Geostatistical Compositional Analysis of Regional Geochemical Stream Sediments of West Java, Indonesia

As analytical tools evolve, more geochemical data are obtained so that effective and robust tools are demanded for capturing detailed information of underlying geological processes such as alteration, mineralization and weathering. The data are formed by a number of elements, whose concentrations are interdependent due to the fact that their total sum is bounded to a constant (e.g. 100%). Classical statistical analysis of this type of data may provide misleading results because it does not consider the closure effects: for instance, one of the correlation coefficients will always be negative regardless of whether the two elements with negative correlation are incompatible or not. Consequently, methods employing variance cannot be used on the raw dataset. To manage this problem, one can adopt Compositional Data Analysis (CoDa) which acknowledges the compositional nature of the data.

In the last decade, CoDa has been widely applied to geochemical data. Thus, combined with geological information, we apply CoDa to the regional geochemical stream sediment data of West Java to address our understanding of the underlying geological processes of the area. The first process carried out is a log-ratio transformation to eliminate both constant-sum constraints and non-negativity problems. Then to distinguish features in the data a Principal Component Analysis (PCA) can be performed. Here we can get linear components that have a large effect on variations in the data. The final stage is the spatial estimation of important PCA components through experimental variogram calculation, variogram modeling, and geostatistical estimation. The interpolated maps can be back-transformed to obtain maps of the original components. In this way, the results will not violate the compositional nature of the data, while spatially representing the captured processes.