Retrospective Reconstruction of the 2019 Midwestern Flood Inundation Dynamics

The record flooding along the Missouri River in the spring of 2019 severely impacted the communities in the Midwest United States resulting in billions of dollars in infrastructural and agricultural damages. Reconstruction of such flood occurrences can assist us to better understand the drivers, causality relationships, degree of flood damage, and develop long-term mitigation and resilience strategies. To accurately represent such large-scale and multi-month flood inundation dynamics, a computationally efficient tool with the ability to resolve the full 2D shallow water equations while ensuring a high spatial resolution for locally relevant information is desired. In this study, we reconstruct this rare flood event by implementing a GPU-accelerated 2D hydrodynamic model, TRITON (https://triton.ornl.gov/). The model is driven by the simulated runoff and streamflow from a calibrated VIC-RAPID hydrologic modeling framework, forced with the NCEP Stage IV hourly quantitative precipitation estimates. The baseline terrain information for TRITON is obtained from the 30m National Elevation Dataset leading to nearly 3.3 billion grid cells that represent 530,000 mi2 of the Missouri River Basin. We use the long-term climatic mean runoff and streamflow to create a steady-state channel stage and velocity that serve as initial conditions for TRITON simulation. We benchmark the simulated flood inundation outputs against high-water marks datasets, remote-sensing-based flood maps, other high-fidelity inundation model simulations, and US Geological Survey stage data. Finally, we present the computational efficiency and scalability of TRITON inundation simulations using the Summit supercomputer at Oak Ridge National Laboratory and discuss the challenges in large-scale hydrodynamic inundation modeling and their implications for real-time inundation forecasting.