ABSTRACT
This study assessed recently launched multispectral sensors to map seagrass properties for a ~ 150 km2 shallow bank in Moreton Bay, Australia. We utilised a previously developed semi-automated, object-based image analysis classification routine for our comparison, utilising field and image data as input. Field data were collected through georeferenced photograph transects which were analysed for species composition and percentage cover. Field data collection occurred close to image capture from Landsat 8 Operational Land Imager (30 m pixel), Sentinel-2 (10 m pixel), Ziyuan-3A (5 m pixel), and WorldView-3 (2 m pixel) sensors. The output maps had average overall accuracies of 66% for species and 57% for percentage cover maps. The study concluded that all sensors tested were suitable for mapping seagrass meadows in clear shallow waters, but the higher resolution sensors provided more detail and were considered more representative. The choice of sensor would depend upon the extent of the seagrass meadow, available funds and frequency of observation.
Acknowledgments
The fieldwork and data collection would not have been possible without the assistance of Moreton Bay Research Station and staff. OBIA image analysis eCognition software support was provided by Trimble. WorldView-3 image data was provided by Digital Globe and Geoimage Pty Ltd. ZY-3A image data was supplied and processed by SASMAC, NASG. Landsat 8 OLI image data was downloaded from Earth Explorer. Sentinel-2 image data was processed by DSITI.