Holistic characterization of water quality parameters to understand the ecological impacts of eutrophication

The naturally oligotrophic waters of the Florida Keys, U.S.A., can be impacted by small increases in nutrients, which are reflected by increased phytoplankton productivity and over time, produce hypoxic conditions, a process called “eutrophication.” Dredged canals can discharge legacy nutrients into the environment that have accumulated in sediments due to decomposition of submerged aquatic vegetation (SAV) in the canal. This study examines a large water quality dataset to determine a context-defined gradient of eutrophic conditions in the Florida Keys. Significant differences in water quality are used for the delineation of strong eutrophic areas as opposed to comparing to a set of previously defined threshold values. The gradient is determined by combining several machine learning techniques including principal component analysis, uniform manifold approximation and projection, Gaussian mixture modeling, and linear discriminate analysis (LDA) to generate a workflow designed to have broad application in context-dependent environmental analyses. Two clusters of sampling sites are defined where one cluster exhibits high chlorophyll-a concentrations and temperatures, low pH and dissolved oxygen saturation, and signs of nutrient pollution. LDA is used to generate a gradient of eutrophication within the dataset based on these clusters which can be interpreted as a water quality score. pH stands out as a parameter that is both significantly different between the two clusters and is the strongest predictor of the overall water quality within this environmental context. This study provides an alternative method for rapid monitoring and analyzing environmental data to identify emergent water quality trends in a large dataset.