Intercomparison of three spatially-resolved, process-based Lake Erie hypoxia models

Lake Erie hypoxia models have been developed for multiple objectives, including 1) to support develop-ment of phosphorus loading targets, and 2) to provide a real-time forecast of hypoxic zone location and movement. Nutrient management objectives have focused on spatially-averaged metrics, including the mean August-September hypoxic area or mean hypolimnetic dissolved oxygen (DO). However, the ability to model spatial patterns of bottom DO is also important for predicting impacts on drinking water intakes, benthic habitat, and biogeochemical cycling of nutrients. A previous binational effort used mul-tiple models to develop load-response curves to guide setting phosphorus load reduction targets to achieve ecosystem objectives under the Great Lakes Water Quality Agreement. Models were assessed individually, using different sets of observations, obscuring differences in modeled spatial patterns of hypoxia. In a first binational effort, we compared three previously developed Lake Erie hypoxia models in terms of predicted hypoxic area, as well as spatial patterns of hypoxia. All models were similar in pre-diction of hypoxic area in 2012, but differed in 2013. Several differences in model constructs likely con-tributed to varying predictions of hypoxia spatial patterns, including sources and sinks of DO in the water column, parameterization of sediment oxygen demand, simulation of vertical mixing, and representation of atmospheric and riverine forcing. This study underscores the importance of comparing models, with various representations of ecological processes, using a common set of visualizations, metrics, and obser-vations before generating ensemble modeling load-response curves to better inform lake management applications. Published by Elsevier B.V. on behalf of International Association for Great Lakes Research.

Automated summary

By comparing multiple Lake Erie hypoxia models, this study found that differences in model constructs can lead to variations in hypoxia predictions, emphasizing the importance of comparing different models before generating ensemble modeling load-response curves.