摘要
滑坡灾害易发性研究在滑坡灾害风险管理与城市规划等方面具有非常重要的现实意义。以滑坡灾害发育较多的三峡库区万州区部分区域为研究区,基于指标因素状态分级和因素相关性分析结果,选取高程、坡度、坡向等10项影响因素作为评价因子,以128个滑坡灾害点为样本数据,在深入分析和讨论当下机器学习模型在小范围区域易发性评价的不足与局限后,提出一种基于稀疏表达重构误差的滑坡易发性评价方法,通过重构误差来衡量某一像元与滑坡特征模式的相似程度,借此进行滑坡易发性的评价,并采取快速聚类法对所得到的易发性结果进行分级。再从统计分析、ROC曲线、制图结果等方面对该模型同常用的信息量模型以及机器学习的BP神经网络模型进行比较与评价。实验结果表明,稀疏表达重构误差模型的AUC高达0.8343,OA和Kappa系数为78.1%和0.562,均优于信息量模型和BP神经网络模型,模型稳定性较好,制图结果与实际更为吻合,具有良好的滑坡预测性能。
The study of landslide disaster susceptibility has very important practical significance in landslide disaster risk management and urban planning.In this paper,some areas of Wanzhou District in the Three Gorges Reservoir area,where landslide disasters are more developed,are taken as the study area.Based on the results of index factor state classification and factor correlation analysis,10 influencing factors such as elevation,slope and slope direction are selected as evaluation factors,and 128 landslide disaster points are taken as sample data.After in-depth analysis and discussion of the shortcomings and limitations of the current machine learning model in small-scale regional vulnerability evaluation,a landslide susceptibility evaluation method based on sparse expression reconstruction error is proposed.The similarity between a pixel and the landslide characteristic pattern is measured by the reconstruction error,and the landslide susceptibility is evaluated.The fast clustering method is adopted to grade the obtained susceptibility results.Finally,this paper compares and evaluates the model with the commonly used information model and the BP neural network model of machine learning from the aspects of statistical analysis,ROC curve and mapping results.The experimental results show that the AUC of the sparse expression reconstruction error model is as high as 83.43%.In addition,the OA and kappa coefficients are 78.1%and 0.562,which are better than the information model and the BP neural network model,and the model has good stability.The mapping results are more consistent with the actual situation than the other two models,and have good landslide prediction performance.
作者
孙晨昊
郑逸榛
李俊斌
霍姝涵
Sun Chenhao;Zheng Yizhen;Li Junbin;Huo Shuhan(School of Geophysics and Space Information,China University of Geosciences(Wuhan),Wuhan,Hubei 430074)
出处
《资源环境与工程》
2022年第5期614-624,共11页
Resources Environment & Engineering
基金
国家级大学生创新创业训练计划(S202110491144)。
关键词
滑坡易发性评价
机器学习
稀疏表达
快速聚类法
landslide susceptibility evaluation
machine learning
sparse expression
fast clustering method
