Comments on “Influence of Autocorrelation and Measurement Errors on Interpretation of Solubility Diagrams”

Recently Xu and Harsh (1994), in discussing the influence of autocorrelation and measurement errors in the interpretation of Al solubility diagrams, referred extensively to a study (Neal et al., 1987) that raised concerns about the use of such diagrams for assessing thermodynamic equilibrium with regard to Al hydrochemistry. Neal et al. (1987) were concerned that the use of plots with dependent axes might produce correlations that have nothing to do with any underlying chemical process. Xu and Harsh (1994) argued that the scatter in the Neal et al. (1987) approach might be overly large and so might distort the true picture. They concluded that, when used with caution, existing solubility methods are reliable and useful in the study of Al chemistry in acidic soils and streams. The Neal et al. (1987) study cannot be construed as a definitive statement on the hydrochemistry of Al in acidic waters. Rather, the work highlighted problems over the thencurrent understanding of Al solubility controls, particularly with regard to stream water. The work not only raised a danger flag but was intended to stimulate development of more rigorous thermodynamic analytical techniques. More importantly, it was also undertaken to encourage further process studies on Al hydrochemistry for acidic soils and streams impacted by acidic atmospheric deposition and land use change. Part of the work of Xu and Harsh (1994) thus represents a natural progression from the Neal et al. (1987) study as well, of course, as progression from other work such as Schecher and Driscoll (1987, 1988). Indeed, several facets of the Xu and Harsh (1994) work are commendable, particularly in regard to assessing error uncertainty, firming up on what "near equilibrium" means, and the use of Monte Carlo simulations, and some of their criticisms of the Neal et al. (1987) study are fair. For example, the scatter in the Neal et al. (1987) data synthetically generated might be larger than observation because the total Al numbers were not generated using a transformation to ensure a normal distribution compared with field observation. During the initial part of the Neal et al. (1987) study, an analysis of concentration distributions was undertaken for field data at Plynlimon. Skewness was observed for both untransformed and log-transformed total Al data — a normal distribution lay somewhere between and it was not clear whether untransformed or log transformations were best. But, this was not reported in Neal et al. (1987) as they considered this aspect to be only an aside to the major thrust of the study. Had work not progressed following the Neal et al. (1987) study, the Xu and Harsh (1994) study would have been of particular significance. However, there are several studies, published in international journals during the late 1980s and early 1990s, which took the Neal et al. (1987) work further but which Xu and Harsh (1994) seem to have missed. These studies demonstrated: