Maximum-likelihood identification of fossils: taxonomic identification of Quaternary marmots (Rodentia, Mammalia) and identification of vertebral position in the pipesnake Cylindrophis (Serpentes, Reptilia)

Abstract

We applied a Maximum-Likelihood (ML) criterion to the problem of identifying unknown specimens using a database of specimens whose identity was known. Our approach was based on shape, quantified using two-dimensional Cartesian landmarks. We applied the technique to two specific problems: (1) identifying Quaternary marmot skulls (Marmota, Sciuridae, Rodentia) to species, and (2) identifying the position of individual elements within the vertebral column of the Red-tailed pipesnake, Cylindrophis ruffus (Serpentes, Alethinophidia). The ML criterion finds the best identity by choosing the sample that best fits the unknown. Cross-validation tests indicated that identifications of unknown marmots were correct about 80%-90% of the time. Fossil marmots from two sites (Meyer Cave, Illinois and Little Box Elder Cave, Wyoming) could be assigned to species (M. monax and M. flaviventris respectively), but marmots from several other localities could not be assigned to a species-level taxon. Snake vertebrae could be allocated to their proper columnar interval more than 80% of the time, with incorrect assignments rarely being more than 10% out of place. Our technique is widely applicable in palaeontology, where the problem of identifying isolated morphological elements can be acute but is often ignored. Our approach allows palaeontologists to base their identifications securely on their morphological data, and to recognize conditions under which a confident identification can or cannot be made.