中巴公路盖孜河谷段形变监测与滑坡易发性评价方法

doi:10.12118/j.issn.1000-6060.2025.641

摘要/Abstract

摘要：

滑坡易发性制图（LSM）是滑坡灾害风险评估的重要环节。针对现有模型在多尺度特征提取方面的不足，提出一种融合多尺度注意力深度可分离卷积与通道优先Transformer的特征提取网络（MADTNet）以充分提取滑坡的局部和全局特征，提升LSM精度。选取中巴公路盖孜河谷段为研究区，利用小基线集干涉合成孔径雷达（SBAS-InSAR）技术获取2019年1月—2024年6月的地表形变作为动态因子，并结合地形地貌、地质条件、水文特征和人类活动等10个静态因子构建滑坡易发性评价模型，并与卷积神经网络（CNN）、残差网络（ResNet）、视觉Transformer（ViT）、FrIT和TransUNet 5种深度学习模型进行对比分析。结果表明：（1）盖孜河谷段雷达视线向形变速率为-52.5~19.4 mm·a^-1，斜坡向形变速率为-70.6~25.7 mm·a^-1。（2） MADTNet在滑坡易发性评价中表现最佳，受试者工作特征曲线下面积（AUC）值达0.959，多项评估指标优于其他模型，具有较好的滑坡预测能力和泛化能力。（3）沙普利加性解释方法（SHAP）分析显示，距河流距离、距道路距离、坡度和高差是影响滑坡发育的主要因子。MADTNet具有较高的预测精度，模型可解释性强，可为滑坡灾害的精准防治提供科学依据。

关键词: 喀喇昆仑公路, InSAR, CNN, Transformer, 滑坡易发性评价, SHAP分析

Abstract:

Landslide susceptibility mapping (LSM) is a critical component of landslide hazard risk assessment. To address the limitations of existing models in multi-scale feature extraction, a hybrid feature extraction network, termed multi-scale attention depthwise separable convolution with a channel-prioritized Transformer network (MADTNet) was developed. This approach effectively captures both local and global landslide features, enabling more accurate LSM. The Gaizi River Valley section of the China-Pakistan highway was selected as the study area. To construct a comprehensive dataset for modeling landslide susceptibility, surface deformation data collected from January 2019 to June 2024 were retrieved using a small baseline subset interferometric synthetic aperture radar (SBAS-InSAR) technique and incorporated as a dynamic factor, along with ten static factors, including topography, geological conditions, hydrological characteristics, and human activities. The performance of MADTNet was systematically compared with that of five benchmark deep learning models convolutional neural network (CNN), residual network (ResNet), vision Transformer (ViT), fractal-based information Transformer (FrIT), and Transformer-based U-net (TransUNet). The results indicate that: (1) The line-of-sight deformation rates in the study area range from -52.5 mm·a^-1 to 19.4 mm·a^-1, while the slope-direction deformation rates range from -70.6 mm·a^-1 to 25.7 mm·a^-1. (2) MADTNet achieves the best overall performance, with an area-under-the-curve of 0.959, demonstrating superior predictive accuracy and generalization capability compared to the other models. (3) Shapley additive explanations (SHAP) analysis reveals that the distance to rivers, distance to roads, slope, and elevation difference are the dominant factors influencing landslide development. Overall, MADTNet achieves high-precision and interpretable landslide susceptibility predictions, offering robust scientific support for targeted landslide mitigation and disaster risk management.

Key words: Karakoram highway, InSAR, CNN, Transformer, landslide susceptibility evaluation, SHAP analysis

王院荣, 火久元, 张耀南. 中巴公路盖孜河谷段形变监测与滑坡易发性评价方法[J]. 干旱区地理, 2026, 49(9): 1788-1802.

WANG Yuanrong, HUO Jiuyuan, ZHANG Yaonan. Deformation monitoring and landslide susceptibility assessment method for the Gaizi River Valley of the China-Pakistan highway[J]. Arid Land Geography, 2026, 49(9): 1788-1802.

图/表 18

图1

表1

图2

图3

图4

表2

图5

表3

图6

图7

图8

图9

表4

图10

图11

表5

表6

图12

参考文献 25

[1]	Li M H, Zhang L, Ding C, et al. Retrieval of historical surface displacements of the Baige landslide from time-series SAR observations for retrospective analysis of the collapse event[J]. Remote Sensing of Environment, 2020, 240: 111695, doi: 10.1016/j.rse.2020.111695.
[2]	张志军, 李有三, 祁万强, 等. 基于GF-2数据的G314沿线奥-塔段地质灾害遥感调查[J]. 合肥工业大学学报(自然科学版), 2021, 44(9): 1249-1257.
	[Zhang Zhijun, Li Yousan, Qi Wanqiang, et al. Remote sensing survey of geological hazards along the Aotai section of G314 based on GF-2 data[J]. Journal of Hefei University of Technology (Natural Science Edition), 2021, 44(9): 1249-1257.]
[3]	汪钰晴, 唐伶俐, 王新鸿, 等. 基于时序InSAR技术的中巴走廊盖孜河谷段地表形变分析[J]. 中国科学院大学学报, 2022, 39(3): 343-351. doi: 10.7523/j.ucas.2020.0046
	[Wang Yuqing, Tang Lingli, Wang Xinhong, et al. Surface deformation analysis of the Gaizi Valley section of the China-Pakistan corridor based on time-series InSAR technology[J]. Journal of University of Chinese Academy of Sciences, 2022, 39(3): 343-351.] doi: 10.7523/j.ucas.2020.0046
[4]	余文秀, 李秀珍, 郑玲静, 等. 基于SBAS-InSAR技术的中巴公路KKH沿线潜在滑坡区识别研究[J]. 工程地质学报, 2024, 32(5): 1597-1606.
	[Yu Wenxiu, Li Xiuzhen, Zheng Lingjing, et al. Identification of potential landslide areas along the China-Pakistan Karakoram highway based on SBAS-InSAR technology[J]. Journal of Engineering Geology, 2024, 32(5): 1597-1606.]
[5]	Hussain S, Pan B, Afzal Z, et al. Landslide detection and inventory updating using the time-series InSAR approach along the Karakoram highway, northern Pakistan[J]. Scientific Reports, 2023, 13(1): 7485, doi: 10.1038/s41598-023-34030-0.
[6]	Zheng D F, Li Y Y, Yan C L, et al. Landslide susceptibility assessment using AutoML-SHAP method in the southern foothills of Changbai Mountain, China[J]. Landslides, 2025, 22(6): 1855-1875. doi: 10.1007/s10346-025-02462-6
[7]	Su X J, Zhang Y, Meng X M, et al. An integrated landslide susceptibility assessment in the Karakoram Mountains based on SBAS-InSAR and machine learning: A case study of the Hunza Valley[J]. Bulletin of Engineering Geology and the Environment, 2025, 84(6): 280, doi: 10.1007/s10064-025-04299-8.
[8]	Zhang M M, Zhao X D, Li W, et al. Cross-scene joint classification of multisource data with multilevel domain adaptation network[J]. IEEE Transactions on Neural Networks and Learning Systems, 2023, 35(8): 11514-11526. doi: 10.1109/TNNLS.2023.3262599
[9]	Liu X Y, Wu Y, Liang W K, et al. High resolution SAR image classification using global-local network structure based on vision transformer and CNN[J]. IEEE Geoscience and Remote Sensing Letters, 2022, 19: 1-5.
[10]	Yu Y, Wang J H, Pedrycz W, et al. Multi-level information fusion transformer with background filter for fine-grained image recognition[J]. Applied Intelligence, 2024, 54: 8108-8119. doi: 10.1007/s10489-024-05584-x
[11]	Zhang Q, He Y, Zhang Y L, et al. A graph-transformer method for landslide susceptibility mapping[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2024, 17: 14556-14574. doi: 10.1109/JSTARS.2024.3437751
[12]	Wang S, Wu W B, Zheng Z Q, et al. CTST: CNN and transformer-based spatio-temporally synchronized network for remote sensing change detection[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2024, 17: 16272-16288. doi: 10.1109/JSTARS.2024.3455261
[13]	Zhao Z Y, Chen T, Dou J, et al. Landslide susceptibility mapping considering landslide local-global features based on CNN and transformer[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2024, 17: 7475-7489. doi: 10.1109/JSTARS.2024.3379350
[14]	蒋磊, 刘小龙, 郭帅, 等. 基于Logistic回归分析的土壤盐渍化易发性评价——以新疆南疆塔里木灌区为例[J]. 干旱区地理, 2023, 46(11): 1858-1867. doi: 10.12118/j.issn.1000-6060.2023.073
	[Jiang Lei, Liu Xiaolong, Guo Shuai, et al. Evaluation of soil salinization susceptibility based on Logistic regression analysis: A case of Tarim irrigation area in southern Xinjiang[J]. Arid Land Geography, 2023, 46(11): 1858-1867.] doi: 10.12118/j.issn.1000-6060.2023.073
[15]	朱颖彦, 潘军宇, 李朝月, 等. 中巴喀喇昆仑公路冰川泥石流[J]. 山地学报, 2022, 40(1): 71-83.
	[Zhu Yingyan, Pan Junyu, Li Chaoyue, et al. Glacial debris flows along the China-Pakistan Karakoram highway[J]. Journal of Mountain Science, 2022, 40(1): 71-83.]
[16]	张世昊, 陶秋香, 刘国林, 等. 联合时序SBAS-InSAR与GACOS的青岛胶州湾大桥形变监测与分析[J]. 科学技术与工程, 2024, 24(5): 2074-2082.
	[Zhang Shihao, Tao Qiuxiang, Liu Guolin, et al. Deformation monitoring and analysis of Qingdao Jiaozhou Bay Bridge by combining time-series SBAS-InSAR and GACOS[J]. Science Technology and Engineering, 2024, 24(5): 2074-2082.]
[17]	莫丹丹, 火久元. 基于SBAS-InSAR技术的喀喇昆仑公路盖孜河谷段形变监测与影响因子分析[J]. 测绘通报, 2025(8): 32-42. doi: 10.13474/j.cnki.11-2246.2025.0806
	[Mo Dandan, Huo Jiuyuan. Deformation monitoring and influencing factor analysis of the Gaizi River Valley section of the Karakoram highway based on SBAS-InSAR technology[J]. Bulletin of Surveying and Mapping, 2025(8): 32-42.] doi: 10.13474/j.cnki.11-2246.2025.0806
[18]	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.
	[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.]
[19]	Li H Y, Wang J, Wu S G, et al. Enhanced landslide detection from remote sensing imagery using an attention-optimized UNet-CBAM[J/OL]. EGUsphere. [2025-05-28]. https://doi.org/10.5194/egusphere-2025-2348.
[20]	Wu H G, Niu J B, Li Y Q, et al. Landslide Susceptibility Prediction Based on a CNN-LSTM-SAM-attention hybrid model[J]. Applied Sciences, 2025, 15(13): 7245, doi: 10.3390/app15137245.
[21]	Wu L, Liu R, Ju N P, et al. Landslide mapping based on a hybrid CNN-transformer network and deep transfer learning using remote sensing images with topographic and spectral features[J]. International Journal of Applied Earth Observation and Geoinformation, 2024, 126: 103612, doi: 10.1016/j.jag.2023.103612.
[22]	丁明涛, 江辉, 李振洪, 等. 融合卷积模糊神经网络与SHAP特征优化的地震诱发滑坡易发性评价[J]. 武汉大学学报(信息科学版), 2026, 51(7): 1404-1422.
	[Ding Mingtao, Jiang Hui, Li Zhenhong, et al. Earthquake-induced landslide susceptibility evaluation based on convolutional fuzzy neural network integrated with SHAP feature optimization[J]. Geomatics and Information Science of Wuhan University, 2026, 51(7): 1404-1422.]
[23]	Yang J, Ding M T, Huang W B, et al. A generalized deep learning-based method for rapid co-seismic landslide mapping[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2024, 17: 16970-16983. doi: 10.1109/JSTARS.2024.3457766
[24]	梁世川, 乔华, 吕东, 等. 伊犁谷地地质灾害分布特征及主控因素分析[J]. 干旱区地理, 2023, 46(6): 880-888. doi: 10.12118/j.issn.1000-6060.2022.458
	[Liang Shichuan, Qiao Hua, Lü Dong, et al. Distribution characteristics and main controlling factors of geohazards in Ili Valley[J]. Arid Land Geography, 2023, 46(6): 880-888.] doi: 10.12118/j.issn.1000-6060.2022.458
[25]	伊学涛, 尚彦军, 邵鹏, 等. 1970-2020年中巴经济走廊典型崩塌、滑坡空间分布及属性数据集[J]. 中国科学数据(中英文网络版), 2021, 6(4)
	[Yi Xuetao, Shang Yanjun, Shao Peng, et al. Spatial distribution and attribute dataset of typical rockfalls and landslides along the China-Pakistan Economic Corridor from 1970 to 2020[J] China Scientific Data (Chinese and English Online Edition), 2021, 6(4): 5-14.]

类别	滑坡因子	数据来源	空间分辨率	年份
地形地貌	高差/m	美国国家航天航空局	30 m	2011
	坡度/(°)	美国国家航天航空局	30 m	2011
	坡向	美国国家航天航空局	30 m	2011
地质条件	岩性	中国地质调查局	1:250000	2019
	距断层距离/m	中国地质调查局	1:250000	2019
	距震中距离/m	中国地质调查局	1:250000	2019
	归一化植被指数	地理空间数据云	30 m	2021
水文特征	年均降水量/mm	青藏高原数据中心	30 m	2019
水文特征	距河流距离/m	开放街道地图	30 m	2019
人类活动	距道路距离/m	开放街道地图	30 m	2019
地表形变	形变速率/mm·a^-1	欧洲航空局	30 m	2019—2024

滑坡因子	VIF	TOL
高差	3.838	0.261
坡度	1.536	0.351
坡向	1.278	0.849
岩性	1.887	0.534
距断层距离	1.835	0.545
距震中距离	1.354	0.738
归一化植被指数	1.275	0.784
年均降水量	4.323	0.244
距河流距离	1.256	0.696
距道路距离	1.358	0.727
形变速率	1.374	0.757

实际类别	预测类别
实际类别	预测正类	预测负类
实际正类	TP	FN
实际负类	FP	TN

模型	AUC	Accuracy	Precision	Recall	F1 Score
CNN	0.937	0.849	0.802	0.921	0.860
ResNet	0.952	0.884	0.867	0.907	0.891
ViT	0.952	0.867	0.890	0.914	0.889
FrIT	0.956	0.872	0.820	0.953	0.882
TransUNet	0.928	0.889	0.887	0.936	0.896
MADTNet	0.959	0.906	0.879	0.946	0.907

评价指标	Fold 1	Fold 2	Fold 3	Fold 4	Fold 5	均值
AUC	0.955	0.960	0.958	0.962	0.956	0.958
Auccary	0.903	0.908	0.904	0.912	0.901	0.906
Precision	0.873	0.885	0.879	0.882	0.876	0.879
Recall	0.943	0.947	0.944	0.950	0.945	0.946
F1 Score	0.904	0.910	0.905	0.912	0.903	0.907