顾及样本敏感性的滑坡易发性评价

滑坡作为一种危害极大的自然地质现象,严重威胁着人民的生命财产安全。因此,科学、准确地评价滑坡体的易发性至关重要。随着机器学习的发展,基于机器学习的滑坡易发性评价逐渐成为研究热点。而在真实情况中,滑坡区域与非滑坡区域面积占比悬殊,这使得机器学习模型的应用存在较严重的样本不均衡问题。本文采用样本敏感性分析方法,综合多个机器学习模型在不同比例的正负滑坡样本集上的表现,以获取最均衡滑坡样本集;并在此样本集基础上采用深度随机森林模型,在示范研究区开展滑坡易发性评价。最终的评价结果接近真实分布,表明本文方法具有较好的有效性。

Landslide susceptibility evaluation considering sample sensitivity

As natural geological phenomena of involving great danger, landslides have seriously threatened people's lives and property. Therefore, it is very important to predict the susceptibility of landslide scientifically and accurately. With the development of machine learning, the prediction of landslide susceptibility based on machine learning has become a research hotspot. But in the real situation, the area ratio of non-landslide and landslide area is very large, which makes the application of machine learning model exist serious sample imbalance problem. In order to obtain the most balanced landslide sample set, the performance of multiple machine learning models on different proportion of positive and negative landslide sample set is analyzed. The multigraded cascade forest model is trained on this sample set and used to predict the landslide susceptibility in the study area. The final prediction results are close to the real distribution, which shows that the method presented in this paper is effective.