Scaled biotic disruption during early Eocene global warming events

. Late Paleocene and early Eocene hyperthermals are transient warming events associated with massive perturbations of the global carbon cycle, and are considered partial analogues for current anthropogenic climate change. Because the magnitude of carbon release varied between the events, they are natural experiments ideal for exploring the relationship between carbon cycle perturbations, climate change and biotic response. Here we quantify marine biotic variability through three million years of the early Eocene that include ﬁve hyperthermals, utilizing a method that allows us to integrate the records of different plankton groups through scenarios ranging from background to major extinction events. Our long time-series calcareous nanno-plankton record indicates a scaling of biotic disruption to climate change associated with the amount of carbon released during the various hyperthermals. Critically, only the three largest hyperthermals, the Paleocene–Eocene Thermal Maximum (PETM), Eocene Thermal Maximum 2 (ETM2) and the I1 event, show above-background variance, suggesting that the magnitude of carbon input and associated climate change needs to surpass a threshold value to cause signiﬁcant biotic disruption.


% total Coronocyclus
suggesting that the isotopic composition and release mechanism of the injected carbon were similar across these events (Stap et al., 2010;Abels et al., 2012).These closely spaced CIEs/climate change events of different magnitudes provide enormous potential for quantifying biological sensitivity to carbon cycle perturbations.However, as new biotic records of these hyperthermals emerge, we require techniques that enable the consistent and quantitative assessment of magnitude and significance of biotic change in multi-taxic datasets.
Here, we apply a modified coefficients of variation technique to quantify levels of variation in a long time-series record of calcareous nannoplankton abundance across multiple hyperthermals, and compare these results with data from other plankton groups and from the mass extinction event at the Cretaceous-Paleogene boundary.

Material and nannofossil data
We generated high-resolution calcareous nannoplankton (nannofossil) assemblage records across a nine-meter section at Ocean Drilling Program (ODP) Site 1209 (32 • 39.11 N, 158 • 30.36 E, present-day water depth 2387 m) in the paleosubequatorial Pacific Ocean (Fig. 1).This section spans five CIEs, the PETM, ETM2, H2, I1 and I2, from ∼53.0 to ∼56.2 Ma (nomenclature following Cramer et al., 2003;Zachos et al., 2010).The CIEs are recorded in the isotopic composition of bulk sediment carbonate (Murphy et al., 2006) and typically correspond to clay-rich dissolution horizons, illustrated by the magnetic susceptibility record (Fig. 2b).Although some dissolution is evident during this interval at Site 1209, it is less severe than, for example, the deeper sites in the Walvis Ridge PETM transect (2717-4755 m water-depth; Zachos et al., 2005).The age model for Site 1209 uses tie-points in the δ 13 C record to correlate with the orbitally tuned stratigraphy and absolute ages of ODP Site 1262, summarized in Zachos et al. (2010) (Fig. 2a).Assemblage data (% abundances) are based on statistically significant abundance counts of ∼600-800 nannofossils per sample (Gibbs et al., 2006b), and were collected at 4-5 cm (∼13 kyr) spacing.Species-level counts were performed on the samples taken across the ETM2, to complement existing data from the PETM (Gibbs et al., 2006b).Species were grouped into genera and ranked according to average abundances, with >97 % of the assemblage typically represented by around 10 genera.For the long time-series samples, we counted generic groups, selecting the genera that include the ten most abundant across the PETM and the ETM2, together 11 genera (see discussion in Sect.3.1 below) (Fig. 1).

Methods, analytical approach and sensitivity tests
Our approach to quantifying and comparing biotic variability across the hyperthermal events required (1) a means of utilizing routinely collected relative abundance data, (2) a technique ideally independent of taxic composition (which may vary with time and space due to evolution and biogeography, as well as between specialists), (3) a means to objectively quantify overall assemblage variability but remove  (Zachos et al., 2003;Murphy et al., 2006), with the CIEs highlighted in grey, and chronostratigraphic tie-points indicated by black crosses.Panel (B) shows magnetic susceptibility (Bralower et al., 2002) dominance-biasing by a small number of taxa, and, (4) the application of an objective means to maximize signal-tonoise.These requirements are met by combining a method of generating the best-smoothed fit (significant zero crossings of derivatives, SiZer; Chaudhuri and Marron, 1999), followed by a quantitative assessment of assemblage variability (summed coefficients of variation, cv ).

Relative abundance and genus level data
For the deep-time geological records studied here, we consider that relative abundance data are the most appropriate as a first-order approximation of relative biological change using the least degraded data, and an objective measure of variance that allows comparison between different stratigraphic levels, different sites and different fossil groups.Relative abundance data are a robust record of population/assemblage dynamics and thus a measure of primary biological interac-tion and response.It is a commonly applied and relatively rapidly produced data type, which facilitates the comparison of published datasets from different locations and time intervals.Although cognisant of closed sum problems, we have tested for these effects by using a ranking analysis, described below.We did not attempt to generate "absolute" numbers per gram or flux-estimates-type data, because such methods require continuous high-resolution age models, are highly sensitive to sedimentation rate changes, and preparation methodology varies widely.This renders the comparison of data problematic, and each methodological step potentially introduces error and degrades the data.
We also chose to adopt a genus-level counting approach to facilitate rapid data capture.Nannofossil genera represent robust groupings of taxa with very similar morphology, and generic classification in nannofossils is typically more stable and less ambiguous than species level, which is particularly important for long time-series studies.Furthermore, morphospecies within genera typically share similar ecologies; e.g.Discoaster are thermophylic and Toweius are cosmopolitan bloom-formers.While it might be argued that combining species reduces the variance signal, there are two points to consider here.First, combining species does not inherently result in loss of variance signal, as integrated records will comprise variance from multiple species that is not automatically lost or cancelled out.Loss may occur if all the species co-varied (but even this could still produce a relatively high level of variance), or if abundance changes occurred at a frequency that cancelled each other out.Second, this is a relatively moot point in our dataset where, of the 11 genera that comprise the Eocene hyperthermals dataset, seven effectively record only one species.Coccolithus and Zygrhablithus are essentially monospecific here, and Campylosphaera, Cruciplacolithus, Chiasmolithus, Coronocyclus and Sphenolithus are represented by few species and dominated by one.The remaining taxa, Fasciculithus, Toweius and Discoaster, are multi-species signals, but species within these groups share similar ecologies.Moreover, the taxonomy of morphospecies within Fasciculithus and Discoaster is particularly inconsistent between authors due to very plastic and intergrading morphotypes; thus genus-level data provide a more robust method of capturing comparable data in these groups.Therefore, the genera we count, regardless of whether they are monospecific or multi-species groups, are valid morphogroups, which are internally consistent through time and provide reliable and comparable taxonomic data.

Combined smoothing and variance techniques
The relative abundance record of each taxon was individually run through the SiZer program to produce a smoothed record that shows the least degradation but highest confidence in signal (Fig. 1).The SiZer technique generates a set of smoothed curves, which use the full range of bandwidths available, and provides criteria by which the most appropriate of these smoothed curves can be chosen (Chaudhuri and Marron, 1999;Chaudhuri et al., 2012;Wagner, 2012; see the application to palaeoceanographic records in Rohling and Pälike, 2005).Smoothing removes part of the noise inherent to these types of data, but using SiZer we introduce a transferable, objective set of criteria by which to increase the signal-to-noise ratio.We have tested for any signal loss by analysing smoothed and non-smoothed records, and similar trends are seen in both results (Fig. 3a).However, the records of summed coefficients of variation ( cv ) from the SiZer smoothed abundances are consistently around half the amplitude of the cv values from the raw relative abundances.As no signal loss results from the SiZer smoothing, we have applied smoothing in order to facilitate comparisons between datasets that may contain different levels of noise.Following SiZer smoothing, we calculated and summed the cv for the most abundant taxa across moving windows.We used windows of 150 kyr, to capture the majority of each CIE (Figs. 2c and 4a), and windows of 25 kyr to resolve patterns within each CIE.This latter window duration is the shortest achievable with the sampling resolution of the data (Fig. 4b).Because we want to directly compare net assemblage change across each event, we also calculated coefficients of variation (CV) across only the duration of each of the CIEs that are shorter than 150 kyr (Figs.2c and 4a).
CVs are calculated by determining the standard deviation of taxon abundance over a given stratigraphic interval, divided by the taxon's mean abundance (Eq. 1, Fig. 3a and  b).This provides a normalised measure of the spread of data about the arithmetic mean.Dividing the standard deviation by the average abundance is necessary in order to remove bias towards high abundance taxa.More abundant taxa can have a disproportionately large standard deviation, but equally, rare taxa that, for example, fluctuate between absence and rare occurrence would have a large standard deviation when divided by their average low abundance.Therefore we have excluded the CVs of rare taxa (<0.7 %) from the Site 1209 record, leaving 11 taxa which contribute >97 % of the assemblage.We have then summed the CVs for the most common taxa to gain a quantified estimate of overall assemblage variance (Eq.1).2c plotted for each peak (using the values highlighted by red circles for H2, I1 and I2) and every intervening 150 kyr.The positions of the cv data points for the PETM and ETM2 are shown against inferred maximum bulk carbonate δ 13 C values (Lourens et al., 2005;Zachos et al., 2005).The level of a suspected unconformity, at ∼700 kyr above the PETM onset, is plotted as an open black circle and the grey area represents a "background" field.Panel (B) shows the cv values across 25 kyr with values plotted for the peak excursions (open black circles) and every intervening 50 kyr.Only data that are spaced at least as widely apart as the CV window are plotted in order for each data point to be independent of its neighbours.Stratigraphically adjacent data points are joined by lines with the arrows indicating the up-section direction.
where t 1 and t 2 are the start and end of the window, n being the number of taxa.
In order to explore the influence of each taxon on the downcore record of cv , we recalculated the cv values multiple times, each time removing a different taxon (Fig. 2d).We can see that three key taxa dominate in different portions of the record: Coronocyclus, Zygrhablithus and Fasciculithus.For example, Fasciculithus dominates the PETM cv values, which is not unexpected given that the abundance and diversity loss of this genus is one of the characteristic features of the event.Therefore, the combination of taxa ensures that we have an integrated picture of assemblage change where variance occurs in different taxa at different stratigraphic levels.
We have also applied the cv technique to other published PETM datasets, including planktic foraminiferal and dinoflagellate cyst records (see Fig. 5a caption), choosing records with the necessary temporal resolution, at least ∼20 kyr sampling interval, to allow for a reasonable comparison of data.We used a window size corresponding to the total duration of the PETM, and have plotted peak and background values in Fig. 5a.Note that, for the Bass River datasets, the resultant cv values may be underestimates as the full duration of the CIE is not recorded.In addition, we also applied the method in two sections that include the Cretaceous-Paleogene boundary (K-Pg, ∼65 Ma), from ODP Sites 1210 and 1262 (Bown, 2005;Bernaola and Monechi, 2007), to provide comparison with the nearest mass extinction event (Fig. 5b).To facilitate direct quantitative comparison, we used the 11 most abundant nannofossil taxa, including both outgoing Cretaceous and incoming Paleocene taxa, but excluded any obvious reworking of Cretaceous taxa in Paleocene sediments.

Sensitivity tests and controls on the summed coefficients of variation record
The magnitude of CVs will be a function of sample window duration, which, ideally, is kept constant.However, sedimentological factors that influence temporal resolution, such as sedimentation rate and hiatuses, may modify the effective window duration.In the Site 1209 analysis, for example, there may be some variation in window duration due to changes in sedimentation rate, occurring at a resolution higher than the applied age model.These biases may impact cv values and need to be considered when interpreting the records.We have therefore tested the sensitivity of cv values by exploring the impact of varying window size and by artificially introducing hiatuses and dissolution.These tests have been applied to the depth record so we can assess how changes in sedimentation rate may influence the signal in the age-domain data.First, we calculated cv using different window durations from 10 cm up to 150 cm (Fig. 3c).As expected, cv values are higher when a greater depth window is applied because greater stratigraphic duration is being incorporated into each window, equivalent to www.biogeosciences.net/9/4679/2012/Biogeosciences, 9, 4679-4688, 2012 4.5 4 0 0.5 1 1.5 2 2.5 3 3.5 4.5 4 Fig. 5. Scatter plots of summed coefficients of variation ( cv ) against magnitude of CIE for different plankton groups and for comparison with the Cretaceous-Paleogene boundary mass extinction event.Panel (A) shows background and peak PETM cv values for multiple PETM sites -nannofossils (black circles, black lines), planktic foraminifera (grey circles, grey lines) and dinoflagellate cysts (grey circles, grey dashed lines).Where possible pre-event cv values are plotted but where missing, then the cv values from the recovery interval are used.1 -ODP onshore drill site Bass River (nannofossils (Gibbs et al., 2010) and dinocysts (Sluijs and Brinkhuis, 2009)); 2 -ODP Site 690, Southern Ocean (nannofossils (Bralower, 2002) and planktic foraminifera (Kelly, 2002)); 3 -Lomonosov Ridge, Arctic Ocean (dinocysts (Sluijs et al., 2008)); 4 -TDP, Tanzanian drilling project Site 14 (nannofossils (Bown and Pearson, 2009)); 5 -ODP Site 1209 (nannofossils -herein -and planktic foraminifera (Petrizzo, 2007)); 6 -ODP Site 1260, Demerara Rise, Atlantic Ocean (nannofossils (Mutterlose et al., 2007)).All cv data use 10 taxa for comparison or, for the foraminifera data, normalised to 10 taxa.Panel (B) illustrates the cv of the PETM and ETM2 from Site 1209 compared with the K-Pg using data from Site 1210 (horizontal black line (Bown, 2005)) and Site 1262 (WR -Walvis Ridge, horizontal grey line (Bernaola and Monechi, 2007)).The K-Pg cv value is not plotted against its respective δ 13 C value as this is not a meaningful measure of environmental perturbation for this event.
decreasing sedimentation rate (Fig. 3c).Achieving an above background cv peak (equivalent to a value of around 1.6 cv , see Fig. 4a) requires a window size increase to just over 100 cm, which is equivalent to a minimum sedimentation rate reduction of approximately 55 % or a hiatus of just over 50 cm.
In the second sensitivity test, we introduced artificial hiatuses and dissolution levels into one interval that has background variability and one that includes a CIE (Fig. 3d and  e).Where the hiatus is introduced into a background interval, we see elevated cv values, but not event-level values, and the longer the hiatus then the greater the apparent increase in cv (Fig. 3d).Where a hiatus was inserted at a CIE, there is little change in the size of cv , but some impact on the structure of the record.
For the dissolution test, we removed 75 % of the nannofossils according to their different dissolution susceptibilities (following Gibbs et al., 2010).For example, we removed fewer Discoaster (which are large and robust nannoliths) than we did Campylosphaera (a less robust coccolith).The resulting cv record suggests very little impact on the background assemblages but some amplification of the cv signal across the CIE (Fig. 3e).This is because the background interval assemblages at Site 1209 are dominated by taxa that have very similar dissolution susceptibilities, and so altering their abundances does not result in a major change in their recalculated relative abundances.In contrast, across the CIE intervals, there is a greater contribution by, for example, Zygrhablithus, Discoaster and calcispheres, which increases the range of dissolution susceptibility in the observed dominant taxa.However, the change in cv is relatively small.Therefore, to produce the values of cv associated with the PETM and ETM2 would require substantially larger hiatuses, dissolution or sedimentation rate changes than tested here.Such large sedimentological changes are usually apparent in these deep-sea sediments and so we are confident that this does not account for the larger cv peaks present here.However, variability in the background record of cv may point to potential hiatuses or sedimentation rate changes that are at a resolution higher than the age model, such as the unexplained variability at ∼700 kyr above the PETM onset (Fig. 2c).
Finally, by looking at the ranked abundance of taxa from one time interval to the next, we can test whether the fluctuations of dominant taxa are affecting the relative abundance of all the other taxa, for example, the fluctuations of Zygrhablithus.If only Zygrhablithus is changing, then the relative rank of all other taxa should not change.In fact the rank order in the non-Zygrhablithus taxa does change for both the PETM and the ETM2 but does not for the I1.It could be that the Zygrhablithus peak is the only major assemblage shift associated with I1, but this does not account for the major assemblage shifts across the PETM and ETM2.

ODP Site 1209 hyperthermal record
The majority of the cv values from Site 1209 cluster between 0.4 and 1.6 cv , supporting the concept of a background range of biotic variability (Figs.2c, 4a).However, several intervals have values above this range, indicating a magnitude of variability that is exceptional or abovebackground biotic change, and supporting the existence of distinct "events".The highest cv value is associated with the PETM and there is a broadly linear trend of declining cv with decreasing size of CIE for ETM2, I1 and H2 (Figs. 2c  and 4a), indicating a scaling of biotic response, which is in line with evidence for scaled temperature change and CIEs (Stap et al., 2010).The I2 CIE interval is less clear-cut, as it has an above background cv value if considered within a 150 kyr window, but a background value if using only the CIE duration (Fig. 2c).Similarly, H2 does not show anomalous values of cv .This may indicate a biotic sensitivity threshold for calcareous nannoplankton at the Site 1209 location that lies between the CIEs of H2 and I1, at approximately 0.6 ‰.Alternatively, ETM2 and H2, and I1 and I2, are relatively closely-spaced (∼100 kyr), paired events and nannoplankton communities may not have had time to revert to background compositions.However, nannofloral communities exhibit response and recovery times that are shorter than 100 kyr elsewhere in our records (e.g. the onset of the PETM), and given the short generation times of these plankton, measured in days, this explanation appears unlikely.
Given that carbonate dissolution is associated with each of the CIEs, the scaling between event magnitude and biotic variance might reflect a simple relationship between carbonate erosion and dissolution-skewed abundance patterns.However, the peak dissolution at each event is decoupled from abundance changes (Fig. 6) and the interval of dissolution is shorter than the window across which we have calculated summed correlation of coefficients.Therefore, it is unlikely that cv is an artefact of co-varying dissolution with δ 13 C, although, based on the sensitivity tests, there is some potential for the cv -δ 13 C relationship to be amplified.
When applied at higher stratigraphic resolution, the cv metric reveals important details of the timing of the environmental versus biotic perturbation (Fig. 4b).The PETM, ETM2, and I1 events all show elevated cv values in the intervals immediately prior to the CIE onset, and each has recovery intervals in which cv values drop back to background levels before the carbon isotope values, a pattern also observed in temperature and CaCO 3 records (Zachos et al., 2003(Zachos et al., , 2005)).This asymmetric structure appears to be a real feature of the events, rather than a data artefact, as The nannofossil-based preservation index uses the ratio of indeterminate Toweius to identifiable Toweius with higher values indicating higher levels of dissolution (Gibbs et al., 2006a(Gibbs et al., , 2010)).The peak in Zygrhablithus associated with each CIE occurs at the start of the carbon isotope recovery in each case and is consistently above the level of peak dissolution (highlighted in yellow).
it is also seen in the original abundance data (Fig. 1).The shifts in nannoplankton assemblages prior to the respective CIEs are similar to abundance trends seen in PETM planktic foraminiferal and dinoflagellate cyst records from other locations (Thomas et al., 2002;Sluijs et al., 2007b), and our data suggest that similar precursor environmental change also occurred prior to the ETM2 and I1 events.

Global levels of variance
To test the wider significance of our Site 1209 results, we have also analyzed published PETM plankton datasets from a range of shelf, slope and open-ocean localities in the Atlantic, Indian and Southern oceans, and included data for nannoplankton, planktic foraminifera and organic-walled dinoflagellate cysts (Fig. 5a).The background-to-peak CIE cv values ( cv ) for nannoplankton are comparable from these other sites but are slightly higher at high latitude oceanic sites.The planktic foraminifera and dinoflagellate data also indicate comparable ranges of variability (Fig. 5a), but they are consistently higher than the nannofossil records at equivalent sites, perhaps suggesting slightly different relative sensitivities to environmental change in the different plankton groups.For example, dinoflagellate cyst abundance records across the PETM exhibit distinct "acmes" and multiple taxa appear and disappear, resulting in higher variance, as environmental thresholds are crossed across a range of parameters including salinity, trophic state and water depth www.biogeosciences.net/9/4679/2012/Biogeosciences, 9, 4679-4688, 2012 (e.g.Sluijs and Brinkhuis, 2009).In contrast, calcareous nannoplankton records are characterized by abundance shifts in dominant species alongside minor shifts across a range of sub-ordinate taxa, but with few appearances and disappearances.These records demonstrate the integrative power of this analytical approach, which enables comparison of diverse biotic data and provides a means to holistically consider responses to multi-stressor environmental changes, specifically the warming, nutrient and sea-level changes that account for the assemblage variations during the PETM (see Sluijs et al., 2007a).The inclusion of organisms from different functional groups and trophic levels -zooplanktonic planktic foraminifera, oceanic nannoplankton, mixotrophic neritic dinoflagellates -also provides a wider-ranging description of the biotic change, and the results are therefore more broadly representative of the marine ecosystem as a whole.
Finally, in order to place the hyperthermal nannoplankton perturbations in the broadest evolutionary context, we have also analysed data from the K-Pg mass extinction event, ∼65 Ma, which saw almost complete extermination of the group (Bown, 2005).The K-Pg scenario can therefore be considered an end-member biotic perturbation; i.e. the pre-and post-event assemblages have virtually no taxonomic similarity.The K-Pg cv values from Shatsky Rise Site 1210 (for direct comparison with our hyperthermals dataset, Bown, 2005) and Walvis Ridge in the south Atlantic (Bernaola and Monechi, 2007) are 17 and 14, respectively, with a background-to-peak change of ∼10-13 (Fig. 5b).The Shatsky Rise cv of 17 is close to the theoretical maximum for complete taxonomic turnover at a single event level.The theoretical maximum value can be estimated using different magnitudes of abundance changes across an event level, ranging from low-level to complete turnover in a highly heterogeneous (i.e.uneven) assemblage to low-level to complete turnover in a more homogeneous assemblage.Higher theoretical values are possible only when abundance declines are added prior to the event level turnover.
The spectrum of cv values resulting from analysis of the hyperthermals and K-Pg mass extinction data suggests that the metric is sensitive to biotic response over an extremely wide range of environmental change.This not only allows discrimination between background and event-level intervals but also provides a measure of event-level perturbation magnitude.Therefore, whilst extinction rate data provide a quantitative means of characterizing biotic response to major events such as the K-Pg, the cv metric allows for characterization of biotic response to events, like the hyperthermals, where evolutionary turnover in plankton is relatively modest (Kelly et al., 1998;Gibbs et al., 2006a).

Threshold behaviour in plankton records
These Paleogene plankton data show threshold behaviour and scaled response to the environmental changes associated with carbon cycle perturbations.But is such behaviour inherent in planktonic ecosystems and does this have any relevance for understanding how modern plankton might respond to future ocean change?Specifically, our data show that nannoplankton assemblage perturbations occur with environmental change associated with CIEs of greater than 0.6 ‰, equating to around 2 • C of global warming, using a proportional relationship between warming and CIE magnitude (Stap et al., 2010).This threshold value, however, may not be directly applicable in the modern ocean.First, rates of carbon cycle change are considerably faster at present, and, second, the modern ocean has different physical baseline conditions (e.g.ocean-atmosphere chemistry and temperature; Zeebe et al., 2009;Goodwin et al., 2009;Ridgwell and Schmidt, 2010), with biological systems today adapted to icehouse climates, rather than the greenhouse climates of the Paleogene.On the one hand the modern ocean system may reach the perturbation threshold more rapidly because of increased rates of change, in addition to the absolute levels of environmental change.By contrast, the greenhouse ocean system may already have been closer to a biologically constrained upper thermal limit and so the threshold would have been reached through relatively smaller environmental changes (see discussion in Huber, 2008).Regardless of the absolute value, this estimate of thermal/carbon-perturbation threshold represents a first-order attempt to place constraints on biological thresholds with the possibility that future biotic response may scale in a similar way to the hyperthermals. Figure1

Fig. 3 .
Fig. 3. Sensitivity tests and comparison of downcore summed coefficients of variation (CV) records calculated using smoothed and original raw percent abundance data.Panel (A) shows cv records from smoothed and raw data.Panel (B) is a comparison of total summed CVs of the smoothed data using CVs divided by the taxon's average across 50 cm (grey line) and 100 cm windows (black line), whereas in Panel (A) the CVs are divided by the taxon's average abundance across the entire interval.Panel (C) is a comparison of background cv values that result from different window sizes, extrapolated as vertical dashed lines; the 50 cm background cv level is highlighted in red.Also marked is the sedimentation rate change equivalent (in percent) to which these background levels correspond.Panels (D) and (E) show the results of artificially introduced hiatuses and intervals of intense dissolution, respectively.The star indicates the levels of a possible natural hiatus.The original cv record is shown in black and the records that include imposed hiatuses and dissolution are shown in red and purple, respectively.Red and purple vertical bars highlight the intervals where hiatuses and dissolution have been introduced.

Fig. 4 .
Fig.4.Scatter plots of summed coefficients of variation ( cv ) against magnitude of carbon isotope excursion.Panel (A) shows the cv values from Fig.2cplotted for each peak (using the values highlighted by red circles for H2, I1 and I2) and every intervening 150 kyr.The positions of the cv data points for the PETM and ETM2 are shown against inferred maximum bulk carbonate δ 13 C values(Lourens et al., 2005;Zachos et al., 2005).The level of a suspected unconformity, at ∼700 kyr above the PETM onset, is plotted as an open black circle and the grey area represents a "background" field.Panel (B) shows the cv values across 25 kyr with values plotted for the peak excursions (open black circles) and every intervening 50 kyr.Only data that are spaced at least as widely apart as the CV window are plotted in order for each data point to be independent of its neighbours.Stratigraphically adjacent data points are joined by lines with the arrows indicating the up-section direction.