The School of Mathematical Sciences of Queen Mary University of London invite applications for a PhD project commencing either in September 2020 for students seeking funding, or in January 2020 or April 2020 for self-funded students. The deadline for funded applications is the 31st of January 2020. The deadline for China Scholarship Council Scheme applications is 12th January 2020.
This project will be supervised by Prof. Christian Beck and Dr. Benjamin Schafer.
Air pollution is the fifth leading cause of mortality worldwide. In a large city like London about 9500 premature deaths are attributed to air pollution every year. Notwithstanding this impact, the time-varying statistics of air pollution are far from being fully understood. Such understanding is critical to design efficient policies mitigating the impact of pollution, e.g. by defining thresholds or reducing the overall exposure.
In this project, we thrive towards a deeper understanding of air pollution using data-driven analysis and statistical methods. Eventually, we aim to provide guidance for policymakers on how to reduce pollution exposure and thereby provide health benefits for city dwellers. To this end, we will apply data analysis, statistical modelling and potentially machine learning techniques.
The PhD student will collect data of various air pollutants, such as nitrogen oxides or particulate matter (PM). As a starting point, London Air provides open data for the Greater London area and more data sources will become available due to international collaborations of the supervisors.
Initially, the data should be analysed in a statistical way: How strong are spatio-temporal correlations of one given pollutant? Does this already allow predictions of when concentrations are going to increase in one location based on measurements of another one? Next, we can move towards the interplay of several pollutants: How are the concentrations of different pollutants linked? Can we identify pre-cursors for the more harmful pollutants by monitoring less harmful ones? What are characteristic speeds with which pollutants propagate through the spatial network?
Complementing this explorative analysis, the PhD student will formulate a data-driven model of pollutant concentrations. Based on preliminary results (Griffin et al 2019), pollutant concentrations seem to follow a superposition of several simple distributions, leading to superstatistics.
We will explore several basic stochastic processes, giving rise to simple distributions, such as exponential or Maxwell-Boltzmann distributions. These local distributions are typically a good approximation for a given location and a small time window of several hours to a few days. Stochastic models producing exponential or Maxwell-Boltzmann statistics will be fitted to the previously explored data to reproduce the empirically observed pollutant concentrations. Aggregating these short predictions over longer periods should re-create the observed pollution statistics with all its rich statistical features.
Depending on the background of the PhD student, we will also pursue machine-learning approaches to directly predict future pollutant concentrations. To this end, recurrent neuronal networks will be trained on past trajectories of pollutant concentrations. Ideally, measurements from the present can then be used to forecast pollutant concentrations for the next hours or days.
The application procedure is described on the School website. For further inquiries please contact Prof. Christian Beck ([email protected]
) or Dr. Benjamin Schaefer ([email protected]
). This project is eligible for full funding, including support for 3.5 years’ study, additional funds for conference and research visits and funding for relevant IT needs. Applicants interested in the full funding will have to participate in a highly competitive selection process.