COMBINING REMOTE SENSING, MACHINE-LEARNING AND MECHANISTIC MODELING TO IMPROVE COASTAL HYDRODYNAMICS AND WATER QUALITY MODELING IN THE LAURENTIAN GREAT LAKES

Large lakes frequently act as early indicators of shifts in the environment and observations within the Great Lakes ecosystems continue to highlight a deterioration in water quality, a surge in the occurrences of algal blooms, and growing threats to indigenous species. Given the inherent complex dynamics of these inland seas and the growing environmental pressures, it is important to understand shifts in the intricate process dynamics governing these systems. Hydrodynamics and temperature, in particular, are fundamental variables that play significant roles as they influence multiple physical, chemical, and biological processes that take place within the lakes and their coastal areas. The goal of this study is to improve models that focus on coastal hydrodynamics and water quality in the Great Lakes. Extensive field datasets were collected in Lake Huron and Lake Erie over multiple years focusing on diverse factors affecting coastal processes including the roles of oscillating, bidirectional exchange flows between Lake Michigan and Lake Huron at the Straits of Mackinac, groundwater upwelling and submerged sinkholes in bays of Lake Huron, and contaminant plumes originating from rivers draining into the lakes. The performance of the models heavily depends on the quality of boundary forcing data and how the domain is discretized. Thus, a systematic assessment was done to improve the models by improving domain discretization through depth-adaptive triangular meshes and nested-grid methods. Additionally, detailed meteorological forcing fields were created with reanalysis and in-situ datasets. High-resolution time series data for water quality variables were generated using machine learning models since traditional monitoring data are notorious for their low temporal resolution, especially for microbiological water quality. The accuracy and performance of the models were tested against in-situ observations. This encompassed data on currents, lake levels, water temperature, and water quality variables (turbidity and Escherichia coli concentrations). High-resolution remote sensing imagery was also incorporated for a comprehensive evaluation of spatial plume dynamics. Novel insights from this research include an understanding of the crucial role played by the exchange flows in the Straits of Mackinac on transport timescales and biophysical processes in the bays of Lake Huron. The exchange flows significantly influence regions as far down as 50-70 km from the straits changing, among other things, bottom currents, which have important implications for biogeochemical processes, including the resuspension of bottom sediment, nutrient availability (e.g., nitrogen and phosphorus), and the growth and sloughing events of benthic algae such as Cladophora. Observed vertical velocities close to the lake bed in Thunder Bay, Lake Huron were found to be an order of magnitude higher compared to simulated vertical velocities of the same system using models that did not explicitly account for groundwater inflow from the karst lake bed. Models and data were used to estimate the upwelling groundwater flux and to quantify the impacts of ignoring groundwater in this system. The study highlights the significant benefits of merging best-available techniques and a fusion of mechanistic modeling, machine learning, and remote sensing to push the envelope of model performance in the Great Lakes. This integration further optimized the utilization of remote sensing imagery, particularly in coastal areas. This research is anticipated to aid management initiatives aimed at enhancing and preserving the resilience of coastal regions.