Origination, extinction, and mass depletions of marine diversity

—In post-Cambrian time, five events—the end-Ordovician, end-Frasnian in the Late Devonian, end-Permian, end-Triassic, and end-Cretaceous—are commonly grouped as the ‘‘big five’’ global intervals of mass extinction. Plotted by magnitude, extinction intensities for all Phanerozoic substages show a continuous distribution, with the five traditionally recognized mass extinctions located in the upper tail. Plotted by time, however, proportional extinctions clearly divide the Phanerozoic Eon into six stratigraphically coherent intervals of alternating high and low extinction intensity. These stratigraphic neighborhoods provide a temporal context for evaluating the intensity of extinction during the ‘‘big five’’ events. Compared with other stages and substages in the same neighborhood, only the end-Ordovician, end-Permian, and end-Cretaceous extinction intensities appear as outliers. Moreover, when origination and extinction are considered together, only these three of the ‘‘big five’’ events appear to have been generated exclusively by elevated extinction. Low origination contributed more than high extinction to the marked loss of diversity in the late Frasnian and at the end of the Triassic. Therefore, whereas the ‘‘big five’’ events are clearly times when diversity suffered mass depletion, only those at the end of the Ordovician, Permian, and Cretaceous periods unequivocally qualify as globally distinct mass extinctions. Each of the three has a unique pattern of extinction, and the diversity dynamics of these events differ, as well, from the other two major diversity depletions. As mass depletions of diversity have no common effect, common causation seems unlikely. Richard K. Bambach and Andrew H. Knoll. Botanical Museum, Harvard University, 26 Oxford Street, Cambridge, Massachusetts 02138. E-mail: [email protected] Steve C. Wang. Department of Mathematics and Statistics, Swarthmore College, 500 College Avenue, Swarthmore, Pennsylvania 19081. E-mail: [email protected] Accepted: 15 February 2004 Introduction: Mass versus ‘‘Background’’ Extinction The idea that mass extinctions stand out as a class of events separate from the range of ‘‘normal’’ or ‘‘background’’ extinctions that characterize most of the geological record originated with the work of Norman Newell (1962, 1963, 1967) and crystallized through the quantitative analysis by Raup and Sepkoski (1982). Using Sepkoski’s compilation of stratigraphic ranges temporally resolved to the stage level for marine families of all animal taxa (Sepkoski 1982), Raup and Sepkoski identified five mass extinctions: the end-Ordovician (Ashgillian); Late Devonian (including the Frasnian/Famennian boundary); endPermian (Guadalupian and Djhulfian together); end-Triassic (Late Norian or Rhaetian); and end-Cretaceous (Maastrichtian). These have become known informally as ‘‘the big * This paper is dedicated to the memory of Stephen Jay Gould (1941–2002). five’’ mass extinctions. Each had been noted earlier by Newell (1967). Raup and Sepkoski plotted the extinction rate (number of families going extinct per million years) for each stage in time order and assumed that apparent outliers on the arithmetic plot were true statistical outliers (and not aberrations produced by errors in the timescale), thus identifying a distinct class of mass extinctions. They also calculated a 95% confidence interval around the remaining ‘‘background’’ extinction stages and demonstrated a secular decrease in extinction rates over time. Quinn (1983), however, showed that the extinction data were highly skewed and correctly pointed out that the assumption of a normal distribution for ‘‘background’’ extinction data was not valid. When log-transformed, the distribution of extinction rates was approximately normal and did not have a suite of outliers that could be categorized unambiguously as a class of mass extinctions separate from the overall ‘‘background’’ distribution. Bambach and Gilinsky (1986) dem523 ORIGINATION, EXTINCTION, AND DIVERSITY DEPELETIONS onstrated that this was also the case for several other extinction metrics, including proportional metrics by interval that are not subject to time-scale error. For all metrics the distribution of extinction intensity grades smoothly from lowest to highest values with no discrete break between ‘‘background’’ and presumed mass extinction intervals. Indeed, Raup (1991) developed his view of the Phanerozoic kill curve on the basis of this continuity of extinction magnitudes. Bambach and Gilinsky (1986) did, however, support the finding that extinction intensities (and origination intensities, as well) declined during the Phanerozoic, a conclusion discussed more fully by Gilinsky (1994). Despite the evidence that intensities of extinction form a continuum, the five events specified by Raup and Sepkoski (1982) continue to be labeled ‘‘mass extinctions’’ (Finney et al. 1999; McGhee 2001; Wignall and Twitchett 1996; Pálfy et al. 2000; MacLeod and Keller 1996). Current threats to biodiversity have even been labeled ‘‘the sixth extinction’’ (Leakey and Lewin 1995). It may be that magnitude alone is sufficient to justify the term ‘‘mass extinction’’ because events with such pronounced loss of diversity are rare, with waiting times of about 100 million years (Raup 1991). But is there any reason to think of these events as a separate class of events, rather than as the uncommon upper tail of a continuous distribution? Wang (2003) recently identified three concepts that must be considered individually when we ask whether putative mass extinctions grade continuously into the range of background extinction: continuity of cause, continuity of effect, and continuity of magnitude. Continuity of cause would be demonstrated if candidate mass extinctions could be shown to be driven by the same processes that are responsible for background extinction, albeit operating at increased intensity or over larger areas. Continuity of effect would be established if background and mass extinctions exhibited common patterns of selectivity on taxonomic, functional, ecological, or other grounds. And continuity of magnitude would exist if the distribution of intensities of mass extinctions graded smoothly and continuously into intensities of background extinction. Although continuity of magnitude appears to have been demonstrated by Quinn (1983) and Bambach and Gilinsky (1986), and was assumed by Raup (1991) in his kill curve analysis for the Phanerozoic, several factors led us to reconsider the status of these intervals. These include (a) the rarity of the largest extinction intensities, (b) the possibility that they do not share continuity of cause or effect with all other intervals, and (c) the fact that origination and extinction have rarely been considered together in examining patterns of diversity change. In the following sections, we reevaluate the continuity or discontinuity of magnitude and then briefly consider whether events that might be regarded as mass extinctions can be unified by effect or cause. We test whether any of the ‘‘big five’’ events are differentiable from the distribution of other extinction intensities, explore the role of origination as well as extinction in diversity changes associated with these five intervals, and comment on the differences, as well as similarities, among the intervals. The upshot will be that although the five intervals in question—the end-Ordovician, Late Devonian, end-Permian, end-Triassic, and end-Cretaceous—are the five intervals with the greatest diversity loss in the Phanerozoic, they share little else in common. Only three were driven predominantly by extinction and even they display distinctly different patterns of diversity change, implying that these events are not related by continuity of effect or cause. Tracking Genus Diversity Figure 1 illustrates the history of marine genus diversity through Phanerozoic time. The data were compiled by using a computerized sorting routine written by Jack Sepkoski and modified by J. Bret Bennington to tabulate genus diversity for 107 stages and substages using Jack Sepkoski’s unpublished tabulation of the stratigraphic ranges of genera as of 1996. Although the Paleontological Research Institution has recently published the raw genus ranges from a later version (1998) of Sepkoski’s compilation (Sepkoski 2002), this publi524 RICHARD K. BAMBACH ET AL. FIGURE 1. Diversity and diversity turnover of marine genera by interval through the Phanerozoic. The five major post-Cambrian diversity depletions are highlighted. The heavy line connects the data on number of genera crossing each interval boundary. The line directly connecting the numbers of boundary-crossing genera follows the path of minimum likely standing diversity, regarded as the minimum diversity because origination and extinction would have to work in exact lock-step to follow that diversity path. The peaked dotted line represents genus turnover within each interval. The rising part of each peak represents all genus originations (first occurrences) reported from the interval. The peak records the total number of genera reported from the time interval. The descending part of the peak represents the number of extinctions (last records) of genera in the interval. The magnitudes of the peaks compared with the minimum standing diversity at interval boundaries represent the degree of faunal turnover in the intervals. cation simply lists all the genera and does not numerically tabulate the data on diversity. Genus diversity follows the general pattern established in the ‘‘consensus paper’’ of Sepkoski et al. (1981) and best known from Sepkoski’s (1981: Fig. 5) widely reproduced family diversity curve. In the Paleozoic we see the ‘‘Cambrian Explosion’’ (the increase in diversity in the Early Cambrian), a Middle and Late Cambrian ‘‘plateau’’ of diversity, and the Ordovician Radiation, followed by the long interval of fluctuating, but non-trending, diversity that began in the Caradocian and lasted for the rest of the era. Diversity changes during this ‘‘Paleozoic Plateau’’ include three of the ‘‘big five’’ diversity depletions, the endOrdovician, the Late Devonian, and the endPermian events. The post-Paleozoic is characterized by nearly continuous diversity increase, interrupted by the other two ‘‘big five’’ diversity depletions, the end-Triassic and the sharp, era-bounding end-Cretaceous events. Note that because we tabulated subgenera of mollusks, resolved the data to the substage level, and emphasized the diversity at interval boundaries rather than the total diversity within each interval, the apparent Cenozoic increase in diversity is proportionately greater than that illustrated by Sepkoski for families. There are some concerns that the apparently large Cenozoic increase in diversity may be partly artifactual. However, recent analyses (Bush and Bambach in press on alpha diversity; Jablonski et al. 2003 on Pull of the Recent; and Bush et al. 2001, in press on techniques of sample-standardization) demonstrate that an increase in Cenozoic diversity is strongly indicated, although the exact amount is still unclear (Jackson and Johnson 2001). The ‘‘consensus paper’’ (Sepkoski et al. 1981) was put together because several of the data sets avoided or mitigated some of the biases, such as imperfections of the geologic record, that were of concern then and still are (Peters and Foote 2001). Although every potential problem should be analyzed and improvement in the data is necessary, it still appears, as was concluded then, that the diversity signal is stronger than the noise. Figure 1 accounts for all the data in the Sepkoski genus compilation (see caption for full explanation). Many diversity curves use total 525 ORIGINATION, EXTINCTION, AND DIVERSITY DEPELETIONS diversity as the recorded data (this is true for published illustrations of Sepkoski’s family curve, for example). If one mentally ‘‘connects the dots’’ of total diversity peaks in Figure 1, it is clear that the general shape of a curve connecting total diversities would be very similar to the boundary-crossing diversity emphasized here. Mid-Devonian diversity would appear higher than Late Ordovician diversity, Carboniferous diversity would fluctuate more, and the upturn in diversity in the mid-Cretaceous would be sharper than shown by the standing diversity plot. The general pattern, however, would be the same. We use the boundary-crossing standing diversity as the preferred representation of diversity and diversity change through time. We know that boundary-crossing diversity is not anomalous compared with total diversity because the trend of boundary-crossing diversity follows the general path of total diversity. Two factors make us prefer it to total diversity. First, it is the only measure we have of actual standing diversity. Total diversity in an interval was not the actual standing diversity at any time because it is unlikely that all originations occurred in an interval before any extinction. However, the recorded diversity at the boundaries of each interval, calculated by subtracting all extinctions in the interval from the total diversity in the interval, is a direct measure of standing diversity at interval boundaries. The other compelling reason is that change in diversity is shown best by comparing diversity at the start of different intervals. Diversity is a function of both origination and extinction. The peaks of turnover within each interval in Figure 1 reveal how much variation of diversity can occur in any interval, but comparing standing diversities at interval boundaries tells us whether origination and extinction are in balance (little or no change of diversity from one boundary to the next) or whether either origination or extinction dominated during an interval (origination dominating if boundary-crossing diversity increases, extinction being more important if boundary-crossing diversity decreases). In plots of total diversity, a predominance of extinction over origination in one interval could be masked by an increase in origination in the next. Total diversity might appear unchanged between the two intervals because origination in a rapid recovery from an extinction event could make total diversity in the succeeding interval equal to that in the previous one, concealing the low diversity at the start of the interval. For example, although extinction exceeded origination in each of the last two intervals of the Silurian and diversity appears to have been lost between the Ludlovian and Pridolian when looking at total diversity, the low point of diversity at the end of the Silurian (end-Pridolian) is masked in the total diversity curve because, in the Gedinnian, the first interval of the Devonian, origination was high, causing total diversity to exceed that of the Pridolian. Boundary-crossing diversity not only approximates standing diversity, but it also gives a clearer representation of the consequences of within-interval evolutionary dynamics than that provided by summed total diversity data for whole intervals. The ‘‘Big Five’’ as Diversity Depletions Inspection of Figure 1 reveals that, although diversity decreased slightly on several other occasions, there are only five post-Cambrian intervals when diversity decreased markedly: (1) at the end of the Ordovician, (2) during the Middle and Late Devonian, (3) during the Late Permian, (4) at the end of the Triassic, and (5) at the end of the Cretaceous. The end-Triassic decrease does not look as large as the other four, but standing diversity throughout the Triassic was lower than at any other time after the mid-Ordovician, so the proportional decrease in the latest Triassic is quite comparable to the other four major diversity depletions. Two Cambrian intervals are also noted on Figure 1 (the late Botomian [LB] and early late Middle Cambrian [ELM]). These were times of relatively small numerical change in diversity but high proportional diversity loss. Because the evolutionary dynamics of the Cambrian and Early Ordovician are unusual we will consider them separately; the bulk of this paper emphasizes the post-Arenig portion of the Phanerozoic. Because it is hard to judge the proportional magnitude of diversity change from a plot of 526 RICHARD K. BAMBACH ET AL. FIGURE 2. Proportion of gain or loss of genus diversity from the Caradoc to the Plio-Pleistocene. The five major diversity depletions (decrease greater than 20%) are numbered. Symmetrical lines are drawn at 213.5% and 113.5% (based on the sixth largest diversity decrease) to indicate the range that might be regarded as ‘‘background’’ fluctuation in diversity. Intervals with greater than 13.5% diversity increase are common only after major diversity depletions. absolute values, a plot of proportion of gain or loss of diversity, rather than a plot of the numbers of genera as such, is desirable. A display of numbers of taxa, as in Figure 1, is useful for illustrating the pattern of change in diversity, but the numbers represent only those taxa discovered in the fossil record and are not a complete record of all the taxa that existed. Proportional diversity change is what we seek to understand here—how episodes of diversity loss affected the whole biota. The question, in effect, concerns the importance of an event in the context of its time, not just how many taxa were involved. Figure 2 shows the proportional gain or loss in the number of genera during each stage or substage interval, starting with the Early Caradocian in the Middle Ordovician after the large proportional increases in diversity associated with the Cambrian Explosion and the Ordovician Radiation were over. The change in diversity is calculated as a proportional change by subtracting the number of genera at the start of each interval (the standing diversity at the boundary between the interval and its preceding interval) from the number of genera at the end of the interval (the standing diversity at the boundary between the interval and its succeeding interval) and dividing that number (the change in the number of genera from the start to the end of the interval) by the number of genera at the start of the interval. For example, if 500 genera pass from interval one to interval two and 600 genera pass from interval two to interval three, then there were 100 more originations than extinctions during interval two, with a gain in diversity of 100 genera, a proportional increase of 10.200. Likewise, a decrease from 600 to 500 genera during an interval (a loss of 100 genera as a result of 100 more extinctions than originations) would be a proportional decrease of 20.167. As noted above, an advantage of determining standing diversity at interval boundaries is that one can follow the balance of origination and extinction as it influences diversity change, something not possible when only tabulating total diversity for each interval. Figure 2 shows that the five intervals already well known as the classic ‘‘big five’’ mass extinctions (with the Guadalupian as well as the end-Permian Djhulfian included in the major later Permian diversity decrease [Stanley and Yang 1994]) are the only postLlandeilian intervals with more than a 20% proportional loss of genus diversity. In fact, there is a gap of 8% between the largest loss 527 ORIGINATION, EXTINCTION, AND DIVERSITY DEPELETIONS of diversity not included in the ‘‘big five’’ and the smallest loss within the ‘‘big five,’’ but no gaps of as much as 2% occur between any smaller values when arranged in rank order. These five intervals were certainly times of mass depletion of diversity, but can we justify regarding any of them as times of mass extinction different in cause, effect, or magnitude from ‘‘background’’ extinction? Large-Magnitude Proportional Increases in Diversity As a side issue, but one related to proportional diversity change and its timing, it is interesting to note that the only times when proportional increase of diversity exceeds 13.5% are during the Cambrian Explosion, the Ordovician Radiation (just ending in the early Caradocian at the start of Fig. 2), in the immediate aftermath of each of the five ‘‘mass depletions’’ of diversity, and briefly (single intervals only) in the Late Cretaceous (Turonian) and Neogene (early Miocene) (Fig. 2). Although origination rates are not unusually high in these intervals (no outliers for origination are found in an analysis of origination proportions at these times), the combination of higher origination and lower extinction during the ‘‘recovery’’ phase after diversity depletion does mark these intervals as times of unusually great proportional increase in diversity. These are not necessarily times of broad transgression or otherwise better representation of the marine record, so higher origination rates in the wake of major diversity depletions may reflect recovery from unusually low diversity and not just the effect of improved record availability, a possibility raised by Peters and Foote (2001). Testing for Continuity of Magnitude of Extinction We tested for continuity or discontinuity of magnitude of extinction (i.e., whether or not the distribution of intensities of apparent mass extinctions grade smoothly and continuously into intensities of background extinction) in two ways. (1) We tested whether there is a smooth continuous distribution of magnitudes of extinction intensity with no strong variation in the upper tail of the distribution, first for the whole Phanerozoic and second for the time after the after the ‘‘Cambrian Plateau’’ of low diversity and high turnover. (2) We compared magnitudes of extinction for each interval against the distribution of magnitudes of extinction in the particular segment of the timescale, based on average high or low extinction rates, to which the interval belongs. Only if an interval satisfies both criteria, that is, if the interval is not part of a continuous smooth distribution of magnitudes, and if the interval also appears as an ‘‘outlier’’ in magnitude compared with the other intervals in its particular segment of the timescale, do we regard it as a ‘‘true’’ global mass extinction, different in magnitude from the bulk of associated stratigraphic intervals. How Continuous Are the Values of Extinction Intensity? As noted above, several analyses have concluded that the distribution of extinction intensities is apparently continuous (Quinn 1983, Bambach and Gilinsky 1986, Raup 1991, Wang 2003). Figure 3A shows this effect for proportion of genus extinction arranged in rank order for the 107 stages and substages of the Phanerozoic as tabulated in our version of Sepkoski’s genus database. Cambrian and Early Ordovician extinction proportions, however, were consistently high (Table 1, Fig. 4). Sixteen of 19 Cambrian and Early Ordovician intervals have extinction intensities that fall within the top quartile of all Phanerozoic intervals (Fig. 3A). Origination was unusually high, as well, during this interval. Thus, whereas taxonomic turnover at the genus level was great, overall diversity did not fluctuate wildly (Fig. 1 and discussion below). These turnover rates are not like those of much of the later Phanerozoic. For example, the decrease in proportion of extinction observed in the Early Ordovician (Fig. 4A) was not produced by the radiation of taxa with lower extinction rates diluting continuing high extinction rates in the trilobites, which dominated Cambrian diversity. Instead, extinction proportions for trilobites, which had consistently exceeded those of the non-trilobite fauna from the origin of the clade in the early Atdabanian though the early Arenigian (Foote 1988), dropped to the same level as non-trilobite proportions of extinction during 528 RICHARD K. BAMBACH ET AL. FIGURE 3. Proportions of genus extinction arranged in rank order by magnitude. A, All 107 intervals of the Phanerozoic. Magnitudes from the Cambrian and Early Ordovician are highlighted. B, Middle Ordovician to Plio-Pleistocene values only. Higher magnitude intervals are labeled. the Late Arenigian and remained at comparable levels thereafter (Fig. 4C). Clearly something changed for trilobite evolutionary dynamics in the later part of the Early Ordovician. Although the change was not as dramatic for the non-trilobite fauna, extinction proportions for that fraction of the fauna, which had been over 30% in two-thirds of the intervals of the Cambrian and Early Ordovician, dropped to levels below 30% and remained low until the Middle Silurian, except for the end-Ordovician late Ashgillian extinction event. Proportions of extinction never returned consistently to the high levels common in the Cambrian and Early Ordovician, even when proportions of extinction increased for extended intervals, such as from the Middle Silurian to the mid-Carboniferous—this held for trilobites and non-trilobites alike. We do not yet understand why turnover rates should have been so high during the Cambrian and Early Ordovician. Perhaps early animals were more vulnerable to extinction for functional reasons—many Cambrian animals belonged to stem rather than crown groups of bilaterian phyla and classes (Budd and Jensen 2000). Perhaps the low diversity of Cambrian and Early Ordovician animals con529 ORIGINATION, EXTINCTION, AND DIVERSITY DEPELETIONS T A B L E 1. D at a on p ro p or ti on s of or ig in at io n an d ex ti n ct io n fo r m aj or in te rv al s of th e P h an er oz oi c b as ed on g en er al m ag n it u d e of ex ti n ct io n . M aj or in te rv al N u m b er of