Geostatistics for compositional data with R

Many real world data sets feature positively valued variables that represent parts of a whole and contain only relative information, in addition to this these variables are often location dependent, so that the data set as a whole is spatial and compositional . Variables of this type are common in geosciences, but also in ecology and geography-based sciences. Traditional statistical techniques are not applicable because of constraints on the data arising from their nature. This book provides practitioner with a consistent treatment, with code and workflows to assist in producing internally consistent results, maps and 3D volume models.

This book is a contribution to the UseR series and provides a guide to the use of R for the geostatistical modelling of compositional data. It provides code based on the R-packages "gstat" and "compositions" together with a customized package "gmGeostats", providing an integrated approach to the modelling of compositional data. This consists of the following aspects: decision criteria to determine whether a compositional treatment is required, statistical and spatial analysis of the input data, modelling of the spatial continuity and more general structure analysis, spatial prediction via estimation or simulation, change of support and validation tools. The methods are illustrated via three geochemical data sets, two from environmental geochemical surveys at two different campaigns (Northern Ireland and Australia) and the other from a mining context.