This website is archived and is not maintained.
For up to date information on eResearch, go to our NEW Manchester e-Research Centre Site, MeRC http://www.merc.ac.uk

Grid enabled spatial regression models (with application to deprivation indices) (GeSRM)

Principal Investigator - Dr Rich Harris (rich.harris @ bris.ac.uk)

Co-Principal Investigators - Prof. Chris Brunsdon, Dr Claire Jarvis, Dr David Kilham, Edward Thomas

Start Date - May 2006

Duration - 12 months

Location - University of Bristol

Standard indexes of poverty and deprivation are rarely sensitive to how the causes and consequences of deprivation have different impacts depending upon where a person lives. More geographically minded approaches are alert to spatial variations but are also difficult to compute using desktop PCs.

The aim of this project is to develop a method of spatial analysis known as ¡®geographically weighted regression' (GWR) to run in the high power computing environment offered by ¡®Grid computation' and e-social science. By exploring the suitability of GWR to the Grid environment and using it to develop a GWR-based index of deprivation in England and Wales, the project will consider the potential of e-social science for freeing spatial analysis from the shackles of computational constraints.

There are a number of options for enabling GWR on the National Grid Service (NGR), reflecting the availability of GWR source code in R (a statistical and computing environment) and Fortran 77, and the potential to recode (in C/C++, for example). Of these, the R route is attractive: in part because an implementation of GWR already is available as an R package (spgwr: see cran.r-project.org for details); more particularly because of research synergies with the GROWL project ¨C a framework to develop and evaluate Grid services in an environment familiar to many scientific programmers (see, for example, the SABRE-R project ).

In collaboration with the GROWL project team, the idea is to develop a ¡®gapply ()' command similar to R's tapply(), allowing the same function to operate on different groups of data (factors) in parallel, by sending each to separate nodes on a computer cluster. Such a command is of generic application beyond GWR: it can usefully be applied to a variety of computationally intensive, spatio-temporal analyses.

Presentations

Developing Grid enabled spatial regression models
R. Harris, M. Jen, D. Kilham, E. Thomas, University of Bristol
C. Brunsdon, C. Jarvis, University of Leicester
2nd International Conference on e-Social Science, Manchester, 28 ¨C 30 June 2006

Presentation

Publications

Developing Grid enabled spatial regression models
R. Harris, M. Jen, D. Kilham, E. Thomas, University of Bristol
C. Brunsdon, C. Jarvis, University of Leicester
Published in the Proceedings of the 2nd International Conference on e-Social Science, Manchester, 28 - 30 June 2006

Paper