采用多模型融合方法评价滑坡灾害易发性:以湖北省五峰县为例

不同的易发性评价模型可以得到有差异的滑坡空间预测结果,选取最优模型甚至综合各模型的优势是提高易发性评价精度的有效方法。为检验模型融合思路的有效性,以鄂西地区五峰县渔洋关镇为研究区,提取坡度、地层、断层、河流、公路等7个滑坡成因条件,分别采用信息量模型、证据权模型和频率比模型进行滑坡易发性评价;并将3种模型分别进行归一化、主成分分析(PCA,Principal component analysis)和优势融合,得到了6幅易发性分区图。结果表明:优势耦合模型精度最高(90.3%),频率比模型次之(89.7%),归一化融合模型和PCA融合模型分别为89.3%和89.1%,以上4种结果的精度均高于证据权模型(87.7%)和信息量模型(87.6%);6幅预测图对应的评价结论与历史滑坡空间分布的实际情况相符。空间一致性对比结论表明,主成分融合模型与优势耦合模型的同格率高达68%,其预测结果避免了单个模型预测结论带来的偶然性和片面性,说明多模型融合方法与优势耦合模型在提高滑坡易发性预测精度上是可行性的,该思路对其他地区滑坡灾害易发性评价具有借鉴意义。

Landslide susceptibility assessment based on multi-model fusion method: A case study in Wufeng County,Hubei Province

Different landslide spatial prediction maps can be worked out from different landslide susceptibility models. It is efficient to choose the best optimal model or to integrate some models together in order to enhance the accuracy of landslide susceptibility. For the sake of testing the effectiveness of fusion models, the information model, the weights of evidence model and the frequency ratio model were used to predict the landslide susceptibility with the landslide controlling factors, such as slope, lithology, fault, river and road, in Yuyangguan Town, Wufeng County, Hubei Province. Then landslide susceptibility maps from three models were fused through normalized fusion method, principal component analysis fusion method and advantage fusion method. Comparatively, the accuracy resulting from advantage fusion method (90.3%) is highest among all that of other landslide susceptibility maps, including frequency ratio method (89.7%), normalized fusion method (89.3%), PCA fusion method (89.1%), weights of evidence mothed (87.7%) and information value method (87.6%). In the result of spatial agreement analysis, 68% area maps have the same class in the maps from advantage fusion method and PCA fusion method, decreasing the contingency and one-sidedness of single model. The study verifies the feasibility of the model fusion, and can provide a reference for landslide evaluation in the other geological environment.