基于BP神经网络和决策树的昆明市东川区滑坡空间易发性评价

Spatial Susceptibility Evaluation of Landslide in Dongchuan District of Kunming Based on BP Neural Network and Decision Tree

滑坡是一种破坏性强、突发性高、诱导条件复杂的地质灾害类型,通过多源数据融合、采用机器学习方法训练有效的滑坡训练模型,对评价滑坡灾害易发性具有重要意义。以云南省昆明市东川区为研究区,选取高程、坡度、坡向、道路、水系等5个评价影响因子,结合实际的滑坡灾害隐患点的空间特征和属性特征数据,分别采用BP神经网络和决策树算法对滑坡易发性构建预测模型,通过ROC曲线进行模型精度验证和比较。结果表明,决策树模型对研究区滑坡易发性更敏感,预测结果可靠度高;用决策树模型生成滑坡易发性分区图,滑坡易发性分区结果可更有效地得出易发性评价,对防灾减灾部门准确评估滑坡易发性、有针对性地提高灾害预测及应急响应工作效率提供了一种有效的计算模型。

Landslide is a type of geological disaster with strong destruction, high burst and complex induction conditions. Training an effective landslide training model through multi-source data fusion and machine learning method is of great significance to evaluate the susceptibility of landslide disaster. Taking Dongchuan District, Kunming City, Yunnan Province as the study area, five evaluation impact factors such as elevation, slope, slope direction,road and water system are selected. Combined with the actual spatial and attribute characteristic data of landslide hazard hidden points, BP neural network and decision tree algorithm are used to construct the prediction model of landslide susceptibility, and the model accuracy is verified and compared through ROC curve. The results show that the decision tree model is more sensitive to the susceptibility of landslide in the study area and the prediction results are reliable. The decision tree model is used to generate the landslide susceptibility zoning map. The landslide susceptibility zoning results can more effectively obtain the susceptibility evaluation, and provide an effective calculation model for the disaster prevention and reduction department to accurately evaluate the landslide susceptibility and improve the efficiency of disaster prediction and emergency response.