基于深度神经网络的滑坡危险性评价——以深圳市为例

Landslide Susceptibility Assessment Based on Deep Neural Network:A Case Study of Shenzhen

该文以提升滑坡危险性评价精度为核心目标,对深度神经网络在滑坡危险性评价中的可行性和适用性进行研究,以期充分发挥深度神经网络强大的非线性学习和拟合能力,取得更加合理的滑坡危险性评价结果。选取滑坡灾害多发的深圳市作为实例,基于深圳市815条历史滑坡数据,开展了深度神经网络建模训练;通过与广义线性模型及分类与回归树模型训练效果的对比,对深度神经网络的建模效果进行了评价,深度神经网络、广义线性模型和分类与回归树模型的AUC值依次是0.908、0.861和0.857。将训练所得的模型应用于深圳市全区,对3种模型输出的滑坡危险性评价成果的合理性和可靠性进行了对比分析,结果表明:深度神经网络建模精度良好,优于常见的广义线性模型和分类与回归树模型,输出的滑坡危险性评价成果具有合理性,适用于滑坡危险性评价工作。

Landslide susceptibility assessment is considered to be a key role in landslide risk management tasks.In order to improve the rationality and accuracy of the landslide susceptibility assessment,deep neural network(DNN) is proposed to be applied in landslide susceptibility assessment to take advantage of its excellent nonlinear learning and generalization ability in this paper,which has been proven to be necessary and crucial in landslide susceptibility model training.Taking Shenzhen City,where landslide hazards occurred frequently,as the study area,a DNN model with 13 layers was designed and trained based on 815 historical landslide occurrence records in the study area.The performance of deep neural network was then analyzed and compared with two typical models,generalized linear model(GLM) and classification and regression trees(CART).The AUC of deep neural network reaches 0.908 while GLM and CART get 0.861 and 0.857 respectively,which means that DNN model outperforms GLM or CART in this case study.Then,all the above three models were applied to the whole study area to carry out further evaluation and comparison of the reasonability and reliability of the outputs by the three models,both in general area and in three typical landform regions.It can be concluded that DNN achieves better accuracy than GLM and CART.The resulting map of DNN is considerately reasonable and credible,which shows deep neural network is overall suitable for landslide susceptibility assessment.