The presentation will introduce the development of meander evolution theory and the non-linear relationship between channel curvature and migration rate. The aim of the research has been to test the applicability of the qualitative and quantitative theories for the River Lugg, UK over a period of 150 years. Results suggest that migration rate was highest in the 1960s and the migration rate increases in a non-linear form as river bends become tighter.

Drug combinations can cause unexpected drug effects, which may remain undetected during clinical trials due to their necessarily limited size and length. After drug launch, observational databases collate information about patients who experience adverse events that could be caused by drug interactions. This research will draw upon expertise in signal processing in order to provide a Bayesian approach to identifying clinically significant drug-drug interactions in the post-market phase.

The quality of a part built by 3D Printing is in large part determined by the build parameters used. Currently, selection of suitable build parameters is a lengthy and expensive process, causing a delay in the release of new materials to market. This project, which is a joint CDT with the EPSRC CDT in Additive Manufacturing, aims to use data from past suitable build parameters to predict an applicable testing region to dramatically reduce the time and expense of the process.

This research project investigates the evolution of ice fronts for 220 tidewater glaciers in Greenland based on NASA Landsat 4-8 satellite imagery for the period 1984 - 2017 and correlates the results with climate forcings. The results show a linear retreating trend of terminus positions for all sectors of the Greenland ice sheet, with a significant new finding that the north-east has been retreating since 1984 and has accelerated in 2008. A correlation between tidewater glacier retreat and increasing air and sea surface temperatures is indicated, however no direct relationship with a single climate forcing could be established.

A preliminary rotorcraft simulator trial was conducted to investigate how motion cues, delivered by a short-stroke motion platform, influence pilot performance and overall simulator fidelity. This involved a vertical reposition task, performed by two test pilots, in which objective measures of pilot performance were taken in conjunction with pilot subjective opinion.

As Cannabis is one the most widely-used drugs across the world, understanding it's effects on brain structure is critical. Current literature suggests heavy cannabis use can impact memory and reward sensitivity networks. To investigate this we supplement a standard meta-analysis with a meta-regression, and a novel technique known as Meta-NSUE.

The aim of this research is to make inferences about the potential outbreak of influenza (flu) in a population using data from each of a number of sources using a particle filter. We consider each individual in the population to belong to a particular state and the population state at a particular time as the counts of the number of individuals in each state.

Due to recent global trends such as the increasing allocation of uncertain in output renewable generators and environmental changes, which are drifting weather scenarios towards extremes, resilience is becoming a major concern for the future power grid. In the last decade, several frameworks were proposed to quantify power grid reliability, risk and vulnerability. However, just a few quantify relevant uncertainties affecting, for instance, weather conditions and renewable production. Furthermore, many did not account for some key elements characterizing a resilient power grid. Specifically, a robust quantification of the relevant uncertainty sources is needed and especially when the available information is of scarce quality, limited or unavailable. Further, self-healing and learning skills define a resilient future grid and, to make the system smarter and more resilient, it has to be equipped with learning capability.
In this work, we investigate generalised Uncertainty Quantification (UQ) methods and Reinforcement Learning (RL) for managing Operation and Maintenance (O&M) of power grids. RL framework is employed to equip the system with learning capabilities in a stochastic environment and the generalised UQ framework quantifies the effect of lack of data and imprecision on the gird resilience score. Generalised UQ methods are powerful tools which can reveal the key input factors driving the output uncertainty. Thus, when additional information needs to be collected (e.g. the level of uncertainty prevents to answer to a resilience question) the UQ framework provides a factor prioritization, maximizing the uncertainty reduction and, at the same time, minimizing data gathering costs.

In this study, a stochastic mortality model is implemented in order to analyse the joint mortality of couples in continuous time. The aim is to identify the presence of short-term dependence between coupled lives and to observe how the impact varies in different cultural settings, enabling improvements within the insurance industry in the accurate pricing and valuation of products which involve mortality assumptions.

The trends of increasing population, food insecurity, climate change, water scarcity, and rising levels of urbanisation have resulted in an emerging need for advanced farming methods. One method, vertical farming, minimises horizontal space requirements with increased yield per unit area and can produce crops year round in soil-less, controlled environments. The vertical farming industry remains in its infancy and has struggled with growing pains, in particular, problems with economic viability and proving claims of environmental sustainability. Starting a vertical farm is a risky venture, with a large proportion of farms that fail within several years. One actionable insight from the diverse and scattered literature on this incipient industry is the need for robust scientific economic analyses of vertical farming. A decision support system is proposed to assist entrepreneurs wishing to develop small-scale vertical farms from early stages towards scalable growth.

The purpose of this research study is to quantify the risk, and the uncertainty of an investment, by predicting the direction of financial market prices using the random forest algorithm. Furthermore, a variable selection method, based on the random forest algorithm, is proposed. The outcome of this research shows that the random forest algorithm based on the selected features was more effective in terms of accuracy in the training and external dataset.

The aim of this study is to evaluate how the epistemic uncertainty is currently captured in state-of-the-art Probabilistic Seismic Hazard Assessments (PSHA) and compare various techniques that have recently been used to capture the appropriate level of uncertainty. Epistemic uncertainty is commonly evaluated using a logic-tree approach following the work of Kulkarni et al., 1984. By running retrospective tests of PSHA, we can compare analysis from the past to data available now and qualitatively and quantitatively assess if the epistemic uncertainty would have been correctly assessed at that time and therefore propose a way to better evaluate the present state of uncertainty. Repeating this process with increasingly modern models, and by decreasing the amount of available data with the use of modern techniques to deal with scarce and limited data will help to understand how epistemic uncertainty affects ground motion estimation in SHA.

Developing suitable priors is essential for accurately applying Bayesian inference to all problems, but in the case of fatigue analysis, where data sets are notoriously expensive to obtain, this is a particularly thorny issue. The inadequacy of a basic Bayesian approach to a very simple crack growth model is demonstrated with four different priors.