| Hydrologic extremes can have devastating consequences on ecosystems, infrastructure, agriculture, and economies. Providing reliable real-time information to stakeholders and decision makers is valuable to mitigate the impact of these events; land surface models, in conjunction with numerical weather forecasting and satellite remote sensing, present a unique opportunity to make this goal a reality. However, uncertainties in the parameterizations, input data, and parameter values in land surface models limit these systems' reliability and applications. The objective of this dissertation is to quantify and reduce a subset of these uncertainties. Chapter 2 assesses the skill of current macroscale land surface models to simulate the observed spatial heterogeneity of land surface processes. Chapter 3 builds on this assessment to develop HydroBloks-a state-of-the-art land surface model that explicitly represents field-scale land surface processes while ensuring computational efficiency. To reduce the uncertainty in model input data, Chapter 4 introduces an algorithm that merges ground truth observations into gridded meteorological forcing datasets. Chapter 5 addresses the lack of field-scale soil data over continental extents by using high performance computing, machine learning, and a century's worth of legacy soil data to develop the Probabilistic Remapping of SSURGO (POLARIS) dataset. Chapter 6 assesses the role of model parameter equifinality in land surface models with an emphasis on understanding the limitations of constraining models with low-quality runoff observations. Finally, Chapter 7 uses the FLUXNET network of eddy covariance sites to improve parameter estimates in the Noah land surface model. In the context of hydrologic monitoring, the results from this thesis show a path towards providing hydrologic information at field scales over continental extents. This work also illustrates how this increase in model complexity requires the implementation of robust ensemble frameworks and novel parameter estimation techniques to account for the unavoidable increase in model parameter and input data uncertainty. |
