Improving understanding of precipitation with medical image registration

Thesis

Changes to the hydrological cycle under global warming are expected to involve both intensification of the hydrological cycle and changes in the location of key features. Thus, differences between General Circulation Models (GCMs) in simulated local precipitation changes may arise in part from location biases in the models' mean climates. Combined with the large internal variability and short spatial scales of precipitation, as well as the limited observational record, such biases would exacerbate the difficulty of robustly evaluating forced regional precipitation changes. While many techniques exist to correct biases in local intensity distributions, few have dealt with errors in location.

In this thesis, therefore, we make use of a technique from medical image registration to remove location biases from simulated precipitation fields, and demonstrate that this reduces disagreement on projected changes. We therefore proceed to introduce a tool tailored towards the removal of location biases from GCMs. We use this tool not only to investigate historical precipitation changes - showing that it marginally increases the detectability of the historical anthropogenic forcing on precipitation - but also to investigate the physical origin of precipitation biases. In particular, we compare the transformations found for different GCMs, and investigate how location biases in the West African Monsoon vary across a perturbed physics ensemble. This study enables us to identify a potential mechanism relating the location of the convergence zone to the parameterisation of the GCM.

By demonstrating the utility of such techniques, we hope that they will continue to be developed and applied in the Atmospheric Sciences, and that other applications may prove valuable.

Authors: Levy, A

• DOI: 10.5287/ora-z6rzvgzem
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford