摘要
针对神经网络模型进行滑坡易发性评价时,传统的随机选取非滑坡单元存在准确性不高的缺点,提出信息量与神经网络结合的易发性评价模型。以江西省上犹县为研究区,首先,基于上犹县滑坡编录与实际调查,选取坡度、高程、坡向、平面曲率、剖面曲率,植被指数(NDVI)、湿度指数(TWI)、距水系距离、距道路距离、土地利用等10个环境因子,其次利用信息量模型对上犹县进行易发性分区,得到上犹县易发性分区图。然后,从信息量模型得出的易发性分区中的低易发区选取非滑坡单元,与滑坡编录中的历史滑坡点组成测试集与训练集,输入神经网络中训练模型,再将上犹县所有栅格输入,预测上犹县栅格的滑坡概率。最后利用自然断点法在上犹县栅格滑坡概率进行分类,得到基于信息量与人工神经网络结合的上犹县易发性分区图。由易发性结果表明:单独的信息量模型的成功率曲线下面积AUC=0.7364,历史灾害点位于高易发区与较高易发区的灾害数占总灾害数的55.6%;基于信息量与神经网络模型的AUC=0.7874;历史灾害点位于高易发区与较高易发区的灾害数占总灾害数的85.8%。信息量–神经网络的评价模型比单独的信息量模型的评价精度提高了5.1%;高易发区与较高易发区所涵盖的灾害数占比高30.2%。信息量–神经网络模型有更好的评价精度,并且证明了在信息量模型中的极低易发区选取非滑坡点具有可行性。
The traditional method of randomly selecting non-slide units has the disadvantage of low accuracy when the evaluation of landslide susceptibility is carried out only based on neural network model.Thus,this paper proposes a susceptibility evaluation model that combines information value with neural network.Taking Shangyou County as a case study area.Firstly,ten environmental factors including slope,elevation,aspect,plane curvature,profile curvature,vegetation index(NDVI),topographic wetness index(TWI),distance to stream,distance to road and land use were employed to perform regional landslide susceptibility evaluation according landslide catalog and actual survey in the study area.Secondly,the susceptibility zoning of Shangyou County was carried out with the information value model,and the susceptibility zoning map of Shangyou County was obtained.Then,the non-landslide units were selected from the low-prone area in the susceptibility zone obtained from the information value model,and the test and training sets with the historical landslide points in the landslide catalog were divided.All the grids of the study area were then input into the resulting models to predict the landslide probability.Finally,the natural breakpoint method was used to classify the probability often grids,and a zoning map of landslides susceptibility based on the combination of information value and artificial neural network was obtained.The susceptibility results show that:the area under the success rate curve of the independent information model is AUC=0.7364,and the number of historical disaster points located in high-susceptibility areas and higher-susceptibility areas accounts for 55.6%of the total number of disasters;In comparison,the combined model based on both information value and neural network achieved a AUC value of 0.7874,and the number of disasters located in the high-prone areas accounted for 85.8%of the total disasters.The evaluation accuracy of the information value-neural network is 5.1%higher than that of the information
作者
陈飞
蔡超
李小双
孙涛
钱乾
CHEN Fei;CAI Chao;LI Xiaoshuang;SUNTao;QIAN Qian(School of Resources and Environmental Engineering,Jiangxi University of Science and Technology,Ganzhou,Jiangxi 341000,China;Jiangxi Key Laboratory of Mining Engineering,Jiangxi University of Science and Technology,Ganzhou,Jiangxi 341000,China)
出处
《岩石力学与工程学报》
EI
CAS
CSCD
北大核心
2020年第S01期2859-2870,共12页
Chinese Journal of Rock Mechanics and Engineering
基金
国家自然科学基金青年基金项目(41702327)
江西省教育厅科学技术研究项目(GJJ16064)。
关键词
边坡工程
信息量模型
神经网络
滑坡易发性
非滑坡单元
slope engineering
information model
neural network
landslide susceptibility
non-landslide unit
