R- vs. Q-mode cluster analysis
Cluster analyses of sites (Q-mode) and taxa (R-mode) were performed and compared to identify major clustering trends among Belly River Group microsites (Fig. 2). Two primary site clusters were identified, with an additional grade of sites between them. The largest site cluster (yellow highlighted component in Fig. 2) contains all Oldman Formation sites (N = 23), along with a majority of the pre-LCZ Dinosaur Park Formation sites (N = 14, out of a possible 18), and one Foremost Formation site (‘SPS’). This cluster contains two large sub-clusters, which broadly group sites based on their sampling region (either DPP or MRM). The second primary site cluster (blue highlighted component in Fig. 2) contains the three stratigraphically lowest Foremost Formation sites (‘PHR-1’, ‘PHR-2’, ‘PHRN’) and both Lethbridge Coal Zone sites (‘BB96’, ‘L2377’). The grade of sites (green highlighted component) situated between the two primary clusters contains one Foremost Formation site (‘PK’) and four of the stratigraphically highest pre-LCZ Dinosaur Park Formation sites (‘BB102’, ‘BB119’, ‘BB108’, ‘BB115’). These stratigraphically high pre-LCZ Dinosaur Park Formation sites (along with ‘BB94’, ‘BB75’, ‘BB54’, and ‘BB120’, situated in the yellow highlighted component of Fig. 2) are positioned near the locally-variable conformable boundary between the informal lower ‘sandy’ and upper ‘muddy’ units within the pre-LCZ Dinosaur Park Formation, a transition thought to indicate the acceleration of the transgressive sequence leading into the LCZ and Bearpaw Formation [12, 50, 52]. The two primary site clusters correspond broadly to the clustering of taxa in the R-mode analysis, with the larger site cluster associated (yellow in Fig. 2) most strongly with lissamphibians (e.g. Caudata + Allocaudata), dinosaurs (e.g. Hadrosauridae, Ceratopsidae, Troodon, Dromaeosauridae, cf. Aves), and actinopterygians (e.g. ‘Holostean A’, Coriops, Teleostei, Esocidae), and the smaller primary site cluster (blue in Fig. 2) most strongly associated with batoids (e.g. Myledaphus + Pseudomyledaphus, Ischyrhiza, Protoplatyrhina, Rhinobatos), sharks (e.g. Hybodus, Archaeolamna, Odontaspidae), and actinopterygians (e.g. Belonostomus, Enchodus, Phyllodontidae). The sub-clusters within the larger (yellow in Fig. 2) of the two primary clusters are broadly similar in taxonomic composition, with the only major difference being the lack of several taxa (e.g. Paronychodon, Richardoestesia, cf. Aves) in Dinosaur Provincial Park sites. The grade of sites (green in Fig. 2) not included in the two primary clusters is associated with aquatic taxa present to varying degrees in both other clusters (e.g. Myledaphus + Pseudomyledaphus, Lepisosteus, Eusuchia, Baenidae, Champsosaurus).
Redundancy analysis
Redundancy analysis (RDA) was carried out on the percent difference dissimilarity matrix computed from the Belly River Group microsite relative abundance dataset, with stratigraphic interval (a proxy for temporal change), depositional setting (site-specific sedimentological characteristics), palaeogeographic sampling location (DPP or MRM), and palaeoenvironment (as reconstructed for the broader area or interval within the geological formation) as explanatory factors (Fig. 3). Broad overlap exists in sites preserved as crevasse splays or in-channel deposits, with these two representing the depositional setting of the vast majority of sites (Fig. 3a, red and green polygons), though sites preserved as shoreface deposits did cluster separately from other sites (Fig. 3a, blue polygon). Palaeogeographic sampling location was effective at separating site clusters in certain situations, such as for sites in the time-equivalent portion of the upper Oldman and pre-LCZ Dinosaur Park formations (Fig. 3b, dark red and blue polygons). When expanded to all sites, sampling location did not produce distinct clusters, and broad overlap was found relating to the position of lower Belly River Group sites from Milk River and upper Belly River Group sites from DPP (Fig. 3b, light red and blue polygons). When the stratigraphic interval of each site was analyzed as a clustering variable, considerable overlap was found and no directional organization could be found that would be consistent with a linear relationship through time (Fig. 3c). Sites from the lower and middle (Comrey sandstone) Oldman Formation (Fig. 3c, purple and yellow polygons) plotted adjacent to one another, with a broad overlapping distribution of sites from the upper Oldman and pre-LCZ Dinosaur Park formations (Fig. 3c, green polygon). Most Foremost Formation sites (Fig. 3c, blue polygon) plotted between pre-LCZ Dinosaur Park Formation sites from high in stratigraphic section (Fig. 3c, sites of green polygon with more negative positions on first RDA axis), near the boundary with the LCZ, and sites from within the Lethbridge Coal Zone itself (Fig. 3c, light blue polygon). The exception to this was the ‘SPS’ site, which plotted most closely to lower Oldman Formation sites. When using palaeoenvironment as a factor, sites were assigned to one of four settings, based on their lithology and predominant fauna: (1) marine, (2) transitional, (3) terrestrial (paralic; lower coastal plain), and (4) terrestrial (alluvial; upper coastal plain). The two terrestrial groupings correspond to the palaeoenvironmental conditions from which the vast majority of dinosaur fossils are known, and represent the two primary environmental regimes discussed in previous studies of dinosaur environmental sensitivity and/or provinciality/endemism [1–7, 9, 19, 20, 22–25, 36, 37, 42–45, 49, 70]. Three non-overlapping grouping were obtained: one for sites with palaeoenvironments reconstructed as marine (Fig. 3d, blue polygon), one for sites reconstructed as transitional and preserving a mix of marine and terrestrial sedimentological features and taxa (Fig. 3d, purple polygon), and a final grouping for sites reconstructed as being terrestrial (Fig. 3d, green and yellow polygons). Sites with terrestrial palaeoenvironments were further subdivided based on their prior associations with more paralic, lower coastal plains (Fig. 3d, green polygon) or more alluvial, upper coastal plains (Fig. 3d, yellow polygon). These further subdivisions followed a clustering pattern consistent with other sites, with more inland terrestrial sites plotting further from marine and transitional sites than more coastal plain terrestrial sites. Despite this trend, considerable overlap exists between more coastally influenced terrestrial sites and more alluvial terrestrial sites, indicating that the two cannot be considered truly distinct for the purpose of site clustering.
Pair-wise site assemblage similarity
Pair-wise Bray-Curtis (percentage-difference) similarity was computed for each consecutive pair of stratigraphically-ordered neighbouring sites from each sampling region (DPP and MRM), along with a visual representation of relative taxonomic group abundance at each site (Fig. 4). The Milk River/Manyberries sites (Fig. 4, red box at right) range stratigraphically from the Foremost Formation to the upper unit of the Oldman Formation (time-equivalent to the pre-LCZ Dinosaur Park Formation in DPP), and the Dinosaur Provincial Park sites (Fig. 4, blue box at left) stratigraphically range from the middle Oldman Formation (‘Comrey sandstone’) to the Lethbridge Coal Zone. Pair-wise similarity curves are relatively stable for much of the sampled intervals in both DPP (Fig. 4, blue curve) and MRM (Fig. 4, red curve), ranging from approximately 50–80% similarity. Two exceptions to this stability exist, one during the regressive phase recorded near the end of the Foremost Formation in MRM sites (Fig. 4, near base of red curve) and the other during the transgressive phase in the Lethbridge Coal Zone near the boundary between the Belly River Group and the overlying Bearpaw Formation in DPP sites (Fig. 4, near top of blue curve). In both of these cases, site similarity dropped to approximately 10–20%. Trends recorded across relative abundances of taxa in individual sites (Fig. 4, DPP sites in blue box at left and MRM sites in red box at right) and in formational average taxon abundance (Fig. 4, top right) both show that the site similarity drop in the Foremost Formation was associated with large reductions in relative abundances of chondrichthyans and large increases in relative abundances of lissamphibians, with the inverse seen in the similarity drop in the Lethbridge Coal Zone.
Sub-sample analyses of time-equivalent sites
Using three subsets of the broader relative abundance dataset, with re-sampled values of dinosaurs from Brinkman et al. [8] in place of Brinkman [12] for the first two, R- vs. Q-mode cluster analyses and pair-wise assemblage similarity analyses were performed for dinosaur-only (Fig. 5), theropod-only (Fig. 6), and non-dinosaur (Fig. 7) components of the assemblage. These subset comparisons were made only for sites in the overlapping stratigraphic intervals of each sampling area, namely the middle (‘Comrey sandstone’) Oldman Formation, and the upper Oldman Formation and pre-LCZ Dinosaur Park Formation.
In the dinosaur sub-sample (Fig. 5), hadrosaurs constituted the vast majority of assemblage relative abundance in almost all sites from both Dinosaur Provincial Park and Milk River. The only exception to this was the DPP ‘BB54’ site, which clustered away from all other sites (Fig. 5A). With the exception of hadrosaurs, no single dinosaur taxon was found to be driving large-scale site clustering, though increased Troodon relative abundance seems to drive the finer-scale clustering of two MRM sites (‘RDS’ and ‘CBC’) and a DPP site (‘BB71’), and relative abundance of ankylosaurs drives the clustering of two DPP sites (‘BB86’ and ‘BB100’) and one MRM site (‘PLS’). Unlike in the cluster analyses of the broader vertebrate assemblages, there is less distinct clustering of sites by sampling region (Fig. 5a, MRM sites indicated by red circles and DPP sites indicated by blue triangles). Pair-wise site similarity for dinosaur assemblages in the sub-sampled interval is relatively stable for both DPP and MRM, ranging approximately 40–80% similarity (Fig. 5b, DPP sites in blue box and represented by blue similarity curve and MRM sites in red box and represented by red similarity curve). The only deviation from this trend is a drop to approximately 20% similarity for sites neighbouring ‘BB54’, which as noted above represents an apparent outlier due to a lower relative abundance of hadrosaurs, and much higher relative abundances of ceratopsians and Richardoestesia (Fig. 5b). Formational average abundances of dinosaurs (Fig. 5c) do not show any major shifts in assemblage between the middle Oldman Formation sites and the sites of the time-equivalent upper Oldman and pre-LCZ Dinosaur Park formations, nor is there considerable difference in assemblages between the DPP and Milk River localities. The only exceptions to this are a moderate increase in ceratopsians in DPP between the middle Oldman Formation (~1% relative abundance) and time-equivalent pre-LCZ Dinosaur Park Formation (~7% relative abundance), a shift also found in the equivalent intervals of the Milk River sites (~4 to ~7% relative abundance, respectively). In the middle Oldman Formation and time-equivalent upper Oldman Formation of Milk River, there were also changes in small theropod relative abundances, with saurornitholestines increasing slightly (~1 to ~5%), and Troodon going from absent to present (with relative abundance in the upper Oldman Formation of ~1%).
The theropod-only sub-sample analyses (Fig. 6) produced similar results to the sub-sample of dinosaurs. Most sites did not show any strong signal from a particular taxon driving clustering patterns (Fig. 6a), though a few clustered together due to their greater association with tyrannosaurids and Richardoestesia (‘BB120’, ‘BB115’, ‘BB75’, BB119) or with cf. Aves (‘BB61’, ‘CN-2’, ‘ORS’, ‘HS’). As with dinosaurs, the site similarity curves of the theropod sub-samples from DPP and MRM are very similar (Fig. 6b), both staying within a range of ~30 to ~80% similarity. The lower bound of that similarity range related to sites neighbouring those with very little theropod material (e.g. ‘BB120’, ‘BB75’, ‘BB115’, and ‘BB119’). The formational average theropod relative abundances (Fig. 6c) in DPP show no appreciable differences, with slight increases in proportions of tyrannosaurids and Richardoestesia, and slight decrease in Troodon. In MRM, there are more considerable differences in formational average relative theropod abundances, with saurornitholestine proportions greatly increasing, Troodon appearing in the upper Oldman Formation while not being found in the middle Oldman Formation, and all other taxa proportionally decreasing slightly.
The non-dinosaur sub-sample analysis (Fig. 7) produced similar results to the non-marine components of the R- vs. Q-mode analysis of all microsite data (Fig. 2, yellow square). Clustering of sites based on their provenience was apparent, though sites did not cluster exclusively based on being from DPP or MRM (Fig. 7a). Sites from MRM clustered more closely to the majority of DPP sites than a number of DPP sites (e.g. ‘BB106’, ‘BB117’, ‘BB121’, ‘BB94’, ‘BB75’, ‘BB119’, ‘BB102’), with those latter sites forming a cluster more similar to each other than to any other site. This cluster was associated the proportion of particular actinopterygians (e.g. Lepisosteus, Paratarpon, Acipenseriformes, Belonostomus), and turtles (e.g. Basilemys, Baenidae, Trionychidae). Two sites, both stratigraphically high in the DPF and close to the LCZ, were associated with batoids (e.g. Myledaphus + Pseudomyledaphus), Basilemys, and in one case sharks (Hybodus, in ‘BB115’). As in other analyses, ‘BB54’ grouped as something of an outlier, and was here associated strongly with ‘Holostean A’ and ‘Holostean B’. Other DPP sites were broadly associated with many taxa, though in particular with actinopterygians (e.g. Coriops, ‘Holostean A’), baenid turtles, and lissamphibians (e.g. Caudata + Allocaudata). Sites sampled from MRM as a whole were distinguished from DPP mainly due to an even stronger association with lissamphibians (e.g. Caudata + Allocaudata) and certain actinopterygians (e.g. esocids, teleosts), though particular MRM sites were also distinguished based on their association with taxa that were either absent or in low abundance at other sites, particularly those from DPP. For example, Adocus and Chiloscyllium are strongly associated with the ‘CS’ site, though absent or in very low abundance in most sites. One MRM site (‘BMC’) clustered as an outlier, and was distinguished through a suite of taxa (e.g. chelydrid turtles, mammals, anurans, squamates). Both of these outlier sites (‘BB54’ and ‘BMC’) have been noted in previous research to be sedimentologically distinct from other sites, possibly representing an exception to the general trend of depositional setting having a relatively small effect on microsite assemblage structure [14, 52]. Site similarity curves for DPP and MRM (Fig. 7b) were very similar, and very stable, both fluctuating around 60% similarity for much of the sampled interval. The only prominent exception to this came at the top of the time-equivalent interval in DPP, where similarity began to steadily drop, reaching approximately 40% similarity by the top of the sampled interval. Overall proportions of major taxonomic groups in DPP and MRM during this interval (Fig. 7c) were similar, with the notable exception being the higher proportion of batoids in DPP (<20%) when compared to MRM (<5%).