A hybridized model based on neural network and swarm intelligence-grey wolf algorithm for spatial prediction of urban flood-inundation

In regions with lack of hydrological and hydraulic data, a spatial flood modeling and mapping is an opportunity for the urban authorities to predict the spatial distribution and the intensity of the flooding. It helps decision-makers to develop effective flood prevention and management plans. In this study, flood inventory data were prepared based on the historical and field surveys data by Sari municipality and regional water company of Mazandaran, Iran. The collected flood data accompanied with different variables (digital elevation model and slope have been considered as topographic variables, land use/land cover, precipitation, curve number, distance to river, distance to channel and depth to groundwater as environmental variables) were applied to novel hybridized model based on neural network and swarm intelligence-grey wolf algorithm (NN-SGW) to map flood-inundation. Several confusion matrix criteria were used for accuracy evaluation by cutoff-dependent and independent metrics (e.g., efficiency (E), positive predictive value (PPV), negative predictive value (NPV), area under the receiver operating characteristic curve (AUC)). The accuracy of the flood inundation map produced by the NN-SGW model was compared with that of maps produced by four state-of-the-art benchmark models: random forest (RF), logistic model tree (LMT), classification and regression trees (CART), and J48 decision tree (J48DT). The NN-SGW model outperformed all benchmark models in both training (E = 90.5%, PPV = 93.7%, NPV = 87.3%, AUC = 96.3%) and validation (E = 79.4%, PPV = 85.3%, NPV = 73.5%, AUC = 88.2%). As the NN-SGW model produced the most accurate flood-inundation map, it can be employed for robust flood contingency planning. Based on the obtained results from NN-SGW model, distance from channel, distance from river, and depth to groundwater were identified as the most important variables for spatial prediction of urban flood inundation. This work can serve as a basis for future studies seeking to predict flood susceptibility in urban areas using hybridized machine learning (ML) models and can also be applied in other urban areas where flood inundation presents a pressing challenge, and there are some problems regarding required model and availability of input data