Temporal monitoring of Zostera marina along the eastern coast of James Bay utilizing multispectral Landsat imagery and random forests classifier
University of New Brunswick
Along the eastern coastline of James Bay, also known to the local Cree as Eeyou-Istchee, exist large subtidal eelgrass meadows. This study assessed the feasibility of evaluating the distribution of eelgrass beds along the entire eastern coastline of James Bay using imagery from the Landsat-8 Operational Land Imager and supervised classification using random forests machine learning algorithm. The methodology was then applied to historical imagery from the Landsat archive (Landsat-5 Multispectral Instrument), and image classifications were evaluated for accuracy using a randomly generated subset of digitized eelgrass distribution maps from Hydro-Quebec. Our classified images from 1988, 1991, 1996, and 2019 achieved overall accuracies ranging from 74.6 – 84.6% when evaluated using our ground-truth datasets. Our supervised classification approach showed the ability to detect eelgrass along the entire coast where turbid water was not present. The total area classified as eelgrass appeared to decrease over the study period (1988 – 2019).