Author(s): Goncalo Jesus; Anabela Oliveira; Joao Rogeiro; Rui Rodrigues; Joao Fernandes

Linked Author(s): Anabela Oliveira, Rui Rodrigues

Keywords: Data quality; Sensor fusion; Machine learning; Sensor networks; Flood forecast

Abstract: For the development of an effective emergency management support system for flash floods in small area watersheds with high slopes, we must consider a real-time monitoring network of both forcing and response variables in the basin to generate timely warnings. This network should be complemented with hydrological modeling and vulnerability analysis of flood prone areas within the watershed. In the case study of the Vinhas Creek basin, a low cost monitoring network comprising multiple sensor stations was deployed along the area covering the pristine headwaters and the highly urbanized downstream floodplain, located at the Tagus estuary mouth. The effectiveness of the forecast procedures and emergency warning for these natural and hazardous events may however be hampered by inconsistent real-time observed data . Ensuring the quality of monitoring data is fundamental to avoid false alarms or to ignore relevant events (Jesus, 2017). In order to increase confidence in the sensory and sensor network technologies, considering that these are subject to potentially harmful environmental factors, a dependable data quality oriented methodology (Jesus, 2021) was applied to the monitoring network datasets. Herein, we present the latest implementation to instantiate the aforementioned methodology in the Vinhas Creek monitoring network. Redundancy was taken as a core aspect of network reliability. In this instantiation, we implemented several machine learning mechanisms to process measurements from the multiple sensors while correlating them according to their geographical position, monitoring timing and the relevant physical processes involved. As an output, we are able to predict the sensor measurements and compare them with the actual sensing value obtained in the monitoring network station. Moreover, in case of any sensor failure, one or more replacement values can be issued. These are important for the correct simulation of the hydrologic and hydraulic processes of the dendritic watershed systems and to predict the inundation characteristics such as levels and flow velocities. Jesus, G., Casimiro, A., & Oliveira, A. (2017). A survey on data quality for dependable monitoring in wireless sensor networks. Sensors, 17(9), 2010. Jesus, G., Casimiro, A., & Oliveira, A. (2021). Using Machine Learning for Dependable Outlier Detection in Environmental Monitoring Systems. ACM Transactions on Cyber-Physical Systems, 5(3), 1-30.

DOI: https://doi.org/10.3850/IAHR-39WC252171192022926

Year: 2022