- Spatially-explicit landscapes IBMs can show how populations change over time in response to management actions
- Development of new Monte Carlo techniques for calibrating IBMs using state-of-the-art statistical techniques
- Use of machine learning methods for reducing computational cost of Monte Carlo approaches
Individual-based models (IBMs) are used to simulate the actions of individual animals as they interact with one another and the landscape in which they live. When used in spatially-explicit landscapes IBMs can show how populations change over time in response to management actions. For instance, IBMs are being used to design strategies of conservation and of the exploitation of fisheries, and for assessing the effects on populations of major construction projects and of novel agricultural chemicals.
There is urgent need to improve methods of calibrating such models: existing methods are too slow, and not always accurate. This project aims to improve the best existing method: Approximate Bayesian Computation, ABC. ABC is currently being used at Reading for statistical inference in a diverse range of applications in ecology, evolution and more widely, including for example: models of elephants in Amboseli; mackerel in the North East Atlantic; local butterfly populations; but also evolution of pathogens; social network analysis; and statistical physics (see Didelot et al. 2011; Prangle et al. 2016; van der Vaart et al. 2016). In most of these cases the challenges of parameter estimation and model comparison are both of importance, but implementation can prove computationally expensive. This project aims to improve ABC methods and to collaborate with model builders to help them in fitting models to data. Initial focus will be on IBMs developed for fisheries management by CEFAS, part of the UK government, https://www.cefas.co.uk/.
ABC compares model outputs with data and is particularly useful for statistical inference where the model is only available as a computer simulator such as an IBM. ABC is a relatively new field of research, and is a hot topic in statistics and several applied fields (Beaumont 2010). There are many open problems in this area, some of which lie at the heart of this project, including:
ABC for high-dimensional parameter spaces. IBMs often have more than 10 parameters that have to be estimated by fitting the model to data: more than in many current applications of ABC.
ABC for computationally expensive simulators. Some IBMs take several minutes to complete a run. This is a problem because existing ABC methods require thousands of runs to obtain reliable results.This project will develop new methods to address these issues, driven by the need for accurate fisheries models to guide fisheries management.
This project will develop new methods to address these issues, driven by the need for accurate fisheries models to guide fisheries management.
Training and Skills
The student will spend time at CEFAS learning how models are used in managing fisheries. At the University of Warwick, the student will learn to program in several languages such as R or python, and to develop new methods in mathematics and biology. They will be part of the Department of Statistics at Warwick, which contains one of the leading groups in computational statistics and machine learning in the UK. The student will become an expert in this field, aided by participating in the departmental training for new PhD students. This training includes focused study groups, and broader seminar programmes.
Year 1: Gain familiarity with models and methods to be used in the project. Perform initial work on calibrating sea bass IBM using existing approaches, paying particular attention to the deficiencies of these approaches. CEFAS will support training in use of the packages and models and to understand the ultimate aims for the models that we are developing. In addition, the student will be encouraged to meet and interact with other marine scientists, learn about the science policy interface and have access to CEFAS training courses.
Year 2: Develop new methods for calibrating IBMs. The main focus is on the following three areas:
Expensive simulators. Develop methods that are computationally feasible despite the use of an IBM that takes a long time to simulate.
High dimensional ABC. Investigate the accuracy of existing approaches to high dimensional ABC, developing improvements where necessary, particularly to enable model comparison.
Estimating model error. To develop methods for estimating statistical models for the error in IBM models; to gain an understanding of this error, and thereby to improve the accuracy with which models are fitted to data.
Year 3: Apply methods developed above to IBMs developed at CEFAS and Reading, including complex multi-species models. Assess the accuracy of their estimates of posteriors using coverage and show how the uncertainty of predictions can be described. Investigate the use of our new methods in other applications such as data assimilation, as used in weather forecasting. Deploy our recommended methods to the ICES secretariat and other agencies.There is additional funding provision for the student to spend up to 3 months each year at CEFAS.The student would take advantage of this when close collaboration with CEFAS would be desirable.
Partners and collaboration (including CASE)
CEFAS is an executive agency of DEFRA and the UK's most diverse applied marine science centre with over 500 members of staff, including modellers, economists, biologists, chemists, physicists, social scientists, and engineers. Cefas helps shape and implement marine policy through internationally renowned science and collaborative relationships that span the EU, UK government, non-governmental organisations.
The student will collaborate with Prof. Richard Sibly, University of Reading: an expert in behavioral and physiological ecology and IBMs.
The supervisory team have been collaborating over >3 years on several projects, e.g., 3 NERC SCENARIO CASE studentships. Current projects encompass IBMs of Mackerel and Bass.
This project has been selected as a CENTA Flagship project. This is based on the projects fulfilment of specific characteristics e.g., NERC CASE support, collaboration with our CENTA high-level end-users, diversity of the supervisory team, career development of the supervisory team, collaboration with one of our Research Centre Partners (BGS, CEH, NCEO, NCAS), or a potential applicant co-development of the project.
Dr Richard Everitt
Department of Statistics,
University of Warwick,
Coventry, CV4 7AL