Influential landmarks

Geometric morphometrics based on two-dimensional landmarks is a powerful tool for distinguishing morphologically similar or cryptic taxa of 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. We demonstrate that small subsets of landmarks can outperform the full landmark sets in species discrimination. Examples are given in 6 insect families: Culicidae, Glossinidae, Muscidae, Psychodidae, Reduviidae and Tabanidae. We compared the classification accuracy of complete landmark configurations (from 10 to 22 landmarks) with smaller subsets of them. 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.