The potential of proxy water level measurements for calibrating urban pluvial flood models

Abstract

Urban pluvial flood models need to be calibrated with data from actual flood events in order to validate and improve model performance. Due to the lack of conventional sensor solutions, alternative sources of data such as citizen science, social media, and surveillance cameras have been proposed in literature. Some of the methods proposed boast high scalability but without an on-site survey, they can only provide proxy measurements for physical flooding variables (such as water level). In this study, the potential value of such proxy measurements was evaluated by calibrating an urban pluvial flood model with data from experimental flood events conducted in a 25 × 25 m facility, monitored with surveillance cameras and conventional sensors in parallel. Both ideal proxy data and actual image-based proxy measurements with noise were tested, and the effects of measurement location and measurement noise were investigated separately. The results with error-free proxy data confirm the theoretic potential of such measurements, as in half of the calibration configurations tested, ideal proxy data increases model performance by at least 70% compared to sensor data. However, image-based proxy data can contain complex correlated errors, which have a complex and predominantly negative effect on performance.