Next Article in Journal
A Preliminary Study on Species Identification of Immature Necrophagous Phorid Flies Based on FTIR Spectroscopy
Previous Article in Journal
Fourier Transform Infrared Spectroscopy to Measure Cholesterol in Goat Spermatozoa
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Animals
Volume 15
Issue 21
10.3390/ani15213108
animals-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Unraveling Fish Community Assembly Rules in Coastal China Seas Based on Hierarchical Modeling of Species Communities

by
Li Lin
1,2,
Yang Liu
3 and
Bin Kang
2,*
1
Department of Animal Husbandry and Fisheries, Guizhou Vocational College of Agriculture, Guiyang 550000, China
2
Key Laboratory of Mariculture (Ocean University of China), Ministry of Education, Qingdao 266003, China
3
College of Ecological Engineering, Guizhou University of Engineering Science, Bijie 551700, China
*
Author to whom correspondence should be addressed.
Animals 2025, 15(21), 3108; https://doi.org/10.3390/ani15213108
Submission received: 22 September 2025 / Revised: 22 October 2025 / Accepted: 23 October 2025 / Published: 26 October 2025
(This article belongs to the Special Issue Ecology and Conservation of Marine Fish)
Browse Figures
Versions Notes

Simple Summary

Coastal China seas’ fish communities face threats like overfishing and climate change, but how these communities react to these threats is unclear. This study aimed to understand what shapes these fish communities using a method of community modelling. We analyzed data on 384 fish species (1980–2018) and environmental factors. The results showed that temperature and salinity mostly determine fish distribution, and fish prefer silt over fine sand habitats. Goby fish have more connections with other fish. The findings help predict coastal fish communities and guide efforts to protect their biodiversity, benefiting ocean health and related human activities.

Abstract

To address uncertainties in how threatened coastal China seas fish communities respond to stressors like overfishing and climate change, this study applied Hierarchical Modelling of Species Communities (HMSC) to disentangle the assembly rules shaping these communities, filling a critical gap in understanding their spatiotemporal dynamics. We analyzed data on 384 fish species (1980–2018) and key environmental factors, with variance partitioning revealing that environmental filtering dominated fish distributions (explaining over 99% of variance), far outweighing random effects (0.60%). Among environmental drivers, sea surface temperature (49.00%) and sea surface salinity (33.25%) were the most influential, while seafloor substrate and water depth played secondary roles; notably, fewer species occupied fine sand habitats, and more preferred silt habitats. Residual species associations—indicative of potential biotic interactions—were most frequent within Gobiidae, likely due to this highly diverse taxon’s specialized resource utilization and wide distribution, highlighting that biotic filtering is concentrated and ecologically relevant within this group. This work demonstrates HMSC’s utility in unraveling coastal fish community assembly, providing a robust basis for predicting community changes and guiding biodiversity conservation efforts that support ocean health and dependent human activities.

Graphical Abstract

1. Introduction

Coastal habitats are essential for ecological processes and provide innumerable benefits and services to society and the entire planet as a whole [1]. These habitats, including soft shores, rocky shores and cliffs, hilly or flat coastal plains, narrow or wide coastal shelves, and a variety of wetlands, constitute various geomorphological features, weather regimes, and biomes, which support rich and indispensable biodiversity on planet earth [2]. The China seas include the Bohai Sea, the Yellow Sea, the East China Sea, and the South China Sea, which extend over 38 degrees of latitude and span from tropical to temperate climate zones, covering almost 5 million km2 and approximately 32,000 km of coastlines [3]. However, over the past half-century, the coastal ecosystems of the China seas have undergone severe degradation in fish community structure, driven by multiple anthropogenic and environmental stressors, including overfishing, marine pollution, biological invasions, and climate change [4,5]. China also has marine regions warming in the top 10% globally [4]. These pressures alter the structure and biodiversity of natural fish communities, thereby potentially affecting the functions and services that coastal ecosystems provide [6]. Therefore, there is an urgent need to understand how fish communities respond to changes in environmental conditions, as this knowledge is foundational to the formulation and prioritization of targeted, science-based conservation strategies.
Effective conservation management of the coastal China seas, where fish community structure has suffered severe deterioration over the past half century, requires knowledge of the rationales underlying how these specific coastal fish communities are assembled [7]. Studies exploring the correlations between communities and environmental conditions can be traced back to niche theory, which defines a species’ niche as a hypervolume, a multidimensional space bounded by the specific environmental characteristics (e.g., temperature, salinity, substrate type) that the species needs to survive, reproduce, and persist in a given habitat [8]. On the other hand, the neutral theory proposes that all individuals are ecologically identical and niche differences are not needed to explain community patterns, i.e., communities arise solely on random events, i.e., chance extinctions balanced by chance speciation [9]. In recent years, the importance of interspecific interactions to explain community structure has also been reinforced [10]. Therefore, not only environmental but also biotic factors and the stochastic nature of the community have been unified into the scope of community assembly rules framework, which incorporates the effects of species dispersal, environmental filtering, and biotic filtering [11]. First, dispersal defines the geographical distance at which a species can colonize [12]. Second, environmental filtering selects species according to their preference and tolerance to local abiotic conditions [13]. Third, species interact with each other through negative or positive biotic interactions [14].
Temperature has frequently been reported to be one of the most important factors driving marine fish community structure [15]. Other environmental factors, including salinity, depth, and water current, have been frequently proven influential at global or local scales [16,17,18]. Biotic filtering has also been proven to be influential for marine fish communities [19]. For instance, in the Pacific coral reefs near Papua New Guinea, in the Pacific coral reefs near Papua New Guinea, sympatric damselfishes Dascyllus aruanus and D. melanurus exhibit local-scale competitive interactions with D. melanurus, restricting the foraging range of D. melanurus to smaller prey in the presence of D. aruanus [20]. Trait-based approaches contribute to the understanding of community assembly [21]. For example, limiting similarity of fish functional traits is correlated with environmental filtering in coastal lagoon fish communities [22]. Fish functional traits and phylogenetic data can be used to better capture species’ responses to abiotic and biotic environments, as well as to improve the performance of community modeling by incorporating trait-phylogeny links to community assembly processes [23]. A plethora of fish resources investigations have been conducted in the coastal China seas tracing back to the 1980s [24,25,26]. These studies mostly focused on patterns of stock evaluation, community structure, or biodiversity in space and/or time, but insufficient efforts were dedicated to disentangling the community assembly rules shaping the spatiotemporal patterns of fish community structures.
Hierarchical Modelling of Species Communities (HMSC) is a multivariate hierarchical generalized linear mixed model designed for community ecology [23]. It has been built intrinsically in such a way that the model components, including fixed effects, random effects, species traits, and phylogenetic relationships, are conceptually related to community assembly processes [27]. Therefore, it is the optimal tool for a complex dataset that integrates occurrence data, multiple environmental variables, functional traits, and phylogenetic information [28]. Critically, these components are conceptually tied to the core community assembly processes, including environmental filtering, biotic filtering, and dispersal-related stochasticity, enabling us to parse their relative contributions to fish community structure in a system as geographically extensive and ecologically heterogeneous as the coastal China seas. With the increasing data availability of environmental variables and historical fish investigations in coastal China seas, this work applied HMSC to model the species responses to the gradients of environmental conditions, aiming to explore fish community assembly rules.

2. Materials and Methods

2.1. Study Area

Fish investigation data were collected at 45 sampling sites distributed along the coastal waters of the China seas, encompassing four major marine regions: the Bohai Sea (latitude range: 37°07′–41°00′ N), the Yellow Sea (latitude range: 31°40′–37°07′ N), the East China Sea (latitude range: 21°54′–31°40′ N), and the South China Sea (latitude range: 16°00′–21°54′ N) (see Figure 1 for spatial distribution). Most observations were made in shallow waters below 40 m deep. Codes, names, and geographic coordinates of these sites are listed in Appendix A—Table A1.

2.2. Data Collection

2.2.1. Species Occurrence Data Compilation

Data of 148 fish investigations in Chinese coastal waters between 1980 and 2018 at these 45 sites were compiled (see the data reference list in the files stated in Data Availability Statement). These investigations occurred independently and were hereinafter referred to as samples. Sampling methods were categorized in accordance with the definitions and classification criteria for fishing gear types established by the Food and Agriculture Organization of the United Nations (FAO) [30]. Most sampling methods were bottom trawls, including bottom otter trawls, bottom pair trawls, beam trawls, combined trawls, and other unspecified bottom trawls (Appendix B—Figure A1a). The sampling year was categorized based on the specific time period during which each fish survey was conducted (Appendix B—Figure A1b). Original fish taxonomic data were double-checked and validated with the records in FishBase to avoid invalid species, synonyms, and homonyms [31]. There were 384 valid fish species identified altogether, with the highest species richness of 250 at the Leizhou Bay (Appendix B—Figure A2a) and the highest species prevalence of 0.78 of the species Konosirus punctatus (Appendix B—Figure A2b).

2.2.2. Environmental Data Compilation

Environmental variables included in this study (Appendix A—Table A2) were selected based on the following ecological rationales: (1) Water depth modulates benthic-pelagic coupling processes and shapes predator-prey interactions, which are critical to fish community structure [17]. (2) Seafloor substrate types directly influence the microhabitat suitability and resource availability for demersal fish species, thereby affecting their distribution and abundance [32]. (3) Water current directly affects temperature and food resources [18]. (4) water temperature, salinity, and pH directly affect fish physiology [16,33]. (5) Water net primary production exerts a fundamental regulatory influence on the structure of marine food webs, as it constitutes the primary energy input that supports the base of trophic hierarchies [34].
Data on mean water depth (depth) were retrieved from the data repository maintained by the National Oceanic and Atmospheric Administration (NOAA) [35]. Data on the types of seafloor surface substrate (substrate) were provided by the State Oceanic Administration of China, which adhered to the classification criteria outlined in the Shepard scheme [36,37]. Data on the northward component of the sea surface current (hereafter abbreviated as “current”), sea surface temperature (hereafter abbreviated as “SST”), and sea surface salinity (hereafter abbreviated as “SSS”) were obtained from the National Marine Data Center, which is part of the National Science & Technology Infrastructure Platform [38]. Data on sea surface water net primary production (NPP) and sea surface pH (pH) were acquired from the Copernicus Marine Environment Monitoring Service (CMEMS) [39]. While bottom temperature/salinity would be ideal, scarce, discontinuous historical data for the coastal China seas prevented their inclusion. Monthly datasets for SST, SSS, NPP, and pH were spatially and temporally averaged to match the specific month and year of each fish investigation. Present-day environmental variable layers were acquired from the Bio-ORACLE (Biological Ocean Atlas of Chemical and Physical Variables) database, a widely recognized repository for marine environmental data in ecological modeling [40]. Depth and substrate were treated as temporally invariant because the magnitude of temporal variation in these two variables was anticipated to be negligible at the same sampling site, and no additional historical datasets had been accessible to support temporal adjustments. Correlation coefficients among the selected environmental variables were all below 0.70 (Appendix B—Figure A3a), and the Variance Inflation Factor (VIF) values for each variable were less than a predefined threshold of 3 (Appendix B—Figure A3b), indicating the dataset exhibited an acceptable level of multicollinearity [41].

2.2.3. Functional Trait Data Compilation

Functional traits were selected based on two criteria: (1) they are associated with different fish functions, including feeding habit, trophic level, swimming capability, habitat preference, and life history; (2) they tend to capture the response of fish species to abiotic and biotic environments, i.e., they are likely to be response traits [42]. We followed Brosse et al. [43] to derive nine morphological traits from their morphological measurements (Appendix B—Figure A4). These morphological measurements were recorded by ImageJ Version 1.53k [44]. Images were primarily sourced from regional fish atlases. Only adult specimens were photographed, with juveniles excluded to account for morphological variations associated with ontogeny. For species exhibiting sexual dimorphism, only male morphological traits were documented, as female specimens were rarely imaged across most species. Nine morphological traits (Appendix A—Table A3) were derived from these morphological measurements, including body elongation, body lateral shape, caudal peduncle throttling, oral gape position, pectoral fin vertical position, pectoral fin size, relative eye size, relative maxillary length and vertical eye position. For those fish with unusual morphologies, we followed Brosse et al. [43] and Villéger et al. [45] to apply the following rules: (1) for species lacking a visible caudal fin (e.g., Sternopygidae, Anguilidae, Plotosidae), the caudal peduncle throttling value was set to 1, with the assumption that caudal fin depth equals caudal peduncle depth; (2) for algivorous species with a subterminal mouth (e.g., Loricaridae, some Balitoridae), both oral gape position and relative maxillary length were assigned a value of 0; (3) for species without pectoral fins (e.g., some Synbranchiformes, Anguilliformes), the pectoral fin vertical position was set to 0; (4) for flatfishes, body depth was measured as body width, given their dorsoventrally flattened body and lateral resting posture. The ecological traits collected from FishBase were reviewed to ensure consistency with the morphological standard to confirm trait alignment.
We further followed the methodology of Trindade-Santos et al. [46] to extract eight additional biologically and ecologically relevant traits (Appendix A—Table A3) from the latest version of the FishBase database [31], including maximum lifespan, generation time, food consumption-to-biomass ratio, maximum body length, water column position, dietary habit, trophic level, and body shape. Given the unavoidable data gaps inherent in large-scale databases such as FishBase, a random forest algorithm was employed to impute missing values in the trait matrix (missing rate < 10%) [47]. Pearson correlation analyses were also conducted among these traits to verify their complementarity (Appendix B—Figure A5). The majority of these traits exhibited no significant correlation, confirming that they provided complementary information.

2.2.4. Phylogenetic Data Compilation

Cytochrome oxidase I (COI) gene sequences were retrieved from the GenBank database [48]. Of the studied fish species, 355 (92.44%) had verified COI sequences, while no records were available for the remaining 29 species as of 1 July 2021. To improve the estimation of community-level phylogenetic information, surrogate sequences for these 29 species were sourced from their phylogenetically and morphologically close congeners (see species details in the Data Availability Statement). Sequences were aligned using Clustal W 2.0 [49]. The Akaike Information Criterion (AIC) was employed to select optimal parameters for constructing the Maximum Likelihood (ML) phylogeny [50]; the substitution model GTR + I + Γ best fit the empirical data based on AIC values (Appendix B—Figure A6), where “GTR” denotes the general time-reversible model, “+I” indicates optimization of the proportion of invariable sites, and “+Γ” represents optimization of the gamma rate parameter within the GTR framework. Finally, the ML phylogenetic tree under the GTR + I + Γ model was constructed using the “PhyML 3.1” executable program in R 4.1.1 [51].

2.3. Hierarchical Modelling of Species Communities

Data were fitted with Hierarchical Modelling of Species Communities (HMSC) [52], which includes a hierarchical layer asking how species respond to environmental covariates [27]. The phylogenetic tree and fish traits were included in the model to improve its performance [28]. Posterior distribution was sampled with four Markov Chain Monte Carlo (MCMC) chains, each of which was run for 300,000 iterations. The first 50,000 iterations were removed as burn-in. The chains were thinned by 1000 to yield 250 posterior samples per chain and so 1000 posterior samples in total. Three competing models were constructed for model validation as follows: (1) model_null, which included only an intercept and three random effects; (2) model_cov, which included an intercept and all predictors but excluded random effects; and (3) model_full, which included an intercept, all predictors, and three random effects. Five-fold cross-validation was performed [53]. Explanatory and predictive powers of these competing models were evaluated in terms of Area Under Curve (AUC) [54] and Tjur’s R2 (TjurR2) [55]. Explained variances were partitioned between fixed and random effects. One-way ANOVA with the method of Tukey’s “Honest Significant Difference” was applied to test if the average explained variances were significantly different between explaining variables at the confidence level of 0.95 [56]. The fine sand was used as the baseline category for the HMSC model’s dummy variable, i.e., substrate. The environmental niche of each species was quantified using the model’s estimated fixed-effect regression coefficients (β parameters) [52]. These β parameters directly describe the magnitude and direction of a species’ response to each environmental gradient, thereby defining its niche. More details for model fitting and model validation are specified in Appendix C.

3. Results

3.1. HMSC Model Validation

The model_full, which incorporated both fixed and random effects, exhibited the highest AUC and TjurR2 values (Table 1, Figure 2), demonstrating superior explanatory and predictive power relative to the other competing models. In contrast, model_cov, which excluded all random effects, had lower explanatory and predictive power than model_full, indicating that spatial and temporal stochasticity contribute to explaining fish community patterns. Model_null, which included only an intercept, showed lower predictive ability compared with model_cov and model_full, confirming that explanatory variables are essential for interpreting community patterns. Additionally, AUC and TjurR2 values for each species in model_full were more convergent both pre- and post-cross-validation, indicating that this model generated the least uncertainty.

3.2. Fixed and Random Effects

Fixed effects of the HMSC model explained 99.40% of the distributions of coastal fish species on average, whereas random effects only explained 0.60% (Figure 3a), indicating strong assembly rules of environmental filtering. SST explained the majority of the variances (49.00%) of the species distributions, followed by SSS (33.25%), seafloor substrate (9.39%), and depth (7.76%) (Figure 3b), indicating that SST and SSS were the most influential drivers.
These results were corroborated by the variance partitioning for the distribution of each species (Appendix B—Figure A7). SST explained the majority of the variances for most species, followed by SSS, depth, substrate, and random effects, indicating that SST was the most important driver for the distribution of most species. Random effects explained only a few variances for most species, indicating that spatial and temporal stochasticity were limited across the range of fish species distributions.

3.3. Species Environmental Niche

For most species, estimated intercepts were below 0 with at least 0.95 posterior probability (Figure 4), indicating that the expected distribution probability for most coastal fish species was smaller than 0.5, i.e., they were not widespread along the whole coastline. Estimated β parameters for the first-order term of SST for most species were below 0, and for the second-order term were higher than 0, indicating that there was an optimal SST value for most species. The mean estimated optimum SST value for these species clustered around 22 °C. The species with significant responses to SSS and depth were fewer than those with significant responses to SST, indicating that most species were more sensitive to SST than to SSS and depth. Estimated β parameters of different substrate categories for most species were positive, indicating that some species were less likely to be distributed in the baseline level of substrate, i.e., fine sand habitat.
Species richness (SR) increased with SST (Figure 5a), which was consistent with the GLMM for the preselection of explaining environmental variables. SR also increased with increasing SSS (Figure 5b), which indicated that most fish species preferred seawater with higher SSS. SR decreased when the water was shallower (Figure 5c), indicating SR was higher in deeper coastal waters. SR was lower in habitats with fine sand (FS), but higher in sandy silt (ST), silty (T), and silty sand (TS) habitats (Figure 5d), corroborating that most species preferred other substrate categories than fine sand habitats.

3.4. Residual Species Associations

Both positive and negative co-occurrence patterns occurred in the fish communities (Figure 6). The percentage of species pairs that showed significant raw associations was 36.54%, while the percentage of species pairs that showed significant residual associations was 9.78%. Residual associations were fewer and more sporadic than raw associations, indicating that most species co-occurrences had been explained in model_full by the fixed and random effects (Figure 6a). The highest positive residual association occurred between mottled skate Beringraja pulchra and stichaeid fish Chirolophis japonicus with a value of 0.91, and the highest negative residual association occurred between stone moroko Pseudorasbora parva and cardinal fish Jaydia lineata with a value of −0.79 (Figure 6b). The most frequent residual associations occurred within or between Gobiidae species and other taxons with a percentage of 13.17% (Figure 6c,d).

4. Discussion

4.1. Performance of HMSC Modeling

Predictor selection is important for regression-based models to minimize the risk of over-parameterization and maximize the capability of extrapolation [57]. Anthropogenic disturbances, particularly marine pollution, overfishing, and management measures such as summer fishing moratoriums in the coastal seas of China, also exert specific impacts on coastal fish communities [4]. These factors are challenging to adequately incorporate into the current HMSC model, primarily due to the limited availability of valid historical data dating back to the 1980s. Consequently, the impacts of these unincluded factors are retained in the residuals or partially accounted for by the random effect of spatial autocorrelation. However, values of AUC were higher than 0.80 for model_full, which can be viewed as a “good” model according to Araujo et al. [58]. Tjur R2 for model_full was quite smaller compared with AUC, because for AUC the baseline that the model prediction is equally good as expected by random is 0.5, whereas for Tjur R2 the same baseline is 0 [52]. Values of Tjur R2 were higher than 0.35, which reassured that the model_full possessed reasonable explanatory and predictive power [55].

4.2. Species Niche and Environmental Filtering

Species thermal niche is one of the most popular subjects of environmental filtering. It is reported that temperature was the only influential factor for marine communities across 13 major taxa at global scales, including animals and plants [59]. In this work, both temperature and salinity were the most influential factors in coastal China seas, possibly because of the existence of many estuaries along the coastlines. China has the most abundant estuarine resources in the world [60]. Over twenty estuaries are characterized by an annual freshwater flux of 5 × 108 m3, including the Yangtze River Estuary, Yellow River Estuary, and Pearl River Estuary. The estuary and its adjacent sea areas are a mixture of fresh and brackish water and therefore present a heterogeneous salinity gradient [61]. The vast gradient of temperature and salinity in the estuarine waters provides variate habitats for fish species and therefore influences the distributions of coastal fish species [15]. Species that adapt to brackish water or seek anadromous/catadromous migration compose different community structures in the estuary than in other salty waters [62]. Therefore, salinity was an important driver, especially for some freshwater/brackish species. For example, half-smooth tongue sole Cynoglossus semilaevis, cardinal fish Jaydia lineata, and silver pomfret Pampus argenteus are estuarine species, and salinity was the only significant driver for their distributions.
Water depth and seafloor substrate were also influential. Some fish adjust their bathymetrical behavior, such as vertical distribution in the water column and habitat selection, to cope with the thermal variation, which possibly correlates to the effect of depth on fish species distributions [63]. Seafloor substrate directly affects demersal habitat conditions. Different substrate types were categorized mainly by rugosity [36,37]. Responses of fish communities to rugosity were not congruent in different ecosystems. For example, in natural coral reefs, fish species richness and abundance were statistically higher in habitats with higher rugosity [64], but in rocky ecosystems, fish communities in substrate habitats with limestones were significantly different from those with granites [32]. There were scarce studies concerning the mechanism that why the species distributions in fine sand habitats were lower than in other habitats, but others have found that this is likely correlated to nursery and foraging habitat availability [32,65]. The availability of specific types of niches in habitats with specific rugosity types can lead to habitat specialization [66].

4.3. Species Associations and Biotic Filtering

It has been underlined that species co-occurrence patterns, i.e., species associations in HMSC, do not necessarily indicate species interspecific interactions [28]. Fish species associations that occurred in the coastal waters of China seas may be caused by a variety of factors, including but not limited to: (1) the environmental niches of different species are similar, and similar environmental filtering causes co-occurrence of species [67]; (2) there are positive or negative interspecific interactions between species, such as competition, parasitism, predation, etc. [68]; and (3) other random historical incidences, such as the formation of Taiwan Strait twenty thousand years ago in the last stage of the Paleolithic era that possibly indirectly influence fish species associations by shaping past habitat connectivity and geographic isolation because of the rising levels of the ocean [69]. However, the residual species associations are indicative of interspecific interactions, especially between those species that have never been studied [70]. For example, mottled skate Beringraja pulchra and stichaeid fish Chirolophis japonicus have a highly positive correlation (correlation coefficient 0.91). They are both small or medium-sized benthic fish feeding on small invertebrates [31]. Although there was no relevant report on the interspecific relationship between them, the significantly positive residual species association provided reasonable hypotheses that although they prey on the same food in the same area, the interspecific competition is weak, or even interspecific facilitation might exist. From the perspective of trophic niche breadth, these two species may differentiate in specific prey types. Beringraja pulchra possibly mainly targets burrowing invertebrates, and Chirolophis japonicus feed on epibenthic invertebrates, which reduces their resource overlap. Pacific cod Gadus macrocephalus is one of the important economic fish in northern coastal China [71]. It is a benthic carnivorous fish, and the adults feed on small fish [72]. Japanese flounder Paralichthys olivaceus is also a benthic fish, feeding on small fish and invertebrates [73]. The body size of Gadus macrocephalus and Paralichthys olivaceus is similar [31], and their negative residual association possibly suggests interspecific competition for food. Gobiidae is one of the largest families of marine fishes, with at least 2000 species described in 200 genera [74]. Gobiidae species, i.e., gobies, are generally benthic and small-sized (often less than 50 mm), and may occupy various niches in the substrate, including the bodies or burrows of invertebrates [75]. Gobiidae was also the most frequent taxon in this work (87 gobies out of 384 fish species). Therefore, more common resource utilization, segregation of habitat preferences, and food web correlations within or between gobies and other taxons were expected, which possibly explained why both positive and negative residual associatons were the most frequent for gobies. For instance, interspecific social interactions to a large extent explained the partial habitat separation between the three gobies under laboratory mud-sand and grass habitats [76]. Residual species association needs to cooperate with other ecological studies or laboratory experiments to further determine if interspecific interactions exist [77]. Nevertheless, residual species association provided us with additional useful information for diversity conservation or interspecific relationship research. For example, proper consideration of interspecific relationships of species can improve the performance of conservation prioritization.
Using static depth and substrate across 39 years is a limitation of the model due to data unavailability, especially given potential coastal sedimentation/erosion over decades. Future studies could integrate long-term data on anthropogenic disturbances (e.g., fishing intensity, marine pollution), depth, and substrate that were not fully accounted for in this work into the HMSC framework, and combine field surveys or laboratory experiments to verify the potential interspecific interactions of Gobiidae and other taxa, thereby further refining the prediction of coastal fish community dynamics and providing more precise guidance for marine biodiversity conservation.

5. Conclusions

The Hierarchical Modelling of Species Communities (HMSC) employed in this study exhibited strong explanatory and predictive power. Environmental filtering drives the spatiotemporal patterns of fish communities in coastal seas. The influence of random effects is limited. Besides SST, other variables, including SSS, seafloor substrate, and water depth, are also significantly influential factors driving coastal fish communities. More interspecific associations occur for Gobiidae species, possibly because of their wide range of distribution and resource utilization. This work provides important information for coastal fish conservation management and is precursory for predictions of coastal fish communities.

Author Contributions

Conceptualization, L.L. and B.K.; methodology, L.L. and Y.L.; validation, L.L. and Y.L.; formal analysis, L.L.; investigation, L.L. and Y.L.; resources, L.L. and B.K.; data curation, L.L. and Y.L.; writing—original draft preparation, L.L.; writing—review and editing, Bing Kang; visualization, L.L. and Y.L.; supervision, B.K.; project administration, B.K.; funding acquisition, B.K. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by the National Natural Science Foundation of China (No. U20A2087 and No. 41976091), the Dongfeng Lake and Liuchong River Basin of Observation and Research Station of Guizhou Province (No. QKHPT YWZ [2025] 002), and the Science and Technology Project of Bijie City of Open Competition Mechanism to Select the Best Candidates (Grant No: BKHZDZX [2023]1). We appreciate the data support by China National Science & Technology Infrastructure—National Marine Data Center and E.U. Copernicus Marine Service Information.

Institutional Review Board Statement

Ethical review and approval were waived for this study due to the sampling process involving no fish capture and the data source being historical investigation records.

Informed Consent Statement

Not applicable.

Data Availability Statement

The data underlying this study are openly available in Figshare at https://doi.org/10.6084/m9.figshare.21864348.v3 [78].

Acknowledgments

We appreciate the data support by China National Science & Technology Infrastructure—National Marine Data Center and E.U. Copernicus Marine Service Information.

Conflicts of Interest

The authors declare no conflicts of interest.

Abbreviations

The following abbreviations are used in this manuscript:
HMSCHierarchical Modelling of Species Communities
SSTSea surface temperature
SSSSea surface salinity
NPPSea surface water net primary production
COICytochrome oxidase I
AICAkaike Information Criterion
MLMaximum Likelihood
MCMCMarkov Chain Monte Carlo chains
AUCArea Under Curve
TjurR2Tjur’s R2
SRSpecies richness
FSfine sand
STsandy silt
Tsilty
TSsilty sand

Appendix A. Tables

Table A1. Sampling site information.
Table A1. Sampling site information.
Site CodeNameLongtitueLatitudeType
YLHYalu_River_Estuary124.270039.9099estuary
QDZBQingduizi_Bay123.313739.7403bay
DLBDalian_Bay121.765638.9624bay
LHELiaohe_Estuary121.812040.8753estuary
FZBFuzhou_Bay121.448439.6933bay
HHEHaihe_Estuary117.881239.0222estuary
QHDQinhuangdao119.588539.8748bay
TSTangshan118.755539.0972bay
CZCangzhou117.744538.4324bay
YRYellow_River_Estuary119.283637.8584estuary
TZBTaozi_Bay121.310037.6280bay
LZBLaizhou_Bay119.345137.6836bay
RCHRongcheng_Bay122.636137.3640bay
RSBRushan_Bay121.482236.6802bay
LSBLaoshan_Bay120.785836.4008bay
JiZBJiaozhou_Bay120.287936.0395bay
HZBHaizhou_Bay119.388134.8817bay
YCYancheng120.821533.4082bay
LSLvsi_Fishang_Ground121.438832.4997bay
YTRYangtze_River_Estuary121.999931.4612estuary
HaZBHangzhou_Bay121.492230.4973bay
ZSBZhoushan_Bay122.226829.8811bay
SMWSanmen_Bay121.467529.0681bay
WZBWenzhou_Bay121.017627.9848bay
SSBSansha_Bay119.813626.5192bay
MJEMin_River_Estuary119.675726.0894estuary
XHBXinghua_Bay119.387925.3941bay
QZBQuanzhou_Bay118.772624.8341bay
JLEJiulongjiang_Esturay117.992724.4342estuary
DSBDongshan_Bay117.543923.8171bay
STBShantou_Bay116.831023.3084Bay
DYBDaya_Bay114.668523.7069bay
PREPearl_River_Estuary113.755422.5379estuary
MYEMoyang_Estuary112.076521.7131estuary
LZWLeizhou_Bay110.525820.9146bay
FCBFangchenggang_Bay108.370021.5411bay
BLEBeilunhe_Estuary108.063821.4557estuary
HKBHaikou_Bay110.282520.0773bay
WQEWanquan_Estuary110.602319.1566estuary
SYBSanya_Bay109.480718.2711bay
CJEChangjiang_Estuary108.933719.5213bay
DSEDanshuihe_Estuary121.410825.1792estuary
WXEWuxi_Estuary120.476024.2046estuary
ZWEZengwenxi_Estuary121.410825.1792estuary
SXEShuangxi_Estuary121.965225.0380estuary
Table A2. Environmental variables that potentially affect fish communities in coastal China seas.
Table A2. Environmental variables that potentially affect fish communities in coastal China seas.
VariableDescriptionUnit/Levels
DepthMean water depth.m
SubstrateTypes of seafloor surface substrate.Fine sand (FS), sand (S), sandy silt (ST), silt (S), silty sand (TS), and clay silt (YT)
CurrentThe north component of sea surface current.m/h
SSTSea surface temperature.°C
SSSSea surface water salinity. PSS
pHSea surface water pH reported on total scale.1
NPPSea surface net primary production of biomass expressed as carbon per unit volume in sea water.mg/m3/day
Table A3. Functional traits which are potentially responsive to environmental variables. Source indicates how the functional traits are compiled. The first nine morphological traits are derived from the morphological measurements. These measurements are standard length (Bl), body depth (Bd), Head depth (Hd), Caudal peduncle depth (CPd), eye diameter (Ed), eye position (Eh), mouth height (Mo), maxillary jaw length (Jl), pectoral fin length (PFl), and pectoral fin position (PFi). For example, Bl/Bd stands for Bl divided by BD. The last nine traits are obtained from FishBase.
Table A3. Functional traits which are potentially responsive to environmental variables. Source indicates how the functional traits are compiled. The first nine morphological traits are derived from the morphological measurements. These measurements are standard length (Bl), body depth (Bd), Head depth (Hd), Caudal peduncle depth (CPd), eye diameter (Ed), eye position (Eh), mouth height (Mo), maxillary jaw length (Jl), pectoral fin length (PFl), and pectoral fin position (PFi). For example, Bl/Bd stands for Bl divided by BD. The last nine traits are obtained from FishBase.
TraitCodeSourceGroupType
Body elongationBElBl/BdMorphologicalContinuous
Vertical eye positionVEpEh/BdMorphologicalContinuous
Relative eye sizeREsEd/HdMorphologicalContinuous
Oral gape positionOGpMo/BdMorphologicalContinuous
Relative maxillary lengthRMlJl/HdMorphologicalContinuous
Body lateral shapeBLsHd/BdMorphologicalContinuous
Pectoral fin vertical positionPFvPFi/BdMorphologicalContinuous
Pectoral fin sizePFsPFl/BlMorphologicalContinuous
Caudal peduncle throttlingCPtCFd/CPdMorphologicalContinuous
Body shapeBodyShapeFishBaseMorphologicalCategorical
Maximum life spanLife_spanFishBaseLife HistoryContinuous
Generation timeGeneration_timeFishBaseLife HistoryContinuous
Food consumption to biomass ratioQ.BFishBaseLife HistoryContinuous
Maximum body lengthMaxLengthTLFishBaseLife HistoryContinuous
Position in water columnDemersPelagIFishBaseEcologicalCategorical
DietDietFishBaseEcologicalCategorical
Trophic levelTrophFishBaseEcologicalContinuous

Appendix B. Figures

Animals 15 03108 g0a1
Figure A1. Basic statistics for the sampling years and sampling methods. Sampling methods (a) are identified according to the definition and classification of fishing gear categories by Food and Agriculture Organization of the United Nations (FAO) [30]. The sampling year (b) indicates the period when the fish investigation was conducted. Length of the horizontal column representing “2018–2018” indicates the number of investigations occurred in that year.
Figure A1. Basic statistics for the sampling years and sampling methods. Sampling methods (a) are identified according to the definition and classification of fishing gear categories by Food and Agriculture Organization of the United Nations (FAO) [30]. The sampling year (b) indicates the period when the fish investigation was conducted. Length of the horizontal column representing “2018–2018” indicates the number of investigations occurred in that year.
Animals 15 03108 g0a1
Animals 15 03108 g0a2
Figure A2. Histograms showing the distribution of (a) species richness (number of species) and (b) species prevalence (frequency of species occurrence) across samples. The x-axis in (a) represents species richness, and in (b) represents species prevalence; the y-axis in both panels represents the count of samples falling into each bin.
Figure A2. Histograms showing the distribution of (a) species richness (number of species) and (b) species prevalence (frequency of species occurrence) across samples. The x-axis in (a) represents species richness, and in (b) represents species prevalence; the y-axis in both panels represents the count of samples falling into each bin.
Animals 15 03108 g0a2
Animals 15 03108 g0a3
Figure A3. Multicollinearity analysis of environmental variables. (a) Correlation matrix heatmap depicting the Pearson correlation coefficients between environmental variables, including the northward component of the sea surface current (Current), sea surface salinity (SSS), Sea surface temperature (SST), net primary production (NPP), and sea surface pH (pH). Cross marks indicate non-significant correlations, while colored cells represent significant correlations (blue for positive, red for negative) with corresponding coefficient values. (b) Variance Inflation Factor (VIF) bar plot showing the VIF values for each environmental variable; a dashed line at y = 3 serves as a threshold for identifying potential multicollinearity.
Figure A3. Multicollinearity analysis of environmental variables. (a) Correlation matrix heatmap depicting the Pearson correlation coefficients between environmental variables, including the northward component of the sea surface current (Current), sea surface salinity (SSS), Sea surface temperature (SST), net primary production (NPP), and sea surface pH (pH). Cross marks indicate non-significant correlations, while colored cells represent significant correlations (blue for positive, red for negative) with corresponding coefficient values. (b) Variance Inflation Factor (VIF) bar plot showing the VIF values for each environmental variable; a dashed line at y = 3 serves as a threshold for identifying potential multicollinearity.
Animals 15 03108 g0a3
Animals 15 03108 g0a4
Figure A4. Morphological measurements of the fish. The measurements include standard length (Bl), body depth (Bd), head depth (Hd), caudal peduncle depth (CPd), eye diameter (Ed), eye position (Eh), mouth height (Mo), maxillary jaw length (Jl), pectoral fin length (PFl), and pectoral fin position (PFi).
Figure A4. Morphological measurements of the fish. The measurements include standard length (Bl), body depth (Bd), head depth (Hd), caudal peduncle depth (CPd), eye diameter (Ed), eye position (Eh), mouth height (Mo), maxillary jaw length (Jl), pectoral fin length (PFl), and pectoral fin position (PFi).
Animals 15 03108 g0a4
Animals 15 03108 g0a5
Figure A5. Correlation matrix heatmap depicting the Pearson correlation coefficients between functional traits. Cross marks indicate non-significant correlations, while colored cells represent significant correlations (blue for positive, red for negative) with corresponding coefficient values.
Figure A5. Correlation matrix heatmap depicting the Pearson correlation coefficients between functional traits. Cross marks indicate non-significant correlations, while colored cells represent significant correlations (blue for positive, red for negative) with corresponding coefficient values.
Animals 15 03108 g0a5
Animals 15 03108 g0a6
Figure A6. Akaike Information Criterion (AIC) for Maximum Likelihood (ML) phylogenetic analysis. The substitution model “GTR + I + Γ” is the one that achieves the lowest AIC, where “GTR” denotes the general time-reversible model; meanwhile, “+I” and “+G” indicate that invariant sites and a gamma distribution of substitution rates have been specified.
Figure A6. Akaike Information Criterion (AIC) for Maximum Likelihood (ML) phylogenetic analysis. The substitution model “GTR + I + Γ” is the one that achieves the lowest AIC, where “GTR” denotes the general time-reversible model; meanwhile, “+I” and “+G” indicate that invariant sites and a gamma distribution of substitution rates have been specified.
Animals 15 03108 g0a6
Animals 15 03108 g0a7
Figure A7. Variance partitioning for the distributions of each species.
Figure A7. Variance partitioning for the distributions of each species.
Animals 15 03108 g0a7

Appendix C. Procedures for HMSC

Appendix C.1. Preselection of Random Effects

Possible random effects were checked and determined in advance by fitting species richness (SR) in a generalized linear mixed model (GLMM) with log link and Poisson error distribution [79,80]. Therefore, three crossed random effects were initially considered according to the data structure (Figure A8) [81], including (1) random intercept of sampling site, to account for spatial non-independence of the samples within sites; (2) random intercept of sampling year, to account for temporal non-dependence of the samples; and (3) random intercept of sampling method, to account for the variance caused by different sampling methods. GLMM was fitted with different combinations of random effects and Akaike information criterion (AIC) was evaluated [82]. The random effect combination which creates the lowest AIC value was random intercepts of sampling site plus sampling year (Table A4). This was corroborated by the influence of random effects on the intercept of GLMM (Figure A9). Values of the intercepts under different sampling methods did not deviate from the expected value of 1 (Figure A9a), but sampling sites and sampling years influenced some values of the intercepts Figure A9b,c), indicating that different sampling methods in the data did not significantly influence the expected values of SR. Therefore, random intercepts of sampling site and sampling year were used in subsequent HMSC models. To account for potential spatial autocorrelation, another random effect of spatial autocorrelation based on geographic coordinates of the sampling sites was also included.
Table A4. Values of the Akaike information criterion (AIC) of the models with different combinations of random effects. No random effect indicates the regression model only includes fixed effects. Site, method, and year represent the random intercept of sampling site, sampling year, and sampling method, respectively. The plus sign “+” indicates the inclusion of different random effects simultaneously, e.g., site + method means the component model includes both random intercepts of sampling site and sampling method.
Table A4. Values of the Akaike information criterion (AIC) of the models with different combinations of random effects. No random effect indicates the regression model only includes fixed effects. Site, method, and year represent the random intercept of sampling site, sampling year, and sampling method, respectively. The plus sign “+” indicates the inclusion of different random effects simultaneously, e.g., site + method means the component model includes both random intercepts of sampling site and sampling method.
CombinationdfAIC
No random effect122315.83
Site131892.60
Method132081.80
Year131490.26
Site + method141764.38
Site + year141382.03
Method + year141424.43
Site + method + year151365.84
Animals 15 03108 g0a8
Figure A8. Demonstration of the data structure. Sample 1, 2, 3 etc. represent different fish investigations. There are altogether 45 sites, 14 levels of sampling methods, 62 levels of sampling years, and 148 samples.
Figure A8. Demonstration of the data structure. Sample 1, 2, 3 etc. represent different fish investigations. There are altogether 45 sites, 14 levels of sampling methods, 62 levels of sampling years, and 148 samples.
Animals 15 03108 g0a8
Animals 15 03108 g0a9
Figure A9. Influence of random effects on fish species richness. The X-axis indicates the standardized intercept of the generalized mixed model fitting species richness, and the Y-axis indicates different categories of (a) sampling method, (b) sampling site, and (c) sampling year. Categories of sampling method are identified according to the definition and classification of fishing gear categories by Food and Agriculture Organization of the United Nations (FAO). Codes of sampling sites are listed in Appendix A—Table A1. The sampling year indicates the period when the fish investigation was conducted. For example, 2017–2018 means the investigation was carried out from 2017 to 2018.
Figure A9. Influence of random effects on fish species richness. The X-axis indicates the standardized intercept of the generalized mixed model fitting species richness, and the Y-axis indicates different categories of (a) sampling method, (b) sampling site, and (c) sampling year. Categories of sampling method are identified according to the definition and classification of fishing gear categories by Food and Agriculture Organization of the United Nations (FAO). Codes of sampling sites are listed in Appendix A—Table A1. The sampling year indicates the period when the fish investigation was conducted. For example, 2017–2018 means the investigation was carried out from 2017 to 2018.
Animals 15 03108 g0a9

Appendix C.2. Preselection of Explaining Variables and Fish Traits

In case of over-parameterization by environmental variables, the significance of environmental variables was checked in the GLMM fitting SR with random sampling site and sampling year. Sea surface temperature (SST), sea surface salinity (SSS), depth, and substrate were those variables that significantly influenced the patterns of SR (Table A5). In case of over-parameterization by fish functional traits, a fourth-corner combined with RLQ analysis was performed [83]. The functional traits which significantly correlated with the environment variables (Figure A10) were pectoral fin size (PFs), caudal peduncle throttling (CPt), food consumption to biomass ratio (Q/B), body shape (BodyShapeI), and position in the water column (DemersPelag). Therefore, these environmental variables and functional traits were used in subsequent HMSC models. We included both the first-order and second-order terms of SST, SSS, and depth, assuming that most species had their optimum preferences for these variables [16,84,85].
Table A5. Significance of different environmental variables. Explaining variables are net primary production (NPP), sea surface pH (pH), sea surface salinity (SSS), sea surface temperature (SST), the north component of sea surface current (current), mean water depth (depth), and types of seafloor surface substrate (substrate). The categories of the substrate are fine sand (FS), sand (S), sandy silt (ST), silt (S), silty sand (TS), and clay silt (YT). FS is treated as a basic level in the model and integrated into the intercept. With the multiplicative Poisson model, incidence rate ratios are the exponents of the regression coefficients, CI refers to the confidence interval at the level of 95%.
Table A5. Significance of different environmental variables. Explaining variables are net primary production (NPP), sea surface pH (pH), sea surface salinity (SSS), sea surface temperature (SST), the north component of sea surface current (current), mean water depth (depth), and types of seafloor surface substrate (substrate). The categories of the substrate are fine sand (FS), sand (S), sandy silt (ST), silt (S), silty sand (TS), and clay silt (YT). FS is treated as a basic level in the model and integrated into the intercept. With the multiplicative Poisson model, incidence rate ratios are the exponents of the regression coefficients, CI refers to the confidence interval at the level of 95%.
PredictorsIncidence Rate RatiosCIp
(Intercept)23.1210.60–50.42<0.001***
Substrate [S]2.351.03–5.370.043*
Substrate [ST]2.231.02–4.870.045*
Substrate [T]2.340.99–5.560.053
Substrate [TS]1.690.73–3.880.219
Substrate [YT]1.980.89–4.390.094
SST1.231.10–1.37<0.001***
SSS1.161.02–1.310.019**
Depth1.161.02–1.330.023*
NPP0.820.72–0.940.065
pH0.920.85–1.000.064
Current1.040.98–1.110.155
Significance codes: *** p < 0.001, ** p < 0.01, * p < 0.05.
Animals 15 03108 g0a10
Figure A10. Correlations between fish traits and environmental variables based on fourth-corner combined with RLQ analysis. (a) Fourth-corner tests between the first two RLQ axes for environmental gradients (AxR1/AxR2) and traits. (b) Fourth-corner tests between the first two RLQ axes for trait syndromes (AxQ1 and AxQ2) and environmental variables. Positive significant associations are represented by red lines and cells, and negative significant associations by blue lines and cells. Variables with no significant associations are shown in grey. p values were adjusted for multiple comparisons using the FDR procedure. Environmental variables include sea surface salinity (SSS), sea surface temperature (SST), net primary production (NPP), mean depth, the north component of sea surface current (current), and sea surface pH (pH). Codes for traits are explained in Appendix A—Table A3.
Figure A10. Correlations between fish traits and environmental variables based on fourth-corner combined with RLQ analysis. (a) Fourth-corner tests between the first two RLQ axes for environmental gradients (AxR1/AxR2) and traits. (b) Fourth-corner tests between the first two RLQ axes for trait syndromes (AxQ1 and AxQ2) and environmental variables. Positive significant associations are represented by red lines and cells, and negative significant associations by blue lines and cells. Variables with no significant associations are shown in grey. p values were adjusted for multiple comparisons using the FDR procedure. Environmental variables include sea surface salinity (SSS), sea surface temperature (SST), net primary production (NPP), mean depth, the north component of sea surface current (current), and sea surface pH (pH). Codes for traits are explained in Appendix A—Table A3.
Animals 15 03108 g0a10

Appendix C.3. HMSC Model Fitting and Validation

Data were fitted with Hierarchical Modelling of Species Communities (HMSC) [27,52,86]. Bayesian generalized linear mixed model with a probit link and binomial error distribution was applied in HMSC models with the R-package Hmsc assuming the default prior distributions [84]. Posterior distribution was sampled with four Markov Chain Monte Carlo (MCMC) chains, each of which was run for 300,000 iterations, of which the first 50,000 were removed as burn-in. The chains were thinned by 1000 to yield 250 posterior samples per chain and so 1000 posterior samples in total. MCMC convergence was checked by examining the potential scale reduction factors of the model parameters [85]. Three competing models were constructed for model validation, including: (1) model_null which only included an intercept and three random effects; (2) model_cov which include an intercept and all predictors, but no random effects; and (3) model_full which included an intercept, all predictors, and three random effects. MCMC chain convergence, which was evaluated in terms of potential scale reduction factors [86], was acceptable for these competing models as the potential scale reduction factors for parameters β, γ, and ω were essentially converged to one (Figure A11).
Animals 15 03108 g0a11
Figure A11. MCMC convergence diagnostics for the competing models in terms of potential scale reduction factors. Potential scale reduction factors are shown for species niches, i.e., coefficients of the explaining environmental variables (β), influences of traits on niches (γ), and residual species associations (ω). The competing models include (1) model_null (a–c) which only included random effects; (2) model_cov (d–f) which include all predictors, but no random effects; (3) model_full (g–i) which included all predictors and random effects; (4) model_joint (j–l) which performed a variable selection jointly for all species by a spike and slab prior process; and (5) model_separate (m–o) which performed a variable selection individually for each species also by the method of spike and slab prior, but species selected these variables separately.
Figure A11. MCMC convergence diagnostics for the competing models in terms of potential scale reduction factors. Potential scale reduction factors are shown for species niches, i.e., coefficients of the explaining environmental variables (β), influences of traits on niches (γ), and residual species associations (ω). The competing models include (1) model_null (a–c) which only included random effects; (2) model_cov (d–f) which include all predictors, but no random effects; (3) model_full (g–i) which included all predictors and random effects; (4) model_joint (j–l) which performed a variable selection jointly for all species by a spike and slab prior process; and (5) model_separate (m–o) which performed a variable selection individually for each species also by the method of spike and slab prior, but species selected these variables separately.
Animals 15 03108 g0a11
Explanatory and predictive powers of these competing models were evaluated in terms of Area Under Curve (AUC) [54] and Tjur’s R2 (TjurR2) [55]. To compute explanatory power, model predictions were projected based on the whole data set. To compute predictive power, a five-fold cross-validation was performed, in which the sampling units were assigned randomly to five folds, and predictions for each fold were based on a model fitted to the remaining four folds [86]. Therefore, AUC and TjurR2 were calculated based on all data sets, indicating the explanatory power of the model, while AUC_CV and TjurR2_CV were based on a five-fold cross-validation, indicating predictive power.

References

  1. Barbier, E.B.; Hacker, S.D.; Kennedy, C.; Koch, E.W.; Stier, A.C.; Silliman, B.R. The value of estuarine and coastal ecosystem services. Ecol. Ecol. Monogr. 2011, 81, 169–193. [Google Scholar] [CrossRef]
  2. Schwartz, M. Encyclopedia of Coastal Science; Springer Science & Business Media: Dordrecht, The Netherlands, 2006. [Google Scholar]
  3. Wang, Y.; Liu, R.; Su, J. Marine Geography of China; China Science Publishing & Media Ltd.: Bejing, China, 2012. [Google Scholar]
  4. Kang, B.; Pecl, G.T.; Lin, L.; Sun, P.; Zhang, P.; Li, Y.; Zhao, L.; Peng, X.; Yan, Y.; Shen, C. Climate change impacts on China’s marine ecosystems. Rev. Fish Biol. Fish. 2021, 31, 599–629. [Google Scholar] [CrossRef]
  5. Liu, J.Y. Status of marine biodiversity of the China Seas. PLoS ONE 2013, 8, e50719. [Google Scholar] [CrossRef]
  6. Beukhof, E.D. Trait-Environment Relationships in Marine Fish: An Exploration of Patterns, Drivers and Assembly Processes Shaping Marine Fish Communities; Technical University of Denmark: Kongens Lyngby, Denmark, 2019; p. 163. [Google Scholar]
  7. Drake, J.A. The mechanics of community assembly and succession. J. Theor. Biol. 1990, 147, 213–233. [Google Scholar] [CrossRef]
  8. Hutchinson, G.E. Homage to Santa Rosalia or why are there so many kinds of animals? Am. Nat. 1959, 93, 145–159. [Google Scholar] [CrossRef]
  9. Hubbel, S.P. The Unified Neutral Theory of Biodiversity and Biogeography; Princeton University Press: Princeton, NJ, USA, 2001. [Google Scholar]
  10. Bruno, J.F.; Stachowicz, J.J.; Bertness, M.D. Inclusion of facilitation into ecological theory. Trends Ecol. Evol. 2003, 18, 119–125. [Google Scholar] [CrossRef]
  11. Keddy, P.A. Assembly and response rules: Two goals for predictive community ecology. J. Veg. Sci. 1992, 3, 157–164. [Google Scholar] [CrossRef]
  12. Zha, Y.; Lindstrom, E.S.; Eiler, A.; Svanback, R. Different roles of environmental selection, dispersal, and drift in the assembly of intestinal microbial communities of freshwater fish with and without a stomach. Front. Ecol. Evol. 2020, 8, 152. [Google Scholar] [CrossRef]
  13. Craven, D.; Hall, J.S.; Berlyn, G.P.; Ashton, M.S.; van Breugel, M. Environmental filtering limits functional diversity during succession in a seasonally wet tropical secondary forest. J. Veg. Sci. 2018, 29, 511–520. [Google Scholar] [CrossRef]
  14. Che, X.L.; Chen, D.J.; Zhang, M.; Quan, Q.; Moller, A.P.; Zou, F.S. Seasonal dynamics of waterbird assembly mechanisms revealed by patterns in phylogenetic and functional diversity in a subtropical wetland. Biotropica 2019, 51, 421–431. [Google Scholar] [CrossRef]
  15. Baptista, J.; Martinho, F.; Martins, R.; Carneiro, M.; Azevedo, M.; Vieira, A.; Gomes, P.; Pardal, M. Water temperature gradient shapes the structure and composition of nearshore marine fish communities in southern Europe. J. Sea Res. 2019, 154, 101807. [Google Scholar] [CrossRef]
  16. Bozzato, D.; Jakob, T.; Wilhelm, C. Effects of temperature and salinity on respiratory losses and the ratio of photosynthesis to respiration in representative Antarctic phytoplankton species. PLoS ONE 2019, 14, e0224101. [Google Scholar] [CrossRef] [PubMed]
  17. Giraldo, C.; Ernande, B.; Cresson, P.; Kopp, D.; Cachera, M.; Travers-Trolet, M.; Lefebvre, S. Depth gradient in the resource use of a fish community from a semi-enclosed sea. Limnol. Oceanogr. 2017, 62, 2213–2226. [Google Scholar] [CrossRef]
  18. Booth, D.J.; Figueira, W.F.; Gregson, M.A.; Brown, L.; Beretta, G. Occurrence of tropical fishes in temperate southeastern Australia: Role of the East Australian Current. Estuar. Coast. Shelf Sci. 2007, 72, 102–114. [Google Scholar] [CrossRef]
  19. Pecuchet, L.; Törnroos, A.; Lindegren, M. Patterns and drivers of fish community assembly in a large marine ecosystem. Mar. Ecol. Prog. Ser. 2016, 546, 239–248. [Google Scholar] [CrossRef]
  20. Limbourn, A.J.; Jones, G.P.; Munday, P.L.; Srinivasan, M. Niche shifts and local competition between two coral reef fishes at their geographic boundary. Mar. Freshw. Res. 2007, 58, 1120–1129. [Google Scholar] [CrossRef]
  21. Weiher, E.; Keddy, P.A. Assembly rules, null models, and trait dispersion: New questions from old patterns. Oikos 1995, 74, 159–164. [Google Scholar] [CrossRef]
  22. Mouillot, D.; Dumay, O.; Tomasini, J.A. Limiting similarity, niche filtering and functional diversity in coastal lagoon fish communities. Estuar. Coast. Shelf Sci. 2007, 71, 443–456. [Google Scholar] [CrossRef]
  23. Tikhonov, G.; Opedal, Ø.H.; Oksanen, J.; de Jonge, M.M.J.; Dallas, T. R Package; Version 3.0-11; Hmsc: Hierarchical Model of Species Communities; rOpensci: Berkeley, CA, USA, 2021. [Google Scholar]
  24. Huang, Y. Species composition and seasonal variations of fishes in Dongshan Bay. J. Fujian Fish. 1999, 2, 55–61. [Google Scholar]
  25. Lian, Z. Research on the fish fauna of the Minjiang River. Chin. J. Zool. 1988, 23, 40–45. [Google Scholar]
  26. Editorial Board of Bays of China. Bays of China; Ocean Press: Beijing, China, 1991; Volume 1. [Google Scholar]
  27. Ovaskainen, O.; Tikhonov, G.; Norberg, A.; Guillaume Blanchet, F.; Duan, L.; Dunson, D.; Roslin, T.; Abrego, N. How to make more out of community data? A conceptual framework and its implementation as models and software. Ecol. Lett. 2017, 20, 561–576. [Google Scholar] [CrossRef]
  28. Tikhonov, G.; Opedal, Ø.H.; Abrego, N.; Lehikoinen, A.; de Jonge, M.M.J.; Oksanen, J.; Ovaskainen, O. Joint species distribution modelling with the R-package Hmsc. Methods Ecol. Evol. 2020, 11, 442–447. [Google Scholar] [CrossRef]
  29. Lin, L.; Liu, Y.; Lin, H.; Kang, B. Considering species functional and phylogenetic rarity in the conservation of fish biodiversity. Divers. Distrib. 2024, 30, e13804. [Google Scholar] [CrossRef]
  30. Nédélec, C.; Prado, J. Definition and Classification of Fishing Gear Categories, 1st ed.; FAO: Fisheries Technical Paper: Rome, Italy, 1990; Volume 222. [Google Scholar]
  31. Froese, R.; Pauly, D. FishBase; World Wide Web electronic publication: Geneva, Switzerland, 2022. [Google Scholar]
  32. Canessa, M.; Bavestrello, G.; Cattaneo-Vietti, R.; Furfaro, G.; Doneddu, M.; Navone, A.; Trainito, E. Rocky substrate affects benthic heterobranch assemblages and prey/predator relationships. Estuar. Coast. Shelf Sci. 2021, 261, 107568. [Google Scholar] [CrossRef]
  33. Hendriks, I.E.; Duarte, C.M.; Alvarez, M. Vulnerability of marine biodiversity to ocean acidification: A meta-analysis. Estuar. Coast. Shelf Sci. 2010, 86, 157–164. [Google Scholar] [CrossRef]
  34. Bailey, S.A.; Horner-Devine, M.C.; Luck, G.; Moore, L.A.; Carney, K.M.; Anderson, S.; Betrus, C.; Fleishman, E. Primary productivity and species richness: Relationships among functional guilds, residency groups and vagility classes at multiple spatial scales. Ecography 2004, 27, 207–217. [Google Scholar] [CrossRef]
  35. Amante, C.; Eakins, B.W. ETOPO1 1 Arc-Minute Global Relief Model: Procedures, Data Sources and Analysis; NOAA technical memorandum NESDIS NGDC-24: Boulder, CO, USA, 2009. [Google Scholar] [CrossRef]
  36. Shepard, F.P. Nomenclature based on sand-silt-clay ratios. J. Sediment. Res. 1954, 24, 151–158. [Google Scholar]
  37. State Oceanic Administration of China. Ocean Atlas of Offshore China: Substrate; Ocean Press: Beijing, China, 2017. [Google Scholar]
  38. National Science & Technology Infrastructure—National Marine Data Center. Regional Reanalysis Data: Temperature, Salinity, and Ocean Current; National Science & Technology Infrastructure—National Marine Data Center: Tianjin, China, 2022. [Google Scholar]
  39. Copernicus Marine Environment Monitoring Service. Global Ocean Biogeochemistry Hindcast; Copernicus Marine Environment Monitoring Service: Toulouse, France, 2019. [Google Scholar] [CrossRef]
  40. Assis, J.; Tyberghein, L.; Bosch, S.; Verbruggen, H.; Serrão, E.A.; De Clerck, O. Bio-ORACLE v2.0: Extending marine data layers for bioclimatic modelling. Glob. Ecol. Biogeogr. 2018, 27, 277–284. [Google Scholar] [CrossRef]
  41. Shrestha, N. Detecting multicollinearity in regression analysis. Am. J. Appl. Math. Stat. 2020, 8, 39–42. [Google Scholar] [CrossRef]
  42. Violle, C.; Navas, M.L.; Vile, D.; Kazakou, E.; Fortunel, C.; Hummel, I.; Garnier, E. Let the concept of trait be functional! Oikos 2007, 116, 882–892. [Google Scholar] [CrossRef]
  43. Brosse, S.; Charpin, N.; Su, G.; Toussaint, A.; Herrera-R., G.A.; Tedesco, P.A.; Villéger, S. FISHMORPH: A global database on morphological traits of freshwater fishes. Glob. Ecol. Biogeogr. 2021, 30, 2330–2336. [Google Scholar] [CrossRef]
  44. Schneider, C.A.; Rasband, W.S.; Eliceiri, K.W. NIH Image to ImageJ: 25 years of image analysis. Nat. Methods 2012, 9, 671–675. [Google Scholar] [CrossRef] [PubMed]
  45. Villéger, S.; Miranda, J.R.; Hernández, D.F.; Mouillot, D. Contrasting changes in taxonomic vs. functional diversity of tropical fish communities after habitat degradation. Ecol. Appl. 2010, 20, 1512–1522. [Google Scholar] [CrossRef]
  46. Trindade-Santos, I.; Moyes, F.; Magurran, A.E. Global patterns in functional rarity of marine fish. Nat. Commun. 2022, 13, 1–9. [Google Scholar] [CrossRef]
  47. Stekhoven, D.J.; Buehlmann, P. MissForest-non-parametric missing value imputation for mixed-type data. Bioinformatics 2012, 28, 112–118. [Google Scholar] [CrossRef] [PubMed]
  48. Benson, D.A.; Cavanaugh, M.; Clark, K.; Karsch-Mizrachi, I.; Lipman, D.J.; Ostell, J.; Sayers, E.W. GenBank. Nucleic Acids Res. 2012, 41, D36–D42. [Google Scholar] [CrossRef] [PubMed]
  49. Larkin, M.A.; Blackshields, G.; Brown, N.P.; Chenna, R.; McGettigan, P.A.; McWilliam, H.; Valentin, F.; Wallace, I.M.; Wilm, A.; Lopez, R.; et al. Clustal W and Clustal X version 2.0. Bioinformatics 2007, 23, 2947–2948. [Google Scholar] [CrossRef]
  50. Posada, D.; Crandall, K.A. Selecting the best-fit model of nucleotide substitution. Syst. Biol. 2001, 50, 580–601. [Google Scholar] [CrossRef]
  51. Guindon, S.; Dufayard, J.F.; Lefort, V.; Anisimova, M.; Hordijk, W.; Gascuel, O. New algorithms and methods to estimate Maximum-Likelihood phylogenies: Assessing the performance of PhyML 3.0. Syst. Biol. 2010, 59, 307–321. [Google Scholar] [CrossRef] [PubMed]
  52. Ovaskainen, O.; Abrego, N. Joint Species Distribution Modelling: With Applications in R; Cambridge University Press: Cambridge, UK, 2020. [Google Scholar]
  53. Rodriguez, J.D.; Perez, A.; Lozano, J.A. Sensitivity analysis of k-fold cross validation in prediction error estimation. IEEE Trans. Pattern Anal. Mach. Intell. 2009, 32, 569–575. [Google Scholar] [CrossRef]
  54. Pearce, J.; Ferrier, S. Evaluating the predictive performance of habitat models developed using logistic regression. Ecol. Model. 2000, 133, 225–245. [Google Scholar] [CrossRef]
  55. Tjur, T. Coefficients of determination in logistic regression models—A new proposal: The coefficient of discrimination. Am. Stat. 2009, 63, 366–372. [Google Scholar] [CrossRef]
  56. Maxwell, S.E.; Delaney, H.D.; Kelley, K. Designing Experiments and Analyzing Data: A Model Comparison Perspective; Routledge: New York, NY, USA, 2017. [Google Scholar]
  57. Blanchet, F.G.; Legendre, P.; Borcard, D. Forward selection of explanatory variables. Ecology 2008, 89, 2623–2632. [Google Scholar] [CrossRef]
  58. Araujo, M.B.; Pearson, R.G.; Thuiller, W.; Erhard, M. Validation of species-climate impact models under climate change. Glob. Change Biol. 2005, 11, 1504–1513. [Google Scholar] [CrossRef]
  59. Tittensor, D.P.; Mora, C.; Jetz, W.; Lotze, H.K.; Ricard, D.; Berghe, E.V.; Worm, B. Global patterns and predictors of marine biodiversity across taxa. Nature 2010, 466, 1098–1101. [Google Scholar] [CrossRef]
  60. Editorial Board of Bays of China. Bays of China; Ocean Press: Beijing, China, 1998; Volume 14. [Google Scholar]
  61. Telesh, I.V.; Khlebovich, V.V. Principal processes within the estuarine salinity gradient: A review. Mar. Pollut. Bull. 2010, 61, 149–155. [Google Scholar] [CrossRef]
  62. Bottom, D.L.; Jones, K.K. Species composition, distribution, and invertebrate prey of fish assemblages in the Columbia River estuary. Prog. Oceanogr. 1990, 25, 243–270. [Google Scholar] [CrossRef]
  63. Freitas, C.; Villegas-Ríos, D.; Moland, E.; Olsen, E.M. Sea temperature effects on depth use and habitat selection in a marine fish community. J. Anim. Ecol. 2021, 90, 1787–1800. [Google Scholar] [CrossRef]
  64. Kuffner, I.B.; Brock, J.C.; Grober-Dunsmore, R.; Bonito, V.E.; Hickey, T.D.; Wright, C.W. Relationships between reef fish communities and remotely sensed rugosity measurements in Biscayne National Park, Florida, USA. Environ. Biol. Fishes 2007, 78, 71–82. [Google Scholar] [CrossRef]
  65. Santoso, P.; Setiawan, F.; Subhan, B.; Arafat, D.; Bengen, D.G.; Iqbal Sani, L.M.; Humphries, A.T.; Madduppa, H. Influence of coral reef rugosity on fish communities in marine reserves around Lombok Island, Indonesia. Environ. Biol. Fishes 2022, 105, 105–117. [Google Scholar] [CrossRef]
  66. Levin, P.; Petrik, R.; Malone, J. Interactive effects of habitat selection, food supply and predation on recruitment of an estuarine fish. Oecologia 1997, 112, 55–63. [Google Scholar] [CrossRef]
  67. Cordero, R.D.; Jackson, D.A. Abiotic factors influence species co-occurrence patterns of lake fishes. J. Anim. Ecol. 2021, 90, 2859–2874. [Google Scholar] [CrossRef]
  68. Masi, M.D.; Ainsworth, C.H.; Kaplan, I.; Schirripa, M. Interspecific interactions may influence reef fish management strategies in the Gulf of Mexico. Mar. Coast. Fish. 2018, 10, 24–39. [Google Scholar] [CrossRef]
  69. Tianjun, G.; Hui, H. Tectonic characteristics and evolution of the Taiwan Strait. Acta Geol. Sin. Engl. Ed. 1995, 8, 1–13. [Google Scholar] [CrossRef]
  70. Tikhonov, G.; Abrego, N.; Dunson, D.; Ovaskainen, O. Using joint species distribution models for evaluating how species-to-species associations depend on the environmental context. Methods Ecol. Evol. 2017, 8, 443–452. [Google Scholar] [CrossRef]
  71. Liu, J.; Ning, P. Species composition and faunal characteristics of fishes in the Yellow Sea. Biodivers. Sci. 2011, 19, 764–769. [Google Scholar] [CrossRef]
  72. Albers, W.; Anderson, P. Diet of Pacific cod, Gadus macrocephalus, and predation on the northern pink shrimp, Pandalus borealis, in Pavlof Bay, Alaska. Fish. Bull. 1985, 83, 601–610. [Google Scholar]
  73. Hamidoghli, A.; Won, S.; Lee, S.; Lee, S.; Farris, N.W.; Bai, S.C. Nutrition and feeding of olive flounder Paralichthys olivaceus: A Review. Rev. Fish. Sci. Aquac. 2020, 28, 340–357. [Google Scholar] [CrossRef]
  74. Thacker, C.E. Systematics of Gobiidae. In The Biology of Gobies; CRC Press: Boca Raton, FL, USA, 2012; pp. 129–136. [Google Scholar]
  75. Thacker, C.E.; Roje, D.M. Phylogeny of Gobiidae and identification of gobiid lineages. Syst. Biodivers. 2011, 9, 329–347. [Google Scholar] [CrossRef]
  76. Wiederholm, A.-M. Habitat selection and interactions between three marine fish species (Gobiidae). Oikos 1987, 48, 28–32. [Google Scholar] [CrossRef]
  77. Zurell, D.; Pollock, L.J.; Thuiller, W. Do joint species distribution models reliably detect interspecific interactions from co-occurrence data in homogenous environments? Ecography 2018, 41, 1812–1819. [Google Scholar] [CrossRef]
  78. Li, L.; Yang, L.; Bin, K. Unraveling fish community assembly rules in coastal China seas based on Hierarchical Modeling of Species Communities. 2023. [Google Scholar] [CrossRef]
  79. Bates, D.; Maechler, M.; Bolker, B.M.; Walker, S.C. Fitting linear mixed-effects models Using lme4. J. Stat. Softw. 2015, 67, 1–48. [Google Scholar] [CrossRef]
  80. Bolker, B.M.; Brooks, M.E.; Clark, C.J.; Geange, S.W.; Poulsen, J.R.; Stevens, M.H.H.; White, J.S.S. Generalized linear mixed models: A practical guide for ecology and evolution. Trends Ecol. Evol. 2009, 24, 127–135. [Google Scholar] [CrossRef] [PubMed]
  81. Faraway, J.J. Extending the Linear Model with R: Generalized Linear, Mixed Effects and Nonparametric Regression Models; Chapman and Hall/CRC: Boca Raton, FL, USA, 2016. [Google Scholar]
  82. Sakamoto, Y.; Ishiguro, M.; Kitagawa, G. Akaike information criterion statistics. In Mathematics and its Applications, 1st ed.; Springer: Dordrecht, The Netherlands, 1986; Volume 81, p. 26853. [Google Scholar]
  83. Dray, S.; Choler, P.; Doledec, S.; Peres-Neto, P.R.; Thuiller, W.; Pavoine, S.; ter Braak, C.J.F. Combining the fourth-corner and the RLQ methods for assessing trait responses to environmental variation. Ecology 2014, 95, 14–21. [Google Scholar] [CrossRef]
  84. Christensen, E.A.; Norin, T.; Tabak, I.; van Deurs, M.; Behrens, J.W. Effects of temperature on physiological performance and behavioral thermoregulation in an invasive fish, the round goby. J. Exp. Biol. 2021, 224, jeb237669. [Google Scholar] [CrossRef]
  85. Gelman, A.; Rubin, D.B. Inference from iterative simulation using multiple sequences. Stat. Sci. 1992, 7, 457–472. [Google Scholar] [CrossRef]
  86. Fushiki, T. Estimation of prediction error by using K-fold cross-validation. Stat. Comput. 2011, 21, 137–146. [Google Scholar] [CrossRef]
Animals 15 03108 g001
Figure 1. Study area and sampling sites [29]. The study area spans a longitudinal range of 105–125° E and a latitudinal range of 16–45° N. Fish survey data were collected at 45 sampling sites distributed along the coastal waters of the China seas encompassing the Bohai Sea, Yellow Sea, East China Sea, and South China Sea (with the Gulf of Tonkin included) over the period from 1980 to 2018. Detailed information on the sampling sites, including their respective codes, geographic coordinates, and additional relevant attributes, is presented in Appendix A—Table A1.
Figure 1. Study area and sampling sites [29]. The study area spans a longitudinal range of 105–125° E and a latitudinal range of 16–45° N. Fish survey data were collected at 45 sampling sites distributed along the coastal waters of the China seas encompassing the Bohai Sea, Yellow Sea, East China Sea, and South China Sea (with the Gulf of Tonkin included) over the period from 1980 to 2018. Detailed information on the sampling sites, including their respective codes, geographic coordinates, and additional relevant attributes, is presented in Appendix A—Table A1.
Animals 15 03108 g001
Animals 15 03108 g002
Figure 2. Validation of the competing models for Hierarchical Modelling of Species Communities (HMSC). The competing models include (1) model_null, which only includes an intercept and three random effects; (2) model_cov, which includes an intercept and all predictors, but no random effects; and (3) model_full, which includes an intercept, all predictors, and three random effects. AUC and TjurR2 in (a) indicate the index of Area Under Curve and Tjur’s R2 evaluated with the whole data set, respectively. AUC_CV and TjurR2_CV in (b) indicate the index of Area Under Curve and Tjur’s R2 with five-fold cross-validation, respectively.
Figure 2. Validation of the competing models for Hierarchical Modelling of Species Communities (HMSC). The competing models include (1) model_null, which only includes an intercept and three random effects; (2) model_cov, which includes an intercept and all predictors, but no random effects; and (3) model_full, which includes an intercept, all predictors, and three random effects. AUC and TjurR2 in (a) indicate the index of Area Under Curve and Tjur’s R2 evaluated with the whole data set, respectively. AUC_CV and TjurR2_CV in (b) indicate the index of Area Under Curve and Tjur’s R2 with five-fold cross-validation, respectively.
Animals 15 03108 g002
Animals 15 03108 g003
Figure 3. Variance partition for fixed and random effects (a), and for each explaining variable (b). Different lower-case letters on top of each box indicate significant differences between groups at the confidence level of 0.95.
Figure 3. Variance partition for fixed and random effects (a), and for each explaining variable (b). Different lower-case letters on top of each box indicate significant differences between groups at the confidence level of 0.95.
Animals 15 03108 g003
Animals 15 03108 g004
Figure 4. Heatmap of estimated parameters β, i.e., species niches. Gradient colors from blue to red show parameters that are estimated to be negative to positive, respectively, with at least 0.95 posterior probability. Symbol “^2” indicates the second-order term of the covariate, while the other counterpart without this symbol indicates the first order. Categorical variables are expanded to their levels. The first level is set as a baseline in the intercept of the model, and the other levels as dummy variables.
Figure 4. Heatmap of estimated parameters β, i.e., species niches. Gradient colors from blue to red show parameters that are estimated to be negative to positive, respectively, with at least 0.95 posterior probability. Symbol “^2” indicates the second-order term of the covariate, while the other counterpart without this symbol indicates the first order. Categorical variables are expanded to their levels. The first level is set as a baseline in the intercept of the model, and the other levels as dummy variables.
Animals 15 03108 g004
Animals 15 03108 g005
Figure 5. Marginal plots for the effects of environmental variables on species richness (SR). Marginal plots are visualized by plotting the response variable against one single explaining variable while maintaining other variables at their mean values. The panels show the expected values of SR along the gradients of sea surface temperature (SST, (a)), sea surface salinity (SSS, (b)), water depth (Depth, (c)), and substrate categories (Substrate, (d)). Red points in (d) indicate the median values.
Figure 5. Marginal plots for the effects of environmental variables on species richness (SR). Marginal plots are visualized by plotting the response variable against one single explaining variable while maintaining other variables at their mean values. The panels show the expected values of SR along the gradients of sea surface temperature (SST, (a)), sea surface salinity (SSS, (b)), water depth (Depth, (c)), and substrate categories (Substrate, (d)). Red points in (d) indicate the median values.
Animals 15 03108 g005
Animals 15 03108 g006
Figure 6. Positive and negative species associations. (a) Raw (upper right) and residual (lower left) species-to-species associations for those species pairs with at least 0.95 support probability for either a positive or negative association. (b) Top 10% positive or negative residual species associations with at least 0.95 support probability. (c) Basic statistics for positive residual associations within or between taxonomic families. (d) Basic statistics for negative residual associations within or between taxonomic families. Numbers in each colored square in (a,b) indicate coefficients of residual species associations. Horizontal axis in (a,b) indicates the counts of residual associations which occurs with at least one species within that family.
Figure 6. Positive and negative species associations. (a) Raw (upper right) and residual (lower left) species-to-species associations for those species pairs with at least 0.95 support probability for either a positive or negative association. (b) Top 10% positive or negative residual species associations with at least 0.95 support probability. (c) Basic statistics for positive residual associations within or between taxonomic families. (d) Basic statistics for negative residual associations within or between taxonomic families. Numbers in each colored square in (a,b) indicate coefficients of residual species associations. Horizontal axis in (a,b) indicates the counts of residual associations which occurs with at least one species within that family.
Animals 15 03108 g006
Table 1. Mean and standard deviation values of the index of Area Under Curve and Tjur’s R2 of different models.
Table 1. Mean and standard deviation values of the index of Area Under Curve and Tjur’s R2 of different models.
AUC *AUC_CV TjurR2 TjurR2_CV
model_cov 0.87 ± 0.08 0.72 ± 0.14 0.26 ± 0.14 0.15 ± 0.14
model_null 0.93 ± 0.6 0.71 ± 0.13 0.13 ± 0.04 0.07 ± 0.08
model_full 0.97 ± 0.04 0.82 ± 0.12 0.49 ± 0.13 0.36 ± 0.13
* The competing models include (1) model_null, which only includes an intercept and three random effects; (2) model_cov, which includes an intercept and all predictors, but no random effects; and (3) model_full, which includes an intercept, all predictors, and three random effects. AUC and TjurR2 indicate the index of Area Under Curve and Tjur’s R2 evaluated with the whole data set, respectively. AUC_CV and TjurR2_CV indicate the index of Area Under Curve and Tjur’s R2 with five-fold cross-validation, respectively.
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Lin, L.; Liu, Y.; Kang, B. Unraveling Fish Community Assembly Rules in Coastal China Seas Based on Hierarchical Modeling of Species Communities. Animals 2025, 15, 3108. https://doi.org/10.3390/ani15213108

AMA Style

Lin L, Liu Y, Kang B. Unraveling Fish Community Assembly Rules in Coastal China Seas Based on Hierarchical Modeling of Species Communities. Animals. 2025; 15(21):3108. https://doi.org/10.3390/ani15213108

Chicago/Turabian Style

Lin, Li, Yang Liu, and Bin Kang. 2025. "Unraveling Fish Community Assembly Rules in Coastal China Seas Based on Hierarchical Modeling of Species Communities" Animals 15, no. 21: 3108. https://doi.org/10.3390/ani15213108

APA Style

Lin, L., Liu, Y., & Kang, B. (2025). Unraveling Fish Community Assembly Rules in Coastal China Seas Based on Hierarchical Modeling of Species Communities. Animals, 15(21), 3108. https://doi.org/10.3390/ani15213108

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop