Title:		Representing uncertainty in hydrologic models

			Abstract

Hydrologic models are used for a variety of engineering applications,
ranging from predicting floods to assessing the impacts of climate and
land use change on water availability. In spite of their uses, the
fidelity of hydrologic model simulations is compromised by uncertainties
in the observations used for model inputs, uncertainties in the governing model equations and their parameter values, and weaknesses in the model time stepping schemes. In this presentation I will argue that traditional methods for quantifying model uncertainty – namely, inverse methods to infer an ensemble of model parameter sets, and multi-model (or multi-physics) ensembles to evaluate differences in model predictions for a range of different modeling configurations – although computationally convenient, provide a poor representation of model uncertainty. I will then present an alternative framework to quantify model uncertainty. I will use a set of case studies from around the world to demonstrate how decomposing models into a set of testable hypotheses – and evaluating those hypotheses with multivariate data from experimental watersheds – can help us understand both the most appropriate modeling options and the ambiguity in selecting among a range of competing model hypotheses.