基于机器学习的长沙市滑坡灾害快速风险评价

Efficient Risk Assessment of Landslide Disasters in Changsha City Based on Machine Learning

提出采用随机森林(RF)和极限梯度提升(XGboost)模型对长沙市滑坡灾害进行快速危险性评价,并利用频率比法对评价结果进行检验和校准。基于层次分析法构建易损性快速评价体系并进行易损性评价。最后,采用数值分级方法集成危险性和易损性评价结果实现快速风险评价。结果表明,RF模型与XGBoost模型性能相近,但经频率比法校正后的XGBoost模型危险性评价结果更加合理;易损性评价中人口密度的权重值最大,高易损区多集中于市区、交通干线等区域。长沙市风险区划结果中较高风险、高风险区域占整个研究区面积的4.6%,主要集中在沟谷、城镇和交通干线等区域。

Random Forest(RF)and eXtreme Gradient Boosting model(XGboost)were adopted to assess the landslide hazard in Changsha City,and frequency ratio(FR)was then used to check and verify the obtained result.A vulnerability assessment system was established based on analytic hierarchy process(AHP)and then adopted to make vulnerability assessment.Finally,efficient risk assessment was realized by integrating the results of hazard assessment and vulnerability assessment with numerical classification method.It is found that RF model is similar to XGBoost model in its evaluation performance,but XGBoost model regulated by FR method can bring a more accurate assessment.Also,the vulnerability assessment has the largest weight value of population density,and high vulnerability areas are mostly concentrated in downtown area and traffic arteries.The areas with relatively higher and high risk account for about 4.6%of the entire study area in Changsha City,which are mainly concentrated in valleys,towns and traffic arteries.