Geometric morphometrics, neural networks and diagnosis of sibling Taterillus species (Rodentia, Gerbillinae)

Abstract

The lack of diagnostic traits in sibling species means that great quantities of biogeographical and ecological data held in museum collections cannot be utilized effectively, leading to underestimates of biodiversity. In this study we applied neural networks (NN) and canonical variates analyses (CVA) to landmark measurements on the skulls of the West African species Taterillus arenarius, T. petteri, T. gracilis and T. pygargus in an attempt to discriminate species previously identified unambiguously from their karyotypes. Among suggested differences, the relationship between inflation of the tympanic bullae and lower population density is discussed. Cross-validated classification rates did not exceed 73%. Two hypotheses are proposed to explain such high phenotypic similarity: morphological plasticity limited by environmental constraints and the possibility that speciation has been too recent to allow significant morphological divergence. © 2002 The Linnean Society of London.