1 Hainan Provincial Key Laboratory of Marine Geology Resources and Environment, Haikou 570206, China
2 Hainan Provincial Geological Survey Institute, Haikou 570206, China
Show less
文章历史+
Received
Published
Revised
2025-05-07
2026-04-08
2026-01-05
Online Date
2026-05-28
PDF (28176K)
摘要
为科学评估热带县域地质灾害易发性,选取地貌特征、地质环境、人类工程活动及自然环境的4类指标12个评价因子,并采用频率比(frequency ratio,FR)将评价因子定量化后作为输入变量,与随机森林(random fores,RF)、逻辑回归(logistic regression,LR)、支持向量机(support vector machin,SVM)进行耦合,构建FR-RF模型、FR-LR模型和FR-SVM模型,对海南省琼中县进行地质灾害易发性评价。结果表明:研究区地质灾害高易发区主要集中于地表起伏剧烈、水力侵蚀显著的山地及沟谷地带;三类模型的曲线下面积(area under curve,AUC)值分别为0.95(FR-SVM模型)、0.94(FR-RF模型)、0.79(FR-LR模型),均高于0.7,其中FR-SVM和FR-RF模型预测性能更优;综合精确率、准确率、召回率及F1分数,FR-RF模型在易发性评价中表现最佳。研究将有效提升小样本条件下的地质灾害预测精度,频率比模型与机器学习算法的融合能够有效提升小样本条件下的地质灾害预测精度,为琼中县及类似区域的灾害防治与规划决策提供了可靠的技术支撑。
Abstract
In order to scientifically assess the susceptibility to geological disasters in tropical counties, 12 evaluation factors covering topographic features, geological conditions, human engineering activities, and natural environments were selected in this study. FR (frequency ratio) method was applied to quantify these factors, and the quantified factors were used as input variables to be integrated with three machine learning models, namely RF (random forest), LR (logistic regression), and SVM (support vector machine). Through this integration, three hybrid models (FR-RF, FR-LR, and FR-SVM) were developed. These models were employed to evaluate the geological disaster susceptibility in Qiongzhong County, Hainan Province.The results show that high-susceptibility zones are mainly distributed in mountainous and gully areas, which are characterized by intense topographic relief and significant hydraulic erosion. AUC (area under curve) values of the three models are 0.95 for FR-SVM, 0.94 for FR-RF, and 0.79 for FR-LR, all of which exceed 0.7. Among them, the FR-SVM and FR-RF models exhibit superior predictive performance. Further comparative analysis based on precision, accuracy, recall, and F1 score demonstrates that the FR-RF model achieves the best overall performance. It is concluded that this study effectively improves the accuracy of geological disaster prediction under small-sample conditions. The integration of the frequency ratio method with machine learning algorithms provides a reliable technical basis for disaster prevention, mitigation, and planning in Qiongzhong County, as well as in other similar tropical regions.
LI Xin, XUE Gui-cheng, LIU Chang-zhu, YANG Yong-peng, YU Hao.
Geological Hazard Susceptibility Assessment Based on Small Sample Data and Frequency Ratio-machine Learning Algorithms: Qiongzhong County[J].
Science Technology and Engineering, 2026, 26(10): 4088-4100 DOI:10.12404/j.issn.1671-1815.2503279
地形地貌因子在不同的分类区间对滑坡等地质灾害的发生有着不同程度的影响,是决定滑坡灾害分布的基础条件,通常地表起伏大、破碎程度高的区域极易发育滑坡[31]。如表3所示,坡度在5°~15°时,灾害数比达0.305,频率比为1.102,表明该坡度区间是滑坡等灾害的相对高发区域;而25°~45°坡度区间频率比较低,说明高陡坡并非地质灾害发生的唯一高风险区域,可能还受到岩土体性质、人类工程扰动、植被覆盖等其他因素影响。高程对地表结构与区域气候参数有显著影响[32],高程在200~400 m 时,灾害数比高达0.649,频率比为1.449,表明该高程区间是地质灾害的主要分布区域;而600 m以上高程区域灾害数比和频率比显著降低,说明高程过高可能由于岩土体风化程度、人类活动影响等因素,反而降低了灾害发生的可能性。
起伏度反映了地表的高低变化程度,对地表径流、土壤侵蚀以及岩土体稳定性都有着重要影响[33];当起伏度在 0~5 m 时,频率比高达82.667,这可能是因为该区域地势相对低洼,容易汇聚地表水,增加了岩土体的含水量,降低了其抗剪强度,从而导致滑坡等灾害发生的频率大幅提高。剖面曲率和平面曲率则从微观角度刻画了地表的弯曲特征[34]。在剖面曲率中,1.95~7.05区间频率比为1.450;平面曲率中,-3.25~1.20区间频率比为1.408,说明特定的曲率变化可能会改变岩土体的应力分布,从而增加滑坡等灾害发生的概率。
相比之下,林地的频率比为0.956,接近1,说明林地对地质环境具有一定的保护作用,有利于维持土体稳定。公路距离是影响地质灾害发生的一个重要因素[38]。如表5所示,在0~200 m 范围内,频率比为2.126,远高于其他距离区间,说明靠近公路的区域地质灾害发生概率显著增加。这可能是因为公路建设过程中的开挖、填方等活动破坏了山体的稳定性,同时公路沿线的人类活动频繁,也会对地质环境造成进一步的扰动。随着距公路距离的增加,频率比逐渐降低,当距离大于1 000 m时,频率比为0,表明该区域受公路建设和人类活动的影响较小,地质灾害发生的可能性极低。
4.2.4 自然环境因子
NDVI指数反映了植被覆盖情况,对地质灾害的发生也有显著影响[39]。如表6所示,在NDVI介于-0.12~0时,频率比达到4.364,说明植被稀少的区域更易发生滑坡等灾害。而NDVI介于0.12~0.24时,频率比为0.889,显示出相对较好的植被覆盖能够降低灾害风险。河流缓冲距离同样对地质灾害发生概率有显著影响[32]。如表6所示,在0~200 m 范围内,频率比为1.888,相对较高,说明靠近水系的区域更容易发生地质灾害。这可能是由于水系的侵蚀作用会削弱山体的稳定性,同时地下水的活动也会对土体产生软化作用,增加滑坡等地质灾害的发生风险。随着距水系距离的增加,频率比总体呈下降趋势,但在大于1 000 m 时,频率比又有所上升,可能是该区域存在其他特殊的地质或水文条件影响了地质灾害的发生。
WuChangrun, JueYuanmei, WangJinliang, et al. Landslide susceptibility assessment in Shuangbai County based on frequency ratio-logistic regression coupled model[J]. Journal of Natural Disasters, 2021, 30(4): 213-224.
YinHao, ChengChao, YanEchuan, et al. Geological hazard modeling in Macheng City based on I + LR method and GIS[J]. Science Technology and Engineering, 2024, 24 (23): 9766-9778.
CunDexin, LinghuChangwei, MaYiqi, et al. Evaluation of susceptibility to geological hazardin in Fuyuan County based on GIS and weighted information model[J]. Science Technology and Engineering, 2024, 24(18): 7563-7573.
ChenYi, FanXuanmei. Post-earthquake geological hazard susceptibility evaluation: a case study of Yingxiu seismic area[J]. Science Technology and Engineering, 2020, 20(9): 3516-3527.
BaoJunfan, ChenJie, YangWentao, et al. Susceptibility identification of loess geological hazards in Kangdian Town, Gongyi City, western Henan Province by using interpretable machine learning models[J]. Science Technology and Engineering, 2025, 25(15): 6200-6219.
LanYingying, GuoChangcheng, ZhuYunfu. A review of methods for geological hazard susceptibility assessment[J]. Geology and Resources, 2024, 33(1): 65-73.
LiuYajing, LiuHongjian. Geological hazard susceptibility assessment in seismic zones based on information quantity-random forest model: a case study of Songpan-Jiaochang seismic zone[J]. Science Technology and Engineering, 2024, 24(1): 143-154.
LiuShaoze, CuiMeijuan, FuXiaoyi, et al. Geological hazard susceptibility assessment using randomforest considering buffer zone optimization: a case study of Fuyang District, Hangzhou, Zhejiang Province[J]. Science Technology and Engineering, 2025, 25(15): 6220-6229.
RuMan, ZhengYan, ZhangBin, et al. Evaluation of geological disaster susceptibility based on SVM-RF model: a case study of Qingtianhe Scenic Area in Bo’ai County, Henan Province[J]. Geology and Resources, 2023, 32(5): 633-641.
LiXin, XueGuicheng, LiuChangzhu, et al. Geological hazard susceptibility assessment in Changjiang County of Hainan based on IV, CF, LRIV and LRCF models[J]. China Earthquake Engineering Journal, 2025(2): 319-330, 341.
LiXin, XueGuicheng, LiuChangzhu, et al. Geological hazard susceptibility assessment in central mountainous area of Hainan Island based on information quantity model and information quantity-logistic regression model[J]. Journal of Geomechanics, 2022, 28(2): 294-305.
LiXin, XueGuicheng, XiaNan, et al. Geological hazard susceptibility assessment in national tropical rainforest parks based on CF, CF-LR and CF-AHP models: a case study of Baoting, Hainan[J]. Geoscience, 2023, 37(4): 1033-1043.
[25]
AdnanM S G, RahmanM S, AhmedN, et al. Improving spatial agreement in machine learning-based landslide susceptibility mapping[J]. Remote Sensing, 2020, 12(20). DOI: 10.3390/rs12203347.
[26]
TrigilaA, IadanzaC, EspositoC, et al. Comparison of logistic regression and random forests techniques for shallow landslide susceptibility assessment in Giampilieri (NE Sicily, Italy)[J]. Geomorphology, 2015, 249: 119-136.
HuangMin. Comparison of multiple machine learning models based on FR in the evaluation of geological disaster susceptibility[J]. Geology of Fujian, 2023, 42(3): 236-243.
[29]
UstaZ, AkıncıH, AkınA T. Comparison of tree-based ensemble learning algorithms for landslide susceptibility mapping in murgul (artvin), turkey[J]. Earth Science Informatics, 2024, 17(2): 1459-1481.
[30]
TruongX L, MitamuraM, KonoY, et al. Enhancing prediction performance of landslide susceptibility model using hybrid machine learning approach of bagging ensemble and logistic model tree[J]. Applied Sciences, 2018, 8(7). DOI: 10.3390/app8071046.
ZhouPing, DengHui, ZhangWenjiang, et al. Landslide susceptibility assessment based on information quantity model and machine learning methods: a case study of Lixian County, Sichuan[J]. Scientia Geographica Sinica, 2022, 42(9): 1665-1675.
WuXueling, YangJingyu, NiuRuiqing. A landslide susceptibility assessment method combining SMOTE and convolutional neural network[J]. Geomatics and Information Science of Wuhan University, 2020, 45(8): 1223-1232.
LiWenbin, FanXuanmei, HuangFaming, et al. Uncertainty in landslide susceptibility modeling based on connection of different environmental factors and prediction models[J]. Earth Science, 2021, 46(10): 3777-3795.
[40]
HuangF, ZhangJ, ZhouC, et al. A deep learning algorithm using a fully connected sparse autoencoder neural network for landslide susceptibility prediction[J]. Landslides, 2020, 17(1): 217-229.
MaoHuarui, SunXiaofei, ZhouYingzhi. Mapping landslide susceptibility of the Three Gorges Reservoir Area in northeast Chongqing using the frequency ratio and projection pursuit model[J]. Science Technology and Engineering, 2022, 22(5): 1803-1813.
[43]
AditianA, KubotaT, ShinoharaY. Comparison of GIS-based landslide susceptibility models using frequency ratio, logistic regression, and artificial neural network in a tertiary region of Ambon, Indonesia[J]. Geomorphology, 2018, 318: 101-111.
[44]
DaiB, GuC, ZhaoE, et al. Statistical model optimized random forest regression model for concrete dam deformation monitoring[J]. Structural Control and Health Monitoring, 2018, 25(6). DOI: 10.1002/stc.2170.
YangKang, XueXicheng, LiShibo. Geological hazard susceptibility assessment based on information quantity integrated into GA-optimized SVM model[J]. Safety and Environmental Engineering, 2022, 29(3): 109-118.
LiYuanyuan, MeiHongbo, RenXiaojie, et al. Geological hazard susceptibility assessment based on certainty factor and support vector machine[J]. Journal of Geo-Information Science, 2018, 20(12): 1699-1709.
NiuRuiqing, PengLing, YeRunqing, et al. Landslide susceptibility assessment based on rough set and support vector machine[J]. Journal of Jilin University (Earth Science Edition), 2012, 42(2): 430-439.
LiChenglin, LiuYansong, LaiSihan, et al. Landslide susceptibility analysis based on coupling model of logistic regression and support vector machine[J]. Journal of Natural Disasters, 2024, 33(2): 75-86.
KongJiaxu, ZhuangJianqi, PengJianbing, et al. Landslide susceptibility assessment in the Loess Plateau based on information quantity and convolutional neural network[J]. Earth Science, 2023, 48(5): 1711-1729.
ZhangXiaodong, LiuXiangnan, ZhaoZhipeng, et al. Comparative study on combined models of information quantity, certainty factor and logistic regression for geological hazard susceptibility assessment[J]. Geoscience, 2018, 32(3): 602-610.
HuXiuyu, QinShengwu, DouQiang, et al. Debris flow susceptibility analysis based on GIS and random forest model: a case study of northern mountainous area in Taonan City, Jilin Province[J]. Bulletin of Soil and Water Conservation, 2019, 39(5): 204-210, 217, 2.
WuRunze, HuXudong, MeiHongbo, et al. Spatial landslide susceptibility assessment based on random forest: a case study of Hubei section in the Three Gorges Reservoir Area[J]. Earth Science, 2021, 46(1): 321-330.
[63]
ZhangW, LiuS, WangL, et al. Landslide susceptibility research combining qualitative analysis and quantitative evaluation: a case study of Yunyang County in Chongqing, China[J]. Forests, 2022, 13(7): 1-11.
[64]
WuY, KeY, ChenZ, et al. Application of alternating decision tree with AdaBoost and bagging ensembles for landslide susceptibility mapping[J]. CATENA, 2020, 187.DOI: 10.1016/j.catena.2019.104396.
ZhangQikai, LingSixiang, LiXiaoning, et al. Comparative study on rapid assessment models for landslide susceptibility in Jiuzhaigou County[J]. Chinese Journal of Rock Mechanics and Engineering, 2020, 39(8): 1595-1610.
[67]
Chowdhury MdS, Rahman MdN, Sheikh MdS, et al. GIS-based landslide susceptibility mapping using logistic regression, random forest and decision and regression tree models in chattogram district, bangladesh[J]. Heliyon, 2024, 10(1). DOI: 10.1016/j.heliyon.2023.e23424.
WangXuedong, ZhangChaobiao, WangCui, et al. Geological hazard susceptibility assessment in Helong City based on logistic regression and random forest[J]. Journal of Jilin University (Earth Science Edition), 2022, 52(6): 1957-1970.