Identifying Dry Intrusion Outflows in Present and Future Climates with Machine Learning

Dry intrusions (DIs) are one of the key airstreams associated with extratropical cyclones. They are known to exacerbate the impacts of mid-latitude weather through mixing with the boundary layer and enhancing atmosphere-surface interactions. Currently, DIs are identified with Lagrangian trajectory analysis, which has enabled studies into the climatology, variability, and characteristics of these airstreams. However, the future of DIs, and the impact of climate change on them, is currently unexplored due to the computational and data demands of this approach. The primary aim of this thesis is to develop a tool that unlocks new research avenues in extratropical cyclone and climate change science, and allows the future of DI outflows to be studied. Through a proof-of-concept and comparison of applied methods, followed by advanced model development, a convolutional neural network -- DI-Net -- is trained to identify DI outflow objects across the Northern Hemisphere, using information on relative and specific humidity, and topography. With only five input fields, DI-Net requires approximately 3% of the data needed to identify DI outflows through Lagrangian trajectory analysis. All the while, DI-Net demonstrates an ability to capture the spatial distribution of historic outflow activity, as well as the variability in DI characteristics and impacts observed across the Northern Hemisphere. Subsequently, DI-Net is applied to climate model data to study the historic representation of DI outflows and investigate the impact of warming on their occurrence and characteristics. DI-Net is able to predict objects in historic climate models that share similar climatological frequencies as the target data. Furthermore, these objects also exhibit known DI characteristics in agreement with existing literature. In future warming scenarios at the end of this century, varying behaviour is observed across forcing experiments and across the Northern Hemisphere. In the most extreme warming scenario, however, the total frequency of DI outflows decreases in the storm tracks, but a strengthening zonal bias leads to an increase in events seen over Europe. This behaviour aligns with future studies of extratropical cyclone track densities in CMIP6 models. Meanwhile, the anomalous properties and impacts of DI outflows are found to be consistent with historic results, showing no significant changes. Overall, this thesis presents the first exploration of future DI activity through the application of machine learning and builds the foundations for continued research in this area.<p></p>

Author dc:creator

• Owain Harris (21052700)

• dry intrusions
• extratropical cyclones
• machine learning
• neural networks
• climate change

Statement dc:rights

• All rights reserved
• Open Access after 2027-06-15

Identifier

10779/exe.32666073.v1

OAI identifier oai:identifier

oai:figshare.com:article/32666073