基于PR-SVM模型的黄陵县滑坡易发性评价

Landslide susceptibility assessment in Huangling County based on probability ratio and support vector machine model

基于概率比率(probability ratio,PR)模型的辅助和对比,探讨同一样本数据两次二分类、多分类和PR耦合分类的支持向量机(support vector machine,SVM)模型评价精度差异,并将最优分类策略应用于黄陵县滑坡易发性评价,旨在进一步探索县域滑坡易发性评价精度的提升路径。研究结果表明:因子多重共线性和相关性分析有效提高了样本参数的估计精度;PR模型的辅助运用有效提高了SVM模型的评价精度和效果;PR-SVM两次二分类模型较SVM多分类模型和PR-SVM耦合多分类模型具有更好的预测效果,115个详查滑坡灾害点中,93.91%落入中易发等级以上区域,63.48%落入高易发等级以上区域,表明该方法在滑坡易发性评价中具有应用推广价值;黄陵县滑坡极高易发区面积占比较低,主要集中分布于半坚硬层状碎屑岩区域和沟壑密度较大的黄土区,高易发区主要集中分布于水力侵蚀系数、降雨不均匀系数和沟壑密度均较大的黄土区。

The assessment accuracy difference of the support vector machine(SVM)for the same sample data under double binary classification,multiclassification,and probability ratio-coupled classification is investigated by aiding these methods with PR model and comparing the results.The optimal classification strategy is applied to examine the landslide susceptibility of Huangling County and to improve the accuracy of landslide susceptibility assessment by SVM.The results show that the multicollinearity and correlation analyses effectively improve the assessment accuracy and performance of the SVM.The PR-SVM double binary classification model also provides better prediction than SVM multiclassification and PR-SVM-coupled multiclassification models.In this study,115 landslide disaster points were surveyed in detail,93.91%of the points fall into medium and higher susceptibility zones,while 63.48%fall into high or very high susceptibility zones.The assessment results indicate that this method can be efficiently used for evaluating the landslide susceptibility.In Huangling County,zones with very high susceptibility are low,scattering in semi-hard stratified clastic and loess regions with high gully densities.High susceptibility zones in Huangling County are primarily found in loess regions with high hydraulic erosion coefficients,rainfall nonuniformity coefficients and gully densities.