Desktop Investigation of Arctic Eelgrass Along the Eastern Coast of James Bay using Multi-date and Multi-sensor Earth Observation data

We conducted a synoptic satellite survey of eelgrass (Zostera marina) presence/absence along approximately 260 km of the eastern coastline of James Bay, Quebec, Canada. Two multispectral satellite platforms RapidEye (RE5) and... [ view full abstract ]

We conducted a synoptic satellite survey of eelgrass (Zostera marina) presence/absence along approximately 260 km of the eastern coastline of James Bay, Quebec, Canada. Two multispectral satellite platforms RapidEye (RE5) and WorldView-2 (WV-2) were used based on image resolution and availability of archived imagery data.  These imagery data were paired with existing groundtruth data to create classified mapping of eelgrass distribution.

Five satellite images were acquired, covering the eastern shoreline of James Bay. Eelgrass was classified using a six-step process including: band mosaicking, water masking, unsupervised classification, application of signal loss mask, application of cloud mask and post processing. This methodology will be discussed including advantages and limitations. Recommendations will be made for mapping eelgrass beds in similar oceanographic conditions.

An overall accuracy of 72.7% was achieved with consumer and producer accuracies respectively of 100% and 68.7% for eelgrass presence data with a kappa value of 0.36. A total of 77 ground truth points distributed throughout the study area were used to determine the accuracy. Overall, the investigation proved useful for large-scale detection of eelgrass using an unsupervised classification applied to multi-year and multi-sensor satellite imagery.