Predicting flood susceptibility using long short-term memory (LSTM) neural network model

Abstract 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 studies ignore 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. Three main contributions of this study are summarized as follows. First, it is a new perspective that the deep learning technique of LSTM is used 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 both attribution information of flood conditioning factors and local spatial information of flood data, but also retain the powerful sequential modelling capability to deal with flood spatial relationship. Experimental results demonstrate that the LSS-LSTM method achieves satisfying prediction performance (93.75% and 0.965) in terms of accuracy and area under the ROC curve.