A BAYESIAN METHOD FOR IDENTIFYING GRAPTOLITE BIOTOPES

For decades paleontologists have recognized that some graptolite taxa (marine pelagic zooplankton of the lower Paleozoic) exhibit recurrent species and lithofacies associations. These groups appear to be ecological in origin. A mesopelagic group, associated with dark shales, probably inhabited the dysaerobic zone, and an epipelagic group, associated with both dark shales and shallow-water facies, probably occupied the photic zone in more oxic settings. Recent work has demonstrated that the two groups differ significantly in their extinction risks, geographic ranges, and sampling frequency, all of which affect their use in biostratigraphy. However, only a small number of species have been assigned to these groups, and those assignments rely heavily on subjective interpretations of the depositional environment represented at individual collection sites. Therefore, different authors have had varying opinions and quantitative assessment of the assignments has been difficult.

We have developed a quantitative, Bayesian method for biotope assignment. Sites are assigned an a-priori probability of being shallow that can be allowed to vary or remain fixed through a run. Taxa in the analysis begin with an equal probability of membership in each biotope and are allowed to switch between the two biotopes searching for stable positions based on likelihood scores. Taxa receive an a-posteri probability based on the proportion of time spent in a particular biotope group. We used a database of 151 graptolite taxa from 30 Middle Ordovician sites in Baltica as a test set for this method. To reduce the effect of uninformative absences, we placed sites into an ordinal composite timescale so that only sites within a taxon’s temporal range were used in the calculations. When site assignments were held fixed approximately two-thirds of taxa were found to remain in a single group at least 90% of the time and thus could be assigned reliably to a biotope. Comparison of results with previous assignments based on expert opinion show significant agreement, suggesting the Bayesian method is able to recognize biotopes with very limited lithological data. The method yields reproducible and explicitly justified results that are suitable for statistical hypothesis testing, including hypotheses about site conditions.