关键词: 易发性分区, BP神经网络, MIV, 空间预测, 滑坡评价

Abstract: To solve the problem of low accuracy in the weight determination of landslide vulnerability evaluation indexes and the susceptibility mapping, a new weight determination method and landslide vulnerability evaluation model was proposed. BP neural network model and MIV method were combined to obtain the optimal number of hidden nodes and optimize the neural network model. On the basis, the weight of landslide vulnerability evaluation index was determined by synthesizing the BP neural network connection matrix and MIV value, and the landslide evaluation model was constructed. The model was applied to landslide susceptibility mapping in Longnan County, and performance of the proposed method was evaluated by ROC curve. The results indicated that MIV-BP model had high accuracy (AUC = 0.8204) and practicality in landslides spatial prediction.

Key words: susceptibility mapping, BP neural network, MIV, spatial prediction, landslide evaluation