关键词: 滑坡敏感性, GIS, 支持向量机, 交叉验证, 霍西煤矿区

Abstract:

Taking Huoxi Coal Mine Area in Shanxi Province as the research area, we conducted numerical modeling and quantitative evaluation of landslide susceptibility using remote sensing and GIS technology. Based on the DEM with spatial resolution of 30 m × 30 m, five topographical parameters were derived: elevation, slope angle, slope aspect, plan curvature and profile curvature. Stratigraphic lithology was digitized based on the geological maps from Department of Geological Survey in 1:50 000 scale. Fault network, drainage network and road were digitized based on the geological maps and other thematic maps from Department of Land Resource in 1:50 000 scale. Then, buffer for faults, drainage, and road were done. Mining disturbance were digitized based on the planning maps of coal resources. If the point falls in the mine area, it is proved to be disturbed by the mining disturbance, otherwise is not affected. NDVI and land-use types interpreted and computed the Landsat TM images. Landslide data was collected by Bureau of Land and Resources and it is represented by the X, Y coordinates of its central point. Then, the correlation characteristics among evaluation factors and the spatial distribution of landslides were acquired by using remote sensing technology and GIS spatial analysis method. Repeated 5-fold cross validation method was adopted in this research and the landslide/non-landslide datasets were randomly split into a ratio of 80:20 for training and validating models. Based on the methods of the 5-fold cross-validation and the fitting accuracy to the constructed the landslide susceptibility assessment model-Radial Basis Function - Support Vector Machine (RBF-SVM), the precision of the models was quantitatively assessed. We calculated the importance of each evaluation factor in the RBF-SVM model. Meanwhile, we obtained landslide susceptibility map of Huoxi Coal Mine Area based on the RBF-SVM model. The landslide susceptibility of Huoxi Coal Mine Area was divided into four scales referencing the quantile law: low (0-0.02), medium (0.02-0.1), high (0.1-0.85) and very high (0.85-1) probability of landslide. The results show that: (1) the fitting accuracy was 87.22% in the modeling phase and 70.12% in the validation phase, respectively, for the RBF-SVM model; (2) it indicated that lithology, distance from road, slope aspect, elevation and land-use types have contribution to each model. Therefore, these five factors are most suitable conditioning factors for landslide susceptibility mapping in this area. Mining disturbance factors have little contribution to the model. The mining method in this area is underground mining and the mining depth is very deep affecting the stability of the slopes. (3) The number of landslides points in the very high region was 316, which account for 93.49% of the total number of landslides points and 50.99% of the total area. This study obtained the spatial distribution characteristics of the Huoxi Coalfield geological disasters and the quantitative evaluation of landslide susceptibility. It provides reference for the investigation about artificial slope in the research area monitoring the rational mining coal resources. It will also provide the reference for the related research in other similar coal region and management work.

Key words: landslide susceptibility, GIS, Support Vector Machines, cross-validation, Huoxi Coal Mine Area