Paradox of otolith shape indices: routine but overestimated use

Abstract

The identification of fish species using otolith shape has been common in many fields of the marine science. Different analytical processes can be applied for the morphological discrimination, but reviewing the literature we have found conceptual and statistical limitations in the use of shape indices and wavelets (contour analysis), being specially worrying in the first case due to their widespread routine use. In the present study, 42 species were classified using otolith shape indices and wavelets and applying traditional and machine learning classifiers and performance measures (accuracy, Cohen’s kappa statistic, sensitivity and precision). Our results were conclusive, wavelets were a more adequate option for the classification of species than shape indices, independently of classifiers and performance measures considered. The artificial neural network and support vector machine provided the highest values for all performance measures using wavelets. In all cases, the measures of sensitivity and precision pointed out a higher confusion between some otolith patterns using shape indices. Therefore, we strongly discourage the routine use of shape indices for the identification of species.

Use of otolith morphology and morphometry for species discrimination of megrims Lepidorhombus spp. in the Central Eastern Adriatic Sea

“…using classical morphometric shape indices (circularity and rectangularity) has already been confirmed (Tuset et al ., 2008). However, classical shape indices lose much information about the real otolith shape, mainly because the ideal geometric form of the shape indices does not necessarily imply the real description of the object, so the otoliths clearly distinctive from the naked eye can have similar values of shape indices (Tuset et al ., 2021). For this reason, classical shape indices are not recommended any more for works related to the identification of fish species, and the Wavelet analysis, as used in this study, is considered a more appropriate and overall better method (Tuset et al ., 2021).…”

Unravelling Stock Spatial Structure of Silverside Odontesthes argentinensis (Valenciennes, 1835) from the North Argentinian Coast by Otoliths Shape Analysis

“…The shape of the otolith depends on the ecological, evolutionary and phylogenetic characteristics of each species [67][68][69][70][71][72]. This can be evident in coastal species that inhabit dynamic environments, as is the case with O. argentinensis, and can be reflected in the morphometry of the otoliths.…”

“…Tuset et al [69] demonstrated that the application of wavelet coefficients on otoliths is a more appropriate methodological option to classify specimens from different localities than morphometric indices. These results are consistent with what was found in this work, since when the wavelet coefficients were incorporated, the percentage of the classification error decreased, while when only the morphometric indices were considered, the classification error was high.…”

Thinking of Fish Population Discrimination: Population Average Phenotype vs. Population Phenotypes

“…A principal component analysis (PCA) built on the variancecovariance matrix was performed to reduce the wavelet functions with no loss of information on the otolith shape (Sadighzadeh et al, 2012;Tuset et al, 2019Tuset et al, , 2021. Significant eigenvectors were detected by plotting the percentage of total variation explained by the eigenvectors vs. the proportion of variance expected under the "broken stick model" (Gauldie and Crampton, 2002).…”

“…Among the different non-parametric classification algorithms, artificial neural network (ANN) was selected for the comparison of phenotypes among origins due to its high accuracy and commonly used in otolith studies, such as fish classification (El Habouz et al, 2016;Tuset et al, 2021), aging (Robertson and Morison, 1999;Moen et al, 2018), and microchemistry (Hanson et al, 2004;Mercier et al, 2011). This classifier is based on a network architecture, where the smallest unit is the neuron.…”