基于地形和遥感影像特征的泥石流流域判识研究

The conditions of material and energy of gullies are the most decisive factors impacting the development and distribution of debris flow; however it is difficult to accurately quantify and evaluate them. The project intends to use downstream areas of the Jinsha River as the study object, where debris flows are very serious and widespread. The research will be embroidered on the relationship between the development of debris flows and the key factors of terrain and remote sensing image features. The conditions of terrain impact the changes of material and energy and remote sensing image features can reflect the changes. The watershed system was divided accurately from DEM of the study area. Some means, such as field investigation, remote sensing image feature extraction and laboratory simulation and analysis, will be adopted. Typical debris flow gullies and non-debris flow gullies are selected as sample training set. The deep learning algorithms and programs will be designed to identify the key factors of terrain and remote sensing image features. Then the coupling relationship among the values of the key factors will be erected. On basis of the coupling relationship a criterion is obtained and an identifying system will be developed to identify debris flow gullies and non-debris flow gullies. By analyzing the developing trends of the combination of key factors of terrain and remote sensing image features, the response mechanism of the development of debris flow gullies to the combination of key factors will be confirmed. Then the potential debris flow gullies will be identified and the future development of debris flows will be predicted. The project will provide a reliable theoretical and practical basis for the research on the development of debris flow gullies based on not grids but watersheds.

物质和能量是影响泥石流发育和分布的最具决定性的条件，由于其复杂性很难准确地量化和评估。本项目拟以泥石流发育且分布广泛的金沙江下游地区为研究对象，围绕影响流域物质和能量变化的流域关键地形特征以及能反映其变化的遥感影像关键特征与泥石流发育的关系，利用DEM划分准确的流域体系，采用野外调查、遥感影像特征提取以及实验室模拟、分析等手段，选择典型泥石流流域和非泥石流流域为样本训练集，设计深度学习算法，识别泥石流流域所特有的关键地形和遥感影像特征，分析泥石流流域各关键特征值间的耦合关系，以此为基础建立判据，开发泥石流流域判识系统，分辨泥石流流域和非泥石流流域；分析流域各关键地形和遥感影像特征组合的发展趋势，明确泥石流发育对流域关键地形和遥感影像特征组合的响应机制，识别未来一定时段可能发生泥石流的潜在流域，对泥石流的未来发育做出预测，为以流域为单元的基于形成机理的泥石流发育研究提供可靠的理论和实践依据。

本项目以泥石流发育且分布广泛的金沙江下游地区（流域面积9.13万km2）为研究对象，围绕影响流域物质和能量变化的流域关键地形特征以及能反映其变化的遥感影像关键特征与泥石流发育的关系，利用研究区1:5万DEM划分准确的流域体系，采用野外调查、遥感影像特征提取以及实验室模拟、分析等手段，选择典型泥石流流域和非泥石流流域为样本训练集，设计机器学习算法并开发相应程序，识别泥石流流域所特有的可以区别于非泥石流流域的关键流域地形和遥感影像特征，分析泥石流流域各关键特征值间的耦合关系，以此为基础建立判据，开发泥石流流域早期判识系统，初步分辨泥石流流域和非泥石流流域；分析流域各关键地形和遥感影像特征组合可能的发展趋势，明确泥石流发育对流域关键地形和遥感影像特征组合的响应机制，识别未来一定时段可能发生泥石流的潜在流域，对泥石流的未来发育做出预测，从而为研究区以流域为单元的基于形成机理的泥石流发育研究提供可靠的理论和实践依据。截止目前，项目已发表论文《Landslides》1篇“Rainfall-induced landslides and debris flows in Mengdong Town, Yunnan Province, China”（2020），国内核心期刊《灾害学》1篇“四川省泥石流灾害的时空分布规律和降水特征”（2017）。申请发明专利4项，其中，国内发明专利3项，国际发明专利1项。