滑坡灾害空间预测支持向量机模型及其应用

Landslide susceptibility mapping using support vector machines

随着GIS技术在滑坡灾害空间预测研究中的广泛应用,滑坡灾害空间预测模型成为研究的热点问题。在总结滑坡灾害空间预测研究现状的基础上,简要介绍了两类和单类支持向量机的基本原理。以香港自然滑坡空间预测为例,采用两类和单类支持向量机进行滑坡灾害空间预测,并与Logistic回归模型进行了比较。结果表明,两类支持向量机模型优于Logistic回归模型,而Logistic回归模型优于单类支持向量机模型。

With the extensive use of GIS techniques in landslide susceptibility mapping, the development of new predictive models for landslide susceptibility mapping has been a hotspot in landslide research. In this paper, the models for landslide susceptibility mapping are first reviewed, and the principle of two-class and one-class Support Vector Machines （SVM） is then briefly introduced. Two-class and one-class SVM methods were used to assess landslide susceptibility in a selected area in a natural terrain of Hong Kong using GIS. The SVM models were developed by training dataset with cross-validation method to obtain the optimum kernel function parameters, and then applied to the study area to derive landslide susceptibility maps. The resulting maps were compared with landslide susceptibility map produced from logistic regression. It is concluded that two-class SVM is more reliable than logistic regression, and that logistic regression is accurate compared with one-class SVM.