Predictive Uncertainty and Parameter Sensitivity of a Sediment-Flux Model:Nitrogen Flux and Sediment Oxygen Demand

摘要

Estimating model predictive uncertainty is imperative to informedenvironmental decision making and management of water resources. This paperapplies the Generalized Sensitivity Analysis (GSA) to examine parameter sensitivityand the Generalized Likelihood Uncertainty Estimation (GLUE) to estimate thepredictive uncertainty of a sediment-flux water quality model with previouslyreported nitrogen and SOD data obtained from Chesapeake Bay sediment samples.The sediment-flux model comprises analytical solutions which describe the processesof ammonia production by anaerobic decomposition of sediment organic matter,nitrate production by nitrification, diffusive transport, denitrification, and sedimentoxygen demand (SOD). Results show that total ammonia flux sensitivity to theparameters is more complex than that of SOD and nitrate flux. The diffusiveboundary-layer thickness, which control mass transfer across the sediment-waterinterface, is identified to be the most sensitive parameter. Preliminary results showthat the posterior model predictive uncertainties are overly underestimated assignificant number of observations fall outside the GLUE-estimated 90% confidencelimits. These results are prelude to further and refined analysis to improve theestimation of model predictive uncertainty and posterior probability distributions ofthe parameters.