Choosing wavelet methods for otolith contour studies

Abstract

Otolith shape analysis has been extensively employed to distinguish stocks and populations of marine fish species. While the majority of studies have employed Elliptic Fourier descriptors (EFD) for this purpose, an alternative approach was introduced in 2005 founded on the Wavelet Transform, using the à trous multiscale signal representation with a B3-Spline function. This approach not only improved the biological and mathematical interpretation but also empowered the user to select a more suitable level based on the contour complexity. Recently, the global adoption of the mathematical R environment enabled the creation of a freely accessible package called shapeR for otolith shape analysis using Daubechies least-asymmetric wavelet. Nevertheless, we have pinpointed certain inconsistencies in this package concerning the biological implications of its results and a deviation from the original application philosophy. To illustrate these inconsistencies, we conducted a study aiming to differentiate populations and morphotypes of the blue jack mackerel, Trachurus picturatus, from the Canary Islands and Madeira. We employed both shapeR package and our original method in the comparison. Furthermore, we evaluated the performance of different parametric and non-parametric classification algorithms using wavelet coefficients obtained from both methodologies. This study has shown that both methods are suitable for population identification, achieving high classification accuracy. However, some inconsistencies were observed between the graphical representation of the intraclass correlation plot and the mean otolith contour reconstruction using the shapeR package. AFORO demonstrates a particular strength in capturing morphological changes, allowing for better identification of contour shape variations along the wavelet signal. Additionally, both methods identified the same number of morphotypes within the overall sample, albeit with differing proportions.