Copy this text
Sentinel-1 Wave Mode Sar Monitoring of Icebergs Around the Antarctica
The high-quality global wave mode synthetic aperture radar (SAR) vignettes routinely collected by Sentinel-1 is today extensively exploited for various oceanic and atmospheric phenomena. Yet, these observations still remain largely untapped for iceberg monitoring in the Southern Ocean. As a follow-up to our previous work (Wang et al., 2019b), a dedicated SAR image classifier is built to detect small-sized icebergs (<5 km) that are commonly underrepresented in current recording systems. It has been fine-tuned from the Inception-v3 deep convolutional neural network using a curated dataset of 2,062 iceberg and 15,338 non-iceberg cases. Independent evaluations, based on three additional datasets, achieve high precision and recall rates above 90%. Applied to all WV images acquired between 2016 and 2018 unveils iceberg occurrences around Antarctica. About ~7.5% of the detected icebergs drift into 40°S to 50°S latitudes, while the majority are concentrated poleward of 55°S. The seasonal patterns of SAR icebergs are generally consistent with altimeter-detection estimates, and exhibit advances over the sea ice regions. Linking these SAR icebergs to the reported large icebergs reveals that small icebergs are more likely located to the east of large iceberg trajectories, suggesting the primary driver of underlying ocean currents to their drift. Although precise identification of the shape and position of these small icebergs remains challenging, WV SAR vignettes provide added values to iceberg investigations at scales beyond current operational reports. Not only relevant for the precise monitoring of icebergs across a wider range of sizes, it can become instrumental for our understanding of iceberg tracking, associated dissolution, along with freshwater transport, and their broader impact on global and local climate processes.
Keyword(s)
Antarctica, Iceberg, Synthetic aperture radar, Sentinel-1 wave mode, Classification
Full Text
File | Pages | Size | Access | |
---|---|---|---|---|
Preprint | 30 | 14 Mo |