|
Abstract:
|
Riverine flooding events occur when river water levels exceed the capacity of their natural or constructed channels, posing challenges to both life and property. Hydraulic models can project riverine flood heights, and these projections can inform risk management policies. The input parameters for hydraulic models can be highly uncertain. Calibration methods use observations to quantify parametric uncertainty. With limited computational resources, calibration often proceeds with either a small number of expensive high resolution model runs or many cheaper low resolution runs. We propose a Gaussian process-based Bayesian emulation-calibration approach that assimilates model outputs and observations at multiple resolutions. We demonstrate our approach on the Lisflood flood hazard model for Selinsgrove, located in a flood-prone region of the Susquehanna River basin in Pennsylvania. Compared to existing single-resolution approaches, our method yields more accurate parameter estimates and flood predictions. The problem of utilizing model runs at different resolutions occurs in many scientific and engineering disciplines. Hence, our methodology has the potential to be broadly applicable.
|
