FURTHER INVESTIGATION OF TECTONIC DEFORMATION IN FOSSILS USING GEOMETRIC MORPHOMETRIC SIMULATIONS

The inclusion of tectonically deformed fossils in morphometric analyses can lead to misinterpretations of morphologic data. Here we expand upon previously presented results using simulations to examine the effects of deformation on geometric morphometric datasets. Our simulations are based on a series of 19 landmarks digitized on the plastron of the extant turtle Emys marmorata, which shows a significant pattern of ontogenetic shape change. Using a small and large turtle as reference points, we generated simulated datasets of 150 specimens with a strong ontogenetic signal. Varying amounts of uniform shear were added to the datasets to simulate the effect of tectonic deformation. In addition to the amount of deformation, we also varied several other simulation parameters, including the range of orientations of the specimens relative to the direction of applied stress and the ratio of deformed and undeformed specimens in the datasets, to examine the effects of deformation in more complex situations.

In general, the amount of variance in a dataset increased rapidly as deformation increased, and the covariance structure of the deformed dataset became progressively less similar to that of the original. It usually was difficult to reject the hypothesis that the ontogenetic signal in the deformed and original datasets was the same, but this seemed to result in part from the increased noisiness of the deformed datasets. MANOVAs usually could not discriminate the original and deformed datasets at low levels of deformation, but discrimination improved slightly with greater deformation. An exception to this generalization was observed in the datasets with limited variation in the orientation of specimens relative to the direction of applied stress. This may result from the fact that the deformation-induced variance reflected the non-random bias relative to cases where the specimens were randomly oriented relative to the applied stress. Attempts to remove the effects of deformation using four published methods generally were not successful. All the retro-deformation methods reduce the total amount of variance in the data, but they uniformly suffer from either removing too much or too little. Furthermore, none significantly improve the similarity of the covariance structure of the deformed dataset relative to the original.