Earth System Models (ESMs) are essential tools for simulating global water and energy cycles and supporting water resource assessments. However, the accuracy of their hydrologic simulations is often limited by parametric uncertainties and process simplifications. In this presentation, I will present my research efforts to enhance the representation of hydrologic and hydrodynamic processes in a fully coupled ESM. However, its coarse resolution (e.g., 50 km – 100 km) further limits their ability to simulate urban flooding dynamics at the scale relevant to human activities and infrastructure. To address this challenge, I contributed to the development of a new river dynamics core that solves two-dimensional shallow water equations, enabling efficient high-fidelity hydrodynamic simulations (e.g., 30m) at large scales. I will demonstrate this new capability using a case study of compound flooding driven by Hurricane Irene. By coupling with a regionally refined atmospheric model, a two-dimensional barotropic ocean model, and a high-resolution land surface model within an ESM, the multi-scale modeling framework captures unprecedented details of flooding processes in complex urban environments. Despite these advances, such actionable-scale flooding simulations remain computationally infeasible for real-time forecasting. To further improve the model efficiency, I proposed a real-time flooding forecast framework based on a machine learning based surrogate model trained on outputs from the physical flooding model prior to flood events. During an event, the surrogate model can rapidly predict flooding extent using forecasted precipitation within seconds on a low-end computer, while also quantifying uncertainties associated with precipitation forecasts.

*Refreshments: 12-12:30 PM, ES&T Atrium