Optimizing Water Quality Modeling in an Uncertain Environment

Optimization of the overall modeling exercise must incorporate consideration of both model aggregation error (level of complexity) and input parameter uncertainty. Focusing on one issue, to the exclusion of the other, will not generally be optimal. Two distinct stream dissolved oxygen model formulations, each representing a different level of simulation complexity, are compared. A non-parametric statistic, the total likelihood (L_total) of successfully verifying a calibrated model, is chosen as the point of comparison. For normal levels of input parameter uncertainty and a minimum required dissolved oxygen accuracy of ±0.25 mg/L, the simpler single parameter model (L_total = 24.3%) had superior performance characteristics in comparison to the complex dual parameter model (L_total = 15.0%). In general, the performance (L _total) of the complex model improved more rapidly than the simple model following reduction of input parameter variability.