1.College of Juncao Science and Ecology,Fujian Agriculture and Forestry;University,Fuzhou,Fujian 350002,China
2.Key Laboratory for Forest Ecosystem Process and;Management of Fujian Province,Fuzhou,Fujian 350002,China
3.Key Laboratory of Mountain Hazards and;Engineering Safety,Institute of Mountain Hazards and Environment,Chinese Academy of Sciences and Ministry of;Water Resources,Chengdu,Sichuan 610213,China
4.University of the Chinese Academy of Sciences,Beijing 100049,China
Objective This study aims to establish a technical pathway of ‘factor selection-model evaluation-mechanism analysis’ to investigate evaluation models with high predictive accuracy for landslide susceptibility, reveal the key driving factors of landslide disasters, and explore the interaction mechanisms among landslide influencing factors under complex geological conditions. It can provide scientific support for disaster risk management and the formulation of ecological protection strategies in the Jiuzhaigou scenic area and similar post-seismic regions with highly concealed landslides. Methods The Jiuzhaigou scenic area was selected as the study area. Landslide susceptibility was evaluated using both traditional methods-analytic hierarchy process (AHP), information value (IV), and certainty factor (CF) and machine learning models (XGBoost, LightGBM, and CatBoost). A systematic evaluation indicator system was constructed based on correlation analysis and collinearity tests. The shapley additive explanations(SHAP) explainable algorithm and the optimal parameter-based geodetector model (OPGD) were used to identify key controlling factors and investigate their interaction mechanisms. Results Among the landslide susceptibility evaluation models, the machine learning models overall outperformed the traditional methods, with the CatBoost model achieving the highest predictive accuracy (AUC=0.927). High-susceptibility zones were concentrated in Panda Lake, Arrow Bamboo Lake, northwestern Danzugou, southwestern Grass Lake, and southeastern Long Lake. Both SHAP and OPGD identified distance to water systems, normalized difference vegetation index (NDVI), slope aspect, and multi-year average annual rainfall as the primary controlling factors. OPGD interaction detection revealed that the interaction between distance to water systems and distance to faults was the strongest (q=0.33), and the relationship between multi-year average annual rainfall and NDVI showed a nonlinear enhancement (q=0.16). Conclusion There are multiple potential zones of high landslide susceptibility within the Jiuzhaigou scenic area. Based on high-accuracy evaluation models, the SHAP algorithm effectively identifies key driving factors. Moreover, the synergistic effects of multiple factors are the key mechanisms for landslide development in this region.
文献参数: 申振宏, 何松膛, 王道杰, 等.基于可解释性机器学习的九寨沟景区滑坡易发性评价及驱动力分析[J].水土保持通报,2025,45(6):213-226. Citation:Shen Zhenhong, He Songtang, Wang Daojie, et al. Landslide susceptibility evaluation and driving force analysis for Jiuzhaigou scenic area based on explainable machine learning [J]. Bulletin of Soil and Water Conservation,2025,45(6):213-226.
本研究采用受试者工作特征曲线(receiver operating characteristic curve, ROC)下面积(area under the curve, AUC)作为模型性能评估的核心指标。AUC值通过整合滑坡与非滑坡样本的预测概率分布,量化模型对滑坡与非滑坡的区分能力,其取值越接近1,表明模型对滑坡易发性的预测精度越高[6]。
YangZhengrong, XiWenfei, ShiZhengtao, et al. Study on the relationship between landslide creep and vegetation cover in mountainous areas with complex vegetation [J]. Science of Surveying and Mapping, 2023,48(1):157-165.
JiShouling, LiJinfeng, DuTianyu, et al. Survey on techniques, applications and security of machine learning interpretability [J]. Journal of Computer Research and Development, 2019,56(10):2071-2096.
HuangRunqiu. Large-scale landslides and their sliding mechanisms in China since the 20th century [J]. Chinese Journal of Rock Mechanics and Engineering, 2007,26(3):433-454.
FanLinfeng, HuRuilin, ZengFengchun, et al. Application of weighted information value model to landslide susceptibility assessment: A case study of Enshi City, Hubei Province [J]. Journal of Engineering Geology, 2012,20(4):508-513.
ZhengYingkai, ChenJianguo, WangChengbin, et al. Application of certainty factor and random forests model in landslide susceptibility evaluation in Mangshi City, Yunnan Province [J]. Bulletin of Geological Science and Technology, 2020,39(6):131-144.
XuChong, DaiFuchu, YaoXin, et al. Gis-based landslide susceptibility assessment using analytical hierarchy process in Wenchuan earthquake region [J]. Chinese Journal of Rock Mechanics and Engineering, 2009,28(S2):3978-3985.
[19]
HongHaoyuan, PourghasemiH R, PourtaghiZ S. Landslide susceptibility assessment in Lianhua County (China): A comparison between a random forest data mining technique and bivariate and multivariate statistical models [J]. Geomorphology, 2016,259:105-118.
[20]
KanungoD P, AroraM K, SarkarS, et al. A comparative study of conventional, ANN black box, fuzzy and combined neural and fuzzy weighting procedures for landslide susceptibility zonation in Darjeeling Himalayas [J]. Engineering Geology, 2006,85(3/4):347-366.
[21]
HuangFaming, YanJun, FanXuanmei, et al. Uncertainty pattern in landslide susceptibility prediction modelling:Effects of different landslide boundaries and spatial shape expressions [J]. Geoscience Frontiers, 2022,13(2):101317.
ZhangXiaoyuan, SuQiaomei, ZhaoCaisheng, et al. A landslide susceptibility evaluation method using Bayesian algorithm to optimize XGBoost [J]. Science of Surveying and Mapping, 2023,48(6):140-150.
HeZhe, ShiYuling, LiFuchun, et al. Landslide susceptibility assessment based on LightGBM model in Linxia County, Gansu Province [J]. Journal of Water Resources and Water Engineering, 2024,35(1):197-205.
ZengTaorui, WangLinfeng, ZhangYu, et al. Landslide susceptibility modeling and interpretability based on CatBoost-SHAP model [J]. The Chinese Journal of Geological Hazard and Control, 2024,35(1):37-50.
[28]
SunDeliang, WangJing, WenHaijia, et al. Landslide susceptibility mapping (LSM) based on different boosting and hyperparameter optimization algorithms:A case of Wanzhou District, China [J]. Journal of Rock Mechanics and Geotechnical Engineering, 2024,16(8):3221-3232.
WangXuan, FanXuanmei, YangFan, et al. Remote sensing interpretation method of geological hazards in lush mountainous area [J]. Geomatics and Information Science of Wuhan University, 2020,45(11):1771-1781.
LuoLuguang, PeiXiangjun, HuangRunqiu, et al. Landslide susceptibility assessment in Jiuzhaigou scenic area with GIS based on certainty factor and logistic regression model [J]. Journal of Engineering Geology, 2021,29(2):526-535.
XuChong, WangShiyuan, XuXiwei, et al. A panorama of landslides triggered by the 8 August 2017 Jiuzhaigou, Sichuan Ms7. 0 earthquake [J]. Seismology and Geology, 2018,40(1):232-260.
CaoJun, YangGeng, DuJie, et al. Study of change of the “stratified lakes and folded waterfalls” landslide in Jiuzhaigou valley scenic spots after the earthquake on Aug. 8, 2017 [J]. Acta Geologica Sichuan, 2021,41(1):161-166.
YaoXin, ZhouZhenkai, LiLingjing, et al. Insar co-seismic deformation of 2017 Ms7.0 Jiuzhaigou earthquake and discussions on seismogenic tectonics [J]. Journal of Geomechanics, 2017,23(4):507-514.
WangJin, ZangMingdong, XuChong, et al. Landslide susceptibility assessment following the 2022 Luding earthquake: A coupled analytic hierarchy process and area under the receiver operating characteristic curve algorithm [J]. Journal of Engineering Geology, 2024,32(5):1696-1711.
[41]
AkibaT, SanoS, YanaseT, et al. Optuna:a next-generation hyperparameter optimization framework [C]∥ Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. Anchorage AK USA. ACM, 2019:2623-2631.
PanWangsheng, ZhaoTianyin, YuXiulian, et al. Landslide susceptibility assessment in Huangling County based on probability ratio and support vector machine model [J]. Journal of Natural Disasters, 2024,33(4):48-59.
[44]
WangJinfeng, LiXinhu, ChristakosG, et al. Geographical detectors‐based health risk assessment and its application in the neural tube defects study of the Heshun Region, China [J]. International Journal of Geographical Information Science, 2010,24(1):107-127.
[45]
SongYongze, WangJinfeng, GeYong, et al. An optimal parameters-based geographical detector model enhances geographic characteristics of explanatory variables for spatial heterogeneity analysis: Cases with different types of spatial data [J]. GIScience & Remote Sensing, 2020,57(5):593-610.
[46]
LundbergS M, LeeS I. A unified approach to interpreting model predictions [C]∥ Proceedings of the 31st International Conference on Neural Information Processing Systems. 2017, Long Beach, California, USA. ACM, 2017:4768-4777.
[47]
LanHuijuan, WangDaojie, HeSongtang, et al. Experimental study on the effects of tree planting on slope stability [J]. Landslides, 2020,17(4):1021-1035.
[48]
AnBangsheng, ZhangZhijie, XiongShenqing, et al. Landslide susceptibility mapping based on ensemble learning in the Jiuzhaigou Region, Sichuan, China [J]. Remote Sensing, 2024,16(22):4218.
YuMingwei, GuoYonggang, SuLibin. Quantitative evaluation of influencing factors for landslide, rockfall and debris flow hazards in the Nyingchi area of Xizang Autonomous Region [J]. The Chinese Journal of Geological Hazard and Control, 2024,35(6):33-43.