Simple and efficient calibration and validation of gridded 2D hydrodynamic models using 1D models, stream gage measurements, and inundation extent maps

The accuracy and precision of grid-based 2D hydrodynamic modeling of pluvial and fluvial urban flooding scenarios are highly sensitive to the spatial resolution of the digital elevation model (DEM) employed. The use of a 1-meter resolution DEM can significantly improve the performance of 2D models relative to coarser resolutions. The sole 1m DEM with extensive coverage of the continental United States is provided by the US Geological Survey (USGS) through the 3DEP program. This dataset is produced by Lidar which is used to generate bare-earth elevations, particularly important for the modeling of urban coastal floodplains.

The challenge in using the USGS 1m DEM for hydrologic modeling is that waterways above a certain width are “hydroflattened”, with the elevation of channels reflecting an averaged elevation of the water surface and providing no information about the underlying channel bathymetry. As river bathymetry data is sparse and inconsistent, this presents an issue. One approach to this problem is adjusting the roughness coefficient, or Manning’s n, of the DEM water surface to reflect very low friction such that discharge “sits” on top of the “solid” base stage of the river and is transported downstream with low resistance. One issue with this approach is that the hydraulic radius is significantly smaller compared to that obtained using the actual channel area, potentially biasing results.

We devise a simplified set of experiments using Manning’s equation for a rectangular channel to efficiently calibrate and validate a 2D hydrologic model based on the USGS 1m DEM. We present the results of a case study of the Charles River watershed in Eastern Massachusetts, USA. The Charles is 129km long and passes through 23 cities and towns before draining into Massachusetts Bay, terminating at the highly urbanized core of the Boston metropolitan area.

Stream gage measurements are used to estimate 100-year return levels for daily average discharge and surface water elevation; these values and Manning’s n (roughness) reported in the literature are then used to estimate channel depth assuming simplified geometry. This in turn is used to estimate Manning’s n for the hydroflattened water surface, which is then substituted into the 2D model. Finally, 2D flood simulation results are evaluated against US Federal Emergency Management Agency (FEMA) 100-year inundation extent maps. This is done using four validation metrics: Probability of Detection, False Alarm Ratio, Critical Success Index, and Bias. This provides a simple and computationally efficient calibration and validation methodology for 2D gridded hydrodynamic models in the absence of known channel bathymetry and roughness; while a number of effective approaches have seen widespread use in different data availability and modeling contexts, our simplest possible methodology is generalizable across a broad range of scenarios and watersheds. This can be particularly useful when 2D inundation mapping is called for in interdisciplinary research contexts in which information on the spatiotemporally explicit evolution of floods is required, such as in the simulation of dynamic disruption and recovery of infrastructure systems in urbanized watersheds during extreme precipitation events.