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

Smoliński, Szymon; Schade, Franziska; Berg, Florian

doi:10.1139/cjfas-2019-0251

Cited by 24 publications

(15 citation statements)

References 64 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Overall, the neural network was slightly more accurate than the LDA model, with the accuracy at classifying the wild individuals being higher in the former, whereas the accuracy of stocked individuals being higher in the later model. The higher accuracy of machine learning methods over more traditional statistical tests observed in this study corroborates the finds of Smoliński et al, (2019), but, in our case, the difference seems to be small and not true for all groups (i.e. stocked fish).…”

Section: Discussionsupporting

confidence: 87%

“…The number of hidden layers were selected based on the number of input variables in the model (2/3 of the size of input layers plus the size of output layers). Machine learning tools outperform traditional statistical classifiers in otolith shape analysis and can improve the accuracy of fish stock discrimination studies (Smoliński et al, 2019). The relative importance of each variable was assessed by the Olden's algorithm (Olden and Jackson, 2002) using the olden function of the NeuralNetTools package (Beck, 2018).…”

Section: Statistical Analysesmentioning

confidence: 99%

“…The number of harmonics used was defined by applying the calibrate_harmonicpower_efourier function. The Elliptical Fourier Analysis is commonly used to describe the fish otolith shapes with a high degree of precision and success in capturing the shape of the otoliths, even those with rough contours (Assis et al, 2020;Gagliano and McCormick, 2004;Karahan et al, 2014;Lestrel, 1997;Mahé et al, 2019;Mérigot et al, 2007;Rashidabadi et al, 2020;Rodgveller et al, 2017;Smoliński et al, 2019).…”

Section: Statistical Analysesmentioning

confidence: 99%

See 2 more Smart Citations

Otolith shape variations between artificially stocked and autochthonous pikeperch (Sander lucioperca)

et al. 2020

View full text Add to dashboard Cite

Fish stocking is one of the most widespread and frequent management strategies in freshwater systems. However, the contribution of stocked fish to the population is seldom investigated, and hence the effectiveness of this strategy is virtually unknown for many populations. Understanding the contribution of stocked fish into the population is crucial because it allows the disentanglement of the confounding effects generated by allochthonous individuals into the estimation of survival and growth rates of the population. To discriminate between the allochthonous and autochthonous individuals in a population, the shape of sagittae otoliths from pikeperch (Sander lucioperca) where compared. Results indicated significant differences among stocked and non-stocked fish, with the former having smaller and wider sagittae otoliths than the later. Our results suggest that this technique can be used to discriminate the natal origin of fish in a much faster and cheaper way than commonly used techniques.

show abstract

Section: Discussionsupporting

confidence: 87%

Section: Statistical Analysesmentioning

confidence: 99%

Section: Statistical Analysesmentioning

confidence: 99%

See 1 more Smart Citation

Otolith shape variations between artificially stocked and autochthonous pikeperch (Sander lucioperca)

et al. 2020

View full text Add to dashboard Cite

show abstract

“…Compared to Fourier descriptors, shape indices are expected to be less sensitive to subtle variation in otolith shape [ 82 , 83 ]. Moreover, Fourier descriptors and shape indices are supposed to be partially redundant information since Fourier descriptors are precise reconstructions of otolith outline whereas shape indices are considered as less integrative, yet accurate, metrics [ 84 ]. To our knowledge, not any study has previously presented spatially heterogeneous shape indices in a single analysis without significant spatial signal in Fourier descriptors.…”

Section: Discussionmentioning

confidence: 99%

Complementarity and discriminatory power of genotype and otolith shape in describing the fine-scale population structure of an exploited fish, the common sole of the Eastern English Channel

et al. 2020

View full text Add to dashboard Cite

Marine organisms show population structure at a relatively fine spatial scale, even in open habitats. The tools commonly used to assess subtle patterns of connectivity have diverse levels of resolution and can complement each other to inform on population structure. We assessed and compared the discriminatory power of genetic markers and otolith shape to reveal the population structure on evolutionary and ecological time scales of the common sole (Solea solea), living in the Eastern English Channel (EEC) stock off France and the UK. First, we genotyped fish with Single Nucleotide Polymorphisms to assess population structure at an evolutionary scale. Then, we tested for spatial segregation of the subunits using otolith shape as an integrative tracer of life history. Finally, a supervised machine learning framework was applied to genotypes and otolith phenotypes to probabilistically assign adults to subunits and assess the discriminatory power of each approach. Low but significant genetic differentiation was found among subunits. Moreover, otolith shape appeared to vary spatially, suggesting spatial population structure at fine spatial scale. However, results of the supervised discriminant analyses failed to discriminate among subunits, especially for otolith shape. We suggest that the degree of population segregation may not be strong enough to allow for robust fish assignments. Finally, this study revealed a weak yet existing metapopulation structure of common sole at the fine spatial scale of the EEC based on genotypes and otolith shape, with one subunit being more isolated. Our study argues for the use of complementary tracers to investigate marine population structure.

show abstract

“…Classification using random forests.-We used random forests to classify fish to genetic lineage, using observed phenotypic traits, based on their proven predictive performance relative to similar classification procedures (Breiman 2001; Maguffee et al 2019;Smolinski et al 2020). We restricted fish inclusion to only those assigned to a tule or upriver bright reporting group; all other groups were either identified readily using unique fin clip markings or were rare among sampled fish.…”

Section: Methodsmentioning

confidence: 99%

Rapid Phenotypic Stock Identification of Chinook Salmon in Recreational Fishery Management

2021

View full text Add to dashboard Cite

Rapid phenotypic stock identification in mixed-stock fisheries can provide a useful alternative to more time-intensive methods (e.g., coded wire tags, genetics) in assessing harvest and informing management decisions. We leveraged local ecological knowledge, existing stock identification methods, and understanding of life history differences to develop rapid stock identification tools for fall-run Chinook Salmon Oncorhynchus tshawytscha encountered in the Buoy 10 recreational fishery at the mouth of the Columbia River. Specifically, we sought to differentiate between the fishery's two dominant genetic lineages: lower river tules and upriver brights. We sampled recreationally landed Chinook Salmon in 2017, 2018, and 2019, assigned sampled individuals to functional reporting groups using a single-nucleotide-polymorphism-based genetic baseline, and collected measurements on phenotypic traits. Using traits including pigmentation patterns (e.g., spotting), fin morphology, characteristics indicative of sexual maturity, and muscle lipid content, random forest classification models provided consistently high classification success across and within genetic groups (i.e., up to 90%). Classification success remained consistent over time within fishery seasons and between years but showed meaningful bias between sexes. Based on observed classification success, we developed and evaluated a categorical visual identification guide capable of facilitating more rapid trait observations and on-site stock identification. The resulting classification key, built using classification trees and visual guide observations from 2019, achieved slightly lower classification success across and within genetic groups and had variable success among samplers. Compared with the existing use of coded wire tags in harvest assessment, phenotypic stock identification methods can provide more rapid and more numerous assignments, albeit with a greater degree of individual assignment error. Applied as a complement to standard methods like coded wire tags, the use of rapid phenotypic stock identification methods offers the potential for increased overall precision and timeliness in harvest assessments.Reliable means of stock identification are critical in successfully monitoring and managing mixed-stock fisheries (Crozier et al. 2004;Cadrin et al. 2014). Stock identification is especially relevant in fisheries for Chinook Salmon Oncorhynchus tshawytscha in the Pacific Northwest, in which numerous stocks co-occur in recreational and commercial fisheries across marine and freshwater environments (McKinney et al. 2017). Recent advances in stock identification measures focus on precise but time-and labor-intensive methods, including genetics

show abstract

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

Cited by 24 publications

References 64 publications

Otolith shape variations between artificially stocked and autochthonous pikeperch (Sander lucioperca)

Otolith shape variations between artificially stocked and autochthonous pikeperch (Sander lucioperca)

Complementarity and discriminatory power of genotype and otolith shape in describing the fine-scale population structure of an exploited fish, the common sole of the Eastern English Channel

Rapid Phenotypic Stock Identification of Chinook Salmon in Recreational Fishery Management

Contact Info

Product

Resources

About