Uncertainty Quantification of Hydrologic Models using Approximate Bayesian Computation (ABC)

Bayesian approach updates the prior probability of a hypothesis when new observation (evidence) is acquired. This method nicely relates to the iterative research cycle of hypothesis formulation, experimentation, data collection, and theory/hypothesis refinement. However, it has only marginally contributed to hypothesis/model improvement. In this talk, the theory, concepts and implementation of the approximate Bayesian computation (ABC) framework are introduced for uncertainty quantification and diagnostic analysis of hydrologic models. The introduced diagnostic approach relaxes the need for an explicit likelihood function in favor of one or more summary statistics, which when rooted in the relevant environmental theory have a much stronger and compelling diagnostic power than a conventional measure of error. The signature based indices, used in this framework, better extract process information from the data, and provide more insights into model behavior. The proposed methodology is statistically coherent, and provides new insights into and guidance on model structural (epistemic) errors, model (hypothesis) refinement, system non-stationarity, and the information content of experimental data.