Geometric morphometrics based on two-dimensional landmarks is a powerful tool for distinguishing morphologically similar or cryptic taxa, an important asset in the fight against medically and veterinary important arthropods. While it is commonly assumed that increasing the number of landmarks should improve discriminatory power by capturing more shape information, our findings challenge this assumption. In terms of shape discrimination, we demonstrate that small subsets of landmarks can outperform full sets of landmarks. Examples are given in 6 insect families: Culicidae, Glossinidae, Muscidae, Psychodidae, Reduviidae and Tabanidae. In all of these examples where landmark-based geometric morphometry was effective in separating morphologically close taxa, the total number of landmarks was not as effective as some significantly smaller subsets. To find such performing subsets, we used a random approach. Thus, for each number of landmarks (subsets), we examined a random sample of their many possible combinations. This random search was compared to a simpler approach, called the hierarchical method, based on the contribution of each landmark to the overall distance between shapes. Both procedures have been integrated into the XYOM online software, providing accessible tools for efficient landmark selection and improved morphometric analysis.
Author summary
Landmark-based geometric morphometrics describes shape in direct relation to the number of landmarks used. It is commonly assumed that increasing the number of landmarks allows for more information about shape, and when discriminating between groups or taxa, this strategy is expected to improve classification accuracy. Our results challenge this assumption. We demonstrate that subsets of landmarks, as small as three or four, can outperform the species classification obtained by the full set of landmarks, and we propose two methods for identifying them. We analyze the possible causes of these counter-intuitive results and the perspectives they could open for morphometric studies.