Fossil Plant Relative Abundances Indicate Sudden Loss of Late Triassic Biodiversity in East Greenland

Science  19 Jun 2009:
Vol. 324, Issue 5934, pp. 1554-1556
DOI: 10.1126/science.1171706


The pace of Late Triassic (LT) biodiversity loss is uncertain, yet it could help to decipher causal mechanisms of mass extinction. We investigated relative abundance distributions (RADs) of six LT plant assemblages from the Kap Stewart Group, East Greenland, to determine the pace of collapse of LT primary productivity. RADs displayed not simply decreases in the number of taxa, but decreases in the number of common taxa. Likelihood tests rejected a hypothesis of continuously declining diversity. Instead, the RAD shift occurred over the upper two-to-four fossil plant assemblages and most likely over the last three (final 13 meters), coinciding with increased atmospheric carbon dioxide concentration and global warming. Thus, although the LT event did not induce mass extinction of plant families, it accompanied major and abrupt change in their ecology and diversity.

Ecological theory shows that relative abundance distributions (RADs) provide important information on the ecological assembly rules for communities in both the present (1, 2) and past (3). The general ecological rules that underpin community assembly are also independent of species composition, thus providing a metric of past diversity that is applicable to communities of disparate composition, phylogenetic history, and age (1, 2). Differences among RADs reflect differences in dominance and rarity as well as richness. RADs describe dominance and rarity more exactly than does evenness (i.e., uniformity of abundances) alone (4). Hypotheses of ecological deterioration make predictions about changes in RADs over time without necessarily predicting extinction (5, 6). Therefore, if prolonged ecological deterioration precedes a mass extinction, then RADs could reveal ecological deterioration better than richness or evenness alone.

We use RADs to examine the pace of diversity loss leading to the Triassic-Jurassic boundary (TJB). The TJB extinction is one of the five greatest in Earth history (7), but the pace of biodiversity loss remains uncertain (812). This hampers our ability to distinguish between competing hypotheses on the causal mechanisms of the TJB mass extinction. Gradual extinction patterns have been reported. RADs offer an opportunity to reexamine the pace of LT biodiversity change in greater detail than provided by either changes in richness or evenness.

We assessed trends in RADs over six taphonomically similar Rhaetian aged fossil plant beds from Astartekløft, East Greenland (10). First, we determined the most likely RAD model for each bed based on the expected number of taxa with x specimens given the observed sample size (3). We considered four RAD models: geometric and the zero-sum multinomial, which are governed largely by ecological succession (1, 2); and lognormal and Zipf, which are governed by increasing ecospace due to facilitation or niche construction (1). Because the different RADs do not represent special cases of each other, we use Akaike’s modified information criterion to choose the best model (3, 13).

Second, we assessed a series of increasingly complicated temporal models of LT plant diversity change. We did this by assessing the likelihood of a range of models, and thus the joint likelihood that 2+ assemblages shared the same RAD. Because not all beds fit the same RAD model, we labeled each model with a more general aspect of diversity: the hypothesized number of genera (S) with frequency greater than 10−6 (Sf>10−6). In order of increasing complexity, we considered (i) uniform diversity over the whole Rhaetian-aged portion of the Astartekløft section (∆Sf>10−6 = 0); (ii) linear diversity decrease over the same interval (∆Sf>10−6 < 0); (iii) static diversity followed by linear decrease in the later Rhaetian portion of the section; (iv) static diversity followed by curvilinear decrease in the later Rhaetian portion of the section.

The simpler temporal models are special cases of the more complicated temporal models. Thus, we can use log-likelihood ratios to test whether a more complicated temporal model is significantly better than a simpler one (14). We tested hypothesized RAD shifts by how well those hypotheses predict observed abundances given the best general RAD model and the hypothesized shift in Sf>10−6, not by how well they predict the best exact model. Second, we reach identical conclusions using Sf>10−5 or Sf>10−4.

Like confidence-interval studies [e.g., (15)], RADs account for unsampled taxa, with lower sample sizes leading to best-fit RADs positing greater numbers of unsampled taxa (3). Although samples from the same RAD over time will show gradual last occurrences of taxa [the Signor-Lipps effect (16)], our likelihood tests will show indistinguishable RADs and thus suggest constant diversity (17). Thus, the Signor-Lipps effect cannot create trends in RADs, even if sampling decreases up-section. The RAD approach has the two additional advantages of implying changes in rarity among (unspecified) unsampled taxa and requiring fewer fossiliferous horizons, as the statistical power comes from the number of sampled fossils rather than the number of occurrences.

We analyze genera rather than species. Because 95% of fossil plant genera in Kap Stewart Group strata have monospecific occurrences within plant beds (10), species-level patterns cannot differ too greatly from genus-level patterns. Moreover, most species determinations for the Kap Stewart flora reflect leaf surface micromorphological traits (18), which might not be taxonomically reliable (10). Genera also add a conservative bias, as the greater taxon numbers provided by species would increase our ability to recognize different RADs.

Taphonomic studies demonstrate that despite some general biases [see (10) for further discussion], leaf litter from temperate and relatively low diversity subtropical floodplain forests provides a relatively accurate indication of both the richness and the dominance-diversity relationship of the live forest community (19). We assume that similar processes affected leaf litter on and preservation from LT floodplain forests of East Greenland (10). RADs of census-collected fossil leaf collections from Greenland should therefore elucidate patterns of changes in LT plant community assembly and diversity for this region. All six LT fossil plants beds at Astartekløft [with the exception of bed 5 (17)] likely represent a geologically instantaneous sample of the standing plant community preserved during river flooding on one or several occasions. RADs from these beds are therefore unlikely to be subjected to any significant time averaging, which can result in lognormal RADs by mixing of different exponential distributions (3).

All beds except bed 3 best fit geometric RADs; bed 3 best fits a lognormal RAD (Table 1 and Fig. 1A). This suggests that LT plant communities were predominantly assembled by simple niche-partitioning rather than more complex assembly rules (3). More important, the best RADs for each bed show a distinct trend toward increasing slope moving up-section and thus decreasing diversity through time (Fig. 1A), especially above 35 m (Fig. 1B). The best hypothesis of continuously decreasing plant diversity (where Sf>10−6 declines by 0.777 per meter; Fig. 1B) is significantly more likely than is the best hypothesis of static diversity (Table 2). The likelihood improves significantly more given a hypothesis of an identical geometric RAD for the first three beds (the first 24 m) and decreasing diversity decreasing markedly over the last three beds [the final 13 m (17)]. Variations on the final model show that we cannot reject the idea that the plant diversity shift is concentrated in the upper 9 m of the Rhaetian aged Astartekløft sedimentary rocks (beds 3 to 5A; Fig. 2), although we can reject the ideas that the shift was distributed over the upper 23 m (beds 1.5 to 5A) or the upper 6 m (beds 4 and 5A). Regardless of the exact timing, a hypothesis of continuous, gradual plant diversity loss suggested by richness decrease alone (10) is not tenable (20).

Table 1

Modified Akaike’s information criteria (AICc) for best examples of each general RAD model. AICc = −2 × lnL[H|data] × n/(nk −1), where H is the best hypothesis from each model, n is the number of specimens, and k is the number of parameters (k = 1 for geometric; otherwise k = 2). The lowest AICc value (bold) gives the best fit (13).

View this table:
Fig. 1

Trends in Late Triassic plant community relative abundance distributions (RADs) and diversity for Astartekløft, East Greenland. (A) Best-fit RAD hypotheses for each bed. (B) Most likely richness of genera with ƒ > 10−6 for each bed (Sf>10−6). Bars give 1-unit support (i.e., lnL ≤ 1 less than maximum); bar width decreases both with increasing sample size and increased fit of general model to the data (14). Lines give predicted diversity given the best hypothesis from three different models of diversity change over meters of sediment. m in the equations gives meters beyond 13 m and 37 m (the heights of apparent shifts). The final hypothesis has one more parameter, as Sf is static until bed 3.

Table 2

Hypotheses of changing diversity going up-section, given here as the changing numbers of taxa (S) with relative abundance >10−6. k gives the number of varying parameters for each hypothesis. lnL gives the log-probability of all observed abundances given the hypothesis and the best model from Table 1. P gives the probability of the difference in log-likelihoods if the simpler hypothesis is correct. Because simple hypotheses are special cases of the complex hypotheses, we assess P using log-likelihood ratio tests. The best 3k hypothesis, linear decrease in diversity over the last 13 m, is omitted as it is worse than the best 2k hypothesis.

View this table:
Fig. 2

Trends in fossil wood stable carbon isotopic composition (δ13Cwood), plant diversity, and atmospheric CO2 concentration. (A) Compilation of published (9) δ13Cwood data from Astartekløft, East Greenland, showing LOESS smoothed trends (using Gaussian kernel function (20). (B) Best hypothesis of plant diversity change (dotted line) plus the best hypothesis postponing that shift (gray line). Sf>10−6 is the hypothesized number of genera with f > 10−6; the same results are achieved with f > 10−5 or f > 10−4. (C) Record of atmospheric CO2 concentration in ppmv derived from stomatal analysis of fossil Ginkgoales leaves from Astartekløft and other localities within Jameson Land (8). δ13Cwood from the same section show a marked decrease, and atmospheric CO2 concentration increases within the stratigraphic range of the plant diversity decrease.

Environmental degradation on ecologic time scales decreases community diversity and complexity and increases RAD slopes (5, 6). We see the same pattern here, but on an evolutionary time scale. If sedimentation for the section was fairly constant and if the Rhaetian is completely represented, then the shift in RADs most likely begins approximately 300,000 to 500,000 years before the TJB. Extraordinarily high sedimentation rates would be required for beds 3 to 5 to represent a snapshot of ecologic time-scale processes. Conversely, large decreases in sedimentation rates over beds 3 to 5 relative to beds 1 to 2 are required to salvage the hypothesis that diversity decreased continuously through the section. There is no sedimentological or taphonomic support for either proposition.

Within uncertainties, the abrupt diversity loss coincides with a moderate +0.95‰ positive excursion in organic matter carbon-isotope composition (Fig. 2) just before the onset of the main TJB negative carbon isotopic excursion (9). These results suggest that the decline in plant diversity at Astartekløft coincided with a major transition between different states in the global carbon cycle from predominantly 12C-sink to 12C-source processes. Available low-resolution CO2 records indicate that the diversity loss occurred between minimum CO2 values reported for the interval [480 ± 160 parts per million by volume (ppmv)] at 32 m and maximum values (1240 ± 400 ppmv) at 47 m (8) and coincided with a major sea-level fall across Europe (11, 21). Concurrently, the inferred global mean surface temperature difference from present (ΔGMST) changed from 2 ± 1.0°C to 7 ± 1.0°C ΔGMST (8). Interpolating between available CO2 estimates indicates that the sudden diversity drop between 33 and 37 m coincided with a mere ~100 to ~350 ppmv rise in CO2 concentration. This was followed by a more protracted diversity decline with a lessening slope approaching the TJB, as atmospheric CO2 (8) and estimated ΔGMST increased to their respective maximum values (21). Therefore, although CO2-induced global warming was likely an important contributory factor to plant species turnover at the TJB (46 m), an alternative or additional triggering mechanism for the abrupt loss of plant diversity between 33 and 37 m at Astartekløft may be possible.

We cannot extrapolate the vegetation responses from one locality globally. However, global hypotheses for the end-Triassic extinctions must predict local patterns; thus, hypotheses failing to predict the Kap Stewart patterns are unlikely as global explanations. Moreover, the most parsimonious explanation for the trends we observe is that they represent regional responses to global environmental change. This argument is supported by the observation that the highest occurrences of plant species at Astartekløft at ~46 m are contemporaneous with those from 12 other Kap Stewart Group localities (18) and that high turnover of fossil plant taxa within the TJB interval has been recorded in North America (22), UK (21), Sweden (18), Spain (23), Austria (24), and Italy (25). Cross-correlation of the stable carbon isotopic profiles from across the globe (9, 21, 24, 25) suggests that the abrupt loss in plant diversity at Astartekløft began at the onset, not the zenith, of the main global carbon isotopic excursion and before turnover of 90% of macrofossil plant species in the Jameson Land region (10, 18). Palynology further supports this temporal correlation as the first appearance of the morphospecies Cerebropollenites thiergartii coincides with the turn to more negative δ13C values in TJB sections at Astartekløft (26), St Audries Bay (UK), and Tiefengraben (Austria) (24). The abrupt loss in Astartekløft plant diversity also coincides with several patterns indicating severe environmental disturbance to the marine environment, including turnover (27) and extinction of shallow marine shelly invertebrates (25), a shallow-water carbonate production crisis (21), and at least localized proliferation of green algal phytoplankton (21, 24).

The abrupt plant diversity loss between 33 and 37 m is consistent with expected plant responses to a catastrophically rapid rather than gradual environmental change and argues against the currently favored extinction mechanisms invoking gradual CO2-induced global warming due to slow release of CO2 from the mantle associated with extrusion of basalt over an area of >10 million km2 (CAMP; Central Atlantic Magmatic Province) (9, 10). Proposed mechanisms of rapid environmental change include a meteorite impact (12), exhalation of thermogenic methane or other gases generated by intrusion of CAMP sill magma (11), sulfur dioxide aerosol release during CAMP eruptions (28), or biogenic methane release from gas hydrates (29). An alternative explanation for the abrupt diversity loss is that it represents a threshold response of LT vegetation to relatively minor increases in CO2 concentration and/or global temperature. High-resolution proxy CO2 and SO2 records, coupled with controlled environment experiments, are required to test further the primary drivers of abrupt LT biodiversity loss.

Supporting Online Material


Figs. S1 to S3

Table S1


References and Notes

  1. Supporting material is available on Science Online.
  2. L. Mander, personal communication (2008).
  3. J.C.M. collected plant paleoecological data, S.P.H collected isotopic and stratigraphic data, and P.J.W. undertook data analyses. J.C.M and P.J.W cowrote the paper, and all authors contributed equally to interpretation. We gratefully acknowledge funding from a Marie Curie Excellence Grant (MEXT-CT- 2006-042531) to J.C.M. and NSF-EAR-0207874 to P.J.W. This is Paleobiology Database contribution 95.
View Abstract

Cited By...


Navigate This Article