studies: a comparative overview of algorithm performance[J]. Earth-Science Reviews, 2020, 207: 103225.
[24] Pawluszek K, Borkowski A. Impact of DEM-derived factors and analytical hierarchy process on landslide susceptibility mapping in the region of Roznów Lake, Poland[J]. Natural Hazards, 2017, 86(2): 919-952.
[25] Yang H, Shi, P J, Quincey D, et al. A heterogeneous sampling strategy to model earthquake-triggered landslides[J]. International Journal of Disaster Risk Science, 2023, 14(4): 636-648.
[26] Tabatabaei M, Salehpour Jam A, Ahmad Hosseini S. Suspended sediment load prediction using non-dominated sorting genetic algorithm Ⅱ[J]. International Soil and Water Conservation Research, 2019, 7(2): 119-129.
[27] 吴辰文, 梁靖涵, 王伟, 等. 基于递归特征消除方法的随机森林算法[J]. 统计与决策, 2017(21): 60-63.
Wu C W, Liang J H, Wang W, et al. Random forest algorithm based on recursive feature elimination[J]. Statistics & Decision, 2017(21): 60-63.
[28] 孙德亮, 陈丹璐, 密长林, 等. 基于随机森林-特征递归消除模型的可解释性缓丘岭谷地貌滑坡易发性评价[J]. 地质力学学报, 2023, 29(2): 202-219.
Sun D L Chen D L, Mi C L, et al. Evaluation of landslide susceptibility in the gentle hill-valley areas based on the interpretable random forest-recursive feature elimination model[J]. Journal of Geomechanics, 2023, 29(2): 202-219.
[29] Cantarino I, Carrion M A, Goerlich F, et al. A ROC analysis-based classification method for landslide susceptibility maps[J]. Landslides, 2019, 16(2): 265-282.
-
图(13)
表(7)
计量
- 文章访问数: 99
- PDF下载数: 35
- 施引文献: 0