SPARROW Applications Using Bayesian Inference Techniques

The parsimonious model structures of semi -empirical conceptual watershed models (e.g., SPARROW, GREEN) offer considerable advantages over physically -based watershed models to incorporate stream water monitoring data and effectively accommodate rigorous error analysis. Nonetheless, even these model structures demonstr ate intrinsic equifinality problems due to multicollinearity of model parameters. Bayesian inference techniques offer a robust and formal statistical calibration methodology to address model equifinality issues with watershed inverse analysis. In our presentation, we first summarise known case- studies of Bayesian inference implementation for semi -empirical watershed models (USA, Canada, China). We then provide an overview of relevant Bayesian statistical formulations to explicitly consider watershed spatial heterogeneity, serial correlation, and inter - and intra- annual dynamics. Finally, we outline the strengths and weaknesses of inverse watershed models within a Bayesian inference context for recursive calibration and data assimilation, including hotspot identification, loading source apportionment, representation of legacy nutrients, and quantification of all major sources of uncertainty