The exploration and potential extraction of deep-sea minerals is emerging as a possible industry on the Norwegian continental shelf. However, knowledge of the geology in the upper layers of the seabed, and of the geological contexts in which mineral deposits occur, remains limited. To close these knowledge gaps, new methods are needed for collecting and interpreting data from the seafloor.
This PhD project investigates how artificial intelligence can be used to improve data collection in deep-sea environments. In particular, it develops methods within reinforcement learning, where an algorithm learns to make decisions through trial and error. A hierarchical approach is applied: a higher-level layer can propose subgoals, while a lower-level layer determines the concrete actions needed to achieve them. In this way, complex tasks can potentially be broken down into simpler parts.
One example is the control of autonomous underwater vehicles (AUVs) exploring the seabed. By combining hierarchical learning with modern sensor technology, AUVs can better adapt to variable environmental conditions and unpredictability in the terrain. This may enable more targeted and robust data collection, thereby strengthening the knowledge base of deep-sea geology on the Norwegian shelf.
Historically, geological data collection on the continental shelf has been driven by the oil and gas industry. Today, Norway faces a possible transition toward new resources and more sustainable alternatives. This research develops digital methods that can support such a shift by enabling faster and more informed decision-making. It may prove valuable for research, resource management, and the broader societal debate on deep-sea minerals.
Deep sea minerals exploration on the Norwegian Continental Shelf (NCS) is gaining traction, with the first licensing round expected to award licenses in Q1 2025. This emerging industry faces significant scientific, resource, and environmental uncertainties. This project aims to address these knowledge gaps and reduce uncertainties affecting the industry and public debate. The primary goal is to develop a prediction model for shallow lithology in deep-sea environments using machine learning. Although datasets on marine mineral systems on the NCS are limited, extensive analogous geological data exist. Training machine learning algorithms on this data will enhance the geological understanding of deep-sea lithology and mineral systems. As new data from NCS exploration emerges, the model will be refined and validated.
The project will leverage various geospatial datasets, including surface data, structural elements, raster data, geophysical datasets, and well data. The research will focus on the shallow lithological zone down to 200 meters below the seabed.
A secondary objective is to establish robust data protocols to ensure transparency, integrity, and reliability. As deep-sea mining is an emerging industry, inconsistent data standards hinder effective cross-industry data utilisation. The project will develop methods for combining geomorphological, geological, and geophysical data, promoting innovative data recycling to minimise environmental impact and reduce costs.
The PhD candidate will work with Adepth and have access to their data and networks, and data from sources like the Norwegian Offshore Directorate and Geological Survey of Norway. The research aims to publish three peer-reviewed articles on data representation methods, a cross-disciplinary study on machine learning for surface data, and a model for sub-surface data. Overall, the project aims to advance the understanding of deep-sea minerals, improving exploration efficiency and sustainability.
