信息量支持下 SVM 模型滑坡灾害易发性评价

摘要

( Weighted-Information)-SVM( Support Vector Machine ) model was adopted to assess the landslide susceptibility on the basis of entropy and SVM models .The SVM’ s training dataset was comprised by the entropy of nine influence factors, including the stratum lithology, the geological structure, the slope gradient, the direction and structure of slope, the land use, the influence of water, and the NDVI (Normalized Difference Vegetation Index ), together with the sum of them .The landslide susceptibility of the whole study area was evaluated , and the result of landslide susceptibility was ranked according to the zero value and abrupt change value of the decision value of model .The landslide susceptibility in Wanzhou district was assessed as an example to validate the WI -SVM model .The research result showed that the accuracy of the training dataset was 81.41%and verification dataset 91.11%, superior to commonly used models.Area with high and relatively high susceptibility accounts for 47.05%of the entire area, mainly in the mainstream and tributaries of the Yangtze River with strong human activities .The results are consist-ent with the distribution of landslides which has been known , indicating that the WI-SVM model has good applica-bility for the study area .%三峡库区是我国滑坡灾害广泛发育的地区之一，滑坡灾害易发性评价对库区的防灾减灾有重要意义。在滑坡灾害易发性指标信息量的基础上，构建了信息量支持下的SVM模型，并对滑坡灾害易发性进行了评价。该模型根据地层岩性、地质构造、坡度、坡向、坡型结构、土地利用类型、水、归一化植被指数，以及上述指标的总信息量，共9类指标组成的数据集进行SVM训练，得到评价模型；运用该评价模型对研究区全区滑坡灾害易发性进行评价，并以模型决策值的零点和突变点确定易发性等级划分标准。并以三峡库区万州主城区为研究区验证模型，研究表明：信息量支持下SVM模型的训练样本精度为81．41％，验证样本精度为91．11％，优于常用的信息量模型，滑坡的高易发区和较高易发区占研究区总面积的47．05％，主要集中在人类工程活动强烈的长江干支流两侧，结果与已知滑坡分布基本一致，表明该模型在研究区具备较好的适用性。