FloodEye: Real-time flash flood prediction system for urban complex water flow

This study presents a water-level sensor system that has multiple functions: real-time river-level monitoring and high accurate flood prediction for urban complex water flow (CWF) flooding caused by Localized Heavy Rain. Traditional water-level sensors for large-scale river can detect water-level rising, though they have limitations in detecting urban river dyke height to alert about flooding. This is a result of various riverside environmental limitations. Additionally, previous prediction methods are difficult to detect CWF flooding with high accuracy because water rising differ depends on a location's environment. Therefore, we propose a detection scheme for CWF by developing a water-level sensor system that works in various installation environments using infrared image processing with both low installation and operation cost. We implemented a CWF prediction system that produces accurate and early predictions using a Linear Regression (LR) of deep learning approach with data assimilation. Evaluating based on actual installations (over 1,000 days, 13 locations), our system proved to accurately monitor and predict water-level rising after 5 minutes.