Chaney, Nathaniel NCBacelar, Luiz2025-07-022025-07-022025https://hdl.handle.net/10161/32813<p>With the continuous advancements in high-resolution environmental datasets (ă1 km) and high-performance computing, making locally relevant hydrological predic- tions at such scales has become increasingly feasible for inclusion in flood forecasting systems. However, uncertainties in input data, model structure, and parameter es- timation continue to pose significant challenges for land surface models (or process- based hydrological models) in accurately representing hydrological processes at these scales. This thesis investigates some of these sources of uncertainty to enhance the applicability and interpretation of locally relevant hydrological predictions. In the first results session, the WRF-Hydro hydrological model was implemented in a flash flood-prone region to explore issues associated with the spatial scaling of flood maps in short-term predictions. Although the methodology showed promise for real-time applications, parameter calibration was crucial for enhancing prediction accuracy, yet it remains constrained by the computational demands required to represent hy- drological processes amid high levels of land heterogeneity. In the second session, the HydroBlocks Multivariate Clustering Scheme is proposed as a strategy to increase the computational agility of WRF-Hydro calibration routines by clustering grids with similar hydrological behavior. During the time it takes for one WRF-Hydro prediction at a 26-meter spatial resolution, the HydroBlocks framework was capable of performing approximately 215 simulations without significant loss of local infor- mation. In the third session, applying HydroBlocks to simulate runoff generation at a 90-meter spatial resolution across CONUS revealed that in some basins, improved representation of land heterogeneity led to better streamflow simulations. This im- provement can largely be attributed to a more accurate spatial representation of soil properties, land cover, and climatic variables that drive runoff-streamflow relation- ships.To begin learning the spatiotemporal parameters that control runoff generation (i.e., vegetation or soil) at such high resolutions, the final chapter of the dissertation analyzes their spatiotemporal variation in four U.S. regions: California, Mississippi, Alabama, and North Carolina. It was found that certain soil parameters—such as saturated hydraulic conductivity—are among the most important for controlling both surface and subsurface flow, while vegetation parameters, although generally secondary, also exhibit significant spatiotemporal variability. This dissertation ad- dresses ongoing challenges in understanding and simulating high-resolution runoff generation in large-scale hydrological models, which is crucial for building reliable runoff-streamflow relationships in a new era of hydrological monitoring systems.</p>https://creativecommons.org/licenses/by-nc-nd/4.0/Hydrologic sciencesMeteorologyUrban planningClimate analysisFlood ForecastHydrological ModelsHydrometeorologyLand HeterogeneityDissertation