Skillful climate prediction does not automatically translate into useful prediction of hydrological extremes like flood and droughts. I am linking GFDL's SPEAR seasonal prediction ensembles with distributed hydrological and hydrodynamic models to identify where and when hydroclimatic extremes are predictable, and how that skill changes as atmospheric information propagates through land and river networks.

Research goal: bridge seasonal climate predictability and decision-relevant hydrological risk, while building toward a unified framework spanning weather-scale forecasts and multidecadal projections.

Large-scale hydrological models often simplify how reservoirs respond to changing inflow, storage, season, and wet/dry conditions. I developed and co-developed a family of interpretable machine learning models that infer decision rules from past operation records and transfer it to reservoirs with limited operating data.

Key contribution: analysis of 400+ major US reservoirs revealed representative operating modules and transition patterns; MODROM then generalized those modules to data-scarce reservoirs using widely available dam attributes.

I integrate data-driven reservoir operations into hydrological models to determine how management changes regulated streamflow, drought propagation, and water availability. This program has progressed from offline benchmarking with the National Water Model and VIC to TigeRes, a two-way coupled system that embeds MODROM within the GPU-accelerated Tiger-HLM.

Current scale: TigeRes is being evaluated across the Mississippi River Basin on a network of approximately 18.6 million river links and 1,700 reservoirs.

I led a proof-of-concept implementation of a coupled Tiger-HLM–TRITON framework that translates atmospheric forcing into runoff, river flow, and two-dimensional flood inundation. The framework combines distributed hydrology with GPU-accelerated hydrodynamics to capture fluvial and pluvial flooding.

Demonstrated capability: 30-meter modeling across the Delaware and Susquehanna River Basins, driven by a 200-member climate ensemble to generate probabilistic inundation information for catastrophe-risk assessment.