This is a 4-year, fully funded project. Funding is provided by the University of Aberdeen Interdisciplinary Research initiative, which will recruit 20 Interdisciplinary Fellows and 12 Interdisciplinary PhD students in total over the next year, across five challenge areas, namely Data and AI, Energy Transition, Environment and Biodiversity, Health Wellbeing and Nutrition, and Social Inclusion and Cultural Diversity (more information here: https://www.abdn.ac.uk/research/explore/). The student will be part of this cohort and will benefit from other activities and interactions.

Climate change is leading to major environmental and socioeconomic changes [1]. Tackling the climate crisis requires approaching the problem from an interdisciplinary perspective. Machine Learning has been identified as a key enabling approach that supports our endeavours to tackle climate change, allowing us to exploit the wealth of data and information at very high scales, granularities and across modalities, with various initiatives already underway [2].

This 4-year interdisciplinary PhD studentship will be based in the Interdisciplinary Centre for Data and AI, and benefit from resources from within the Schools of Natural and Computing Sciences and Geosciences (e.g. facilities/HPC). The aim is to conceptualise new deep learning approaches, with a particular focus on self-supervised transformer and diffusion models for representation learning that will exploit large-scale multimodal datasets from remote sensing and other sources, such as satellites, drones and sensor networks. From a technical standpoint, it is intended that this PhD will build upon approaches described in [3-6], to contribute to foundational research in deep learning.

The techniques will be developed alongside considering international initiatives, such as the Svalbard Integrated Arctic Earth Observing System (SIOS) and Greenland Arctic Earth Observing System (GIOS), which are large repositories of data acquired from a range of platforms (e.g. satellites, drones, in situ sensors) collected over Svalbard and Greenland, respectively, to measure a range of Essential Climate Variables (ECVs). There is now huge potential to exploit this untapped resource, together with big data archives collected from satellites, to better understand how rapidly changing regions, such as the Arctic, are responding to climate change. New self-supervised learning approaches could be extremely helpful to leverage all the data available (unlabelled and labelled). In particular, developing data fusion and analysis techniques in a machine learning framework will be used to bridge gaps between previously distinct disciplines e.g. assimilating 2D imagery with 3D models to understand glacier surface changes.

It is anticipated that some early-stage developments will require the use of large-scale benchmark vision datasets, e.g. Imagenet.

The first 6 months of the project will be primarily spent exploring avenues and open problems within self-supervised learning, multimodal deep learning, and remote sensing, and establishing a deep understanding of how it can be used to improve our understanding of environmental change in locations such as the Arctic. A PhD proposal will be due on month 9.

It is essential that the PhD student is self-driven, curious, interested in working across disciplines and exploring new areas, as well as eager to work as part of a large interdisciplinary team.

The student will be expected to engage in public engagement activities, training events, and seminars and champion this area in collaboration with the supervisors, including international collaborators.