Can simple phosphorus mass balance models guide management decisions? A case study in the Bay of Quinte, Ontario, Canada

We test the capacity of an existing simple mass-balance total phosphorus (TP) model to evaluate nutrient loading scenarios in the Bay of Quinte, Ontario, Canada. Our study examines whether model parameters and loading inputs are well characterized and relevant to the current conditions in the Bay of Quinte and its drainage areas. We also identify critical data gaps and influential assumptions in regard to the uncertainty of model outputs and the credibility of predictive statements about the achievability of delisting objectives of the system. Our analysis shows that the model closely reproduced the observed variability of the TP seasonal averages during the calibration period 1972–2001, but its performance was significantly reduced when the actual predictive capacity was assessed in the 2002–2009 validation period. The most troublesome result is the inability of the model to reproduce the observed TP variability at temporal scales that are more meaningful from an environmental management point of view (i.e., monthly averages or daily snapshots from the system). Sensitivity analysis shows that several parameters associated with the role of the sediments were significant drivers of the model outputs, suggesting that considerable uncertainty exists in regard to the characterization of the sediments. The loadings from Trent River and the TP levels of the inflowing water masses from Lake Ontario predominantly shape the variability in the upper and lower segments of the Bay of Quinte, respectively. We also present a critical review of the suitability of the existing water quality criteria to depict the trophic status throughout the system. Our study contends that the summer average TP concentrations do not adequately reflect the prevailing conditions and that the development of proper water quality criteria should place more emphasis on inshore sites, where the eutrophication problems are more frequently manifested. Finally, we pinpoint factors unaccounted for by the original model that are likely to modulate the response of the system in its present state. We also discuss important directions of model structure augmentation and ways to optimize the spatial segmentation.