The H1N1 pandemic in 2009 has been the catalyst which has shown how important locally based human observers are for outbreak detection, verification and pinpoint geocoding. Given the limited resources of public health agencies, the paramount challenge now is to harness the power of Web-based reports. Global bio-surveillance systems such as GPHIN, MediSys, BioCaster and HealthMap trawl through high volume, high velocity news reports in real time and are now seen as adding substantial value to the efforts of national and transnational public health agencies. With concerns about the rapid spread of newly emerging diseases such as A(H5N1), re-emerging diseases such as Ebola and the threat of bioterrorism there has been increasing attention on bio-surveillance systems which can complement traditional indicator networks by detecting events on a global scale so that they can be acted on close to source. The goal of this research project (‘Panda Alert’) is to produce a robust computational model that understands unstructured natural language signals in online news media reports for early risk alerting of adverse health events. Key beneficiaries will be the public health and animal health surveillance communities. The project will build on substantial groundwork already done in the BioCaster project (2006-2012, PI: Collier) and inter-disciplinary collaborations with the international public health community.

The project will involve inter-disciplinary collaboration with experts at the Joint Research Centre (JRC) at Ispra (Institute for the Protection and Security of the Citizen) who provide the European Commission’s MediSys bio-surveillance system. The JRC group will provide (1) expert background information into the context of biomedical surveillance, (2) an introduction to the state of the art techniques used in Medisys, (3) guidance on the selection of real world agency interest for analysis, (4) feedback on system performance.

The Panda Alert project is aligned with DREAM themes in several areas:

1. Approaches and tools for the identification of sources of risks, their drivers and impacts within complex systems.
Previous research by the lead supervisor in the BioCaster project (2006-2012) pioneered the use of time series analysis for alerting epidemics at the country level (e.g. using CDC’s Early Aberration and Reporting method for syndromic surveillance) using hand crafted features mined from news media reports. Similarly JRC’s MediSys pioneered Boolean rule alerting and Harvard University’s HealthMap team has recently applied Incidence Decay and Exponential Adjustment to project growth of outbreaks based on news data. Panda Alert proposes to improve the state of the art in health risk detection by incorporating richer and more complex features into the language models than features which are known a priori to indicate risk to human health (e.g. see the International Health Regulations (IHR) 2005 Annex B decision instrument).

2. Tools for developing, managing and analyzing ‘Big Data’, to understand risk better, and to apply modern cloud computing approaches.
As noted above, our methods will develop the use of statistical machine learning on natural language texts (e.g. recursive neural networks, kernel methods) to classify news reports for public health risk. All tools and data sets from this research will be available for download from public repositories such as GitHub and Zenodo. The software where appropriate will be cloud ready and tested on our own group HPC cluster (SLURM middleware, currently 32 cores, 512 GB RAM expanding soon to 48 cores, 756 GB RAM). The techniques will be tested on real-world high volume, high veracity data in the open source GDELT data set. We estimate that real-time analysis will have to scale to handle approximately 300,000 news items per day in >40 languages with surge capacity during epidemics such as AH1N1 (2009) or SARS (2002). An integral part of the project will be to develop a fully functioning system for real-time alerting and mapping of risk based on the new techniques.

3. New tools and approaches to multi-hazard assessment and interconnected risks (cascade effect).
The use of distributed semantic representations for event reporting brings with it the powerful possibility to generate compositional representations that can be tested for alerting severity. We propose to explore the space of distributed semantic compositions, bounded within a time window, in order to find potential interconnected risks.

Epidemics such as SARS, Ebola and pandemic influenza threaten lives and livelihoods across the globe. Based on the PI’s nine years of experience in epidemic intelligence from social sensors the goal of this project is to build a high quality risk alerting system using distributed semantic representations of news about public health hazards. The PI’s previously reported experiments against human standards have shown that health news is effective at outbreak event detection (F1=0.56) measured against ProMed-mail alerts. The resulting BioCaster system was in regular use by national and international public health agencies such as the World Health Organization. This project aims to build on this legacy by investigating novel methods for event alerting using the latest advances in natural language processing technology by groups at Stanford University (PI: C. Manning) and McGill University (PI: Y. Bengio), i.e. in distributed semantic feature representations and deep learning.

Importantly we expect the distributed semantic techniques we explore to be able to learn novel health risk features using deep machine learning. This is a key step as it avoids the fragility of hand-built feature selection upon which all previous operational systems have been designed. The distributed features will be learnt in an unsupervised mode from high volumes of un-annotated news data over a 10 year period. Examples of a priori features used in existing systems include international travel, infection of healthcare workers and mention of deliberate release of agents. Examples of complex linguistic features we might see learnt could include linguistic clues related to the geography/environment, language of event reporting by officials as well as indicators of economic/social disruption and their combinations. We expect the result to be a system that automatically learns the most relevant clues for disease alerting and which is robust to a range of event types (e.g. early stages of novel asyndromic outbreaks) and reporting styles.

This project will conduct bioinformatics analyses of the genomes of Antarctic fish to assess the risk of the potential for climate change induced extinction of high latitude fish species, and the impact on high latitude Southern Ocean fisheries.

Antarctic fish have evolved in isolation for millions of years in a very stable, near-freezing environment and have developed a number of well-documented adaptions to cope with life in the cold (e.g. antifreeze glycoproteins, large muscle fibres and lack of haemoglobin). Unlike their temperate counterparts, these fish are remarkably sensitive to even small changes in temperature: they are highly stressed by just a 3^oC rise in their environment and die at temperatures 5^oC above normal physiological ranges. There is strong evidence that this may be caused by the instability of the proteins at around 0^oC. It is known that protein folding is more problematic at low temperatures. In the few examples studied to date, amino acid substitutions have been characterized that produce a more flexible protein structure which works more efficiently in the cold. Additionally, in all Antarctic fish species studied to date, the 70-kDa heat-shock protein chaperones (which aid protein folding) are constitutively expressed at much higher levels than their temperate relatives. These organisms also have higher RNA:protein ratios, dedicate more of their energetic budget to protein synthesis, and retain only 30% of the protein they produce, (compared with over 70% for similar tropical species). Taken together, this indicates that protein folding is only marginally successful and that the Antarctic marine fauna are ‘clinging on’ to life in their environment (http://www.sciencemag.org/content/347/6226/1259038.full.pdf).

There has, however, been no comprehensive analysis of the proteome of these species to identify if this protein instability is restricted to certain key proteins or is more widespread. Given that the waters around the Antarctic Peninsula and related island archipelagos (a key fishing ground) is warming at one of the most rapid rates on the planet and that the Antarctic fisheries are the only large-scale exploitation in the region, there is a real risk of extinction for key Antarctic fish species and collapse of the Southern Ocean fisheries industry. mRNA-seq has already been performed using Illumina Hi-seq on 2-12 tissues in different individuals from 8 Antarctic fish species (either in-house or via US collaborators). The aim here is to mine this extensive dataset to identify how extensive cold-adapted changes are in Antarctic fish, perform in silico analyses to determine how these changes affect protein folding and functioning and therefore predict future vulnerabilities.

The transcriptomes of 8 Antarctic fish species (multiple tissues for each species, several individuals for each species and sequenced via Illumina 100bp paired end reads) are already available. These represent a considerable dataset (Tb of sequence data: Big Data). Detailed analysis across species will require the development of bespoke tools to comparatively map both the genes identified in each species with each other, but also to the publicly available genomes and transcriptomes of temperate fish species (e.g. the Japanese pufferfish, zebrafish, tilapia etc). Tools will need to be developed to be able to discriminate between species-specific amino acid changes with more generic cold-adapted changes, thus enabling the evaluation of the level of vulnerability for each species.

The need to get a better understanding of species vulnerability to climate change; to identify how species have adapted to low temperatures and provide background information to allow assessments to be made of sensitivities/resilience to climate change. We aim in future or related projects, to expand this approach beyond fish through to a range of polar invertebrate species to develop more ecosystem level analyses of climate change effects. We want to train high quality mathematically skilled students in the field of bioinformatics, specifically as it applies to extreme environments and non-model organisms. The lack of scientists with such skills currently limits progress in understanding in this field.

The research seeks to identify the extent of cold-adapted amino acid changes in the genomes of Antarctic fish species to enable estimation of their sensitivity/vulnerability in the face of climate change. Additionally identify whether there are particular biochemical pathways affected which could specifically impact on life history traits such as reproductive abilities which could further compromise recruitment and population sustainability.

It is estimated that the world’s food demand will increase by 70 % by 2050. Terrestrial, airborne and high resolution satellite spectral imaging systems can enable effective and timely crop monitoring, helping agriculturalists manage resources more effectively to meet this requirement: monitor crop yields and optimise decisions relating to irrigation, disease control, harvesting, fertiliser and agro-chemical applications. The project aims to i) develop an extensible, framework for analysing data incorporating machine learning and human expertise, ii) using i), build a database of spectral signatures of different plant species, accounting for plant life cycles and health status, and iii) use i) and ii) to provide near real-time summary information about crop status to support on-the-ground decision-making.

The proposed project fits within the remit of big data, since not only will large quantities of data be analysed, but also a sparse model of the data will be developed to aid deployment in the field. The project will develop the technical expertise in data analysis and machine learning as applied in remote sensing and plant sciences. This project aims to be able to alert to risks to crop so that these can be mitigated.

This project is a continuation of two projects undertaken within the MPhil in Scientific Computing at Cambridge University supervised by Faul. One project was entitled “Deep Learning to Exploit Multi-Spectral Satellite Data for Landcover Mapping and Crop Classification”. The other project is ongoing and explores deep neural networks and parallelization, since it is envisaged that large quantities of data need to be processed.

The project also builds on the ongoing collaboration between Coomes and Schönlieb, who have co-supervised one doctoral student (to completion), two postdoctoral researchers and a summer intern student within the broad area of forest assessment from airborne imaging data.

References

ESA’s sentinel satellite constellations will eventually gather 4 terabytes of data per day. This presents an opportunity to extend the automation of landcover mapping and vegetation classification.

The aim is to not just have a database of the multi-(hyper-)spectral signatures of different plant species, but also to incorporate phenology and health status of the plants. With regards to crops this will help mitigate the risks of droughts and diseases. Irrigation can be directed where needed and fertilizer used more effectively. Early intervention can stop diseases spreading and there will be less use of pesticides and fungicides. Being able to choose an optimal time to harvest will lead to less food wastage.

Using sophisticated machine learning techniques, sparse models of the landcover can be created. These models will help where there is limited up- and downlink bandwidth as there are with airborne technologies as well as with satellites. The device gathering the data can carry a sparse model and only where new data is significantly different to the model action is necessary. In the first instance this action will be an alert to an anomaly. Further analysis is then necessary whether the anomaly is expected due to e.g. change of the season, or the anomaly needs intervention or the model needs updating (e.g. a change of crop).

The cilium is present in most mammalian cells and ciliates. Ciliates are an important group of protists (» 3,500 species), common almost everywhere there is water. The cilium acts as an antenna and central processing unit that receives diverse signals from the extracellular environment, such as light, proteins, and mechanical stimuli.

The project will use multi omic data from Antarctic and temperate region living ciliates and available mammalian data on cilia to 1) build the first multi omic metabolic model of a ciliate, 2) integrate cilia mechanics with metabolic models; 3) calibrate ciliate models as environmental change responders on a number of scenarios; 4) identify analogous sensor functions dynamics of cilia in mammalian cells; understand linearity and non-linearity of physiological responses to network variations.

The H1N1 pandemic in 2009 has been the catalyst which has shown how important human observers are for outbreak detection, verification and pinpoint geocoding. Given the limited resources of public health agencies, the paramount challenge now is to harness the power of citizen reports. Global bio-surveillance systems such as GPHIN, MediSys, BioCaster and HealthMap trawl through high volume, high velocity news reports in real time and are now seen as adding substantial value to the efforts of national and transnational public health agencies. With concerns about the rapid spread of newly emerging diseases such as A(H5N1), re-emerging diseases such as Ebola and the threat of bioterrorism there has been increasing attention on bio-surveillance systems which can complement traditional indicator networks by detecting events on a global scale so that they can be acted on close to source. Whilst volume and velocity in Big news Data have to some extent been addressed in the previous projects the problem of veracity remains a challenging issue. The goal of this research project (‘Rumour Mill) is to produce a robust computational model that can measure the likelihood of citizen’s assertions made about breaking news events. Key beneficiaries will be the public health and animal health surveillance communities. The project will build on substantial groundwork already done in the BioCaster project (2006-2012, PI: Collier), inter-disciplinary collaborations with the Joint Research Centre (JRC) at Ispra and a new shared task data set (RumourEval at SemEval 2017).

The JRC group will provide (1) expert background information into the context of biomedical surveillance, (2) an introduction to the state of the art techniques used in Medisys, (3) guidance on the selection of real world agency interest for analysis, (4) feedback on system performance.

The Rumour Mill project is aligned with DREAM themes in several areas:
1. Tools for developing, managing and analyzing ‘Big Data’, to understand risk better, and to apply modern cloud computing approaches.
Our methods will develop the use of statistical machine learning on social media texts (e.g. Twitter) to identify natural hazard rumours and classify the veracity of their assertions. Machine learning methods will include deep learning to provide an integrated representation of words and concepts using distributed vector spaces using samples of language from large scale generic corpora (e.g. Google News), social media authors and their social networks. All tools and data sets (including Twitter messages IDs) from this research will be available for download from public repositories such as GitHub and Zenodo. The software will be tested on our own group HPC cluster (SLURM middleware, currently 64 cores). The techniques will be tested on the 2017 SemEval RumourEval data set as well as a new gold standard set of natural hazard rumours to be constructed within the project. An integral part of the project will be to develop a fully functioning cloud-ready prototype for real-time analysis of rumour veracity based on the new techniques.

2. Measurement, characterisation and handling of uncertainty, including within model chains.
Previous research by the lead supervisor in the BioCaster project (2006-2012) pioneered the use of time series analysis for alerting epidemics at the country level (e.g. using CDC’s Early Aberration and Reporting method for syndromic surveillance) using hand crafted features mined from news media reports. Similarly JRC’s MediSys pioneered Boolean rule alerting and Harvard University’s HealthMap team has applied Incidence Decay and Exponential Adjustment to project growth of outbreaks based on news data. Rumour Mill proposes to improve the state of the art in natural hazard alerting by providing a measure of veracity. Within a deep learning framework, features will include (a) linguistic content from the rumour, (b) the graph structure of the social network involved in spreading the rumour, and (c) real world knowledge graphs, e.g. in the form of BabelNet. Since each new rumour event is unique, the project also expects to look at recent advances in one shot learning using e.g. memory augmented neural networks to try and leverage small amounts of new data for improved rumour veracity classification/quantification performance.

3. Risk perception, communication, decision making and management
After extracting messages that contain rumours about a natural hazard event, risk perception will be done in two stages following the RumourEval share task approach. The first stage will be to determine the stance of the assertion, i.e. whether a rumour about an unsubstantiated event either supports, denies, questions or comments on the rumour. The second stage will be to determine to what degree a rumour about an unsubstantiated event is true.

Epidemics such as SARS, Ebola and pandemic influenza threaten lives and livelihoods across the globe. Based on the PI’s nine years of experience in epidemic intelligence from social sensors the goal of this project overall goal is to contribute to a new generation of high quality risk alerting systems that can incorporate an understanding of the veracity of the event. The PI’s previously reported experiments against human standards have shown that temporal patterns in health news are effective at outbreak event detection (F1=0.56) measured against ProMed-mail alerts. The resulting BioCaster system was in regular use by national and international public health agencies such as the World Health Organization. This project aims to build on this legacy by investigating novel methods for measuring veracity. This will build on (a) recent community challenges in rumour veracity detection in RumourEval, and (b) the latest advances in natural language processing technology by groups at Stanford University (PI: C. Manning) and McGill University (PI: Y. Bengio), i.e. in distributed semantic feature representations and deep learning.

Importantly we expect the deep learning techniques we explore to be able to learn novel veracity features. This is a key step as it avoids the fragility of hand-built feature selection upon which all previous operational systems have been designed. The distributed features will be learnt in an unsupervised mode from high volumes of un-annotated news and Twitter data. Examples of a priori features used in existing systems include international travel, infection of healthcare workers and mention of deliberate release of agents. We expect the result of our new approach to be a system that automatically learns the most relevant clues for assessing rumour veracity and which is robust to a range of event types (e.g. early stages of novel asyndromic outbreaks) and reporting styles.

This study will develop a data-driven regional scale model of the interaction between the physical dynamism of the coastal landscape and human responses as well as drivers of such dynamism. It will achieve this though an integration of state-of-the-art satellite derived earth observation products, aerial photography, and spatial datasets on socio-economic indicators and change to investigate the link between socio-economic and natural system dynamics in the coastal zone. Particular attention will be given to how the modelled dynamism of this society-nature coupled system affects the extent, value, and use of regional natural capital and ecosystem services. Validation/modification will be achieved through a comparison of model predictions with an in-depth analysis of change derived from 26 years of annual aerial photography of the UK coast (Environment Agency photography archive) and historic socio-economic data on coastal protection and other infrastructures and services. The model development will be regionally focused on the UK East coast but this is seen very much as a prototype for future application across the globe.

Some £150 billion of assets and 4 million people are currently at risk from coastal flooding in the UK. Protecting lives, livelihoods and infrastructure with ‘hard’, engineered defences is costly; these costs will rise with climate change and increasing human occupation and use of the coastal zone. Within this context, there has been considerable interest in the role of vegetated foreshores in providing natural coastal protection, both through the dissipation of storm wave energy and in increasing the stability of intertidal surfaces. The UK National Ecosystem Assessment (2011) provisionally estimated that for England this buffering role provides £3.1 – £33.2 billion worth of capital savings in sea-defence costs. However, a much more systematic approach is now required, one which looks at the value of coastal natural capital assets against artificial assets, enabling decision-makers to make a strong case for investment in coastal foreshore restoration and creation. This approach is particularly needed along the low lying, storm surge-threatened coastline of eastern England where coastal habitats are essential to the economy, but are under severe threat from erosion, flooding and climate change (sea level rise and changing storminess).

Machine learning for satellite imagery is now developing into a methodology that can be used for pattern recognition and analysis across large spatial and temporal scales. Historical satellite data allow this to be extended back more than 30 years, while newer satellites with higher resolution and temporal frequency offer the prospect of improved predictive capabilities for estimation of future coastal risk. Tools such as Google Earth Engine now make it possible to apply this kind of learning technique rapidly to petabytes of data and to test the effects of multiple different algorithms, including deep learning frameworks such as tensorflow. At the same time agent-based modelling is maturing as a technique to represent the dynamic effects of human-environment interaction and to assess the likely effects of different impacts on the built environment, using intelligent agents that are able to learn from the past and adapt to future change. Combining these techniques to look at risk in vulnerable coastal environments would allow better understanding of the economic impacts of coastal change processes.

Geological Disposal Facilities (GDFs) are, at present, the preferred option for the safe disposal of high and intermediate level waste. The internationally accepted conceptual model that has been firmly established for the last 30 years is based upon the 'multi-barrier system', whereby a series of engineered and natural barriers act together to isolate the wastes and contain the associated radionuclides. One of the disposal concepts under consideration is the KBS-3V disposal concept, a multi-barrier system consisting of a metal container, copper overpack, buffer/backfill and a high strength natural rock. A key part of this concept is the use of bentonite clay as the engineered barrier, it swells in contact with water to provide a mechanical buffer to protect the metal container and to mop up any radionuclides that may might leak from the container.

Water seeping from the natural geological environment produces bentonite hydration, swelling and the formation of a clay gel that can penetrate into pores and fine fractures of the rock. Chemical or physical erosion processes on the clay surface gel may generate mobile nano-sized colloidal particles which are potential carriers for radionuclide contaminants. Recent studies have shown that the in-situ migration of actinides (i.e. Am and Pu) were strongly facilitated by bentonite colloids. For this reason, it is important to study colloid generation mechanisms, to establish their role on radionuclide transport in the environment and to evaluate the associated risks.

The safety case for the implementation of a GDF requires the consideration of processes evolving over a long timescale, these necessitate the development of sound predictive modelling tools that are based on rigorous short term experiments and sensible extrapolation over the lifetime of a GDF. The technical approach that will be adopted in this study therefore uses a variety of methods (laboratory experimentation and modelling) to build evidence for a greater understanding and use of bentonite buffers that will underpin the required safety case for a GDF. Amongst the questions we will attempt to address are: (1) How is radionuclide uptake and mobility over rock affected by clay colloid particle size, morphology and surface charge? (2) What clay colloid particles are released by bentonite materials, what is their radionuclide uptake and mobility? (3) Can colloid mobility be successfully modelled under different hydrogeochemical conditions at the level of an intact natural fracture core sample in the laboratory? (4) Can this then be represented appropriately in field-scale numerical models?

The reconstruction of the Alberta oil sands represents one of the largest restoration challenges on the planet. Such constructed landscapes must be designed to develop rapidly over time, producing expansive productive forests that maximize the economic potential of the landscape and maintain high water fluxes to flush ecologically harmful salts. At the same time, the landscape design must limit the risk of failure, forming resilient environments which follow planned trajectories within the dry sub-humid climate of the Western Boreal. This research project integrates sound mathematical and statistical analysis of big data to identify the sources of such risk, their drivers and impacts within these complex boreal systems.

A system dynamics model informed from current process based understanding will be developed representing hydrological stores and transfers within the Western Boreal Plain. These landscapes have formed the focus of close to two decades of research by the project partners; legacy data which is being compiled into a readily available data rich achieve (measurements from ~15000 wells in addition to a range of supplementary hydrological, meteorological and ecological data). The models will be optimised and evaluated against this hydrological data. The modelling approach will be developed within STELLA, an intuitive icon-based graphical interface that enables detailed exploration and development of system dynamics. It will offer the opportunity to develop multi-level, hierarchical model structures that can serve as building blocks for large complex systems. The top down approach will enable key leverage points to be identified and the analysis of different system organisations and structures. The modelling framework will be applied to explore how the climate cycle, a superposition of varying climate signals of different intensities and phases, provides the overarching driver of the ecohydrology behaviour of these landscape. Further, how this climate cycle cascades through landscape storage units that vary in configuration, extent and scale of connectivity. It will determine how such complex interactions trigger periods of water scarcity over varying spatial and temporal scales that place individual ecosystem at risk of failure. The project will identify how such collapses and the resultant shift in hydrological function result in a loss of system resilience that may cascade through the landscape leading to its potential large scale failure.

The research will also involve integration within the Water Science group and Birmingham Institute of Forest Research (BIFoR), joining the core team of researchers within the Hydrology Ecology and Disturbance (HEAD3) project funded in partnership by industrial partners Syncrude and Canadian Natural Resource Ltd. This consortium composes of the University of Birmingham, University of Alberta, McMaster University and University of Waterloo.

This project aims to explore the relationship between (tele)commuting and weather. Researchers have spent significant effort in modelling the effects of weather conditions and also extreme weather events on commuting and transport infrastructure. Also, prior research has tried to understand the role that Information and Communication Technologies (ICTs) can perform as an enabling platform for working remotely and avoiding or decreasing physical commuting. This PhD project will build upon these two streams of research and also incorporate a risk dimension which is related to extreme weather and climate change. For instance, changes to the daily commute can be made during extreme weather (e.g. floods, heatwaves, snow), allowing commuters to select the mode of transport which is most resilient for the conditions. At the extreme, telecommuting can be seen as a powerful tool to increase resilience.

Cities are organised in space as complex urban networks, which are connected together through various diverse layers of infrastructure (from transport to digital infrastructure). These infrastructural layers vary from city to city and will affect the capacity of individuals to commute. With respect to telecommuting, the complexity of the above argumentation increases if we consider labour and housing markets. For instance, not every industry can support and take advantage of telecommuting opportunities. Similarly, people whose occupation enables telecommuting may reside in close proximity or in areas of similar socio-economic profile. For instance, problems with Internet broadband connectivity in rural areas might still be a deteriorating factor for working from home and avoiding physical commuting. This PhD project will build upon the above narrative and answer research questions related with the capacity of places and individuals to telecommute, the relation of telecommuting with with weather and extreme weather events, and the link between infrastructure - both digital and transport infrastructure - with telecommuting.

Ice shelves comprise floating extensions of the inland ice of the East, West and Antarctic Peninsula ice sheets. They provide crucial buttressing forces holding back the flow of the ice sheets towards the sea, thus regulating rates of global sea-level rise. In recent years, ice shelves in the Antarctic Peninsula have been observed to substantially retreat and even catastrophically collapse. These major global change episodes have been linked to a variety of causal mechanisms, yet no single clear explanation has emerged, making physically-based forecasts of future change problematic.

This project will provide expertise and training in satellite remote sensing, expert elicitation, Bayesian methods and risk assessment to address the problem of ice shelf collapse. New satellite data (microwave and optical imagery) will be analysed to assess ice shelf retreat and collapse since 2010, placing these new observations in the context of the last half decade of observational ice shelf history. An expert elicitation exercise will quantitatively assess expert opinion of ice shelf collapse risk in the next 100 years.

These datasets will then be combined with existing environmental, geophysical and glaciological ‘Big Data’-sets in a Bayesian nonparametric statistical model framework to calculate the probabilities of ice shelf collapse risk during the next 100 years. The candidate will gain expertise and experience in glaciology, satellite remote sensing analysis, expert elicitation, and Bayesian numerical methods. This combination of skills is unique and in high demand and is expected to result in a number of high-impact outputs. The collapse timing estimates generated by this project may then be used by ice sheet modellers to more accurately forecast the future contributions of the Antarctic ice sheets to global sea-level rise.

The project aims to develop a holistic strategy for UK flood defence assets in order to limit the impacts of fluvial flooding under future climate and land use scenarios. Assessment of the success of international flood defences and their suitability in application to the UK system will be key to the project. The PhD will focus particularly on examples from The Netherlands. Here, the success of Dutch techniques such as the Polder system, Dykes and Dams, and “Room for the River” will be assessed. Application of Dutch water management techniques to the UK system will be considered. Through the use of computer modelling, map analysis, private/public/government and community engagement, adaptations to Dutch techniques will be developed so as to fit with UK hydrological processes and requirements.

Both soft and hard engineering techniques and community engagement will be considered and assessed equally. Through analysis, adaptation and modelling of international measures to limit fluvial flooding, the project will produce a flood defence strategy ready for application to the UK.

To allow for more detailed and accurate analysis and modelling of flood defence recommendations, the project may be developed to focus on a specific area of the UK. However, to account for multi-hazard or cascade effects, the scale should be reduced no further than River Basin level. At this scale, general geology/hydrology/ meteorological etc. trends can be accounted for but the diversity and interconnectivity of processes (e.g influence of tributaries on main channel) is not lost.

Big Data – Meteorological (precipitation, temperature, air moisture etc.), River Stage, Soil moisture and Groundwater data for the study area (River Basin/UK) will all be assessed to determine long term hydrological trends and changes to flood intensity/ frequency. Historical and contemporary map analysis will be undertaken to look at land use change. This will place modern flooding and processes in a historical context and help understand the severity of modern day flood risk.

Datasets available from the relevant UK organisations extend at least 40 years, there is expected to be a significant amount of secondary data to be used within this project. Multi-hazard modelling of future climatic, hydrological and land use scenarios will also be undertaken and will help determine and understand future flood risk. Importantly, multi-defence modelling will also be undertaken to determine the influence of flood defence assets on future flood risk.

Risk – Big Data analysis of historical flooding and modelling of future flooding and map analysis will help visualise risk and put it into historical context. Maps will be produced, indicating the extent of future flooding with and without mitigation and will be available for the study area community to access. Community/private/government organisations will be included in discussion regarding flood defence proposals, the maps and model outputs will improve understanding of future flood risk and will support positive discussion regarding mitigation. The researcher will take on community/stakeholder concerns to develop mitigation recommendations which meet both hydrological and public requirements.

Mitigation - The ultimate aim of the project is to develop a holistic plan for fluvial flood mitigation within a River Basin (RB)/ UK. Recommendations will be based on extensive international research, analysis of long term hydrological trends and future climate scenarios, and modelling of the influence of flood defence assets. Due to the extent of research used in development of recommendations it is highly likely that a successful project will effectively help mitigate against flood risk on a large, River Basin of National scale.

Flooding is the most significant natural hazard in the UK and under future climate scenarios is projected to increase in frequency and magnitude. The extensive environmental, social and economic cost of flooding have driven traditional implementation of point based, hard engineering flood defence structures. These structures are now recognised to limit natural hydrological processes and exacerbate flooding downstream of the structure. There is now a paradigm shift in flood mitigation away from traditional measures towards holistic, catchment based approaches which focus on green engineering and reinstatement of natural hydrological functions. This approach, driven by incentives such as the EU Water Framework Directive and DEFRA Making Space for Water, has multiple hydrological and ecological benefits and is widely perceived as the most sustainable approach to managing catchment water balance and flooding under future climate scenarios.

The UK has made some progress within this field and 12 River Basin Management Plans (RBMPS) are now in place within England, Wales and Scotland. First implemented in 2009 these RBMPS offer the first significant approach to holistic water management however, they are yet to achieve their full potential, localised/disconnected water management and flooding remains a significant issue across the UK.

Innovative and proactive work is needed in order to prepare the UK for more significant flooding under future climate and land use scenarios. RBMPs offer a good base but much more work is needed, flooding and RBMPs need to be highlighted and supported in the public and political domain but in order to do this, each RBMP needs to have clear and achievable objectives. RBMPS and effective flood mitigation are poorly understood, and agencies responsible for implementation of RBMPS are often curtailed by finance, time politics or lack of invention. It is therefore essential that academia supports the development of RBMPS through provision of new concepts and knowledge. Only together will we truly be able to prepare society and our environment for the hydrological challenges that lie ahead.

This research will provide new and tested techniques for sustainable fluvial flood mitigation. Working closely with, and providing recommendations to government partners such as the EA this work will drive the advancement and development of effective RBMPS supporting a holistic approach to UK flood mitigation.

This PhD project will investigate the effect of enhanced carbon dioxide (CO₂) concentrations, caused by anthropogenic climate change, upon ecological networks within oak woodlands. It will characterise the response of ecological networks to CO₂ enhancement by measuring species networks under both enhanced (550 ppm) and non-enhanced (ambient ca. 400 ppm) CO₂ concentrations. This CO₂ enhancement is highly likely to occur in the next 50-100 years.

Oak species dominate the UK’s lowland woods and, like many native trees, are under threat from emerging pests and disease, particularly oak processionary moth (OPM) and acute oak decline (AOD). These new pests and diseases act alongside more established environmental stresses that are often associated with climate change. They add to the constant pressure of established and endemic threats such as oak mildew and honey fungus.

To determine the interaction pathways between oak (and other woodland species), invertebrate pests and tree pathogens, the student will construct the most comprehensive and highly-resolved oak woodland ‘networks of ecological networks’ using state-of-the art molecular ‘community metagenomic approaches. These networks together with environmental data will then be applied in environmental modelling to predict long-term risks associated with tree pathogens.

A ground-breaking ‘Free-Air Carbon Dioxide Enrichment' (FACE) facility, currently being installed in the research forest (Mill Haft) of the University of Birmingham, will be used to investigate the effect of enhanced CO₂ on these sensitive ecological networks. The Mill Haft site enables world-leading scientists to explore the forest thoroughly, taking measurements everywhere from deep within the soil to above the tree canopy. The FACE experiment, applied to oak and other woodland tree species, will provide a formidable addition to the Mill Haft site’s capability to probe the biosphere’s response to increases in atmospheric CO₂.

The transport of people and goods is key to the success of the modern city. Hence good transportation is a pre-requisite for a successful city. However, transport has a darker side in its generation of atmospheric pollutants such as nitrogen oxides and particulate matter. Ambient air pollution is a major risk factor for premature deaths, years of life lost, and years of life lost to disability. The cost of air pollution through loss of life and ability to work has a very detrimental impact upon the worldwide economy; in the UK this economic impact is estimated to be £54B per year (OECD). In particular air pollution is a key risk factor for cancers, lower respiratory infections, cardiovascular and cerebrovascular disease. Worldwide air pollution kills in excess of 7 million people each year (WHO). In the UK, the major emitters of the key air pollutant (PM and nitrogen oxides) are from vehicle use. The Department of Health’s Committee on the Medical Effects of air Pollutants (COMEAP) estimate that long term exposure to air pollutants result in 29,000 deaths in the UK each year and is instrumental in triggering more than 200,000 heart attacks.

This project will investigate the relationships between traffic fleet composition and movement within the city of Birmingham, UK, and how this affects the spatial distribution and concentration of air pollutants. Telematics data will provide real time data on fine grained vehicle flow in the city allowing an understanding of precise acceleration and braking behaviour and behavioural aspects of fine grained mobility.

The objectives of the project are as follows:

1. Identify and combine all pre-existing big data streams of telematics, meteorology and air pollutants for the city of Birmingham.
2. Through advanced analytical approaches, interrogate the dataset for causal links between driving conditions and behaviour with air pollution.
3. Identify locations within the city where interventions, such as changing traffic light design, could lead to lowering air pollution.
4. Disseminate the information to all relevant stakeholders through both traditional academic formats (scientific papers and international conferences) and by engaging directly with stakeholders including BCC.
5. The project will work with the CASE partner ‘The Floow Limited’ who are world leading experts in telematics and the real time mobility understanding of the traffic fleet.

As cities become increasingly congested with growing traffic, the increase in transport-related emissions has raised concerns over the risk on human health and urban environment. Street canyons are hotspots of traffic-related air pollution, because at this spatial level significant variation of traffic volumes and the pollutant transport to receptors (i.e. exposed population) are both at a time scale of minutes. Many emission pollutants are chemically reactive, and fast reactions can take place at second-to-minute timescales to generate secondary pollutants. As a result, the variation of urban air quality can only be understood by considering the combined effects of fleet composition, traffic-produced turbulence, street and building geometries, and meteorological conditions in real time (i.e., a time scale of minutes). Fundamental scientific questions remain, for example, how to model real-time dynamics of the cause-and-effect process from transport activity to distribution of air pollution within street canyons, and how to account for both the heterogeneous distribution of air pollution across an urban street network and its variation in abundance with time. Answers to these questions are of utmost importance to better assess individual or population exposure and the potential risks on human health.

To model this cause-and-effect chain, an integrated modelling approach is built upon bridging research on three complex systems: (i) dynamical changes in travel demand, (ii) vehicle emissions, and (iii) dispersion of air pollutants in street canyons. This approach is thus, not only able to provide detailed indication of air quality in street canyons, but also can identify street sections with high volumes of vulnerable travellers, such as pedestrians and cyclists. These together contribute to a better quantification of potential health risks that could impose on a population in question. Furthermore, travel demand changes respond to different policies, such as speed limit and road pricing in Clean Air Zones; the effects propagate from varying traffic to emissions and cause the change in pollution concentrations. It is through the causes and effects reflected by this model chain that transport policies aiming at risk mitigation can be evaluated in regards to their effectiveness in reducing adverse impacts on human health and environment.

This project inherently deals with issues of Big Data, risk and mitigation in the integrated modelling approach. PhD students will receive related training from three aspects: 1) urban transport and air quality modelling and applications, 2) programming, computing and data handling, and 3) uncertainty and risk related analysis.

Urban shrinkage is a common phenomenon throughout the world despite urbanisation being a well-established trend. With increasing globalisation, cities in both developed and developing counties experience economic downturn, population decline, de-urbanisation. Reasons and solutions of urban shrinkage have been discussed and documented extensively for developed countries (e.g. UK, US, Germany, and Japan). However, deeper understanding of urban shrinkage issues and how to resolve them in the developing world, especially in China with a large number of fast growing cities, is still lacking. Insights from developed countries could be learned in order to better address the challenges for building resilience into shrinking cities of the developing world. Northeast China provinces, including Liaoning, Jilin, and Heilongjiang, now known as the “rust belt” in China, have topped the chart in the number of shrinking cities due to resource depletion, deindustrialization, and demographic changes. Similarly, most of the top UK declining cities are in the north of England as a strong indication of the North-South divide. Core cities of North England, such as Liverpool, Manchester, Leeds, Sheffield and Newcastle, share some common characteristics with their counterpart in Northeast China in terms of industrial legacy, aging population, and loss of growth power to support surrounding areas. Insights could be gained for those cities in both countries by a comparative study of their resilience to internal and external changes.

With a focus on Northeast China cities, this project seek to 1) identify and better understand the spatial, economic and social issues of shrinking cities and the underpinning mechanisms in relation to other Chinese cities, and 2) design adaptive strategies to build resilience into these cities through a comparative study of urban shrinkage in China and UK. This project will expand the existing research by combining the spatial, economic and social dimensions of human mobility and urban interactions and considering the interplay of all three dimensions in defining a multidimensional measurement and assessment of urban resilience. Furthermore, this project will promote the collaboration between the UoB research team and the Chinese stakeholders in order to incorporate local interests and benefit decision-makers with both general and place-based strategies in policy-making.

This project will facilitate the identification and acquisition of various traditional and novel sources of data, which can be leveraged them to gain better insights by leading-edge big data analytics and AI techniques. The substantive and methodological knowledge that this PhD project will generate will directly contribute to the UK Industrial Strategy Grant Challenge of Artificial Economy and the Data Economy as well as on the Key Policies on Infrastructure and Places. Moreover, this PhD project will contribute to the research objectives of the Alan Turing Institute, which the University of Birmingham recently joined. The latter signifies the broader recognition of AI and the Data Economy as a research priority for the University of Birmingham.