Estimation of Channel Characteristics in Compound Flooding Prediction Models

Storms can simultaneously trigger storm surges, fluvial, and pluvial flooding, leading to compound flooding events. These events result in elevated water levels, larger flood extents, and more severe damage compared to the impacts of the individual flood mechanisms. Therefore, it is crucial for our better preparedness to assess increasing risks under climate change, evaluate mitigation measurements and produce forecasts using numerical prediction models. To build a prediction model that is robust and reliable under various situations with different meteorological, riverine and other forcing conditions, it is important to accurately represent the hydrodynamics of water bodies in the ocean, coastal areas, channels and on floodplains. Furthermore, the model must capture interactions between water bodies, such as overtopping from a channel to a floodplain and draining from a floodplain to a channel.

We have developed and applied the ADCIRC hydrodynamic model to predict storm surges in coastal areas for various purposes and scenarios. Recently, we extended ADCIRC’s capability to embed two-dimensional (2D) or one-dimensional (1D) channels in a 2D unstructured mesh, allowing seamless interactions between channel flows and floodwaters. Unstructured grids for channels and channel networks are generated based on their geometrical and topographical characteristics, including their centerline locations, widths, bottom elevations, and connectivity. Another critical parameter is Manning’s roughness coefficient, which represents losses due to bottom friction and all other factors not explicitly modeled in the governing equations.

The methodology for determining these characteristics in the context of compound flood predictions has not been established and is the focus of this presentation. Our test case is the prediction of compound flooding in North Carolina during Hurricane Florence 2018. To determine the geometrical and topographical characteristics, we use various digital elevation model products such as NOAA CUDEM, USGS CoNED, and 3DEP, along with HECRAS model dataset provided by the state government of North Carolina, the centerline polylines and water area polygons from USGS NHDPlus. To determine the appropriate Manning’s roughness coefficients and channel bottom elevations for better prediction accuracies, we conduct parameter studies comparing stage-discharge curves at multiple USGS stations with model results along Tar River, Neuse River, and Trent River. The stage-discharge curve analysis provides insights not only for determining parameter values but also understanding the capabilities and limitations of our 2D/1D channel representation with rectangular cross-sections. The approximation of channels with rectangular cross-sections sufficiently represents channel flows with medium to high discharges but underpredicts water levels with low discharges along smaller channels.

Hurricane Florence hindcast simulations are conducted using discharge data from USGS station data and NOAA National Water Model Retrospective Dataset. The channel parameters determined using this approach yield improved prediction accuracy of both maximum water levels and water level time series. It is emphasized that achieving excellent accuracy in predicting time series - and thus creating a robustly accurate model - requires extensive parameter estimation efforts compared to predicting maximum water levels during an event.