Assessing the Relative Contributions of Input, Structural, Parameter, and Output Uncertainties to Total Uncertainty in Hydrologic Modeling

• Main Author: Pokorny, Scott
• Format: Thesis
• Language: unknown
• Published: 2021
• Subjects: Hudson Bay; Modeling; Hydrology; Hudson
• Online Access: http://hdl.handle.net/1993/36102

The simulation of physical environments by hydrologic models has become common as computational power has increased. It is well known that, to simulate the hydrology of a physical environment, simplifications of that environment are needed. The simplified versions of hydrologic processes generate uncertainty, in addition to ingesting uncertainty from input data. The uncertainty from one modeling step affects the next through propagation. Although computational power has increased through time, the computational demand for uncertainty analysis still remains a common limiting factor on the level of detail an uncertainty analysis can be conducted with. This thesis generates an estimate of total uncertainty propagated from input, structural, and parameter uncertainties for the Nelson River in the Lower Nelson River Basin near the outlet to Hudson Bay, as part of the BaySys project. Each source of uncertainty was relatively partitioned for determination of the most valuable source of uncertainty for consideration in an operational environment with a limited computational budget. The results of this thesis show the complex spatial and temporal variation present in gridded climate data. This thesis also presents an ensemble-based methodology to account for the input uncertainty associated with gridded climate data subject to propagation. The ensemble of input data was propagated through an ensemble of hydrologic models. Relative sensitivities of model parameters were shown to vary temporally and based on performance metrics, suggesting that aggregated performance metrics obscure information. Lastly, relative partitions of uncertainty were compared through cumulative distribution functions. Accounting for all sources of uncertainty appeared valuable towards improving streamflow predictability, however, structural uncertainty may be the most valuable in an operational environment with a limited computational budget followed by input, and parameter uncertainty.