Using CFD in a GLUE framework to model the flow and dispersion characteristics of a natural fluvial dead zone

Monte Carlo simulations of a two-dimensional depth-averaged distributed bed -roughness flow model, TELEMAC-2D, are used to model a detailed tracer disp ersion test in a complex reach of the River Severn in the Generalized Likel ihood Uncertainty Estimation (GLUE) framework. A time efficient, zero equat ion, spatially distributed eddy viscosity model is derived from physical re asoning and used to close the flow equations. It is shown to have the prope rty of low numerical diffusion, avoiding recourse to a globally large value of the eddy viscosity. For models of complex river flows, there are typica lly so many degrees of freedom in the specification of distributed paramete rs owing to the limitations of field data collection, that the identificati on of a unique model structure is unlikely. The data used here to constrain the model structure come from a continuous tracer injection experiment, co mprising six spatially distributed time series of concentration measurement s. Several hundred Monte-Carlo simulations of different model structures we re investigated and it was found that multiple model structures produced fe asible simulations of the tracer mixing, giving rise to the phenomenon of e quifinality. Rather than optimizing the model structure on the basis of the constraining data, we derive relative possibility measures that express ou r relative degree of belief in each model structure. These measures can the n be used as weights for assessing predictive uncertainty when using a rang e of model structures, to estimate the flow distribution under varying stag es, or for providing maps indicating fully distributed confidence limits in the risk assessments process. Such an approach is used here, and helps to identify the circumstances under which two-dimensional modelling can be use ful. The framework is not limited to the model structures that are develope d herein, and more advanced process representation techniques can be includ ed as computational efficiency increases. Copyright (C) 2001 John Wiley & S ons, Ltd.