论文部分内容阅读
滑坡负样本在统计型滑坡危险度制图中具有重要作用,能抑制统计模型对滑坡危险度的高估。当前滑坡负样本采样方法采集的负样本可信度未知,在负样本采样过程中,极有可能将那些潜在滑坡点错选为负样本,这些假的负样本会降低负样本集的质量和训练样本集的质量,进而影响统计模型的精度。本文基于“地理环境越相似、地理特征越相似”的地理学常识,认为与正样本有着相似地理环境的点极有可能是未来发生滑坡的点;与正样本的地理环境越不相似的点,则越有可能是负样本。基于此假设提出一种基于地理环境相似度的负样本可信度度量方法,将该方法应用于滑坡灾害频发的陇南山区油房沟流域,对油房沟进行滑坡负样本可信度评价制图;使用油房沟流域的滑坡发生初始面来验证该方法的有效性。结果发现:滑坡发生初始面上所有栅格点的负样本可信度平均值为0.26,超过95%的栅格点的负样本可信度都小于0.5,说明本文提出的负样本可信度度量方法合理。
The negative sample of landslide plays an important role in the statistical landslide risk map, which can restrain the statistical model from overestimating the risk of landslide. At present, the reliability of the negative samples collected by the negative sample sampling method for landslide is unknown. In the process of negative sample sampling, it is very likely that those potential landslide points will be wrongly selected as negative samples. These false negative samples will reduce the quality and training of negative sample sets The quality of the sample set, which in turn affects the accuracy of the statistical model. Based on the common sense of geography that “the more similar the geographical environment is and the more similar the geographical features are,” the article considers that the point with similar geographical environment to the positive sample is most likely to be the point where the landslide will occur in the future; the more dissimilar to the positive geographical environment of the sample Point, the more likely it is a negative sample. Based on this assumption, this paper proposes a new method to measure the reliability of negative samples based on the similarity of geographical environment. This method is applied to the oil-fanggou watershed in Longnan mountainous area where landslide is frequent, The validity of the proposed method is verified by using the initial landslide occurrence in the oil and gas drainage area. The results show that the average reliability of negative samples for all grid points on the initial landslide is 0.26, and the reliability of negative samples for grid points over 95% is less than 0.5, which indicates that the reliability of negative samples proposed in this paper The method is reasonable.