The challenges of regional geochemistry to compositional data analysis from a methodological viewpoint

Regional Geochemistry is an important tool for the detect on of geopotenials (e.g. deposits) and risks (e.g. polution sources) and is an important source of geological insight on large scale. It is mainly concerned with geochemical data, which is inherently compositional. Modern compositional data analysis (CoDa) provides a lot of tools like distribution models, transforms, graphics, compositional geostatistics, imputation, compositional regression and linear models, outlier detection and robustness. However it does not yet provide tools for typical tasks in regional geochemistry, which among others are: maps of single components, anomaly detection and background definition, dealing with below detection limit, dealing with spatially varying geology and land use, working with surveys with too many components to explore all pairwise log ratios, calibration of instruments, collocated compositions, etc.. Ideally CoDa methods should be superior to classical statistical methods for geochemical data and it should thus be possible to simply replace the statistical methods in state-of-the-art geochemical practice by corresponding CoDa tools.The aim of the talk is to give a systematic account of how and why this is not yet occuring. For instance, single components maps are considered a key information in geochemistry, but spurious according to the doctrine of the Aitchison simplex. The compositional alternative would be to work with pairwise log ratios. However such compositional tools have other drawbacks, like e.g. too many pairs, mixing of different information on and no standard literature on their interpretation. Anomaly detection and spa! al factors are not yet sufficiently developed in the methodology research on CoDa. O$ en the application on of standard CoDa tools generates practical problems, like e.g. the identification of anomalies in a multivariate compositon will show a multitude of kinds of anomalies and we are confronted with many different sources and reasons for their occurrence. Par! ally CoDa methods need to be developed for tasks specific to geochemistry, and partly geochemists need to to develop a new thinking for interpreting the results of CoDa methods.From this systematic analysis we have deduced a set of key issues:

• The composition as a whole holds too much information at once. We need efficient methods to extract informative summaries with respect to geochemical tasks. This includes developing readable CoDa graphics and summaries for more than 30 components and multiple layers.
• The single component is understood in a completely different way by geochemists and CoDa-statistians. It is necessary to generate a joint view of this problem and then solve it.
• Enabling CoDa methods for below detection limit and measurement error issues including taylored callibration on for the needs of regional geochemistry.
• High level key publications of the proper use of compositional methods in a regional geochemistry context as reference for future geochemical publications.
• Specialized CoDa based so$ ware for geochemistry powerful enough to replace the existing tools.