Assessing the Utility of Hydrologic Model Diagnostics for Decision Support

Abstract

Theoretical, computational and experimental advances have led to easier access to more complex and robust hydrologic models. These hydrologic models may be used to support decision making by water managers and stakeholders. Modeler may choose to utilize a various combination of model diagnostics on different hydrologic data available to describe the model performance. The “goodness” of a specific diagnostic may depend on multiple factors (hydrologic complexity of basin, data availability, data used for evaluation, resources spent on model, validation methods, and intended use of model). Through the DCT, which explicitly evaluates a model’s skill at informing specific decisions, different model diagnostics are correlated to a model’s decision-support capability. In this thesis, a hydrologic model is used to evaluate three reservoir operation rule curves in the Lake of the Woods Watershed, based on ecological and economic impacts. Synthetic realities are generated through random sampling of parameters. Each synthetic reality is operated using all rule curves to determine the preferred rule curve for a given parameter set. Then, the model is calibrated to the synthetic realities’ using various calibration formulations. For each calibration, the model is evaluated on whether the model prefers the same rule curve preferred by the synthetic reality. After many of parameter set realizations, each incremental value of calibration formulation is assigned a similarity score to describe the probability of informing the correct decision. Using the correlation, the model’s capabilities and uncertainties may be more readily quantified and communicated to stakeholders. Results indicate specific calibration formulation may be beneficial to support specific decisions.