AI-enabled Mapping of High Tide Flooding: A Case Study of the 2020 Flood in Newport Beach, California

Abstract:

Driven by sea-level rise, the past benign high tidal levels could translate into flooding with moderate rainfalls and/or ocean waves, or even without any weather events. As a kind of “nuisance flooding”, High Tide Flooding (HTF) disrupts transportation systems, poses risks to public health and safety, and causes property damages. Extracting the flood extent, water depth, and flow rate of a HTF event is challenging, because the traditional monitoring infrastructures are insufficient. For example, satellite imaging is limited by long scanning time intervals to cover the flooding event at the right time; and sensor networks installed sparsely along rivers and coastlines are limited in area coverage. Recently, volunteered data (or citizen science data) emerges as a new means of flood monitoring. Since the volunteered data is contributed by the affected communities, this data source could reach a wide geolocation coverage and continuous monitoring. Furthermore, the community contributed data provides the first-hand and exclusive witness of the flooding site that other data sources such as public media and local emergency warnings cannot replace. So this data source is considered a promising method of nuisance flooding observing. However, the data quality could be challenging and inconsistent for scientific studies and the quantitative data is usually difficult to extract because the volunteered data is not collected with the consideration of scientific purposes. The present study demonstrate that the Artificial Intelligence techniques could address this issue. We present a study to use a drone footage posted on Youtube by a local TV station to extract the flood extent, depth and flowrate of a flood event in Newport Beach, California in July 2020. The study shows that by recognizing the car and other reference objects, flood depth could be extracted from the footage; the flood extent can be reconstructed using a flood/dry land recognition algorithm; and the flood flowrate can be obtained through fluid dynamics analyses. The study shows that the emerging new method provides a possibility to virtually investigate the site for disaster management. The generated high quality data proves to be valuable to support the validation of numerical modeling.