Abstract: Three statistical methods, the Logistic regression, generalized additive models (GAM), classification and regression tree (CART), were applied to the analysis of landslide susceptibility in Shenzhen using. By calculating the Kappa value and the area under the ROC curve (AUC) through 10-fold cross validation, the authors made comparison of the three methods and evaluated the credibility of the landslide susceptibility maps derived. It can be concluded that under the condition of the study area, GAMis the best method while the performances of the Logistic regression and CART are approxi mately same. The Logistic regression and CART can be used to automatically detect the important factors, but by GAMthe relationship between every influencing factor and the dependent variable can be visualized.

Key words: landslide susceptibility analysis, Logistic regression, generalized additive models, classification and regression tree

摘要： 采用Logistic回归模型、广义加法模型(GAM)和分类与回归树(CART)3种统计方法， 对深圳市的研究区域进行了滑坡灾害的危险性评价；利用Kappa值和ROC曲线，结合危险性评价结果图对3种方法的效果进行了比较，并分析了3 种方法在选取主控因子以及确定因子影响程度等方面各自具有的特点。在研究区域的条件下，GAM 的效果优于Logistic回归模型和CART，Logistic回归模型和CART的效果大致相当。Logistic回归模型和CART可自主选择主控因子，通过GAM 可定量研究因子的影响程度以及变化趋势。

关键词: 滑坡灾害危险性评价, Logistic回归, 广义加法模型, 分类与回归树