联合InSAR形变的四种不同泥石流灾害易发性评价模型对比研究——以小江流域为例
Comparative Study of Four Different Mudslide Disaster Susceptibility Evaluation Models with Joint InSAR Deformation—A Case Study of Xiaojiang River Basin

传统泥石流易发性评估方法主要依赖于静态数据，限制了对泥石流动态特征的深入理解，从而影响了评估结果的可靠性。InSAR技术能够提供泥石流区域地表形变的连续监测数据，为评估泥石流易发性提供了新的视角。本文将InSAR形变作为重要影响因子，结合地形、地质、水文和人类活动形成五大类影响因子，具体分为10个影响因子；并运用信息量模型(IM)、随机森林(RF)、支持向量机(SVM)和反向传播(BP)神经网络四种包含传统与机器学习的评估模型，进行泥石流易发性评价。实验结果显示，当模型中包含InSAR形变数据作为动态因子时，评估模型的各项评价指标均显示出更高的精度。在四种模型中，BP神经网络模型对本研究区域泥石流易发性识别具有最好的效果，模型精度达到89.5%，更适合该区域的泥石流易发性建模。此外，研究发现小江流域泥石流高易发性区域主要分布在大沟、老凹沟和蒋家沟等地区，应加强实时监测。本研究结果可为小江流域泥石流灾害预测预报和防灾减灾工作提供参考。
Traditional mudslide susceptibility assessment methods mainly rely on static data, which limits the in-depth understanding of the dynamic characteristics of mudslides and thus affects the reliability of the assessment results. The InSAR technology can provide continuous monitoring data of surface deformation in mudslide areas, which provides a new perspective for assessing mudslide susceptibility. In this paper, InSAR deformation is taken as an important influence factor and combined with topography, geology, hydrology, and human activities to form five categories of influence factors, which are specifically categorized into ten influence factors; and four assessment models containing traditional and machine learning, namely, informative modeling (IM), random forest (RF), support vector machine (SVM) and back-propagation (BP) neural network, are used to evaluate mudslide susceptibility. The experimental results show that all evaluation metrics for assessing the model show higher accuracy when the model includes InSAR deformation data as a dynamic factor. The BP neural network model is effective in identifying the mudslide susceptibility in this study area, and the model accuracy reaches 89.5%, which indicates that the BP neural network model has higher accuracy and reliability in the Xiaojiang River Basin, and is more suitable for modeling the mudslide susceptibility in this area. In addition, it was found that the high susceptibility of the Xiaojiang basin is mainly distributed in areas such as Dagou, Lao Au Gou, and Jiangjia Gou along the Xiaojiang basin, and real-time monitoring should be strengthened. The study’s results can provide a reliable reference for predicting and forecasting mudslide disasters in the Xiaojiang basin and for good disaster prevention and mitigation work.