A computationally efficient 2D hydraulic approach for global flood hazard modeling

Abstract:

We present a physically-based flood hazard model that incorporates two main components: a hydrologic model and a hydraulic model. For hydrology we use TOPNET, a more comprehensive version of the original TOPMODEL. To simulate flood propagation, we use a 2D Godunov-type finite volume shallow water model. Physically-based global flood hazard simulation poses enormous computational challenges stemming from the increasingly fine resolution of available topographic data which represents the key input. Parallel computing helps to distribute the computational cost, but the computationally-intensive hydraulic model must be made far faster and agile for global-scale feasibility.

Here we present a novel technique for hydraulic modeling whereby the computational grid is much coarser (e.g., 5-50 times) than the available topographic data, but the coarse grid retains the storage and conveyance (cross-sectional area) of the fine resolution data. This allows the 2D hydraulic model to be run on extremely large domains (e.g. thousands km²) with a single computational processor, and opens the door to global coverage with parallel computing. The model also downscales the coarse grid results onto the high-resolution topographic data to produce fine-scale predictions of flood depths and velocities. The model achieves computational speeds typical of very coarse grids while achieving an accuracy expected of a much finer resolution. In addition, the model has potential for assimilation of remotely sensed water elevations, to define boundary conditions based on water levels or river discharges and to improve model results.

The model is applied to two river basins: the Susquehanna River in Pennsylvania, and the Ogeechee River in Florida. The two rivers represent different scales and span a wide range of topographic characteristics. Comparing spatial resolutions ranging between 30 m to 500 m in both river basins, the new technique was able to reduce simulation runtime by at least 25 fold, while keeping flood extent discrepancy within 10% and water height differences mostly in the range 10-25%.