A conceptual precipitation-runoff modeling suite : model selection, calibration and predictive uncertainty assessment

Abstract

In Montana and much of the Rocky Mountain West, the single most important parameter in forecasting the controls on regional water resources is snowpack. Despite the heightened importance of snowpack, few studies have considered the representation of uncertainty in coupled snowmelt/hydrologic conceptual models. Uncertainty estimation provides a direct interpretation of the risk associated with predictive modeling results. Bayesian inference, through the application of Markov chain Monte Carlo methods, provides a statistical means of approximating uncertainty associated with both the parameters and the model structure. This thesis addresses the complexities of predictive modeling in hydrology through the development, implementation and analysis of a suite of conceptual hydrologic models under a Bayesian inference framework. The research is presented in four main sections. First, a comparative assessment of three recently developed Markov chain Monte Carlo algorithms, based on their performance across two case studies, is performed. This study has revealed that the extreme complexity of the parameter space associated with simple, conceptual models is best explored by the Delayed Rejection Adaptive Metropolis algorithm. Second, a complete description of the models and study site incorporated in the study are presented, building on established theories of model development. Third, an investigation of the value of each model structure, considering predictive performance, uncertainty and physical realism is presented. This section builds on results of the first section, through the application of the Delayed Rejection Adaptive Metropolis algorithm for model calibration and uncertainty quantification under Bayesian principles. Finally, a discussion of the Simulation and Prediction Lab for Analysis of Snowmelt Hydrology, developed to incorporate the tasks of model selection, calibration and uncertainty analysis into a simple graphical user interface is explained. The application of a complete modeling framework from model selection to calibration and assessment presented in this thesis represents a holistic approach to the development of improved understanding of snow-dominated watersheds through prediction by coupled snowmelt/hydrologic modeling strategies.