Parameter distributions for uncertainty propagation in water quality modeling

Complex simulation models of lake eutrophication processes are commonly used to aid scientific understanding and to guide management decisions. Confidence in large models for these purposes depends on uncertainty in model equations and on effects of input uncertainties on model outputs. The objectives herein are to investigate the scientific basis for 20 equation parameters of a WASP-based mechanistic water quality model of Lake Okeechobee, Florida (USA) and to examine methods of estimating parameter distributions for Monte Carlo uncertainty propagation.; A review of the literature supporting parameter estimation reveals uncertainties about the applicability of reported laboratory and calibration values. Model homogeneity assumptions confound transferral of equations to natural systems, and wide ranges of reported values necessitate relatively arbitrary parameter judgments.; Estimates of parameter uncertainty which fail to account for parameter correlations do not accurately represent modeler uncertainty. Results show that independent distributions centered on calibration values generate unacceptable predictions of lake behavior. Alternative estimates are obtained in a regionalized sensitivity analysis, where parameter uncertainty is restricted to vectors whose predictions lie within the range of plausible observed behaviors. Results show that wide ranges of parameter values are behavior-giving subject to complex multivariate relationships. Preliminary investigations suggest Metropolis-Hastings sampling yields similar estimates but more efficiently combines information in the literature, the model, and data.; Propagation of behavior-giving parameter vectors in speculative forecasts predicts small lake responses to 40% reductions in external total phosphorus loads. Predicted decreases in Lake Okeechobee annual average chlorophyll {dollar}alpha{dollar} and total phosphorus concentrations are less than 2% and 6% respectively, regardless of parameter uncertainty.; In conclusion, estimates of model variability due to parameter uncertainty may be reduced if new site-specific data are obtained which reduce uncertainty in lake behaviors or in plausible parameter ranges. Reductions in overall uncertainty, however, are limited by inadequate model structure. For example, results illustrate that predictions of Lake Okeechobee total nitrogen are poor, regardless of parameter values. Uncertainty propagation with large mechanistic models may prove more useful in identifying such research needs than for predicting the distribution of possible lake responses for management decision-making.