Geohazard Susceptibility Assessment of Riverside Highway Zones under Multiple Feature Spaces Adaptation Network
-
摘要:
公路临水区因邻近水体、地形陡峭及频繁人类活动，常面临滑坡等地质灾害的高风险.然而，当前基于单源域的迁移学习方法面临源域与目标域在临水区特有的水文条件（如河流密度、降雨集中度）及工程活动（如公路建设）等特征差异较大时，易引发负迁移问题，难以准确预测目标区域的地质灾害易发性.运用一种基于多特征空间自适应网络（Multiple Feature Spaces Adaptation Network，MFSAN）的多源域迁移学习框架，以福建省3个公路临水地区为例，提取9个相关环境因子（含公路密度、河流密度等临水区核心特征）建立滑坡空间数据库，将安溪县（源域1）和德化县（源域2）两个易发性模型迁移至无标签样本的尤溪县（目标域）进行预测，实现多源域跨区域滑坡易发性评价.与无迁移预测模型（Non-Transferable Learning Model，NTL）以及单源域迁移预测模型（Domain Adaptive Neural Network，DANN）进行精度比对，结果显示：（1）相比于单源域地质灾害易发性迁移模型，MFSAN模型的跨区域预测精度为0.851，其准确率提高3.61%，AUC值提高1.91%，综合评估指标OA提高了9.64%；（2）通过历史滑坡验证其落入高、极高易发性区间的滑坡频率比占比最高（79.2%）；（3）MFSAN模型对临水区特有的水文-地质耦合效应捕捉能力更强，如公路3 km范围内隐患点集中现象（占比70%~83%）在预测结果中得以精准反映.可见MFSAN模型能够整合不同源域数据的空间特征和灾害发育规律，更全面捕捉区域异质性特征，为跨区域地质灾害预测提供了更优的解决方案，具备更强的泛化能力.
Abstract:Riverside highway zones are prone to high risks of geohazards such as landslides due to proximity to water bodies, steep terrain, and frequent anthropogenic activities. However, current single-source domain transfer learning methods face limitations in geohazard susceptibility prediction when significant discrepancies exist between source and target domains in hydrogeological conditions (e.g., river density, rainfall intensity) and engineering disturbances (e.g., highway construction), often leading to negative transfer issues and reduced model generalizability. This study proposes a multi-source domain transfer learning framework based on a multiple feature spaces adaptation network (MFSAN). Focusing on three riverside highway zones in Fujian Province, China, nine environmental factors (including highway density and river density as core hydrogeological features) were extracted to construct a landslide spatial database. The susceptibility models from Anxi County (source domain 1) and Dehua County (source domain 2) were transferred to Youxi County (target domain) with unlabeled samples for cross-regional landslide susceptibility evaluation. Comparative analyses were conducted against non-transferable learning models (NTL) and single-source domain adaptive models (domain adaptive neural network, DANN). The results demonstrate: (1) The MFSAN model achieved a cross-regional prediction accuracy of 0.851, outperforming single-source transfer models with improvements of 3.61% in accuracy, 1.91% in AUC, and 9.64% in overall assessment metric (OA). (2) Historical landslide validation revealed that 79.2% of landslides occurred within high-to-extreme susceptibility zones predicted by MFSAN, the highest among all models. (3) MFSAN exhibited superior capability in capturing hydrogeological coupling effects unique to riverside environments. For instance, the concentration of hazard-prone sites within 3 km of highways (70%-83%) was accurately reflected in predictions. The MFSAN framework effectively integrates spatial features and disaster development patterns from multiple source domains, comprehensively capturing regional heterogeneity and providing an optimized solution for cross-regional geohazard susceptibility prediction. This approach demonstrates enhanced generalization capability and practical value for mitigating landslide risks in complex engineering environments.
-
图 2 研究区地理信息概况及滑坡分布
a.福建省区划图；b.尤溪县研究区；c.安溪县研究区；d.德化县研究区
Fig. 2. Geographical information and landslide distribution in the study area
图 3 安溪县滑坡环境因子示意图
Fig. 3. Schematic diagrams of landslide environmental factors in Anxi County
图 4 德化县滑坡环境因子示意图
Fig. 4. Schematic diagrams of landslide environmental factors in Dehua County
图 5 尤溪县滑坡环境因子示意图
Fig. 5. Schematic diagrams of landslide environmental factors in Youxi County
图 8 源域与目标域样本余弦相似度对比
Fig. 8. Comparison of cosine similarity between source and target domain samples
图 9 源域与目标域样本分布的最大均值差异(MMD)可视化分析
Fig. 9. Visualization of maximum mean discrepancy (MMD) between source and target domain sample distributions
图 12 易发性指数分布
Fig. 12. Distribution of susceptibility index
a.NTL-KNN; b.NTL-LOG; c.NTL-BAYES; d.TCA; e.DANN; f.MFSAN
图 13 易发性分区
Fig. 13. Zonation of susceptibility
a.NTL-KNN; b.NTL-LOG; c.NTL-BAYES; d.TCA; e.DANN; f.MFSAN
表 1 地质灾害隐患点距公路分布统计表
Table 1. Distance distribution statistics of geohazard risk sites from roads
距离 安溪 安溪灾害
点占比德化 德化灾害
点占比尤溪 尤溪灾害
点占比0~1 000 1 449 47.95% 894 45.08% 897 45.33% 1 000~2 000 581 19.23% 497 25.06% 272 13.74% 2 000~3 000 465 15.39% 264 13.31% 229 11.57% 3 000~4 000 224 7.41% 162 8.17% 228 11.52% 4 000~5 000 189 6.25% 84 4.24% 139 7.02% 5 000~6 000 60 1.99% 39 1.97% 93 4.70% 6 000~7 000 33 1.09% 20 1.01% 52 2.63% > 7 000 21 0.69% 23 1.16% 69 3.49% 
下载: 导出CSV
表 2 基础数据来源信息
Table 2. Basic data sources
数据类型 分辨率 数据来源 DEM 30 m×30 m 地理空间数据云(gscloud.cn) Landsat 8 OLI_TIRS影像 全色15 m，多光谱30 m 地理空间数据云(gscloud.cn) 地质图 1︰20万 全国地质资料馆(ngac.org.cn) 地表覆盖数据 30 m×30 m 全国地理信息资源目录服务系统
(webmap.cn)地质灾害隐患点编录数据 - 遥感识别、实地调查、地质灾害点编录数据等 降雨量数据 1 km×1 km 国家地球系统科学数据中心
(geodata.cn)全国基础地理资料 1︰100万 全国地理信息资源目录服务系统(webmap.cn)
下载: 导出CSV
表 3 基于混淆矩阵的评价指标
Table 3. Evaluation metrics based on confusion matrix
指标 计算公式 解释 准确率
(Accuracy)Accuracy=$ \frac{TP+TN}{TP+TN+FP+FN} $ 正确预测的样本数占总样本数的比例 精确率
(Precision)Precision=$ \frac{TP}{TP+FP} $ 预测为正例的样本数占正例样本的比例 召回率
(Recall)Recall=$ \frac{TP}{FN+TP} $ 正确预测为正例样本占实际为正例样本的比例 F1值
(F1-score)F1=$ \frac{2\times \mathrm{P}\mathrm{r}\mathrm{e}\mathrm{c}\mathrm{i}\mathrm{s}\mathrm{i}\mathrm{o}\mathrm{n}\times \mathrm{R}\mathrm{e}\mathrm{c}\mathrm{a}\mathrm{l}\mathrm{l}}{\mathrm{P}\mathrm{r}\mathrm{e}\mathrm{c}\mathrm{i}\mathrm{s}\mathrm{i}\mathrm{o}\mathrm{n}\times \mathrm{R}\mathrm{e}\mathrm{c}\mathrm{a}\mathrm{l}\mathrm{l}} $ 精确率和召回率的调和平均数 马修斯系数
(Mattews correlation coefficient, Mcc)Mcc=$ \frac{TP\times TN-FP\times FN}{\sqrt{(TP+FP)\times \sqrt{TP+FN}\times \sqrt{TN+FP}\times \sqrt{TN+FN}}} $ 解决准确率的局限性，综合衡量二分类模型性能的指标 综合评估指标(OA) OA=Accuracy+F1+Mcc 综合考虑4个分类的结果，评估模型性能 AUC值 ROC曲线下与坐标轴围成的面积 模型对正负样本区分的能力 注：TP表示分类器正确预测为隐患点的样本数量；TN表示正确预测为非隐患点的样本数量；FP表示错误预测为隐患点的样本数量；FN表示错误预测为非隐患点的样本数量.准确率、精确率、召回率、F1值与AUC值取值范围在0至1之间，马修斯相关系数取值范围在-1至1之间，越接近于1说明模型性能越好，综合评估指标OA越大说明模型性能越好.
下载: 导出CSV
表 4 模型易发性评价指标对比表
Table 4. Comparison of susceptibility evaluation metrics
方法 Accuracy Precesion Recall F1 Mcc AUC OA NTL-KNN 0.638 0.593 0.876 0.707 0.313 0.751 1.658 NTL-LOG 0.689 0.625 0.944 0.752 0.439 0.827 1.880 NTL-Bayes 0.688 0.624 0.946 0.752 0.440 0.827 1.880 TCA 0.747 0.766 0.711 0.738 0.496 0.804 1.981 DANN 0.774 0.778 0.767 0.773 0.549 0.835 2.095 MFSAN 0.802 0.814 0.784 0.799 0.696 0.851 2.297
下载: 导出CSV
表 5 模型易发性分区频率比
Table 5. Frequency ratios of model susceptibility zonation
易发性水平 NTL-KNN NTL-LOG 面积占比(%) 隐患点占比(%) 频率比 面积占比(%) 隐患点占比(%) 频率比 极低易发区 6.50 0.96 0.15 15.33 1.41 0.09 低易发区 23.76 5.85 0.25 28.48 6.81 0.24 中易发区 24.79 20.87 0.84 25.05 12.51 0.50 高易发区 25.53 22.80 0.89 15.62 29.02 1.86 极高易发区 19.42 49.52 2.55 15.52 50.26 3.24 易发性水平 NTL-Bayes TCA 面积占比(%) 隐患点占比(%) 频率比 面积占比(%) 隐患点占比(%) 频率比 极低易发区 10.67 1.33 0.12 19.76 3.40 0.17 低易发区 25.92 5.55 0.21 31.50 8.88 0.28 中易发区 26.51 15.10 0.57 22.79 18.73 0.82 高易发区 15.11 18.28 1.21 9.13 16.58 1.82 极高易发区 21.80 59.73 2.74 16.82 52.41 3.12 易发性水平 DANN MFSAN 面积占比(%) 隐患点占比(%) 频率比 面积占比(%) 隐患点占比(%) 频率比 极低易发区 24.77 1.63 0.07 23.91 1.18 0.05 低易发区 28.80 9.18 0.32 29.38 6.37 0.22 中易发区 18.85 14.95 0.79 16.66 13.25 0.80 高易发区 12.06 24.72 2.05 17.48 35.16 2.01 极高易发区 15.52 49.52 3.19 12.57 44.04 3.50
下载: 导出CSV
-
Ai, X., 2021. Construction of Earthquake Landslide Susceptibility Assessment Model Based on Machine Learning: A Case Study of Beijing Mountainous Area(Dissertation). Institute of Engineering Mechanics, China Earthquake Administration, Harbin (in Chinese with English abstract). Ayalew, L., Yamagishi, H., 2005. The Application of GIS-Based Logistic Regression for Landslide Susceptibility Mapping in the Kakuda-Yahiko Mountains, Central Japan. Geomorphology, 65(1-2): 15-31. https://doi.org/10.1016/j.geomorph.2004.06.010 Chen, L., Ding, Y. L., Pirasteh, S., et al., 2022. Meta-Learning an Intermediate Representation for Few-Shot Prediction of Landslide Susceptibility in Large Areas. International Journal of Applied Earth Observation and Geoinformation, 110: 102807. https://doi.org/10.1016/j.jag.2022.102807 Du, J., Glade, T., Woldai, T., et al., 2020. Landslide Susceptibility Assessment Based on an Incomplete Landslide Inventory in the Jilong Valley, Tibet, Chinese Himalayas. Engineering Geology, 270: 105572. https://doi.org/10.1016/j.enggeo.2020.105572 Fu, Z. Y., Li, D. Q., Wang, S., et al., 2023. Landslide Susceptibility Assessment Based on Multitemporal Landslide Inventories and TrAdaBoost Transfer Learning. Earth Science, 48(5): 1935-1947(in Chinese with English abstract). Ghifary, M., Kleijn, W. B., Zhang, M. J., 2014. Domain Adaptive Neural Networks for Object Recognition. PRICAI 2014: Trends in Artificial Intelligence. Springer International Publishing, Cham: 898-904. https://doi.org/10.1007/978-3-319-13560-1_76 Gretton, A., Borgwardt, K. M., Rasch, M. J., et al., 2012. A Kernel Two-Sample Test. Journal of Machine Learning Research, 13(1): 723-773. https://doi.org/10.5555/2503308.2188410 Huang, F. M., Ye, Z., Yao, C., et al., 2020. Uncertainties of Landslide Susceptibility Prediction: Different Attribute Interval Divisions of Environmental Factors and Different Data-Based Models. Earth Science, 45(12): 4535-4549(in Chinese with English abstract). Huang, J. B., 2013. Preliminary Studying of Landslide Critical Rainfall in Dehua County. Geology of Fujian, 32(1): 65-69(in Chinese with English abstract). Huang, J. H., 2020. Research on Data Mining Methods and Application of Main Control Factors of Collapse, Landslide and Debris Flow (Dissertation). Chongqing University, Chongqing(in Chinese with English abstract). Li, C. L., Liu, Y. S., Lai, S. H., et al., 2024. Landslide Susceptibility Analysis Based on the Coupling Model of Logistic Regression and Support Vector Machine. Journal of Natural Disasters, 33(2): 75-86(in Chinese with English abstract). Lv, J. C., Zhang, R., Shama, A., et al., 2024. Exploring the Spatial Patterns of Landslide Susceptibility Assessment Using Interpretable Shapley method: Mechanisms of Landslide Formation in the Sichuan-Tibet Region. Journal of Environmental Management, 366: 121921. https://doi.org/10.1016/j.jenvman.2024.121921 Lyu, C. H., Cheng, J. J., Hu, Y. G., et al., 2022. Online Fault Diagnosing of Rudders Based on Multi-Source Domain Deep Transfer Learning. Journal of Ordnance Equipment Engineering, 43(9): 60-67(in Chinese with English abstract). Ma, Y. B., Li, H. R., Wang, L., et al., 2022. Application of Machine Learning Method in Landslide Susceptibility Evaluation. Journal of Civil and Environmental Engineering, 44(1): 53-67(in Chinese with English abstract). Qiao, X. X., 2006. GIS-Based Geological Hazard Risk Assessment along Complex Mountainous Highways: A Case Study of the Ankang-Ziyang Section of GZ40 Expressway(Dissertation). Changan University, Xi'an (in Chinese with English abstract). Su, X. L., 2009. Discussion on the Environmental Factors of Geologic Hazards Caused by a Creep Landslip in Youxi County. Geology of Fujian, 28(4): 335-340(in Chinese with English abstract). Su, Y., Huang, S. X., Lai, X. H., et al., 2024. Evaluation of Trans-Regional Landslide Susceptibility of Reservoir Bank Based on Transfer Component Analysis. Earth Science, 49(5): 1636-1653(in Chinese with English abstract). Sun, D. L., 2019. Study on Landslide Susceptibility Zoning and Rainfall-Induced Landslide Prediction and Early Warning Based on Machine Learning(Dissertation). East China Normal University, Shanghai (in Chinese with English abstract). Sun, D. L., Gu, Q. Y., Wen, H. J., et al., 2023. Assessment of Landslide Susceptibility along Mountain Highways Based on Different Machine Learning Algorithms and Mapping Units by Hybrid Factors Screening and Sample Optimization. Gondwana Research, 123: 89-106. https://doi.org/10.1016/j.gr.2022.07.013 Sun, X. L., Ji, W. D., Wang, X., 2022. Natural Computation Method Based on Cosine Similarity Opposition Strategy. Information and Control, 51(6): 708-718(in Chinese with English abstract). Wang, H. J., Wang, L., Zhang, L. M., 2023. Transfer Learning Improves Landslide Susceptibility Assessment. Gondwana Research, 123: 238-254. https://doi.org/10.1016/j.gr.2022.07.008 Wang, Z. H., Goetz, J., Brenning, A., 2022. Transfer Learning for Landslide Susceptibility Modeling Using Domain Adaptation and Case-Based Reasoning. Geoscientific Model Development, 15(23): 8765-8784. https://doi.org/10.5194/gmd-15-8765-2022 Wattenberg, M., Viégas, F., Johnson, I., 2016. How to Use T-SNE Effectively. Distill, 1(10): e2. https://doi.org/10.23915/distill.00002 Wu, L. Y., Zeng, T. R., Liu, X. P., et al., 2024. Landslide Susceptibility Assessment Based on Ensemble Learning Modeling. Earth Science, 49(10): 3841-3854(in Chinese with English abstract). Wu, R. Z., Hu, X. D., Mei, H. B., et al., 2021. Spatial Susceptibility Assessment of Landslides Based on Random Forest: A Case Study from Hubei Section in the Three Gorges Reservoir Area. Earth Science, 46(1): 321-330(in Chinese with English abstract). Xue, M. M., 2020. Evaluation Model of Random Forest and Support Vector Machine for Landslide Prone along Mountain Road (Dissertation). Chongqing University, Chongqing (in Chinese with English abstract). Yang, S. K., Kong, X. G., Wang, Q. B., et al., 2022. Mechanical Fault Diagnosis Based on Multi-Source Domain Deep Transfer Learning. Journal of Vibration and Shock, 41(9): 32-40(in Chinese with English abstract). Yao, J. Y., Qin, S. W., Qiao, S. S., et al., 2022. Application of a Two-Step Sampling Strategy Based on Deep Neural Network for Landslide Susceptibility Mapping. Bulletin of Engineering Geology and the Environment, 81(4): 148. https://doi.org/10.1007/s10064-022-02615-0 Ye, L. Z., 2011. Characteristics of Geo-Hazards and Their Influence Factors in Anxi, Fujian Province. Journal of Geological Hazards and Environment Preservation, 22(2): 46–49(in Chinese with English abstract). Zhu, Y. C., Zhuang, F. Z., Wang, D. Q., 2019. Aligning Domain-Specific Distribution and Classifier for Cross-Domain Classification from Multiple Sources. In: Proceedings of the 33rd AAAI Conference on Artificial Intelligence. AAAI Press, Honolulu, Hawaii, USA, 5989-5996. https://doi.org/10.1609/aaai.v33i01.33015989 .艾骁, 2021. 基于机器学习的地震滑坡易发性评估模型构建: 以北京市山区为例(博士学位论文). 哈尔滨: 中国地震局工程力学研究所. 付智勇, 李典庆, 王顺, 等, 2023. 基于多时空滑坡编录和TrAdaBoost迁移学习的滑坡易发性评价. 地球科学, 48(5): 1935-1947. doi: 10.3799/dqkx.2023.013 黄发明, 叶舟, 姚池, 等, 2020. 滑坡易发性预测不确定性: 环境因子不同属性区间划分和不同数据驱动模型的影响. 地球科学, 45(12): 4535-4549. doi: 10.3799/dqkx.2020.247 黄健豪, 2020. 崩滑流灾害主控因子数据挖掘方法与应用研究(硕士学位论文). 重庆: 重庆大学. 黄俊宝, 2013. 德化县滑坡成灾临界降雨量研究. 福建地质, 32(1): 65-69. 李成林, 刘严松, 赖思翰, 等, 2024. 基于逻辑回归和支持向量机耦合模型的滑坡易发性分析. 自然灾害学报, 33(2): 75-86. 吕丞辉, 程进军, 胡阳光, 等, 2022. 基于多源域深度迁移学习的舵机在线故障诊断. 兵器装备工程学报, 43(9): 60-67. 马彦彬, 李红蕊, 王林, 等, 2022. 机器学习方法在滑坡易发性评价中的应用. 土木与环境工程学报(中英文), 44(1): 53-67. 乔晓霞, 2006. 基于GIS的复杂山区高速公路沿线地质灾害危险性评价研究: 以GZ40高速公路安康至紫阳山区段为例(硕士学位论文). 西安: 长安大学. 苏兴来, 2009. 尤溪县蠕动滑坡地质灾害环境因素探讨. 福建地质, 28(4): 335-340. 苏燕, 黄绍翔, 赖晓鹤, 等, 2024. 基于迁移成分分析的库岸跨区域滑坡易发性评价. 地球科学, 49(5): 1636-1653. doi: 10.3799/dqkx.2022.453 孙德亮, 2019. 基于机器学习的滑坡易发性区划与降雨诱发滑坡预报预警研究(博士学位论文). 上海: 华东师范大学. 孙小琳, 季伟东, 王旭, 2022. 基于余弦相似度反向策略的自然计算方法. 信息与控制, 51(6): 708-718. 邬礼扬, 曾韬睿, 刘谢攀, 等, 2024. 基于集成学习建模的滑坡易发性评价. 地球科学, 49(10): 3841-3854. doi: 10.3799/dqkx.2022.451 吴润泽, 胡旭东, 梅红波, 等, 2021. 基于随机森林的滑坡空间易发性评价: 以三峡库区湖北段为例. 地球科学, 46(1): 321-330. doi: 10.3799/dqkx.2020.032 薛蒙蒙, 2020. 山区道路沿线滑坡易发性随机森林和支持向量机评价模型(硕士学位论文). 重庆: 重庆大学. 杨胜康, 孔宪光, 王奇斌, 等, 2022. 基于多源域深度迁移学习的机械故障诊断. 振动与冲击, 41(9): 32-40. 叶龙珍, 2011. 福建省安溪县地质灾害发育特征及影响因素分析. 地质灾害与环境保护, 22(2): 46-49. -
点击查看大图
计量
- 文章访问数: 65
- HTML全文浏览量: 8
- PDF下载量: 5
- 被引次数: 0
