基于GIS和K-means聚类方法的滑坡气象预警模型

Landslide Meteorological Warning Model Based on GIS and K-means Clustering

针对滑坡气象预警问题,开展了GIS和K-means算法在滑坡气象预警中的应用研究。由于单一临界降雨判据的预警准确性不高,很难反映不同类型滑坡对降雨的敏感性不同,因此,采用加权欧式距离的K-means算法,根据滑坡的物质组成、斜坡结构、成灾层位、规模、地理位置、诱发因素等要素将滑坡分类,然后采用统计分析方法,得到滑坡前期有效降雨量和当日降雨量的散点图,针对每类滑坡构建临界降雨判据。并提出根据滑坡的风险性等级,即综合考虑滑坡发生的可能性和危险性,做出相应的滑坡预警。选取湖南某县为典型研究区,采用加权K-means算法进行聚类,将该县滑坡分为三类,根据滑坡的前期有效降雨量和当日降雨量,构建每类滑坡的临界降雨判据,使用GIS叠加滑坡分布图和危险性分区图,划分滑坡气象预警区。相对于单一临界降雨判据,该方法预测准确度更高;而且综合考虑滑坡发生的可能性和危险性,在实际预警中更加合理。

Aiming at the problem of landslide meteorological warning,the application of GIS and K-means clustering into landslide meteorological warning was researched. Due to low warning accuracy of single critical rainfall criterion,it is difficult to reflect the different sensitivity of different types of landslides on rainfall. Hence,the K-means clustering algorithm based on weighted Euclidean distance is adopted to classify the landslides according to the material composition,slope structure,formation,scale,location and induced factors of landslides. The scatter diagrams about effective rainfall before landslides and rainfall at the same day of landslide are obtained based on statistics analysis. Then,the critical rainfall criterion is established on the basis of different landslides. It is proposed to make landslide early warning according to the risk grade of landslide,that is,comprehensively considering the possibility and risk of landslide. With a county of Hunan Province chosen as the typical study area,the K-means clustering algorithm based on weighted Euclidean distance is adopted to classify the landslides into three types. The critical rainfall criterion for each type of landslide is constructed based on antecedent effective rainfall and rainfall at the same day of landslide. The landslide distribution map and the risk zone map were superposed by GIS,and the landslide meteorological warning district is obtained. Compared with the single critical rainfall criterion,this method has a higher prediction accuracy. Moreover,considering the possibility and danger of landslides,it is more reasonable in actual early warning.