The thermodynamic phase of a cloud affects how it interacts with the environment: whether clouds are comprised of liquid droplets, ice crystals or a mixture of the two impacts their dynamical development, radiative properties and precipitation efficiency. Given that mixed-phase clouds are ubiquitous in the part of the atmosphere residing below 0 C, it is of particular interest to know the role of these clouds when establishing how radiation and hydrological budgets are affected. Mixed-phase clouds are particularly difficult for numerical models to represent well due to the scales of motion required, the interactions between liquid and ice particles, and their effects on radiative energy transfer in the atmosphere. As well as important for estimating fall-speeds and mass fluxes, the determination of ice crystal shape is critical in estimations of the radiative impact of ice clouds, since an ice crystal's shape, or habit, affects its scattering properties. Efforts to image and classify cloud particles using aircraft, and ground-based, measurements have been numerous. The challenge arises in processing the multitude of images obtained and subsequently in the classification of the particles imaged. In the first part of this thesis, the above problem is tackled by using an up-to-date deep learning algorithm, known as deep embedded clustering (DEC), on a large data set of 3V-CPI images. It is demonstrated that, when used with convolutional neural networks (CNNs), this method of unsupervised clustering provides useful information about in-situ cloud microphysical properties and can offer a valuable insight into the evolution of different cloud types. Meanwhile, past studies have found that ambient ice nucleating particles (INPs) are often insufficient by a few orders of magnitude to explain observed ice particle concentrations, suggesting that there are still gaps in our understanding of ice formation processes. This study re-visits the supercooled layer cloud conundrum to determine how it is possible for ice crystals to continually nucleate and precipitate for a prolonged period of time when, in theory, the INPs should run out. In order to investigate the possible causes and contrasting theories for the persistence of mixed-phase layer clouds, here a large-eddy simulation model with a bulk cloud microphysics scheme has been developed. The second part of the thesis describes its novel developments, which include the treatment of aerosols, variable ice particle habits, accurate ice particle fall-speeds and the inclusion of secondary ice production mechanisms, other than the well-known Hallett-Mossop process of rime splintering. Initial simulations presented within the third results chapter show that the new model is able to reproduce an observed shallow mixed-phase layer cloud. Further modelling of the case study finds that the recycling of primary INPs is of key importance, while secondary ice mechanisms have only a small effect. A successful simulation is also achieved by assuming that the INPs act in a slowly stochastic manner, as previously proposed. However, this is not necessarily required as very similar results are obtained by assuming singular nucleation. Knowledge of the implications of ice-containing clouds is essential for ultimately improving both numerical weather prediction and global climate models. Through the use of numerical modelling and deep learning techniques, the findings of this thesis highlight the need to consider the many complex aspects of mixed-phase layer clouds for their accurate representation in models.
|Date of Award
|31 Dec 2022
- The University of Manchester
|Jonathan Crosier (Supervisor) & Paul Connolly (Supervisor)