Date & time: Tuesday 18 August 2015, 09.00-16.30
Objective: Following a brief overview of Gaussian random fields and geostatistical models, facilities in the geostatsp package for working with these models in R will be demonstrated. Both Gaussian models (maximum likelihood estimate, Kriging) and non-Gaussian models (Bayesian inference using INLA) will be considered.
General description: Geostatistical models assume that data observed at different locations are generated from the sum of: the combined effects of know explanatory variables (or covariates); a continuously varying random spatial surface; and micro-scale variation or observation errors. Analyzing a spatial dataset with a geostatistical model involves estimating the model parameters (coefficients on covariates, spatial range, variances) and making spatial predictions of the underlying surface throughout the region of interest. Inference on Gaussian (or continuous) data can be done using Maximum Likelihood Estimation, where as non-Gaussian (discretely-valued) data requires models for which Bayesian inference is well suited. Facilities in the geostatsp package will be demonstrated on datasets including arsenic contrentations in groundwater, locations of murders in Toronto, and coffee consumption.
Required back-ground knowledge:
Software / R packages required: Basic understanding of statistical modelling, maximum likelihood estimation will be assumed and familiarity with Bayesian inference would be helpful. Knowledge of the R functions lm and glm, and familiarity with the sp package will be assumed.
Materials / slides:
Brown, PE. (2015).
“Geostatistics the easy way”.
Journal of Statistical Software 63.
Diggle, Peter, and Paulo Justiniano Ribeiro. ( 2007)
Springer Science & Business Media,.