Our Researchers

I work in operational research, developing algorithms to solve optimization problems which are then applied to real-world issues. For the last ten years I’ve developed models to solve problems concerning energy and electricity: to use existing networks more efficiently, incorporate renewable energy, such as wind and solar, and store energy until it is needed. I focus on both long-term and short-term problems that, while asking different questions, use the same mathematics. So how does this relate to sustainability? We want to move to a system where we use gas and coal less, but sometimes a lack of sustainable energy sources (for example if there is no wind) means that energy needs to be obtained urgently from other less sustainable sources. This is very expensive, as the short notice means that other power stations or large-scale energy storage units need to be on standby constantly, which drives up the cost. If prices continue to rise, many could resort to generating their own electricity at home, which would further increase carbon emissions. It’s therefore very important to keep the price of energy down, and optimization models play a large part in this.

I’m currently involved in a project with National Grid about how to plan the maintenance of the transmission system, as to maintain the lines you sometimes have to switch the power off! There is the possibility that this will prevent wind generated power from entering the system, which would likely be replaced with fossil fuel generation currently, the problem is about making sure the power keeps flowing, and not distinguishing between different types of power. In a future phase I hope to co-ordinate these actions to prioritise the entry of wind generated energy.

Climate applications often provide a particularly complex environment for having to manage data from different sources and that of course makes it a very good demonstrator for the interlocking of mathematics, data science and climate prediction and energy forecasting, because it's a very rich environment for studying how different kinds of data and modelling need to be brought together.

In terms of understanding how climate model outputs can best be used at decision making, and perhaps the most interesting thing for us, both as researchers and in our work supporting policy and decision-making, we’re interested in what the ultimate best practice here might be? That would mean, how might the next generation of climate models, the next generation of the UK climate projections studies, how those could those be designed to support decision making in a better way across the huge range of applications for which they are used?

The structure of our buildings acts as a sort of thermal storage, and it’s charging cycles can be optimised to achieve any goal. There are many factors which affect heating demand; therefore, we propose an optimisation framework that represents the brain of smart buildings. We call this TEMS (Thermal Energy Management System), and it aims to be an affordable solution to reduce energy consumption and decarbonise cities by providing flexibility to the grid. We also study building refurbishment; this is important because of the need to reduce our energy consumption. We model a framework that outputs the optimal actions, such as optimal values of insulation, to renovate a building. This is intended to be a tool used by both architects and engineers. I was able to test my knowledge in real projects whilst working as an intern with the University of Oxford on project LEO (Local Energy Oxfordshire). The aim of project LEO is to find evidence of the technological, market and social conditions needed for a more sustainable and affordable energy system. Alongside my PhD, I also work as a sustainable engineer intern in the IES company (Integrated Environment Solution). IES is a software and consultancy company that work to make our cities more sustainable.

There are lots of different models and data sets that we can use to learn about the climate and energy use, but to relate those models and datasets to the real-world we need to understand and quantify the underlying probabilistic structure.  That is something that statisticians and mathematicians can contribute to because there are lots of different uncertainties that need to be combined together in a principled way. For example, uncertainties about the inputs to the models; the models themselves; and the data sets (e.g. measurement error).

As an applied statistician I often work with large computer simulators (or models) developed by other disciplines. These are physical models and they are based on very good science, but at the end of the day they are models, simplified versions of reality. If you want to use them to say anything about reality, then you need probability, because you need to understand the link between the two. You can do that if you have real world data, and you have a model because you can compare the two and that tells you something about how close your model is to the real world. But that kind of measure of how close, that's a probabilistic thing and we need statistics and uncertainty to understand that.

The model was also used for a simulation study on the optimal grid resolution of the German monitoring survey; these reports have a major impact on regional and national strategic policy decisions on forest maintenance. My work on this (Augustin et al, 2009) has been successful in the sense that the proposed model produced results which secured the continuation of the monitoring programme, and the methodology is used for official yearly reporting on the forest health status of Baden-Wuerttemberg. In addition, the methodology has recently been applied to forest health data of the whole of Germany for official reporting.

My work in space-time modelling of blue ling, a species of cod, for fisheries’ stock management was a collaboration with scientists at the French Institute for Exploration of the Sea (IFREMER) and resulted in Augustin et al, 2013. The method uses Generalised Additive Mixed Models (GAMMs) similar to the above, but with a soap film smooth which allows us to define aerial boundaries. The methodology has been taken up in fisheries and other areas and I was invited to give a keynote speech on this at a workshop organised by the Center for the Advancement of Population Assessment Methodology (CAMPAM) in California. In addition, the results were used for setting fisheries’ quotas on blue ling.

Augustin, N., Musio, M., von Wilpert, K., Kublin, E., Wood, S. and Schumacher, M., 2009. Modeling Spatiotemporal Forest Health Monitoring Data. Journal of the American Statistical Association, 104(487), pp.899-911.

Augustin, N., Trenkel, V., Wood, S. and Lorance, P., 2013. Space-time modelling of blue ling for fisheries stock management. Environmetrics, 24(2), pp.109-119.

Estimating the population sizes of wildlife species within a given area can be very important for conservation and management purposes. To estimate abundance a variety of different survey techniques may be employed, including capture-recapture-type approaches whereby the population is repeatedly sampled over a period of capture occasions. Models can then be developed, dependent on the specific survey protocol, to analyse the data to obtain an estimate of the total population size, amongst other interpretable quantities. The ability to estimate such quantities can be particularly important for endangered species, in terms of not only providing an estimate of absolute abundance but also to monitor relative abundance over time. Further, estimating the demographic parameters that influence their population dynamics over time (such as birth/death rates and associated factors that may influence these) may help within conservation management and/or assessment of policies. With advancing technology, new survey techniques and associated data are becoming more available such as motion sensor cameras, drones and even earth observation data using satellites. These new forms of data provide new associated statistical challenges, but also the potential for greater insight into particularly difficult to observe populations that are under threat from the many challenges such as habitat loss and climate change.

I am interested in the various mathematical ways of dealing with images and sequences of imaging, to extract information about the scene being imaged. For Earth observation data the image sequences are usually taken from a satellite either working with reflected light (visible or infra-red light like a normal sort of photo), or involve radar where a satellite beams radio waves at the ground and images the reflection. Satellite data of these types is becoming more frequently collected, and higher resolutions are becoming available more routinely. This leads to a proliferation of data to analyse, and enables new types of questions to be answered. This requires expertise from across computing, applied mathematics, statistics and geosciences and sometimes also from an application area specialist. These interdisciplinary teams are exciting environments for mathematicians to work in and contribute to.

Applications of this data I am interested in and have worked on include using satellite data to determine tropical forest loss and degradation, trying to determine areas of tropical rainforest where deforestation is happening. I am also interested in temperature measurements and imaging to determine heat stress on forest or plants that is being exacerbated by global warming. Also I am exploring ways of assessing animal populations (such as penguin colonies for example) using satellite imaging, and linking these to other types of data to monitor the numbers of endangered species.

What does it mean to quantify diversity? Briefly, it is to take a biological community and extract from it a numerical measure of its “diversity”. There are major practical and statistical challenges here, but my focus is on the fundamental conceptual problem, assuming that we have complete and perfect data. In both news media and scientific literature, the most common meaning given to the word “diversity” is simply the number of species present. This measure of diversity makes rare species count for as much as common ones: every species is precious.  Certainly, this is an important quantity. However, it is not always very informative. For instance, the number of species of great ape on the planet is 8, but 99.99% of all great apes belong to just one species: us. In terms of global ecology, it is arguably more accurate to say that there is effectively only one species of great ape. But there is an opposing viewpoint that prioritizes the balance of communities. Common species are important; they are the ones that exert the most influence on the community.

The theory of diversity measurement is driven by both abstract mathematical questions and biological reality. It is the subject of my book “Entropy and Diversity: The Axiomatic Approach”, which builds on the Ecology paper “Measuring Diversity: the importance of species similarity” (joint with Christina Cobbold). In all of this, abstract mathematical aesthetics and faithfulness to biological reality are seen to work in concert, rather than being in conflict.

The kind of data produced by environmental and ecological studies can pose challenges for statisticians. Such data often show a high amount of spatial and temporal autocorrelation, and are subject to various limitations due to difficulties in sampling and data collection, such as ad hoc sampling, missing data, and reliance on proxy variables. Working with ad hoc data collected using different sampling methods was a problem I encountered when working with environmental researchers on a project for the Scottish Government on estimating the number of houses in Scotland with internal lead piping.

Lead is an important environmental contaminant which can be detected in tap water across Scotland due to the presence of internal lead piping in people’s homes. The Scottish Government would like to estimate the cost of replacing internal lead piping to reduce the population’s exposure to lead. We developed a modelling framework to predict the number of houses with lead pipes per postcode using data collated from various sources. We are currently examining ways of improving our predictions by accounting for spatial autocorrelation in a Poisson hurdle model and developing a new sampling regime to validate the model.

Model validation is also the focus of an ecological research project that I am working on with colleagues in the School of Mathematics and Biomathematics and Statistics Scotland. In this project, we have located unique seabird tracking datasets that have been validated using observation data. We are using these data to examine the accuracy of hidden Markov models (HMMs) fitted to GPS data recording seabird movement during foraging trips. This work has important implications for marine conservation as HMMs fitted to GPS animal tracking data are used to identify important offshore foraging areas and advise on the location of marine protected areas and offshore developments.    

Food waste is a huge problem related to sustainability. Recycling is currently the main solution to this, and is done on a global scale. IntelliDigest are developing a bioreactor to convert food waste to useful chemicals. People could have this bioreactor at home and then sell these chemicals to those who could use them, creating an individual solution to the problem. Ideally for this process, the food waste needs to be broken down into smaller pieces to increase surface area and speed up the reaction. IntelliDigest want to understand how fluids with particles of a variety of different shapes and sizes suspended in them, i.e. the food waste, act under different physical conditions. Existing mathematical models assume that all these particles are spherical and of the same size, however different particles may act differently. For example, it is relatively easy to create mathematical models for spherical particles as their geometry remains the same as they spin, and you can easily model spheres bouncing off one another. However, if we consider two lozenge shaped particles (ellipsoids), their geometry changes as they rotate, so the angle at which they collide is important. I am working to extend the existing models to consider a variety of different particles. These models have other potential applications, for example in different bioreactors, to separate materials, or even in the brewing of beer.

Raising cattle is commonly viewed as being bad for the environment, due to emissions like methane released by cows, as well as the deforestation associated with raising them. However, our model of beef production in the Cerrado demonstrated that better pasture management could lead to greater carbon sequestration and reduced deforestation in the context of more intensive beef production. This is because as demand for beef increases, so does the incentive for farmers to take better care of their pastureland. In the Brazilian Cerrado, the grass genus, Brachiaria, which most pasturelands are made up of, is especially effective at absorbing the carbon dioxide in the air and storing it in its roots, due to their depth in the soil and length. This means that well-kept pastureland is more effective at absorbing carbon dioxide. We carried out the study by using our model to simulate different potential scenarios of increased or decreased demand for beef in 2030 and found that if deforestation rates are kept fixed by effective policy, then by 2030, if demand for beef increases by 30%, net emissions could decrease by 10%.  

Articles about our model were published in Nature Climate Change in 2016, and Agricultural Systems in 2017, afterwards being covered by various media outlets. At the time, the media measure for the article was the second highest in that the School had achieved!

If you would like to read more about our study, here are a couple of links to the media pages that covered it:

Carbon Brief; Clear on Climate article. "Higher beef producation could lower Brazil's emissions, study says"

Phys.org article. "Eating less meat might not be the way to go green, say researchers"

University of Edinburgh Research. "Rafael Silva finds beef-producing areas could boost greenhouse gas emissions"