Isolating aerosol effects in multiple heterogeneous satellite datasets

Project

In this project we will develop new machine learning techniques for the identification and causal attribution of aerosol cloud interactions fusing multiple heterogeneous satellite observations. 

This project will develop novel data-fusion tools, including exploring multitask learning approaches, in combination with state-of-the-art machine-learning-based feature learning and tracking tools based on modern deep convolutional neural network architectures, such as ResUNet, designed for feature detection in remote sensing imagery with minimal training data.

The results will serve as novel observational constraints on the highly uncertain representation of such effects in high-resolution and climate models.

Supervisors

Philip Stier (supervisor), Dino Sejdinovic (co-supervisor), Javier Gonzalez Hernandez (Amazon, non-academic advisor)

Planned secondments

3 months at The Alan Turing Institute & 3 months at Amazon Seattle

Enrolment in Doctoral degree

PhD in Physics at University of Oxford