基于自动机器学习的全球尺度滑坡灾害易发性预测  

Global Landslide Susceptibility Prediction Based on Automatic Machine Learning

作　　者：王毅[1] 陈曦 唐贵希 方志策 李朋磊 Wang Yi;Chen Xi;Tang Guixi;Fang Zhice;Li Penglei(China University of Geosciences,Wuhan,Hubei 430074;Hubei Geological Survey,Wuhan,Hubei 430034;Remote Sensing Application Technology Center of Hubei Geological Bureau,Wuhan,Hubei 430034)

机构地区：[1]中国地质大学(武汉),湖北武汉430074 [2]湖北省地质调查院,湖北武汉430034 [3]湖北省地质局遥感应用技术中心,湖北武汉430034

出　　处：《资源环境与工程》2022年第5期604-613,共10页Resources Environment & Engineering

基　　金：国家自然科学基金联合基金项目——地质环境遥感大数据智能解译(U21A2013);国家自然科学基金项目——高光谱遥感影像分解模型研究(61271408);湖北省地质局科技项目——基于多源数据的自然资源遥感监测体系研究(KJ2022-7),多灾种应急动态风险区划展示平台建设(KJ2022-58)。

摘　　要：基于机器学习的滑坡灾害易发性预测模型由于其固有缺陷,难以获得更为可靠的评价结果。自动机器学习作为新兴的智能学习方法,拥有自动选择特征、模型和参数的优良性能,并最大程度减少模型选择和训练成本,因此在滑坡灾害易发性预测方面具有广阔前景。为了深入研究该技术在全球尺度滑坡灾害易发性预测中的可行性,利用Auto-PyTorch自动机器学习模型构建全球尺度滑坡易发性预测处理框架,并与经典机器学习模型——随机森林(RF)模型和朴素贝叶斯(NB)模型进行性能对比。在此基础上,以全球60°N-60°S纬度范围作为研究区,进行全球尺度滑坡灾害易发性制图。实验结果表明,相较于经典机器学习方法,基于自动机器学习的滑坡易发性分析能够获得性能更鲁棒、精度更优越的预测结果,可以为全球性地质灾害风险评估和管理提供坚实的科学依据。The landslide disaster susceptibility prediction model based on machine learning is difficult to obtain more reliable evaluation results due to its inherent defects.As an emerging intelligent learning method,automatic machine learning has excellent performance in automatic selection of features,models and parameters,and minimizes the selection and training costs of models,so it has broad prospects in landslide disaster susceptibility prediction.In order to deeply study the feasibility of predicting the susceptibility of landslide disasters at the global scale,this paper uses Auto-PyTorch automatic machine learning models to construct a global scale landslide susceptibility prediction processing framework,and compares the performance with the classical machine learning model,Random Forest(RF)model and Naive Bayes(NB)model.On this basis,the global latitude range of 60°N-60°S was used as the study area to map the susceptibility of landslide disasters at a global scale.Experimental results show that compared with classical machine learning methods,landslide susceptibility analysis based on automatic machine learning can obtain more robust and accurate prediction results,and can provide a solid scientific basis for global geological disaster risk assessment and management.

关 键 词：滑坡 易发性预测 全球尺度 自动机器学习 灾害评估与管理 

分 类 号：P642.22[天文地球—工程地质学]