Population structure of Japanese Spanish mackerel Scomberomorus niphonius in the Bohai Sea, the Yellow Sea and the East China Sea: evidence from random forests based on otolith features

Abstract

demographic properties, life history pattern and response to exploitation. Ignoring underlying spatial stock structure in a fishery management system can increase risk of local depletion, resulting in loss of genetic diversity [1][2][3]. Therefore, a thorough understanding of stock structure is the starting point of effective management in multi-stock fisheries [4,5].Otolith shape is subjected to a combination of genetic and environment effects and demonstrates stock-specific features [6]. It has also been deemed to be an ideal natural marker for spatial stock, and is widely used in stock identification. The introduction of multiple image analysis techniques improves the ability to visualize and compare otolith shape, such as Fourier analysis [7,8], curvature scale space [9], wavelet transform [10, 11] and partial reflection [12]. However, few efforts have been put into the improvement of classification methods, whereas machine-learning methods (e.g. random forests, artificial neural networks) have been applied in fish parasites and otolith microchemistry, and have outperformed discriminant analysis [13][14][15]. Discriminant analysis (linear, quadratic and canonical) is commonly used in population studies based on otolith shape, and multivariate normality and independence of the predictor variables are required for this method. Appropriate data transformations (log, square root, etc.) are often required to reach a normal distribution. As to multi-collinearity, there are two options: a principle component analysis or excluding variables with significant multi-collinearity. Algorithmic advantages of random forests enable researchers to eliminate the above steps, maximize the number of variables inputted for discrimination and retain the original information of each variable. However, an increase in classification accuracy has its own limits with the variables used, and many variables increase intrinsic classification accuracy but also decrease extrinsic accuracy in Abstract In this study, the population structure of Japanese Spanish mackerel Scomberomorus niphonius were investigated based on the geographical variation of otolith features, and the concept of random forests was introduced as a classifier. Samples were collected from eight main spawning grounds using commercial gill nets, covering the distribution range along the coast of China. Otolith shape was described by the shape variables and principle components of Fourier coefficients. Every possible combination of otolith variable was tested to search for an optimal variable combination. An intermediate number of variables (8 out of 13) produced the most discriminating signal, and then the eight grounds were divided into three stocks with 64.5 % global accuracy. The discriminant result provided phenotypic evidence for the common migration trajectory, and further confirmed the existence of a metapopulation in Japanese Spanish mackerel at a larger scale.

Otolith microchemistry: a useful tool for investigating stock structure of yellowfin tuna (Thunnus albacares) in the Indian Ocean

“…Rooker et al 2008bRooker et al , 2016Wells et al 2012;Macdonald et al 2013). However, in recent years, machine learning techniques have emerged as promising classification tools in otolith-related studies (Zhang et al 2016;Tournois et al 2017;Bouchoucha et al 2018). The accuracy of each technique will depend on the nature of the data analysed, and the results of this study agree with those of recent studies encouraging the use of machine learning methods when otolith chemical data are not multivariate normal or exhibit skewed distributions (Mercier et al 2011;Jones et al 2017).…”

Assessing the performance of statistical classifiers to discriminate fish stocks using Fourier analysis of otolith shape

“…S1 and S2 1 ). Following Mercier et al (2011) and Zhang et al (2016), a fourfold cross-validation resampling method was used to provide the data for the assessment of the performance of each classifier. This validation method is advised as a reasonably stable and low biased measure of model performance (Hastie et al 2009), but typically indicates lower accuracy of the evaluated algorithms than most often applied leave-one-out cross-validation.…”

“…Studies that applied only DA constituted ϳ92%, while one study (<1%) used DA and RF in parallel (Jones and Checkley 2017). The remaining ϳ7% of the publications applied classifiers other than DA to assign samples to their respective class (e.g., SVM or KNN classifier (Reig-Bolaño et al 2010b;Benzinou et al 2013), boundary-based shape classification (Nasreddine et al 2009), between-class correspondence analysis (Ponton 2006), or RF (e.g., Zhang et al 2016)).…”

“…The literature review emphasized that traditional DA was used in most of the studies for the classification of fish groups based on the elliptical Fourier descriptors of otolith shape, while application of alternative classifiers was less common. For example, Zhang et al (2016) used RF to discriminate stocks of the Japanese Spanish mackerel (Scomberomorus niphonius) based on Fourier descriptors of otolith shapes, but no comparison with other classifiers was reported. Mapp et al (2017) used nine ML algorithms for fish stock separation of two clupeid species using otolith shapes.…”

“…Few studies used ML algorithms and Fourier analysis of otolith shape to discriminate fish stocks (e.g., Zhang et al 2016;Mapp et al 2017). However, these studies did not compare the ML performance with traditional classifiers like linear discriminant analysis.…”