An Integrated Framework to Predict Peak Flood and Map Inundation Areas in the Chesapeake Bay Using Machine Learning Methods with High-Resolution Lidar DEM and Satellite Data

Abstract:

As the intensity and frequency of storm events are projected to increase due to climate change, local agencies are in urgent need of a timely and reliable framework for flood forecast which can downscale from watershed to street level in urban areas. Better-quality and finer-resolution data in hydrology, topography, and meteorology are available now. These data can potentially improve the accuracy of forecasting flood peak and the time to peak flood from the initiation of an individual storm event, but predictions require upgrading methodology which has previously been designed to operate in much more data-poor environments.

In this study, we propose a framework that uses machine learning methods to quickly and accurately predict the peak flood stage, time-to-peak, and maps peak inundated areas. We test the ability of this framework to forecast extreme flow events given knowledge of antecedent conditions and storm precipitation patterns. We develop a workflow utilizing USGS and municipal observations of gage height, reach-level cross-section of streams and riparian areas from 1-m Lidar Digital elevation model (DEM), real-time event precipitation quantity and intensity patterns, antecedent watershed soil moisture from satellite and a hierarchy of models, land cover patterns, and drainage network information to predict peak flood height and map inundation areas rapidly. We test the performance of the workflows using a set of urban to rural watersheds in the Baltimore Ecosystem Study.