896 Detecting and Tracking Iceberg A-76A from VIIRS Observations with U-Net Deep Learning Model

Thursday, 1 February 2024
Hall E (The Baltimore Convention Center)
Tiancheng Steven Shao, Montgomery Blair High School, Silver Spring, MD; and B. Zhang, S. Uprety, and J. Dong

Tracking iceberg movement aids in assessing potential threats to maritime safety, societal infrastructure and ecosystems, contributing to climate and environmental monitoring efforts. In this study, A U-Net based deep learning model was developed to process images from Suomi-NPP (SNPP) and NOAA-20 Visible Infrared Imaging Radiometer Suite (VIIRS) observations for detecting and tracking Iceberg A-76A. In May 2021, Iceberg A-76 broke from the Ronne Ice Shelf in Antarctica and was the largest iceberg at the time. Within a month, it fractured into three pieces, with the largest fragment being named as A-76A. Over the course of its journey, A-76A traversed nearly 2000 kilometers, eventually being dragged by coastal current along western Weddell sea to be within the Drake Passage/Scotia Sea by October 2022. Being in the relatively warm waters of Antarctic Circumpolar Current starting in March, 2023, Iceberg A-76A broke apart into multiple pieces near South Georgia Island. The VIIRS instruments onboard SNPP and NOAA-20 missions are designed to provide moderate-resolution, radiometrically accurate global images in 22 spectral bands ranging from 0.41 to 12.5 μm. In this study, SNPP and NOAA-20 VIIRS observations in three moderate-resolution visible bands from October, 2022 to April, 2023 are resampled onto uniformly spaced pixels to assemble RGB true color images. A subset of the VIIRS imagery dataset was extracted for Iceberg boundary tagging to form the training and testing datasets. The U-Net deep learning architecture is a deep convolutional neural network (CNN) and has a unique design consisting of a contracting path that captures context and a symmetric expanding path that enables precise localization. The U-Net model is particularly suitable for differentiating objects such as icebergs from their backgrounds. Once trained, the U-Net model processes unlabeled VIIRS images and generates segmentation maps of the icebergs, enabling the identification of the boundaries of icebergs. The detected boundaries of the Iceberg A-76A from U-Net were further analyzed to estimate its size, location, and track the drift of the iceberg from October, 2022 to April, 2023. Furthermore, the multiple daily observations of Iceberg A-76A from both VIIRS sensors provides the opportunity to quantitatively monitor the daily rotation and drift of the iceberg in a short time scale. For example, it was found that the Iceberg A-76A rotated about 35.1 degrees over three days from 2022-11-16 to 2022-11-19. Utilizing the U-Net deep learning architecture in conjunction with VIIRS observations for accurate iceberg detection and tracking provides an invaluable tool in understanding and mitigating the impacts of these icebergs on navigation, climate, and ecosystems. Tracking the iceberg movements with observations from constellation satellites can also help understanding the complex structure of the Antarctic Circumpolar Current with multiple frontal jets in different time scales. Finally, the presence of false positives in heavy cloud cover will also be discussed.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner