Understanding uncertainty in process-based hydrological models

Building an environmental model requires making a series of decisions regarding the appropriate representation of natural processes. While some of these decisions can already be based on well-established physical understanding, gaps in our current understanding of environmental dynamics, combined with incomplete knowledge of properties and boundary conditions of most environmental systems, make many important modeling decisions far more ambiguous. There is consequently little agreement regarding what a 'correct' model structure is, especially at relatively larger spatial scales such as catchments and beyond. In current practice, faced with such a range of decisions, different modelers will generally make different modeling decisions, often on an ad hoc basis, based on their balancing of process understanding, the data available to evaluate the model, the purpose of the modeling exercise, and their familiarity with or investment in an existing model infrastructure. This presentation describes development and application of multiple-hypothesis models to evaluate process-based hydrologic models. Our numerical model uses robust solutions of the hydrology and thermodynamic governing equations as the structural core, and incorporates multiple options to represent the impact of different modeling decisions, including multiple options for model parameterizations (e.g., below-canopy wind speed, thermal conductivity, storage and transmission of liquid water through soil, etc.), as well as multiple options for model architecture, that is, the coupling and organization of different model components (e.g., representations of sub-grid variability and hydrologic connectivity, coupling with groundwater, etc.). Application of this modeling framework across a collection of different research basins demonstrates that differences among model parameterizations are often overwhelmed by differences among equally-plausible model parameter sets, while differences in model architecture lead to pronounced differences in model simulations at larger spatial scales. Work is ongoing to use this modeling framework to understand differences among existing models, especially, to understand why different hydrologic models have a very different portrayal of the impacts of climate change on water resources.