When can we declare a success? A Bayesian framework to assess the recovery rate of impaired freshwater ecosystems.

Evaluating the degree of improvement of an impaired freshwater ecosystem resembles the statistical null-hypothesis testing through which the prevailing conditions are compared against a reference state. The pillars of this process involve the robust delineation of what constitutes an achievable reference state; the establishment of threshold values for key environmental variables that act as proxies of the degree of system impairment; and the development of an iterative decision-making process that takes advantage of monitoring data to assess the system-restoration progress and revisit management actions accordingly. Drawing the dichotomy between impaired and non-impaired conditions is a challenging exercise that is surrounded by considerable uncertainty stemming from the variability that natural systems display over time and space, the presence of ecosystem feedback loops (e.g., internal loading) that actively influence the degree of recovery, and our knowledge gaps about biogeochemical processes directly connected to the environmental problem at hand. In this context, we reappraise the idea of probabilistic water quality criteria, whereby the compliance rule stipulates that no more than a stated number of pre-specified water quality extremes should occur within a given number of samples collected over a compliance assessment domain. Our case study is the Bay of Quinte, Ontario, Canada; an embayment lying on the northeastern end of Lake Ontario with a long history of eutrophication problems. Our study explicitly accounts for the covariance among multiple water quality variables and illustrates how we can assess the degree of improvement for a given number of violations of environmental goals and samples collected from the system. The present framework offers a robust way to impartially characterize the degree of restoration success and minimize the influence of the conflicting perspectives among decision makers/stakeholders and conscious (or unconscious) biases pertaining to water quality management.

A Bayesian risk assessment framework for microcystin violations of drinking water and recreational standards in the Bay of Quinte, Lake Ontario, Canada.

Freshwater ecosystems can experience harmful algal blooms, which negatively impact recreational uses, aesthetics, taste, and odor in drinking water. Cyanobacterial toxins can have dire repercussions on aquatic wildlife and human health, and the most ubiquitous worldwide are the hepatotoxic compounds known as microcystins. The factors that influence the occurrence and magnitude of cyanobacteria blooms and toxin production vary in space and time and remain poorly understood. It is within this context that we present a suite of statistical models, parameterized with Bayesian inference techniques, to link the retrospective analysis of important environmental factors with the probability of exceedance of threshold microcystin levels. Our modelling framework is applied to the Bay of Quinte, Lake Ontario, Canada; a system with a long history of eutrophication problems. Collectively, 16.1% of the samples of the system collected during the study period (2003-2016) exceeded the drinking water guideline of 1.5 µgL-1, while approximately 3% of recorded values exceeded the recommended recreational threshold of 20 µgL-1. Using a segmented regression model with a stochastic breakpoint of microcystin concentrations estimated at 0.54 µg L-1, we demonstrate that the environmental conditions associated with increased probability of exceedance of the drinking water standard are chlorophyll a concentration ≥7 µg L-1, water temperature ≥20 °C, ammonium concentration ≤40 µgL-1, total phosphorus concentration ≥25 µg L-1, and wind speed ≤37 km h-1. Considering the multitude of factors that can influence the ambient levels of toxins, our study argues that the adoption of probabilistic water quality criteria offers a pragmatic approach to accommodate the associated uncertainty by permitting a realistic frequency of violations. In this context, we present a framework to evaluate the confidence of compliance with probabilistic standards that stipulate less than 10% violations of microcystin threshold ambient levels.

Phosphorus retention and internal loading in the Bay of Quinte, Lake Ontario, using diagenetic modelling.

Internal phosphorus (P) loading significantly contributes to hysteresis in ecosystem response to nutrient remediation, but the dynamics of sediment P transformations are often poorly characterized. Here, we applied a reaction-transport diagenetic model to investigate sediment P dynamics in the Bay of Quinte, a polymictic, spatially complex embayment of Lake Ontario, (Canada). We quantified spatial and temporal variability of sediment P binding forms and estimated P diffusive fluxes and sediment P retention in different parts of the bay. Our model supports the notion that diagenetic recycling of redox sensitive and organic bound P forms drive sediment P release. In the recent years, summer sediment P diffusive fluxes varied in the range of 3.2-3.6 mg P m-2 d-1 in the upper bay compared to 1.5 mg P m-2 d-1 in the middle-lower bay. Meanwhile sediment P retention ranged between 71% and 75% in the upper and middle-lower bay, respectively. The reconstruction of temporal trends of internal P loading in the past century, suggests that against the backdrop of reduced external P inputs, sediment P exerts growing control over the lake nutrient budget. Higher sediment P diffusive fluxes since mid-20th century with particular increase in the past 20 years in the shallower upper basins, emphasize limited sediment P retention potential and suggest prolonged ecosystem recovery, highlighting the importance of ongoing P control measures.

Delineation of the role of nutrient variability and dreissenids (Mollusca, Bivalvia) on phytoplankton dynamics in the Bay of Quinte, Ontario, Canada.

The Bay of Quinte, a Z-shaped embayment at the northeastern end of Lake Ontario, has a long history of eutrophication problems primarily manifested as spatially extensive algal blooms and predominance of toxic cyanobacteria. The purpose of this study was to identify the structural changes of the phytoplankton community induced by two environmental alterations: point-source phosphorus (P) loading reduction in the late 1970s and establishment of dreissenid mussels in the mid-1990s. A combination of statistical techniques was used to draw inference about compositional shifts of the phytoplankton assemblage, the consistency of the seasonal succession patterns along with the mechanisms underlying the algal biovolume variability in the Bay of Quinte over the past three decades. Based on a number of diversity and similarity indices, the algal assemblages in the upper and middle segments of the Bay are distinctly different from those typically residing in the outer segments. Our analysis also identified significant differences among the phytoplankton communities, representing the pre- and post-P control as well as the pre- and post-dreissenid invasion periods. Recent shifts in phytoplankton community composition were mainly associated with increased frequency of occurrence of toxin-producing Microcystis outbreaks and reduced biovolume of N2 fixers, such as Aphanizomenon and Anabaena. Bayesian hierarchical models were developed to elucidate the importance of different abiotic factors (light attenuation, water temperature, phosphorus, and ammonium) on total cyanobacteria, Microcystis, Aphanizomenon, and Anabaena relative biovolume. Our modelling exercise suggests that there is significant spatial heterogeneity with respect to the role of the factors examined, and thus total phosphorus alone cannot always explain the year-to-year variability of cyanobacteria succession patterns in the system. The lessons learned from the present analysis will be helpful to the water quality criteria setting process and could influence the management decisions in order to delist the system as an Area of Concern.