2 Frontiers Science Center for Deep Ocean Multispheres and Earth System(FDOMES), Ocean University of China, Qingdao 266100, China;
3 Laboratory for Marine Fisheries Science and Food Production Processes, Qingdao National Laboratory for Marine Science and Technology, Qingdao 266071, China
Species identification is a prerequisite for analyzing fishery resources; the process aims to identify individuals which are completely or partially discrete from congeners in space or time (Cadrin et al., 2014); it can effectively control overfishing and improve sustainable fisheries development—both of which are important for scientific assessment and management of fishery resources (Crandall et al., 2000; Cadrin and Silva, 2005). Species identification is also important to understand biological relationships and plays an important role in the protection of biodiversity and maintenance of global ecosystem stability. Species identification methods mainly include morphological and molecular markers (Ren et al., 2015). Cephalopods are characterized by rapid growth, short lifespans, high turnover rates, and strong phenotypic plasticity (Anderson and Rodhouse, 2001; Pecl et al., 2004; Pecl and Jackson, 2008); they are recognized as major components of many marine ecosystems (de la Chesnais et al., 2019). Cephalopod identification traditionally relies on examination of soft and hard tissues (such as fin shape and size, sucker ring dentition, tentacle and club morphology, and statolith morphology) (Dong, 1993). However, some morphological characters commonly used in identification are lost or damaged during fishing, and are affected by specimen freshness and life history characteristics, thereby complicating species identification (Chapela et al., 2003; Goud and De Heij, 2012). Cephalopod hard tissues (such as statoliths, gladius, and beaks) are more suitable and effective for species identification than soft tissues (such as fins and tentacles) due to their relative stability (Borges, 1995; Hu et al., 2017). Characters that enable accurate identification are of value for managing cephalopod populations.
China's coastal seas and ecosystems are overexploited, and its waters are warming rapidly (Belkin, 2009). Local catches of coastal cephalopods, especially squid, have increased since the 1990s (Pang et al., 2018). Squids of the family Loliginidae are the main commercial targets of cephalopod fisheries (Díaz-Santana-Iturrios et al., 2019); they are typically caught by bottom trawl, squid jig, trap net, or purse seine net (Jereb and Roper, 2010; Arkhipkin et al., 2015). The population dynamics of these species and their dominance vary (Pang et al., 2018). Accurate species identification will contribute to improved understanding of the population dynamics of species, and facilitate sustainable development and fishery management.
Loliginid squid identification is traditionally based on the characteristics of horny rings encircling arm suckers (Dong, 1988). However, these rings are easily lost, and other traditional discriminating features (such as shape and size of fins, mantle, color, etc.) often overlap among species, making identification difficult and time-consuming (Jereb and Roper, 2006). The potential of hard tissues, particularly the beaks (Yang et al., 2012; Jin et al., 2017, 2018) and statoliths (Lombarte et al., 2006) to assist in identification has been examined. Because the taxonomy and distribution of main loliginid taxa in China's seas are not well known, a survey of loliginid statoliths would determine if these structures are of systematic value.
Statoliths, paired calcareous structures composed of aragonite and calcite, are located in the cartilaginous cranium of cephalopods (Arkhipkin, 2005). Their morphology is of taxonomic value, and they have been used to identify squid more accurately than the gladius and beaks (Clarke, 1978). The shape of the statolith is also physiologically informative, which makes it possible to reconstruct aspects of a squid's behavior and movement (Arkhipkin, 2005). Their shape is also phylogenetically informative and is thought to be related to adaptive capacity in connection with mechanoreception (Lombarte et al., 2006). Advanced image processing techniques (Fourier analysis, wavelet transformation, and curvature scalespace analysis) enable quantitative comparison of shape contours (Lombarte et al., 1997; Parisi-Baradad et al., 2010; Reig-Bolaño et al., 2010). Fourier analysis is suitable for capturing information on statolith shape to compare between individuals (Smith et al., 2002; Green et al., 2015; Zhao et al., 2017). Analysis of statolith shape could is of great significance for cephalopods study and provide an alternative character for large-scale rapid identification, which will be routine work in studies of cephalopod biology.
Eight loliginid species occur in China's coastal waters, of which five (Uroteuthis (Photololigo) duvaucelii, U. edulis, U. chinensis, Loliolus beka, and L. japonica) with tropical and temperate distributions are commercially exploited (Chen et al., 2013). The high overlap in habitat distribution and morphological characteristics similarity of these species complicates their identification (Sin et al., 2009; Jin et al., 2017), which in turn hampers resource management and development. In this study, the five commercial Loliginidae squid species were identified using statolith shape analysis, the aim was to provide an effective identification method for Loliginidae squid species by comprehensive statolith comparison of these five species from the Chinese waters, covering their distribution range. In addition, the possible influence of genetic and environmental variables on statilith morphology was appraised preliminarily.2 MATERIAL AND METHOD 2.1 Sample collection
Squids collected from April to August 2019 off the coast of China (Fig. 1; Table 1) were randomly selected from the survey samples or local commercial landings, caught by bottom trawl, light luring or squid jig vessel. All of the samples were stored ice-frozen and transported to the laboratory for biological experiments. Species identification was based on anatomical characters, such as sucker ring dentition, following Dong (1988) and Okutani (2015). Dorsal mantle length (DML, mm) and weight (0.1 g) were measured for each individual, and the developmental stage of the gonad was visually appraised. Statoliths were extracted, cleaned with ultrasonic cleaner, and preserved in 75% ethanol.2.2 Statolith imaging and shape analysis
To exclude allometric effects, only maturing individuals (gonad stages Ⅲ and Ⅳ) were selected for statolith shape analysis. DML range was also restricted to an appropriate range for each species, to minimize statolith variability caused by size effects. Because the shape and position of statoliths in loliginids does not differ significantly between males and females, and morphological parameters of left and right statoliths are very similar in microstructural appearance, one statolith from each pair was randomly chosen for photography and image analysis (Natsukari et al., 1988; Moustahfid, 2002). Intact statoliths were placed with their posterior side facing up and their rostrum oriented horizontally (Fig. 2). A stereo microscope fitted with a video camera (Nikon SMZ18) took statolith images under reflected light on a black background.
Images were analyzed using an R package "shapeR" designed specifically for otolith/statolith shape analysis (Libungan and Pálsson, 2015). The package "shapeR" has built-in functions enabling users to extract statolith outlines from images, visualize and generate shape data, and transform outlines into independent Normalized Elliptic Fourier or Discrete Wavelet coefficients. In this study, 45 Normalized Elliptic Fourier coefficients were generated for shape analysis. Statolith length, width, and area were also measured by "shapeR, " and then Sw/Sl (statolith width/ length) and standard statolith area (Ssa=1 000×(statolith area/DML2)) were calculated.2.3 Statistical analysis
A one-way ANOVA detected differences among species in Sw/Sl and Ssa; significant results were subjected to a Tukey's Honest Significant Difference (HSD) test. Random forest (RF) was used to classify species because it: 1) outperforms other classifiers; 2) does not require any distributional assumption and is independent of predictor variables; and 3) does not overfit (Pérez-del-Olmo et al., 2010; Zhang et al., 2016). The RF algorithm selects bootstrap samples from the data, each of which contain 63% of original observations, to generate a classification (Strobl et al., 2009). The remainder are referred to as OOB (out-ofbag) observations; these represent an independent bootstrap data sample to fit each tree and unselected data set in the bootstrap sample as a test set for providing a cross-validation estimate of generalization errors to determine their unbiased classification rate (Breiman, 2001). RF also provides a global measure of the contribution of each variable by calculating the decrease in a Gini index (GI), which ranges from 0 (all variables contribute equally to the RF classifier) to 100 (a single variable contributes 100% to the RF classifier). Two thousand trees were built to ensure that each individual was predicted several times. The mean decrease in GI was calculated to measure the contribution of these variables to the optimal classifier. A visual shape similarity among species was plotted by graphical analysis of models using multidimensional scaling (MDS). The RF package R.5.03 was used for model development.3 RESULT
The Sw/Sl ratio varied significantly across species (Table 2), with the highest values observed for U. edulis and U. chinensis, and the lowest for L. beka. Ssa also differed significantly among species (Table 2), being highest in L. japonica, and lowest in U. chinensis and U. duvaucelii. An intermediate value was observed in U. edulis.
The RF model constructed by Sw/Sl, Ssa, and Fourier coefficients correctly classified 84.8% of individuals (Table 3) to species, with Ssa most accurately classifying species (Fig. 3). The highest classification success was recorded for L. beka, and the lowest for U. chinensis and L. japonica. The MDS plot depicts relationships between taxa, with three apparent 'branches' (Fig. 4). Of the five species, L. beka and L. japonica each represent a single branch, and the three Uroteuthis species a third; U. duvaucelii was the most variable of the three Uroteuthis species. Reconstructed average outlines reveal shape differences among these five species (Fig. 5), for which the main differences occur in the lateral and dorsal dome areas: U. chinensis and U. edulis have a raised lateral dome, L. beka and L. japonica have a pointed dorsal dome, and an intermediate morphotype occurs in U. duvaucelii.4 DISCUSSION
Statolith morphology is regarded as more reliable for recent and fossil species identification than the beaks or gladius (Clarke, 1978). By introducing shape variables (Sw/Sl and Ssa) and applying a more efficient classification method (random forest), we achieve a higher classification success rate than Jin et al. (2017) and higher classification rate for L. beka and U. duvaucelii. This suggests that statoliths have considerable value for species and population identification. Many reasons affect classification success rate, such as sample size, measurement errors, and data processing methods. Because of differences in studied species, sea areas, and measurement methods, comparing results between studies is difficult. More basic statolith data should be accumulated, statolith measurements and descriptions should be standardized, and new technologies (and software) capable of analyzing statolith shape should be exploited to better understand the merits of statolith morphology for species identification. Statolith shape differs within a species and can be used for stock identification (Fang et al., 2014; Green et al., 2015). Because our samples were collected throughout the distribution of these species in coastal China, our data also include population-level variation. Loliginid squid can be rapidly and reliably identified using statolith shape, and this technique can be used for more effective fishery management.
The shape of statoliths, like those of fish otoliths, is related to phylogeny (Clarke, 1978; Dommergues et al., 2000); these structures may be adapted for mechanoreception purposes (Arkhipkin and Bizikov, 2000; Arkhipkin, 2003). Statoliths of near-bottom species are larger than those of pelagic species of the same body size (Arkhipkin and Bizikov, 2000; Arkhipkin, 2003)—a trend reported also for fishes (Lombarte and Cruz, 2007). For vertically migrating squid, relative statolith size decreases with increased habitat depth, consistent with the expectation that inactive species need larger statoliths than active species to detect movement accelerations (Hanlon and Messenger, 1996). The two deeper-dwelling U. chinensis and U. edulis migrate further through the water column than the three other taxa. Of the three inshore species, U. duvaucelii lives in warmer waters and has a smaller Ssa than L. beka and L. japonica, while L. japonica, a cold water species, has the highest Ssa. Further study on relationships between statolith size and mantle length is needed to understand species ecology, considering the extensive distribution of cephalopod species globally.
Discrimination of these five Loliginidae species reveals a likely genetic basis to statolith shape, as the taxa clustered into two groups, namely the genera Uroteuthis and Loliolus, consistent with phylogenetic relationships revealed by COI gene sequences (Ho, 2005; Du, 2016). Genetic studies also clearly demonstrate U. chinensis and U. edulis are different species (Sin et al., 2009). Of the three Uroteuthis taxa, U. duvauceliis differs more from the other two Uroteuthis species than they differ from each other, which is also consistent with phylogenetic relationships where U. duvaucelii diverges from a clade comprising Loliolus and other Uroteuthis species (Jiang et al., 2018).
Statolith shape in five Loliginidae species demonstrated consistent determinants compared with fish otoliths, namely environment induces an overall change in shape, and genetically induced changes locally (Vignon and Morat, 2010). The good performance of shape variables (Sw/Sl and Ssa) associated with ecological characteristics suggests these variables would be valuable for species and population identification. Because of the rapid response of cephalopods to changing environmental conditions, studies on the relationship between statolith shape and species life histories would improve understanding of population structure, especially for species with wide distributions or multiple ecotypes.5 CONCLUSION
Five commercially Loliginidae squid species (Uroteuthis (Photololigo) duvaucelii, U. edulis, U. chinensis, Loliolus beka, and L. japonica) inhabiting the coastal waters of Chinese waters can be differentiated by comprehensively comparing their statoliths using a statolith shape analysis. A higher classification success rate of 84.8% was obtained. Statolith relative size decreases with increased habitat depth for vertically migrating squids, and environment induces an overall change in shape and genetically induced changes locally. We report statolith shape to be of value in species identification when assessing impact of single species management systems or development of multi-species fisheries. Identification of squids by statolith shape analysis is also fast and accurate, and may assist with future development and management of Loliginidae squid fisheries in Chinese waters.6 DATA AVAILABILITY STATEMENT
The data generated or analyzed during the current study are available from the corresponding author on reasonable request.7 ACKNOWLEDGMENT
We thank Yanyu CHEN (Beijing Normal University-Hong Kong Baptist University United International College) and Haozhan WANG (Sansha Marine Environment Monitoring Center Station, State Oceanic Administration) for their great assistance in sample collection.
Anderson C I H, Rodhouse P G. 2001. Life cycles, oceanography and variability: ommastrephid squid in variable oceanographic environments. Fisheries Research, 54(1): 133-143. DOI:10.1016/S0165-7836(01)00378-2
Arkhipkin A I, Bizikov V A. 2000. Role of the statolith in functioning of the acceleration receptor system in squids and sepioids. Journal of Zoology, 250(1): 31-55. DOI:10.1111/j.1469-7998.2000.tb00575.x
Arkhipkin A I, Rodhouse P G K, Pierce G J, Sauer W, Sakai M, Allcock L, Arguelles J, Bower J R, Castillo G, Ceriola L, Chen C S, Chen X J, Diaz-Santana M, Downey N, González A F, Amores J G, Green C P, Guerra A, Hendrickson L C, Ibáñez C, Ito K, Jereb P, Kato Y, Katugin O N, Kawano M, Kidokoro H, Kulik V V, Laptikhovsky V V, Lipinski M R, Liu B L, Mariátegui L, Marin W, Medina A, Miki K, Miyahara K, Moltschaniwskyj N, Moustahfid H, Nabhitabhata J, Nanjo N, Nigmatullin C M, Ohtani T, Pecl G, Perez J A A, Piatkowski U, Saikliang P, Salinas-Zavala C A, Steer M, Tian Y J, Ueta Y, Vijai D, Wakabayashi T, Yamaguchi T, Yamashiro C, Yamashita N, Zeidberg L D. 2015. World squid fisheries. Reviews in Fisheries Science & Aquaculture, 23(2): 92-252. DOI:10.1080/23308249.2015.1026226
Arkhipkin A I. 2003. Towards identification of the ecological lifestyle in nektonic squid using statolith morphometry. Journal of Molluscan Studies, 69(3): 171-178. DOI:10.1093/mollus/69.3.171
Arkhipkin A I. 2005. Statoliths as 'black boxes' (life recorders) in squid. Marine and Freshwater Research, 56(5): 573-583. DOI:10.1071/MF04158
Belkin I M. 2009. Rapid warming of large marine ecosystems. Progress in Oceanography, 81(1-4): 207-213. DOI:10.1016/j.pocean.2009.04.011
Borges T C. 1995. Discriminate analysis of geographic variation in hard structures of Todarodes saguttatus (Lamarek 1798) from North Atlantic Ocean. ICESMarine Science Symposium, 199: 433-440.
Breiman L. 2001. Random forests. Machine Learning, 45: 5-32. DOI:10.1023/A:1010933404324
Cadrin S X, Kerr L A, Mariani S. 2014. Stock Identification Methods: Applications in Fishery Science. 2nd edn. Academic Press, London. p. 1-10.
Cadrin S X, Silva V M. 2005. Morphometric variation of yellowtail flounder. ICES Journal of Marine Science, 62(4): 683-694. DOI:10.1016/j.icesjms.2005.02.006
Chapela M J, Sotelo C G, Pérez-Martín R I. 2003. Molecular identification of cephalopod species by FINS and PCR-RFLP of a cytochrome b gene fragment. European Food Research and Technology, 217(6): 524-529. DOI:10.1007/s00217-003-0788-y
Chen X J, Wang G Y, Qian W G. 2013. Important Economic Resources and Fisheries of Cephalopod in Coast of China. Science Press, Beijing, China. p. 67, 101-120. (in Chinese)
Clarke M R. 1978. The cephalopod statolithan-introduction to its form. Journal of the Marine Biological Association of the United Kingdom, 58(3): 701-712. DOI:10.1017/S0025315400041345
Crandall K A, Bininda-Emonds O R P, Mace G M, Wayne R K. 2000. Considering evolutionary processes in conservation biology. Trends in Ecology & Evolution, 15(7): 290-295. DOI:10.1016/S0169-5347(00)01876-0
de la Chesnais T, Fulton E A, Tracey S R, Pecl G T. 2019. The ecological role of cephalopods and their representation in ecosystem models. Reviews in Fish Biology and Fisheries, 29(2): 313-334. DOI:10.1007/s11160-019-09554-2
Diaz-Santana-Iturrios M, Salinas-Zavala C A, Granados-Amores J, de la Cruz-Agüero J, Garcia-Rodriguez F J. 2019. Taxonomic considerations of squids of the family Loliginidae (Cephalopoda: Myopsida) supported by morphological, morphometric, and molecular data. Marine Biodiversity, 49(5): 2 401-2 409. DOI:10.1007/s12526-019-00979-3
Dommergues J L, Neige P, Boletzky S V. 2000. Exploration of morphospace using procrustes analysis in statoliths of cuttlefish and squid (Cephalopoda: Decabrachia)-evolutionary aspects of form disparity. The Veliger, 43(3): 265-276.
Dong Z Z. 1988. Fauna Sinica, Phylum Mollusca, Class Cephalopoda. Science Press, Beijing, China. p.111-114. (in Chinese)
Dong Z Z. 1993. Morphological comparison of the several structures of cephalopods. Acta Zoologica Sinica, 39(4): 348-354. (in Chinese with English abstract)
Du T F. 2016. Resources Assessment for Cephalopod in Offshore Water of China and Classification of Genus Level of Squids. Shanghai Ocean University, Shanghai, China. p.47. (in Chinese with English abstract)
Fang Z, Liu B L, Li J H, Su H, Chen X J. 2014. Stock identification of neon flying squid (Ommastrephes bartramii) in the North Pacific Ocean on the basis of beak and statolith morphology. Scientia Marina, 78(2): 239-248. DOI:10.3989/scimar.03991.06A
Goud J, De Heij A. 2012. Mediterranean Sepiola aurantiaca Jatta, 1896, versus the NE Atlantic Sepiola pfefferi Grimpe, 1921 (Cephalopoda, Sepiolinae). Basteria, 76(1-3): 1-11.
Green C P, Robertson S G, Hamer P A, Virtue P, Jackson G D, Moltschaniwskyj N A. 2015. Combining statolith element composition and Fourier shape data allows discrimination of spatial and temporal stock structure of arrow squid (Nototodarus gouldi). Canadian Journal of Fisheries and Aquatic Sciences, 72(11): 1 609-1 618. DOI:10.1139/cjfas-2014-0559
Hanlon R T, Messenger J B. 1996. Cephalopod Behaviour. Cambridge University Press, Cambridge, England. 232p.
Ho S C. 2005. Taxonomic Study of the Taiwan Inshore Squids. Taiwan Ocean University, Keelung, China. 93p. (in Chinese with English abstract)
Hu F F, Chen X J, Liu B L, Li J H. 2017. Review on identification of cephalopods classification. Marine Fisheries, 39(1): 110-120. (in Chinese with English abstract)
Jereb P, Roper C F E. 2006. Cephalopods of the Indian Ocean. A review. Part I. Inshore squids (Loliginidae) collected during the international Indian Ocean Expedition. Proceedings of the Biological Society of Washington, 119(1): 91-136. DOI:10.2988/0006-324x(2006)119[91:cotioa]2.0.co;2
Jereb P, Roper C F E. 2010. Cephalopods of the world. An annotated and illustrated catalogue of cephalopod species known to date. Volume 2. Myopsid and Oegopsid squids. In: FAO Species Catalogue for Fishery Purposes. Rome, FAO. p. 38-117.
Jiang L H, Kang L S, Wu C W, Chen M, Lü Z M. 2018. A comprehensive description and evolutionary analysis of 9 Loliginidae mitochondrial genomes. Hydrobiologia, 808(1): 115-124. DOI:10.1007/s10750-017-3377-y
Jin Y, Liu B L, Chen X J, Staples K. 2018. Morphological beak differences of loliginid squid, Uroteuthis chinensis and Uroteuthis edulis, in the northern South China Sea. Journal of Oceanology and Limnology, 36(2): 559-571. DOI:10.1007/s00343-017-6285-0
Jin Y, Liu B L, Li J H, Chen X J. 2017. Identification of three common Loliginidae squid species in the South China Sea by analyzing hard tissues with geometric outline method. Journal of Ocean University of China, 16(5): 840-846. DOI:10.1007/s11802-017-3218-7
Libungan L A, Pâlsson S. 2015. ShapeR: an R package to study otolith shape variation among fish populations. PLoS One, 10(3): e0121102. DOI:10.1371/journal.pone.0121102
Lombarte A, Cruz A. 2007. Otolith size trends in marine fish communities from different depth strata. Journal of Fish Biology, 71(1): 53-76. DOI:10.1111/j.1095-8649.2007.01465.x
Lombarte A, Rufino M M, Sanchez P. 2006. Statolith identification of Mediterranean Octopodidae, Sepiidae, Loliginidae, Ommastrephidae and Enoploteuthidae based on warp analyses. Journal of the Marine Biological Association of the United Kingdom, 86(4): 767-771. DOI:10.1017/S0025315406013683
Lombarte A, Sanchez P, Morales-Nin B. 1997. Intraspecific shape variability in statoliths of three cephalopod species. Vie et Milieu-Life and Environment, 47(2): 165-169.
Moustahfid H. 2002. Age and growth of arrow squid Todarodes sagittatus (Cephalopoda: Ommastrephidae) sampled in summer in Atlantic Moroccan waters. Bulletin of Marine Science, 71(1): 535-543.
Natsukari Y, Nakanose T, Oda K. 1988. Age and growth of loliginid squid Photololigo edulis (Hoyle, 1885). Journal of Experimental Marine Biology and Ecology, 116(2): 177-190. DOI:10.1016/0022-0981(88)90054-8
Okutani T. 2015. Cuttlefishes and Squids of the World (New Edition). Tokai University Press, Tokyo, Japan. p. 73-227.
Pang Y M, Tian Y J, Fu C H, Wang B, Li J C, Ren Y P, Wan R. 2018. Variability of coastal cephalopods in overexploited China Seas under climate change with implications on fisheries management. Fisheries Research, 208: 22-33. DOI:10.1016/j.fishres.2018.07.004
Parisi-Baradad V, Manjabacas A, Lombarte A, Olivella R, Chic O, Piera J, Garcia-Ladona E. 2010. Automated Taxon Identification of Teleost fishes using an otolith online database-AFORO. Fisheries Research, 105(1): 13-20. DOI:10.1016/j.fishres.2010.02.005
Pecl G T, Jackson G D. 2008. The potential impacts of climate change on inshore squid: biology, ecology and fisheries. Reviews in Fish Biology and Fisheries, 18(4): 373-385. DOI:10.1007/s11160-007-9077-3
Pecl G T, Moltschaniwskyj N A, Tracey S R, Jordan A R. 2004. Inter-annual plasticity of squid life history and population structure: ecological and management implications. Oecologia, 139(4): 515-524. DOI:10.1007/s00442-004-1537-z
Pérez-del-Olmo A, Montero F E, Fernández M, Barrett J, Raga J A, Kostadinova A. 2010. Discrimination of fish populations using parasites: random forests on a 'predictable' host-parasite system. Parasitology, 137(12): 1833-1847. DOI:10.1017/S0031182010000739
Reig-Bolaño R, Marti-Puig P, Lombarte A, Soria J A, Parisi-Baradad V. 2010. A new otolith image contour descriptor based on partial reflection. Environmental Biology of Fishes, 89(3-4): 579-590. DOI:10.1007/s10641-010-9700-3
Ren Y, Hou R, Feng H, Wang L, Feng C L. 2015. DNA analysis methods in species identification. Shaanxi Journal of Agricultural Sciences, 61(10): 61-64. (in Chinese)
Sin Y W, Yau C, Chu K H. 2009. Morphological and genetic differentiation of two loliginid squids, Uroteuthis (Photololigo) chinensis and Uroteuthis (Photololigo) edulis (Cephalopoda: Loliginidae), in Asia. Journal of Experimental Marine Biology and Ecology, 369(1): 22-30. DOI:10.1016/j.jembe.2008.10.029
Smith P J, Robertson S G, Horn P L, Bull B, Anderson O F, Stanton B R, Oke C S. 2002. Multiple techniques for determining stock relationships between orange roughy, Hoplostethus atlanticus, fisheries in the eastern Tasman Sea. Fisheries Research, 58(2): 119-140. DOI:10.1016/s0165-7836(01)00389-7
Strobl C, Malley J, Tutz G. 2009. An introduction to recursive partitioning: rationale, application, and characteristics of classification and regression trees, bagging, and random forests. Psychological Methods, 14(4): 323-348. DOI:10.1037/a0016973
Vignon M, Morat F. 2010. Environmental and genetic determinant of otolith shape revealed by a non-indigenous tropical fish. Marine Ecology Progress Series, 411: 231-241. DOI:10.3354/meps08651
Yang L L, Jiang Y Z, Liu Z L, Lin N, Li S F, Cheng J H. 2012. Analysis of beak morphology of Loligo beka in the East China Sea. Journal of Fishery Sciences of China, 19(4): 586-593. (in Chinese with English abstract) DOI:10.3724/SPJ.1118.2012.00586
Zhang C, Ye Z J, Li Z G, Wan R, Ren Y P, Dou S Z. 2016. 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. Fisheries Science, 82(2): 251-256. DOI:10.1007/s12562-016-0968-x
Zhao B, Liu J H, Song J J, Cao L, Dou S Z. 2017. Evaluation of removal of the size effect using data scaling and elliptic Fourier descriptors in otolith shape analysis, exemplified by the discrimination of two yellow croaker stocks along the Chinese coast. Chinese Journal of Oceanology and Limnology, 35(6): 1482-1492. DOI:10.1007/s00343-017-6012-x