Coastal Zone Mapping of the Great Lakes: A Machine Learning Approach

Coastal systems on the Laurentian Great Lakes are affected by meter scale water level fluctuations that can occur over hourly to decadal time scales which can cause erosion and damage to infrastructure. Changes in water levels, ice coverage, and storm activity have been projected under various climate change scenarios and will likely alter historic coastal processes across the region. Monitoring changes in the descriptive morphometrics of coastal systems, such as coastal bluffs and shorelines, over time can offer insight into how these systems might respond to projected climate scenarios. With increasing access to geospatial data through open- source aerial and satellite imagery, an improved observational record will enhance our understanding of the spatiotemporal scales and environmental and anthropogenic controls on coastal dynamics of the Great Lakes. To take full advantage of these large data sets, this study utilizes a machine learning (ML) approach to automate the identification of coastal features, such as the shoreline indicated by the wet/dry boundary, vegetation, and coastal bluffs. Random Trees (RT) classifier models were trained using manual classifications of satellite imagery of the northwestern shore of Lake Erie. Results demonstrate that coastal features can be sufficiently identified using independent ML models, indicating the potential for an effective tool to map coastlines across large spatial scales. Over time, a similar methodology can be used to monitor coastal system dynamics to inform coastal managers and policy makers on the response of the Great Lakes coastal systems to climate change