An experimental comparison of cokriging of regionalised compositional data using four different methods: case study3A Bauxites in Hungary-2

Abstract

An important problem in the geosciences is the etimation or prediction or regionalized compositions. In fact, it is usual to deal with data such as percentages, concentrations, ppm,..., and use them to estimate values in other locations. Compositional data have been regarded as difficult to work with because of the so-called constant sum constraint. Following Aitchison(1986), any meaningful satatement abaut a composition can be expressed in terms of logratios, but those transformations, and their backtransformations, are not always easy to deal with. The aim of this paper is to compare results obtained applying different methodologies developed in geostatistics, with samples of compositional data from a bauxite deposit in Halimba II (Hungary). Firstly, a classical geostatistics study is done usin raw data; secondly applyin two wellknown transormations in compositional data analysis: additive logratio (ALR) and centered logratio (CLR); thirdly, the Fast Fourier Transform (FFT) methodology to calculate the spatial variance-covariance matrix is used in cokriging. To be able to compare predictive values and kriging erros respective backtransforms are found. At last, results obtained with the different approaches are discussed and compared.