Fig 1. Schematic diagram of integrative taxonomy applied to Ophiothrix spp. (A) A priori classification inferred from morphological characters. (B) Morphological characters (B1) external morphology and (B2) arm microstructures morphology. (C) Morphometry: measurements and linear discriminant analysis (LDA). (D) Molecular characters: (D1) phylogenetic analyses (Maximum Parsimony, Bayesian Inference, and Maximum Likelihood) and (D2) genetic diversity and species delimitation test (genetic distances, AMOVA, haplotype network, bPTP). (E) Congruence framework of integrative taxonomy. https://doi.org/10.1371/journal.pone.0210331.g001
Figure 5. Geographical distribution of Plagiobrissus grandis in America. The dots represent the previous and new records of the species. Map background generated with package “ggmap” using Google Maps API. http://dx.doi.org/10.11606/1807-0205/2023.63.026
Fig. 3. Schematic diagram of the statistical analysis using the machine learning approach. Step A: A priori classification of quadrat images into one of six subenvironments: mesolittoral (ML), drift zone (DZ), supralittoral (SL), incipient foredunes (IF), established foredunes (EF), and restinga (RE). Step B–D: Pre-processing of numeric predictor variables to select the final model: sub-environments ∼ sector (SEC) + local slope (SLO) + segmented cover vegetation (SVC) + drift cover rate (DCR) + entropy of texture (ENT) + average vegetation height (AVH) + number of sites with sand or other marine litter (NSC) + number of points with Ipomoea pes-caprae (ipe) + number of points with Blutaparon portulacoides (blpo) + number of points with Panicum racemosum (pan) + number of points with herbaceous plants (herb) + number of points with grass plants (grass) + species richness (S). Step E: Implementation of the candidate machine learning algorithms: classification and regression trees (rpart), random forests (RF), stochastic gradient boosting (GBM), support vector machines with radial basis function kernel (SVM), C5.0 decision trees and rule-based models (C50), penalized multinomial regression (MREG), variational Bayesian multinomial probit regression (VBMP). Step F–H: Computation of statistics to aid selection of the best algorithm. Step I: Assessment of the results from the best-performing machine learning algorithm. http://dx.doi.org/10.1016/j.ecss.2018.04.030
