摘要
【目的】以重庆市武隆区滑坡与非滑坡样本作为研究对象,探究不同滑坡与非滑坡样本比例与超参数优化对滑坡易发性模型准确性的影响。【方法】选取距河流距离、地形湿度指数、多年平均降水量、坡向、曲率、地形起伏度、距道路距离、坡度、POI核密度、归一化植被指数、高程等11个因子作为影响因子,在1∶1~1∶10作为正负样本比例范围内,采用网格搜索法与贝叶斯优化,基于XGBoost模型对研究区样本的滑坡易发性进行评价。【结果】随着样本数量中非滑坡数量的逐渐增加,XGBoost模型准确率逐步提升,曲线下面积(AUC)未发生明显改变,且较高、高易发区面积逐步减少,低、较低易发区面积逐步增加;基于超参数优化后XGBoost模型AUC值均得到提升,且贝叶斯优化后的XGBoost模型的滑坡易发性评价精度与运行速率更优。【结论】以1∶10作为正负样本比例,通过贝叶斯优化所构建的基于XGBoost模型的武隆区滑坡易发性模型具有更好的预测能力与预测稳定性。
[Purposes]Taking the landslide and non landslide samples in Wulong district of Chongqing municipality as the research object, it explores the influence of different landslide and non landslide sample proportions and superparametric optimization on the accuracy of landslide susceptibility model. [Methods]11 landslide factors including distance from river, TWI, average annual precipitation, aspect, curvature, relief amplitude, distance from road, slope, POI kernel density, NDVI, to elevation are selected as impact factors. Within the range from 1∶1 to 1∶10 as positive and negative sample proportions, grid search method and Bayesian optimization algorithm are used to evaluate the landslide susceptibility of samples in the study area with XGBoost algorithm as the main model. [Findings]With the gradual increase of the number of non landslides in the number of samples, the accuracy of XGBoost model gradually improved, the AUC value did not change significantly, and the area of high and high prone areas gradually decreased, and the area of low and low prone areas gradually increased;The AUC value of XGBoost model based on super parameter optimization is improved, and the accuracy and running speed of landslide susceptibility evaluation of XGBoost model based on Bayesian optimization are better. [Conclusions]Taking 1∶10 as the positive and negative sample ratio, the landslide susceptibility model based on XGBoost model constructed by Bayesian optimization in Wulong district has better prediction ability and stability.
作者
张军以
丁悦凯
孙德亮
ZHANG Junyi;DING Yuekai;SUN Deliang(School of Geography and Tourism,Chongqing Normal University,Chongqing 401331,China;Chongqing Key Laboratory of Surface Process and Environment Remote Sensing in the Three Gorges Reservoir Area,Chongqing Normal University,Chongqing 401331,China;Chongqing Key Laboratory of GIS Application,Chongqing Normal University,Chongqing 401331,China)
出处
《重庆师范大学学报（自然科学版）》
CAS
北大核心
2022年第5期47-57,共11页
Journal of Chongqing Normal University:Natural Science
基金
国家自然科学基金(No.41901214)
重庆英才计划“包干制”项目(No.CSTC2021ycjhbgzxm0109)
重庆市教育委员会人文社会科学研究重点项目(No.22SKGH090)。
关键词
武隆区
样本比例
超参数优化
滑坡易发性
Wulong district
sample proportion
superparametric optimization
landslide susceptibility
