Flood inundation modelling by a machine learning classifier

ABSTRACT The present study proposes and evaluates a machine-learning classifier to simulate the flood inundation area in which adaptive neuro fuzzy inference system was applied to classify the simulated domain into flooded and non-flooded areas. Particle swarm optimization was utilized in the training process of the data-driven model. Moreover, the outputs of simulating floods by the two-dimensional numerical hydraulic model were used in the training and testing process. However, aerial images of observed floods could be used as well. Based on the results in the case study, the proposed data-driven classifier is able to reduce the computational complexities of the flood inundation modelling including runtime and CPU usage. The proposed model is highly reliable and robust for generating maximum flood inundation map in the major floods. The results indicated that the rate of incorrect assessment is less than 7% in all tests. It is recommendable to apply the proposed method in the future flood engineering projects in which numerous simulations of the maximum flooded area are required. The developed model considerably reduces the computational costs in the projects.