Integrating Machine Learning and Subsurface Characterization for Improved Urban Flood Modelling and Parameter Optimization

Abstract

Flooding is one of the worst natural disasters which endangers lives and hampers economic activities across the globe. Construction and calibration of physical models which includes computational resources and data are the primary components of urban flood modeling. In this research, a paradigm shift is proposed by combining subsurface data with ML methodologies which advance urban flood modeling techniques as well as parameter tuning. Model accuracy and robustness are achieved through parametric value enhancement. The findings obtained shed light on the potential improvements that were observed in urban flood modeling. Such changes will facilitate better management in the contexts of urban infrastructure development, risk assessment, and hydrological resource optimization. The strategic evaluation optimization technique relies on the K-means ANN driven Genetic Algorithm when dealing with urban inundation parameters. This technique identifies sensitive parameter values while also providing optimal results to enhance model effectiveness. Subsequently, the designed inundation model underwent validation through simulating real-world rainfall data followed by an evaluative approach to test the accuracy of the proposed predictive methodology.