An Updated Antarctica Coastline Extracted Automatically from Remote Sensing Data

In Antarctica, extraction and tracking of coastline changes have become of high importance because it has a significant impact on the climate change. The temporal and spatial change of coastline is a significant indicator of mass balance of ice shelves as well as their calving. In this study, we presented three image mosaics of surface morphology for the Antarctica continent, assembled from optical and SAR data in 2005, 2010 and 2017. Multiple source remote sensing images were pre-processed and then combined to improve the spatial resolution and accuracy of the mosaics. The original Canny algorithm was improved in this study by automatically calculating the high and low thresholds through the bimodality of the intensity histogram from the images, to eliminate the subjectivity and the error effect generated by setting thresholds manually. The improved Canny edge detection algorithm with minimized human intervention then was used to extract high-resolution (30m) coastline obtained from the mosaics of the Antarctica coastline. The results show that the extracted coastlines are highly consistent with the mosaics of coastlines from remote sensing data, and the accuracy of planimetric position is better than 75m. The coastline length of Antarctica has decreased from 35120 km in 2005 to 34928 km in 2017. The mean decrease rate is about 16 km/year. Meanwhile, the total area of the Antarctica continent is estimated to be 13625572 km² in 2005 and 13526892 km² in 2010 and 13669433 km² in 2017. We found that the decline of coastline area between 2005 and 2010 is greatly relevant with the significant collapse of the Antarctic ice shelf, it is mainly occured in the Antarctica Peninsula and East Antarctica, like the collapse of Wilkins Ice Shelf in March, 2008. This study would be an update of the MOA 2004 and 2009 products. Detailed analysis of the temporal and spatial change of Antarctic coastline and mass balance of ice shelves in the last decade will be carried out in the future based on our dataset.