Predicting flood susceptibility using LSTM neural networks

JOURNAL OF HYDROLOGY

卷:594

Article

Fang, Zhice¹  Wang, Yi¹  Peng, Ling²  Hong, Haoyuan³ 

[1] China Univ Geosci, Inst Geophys & Geomat, Wuhan 430074, Peoples R China

[2] China Inst Geoenvironm Monitoring, Beijing 100081, Peoples R China

[3] Univ Vienna, Dept Geog & Reg Res, A-1010 Vienna, Austria

关键词: Flood susceptibility prediction;    Long short-term memory neural network;    Deep learning;    Feature engineering;   

DOI  :  10.1016/j.jhydrol.2020.125734

来源: Elsevier

【 摘 要 】

Identifying floods and producing flood susceptibility maps are crucial steps for decision-makers to prevent and manage disasters. Plenty of studies have used machine learning models to produce reliable susceptibility maps. Nevertheless, most research ignores the importance of developing appropriate feature engineering methods. In this study, we propose a local spatial sequential long short-term memory neural network (LSS-LSTM) for flood susceptibility prediction in Shangyou County, China. The three main contributions of this study are summarized below. First of all, it is a new perspective to use the deep learning technique of LSTM for flood susceptibility prediction. Second, we integrate an appropriate feature engineering method with LSTM to predict flood susceptibility. Third, we implement two optimization techniques of data augmentation and batch normalization to further improve the performance of the proposed method. The LSS-LSTM method can not only capture the attribution information of flood conditioning factors and the local spatial information of flood data, but also has powerful sequential modelling capabilities to deal with the spatial relationship of floods. The experimental results demonstrate that the LSS-LSTM method achieves satisfactory prediction performance (93.75% and 0.965) in terms of accuracy and area under the receiver operating characteristic (ROC) curve.

【 授权许可】

Free