The study area is situated along the main stem of the Dahan River in northern Taiwan (Fig. 1A, B). This region exposes a thick succession (~2,500 m) of foreland deposits that accumulated during the continuous infilling of a wedge-shaped basin associated with the Taiwan orogeny (Pan et al. 2015). From base to top, the Neogene to Quaternary stratigraphy in the study area comprises the Tapu Formation (Late Miocene), Erchiu Formation (Early Pliocene), Chinshui Shale (Late Pliocene), Cholan Formation (Early Pleistocene), and Yangmei Formation (Early–Middle Pleistocene) (Lin et al. 2003; Fig. 2).

Among these, the Erchiu and Cholan formations are most relevant to the present study, as they contain the targeted fossiliferous intervals. The Erchiu Formation (locally also recognized as the Upper Kueichulin Formation) represents a Lower Pliocene succession characterized by intensely bioturbated muddy sandstone, thick-bedded mudstone, and thin interbeds of sandstone and mudstone (Pan et al. 2015; Chen 2025). These lithofacies indicate a wave- and tide-influenced open-marine depositional environment.

The Cholan Formation represents a Lower Pleistocene succession that can be divided into two distinct intervals (Lin et al. 2003; Pan et al. 2015). The lower portion consists of fine-grained sandstone with planar cross-stratification, tidal bundles, and wavy to flaser bedding, alternating with bioturbated mudstone and reflecting a wave- and tide-dominated coastal setting. Conversely, the upper portion is primarily composed of very fine- to fine-grained sandstone and mudstone interbeds deposited in a fluvial system (Pan et al. 2015). Regionally, the Cholan Formation is renowned for its diverse fossil assemblages, including ostracods (Hu 1976), mollusks (Hu 1992), decapod crustaceans—particularly the crab Galene granulifera Lin, 1947 (Hu 1980; Hu and Tao 1996)—diverse marine fish elements (Lin et al. 2021), and terrestrial mammals such as Mammutidae (Chen and Hsu 2010).

Overall, this stratigraphic framework records well-documented shallow-water deposits formed within comparable coastal settings (Pan et al. 2015; Zhu 2020). Fossils examined in this study were recovered from shell fragment beds in three productive outcrops within this sequence: Tachi-1, Tachi-2, and Sanxia (see below).

Tachi-1

The Tachi-1 site is located west of Tachi Township in Taoyuan County (Fig. 1B, C) and exposes the lower, coastal portion of the Cholan Formation (Fig. 2). Studied fossils and sediments were collected from a 1 m-thick shell fragment bed (Fig. 3A, B), which is interpreted as a transgressive lag deposit (Pan et al. 2015; Zhu 2020). Based on sequence stratigraphy and correlations with Milankovitch cycles, this horizon is dated to approximately 2.0 Ma (Pan et al. 2015; Zhu 2020). Alongside the dense concentration of otoliths and mollusk shells, the bed yielded fossil echinoids, shark teeth, and corals.

Tachi-2

The Tachi-2 site is located near Tachi-1, west of Tachi Township in Taoyuan County (Fig. 1B, C), and exposes the open-marine facies of the Erchiu Formation (Fig. 2). Fossil specimens were collected from a 1 m-thick sandy shell fragment bed (Fig. 3D). This bed is overlain by a thick mudstone interval containing abundant fossiliferous nodules. These nodules frequently encapsulate exceptionally preserved decapod crustaceans (Hu and Tao 1996) and articulated fish skeletons, including the giant basslet Stereolepis arcanum Přikryl & Lin, 2024 (Přikryl et al. 2024).

Sanxia

The Sanxia site is located west of Sanxia District in New Taipei City (Fig. 1B, D). Like Tachi-2, it samples the open-marine facies of the Erchiu Formation (Fig. 2). The studied fossils were recovered from two 1 m-thick shell fragment beds (Fig. 3C) in the lower and upper parts of the outcrop, respectively. The two beds exhibit similar taphonomic and sedimentological characteristics and are only slightly different in age, despite being separated by approximately 50 m stratigraphically. The majority of the samples were collected from the lower bed, with only one bulk sample obtained from the upper bed. In contrast to Tachi-2, the macrofossils at Sanxia, including echinoids and crustaceans, are not preserved within nodules. Importantly, this study represents the first formal publication of paleontological materials from the Sanxia outcrop, highlighting the previously overlooked fossil potential of Pliocene strata in northern Taiwan.

A total of 151 bulk sediment samples, collectively weighing over 747.9 kg, were collected across the three studied localities, with individual bulk samples averaging approximately 5 kg each (Table 1). Prior to processing, all sediment samples were thoroughly dried and weighed. The sediments were then soaked in water and subsequently wet-sieved through a 500 μm mesh screen. All otoliths retained on the 500 μm sieve were hand-picked under a stereomicroscope. High-resolution micrographs of diagnostic specimens were captured using a digital camera mounted on a Nikon SMZ1270 stereomicroscope, with extended depth-of-field composites generated using Helicon Focus software. For broken or poorly preserved material, only specimens retaining diagnostic features (such as a largely intact sulcus acusticus) were included in subsequent taxonomic and quantitative analyses.

Taxonomic identification was carried out by consulting regional literature and comparative otolith guides, including Ohe (1977, 1981, 1985), Yu and Shen (1987), Schwarzhans (1993, 1999, 2019a, 2019b, 2019c), Rivaton and Bourret (1999), Lin and Chang (2012), Nolf (2013), Mitsui et al. (2021, 2024), Kocsis et al. (2024), and Ng et al. (2024). In addition, extensive gobiiform otolith illustrations provided by Schwarzhans et al. (2020a, 2020b), Carolin et al. (2023), and Schwarzhans and Aguilera (2024) were used. Identifications were further verified through direct comparison with modern comparative otolith material housed in our laboratory’s reference collection (registration code prefix CHLOL) and micro-3D-scanned morphological data of Sciaenidae (Mediodia et al. 2026a). Wherever diagnostic characters permitted, specimens were identified to the species level; otherwise, open nomenclature was applied.

All recovered paleontological material is permanently deposited in the collections of the Biodiversity Research Museum, Academia Sinica, Taiwan (BRMAS). Imaged specimens are systematically cataloged under the registration code prefix ASIZF.

Otolith diversity was quantified using Hill numbers (Hill 1973), calculated for three orders of diversity: q = 0, q = 1, and q = 2, corresponding to taxonomic richness, the exponential Shannon diversity, and the inverse Simpson diversity, respectively. These metrics represent the effective number of otolith-based taxa under increasing sensitivity to taxon abundance. Because diversity estimates are sensitive to differences in sample size, observed Hill numbers were standardized using specimen-based rarefaction, and uncertainty was quantified using 1,000 bootstrap replicates to construct 95% confidence intervals (see details in Lin et al. 2023a).

To evaluate temporal changes in fish assemblage composition along the Dahan River section, otolith-based records from the Miocene to the Pleistocene were compiled into a presence/absence sample-by-nominal-species matrix (excluding genus- and family-level indeterminate taxa; Suppl. material 1). The Late Miocene assemblage from Lin and Chien (2022) was included, with several taxonomic revisions updated following Kocsis et al. (2024). After excluding samples devoid of otoliths, the resulting dataset comprised 38 Miocene, 62 Pliocene, and 80 Pleistocene bulk samples and was used for all subsequent compositional analyses.

Shifts in assemblage composition among epochs were assessed using permutational multivariate analysis of variance (PERMANOVA; Anderson 2001) based on Jaccard dissimilarity. Pairwise PERMANOVA tests were performed for all epoch pairs. Because PERMANOVA can be influenced by differences in within-group dispersion, homogeneity of multivariate dispersion among epochs was evaluated using permutational analysis of multivariate dispersions (PERMDISP; Anderson 2006). Temporal variation in fish assemblage composition was visualized using principal coordinates analysis (PCoA).

To further interpret the temporal compositional changes, total Jaccard beta diversity (β_jac, pairwise Jaccard dissimilarity) was partitioned into turnover and nestedness-resultant components following Baselga (2010, 2012) and Ulrich and Almeida-Neto (2012). The turnover component (β_jtu, Jaccard-based turnover) reflects taxonomic replacement between assemblages, whereas the nestedness-resultant component (β_jne, Jaccard-based nestedness) reflects asymmetric taxonomic loss or gain without replacement. In ecological terms, this partitioning distinguishes whether compositional shifts across epochs were driven by true species replacement (e.g., environmental turnover or evolutionary radiation) or whether poorer assemblages were merely nested subsets of richer ones (e.g., habitat filtering or shoaling-induced biodiversity loss). The relative importance of these components was assessed using Wilcoxon signed-rank tests (Wilcoxon 1945).

In addition, taxa characteristic of each temporal assemblage were identified using indicator species analysis based on the IndValg statistic (De Cáceres and Legendre 2009; De Cáceres et al. 2010), with statistical significance assessed by permutation tests. The strength of association between each taxon and a given epoch was quantified by its indicator value (IndVal), ranging from 0 to 1, where higher values indicate a stronger and more exclusive association with that epoch. Rare taxa occurring in fewer than three bulk samples were excluded to reduce the influence of sporadic occurrences. For analyses involving multiple pairwise or taxon-level tests, p-values were adjusted using the Benjamini–Hochberg false discovery rate procedure (Benjamini and Hochberg 1995).

All statistical analyses were conducted in Python v. 3.10.8 (Python Software Foundation 2022) and R v. 4.2.2 (R Core Team 2022). Diversity rarefaction and Hill number estimations were generated using the R package iNEXT (Chao et al. 2014; Hsieh et al. 2016). Compositional analyses, including Jaccard dissimilarity, PERMANOVA, PERMDISP, PCoA, and beta diversity partitioning, were performed using the Python library scikit-bio (Aton et al. 2025). Wilcoxon signed-rank tests were conducted using the Python library SciPy (Virtanen et al. 2020); indicator species analysis was implemented in the R package indicspecies (De Cáceres and Legendre 2009); and multiple-testing corrections were applied using the Python library statsmodels (Seabold and Perktold 2010).

Given the high diversity and abundance of sciaenid otoliths in Neogene and Quaternary coastal to shallow-marine deposits across the Dahan River section and the broader western Pacific, we compiled genus-level sciaenid stratigraphic ranges to (1) evaluate their utility for regional biostratigraphic correlation and (2) provide empirical fossil constraints for comparison with existing molecular-clock divergence-time estimates.

To establish a comprehensive otolith-based chronostratigraphic framework, published sciaenid otolith records were compiled from Japan (Ohe 1975a, 1975b, 1976, 1977, 1981, 1983, 1991, 1996, 2011; Takahashi 1976; Takayasu and Takahashi 1988; Ohe et al. 2016a, 2016b; Mitsui et al. 2024), Taiwan (Lin et al. 2018, 2021, 2022, 2023b; Lin and Chien 2022), Thailand (Jirapatrasilp et al. 2024), and Brunei Darussalam (Stinton 1962; Schwarzhans 1993; Kocsis et al. 2024) (Suppl. material 2). Records with unresolved taxonomic identities, uncertain stratigraphic placement, or questionable biogeographic affinities were excluded (see Suppl. material 2). For each verified sciaenid genus, the fossil occurrences were compiled and visualized to reconstruct stratigraphic ranges; regional biostratigraphic patterns were then evaluated, and these fossil ranges were compared with current molecular divergence estimates across the western Pacific.

Otoliths of Ariosoma are readily recognizable by their distinct S-shaped sulcus and a short, dorsally oriented ostial channel (Schwarzhans 2019b: pl. 2). Specimens representing various growth stages are abundant in the collection. The majority of specimens exhibit a prominently elevated dorsal dome, resembling those of A. anagoides (Schwarzhans 2019b: pl. 2, fig. 2; see also Lin and Chien 2022: fig. 4a–f). Although approximately eight species of Ariosoma are currently recorded from Taiwan, incomplete modern comparative reference collections and published otolith images preclude assignment to named species at present. At least two species are represented, one characterized by a more compact outline and a less extended posterior rim (Fig. 4A); however, species boundaries across the entire collection remain insufficiently resolved, and additional species may be represented. The material is therefore retained collectively as Ariosoma spp.

A broken otolith is assigned to the genus Muraenesox based on its characteristic sulcus morphology and a prominent postero-dorsal angle. Two Muraenesox species, M. bagio (Hamilton, 1822) and M. cinereus (Forsskål, 1775), currently occur in Taiwanese waters, but their otoliths are morphologically very similar and cannot be reliably distinguished (Fig. 4C–E).

The specimens are thick and robust and occur more frequently at the Tachi-1 site. Morphologically, they closely resemble otoliths of extant Coilia mystus (Linnaeus, 1758) (Nolf 2013: pl. 30; Fig. 4G), but the posterior rim is vertically straight in the latter species; hence the tentative assignment here. The only other reported fossil record of this genus is Coilia planata Stinton, 1962, from the Late Miocene of Brunei, which is also morphologically similar to the present specimens. However, the holotype of C. planata represents a smaller, slightly worn individual. Furthermore, as discussed by Kocsis et al. (2024), C. planata exhibits a slightly thinner profile, a shallower ventral area, and a more elongated rostrum, which differ from our specimens.

Six lapilli (utricular otoliths) of Plotosus lineatus were recovered. These lapilli are circular to subrectangular in outline and easily distinguished by a thin outer margin featuring a posteriorly opening groove extending along approximately one-third of the entire outline (Rivaton and Bourret 1999; Lin and Chang 2012). The species was recently reported from the Lower Pliocene strata of Cabarruyan Island in the northern Philippines (Mediodia et al. 2026b).

According to a recent taxonomic study by Lin et al. (2025b), three species, Arius maculatus, Netuma bilineata (Valenciennes, 1840), and Plicofollis nella (Valenciennes, 1840), occur in Taiwanese waters based on external morphology and molecular evidence. Detailed documentation of their lapilli reveals that these species can be reliably distinguished (Ohe 2006; Lin and Chien 2022; Lin et al. 2022, 2025b). Although the margins of the lapilli in the collection are somewhat worn, the gently curved posterior rim and flattened mesial rim agree well with those of A. maculatus. Notably, the otolith figured in Lin et al. (2018: pl. 1, fig. 10) represents N. bilineata rather than Plicofollis, as previously suggested by Lin and Chien (2022).

A slightly longer caudal colliculum and a distinct postero-ventral denticle allow assignment to Benthosema pterotum (Schwarzhans 2019a; Lin et al. 2023b: fig. 7). Otoliths of B. pterotum are exceptionally rare in the samples, which is notable given that mesopelagic taxa are typically scarce in shallow coastal depositional environments, in contrast to their dominance in upper-slope or outer-shelf settings (Lin et al. 2023b). Their sporadic presence in these shallow-water strata was likely driven by trophic transport via predator excretion (Lin et al. 2019).

A single otolith closely matches those of extant Sirembo imberbis, the only extant representative of the genus in Taiwanese waters (Shao 2026). Although moderately preserved, the specimen is readily recognized by its thickset, stocky profile and a wide, shallow sulcus that is strongly constricted only along the crista inferior at the collum. Specimens previously reported as Sirembo sp. by Lin et al. (2018) are here assigned to S. imberbis.

Jaydia and Ostorhinchus are the two apogonid genera most commonly found in modern waters off western Taiwan (Shao 2026). Their otoliths can be distinguished morphologically: otoliths of Jaydia have a more elevated, dome-shaped antero-dorsal rim followed by a deeper dorsal notch, as well as a more pronounced dorsal ostial lobe (see examples in Fig. 7O–R). However, confident assignment to either genus is restricted to well-preserved adult specimens, leaving many smaller or abraded otoliths in open nomenclature as Apogonidae indet.

Otoliths assigned to this family include Taenioides sp., Trypauchen vagina (Bloch & Schneider, 1801), and an unidentified oxudercid. Otoliths of Taenioides are characterized by a sharply inclined dorsal rim, a robust, broad postero-ventral angle, and a distinctly convex inner face, distinguishing them from other similarly sized gobiiform otoliths in the collection. These morphological features align closely with those of extant Taenioides cirratus (Blyth, 1860) (Fig. 8E), but otoliths from other congeners are required for species assignment. Re-evaluation of published material indicates that the otoliths designated as “Gobiidae indet. 5” in Lin and Chien (2022: fig. 7g) belong to Taenioides sp.

Otoliths of T. vagina are highly distinctive due to their boomerang-shaped outline. However, they exhibit pronounced ontogenetic variation: otoliths of smaller individuals are relatively slender, featuring a more erect and straight ventral area, whereas larger specimens develop broader, more strongly sloping antero-ventral rims. The fossils are nearly identical to modern otoliths of T. vagina (Fig. 8I–M). Furthermore, the slightly damaged specimen previously identified as “Microdesmidae indet.” by Lin et al. (2023b: fig. 13k) is here confidently reassigned to T. vagina.

The unidentified oxudercid otolith shows overall outline similarities to Tridentiger, Pterogobius, and Chaenogobius, particularly in possessing a ventral rim that is noticeably longer than the dorsal one (Fig. 9E). However, due to limited comparative material, generic-level assignment is precluded at present.

Gobiid otoliths are the most common and abundant among the families found across the studied sites (Table 2). However, species- or even genus-level identification within Gobiidae remains challenging due to the still-incomplete comparative otolith collections for modern Indo-Pacific species (W. Schwarzhans pers. comm. 2025) and the high proportion of very small juvenile specimens, particularly at the Tachi-2 and Sanxia sites. We therefore adopted a conservative approach, retaining many unassignable specimens in open nomenclature, though continued field sampling and expansion of modern reference collections will likely resolve these issues in the future.

Nevertheless, several taxa can be reasonably identified based on distinct characteristics. Amblychaeturichthys hexanema (Bleeker, 1853) has a flat inner face, a pronounced postero-dorsal extension, and a slightly more compact outline compared with those of Chaeturichthys and Parachaeturichthys, though these distinctions are indeed very subtle (see also Mitsui et al. 2021; Lin and Chien 2022; Fig. 9F–H). Furthermore, the ostial crista superior in both Chaeturichthys and Parachaeturichthys appears to be bifurcated or strongly notched, whereas that of A. hexanema is straight (Fig. 9F–H). It should be noted that the otoliths designated as “Gobiidae indet. 1” in Lin and Chien (2022: fig. 6g–m) are here reassigned to A. hexanema.

Another related taxon identified here is Parachaeturichthys sp., whose otoliths exhibit an even more protruding postero-dorsal extension than those of A. hexanema and a more elongated outline. This characteristic is usually more pronounced in larger individuals of Parachaeturichthys. We note that this taxon is distinct from Parachaeturichthys sp. in Lin and Chien (2022: fig. 6a–f), which possesses a more squarish outline and an inclined dorsal rim.

Otoliths of Myersina filifer (Valenciennes, 1837) and Priolepis sp. conform to the descriptions provided by Lin et al. (2023b), which we follow here. Finally, the assignment of abundant otoliths to Silhouettea sp. remains tentative (W. Schwarzhans pers. comm. 2025); these specimens are typically small and subquadrate, characterized by a slightly inclined dorsal rim, a convex inner face, and a shallow yet somewhat delineated sulcus. In our view, they show characters most closely resembling those of extant Silhouettea (Schwarzhans 2024: fig. 8).

The two recovered specimens are broken at their anterior tips, but both show morphological features diagnostic of M. maculata: a typical pyriform outline with a nearly vertical posterior rim and an ostium that widens gradually anteriorly while curving toward the outer face (Lin and Chang 2012: pl. 102). This discovery is the first fossil record of the species in the western Pacific.

As with gobiid otoliths, pleuronectiform otoliths in the collection consist predominantly of small juvenile specimens, presenting taxonomic challenges. Here, we follow the benchmark work by Schwarzhans (1999) and our regional extant comparative collection covering the most common species. Only specimens identified with high confidence are assigned to the species or genus level, although this means that true species diversity is likely underestimated, particularly given the numerous unassignable juvenile bothid and soleid otoliths.

Tarphops oligolepis (Bleeker, 1858) otoliths are frequently encountered and show a subquadrate outline resembling that of Pardachirus (Soleidae), but the sulcus in T. oligolepis is straight and undivided (Schwarzhans 1999: pp. 134–135, figs 240–242; Mitsui et al. 2021: fig. 8D; Lin and Chien 2022: fig. 8d, e). Specimens identified as Chascanopsetta cf. C. lugubris by Ohe from the Pliocene of central Japan (Ohe 1977: fig. 16; Ohe 1981: pl. 9–9; Ohe 1983: pl. 14, fig. 4) likely belong to T. oligolepis.

Bothid otoliths are characterized by thick, elevated cristae and a narrow, well-divided sulcus that is mainly constricted along the crista inferior (Schwarzhans 1999). In Cynoglossus otoliths, the high-bodied, triangular outline and shallow, hammer-shaped sulcus render them readily distinguishable. However, a comprehensive modern reference collection covering all regional species is currently lacking, precluding further species-level assignments.

Soleidae is the most abundant and diverse pleuronectiform family in the samples (Table 2); its otoliths usually feature a less differentiated sulcus (Schwarzhans 1999). Otoliths of Aseraggodes and Pardachirus are characterized by high-bodied outlines, whereas those of Heteromycteris are horizontally elongated. In Aseraggodes, the dorsal rim is elevated with a gently curved postero-dorsal region, whereas in Pardachirus, the dorsal and posterior rims form a distinct angle. Two morphotaxa within the genus, Pardachirus sp. 1 and Pardachirus sp. 2, are moderately common (particularly the latter taxon) and exhibit morphological features most closely resembling those of extant Pardachirus pavoninus (Lacepède, 1802). Both taxa were previously reported from the Late Miocene shallow-marine deposits of northern Taiwan by Lin and Chien (2022), and the descriptions in that study are followed here.

Two tiny, spindle-shaped otoliths with an elongated and tapering posterior part and a shallow, rather unspectacular sulcus are identified as M. fusiformis, based on the description by Schwarzhans (2019c: pl. 3, fig. 13). This extant species is known to inhabit shallow, sandy bottoms across the western Pacific (Matsuura 1991). However, extant specimens from Taiwan were only recently reported (Wang et al. 2024), suggesting that a larger population of the species may have existed regionally during the Plio–Pleistocene.

Most Pomadasys otoliths in the collection represent juveniles and cannot be reliably distinguished at the species level. Four extant Pomadasys species are known from Taiwan, and their adult otoliths are differentiated by the curvature of the cauda and the presence of a notch at the excisura (Lin et al. 2013). Specifically, a strongly curved, flexed cauda and an excisural notch are diagnostic of P. maculatus, whereas P. argenteus has a more gently curved cauda without a distinct caudal flexion.

Sciaenidae represents the most diverse and second most abundant family in the collection (Table 2). Many taxa were described in detail by Lin and Chien (2022), and their descriptions are not repeated here. The otoliths of Collichthys lucidus (Richardson, 1844) and Larimichthys polyactis (Bleeker, 1877) are morphologically similar, particularly in smaller individuals (Fig. 14A). However, in C. lucidus, the dorsal rim is prominently elevated and anteriorly expanded and frequently exhibits an irregular margin, which contrasts with the smooth, flattened dorsal rim of L. polyactis.

Johnius possesses highly distinctive otoliths that are readily distinguished from those of all other teleosts observed to date (Fig. 14B). However, species-level distinction based on these unique otoliths remains challenging due to marked morphological similarity among congeners (Fig. 15A–F).

Two species of Nibea, N. albiflora (Richardson, 1846) and N. chaoi Lin & Chien, 2022, along with six poorly preserved Nibea otoliths, were identified in the collection (Fig. 14F–H). Otoliths of the extinct N. chaoi were recently described by Lin and Chien (2022) as having a flattened yet anteriorly expanded dorsal rim, which contrasts with the shallow, unexpanded dorsal rim and more elongated outline of extant N. albiflora. More recently, Kocsis et al. (2024) described another fossil species, N. ambugensis Kocsis et al., 2024, from the Late Miocene strata of Brunei. Otoliths previously reported as N. albiflora (Richardson, 1846) by Lin and Chien (2022: fig. 12j–l) from the Late Miocene of northern Taiwan are here reassigned to N. ambugensis.

In Pennahia, the four extant species recorded in Taiwan can be distinguished by a unique combination of otolith morphological characters (Fig. 15G–M): P. macrocephala (Tang, 1937) and P. pawak (Lin, 1940) both have high-bodied otoliths with a deep ventral area and a slight constriction above the ventral rim, whereas in P. anea (Bloch, 1793) and P. argentata (Houttuyn, 1782), the ventral area is flatter and lacks a constriction above the ventral rim. An expanded ventral lobe in the ostium is present in P. argentata and P. pawak, whereas this lobe is only slightly expanded or nearly equal in size to the dorsal lobe in P. anea and P. macrocephala. The cauda is gently curved downward and flexed approximately one-third from its posterior end in P. anea and P. argentata; it is flexed downward only at the very end of the cauda in P. macrocephala; and it is almost straight without conspicuous flexure in P. pawak.

The Sillago otoliths in the collection are assigned to S. japonica Temminck & Schlegel, 1843 and S. sihama (Fabricius, 1775), the two most common extant species in western Taiwan. Compared with other congeners, otoliths of S. japonica are more rounded (Fig. 16H, I). In S. sihama, the postero-dorsal part is distinctly extended and tapers dorsally, which imparts a sharp triangular outline; however, compared with other congeners featuring a posterior extension, its dorsal rim is flat, and the ventral area is short and shallow (Fig. 16K).

A total of 4,893 otoliths were analyzed across the three localities (Tachi-1: 2,298; Tachi-2: 1,654; Sanxia: 941), revealing the presence of 80 taxa (Table 2). Diversity rarefaction and extrapolation curves based on Hill numbers indicate that sampling was sufficient to characterize common and dominant taxa while also showing differences in community evenness among sites (Fig. 17). Total observed taxonomic richness across the pooled dataset reached 80 taxa (⁰D = 80; extrapolated to 93 taxa at doubled sampling effort), with observed richness being highest at Tachi-2 (⁰D = 54), followed by Tachi-1 (⁰D = 53) and Sanxia (⁰D = 48). The continued upward trajectory of ⁰D curves reflects the presence of rare taxa. In contrast, exponential Shannon diversity (¹D, common taxa) and inverse Simpson diversity (²D, dominant taxa) asymptote rapidly (< 650 otolith specimens) for all sites, leveling off globally approximately at ¹D = 22 and ²D = 12. Notably, Tachi-1 exhibits higher common and dominant diversity than Tachi-2 and Sanxia. Although total taxonomic richness (⁰D) is comparable across sites, the lower ¹D and ²D values at Tachi-2 and Sanxia indicate lower community evenness, driven by strong dominance of a few abundant taxa.

Taxonomic compositions of fish assemblages differed significantly among all epoch pairs (PERMANOVA, p < 0.05; Table 3), indicating pronounced temporal restructuring of fish communities from the Miocene to the Pleistocene. Principal coordinates analysis (PCoA) showed separation among samples from the three epochs, resolving the samples into three non-overlapping, epoch-defined clusters along the first two axes, which explain 18.1% and 13.9% of the total variance, respectively (Fig. 18). Taxon vector correlations ≥ 0.5 indicated that this ordination was driven by distinct species associations: positive loadings along PCoA 1 were strongly aligned with the extinct sciaenid †Taosciaena jiangi in the Miocene, whereas negative loadings along PCoA 1 and PCoA 2 were driven by modern shelf taxa in the Pleistocene, including Coilia, Collichthys, Larimichthys, and Pennahia (Fig. 18). Pliocene assemblages were differentiated along the positive direction of PCoA 2, associated with Sillago japonica (Fig. 18). Although PERMANOVA confirmed clear separation of epoch centroids, significant PERMDISP results (Table 3) indicated that multivariate dispersion—specifically, higher compositional variance among Miocene samples relative to tighter Plio–Pleistocene clustering—also contributed to the observed ordination structure.

Partitioning of beta diversity showed that assemblage differences were driven predominantly by taxonomic turnover rather than nestedness. In the Miocene–Pliocene comparison, turnover dissimilarity greatly exceeded nestedness dissimilarity, with a near absence of nestedness (median β_jtu = 0.97, median β_jne = 0.01, Wilcoxon test p < 0.001). Similarly, turnover remained the dominant component in the Pliocene–Pleistocene comparison (median β_jtu = 0.81, median β_jne = 0.10, Wilcoxon test p < 0.001). These results indicate that community succession was primarily driven by taxonomic replacement rather than selective loss or gain of taxa, reflecting substantial compositional replacement over time.

To formally identify diagnostic taxa for each epoch, indicator species analysis revealed distinct taxonomic compositions among epochs, with sciaenids accounting for the majority of indicator taxa (Table 4). The Miocene and Pleistocene were each characterized by multiple significant indicator species. The Miocene assemblage was characterized by three extinct sciaenids: †Nibea chaoi, †Taosciaena hui, and †Taosciaena jiangi, the latter showing the strongest association with this epoch (IndVal = 1.00). The Pliocene was characterized by Bregmaceros nectabanus (Bregmacerotidae; IndVal = 0.51) and Myersina filifer (Gobiidae; IndVal = 0.53), whereas Sillago japonica (Sillaginidae) was identified as a shared indicator for the Pliocene and Pleistocene (IndVal = 0.66). In contrast, the Pleistocene assemblage was characterized by Coilia cf. mystus (Engraulidae) and three extant sciaenids: Collichthys lucidus, Larimichthys polyactis, and Pennahia pawak. Among these, Larimichthys polyactis exhibited the strongest association with the Pleistocene (IndVal = 0.93).

A total of 5,083 fossil and archaeological occurrences representing 19 genera were compiled to reconstruct the Neogene–Quaternary biostratigraphic history of Sciaenidae across the western Pacific (Fig. 19, Suppl. material 2). During the Late Miocene, 13 of the 19 sciaenid genera were present. Among these, four extinct genera were stratigraphically restricted to the Miocene, whereas nine extant genera persisted to the present (Fig. 19A). With the exception of the monotypic Early Miocene genus †Paleosciaena from Japan, all Miocene genera first appeared during the Late Miocene. In contrast, four extant genera, Collichthys, Johnius, Otolithes, and Pennahia, first appeared during the Pliocene, whereas Chrysochir shows the shortest biostratigraphic range, extending from the Late Pleistocene to the present (Fig. 19A). Geographically, these occurrences span localities from Japan to Southeast Asia, indicating a continuous occurrence of diverse sciaenid assemblages along the western Pacific margin since the Late Miocene (Fig. 19B, Suppl. material 2).

The rapid stabilization of rarefaction curves indicates that overall sampling captured representative fish community structures across the studied localities (Gotelli and Colwell 2001; Chao et al. 2014). The Hill number diversity profiles reveal differences in community structure among the studied localities despite comparable total taxonomic richness (⁰D = 48–54). While total observed richness suggests broadly comparable local taxonomic pools, differences in exponential Shannon and inverse Simpson diversity between Tachi-1 and the Tachi-2 and Sanxia sites reflect differences in local paleoecological conditions and depositional environments.

At Tachi-1, the higher community evenness is consistent with a high-energy setting with estuarine to deltaic influence. In such environments, heterogeneous coastal microhabitats may support a more balanced community without dominance by a few opportunistic taxa (Blaber 2000). In particular, the site is characterized by a diverse sciaenid assemblage, supporting its interpretation as a shallow-water coastal lag deposit formed in an estuarine sink or delta-front setting (Zhu 2020; Lin and Chien 2022; Kocsis et al. 2024; Lin et al. 2025a).

In contrast, the lower evenness and high ecological dominance at Tachi-2 and Sanxia are consistent with sedimentary evidence indicating prodelta, shallow inner-shelf lag deposits with greater exposure to open-marine waters. At both sites, the high proportions of juveniles and small demersal fishes, such as gobies (Gobiidae) and flatfishes (Pleuronectiformes), are notable and may reflect local nursery-ground dynamics, as also seen in modern demersal fish assemblages along the shallow open-shelf and estuarine systems of western Taiwan (Chen et al. 2022a, 2022b). However, the co-occurrence of oceanic taxa, particularly rare lanternfishes (Myctophidae) and codlets (Bregmacerotidae), likely reflects oceanographic transport or nearshore intrusion of mesopelagic and epipelagic taxa into shallow shelf environments rather than deep-water deposition.

By documenting these newly collected Plio–Pleistocene assemblages, the present study completes a critical temporal sequence along the Dahan River section, extending the fish record from the Upper Miocene (Lin and Chien 2022) into the Quaternary. This enables a long-term assessment of subtropical coastal fish community dynamics over the past 8 million years. While traditional fossil otolith studies predominantly focus on taxonomic (alpha) diversity and standard ordination, here we apply formal beta diversity partitioning (Baselga 2010) to quantitatively decouple species replacement (turnover) from compositional richness gradients across time.

Ordination (PCoA) and PERMANOVA confirm significant community restructuring across the Dahan River sequence, and beta diversity partitioning helps identify the main component of this compositional change. Across both the Miocene–Pliocene (β_jtu = 0.97; β_jne = 0.01) and Pliocene–Pleistocene (β_jtu = 0.81; β_jne = 0.10) transitions, dissimilarity was driven mainly by species replacement rather than simple species attrition or selective loss. From a macroecological perspective, the low nestedness values indicate that the observed assemblage changes were not primarily caused by nested taxonomic loss, thereby making degradation-driven community contraction or collapse an unlikely primary explanation. Instead, the dominant turnover pattern suggests continued faunal replacement within shallow-marine communities, where shifting shelf environments likely drove biogeographic and ecological reorganization while maintaining broadly comparable community structure (Legendre 2014).

Furthermore, significant differences in multivariate dispersion (PERMDISP) highlight an underlying shift in community variability across time. The wider dispersion in Miocene samples reflects higher compositional patchiness across coastal habitats, whereas tighter Plio–Pleistocene clustering indicates more homogeneous assemblage composition, possibly associated with more comparable or stabilized shelf settings.

We note that when evaluating these temporal patterns, potential facies dependence must be considered. Because the studied horizons along the Dahan River generally represent shallow-water lag deposits formed within comparable inner-shelf bathymetric zones, major bathymetric shifts are unlikely to be the sole driver of community change (Patzkowsky and Holland 2012). This interpretation is supported by clear taxonomic replacement within shared ecological guilds and families across epochs. The most prominent example occurs within Sciaenidae: extinct Miocene shelf taxa, including †Taosciaena, are completely absent from Plio–Pleistocene strata, where they are replaced by modern shelf-dwelling genera (e.g., Johnius and Pennahia). A similar replacement occurs in Sillaginidae, with Sillago japonica emerging as a key Plio–Pleistocene indicator.

This within-family turnover indicates that regional faunal restructuring occurred within the northern Taiwan coastal ecosystem. The drivers behind this replacement were likely mediated by local basin development during the Taiwan Orogeny alongside late Neogene climate transitions, sea-level oscillations, and, perhaps more importantly at the regional scale, the establishment of the modern-style Kuroshio Current during the Pliocene (e.g., Gallagher et al. 2009, 2015; Nagel et al. 2013; Herbert et al. 2016; Dashtgard et al. 2026). Nevertheless, because non-continuous sedimentary sections can conflate local environmental shifts with global signals, these paleoceanographic drivers remain working hypotheses. Importantly, decoupling environmental tracking from regional community turnover will require further sampling across facies-standardized sequences in the western Pacific.

Together, the quantitative community analyses reveal a gradual succession in subtropical teleost assemblages toward increasingly extant-dominated coastal faunas rather than an abrupt extinction/recovery-driven community shift. The Plio–Pleistocene transition therefore represents an important interval of faunal replacement and establishment of modern shelf lineages in Taiwan.

The comprehensive biostratigraphic synthesis of Sciaenidae reveals a stepwise temporal pattern that supports the use of the family as a regional biochronological framework for Neogene–Quaternary shelf environments in the western Pacific (Fig. 19). The regional history of Sciaenidae is characterized by successive faunal phases. The initial Late Miocene radiation introduced a diverse array of shelf specialists, including several endemic lineages, such as †Paleosciaena in Japan, †Taosciaena in Taiwan, and †Advenasciaena and †Bruneisciaena in Brunei, that do not appear to persist into younger regional records. This early phase was followed by successive Plio–Pleistocene appearances of extant genera, during which modern commercial taxa, including Johnius, Otolithes, Pennahia, and Chrysochir, became established in the regional fossil record (Fig. 19A). Because these fossil records are temporally structured and geographically coherent across the East Asian margin to Southeast Asia (Fig. 19B), sciaenid otoliths provide useful biochronological markers for correlating Neogene–Quaternary coastal and inner-shelf facies.

Furthermore, the recognized diversity of extinct Miocene sciaenids in the western Pacific is likely underestimated due to several historical taxonomic assignments. Specifically, the compilation reveals that Miocene occurrences from Japan reported in older literature, such as records of Nibea and Umbrina (Takahashi 1976), Genyonemus (Ohe 1976), or cf. Leiostomus (Ohe 1996), likely represent undescribed, extinct taxa rather than modern or New World generic representatives (Suppl. material 2). These legacy identifications predate modern otolith comparative frameworks and updated regional stratigraphic constraints. Therefore, systematic re-examination of these collections using modern sciaenid otolith systematics will be critical to fully resolving the early diversification phase.

Comparing the empirical fossil record with age-calibrated molecular phylogenies (e.g., Lo et al. 2015, 2017) reveals temporal discrepancies in both directions: molecular estimates appear to underestimate the age of origin of the family Sciaenidae while assigning older divergence ages to several modern Indo-West Pacific lineages than are currently documented by the regional otolith record. Specifically, multigene molecular-clock models (e.g., Lo et al. 2015) propose an origin of Sciaenidae in Tropical America during the Late Oligocene to Early Miocene (~25–30 Ma). However, this young-origin hypothesis is difficult to reconcile with the global Paleogene otolith record. Rich and diverse sciaenid assemblages were already established in shallow-water shelf environments throughout the Eocene (~33–50 Ma; Schwarzhans 1993; Müller 1999; Aguilera et al. 2016; Ebersole et al. 2019; Lin and Nolf 2022), with the earliest record tracing back to the early Eocene U.S. Gulf and Atlantic coastal plains (see details in Lin et al. 2024).

Conversely, for extant western Pacific clades, molecular models place origination and divergence dates deep in the Early Miocene (~18–20 Ma; Lo et al. 2015, 2017). Our empirical biostratigraphic record across the western Pacific indicates that currently recognized extant genera are absent from available Early and Middle Miocene otolith records, which are instead dominated by extinct endemic forms such as †Paleosciaena (Ohe 1976). Extant genera are first documented in the regional record in the Late Miocene, including Argyrosomus, Larimichthys, and Nibea, whereas several dominant commercial genera, including Johnius and Pennahia, have their currently documented first appearances from the Pliocene onward, with Chrysochir making its initial appearance in the Pleistocene. In our view, the 18–20 Ma divergence ages proposed by molecular-clock hypotheses would imply unsampled ghost lineages of approximately 10–15 Myr for multiple extant western Pacific genera.

Nevertheless, the first documented regional occurrences of extant genera need not coincide with their evolutionary origins. Incomplete sampling and taphonomic loss may obscure earlier occurrences, while earlier origination elsewhere followed by dispersal into the western Pacific provides another possible explanation. These alternatives caution against attributing the observed temporal gaps solely to overestimated molecular divergence ages. The comparison therefore highlights the need for further systematic sampling of older shallow-marine deposits and taxonomic re-examination of historical collections in this underexplored region (Lin et al. 2021; Mediodia et al. 2025).