Smoothing-Based and Gaussian-Process-Based Methods for Non-Parametric Regression in Environmental Problems

Clive Anderson, University of Sheffield
Vic Barnett, Nottingham Trent University
Adrian Bowman, University of Glasgow
Marian Scott, University of Glasgow
Richard Smith, University of North Carolina
Ron Smith, CEH Edinburgh
Antonia Turkman, University of Lisbon

About:

The workshop provided an overview of the current state of research in smoothing and Gaussian process based approaches to non-parametric regression with applications in environmental sciences. A brief summary of the content is as follows: six speakers gave a series of presentations that reflected new developments in Bayesian and Gaussian process modeling, four speakers outlined the challenges in environmental science; three talks discussed connections between the three main areas of the workshop; and six speakers addressed new developments and applications. For example, Professor Marian Scott and Vic Barnett opened the workshop with a brief introduction to SPRUCE (co scientific sponsor) and to the themes of the workshops, these being then followed by more technical introductions to smoothing, Gaussian process and the nature of environmental problems. Mark Hallard (SEPA), Gavin Simpson (UCL- Environmental change) and Jan Dick and Peter Levy (CEH) all spoke about their needs as environmental scientists. The meeting brought together leading experts in modern smoothing techniques in data analysis, Gaussian process based Bayesian statistical methodology and environmental statistics.

Programme

The objectives were to:

Explore the relationship between smoothing-based and Bayesian-Gaussian-process-based approaches to non-parametric regression and model uncertainty, and assess the potential for transfer of computational tools and theoretical constructs between the two
Confront non-parametric regression methodology with large scale environmental problems of trend detection and assessment, and of model uncertainty, and to formulate proposals for practical solutions
Explore the power of non-parametric regression techniques in spatial and spatio-temporal contexts and consider advantages and disadvantages of the techniques with any computational issues

Discussion themes included:

Challenges in modern smoothing techniques in regression and time series: model-based opportunities?
Gaussian processes and Bayesian methodology: strategies for implementation in high-dimensional environmental contexts.
Environmental science problems and questions needing a statistical solution. Too much data, too little information?
Spatial and spatio-temporal non-parametric modelling
Computational challenges