刘坚, 李树林, 陈涛. 基于优化随机森林模型的滑坡易发性评价[J]. 武汉大学学报 ( 信息科学版), 2018, 43(7): 1085-1091. DOI: 10.13203/j.whugis20160515
引用本文: 刘坚, 李树林, 陈涛. 基于优化随机森林模型的滑坡易发性评价[J]. 武汉大学学报 ( 信息科学版), 2018, 43(7): 1085-1091. DOI: 10.13203/j.whugis20160515
LIU Jian, LI Shulin, CHEN Tao. Landslide Susceptibility Assesment Based on Optimized Random Forest Model[J]. Geomatics and Information Science of Wuhan University, 2018, 43(7): 1085-1091. DOI: 10.13203/j.whugis20160515
Citation: LIU Jian, LI Shulin, CHEN Tao. Landslide Susceptibility Assesment Based on Optimized Random Forest Model[J]. Geomatics and Information Science of Wuhan University, 2018, 43(7): 1085-1091. DOI: 10.13203/j.whugis20160515

基于优化随机森林模型的滑坡易发性评价

Landslide Susceptibility Assesment Based on Optimized Random Forest Model

  • 摘要: 以三峡库区沙镇溪镇-泄滩乡为研究区,探索基于最短描述长度原则的信息增益法对滑坡连续型因子进行离散的效果,计算皮尔森系数去除高相关因子。利用信息量法预测的极低、低易发区随机抽取非滑坡样本点。通过迭代计算袋外误差估计确定较优的随机特征及其数目,将优化后的随机森林对研究区滑坡进行易发性评价,并与逻辑回归等方法进行比较。绘制各算法预测结果的接收灵敏度曲线,其中优化后的随机森林预测结果的曲线下面积较高,达91.8%,表明优化随机森林模型在滑坡易发性评价中具有较高的预测能力。

     

    Abstract: The research area is located in Shazhenxi town and Xietan town of Three Gorges reservoir area in this paper. In order to obtain better results that discrete the continuous factors of landslide, entropy based on minimal description length principle(Ent-MDLP) method is used. To avoid the influence of correlation between factors, we calculate the Pearson correlation coefficient to remove high correlation factor. In order to obtain more accurate non-landslide sample points, the non-landslide sample points are randomly selected from the very low and low susceptible regions predicted by the entropy method. For the optimized random forests model, the optimal random features and its number are determined by iterative calculation of out-of-bag error estimation. Then the optimized random forest is evaluated for the landslide of the study area, and the landslide susceptibility level is divided. The model is compared with the methods of logistic regression, support vector machine and non-optimized random forest. The accuracy of each model is evaluated by plotting the receiver sensitivity curve of each algorithm. The optimized random forest's area is the highest, which the area under the curve is 91.8%. These show that the random forest model is optimized with more high-predictive power in landslide-prone assessment.

     

/

返回文章
返回