A novel hybrid approach based on a swarm intelligence optimized extreme learning machine for flash flood susceptibility mapping

• Published in: Catena (Giessen) Vol. 179; pp. 184 - 196
• Main Authors: Bui, Dieu Tien, Ngo, Phuong-Thao Thi, Pham, Tien Dat, Jaafari, Abolfazl, Minh, Nguyen Quang, Hoa, Pham Viet, Samui, Pijush
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.08.2019
• Subjects: case studies; decision support systems; Environmental modeling; Extreme learning machines; Flash flood; floods; model validation; neural networks; normalized difference vegetation index; Particle swarm optimization; prediction; rain; soil types; spatial data; statistics; streams; support vector machines; topography; typhoons; Vietnam; Vietnam; Vietnam; Extreme learning machines; Flash flood; Environmental modeling; Particle swarm optimization
• ISSN: 0341-8162; 1872-6887
• DOI: 10.1016/j.catena.2019.04.009

Flash flood is a typical natural hazard that occurs within a short time with high flow velocities and is difficult to predict. In this study, we propose and validate a new soft computing approach that is an integration of an Extreme Learning Machine (ELM) and a Particle Swarm Optimization (PSO), named as PSO-ELM, for the spatial prediction of flash floods. The ELM is used to generate the initial flood model, whereas the PSO was employed to optimize the model. A high frequency tropical typhoon area at Northwest of Vietnam was selected as a case study. In this regard, a geospatial database for the study area was constructed with 654 flash flood locations and 12 influencing factors (elevation, slope, aspect, curvature, toposhade, topographic wetness index, stream power index, stream density, NDVI, soil type, lithology, and rainfall). The model performance was validated using several evaluators such as kappa statistics, root-mean-square error (RMSE), mean absolute error (MAE), coefficient of determination (R2), and area under the ROC curve (AUC-ROC) and compared to three state-of-the-art machine learning techniques, including multilayer perceptron neural networks, support vector machine, and C4.5 decision tree. The results revealed that the PSO-ELM model has high prediction performance (kappa statistics = 0.801, RMSE = 0.281; MAE = 0.079, R2 = 0.829, AUC-ROC = 0.954) and successfully outperformed the three machine learning models. We conclude that the proposed model is a new tool for the prediction of flash flood susceptibility at high frequency tropical typhoon areas. •PSO-ELM is proposed and verified for flash flood susceptibility modeling.•PSO-ELM has high prediction performance.•PSO-ELM performs better than ANN, SVM, and the C4.5 decision tree.