Deep Learning

Improving Antarctic Sea Ice Detection with CubeSat GNSS-R
A novel application of passive GPS signals (GNSS-R) picked up by cube satellites to detect the location and age of Antarctic sea ice. A simple Bayes rule based update strategy has been implemented to improve estimates of various sea ice types using both a gradient boosting machine learning model and a robust mixture discriminent analysis statistical model as the likelihood.
Steefan Contractor, Shane Keating, Jessica Cartwright, Alex
Improving Antarctic Sea Ice Detection with CubeSat GNSS-R
Gap-filling of Ocean Temperature Timeseries
Probabilistic deep learning (bidirectional LSTMs) with monte-carlo dropout for filling up to three month long gaps in daily temperature observations around the Sydney harbour and the East coast of Australia. As part of a postdoctoral role in the UNSW coastal oceanography group, part of the Integrated Marine Observing System Australia.
Gap-filling of Ocean Temperature Timeseries
Kelvin Wake Detection

We use fluid dynamics to simulate realistic images of the ocean surface with the triangular wave patterns behind moving objects (kelvin wakes) in water overlayed on wind generated waves. A deep learning (u-net) algorithm was trained to detect the presence absence of the kelvin wakes in the images. The prediction skill is further improved with the use of diffusion models to improve the signal to noise ratio between the kelvin wakes and the wind generated waves. This project was in collaboration with Naval Group Australia.

Photo by Martin Adams on Unsplash

Steefan Contractor, Shane Keating
Kelvin Wake Detection