Uncertainty quantification in watershed hydrology: Which method to use?

Abstract

Different paradigms have emerged in watershed hydrology to deal with the uncertainties associated with modeling with both similarities and differences in philosophies and implementation.. A review of different paradigms, which include information-theoretic, formal Bayesian, informal Bayesian (GLUE), limits-of-acceptability (LOA), and others, is provided with a focus on their philosophical/theoretical underpinnings. Subsequently, the role of a relatively new paradigm, LOA is explored in the context of rainfall-runoff modeling. Ideally, LOA should be such that it captures measurement uncertainty due to errors in rainfall and streamflow measurements, however, it is difficult to specify an appropriate LOA. In this paper, a few methods of specifying LOA that have been proposed in the literature are discussed in the context of separation of structural and measurement uncertainties. It is concluded that the LOA method combined with the formal Bayesian method is a promising strategy whenever it is possible to construct LOAs.