Paleobiological data provide a key historical record of global biodiversity dynamics, but their interpretation is controversial due to geological and sampling biases. Raw data suggest that marine metazoans diversified dramatically during the late Mesozoic and Cenozoic, whereas bias-corrected analyses based on occurrence-level data in the Paleobiology Database (PBDB) have indicated much less Cenozoic diversification. These standardized analyses are cited as evidence that biases strongly conceal underlying patterns in the global fossil record. However, we show that marine diversity did increase substantially and continuously from the Jurassic to the Neogene, even after correcting for biases in PBDB data. Previous standardized analyses did not capture this diversification in full because they were based on incomplete data. In the Cenozoic, observed richness rose to twice the Paleozoic average, which is within the range of values seen in analyses of raw data, suggesting that even the raw global marine fossil record preserves first-order signals of diversity history.
Paleontological data provide a direct, historical record of biodiversity dynamics from which causal processes are inferred (Sepkoski, 1984; Benton and Emerson, 2007; Stanley, 2007; Harmon and Harrison, 2015; Rabosky and Hurlbert, 2015), but geological and paleontological biases may obscure actual diversity history. For example, biodiversity could appear to increase due to increased exposure of sedimentary rocks (Raup, 1972, 1976; Smith, 2001; Peters, 2005; but see Hannisdal and Peters, 2011), increased ease of fossil extraction (Kowalewski et al., 2006; Hendy, 2009; Sessa et al., 2009), or decreased effects of diagenetic processes like dissolution (Cherns and Wright, 2000; Kowalewski et al., 2006).
Despite differences in temporal resolution, taxonomic resolution, and taxonomic coverage, numerous compilations of raw data indicate that the diversity of marine animals rose substantially from the Paleozoic to the Cenozoic (e.g., Raup, 1972; Sepkoski et al., 1981; Benton, 1995; Sepkoski, 1997; Peters et al., 2014; and references therein). For example, Sepkoski’s genus compendium (Sepkoski, 2002) shows Neogene richness exceeding the middle–late Paleozoic average by a factor of 2.1–4.2, depending on the analysis (e.g., Sepkoski, 1997; Bambach et al., 2004) (Fig. 1A), and trends in the alpha diversity of benthic assemblages support an increase of this magnitude (Powell and Kowalewski, 2002; Bush and Bambach, 2004; Kowalewski et al., 2006). However, the observed increase in diversity arguably reflects preservational biases like rock availability (Raup, 1972, 1976; Alroy et al., 2008), and the Paleobiology Database (PBDB, paleobiodb.org) has been compiled in part to permit the removal of the effects of these biases by standardizing data quantity and quality. Recent standardized curves resemble older curves in some respects (Alroy et al., 2008; Foote, 2010; Alroy, 2010a, 2010b, 2014), with diversity increasing at least somewhat during the Cretaceous and/or early Cenozoic (e.g., Fig. 1B).
Despite these similarities, the magnitude and duration of Cenozoic diversification still differ strikingly between raw and standardized curves published to date. In raw curves, diversification extends into the Neogene (Fig. 1A), whereas it stalls or reverses in the Late Cretaceous or Paleogene in the standardized curves (e.g., Fig. 1B), with final diversity barely higher than the Paleozoic maximum (Alroy et al., 2008; Alroy, 2010a, 2010b, 2014). Although Foote (2010), using different methods, showed standardized richness increasing to a greater degree, he felt his Neogene values were compromised by edge effects, leaving Neogene diversity history unaddressed.
These key differences between raw and standardized biodiversity curves lead to critically different interpretations. Did the living marine fauna evolve under conditions of extended (and perhaps exponential) growth (Valentine, 1973; Benton, 1995; Bambach et al., 2004; Stanley, 2007), or was diversification limited by density-dependent processes (Alroy, 2008, 2010a, 2010b, 2014; Foote, 2010)? Moreover, this debate touches on a fundamental question in paleontology: How good is the fossil record? Must it be interpreted through multiple levels of analytical correction, or is the biological signal strong enough to overcome geologic biases (Sepkoski et al., 1981; Benton et al., 2000)?
Here, we show that the bias-standardized diversity of marine animals increased substantially and relatively continuously for the past ∼200 m.y., in contrast with previous standardized analyses, indicating that signals in the global fossil record are more robust than often portrayed.
We generated standardized diversity curves from the PBDB using shareholder quorum subsampling (SQS) at a sampling quorum of 0.60 (Alroy, 2010a, 2010b, 2014; Chao and Jost, 2012). SQS appears to represent an advance over older methods and was used previously in arguments for limited Cenozoic diversification. Specific parameter values and methodological details are provided in the GSA Data Repository1.
We tested the effects of standardizing sampling intensity by comparing standardized and non-standardized diversity curves. We tested the effects of taphonomic biases by comparing curves that include and exclude modes of preservation that are concentrated in specific portions of Phanerozoic time (“temporally restricted modes of preservation”): fossils preserved as original aragonite, preserved in unlithified sediments, and/or sieved from the sediment without extensive processing (soft-part preservation was excluded from all analyses). Including such fossils is believed to permit the measurement of higher diversity in the Cenozoic (Alroy et al., 2008), even after standardizing overall sampling intensity. Unlike previous works, we excluded silicified fossils because they are also temporally restricted (Schubert et al., 1997) and can be extracted with relative ease. We tested the effects of taxonomic scope by comparing curves that included all marine metazoans and those that excluded marine tetrapods, which were excluded in previous analyses of the PBDB (Alroy et al., 2008; Alroy, 2010a, 2014).
Trends in unstandardized, sampled-in-bin diversity in the PBDB generally resemble trends in Sepkoski’s compendium (Sepkoski, 1997; Bambach et al., 2004), with fluctuations during the Paleozoic and early Mesozoic and rising values thereafter (Fig. 2A; Table DR1 in the Data Repository). Excluding temporally restricted preservational modes decreased the magnitude of radiation in the Cenozoic in the raw PBDB data (Fig. 2A). These modes are abundant in the late Cenozoic (Fig. 2B), consistent with published concerns about preservational biases (Alroy et al., 2008).
When sampling intensity was standardized using SQS, with all preservational modes and all marine metazoans included, richness increased throughout the Cenozoic, although less than in the raw data (Fig. 2C, blue line; Table DR2). Surprisingly, removing temporally restricted preservational modes did not greatly change the standardized curve; the Triassic appeared less diverse due to the removal of silicified fossils, and the trajectory of the Cenozoic radiation was altered only slightly (Fig. 2C, black line). However, removing tetrapods had a pronounced effect on the form of the curve (Fig. 2C, red line) because richness was lost mostly from the Neogene, reflecting the diversification of marine mammals at this time (Kelley and Pyenson, 2015; Marx and Fordyce, 2015). Excluding tetrapods creates a data set and a result that are similar to those of Alroy (2010a, 2010b, 2014), who claimed diversification ceased by the Paleogene (e.g., Fig. 1B).
Removing temporally restricted modes of preservation reduced the apparent strength of the Cenozoic radiation in the raw PBDB curve (Fig. 2A), suggesting that preservational heterogeneity does bias unstandardized global diversity curves. About half of the occurrences in the final bin of the Cenozoic belong to these modes (Fig. 2B). Interestingly, simply standardizing overall sampling intensity using SQS reduced the magnitude of the radiation from about four times the Paleozoic average to two times (Fig. 2C). Together, these points indicate that preservational heterogeneity affects raw richness at least in part by controlling the quantity of data, irrespective of effects on data quality. For example, disaggregating sediments and sieving fossils from the Cenozoic can permit the measurement of greater diversity because large sample sizes are easily achieved. However, omitting temporally restricted modes of preservation had limited additional effects on the magnitude and duration of the Cenozoic radiation (Fig. 2C). In other words, varying quality of preservation had little effect on measured diversity once overall sampling intensity was standardized. Thus, standardizing sampling intensity is valuable, but geological and preservational biases did not otherwise have strong effects at this scale.
Preservational heterogeneity can affect comparisons of diversity at the local level (Cherns and Wright, 2000; Kowalewski et al., 2006; Hendy, 2009; Sessa et al., 2009), but its effects at the global scale appear to be largely overwhelmed by preservational redundancy and the amalgamation of samples from many geographic, environmental, and taphonomic regimes (Bush and Bambach, 2004; Kidwell, 2005). Also, removing modes of preservation may introduce unintended biases by altering the paleoenvironmental distribution of samples. For example, siliciclastic sediments lithify more slowly than carbonates on average and are more likely to be unlithified (though less subject to aragonite loss; Foote et al., 2015).
Previous standardized analyses of marine animal diversity excluded marine tetrapods (Foote, 2010; Alroy, 2010a, 2014), and removing them from our analysis damped diversity increase as well. With tetrapods, the Cenozoic radiation extended into the Neogene, and without them, it petered out in the Eocene and total diversification was less (Fig. 2C, red line).
Excluding tetrapods renders the data set paraphyletic and the analysis incomplete if one is interested in total marine metazoan diversity. In justifying their exclusion, Alroy et al. (2008) argued that tetrapods can be larger than other taxa, are sampled differently, are overstudied, and are not numerically important. However, non-tetrapod fossils also vary enormously in size, sampling method, and abundance. Some fish, mollusks, and colonial organisms are also very large, and the crucial parts for genus identification of tetrapods are commonly no larger than many large invertebrate fossils. Also, other taxa (e.g., those of biostratigraphic importance) are overstudied relative to coeval taxa, and these have never elicited similar treatment. Alroy (2010b) suggested that variations in sampling intensity among groups could be controlled by dividing the data, analyzing each group separately with SQS, and summing the resulting diversity curves. When tetrapods and non-tetrapods are treated this way, the result is similar to the diversity curve generated from the combined data set (Fig. 2C, green dashed line), so oversampling of tetrapods is not a problem, at least in the simple sense of data quantity.
We further examined the purported overstudy of tetrapods by comparing the proportional genus richness of several higher taxa in the Neogene in the PBDB and Sepkoski’s compendium with similar data for the modern oceans taken from the World Registry of Marine Species (marinespecies.org; see Data Repository for methods). Proportionally, tetrapods are more diverse in the PBDB than the other data sets (Table DR3), possibly because of careful database entry and vetting (e.g., Uhen, 2015) (cf. Jablonski et al., 2003), but several lines of evidence argue against their ad hoc removal. (1) Given that they have robust hard parts, tetrapods should be overrepresented relative to the living fauna, as are several other taxa in the PBDB (conversely, taxa with low preservation potential are underrepresented; Table DR3). (2) Tetrapods may have been more diverse in the recent geologic past than in the modern oceans (Marx and Fordyce, 2015). (3) Vertebrates as a whole are actually underrepresented in the PBDB relative to the living fauna due to poor representation of fish (Table DR3), which also radiated in the later Phanerozoic (Friedman and Sallan, 2012).
To test the potential effects of the “overstudy” of tetrapods, we removed half their observed richness from the late Cenozoic portion of the curve by averaging the curves produced from the full data set and the non-tetrapod data set (Fig. 2C, dashed red and black line). Halving their richness is somewhat arbitrary, but it means that they are no longer the most overrepresented group in the PBDB data set relative to the living fauna (Table DR3). In this treatment, richness still increases throughout the Cenozoic, although total diversification is slightly less.
Other Potential Biases
Variations in observed richness could reflect variations in the latitudinal distribution of collections, but there are time bins with good latitudinal coverage in the PBDB in both the Paleozoic and Cenozoic (Vilhena and Smith, 2013), so latitudinal coverage is unlikely to drive the long-term trend of increasing diversity. However, Valentine et al. (2012) argued that additional tropical data need to be added to the PBDB for the Cenozoic, especially from the diverse Indo-Pacific; better representation of this fauna in the PBDB would lead to even greater observed diversification.
PREFERRED ESTIMATE OF DIVERSITY HISTORY
Our preferred estimate of metazoan diversity history given current data (Fig. 2D) is standardized for sampling intensity, which is reasonable given the great variation in this parameter (Fig. 2B) (but see Hannisdal and Peters, 2011). Further standardization by excluding some modes of preservation may not be necessary because preservational heterogeneity appears to influence the diversity curve strongly only via its effects on sample size (Figs. 2A–2C), but we did so anyway for comparability with previous analyses. We included marine tetrapods, which is appropriate if the topic of study is the total diversity of marine metazoans. As noted above, reducing the contribution of tetrapods to the total signal by a reasonable amount would slightly damp, but not eliminate, the taxonomic radiation in the Cenozoic (Fig. 2C).
Standardization does not eliminate the basic form of the diversity curve as seen in multiple compilations of raw data produced over the past 150 years: diversity fluctuated from the Ordovician to mid-Mesozoic, then increased fairly steadily thereafter to new heights (Fig. 2D). This radiation corresponds with changes to diversity structure at numerous scales (Powell and Kowalewski, 2002; Bush and Bambach, 2004; Wagner et al., 2006; Holland and Sclafani, 2015). Peters (2005) showed that genus richness in Sepkoski’s compendium generally tracked rock availability during most of the Phanerozoic, but the two records diverged in the Cenozoic, with richness alone increasing. Our standardized results suggest that this increase in diversity is real (at least in part), and an increase in alpha diversity (Powell and Kowalewski, 2002; Bush and Bambach, 2004; Kowalewski et al., 2006) would serve to decouple global richness and rock availability during the later Phanerozoic. However, despite a lack of trend in rock availability (Peters, 2005), sampling of fossils increased greatly in the later Cenozoic (Fig. 1B), indicating that biases other than rock availability influence sampling. Differential ease of fossil extraction likely contributes to the large sample sizes in the Cenozoic that necessitate sampling standardization.
In this analysis, Neogene genus richness rises to 2.05 times the middle–late Paleozoic average (Silurian–Permian). Standardizing sampling intensity does reduce the apparent magnitude of the Mesozoic–Cenozoic radiation compared to raw data (Fig. 2A), but long-term changes in richness are within the range of expectations based on older analyses (e.g., Sepkoski, 1997; Fig. 1A). The curve in Figure 2D is similar to Sepkoski’s (1997) family curve, which should be less influenced by biases than genus-level curves. Some unstandardized genus richness curves do show a greater increase (Bambach et al., 2004; Stanley, 2007), and standardization is valuable in demonstrating that these curves (and an interpretation of exponential growth) may be too extreme, at least given current data (further data collection could alter the observed magnitude of diversity change; Valentine et al., 2012). However, at this scale of analysis, biases of sampling and preservation do not overprint the first-order pattern of long-term diversification in the fossil record (also see Foote, 2010). In fact, the Mesozoic–Cenozoic radiation is the only substantial, permanent increase in diversity since the Ordovician (Fig. 2D).
In our best estimate given current data, the bias-standardized genus richness of marine animals fluctuated during the Paleozoic and early Mesozoic and subsequently increased to new heights. This radiation lasted into the Neogene, elevating diversity to about double the Paleozoic average. This diversity curve is of interest compared to many previous standardized curves because it is based on more complete data (i.e., all marine metazoans). The presence of a clear Cenozoic radiation in the standardized curve validates the general form of older, unstandardized curves, and suggests that the fossil record contains a strong signal of biodiversity that was emergent even in older compilations, despite great heterogeneity in preservation and collection. Although the Cenozoic radiation appears to be a robust feature of marine animal diversity history, the exact magnitude of increase in standardized analyses will certainly change as the data set improves, particularly through more complete tabulations of understudied clades, time intervals, and geographic regions, as well as taxonomic vetting of existing data.
Biodiversity dynamics shifted dramatically during the middle of the Mesozoic: from the Ordovician to the Triassic, biodiversity merely fluctuated, but from the Jurassic to Neogene, there was a sustained taxonomic radiation. This radiation is one of the key features of marine animal diversity history—not since the early Paleozoic had diversity changed substantially and permanently.
For helpful comments, thanks to D. Jablonski, C. Marshall, and A. Knoll. Thanks to M. Kowalewski, S. Peters, and another anonymous reviewer. Special thanks to J. Alroy for providing computer code and advice. Thanks to PBDB data contributors, particularly W. Kiessling, A. Hendy, M. Clapham, A. Miller, M. Foote, J. Alroy, M. Aberhan, M. Kosnik, M. Patzkowsky, and P. Wagner. This is Paleobiology Database publication 238.