Variation diagrams to statistically model the behavior of geochemical variables: Theory and applications

•New variation diagrams to investigate geochemical processes were proposed.•A new method for evaluates compositional changes was developed.•The method has been tested to investigate world rivers chemistry.•Results indicate that only some chemical variables trace weathering processes.•Chemostatic behavior was found an important feature on the observed scale.

SummaryVariation diagrams are revisited from a compositional perspective. An alternative, consistent with the Aitchison geometry of the simplex, is proposed. The new multiple binary diagrams compare a reference balance, b1, fixed on the x-axis, and a set of secondary balances, b2, on the y-axis. The diagrams fulfill compositional principles of invariance under change of scale, guaranteeing consistency of interpretations and working better than diagrams based on raw concentrations of chemical species. The new diagrams allow us to investigate the nature of compositional changes, and inform about the elements or chemical species responsible for the variation when data are a priori ranked following some compositional (or non-compositional) criteria. An application example using the chemistry of large rivers illustrates the procedure. Data are extracted from the Global Register of river inputs, which contains the water quality of about 550 rivers having exorheic basins exceeding 10,000km2 and water discharge exceeding 10km3/yr. The chemistry of only 55 large rivers of the world (in μmol/L, content of HCO3-, Cl-, SO42-, K+, Na+, Mg2+, Ca2+ and SiO2(aq)) as well as their discharge values are considered. The choice is based on the completeness of analytical data and representativeness with respect to water/rocks interaction processes. Results indicate that only the balances b1(Ca2+) and b2(HCO3-) have a high, significant correlation. They are the only two chemical components able to trace weathering processes on this large scale. On the contrary, large data scattering is found in all other cases, indicating a higher sensibility to different environmental parameters, like local lithology, climate or pollution. When all the balances are analyzed as a function of discharge values very low correlation is found, indicating that the chemostatic behavior appears to be a common condition. When samples are ranked for increasing values of b1(Ca2+), or alternatively by discharge values, the statistical investigation of chemical changes reveals interesting features of the analyzed compositions.