基于PSO_RF与SHAP的地质灾害易发性评价

Geological disaster susceptibility assessment based on PSO_RF and SHAP

摘要

为寻找适宜山区的最佳地质灾害易发性评价方法与解释模型输入到输出的黑箱问题，构建了粒子群优化随机森林的评价模型并量化了影响因子对评价结果的影响。以石泉县为研究区，最终选取高程、坡度、坡向、工程地质岩组等12个相关性程度较低的因子并以支持向量机(support vector machine，SVM)、极端梯度提升(eXtreme gradient boosting，XGBoost)、随机森林(random forest，RF)与粒子群优化随机森林(particle swarm optimized random forest，PSO_RF)开展地质灾害易发性评价，采用多种指标评价模型的精度，并将研究区划分为极高易发区、高易发区、中易发区、低易发区和极低易发区共5个分区。研究结果表明，经过优化后的模型(PSO_RF)曲线下面积(area under curve，AUC)值最大，为0.878，各项指标明显大于其他模型，模型分区结果均划分合理，高、极高易发区面积主要分布于研究区西北部、中部、东部及南部少量区域。利用沙普利加性解释(SHapley Additive exPlanations，SHAP)最优模型结果显示，特征重要性的前4位分别为岩组、归一化差分植被指数(normalized difference vegetation index，NDVI)、道路与高程，其中岩组对易发性的结果的影响作用最为强烈。结果表明，构建的耦合模型(PSO_RF)表现最佳，更适合山区易发性评价研究，因子的影响程度可视化有助于理解模型预测的黑箱问题。研究结果可为评价因子与模型的选择提供思路，为当地的防灾减灾提供一定参考。

Abstract

In order to identify the optimal method for geological disaster susceptibility assessment in mountainous areas and address the black box issue of model input and output, a particle swarm optimized random forest (PSO_RF) evaluation model was developed, and the impact of influencing factors on the assessment results was quantified. Shiquan County was selected as the study area, where 12 factors with low correlation, such as elevation, slope, aspect, and engineering geological rock groups, were ultimately chosen. Geological disaster susceptibility assessments were conducted using support vector machine (SVM), eXtreme gradient boosting (XGBoost), random forest (RF), and particle swarm optimized random forest (PSO_RF). Multiple indicators were used to evaluate the accuracy of these models, and the study area was divided into five zones: very high susceptibility, high susceptibility, moderate susceptibility, low susceptibility, and very low susceptibility. The results indicated that the optimized model (PSO_RF) achieved the highest AUC (area under curve) value of 0.878, significantly outperforming the other models. The zoning results of the model were all reasonable, with high and very high susceptibility areas mainly distributed in the northwest, central, eastern, and a small part of the southern regions of the study area. Using SHAP (SHapley Additive exPlanations) to interpret the results of the optimal model revealed that the top four most important features were rock group, NDVI (normalized difference vegetation index), roads, and elevation, with the rock group having the most impact significant on susceptibility results. The results indicate that the coupled model (PSO_RF) performs the best and is more suitable for susceptibility evaluation in mountainous areas. The visualization of the impact of factors helps to understand the black box problem of model prediction. These findings can provide insights for the selection of evaluation factors and models and offer references for local disaster prevention and mitigation efforts.

关键词

石泉县 / 地质灾害 / 易发性评价 / PSO_RF / ROC曲线

Key words

Shiquan County / geological hazards / susceptibility assessment / PSO_RF / ROC curve

参考文献

[1] 董秀军, 邓博, 袁飞云, 等 . 航空遥感在地质灾害领域的应用: 现状与展望[J]. 武汉大学学报(信息科学版), 2023, 48(12): 1897-1913.

[2] Dong Xiujun, Deng Bo, Yuan Feiyun, et al. Application of aerial remote sensing in geological hazards: current situation and prospects[J]. Geomatics and Information Science of Wuhan University, 2023, 48(12): 1897-1913. (in Chinese)

[3] 许强, 郭晨, 董秀军. 地质灾害航空遥感技术应用现状及展望[J]. 测绘学报, 2022, 51(10): 2020-2033.

[4] Xu Qiang, Guo Chen, Dong Xiujun. Application status and prospect of aerial remote sensing technology for geohazards[J]. Acta Geodaetica et Cartographica Sinica, 2022, 51(10): 2020-2033. (in Chinese)

[5] 张永双, 李长冬, 张帆宇, 等 . 自然灾害基础理论体系研究现状与展望[J]. 中国科学基金, 2025, 39(6): 872-879.

[6] Zhang Yongshuang, Li Changdong, Zhang Fanyu, et al. The progress and prospects of the basic theoretical system on natural disasters[J]. Bulletin of National Natural Science Foundation of China, 2025, 39(6): 872-879. (in Chinese)

[7] 王伟中, 曹小红, 汪鑫, 等 . 基于信息量法的新疆伊宁县滑坡灾害易发性评价[J]. 新疆地质, 2022, 40(4): 566-573.

[8] Wang Weizhong, Cao Xiaohong, Wang Xin, et al. Assessment of the susceptibility to landslide hazards in Yining County based on information value method[J]. Xinjiang Geology, 2022, 40(4): 566-573. (in Chinese)

[9] Lu Jiangang, He Yi, Zhang Lifeng, et al. Ensemble learning landslide susceptibility assessment with optimized non-landslide samples selection[J]. Geomatics, Natural Hazards and Risk, 2024, 15(1): 2378176.

[10] 王毅, 方志策, 牛瑞卿, 等 . 基于深度学习的滑坡灾害易发性分析[J]. 地球信息科学学报, 2021, 23(12): 2244-2260.

[11] Wang Yi, Fang Zhice, Niu Ruiqing, et al. Landslide susceptibility analysis based on deep learning[J]. Journal of Geo-Information Science, 2021, 23(12): 2244-2260. (in Chinese)

[12] 孟晓捷, 张新社, 曾庆铭, 等 . 基于加权信息量法的黄土滑坡易发性评价: 以1∶5万天水市麦积幅为例[J]. 西北地质, 2022, 55(2): 249-259.

[13] Meng Xiaojie, Zhang Xinshe, Zeng Qingming, et al. The susceptibility evaluation of loess landslide based on weighted information value method: taking 1∶50000 map of Maiji District of Tianshui City as an example[J]. Northwestern Geology, 2022, 55(2): 249-259. (in Chinese)

[14] 陈志波, 李鼎兴, 陈澄, 等 . 基于机器学习和InSAR技术的滑坡易发性动态评价[J]. 自然灾害学报, 2025, 34(2): 56-65.

[15] Chen Zhibo, Li Dingxing, Chen Cheng, et al. Dynamic evaluation of landslide susceptibility based on machine learning and InSAR technology[J]. Journal of Natural Disasters, 2025, 34(2): 56-65. (in Chinese)

[16] 林沛文, 何书, 鲜木斯艳·阿布迪克依木, 等 . 基于逻辑回归-模糊层次分析模型的公路沿线滑坡易发性评价[J]. 自然灾害学报, 2024, 33(6): 27-36.

[17] Lin Peiwen, He Shu, Abudikeyimu Xmsy, et al. Landslide susceptibility assessment along highways based on LR-FAHP model[J]. Journal of Natural Disasters, 2024, 33(6): 27-36. (in Chinese)

[18] 王守华, 王睿菘, 孙希延, 等 . 基于CF-CNN-LSTM模型的滑坡易发性评价[J]. 自然灾害学报, 2024, 33(5): 84-95.

[19] Wang Shouhua, Wang Ruisong, Sun Xiyan, et al. Landslide susceptibility evaluation based on CF-CNN-LSTM model[J]. Journal of Natural Disasters, 2024, 33(5): 84-95. (in Chinese)

[20] Huang Faming, Xiong Haowen, Yao Chi, et al. Uncertainties of landslide susceptibility prediction considering different landslide types[J]. Journal of Rock Mechanics and Geotechnical Engineering, 2023, 15(11): 2954-2972.

[21] Liu Zaobao, Shao Jianfu, Xu Weiya, et al. Comparison on landslide nonlinear displacement analysis and prediction with computational intelligence approaches[J]. Landslides, 2014, 11(5): 889-896.

[22] 赵恒梓, 李素敏, 郭庆, 等 . 样本优化的集成学习模型在滑坡易发性评价中的应用[J]. 测绘科学, 2024, 49(12): 132-141.

[23] Zhao Hengzi, Li Sumin, Guo Qing, et al. Application of ensemble learning model based on sample optimization in landslide susceptibility evaluation[J]. Science of Surveying and Mapping, 2024, 49(12): 132-141. (in Chinese)

[24] 张向营, 张春山, 孟华君, 等 . 基于Random Forest和AHP的贵德县北部山区滑坡危险性评价[J]. 水文地质工程地质, 2018, 45(4): 142-149.

[25] Zhang Xiangying, Zhang Chunshan, Meng Huajun, et al. Landslide hazard evaluation in the northern mountainous area of Guide County based on Random Forest and AHP[J]. Hydrogeology & Engineering Geology, 2018, 45(4): 142-149. (in Chinese)

[26] 崔婷婷, 安雪莲, 孙德亮, 等 . 基于SHAP的可解释机器学习的滑坡易发性评价模型[J]. 成都理工大学学报(自然科学版), 2025, 52(1): 153-172.

[27] Cui Tingting, An Xuelian, Sun Deliang, et al. Explainable machine learning models for landslide susceptibility mapping based on SHAP[J]. Journal of Chengdu University of Technology (Science & Technology Edition), 2025, 52(1): 153-172. (in Chinese)

[28] 马祥龙, 文海家, 张廷斌, 等 . 自动可解释机器学习滑坡易发性评价模型[J]. 北京师范大学学报(自然科学版), 2024, 60(6): 806-818.

[29] Ma Xianglong, Wen Haijia, Zhang Tingbin, et al. An automated and explainable machine learning model for landslide susceptibility mapping[J]. Journal of Beijing Normal University (Natural Science), 2024, 60(6): 806-818. (in Chinese)

[30] 许力. 石泉县地热资源潜力分析[D]. 西安: 长安大学, 2021.

[31] Xu Li. Potential analysis of geothermal resources in Shiquan County[D]. Xi’an: Chang’an University, 2021. (in Chinese)

[32] 王启盛, 熊俊楠, 程维明, 等 . 耦合统计方法、机器学习模型和聚类算法的滑坡易发性评价方法[J]. 地球信息科学学报, 2024, 26(3): 620-637.

[33] Wang Qisheng, Xiong Junnan, Cheng Weiming, et al. Landslide susceptibility mapping methods coupling with statistical methods, machine learning models and clustering algorithms[J]. Journal of Geo-Information Science, 2024, 26(3): 620-637. (in Chinese)

[34] 刘伊铭, 徐胜华, 刘春阳, 等 . 顾及多方法集成特征选择与负样本优化的滑坡易发性评价[J]. 测绘通报, 2024(9): 74-79.

[35] Liu Yiming, Xu Shenghua, Liu Chunyang, et al. Landslide susceptibility assessment considering multi-method integrated feature selection and negative sample optimization[J]. Bulletin of Surveying and Mapping, 2024(9): 74-79. (in Chinese)

[36] 孙聿卿, 葛永刚, 陈兴长, 等 . 金沙江流域泥石流危险性评价[J]. 地质通报, 2025, 44(2/3): 377-391.

[37] Sun Yuqing, Ge Yonggang, Chen Xingchang, et al. Hazard assessment of debris flow in Jinsha River[J]. Geological Bulletin of China, 2025, 44(2/3): 377-391. (in Chinese)

[38] 袁旭山, 刘京会, 黄龙生, 等 . 基于机器学习模型的新源县滑坡易发性评估[J]. 科学技术与工程, 2025, 25(5): 1815-1826.

[39] Yuan Xushan, Liu Jinghui, Huang Longsheng, et al. Assessment of landslide susceptibility in Xinyuan County based on machine learning models[J]. Science Technology and Engineering, 2025, 25(5): 1815-1826. (in Chinese)

基金资助

国家自然科学基金项目(41807243)

陕西省自然科学基金项目(2023-JC-YB-231)

中央高校基金项目(300102264909)