Replicate sampling along stratigraphic horizons in one outcrop of Ordovician limestones of the Nashville Dome reveals considerable variation in δ13C, δ18O, and Sr concentration among seemingly identical samples. Ten vertical transects were sampled, each consisting of samples taken at 10 cm intervals from 2 m below to 1 m above the M4–M5 sequence boundary of Holland and Patzkowsky (1997, 1998). The 10 transects were each 2 m apart. The resulting 10 replicate and thus seemingly interchangeable samples from each horizon have ranges of δ13C as great as 2.4‰ and ranges of δ18O as great as 2.8‰, and Sr concentrations that vary by as much as a factor of four. Variance is great both just below and just above surfaces of subaerial exposure. As a result, maxima in variance in δ13C, as well as or (in some cases) instead of minima in mean δ13C, can be used to test such surfaces for exposure. Maxima in variance and minima in mean δ13C may be found either below a surface of subaerial exposure in the meteorically altered sediment, or above that surface in the transgressive lag where reworked pre-exposure sediment is mixed with unaltered post-exposure sediment.
The large variances and ranges found in this study, and the dissimilarity of geochemical profiles from the 10 vertical transects, indicate that single samples of stratigraphic horizons, and thus single vertical transects, can yield misleading characterizations of stratigraphic variation of geochemical parameters. Instead, replicate sampling along horizons better characterizes isotopic and elemental variation through stratigraphic sections than does existing single-sample strategies, and it will be more useful in future testing of surfaces of subaerial exposure. For example, the replicate geochemical data used in this project confirm the presence of three surfaces of subaerial exposure, whereas only two had been recognized as exposure surfaces before. Data from replicate samples also make unnecessary the use of arbitrary criteria for inferring exposure, such as the predetermined numerical cutoffs used by Railsback et al. (2003).