基于Logistic回归的陕南秦巴山区降雨型滑坡预测方法

通过建立陕南秦巴山区降雨型滑坡灾害数据库，分析了不同降雨因子的雨强分布，计算了降水突发型滑坡灾害、降水滞后型滑坡灾害的雨强与滑坡发生概率的相关系数，采用Logistic回归方法确定不同时效降雨因子，得到陕南秦巴山区降雨型滑坡预测模型，并利用滑坡灾害实例，运用ROC曲线和kappa系数法对模型进行了验证。结果表明：滑坡前第m日降雨量R_dm(m=0,1,2)及综合雨量R_c四个降雨因子为诱发降雨型滑坡较为显著的因子。当降雨强度≥75 mm·h^-1时，最易引起突发型滑坡；当连续降水达到2 d，且24小时雨量达到小雨或中雨时，应警惕滞后型滑坡灾害的发生。模型预测准确率达82.1%，ROC曲线的AUC值为0.836，kappa系数为0.616，验证结果显示该模型可靠。研究成果可作为陕南秦巴山区降雨型滑坡预报预警研究工作的重要参考。

Prediction method of rainfall-induced landslides in Qinba Mountains of south Shaanxi Province based on Logistic regression

By establishing the database of rainfall-induced landslide hazards in Qinba Mountains of south Shaanxi Province, the rain intensity distribution of different rainfall factors is ascertained and the correlation coefficients between probability of occurrence of landslide and rainfall intensity factors are analyzed. Different rainfall factors are determined by Logistic regression methods. This research gets the regression model of rainfall-induced landslides in Qinba Mountains of south Shaanxi Province. Using the cases of landslide disasters, the model is validated by ROC curve and kappa coefficient method. The results show that the rainfall factors of R_dm (m=0,1,2) and R_c have the highest correlation with landslides. When the rainfall intensity ≥ 75 mm·h^-1, sudden landslide is most likely to burst. When precipitation lasts for two days and 24-hour rainfall reaches light or moderate rain level, the outburst of lagged landslide disaster should be on guard against. The accuracy of the model may reach 82.1%. The AUC value of the ROC curve is 0.836 and the kappa coefficient is 0.616. The verification results showed that the model was reliable. This study is useful to predict the rainfall-induced landslides in Qinba Mountains of south Shaanxi Province.