Seagrasses, which provide many ecosystem functions to shallow coastal areas, are facing increased anthropogenic and climate disturbances, disrupting their ability as ecosystem stabilizers. Thus, understanding trends in seagrass abundance is vital to ensure sustainability of this resource. Despite the challenges associated with extracting spectral and spatial information over water, this work seeks to demonstrate the potential of satellite remote sensing of shallow coastal waters to map and monitor seagrass habitats at a large spatial and temporal scale. In this study, we used a combination of a basic atmospheric approach and a machine learning classification technique to correct cloud-free Landsat 8 imagery, classify seagrass, and quantify their total area in St. Joseph Bay (SJB), FL from 2013 to 2019.
Results/Conclusions
The efficacy of the dark object subtraction atmospheric correction method for Landsat 8 imagery was found to be sufficient for seagrass detection by the convolutional neural network (CNN) machine learning classification algorithm. A 93.1% agreement between a 2013 CNN classified image and a 2010 aerial survey from the Florida Fishery and Wildlife Conservation Commission was achieved, reinforcing the potential of using medium-resolution satellite imagery to map seagrass. Seagrass area in SJB ranged from a maximum of 28 km2 in 2013 to a minimum of 20 km2 in 2019. This study will address the current gap in spatial and temporal assessment of seagrass cover resulting from the challenges associated with land-based mapping or aerial surveys.