New Insights into Geometric Morphometry Applied to Fish Scales for Species Identification

Simple Summary The identification of fish species from a single dermal scale is intriguing in various contexts, including ecology, commerce/forensics, archaeology, and others. While molecular methods serve as the gold standard for species attribution in some cases, they may not always be feasible and, moreover, they are expensive and need specialized personnel. The prospect of attributing fish species through the analysis of dermal scale shape presents a cost-effective alternative that, after the development of dedicated software, could make the process nearly fully automated. The term "geometric morphometry" encompasses a range of techniques designed to quantify and statistically analyze shapes. In this study, after reviewing previous literature on geometric morphometry applied to fish scales, two distinct methods of geometric morphometry were employed on scales from five different fish species. The advantages of both methods were compared. While one method, referred to as landmark-based and commonly utilized in previous literature, was optimized through the introduction of technical enhancements. The second method, known as outline-based and less prevalent in the literature, demonstrated superior performance and holds promise for future automation.Abstract The possibility of quick and cheap recognition of a fish species from a single dermal scale would be interesting in a wide range of contexts. The methods of geometric morphometry appear to be quite promising, although wide studies comparing different approaches are lacking. We aimed to apply two methods of geometric morphometry, landmark-based and outline-based, on a dataset of scales from five different teleost species: Danio rerio, Dicentrarchus labrax, Mullus surmuletus, Sardina pilchardus, and Sparus aurata. For the landmark-based method the R library "geomorph" was used. Some issues about landmark selection and positioning were addressed and, for the first time on fish scales, an approach with both landmarks and semilandmarks was set up. For the outline-based method the R library "Momocs" was used. Despite the relatively low number of scales analyzed (from 11 to 81 for each species), both methods achieved quite good clustering of all the species. In particular, the landmark-based method used here gave generally higher R2 values in testing species clustering than the outline-based method, but it failed to distinguish between a few couples of species; on the other hand, the outline-based method seemed to catch the differences among all the couples except one. Larger datasets have the potential to achieve better results with outline-based geometric morphometry. This latter method, being free from the problem of recognizing and positioning landmarks, is also the most suitable for being automatized in future applications.