Optimizing Landslide Prevention Efficacy Based on Environmental Factor Connection Method and Model Selection:A Case of Hanbin District in Ankang,Shaanxi
1.Shaanxi Nuclear Industry Engineering Survey Institute Co. ,Ltd,Xi’an,Shaanxi 710054,China
2.School of Engineering and Technology,China University of Geosciences (Beijing),Beijing 100083,China
Show less
文章历史+
Received
Published
2025-06-23
2026-04-25
Issue Date
2026-06-11
PDF (16657K)
摘要
滑坡易发性评价(LSA)是区域地质灾害风险防控的关键环节。为优化资源配置并提升预警效率,本研究探究了环境因子联接方式与评价模型对LSA的影响规律,以期为提高风险防控效能提供理论依据。以陕西省安康市汉滨区主城区为例,基于260处滑坡及10种环境因子数据,采用原始输入、滑坡密度、频率比(FR)、信息量(IV)、确定性系数(CF)5种联接方式,分别与随机森林(RF)、人工神经网络(ANN)、极端梯度提升(XGBoost)3种机器学习模型耦合,构建15种组合。同时将FR、IV、CF的预测结果作为对比基准,采用分区统计、ROC曲线对上述18种结果进行检验。结果表明:(1)环境因子联接方式对“基于树”的模型(RF、XGBoost)无明显影响;(2)“函数拟合型”模型 (ANN)对联接方式较为敏感,其精度取决于联接方式的优劣;(3)机器学习模型整体优于统计学模型,模型整体性能表现为:RF > XGBoost > ANN > FR = IV > CF。因此,从防控实践出发,在难以确定最佳联接方式时优先采用RF模型,可保证评价精度的同时,有效提升防控资源的靶向性。
Abstract
Landslide Susceptibility Assessment (LSA) plays a crucial role in regional geological hazard risk prevention and control. To optimize resource allocation and enhance early warning efficiency, this study investigates the influence of environmental factor integration methods and evaluation models on LSA performance, with the aim of providing a theoretical basis for improving risk management effectiveness. Focusing on the main urban area of Hanbin District, Ankang City, Shaanxi Province, we utilized data from 260 historical landslides and ten environmental factors. Five integration methods, namely Original Input, Landslide Density, Frequency Ratio (FR), Information Value (IV), and Certainty Factor (CF), were coupled with three machine learning models: Random Forest (RF), Artificial Neural Network (ANN), and Extreme Gradient Boosting (XGBoost), resulting in 15 combined approaches. The predictive results of the FR, IV, and CF models alone were also included as benchmark comparisons. All 18 susceptibility results were evaluated using zonal statistics and ROC curve analysis. The findings indicate that: (1) The choice of connection method has no significant impact on tree-based models (RF and XGBoost); (2) The function-fitting model (ANN) is sensitive to the connection method, and its accuracy depends on the appropriateness of the method selected; (3) Machine learning models generally outperform statistical models, with overall performance ranked as: RF > XGBoost > ANN > FR = IV > CF. Therefore, from a practical prevention and control perspective, adopting the RF model when the optimal integration method is uncertain can ensure assessment accuracy while effectively improving the targeting of prevention resources.
本研究使用的数据主要包括环境因子与滑坡清单两类(表1)。环境因子数据源自多源平台:30 m分辨率DEM用于提取地形地貌因子;工程地质岩组与断裂信息从1∶5万地质图中提取;土地利用与归一化植被系数(NDVI)数据分别来自杨杰等[21]发布的中国30 m土地利用数据集与资源环境科学数据平台;基于Open Street Map(2021年)道路数据与第三次全国国土调查(2019年)河流数据,通过欧氏距离算法生成距道路与河流距离因子。
LIM X, WANGH Y, CHENJ L, et al. Assessing landslide susceptibility based on the random forest model and multi-source heterogeneous data[J]. Ecological Indicators, 2024, 158: 111600.
XUC, XUX W, YAOX, et al. Three (nearly) complete inventories of landslides triggered by the May 12, 2008 Wenchuan Mw 7.9 earthquake of China and their spatial distribution statistical analysis[J]. Landslides, 2014, 11: 441 - 461.
CUTLERA, CUTLERD R, STEVENSJ R. Random forests[J]. Ensemble machine learning: Methods and applications, 2012: 157 - 175.
[31]
李贞贵. 随机森林改进的若干研究[D]. 厦门:厦门大学, 2013.
[32]
SAHAS, MAJUMDARP, BERAB. Deep learning and benchmark machine learning based landslide susceptibility investigation, Garhwal Himalaya (India)[J]. Quaternary Science Advances, 2023, 10: 100075.
[33]
LIUY R, MENGZ Q, ZHUL, et al. Optimizing the sample selection of machine learning models for landslide susceptibility prediction using information value models in the Dabie mountain area of Anhui, China[J]. Sustainability, 2023, 15(3): 1971.
AYALEWL, YAMAGISHIH. The application of GIS-based logistic regression for landslide susceptibility mapping in the Kakuda-Yahiko Mountains, Central Japan[J]. Geomorphology, 2005, 65(1/2): 15 - 31.
KEC Y, HES, QINY G. Comparison of natural breaks method and frequency ratio dividing attribute intervals for landslide susceptibility mapping[J]. Bulletin of Engineering geology and the Environment, 2023, 82(10): 384.
ABDELKADERM M, DAOUDA M A, ABDEL-GAWADA G, et al. Improving Landslide Susceptibility Mapping: The Role of Non-Landslide Sample Selection in Machine Learning Models[J]. Remote Sensing Applications: Society and Environment, 2025: 101759.
[53]
LIUQ, TANGA P, HUANGD L. Exploring the uncertainty of landslide susceptibility assessment caused by the number of non-landslides[J]. Catena, 2023, 227: 107109.
[54]
HUANGF M, YEZ, JIANGS H, et al. Uncertainty study of landslide susceptibility prediction considering the different attribute interval numbers of environmental factors and different data-based models[J]. Catena, 2021, 202: 105250.