Spatially Referenced Watershed Models for the Red-Assiniboine River Basin Using Bayesian and Frequentist Parameter Estimation Methods: A Critical Binational Comparison

Excess nutrient loading to aquatic ecosystems remains a leading cause of declining water quality in lakes, estuaries, and coastal waters worldwide, with estimated global economic costs ranging from US$200 billion to US$2 trillion annually due to impacts on fisheries, tourism, freshwater resources, and water treatment. Within this context, our study focuses on Lake Winnipeg and its binational Red-Assiniboine River Basin, where sustained nutrient inputs have led to significant water-quality degradation and increasingly frequent cyanobacterial blooms. These changes pose substantial ecological risks, including algal proliferation and hypoxia, as well as public-health threats from toxin production and economic consequences for drinking water supply, fisheries, and tourism. A key feature of our study is the application of a spatially referenced watershed model with a parsimonious hybrid statistical/mechanistic foundation that partitions annual nutrient loads at watershed outlets into nutrient export rates from land uses, land-to-water delivery, and in-reservoir decay. We compare models calibrated using both Bayesian and traditional frequentist approaches. In the traditional model, significant source variables included agricultural inputs, forests and wetlands, and stream channels, precipitation as a measure of land-to-water delivery, and reservoir losses, with the coefficients for wastewater directly input into the model because it was not statistically significant. However, calibrations of all variables were able to be achieved using the Bayesian approach. Model results provide critical planning information by delineating nutrient-export hotspots across the broader watershed. Results indicate that 54–62% of phosphorus originates from the U.S. portion of the basin, and modelling scenarios reveal pronounced differences in agricultural nutrient sources, which in average constitute 67% of the sources – highlighting the importance of agriculture-focused BMPs. Given the global relevance of nutrient-driven water-quality challenges and the advantages of Bayesian calibration for robust risk assessment, our analysis offers a roadmap for designing and implementing adaptive nutrient management.