Urban flood vulnerability analysis and prediction based on the land use using Deep Neural Network

Abstract

Because of the uncertainties of climate change and the characteristics of flood damage, such as its large scale, occasionality, and complexity, there are limitations in preventing flood damage with structural measures alone. This study emphasizes non-structural land utilization measures to minimize urban flood damage and establish appropriate measures, and seeks to analyze and forecast urban flood vulnerability using a new hybrid fuzzy and DNN approach. Vulnerability to flood damage was divided into four classes. The analysis showed highest vulnerability in commercial areas, followed by residential areas, industrial areas, green areas, management areas, and agricultural and forest areas. The predictive analysis showed that the mean squared error was 1.95E-05, thereby demonstrating both the reliability of the model and the possibility of forecasting future flood-prone areas. The results of this study permit the identification of areas with potentially high vulnerability to flooding and intuitive determination of their current status. It is also possible to measure relative vulnerability among regions and select highly vulnerable areas first to devise intensive preventive measures. This study is significant in developing ways to minimize and reduce flood damage by making a land utilization plan to ensure safety in urban planning.