Uncertanity Analysis in Parameter Estimation of Coupled Bacteria-Sediment Fate and Transport in Streams

Abstract:

E. coli is widely used as an fecal indicator bacteria in streams. It has been shown that the interaction between sediments and the bacteria is an important factor in determining its fate and transport in water bodies. In this presentation parameter estimation and uncertainty analysis of a mechanistic model of bacteria-sediment interaction respectively using a hybrid genetic algorithm and Makov-Chain Monte Carlo (MCMC) approach will be presented. The physically-based model considers the advective-dispersive transport of sediments as well as both free-floating and sediment-associated bacteria in the water column and also the fate and transport of bacteria in the bed sediments in a small stream. The bed sediments are treated as a distributed system which allows modeling the evolution of the vertical distribution of bacteria as a result of sedimentation and resuspension, diffusion and bioturbation in the sediments. One-dimensional St. Venant's equation is used to model flow in the stream. The model is applied to sediment and E. coli concentration data collected during a high flow event in a small stream historically receiving agricultural runoff. Measured total suspended sediments and total E. coli concentrations in the water column at three sections of the stream are used for the parameter estimation. The data on the initial distribution of E. coli in the sediments was available and was used as the initial conditions. The MCMC method is used to estimate the joint probability distribution of model parameters including sediment deposition and erosion rates, critical shear stress for deposition and erosion, attachment and detachment rate constants of E. coli to/from sediments and also the effective diffusion coefficients of E. coli in the bed sediments. The uncertainties associated with the estimated parameters are quantified via the MCMC approach and the correlation between the posterior distribution of parameters have been used to assess the model adequacy and parsimony.