About the event
The rapid and accurate detection of Antarctic sea ice is important for the safe navigation of polar ships, understanding ecosystem dynamics, and determining seasonal to decadal scale changes in sea ice response to a warming climate. Antarctic sea ice can be detected within passive microwave, multispectral and synthetic aperture radar (SAR) satellite imagery, each image types has limitations when used for this task. This talk presents two machine-learning approaches to leveraging multi-modal data to detect Antarctic sea ice extent (i) deep learning image segmentation combining concurrent multispectral imagery from the MODIS platform and Sentinel-1 SAR imagery, and (ii) convolutional neural processes for downscaling passive microwave derived sea ice concentration data using MODIS imagery. The talk will discuss the opportunities and limitations of both approaches.
This event is part of the Environment & Sustainability Seminar Series.
