Geometric morphometrics analysis of the pitcher shapes variation among the species of Nepenthes L. (Nepenthaceae)

Abstract

Nepenthes pitcher shapes are diverse and important for identifying the species. However, species dentification often found difficulties in understanding the lexical description of this pitcher characteristic. Modern morphometrics encompasses an array of mathematical approaches that turn shapes into quantitative variables. This study aims to analyze quantitatively the variation of pitcher shapes among Nepenthes species, the similarity between hybrids with its parents, and the comparison between the images from herbarium materials with the images from living plants using geometric morphometrics. And to development of an automated identification system using machine learning algorithms. Geometric morphometrics analysis was conducted in three main steps: feature extraction using elliptical Fourier (EF) analysis for obtaining the quantitative data from the outline images, statistical analysis for evaluating the variation of pitcher shapes among seven Nepenthes species, the similarity of hybrids with its parents, the differences between herbarium materials and fresh materials, and building an automated identification system using machine learning algorithms (linear discriminant analysis (LDA) with leave-one-out cross-validation (LOOCV). A total of 252 images of living pitchers, taken from three different perspectives, top view, front view, and side view were used in this study. We used five species (N. gracilis, N. mirabilis, N. reinwardtiana, N. bicalcarata, and N. rafflesiana) and two hybrids (N. mirabilis x N. rafflesiana and N. x hookeriana) from Central and West Kalimantan. This study shows that geometric morphometrics, with an outline analysis using Elliptical Fourier Transform, capable of distinguishing Nepenthes species and offer a powerful approach to provide an objective description of pitcher shape. The scatter plot of the PCA shows that the shapes are separated and grouped based on species, even there is still little overlap among species. Analysis Multivariate of variance (MANOVA) shows that all of the pitcher shapes among species on the side and front view are distinct with a value of MANOVA (p <0.05), except N. mirabilis x N. rafflesiana and N. reindwardtiana on the side view (p> 0.05). At the same time, the lid (top view) of some species is not different (p> 0.05). The hybrids analysis shows that N. mirabilis x N. rafflesiana has an intermediate pitcher shape between its parents (N. mirabilis and N. rafflesiana). Meanwhile, the images from herbarium materials are entirely different from the images from living plants. An automated identification system using LDA with LOOCV achieves an accuracy of identification as high as 84.5% (side view), 90.5% (front view), and 85.7% (top view). The current analysis shows that the front view is the best character that can be us as a taxonomical feature. The system developed in this study is able to classify the Nepenthes species according to pitcher shape with acceptable accuracy.