Changes in benthic ecosystems and ocean circulation in the Southeast Atlantic across Eocene Thermal Maximum 2

Eocene Thermal Maximum 2 (ETM2) occurred ~1.8Myr after the Paleocene-Eocene Thermal Maximum (PETM) and, like the PETM, was characterized by a negative carbon isotope excursion and warming. We combined benthic foraminiferal and sedimentological records for Southeast Atlantic Sites 1263 (1500m paleodepth) and 1262 (3600m paleodepth) to show that benthic foraminiferal diversity and accumulation rates declined more precipitously and severely at the shallower site during peak ETM2. As the sites are in close proximity, differences in surface productivity cannot have caused this differential effect. Instead, we infer that changes in ocean circulation across ETM2 may have produced more pronounced warming at intermediate depths (Site 1263). The effects of warming include increasedmetabolic rates, a decrease in effective food supply and increased deoxygenation, thus potentially explaining the more severe benthic impacts at Site 1263. In response, bioturbation may have decreased more at Site 1263 than at Site 1262, differentially affecting bulk carbonate records. We use a sediment-enabled Earth systemmodel to test whether a reduction in bioturbation and/or the likely reduced carbonate saturation of more poorly ventilated waters can explain the more extreme excursion in bulk δC and sharper transition in wt % CaCO3 at Site 1263. We find that both enhanced acidification and reduced bioturbation during the ETM2 peak are needed to account for the observed features. Our combined ecological and modeling analysis illustrates the potential role of ocean circulation changes in amplifying local environmental changes and driving temporary, but drastic, loss of benthic biodiversity and abundance.


Introduction
Potential environmental impacts of increasing atmospheric CO 2 concentrations include warming, increased intensity of the hydrological cycle, and nutrient influx into the oceans, ocean stratification, ocean acidification, and increased hypoxia [Caldeira and Wickett, 2003;Hutchins et al., 2007;Solomon et al., 2009;Coma et al., 2009;Keeling et al., 2010;Durack et al., 2012;Pörtner et al., 2014], any or all of which may affect organisms and ecosystems. However, anticipating the biotic response to these multiple, potentially synergistic environmental parameters is challenging [Bopp et al., 2013;Melzner et al., 2013;Norris et al., 2013;Pörtner et al., 2014]. The response of species and ecosystems to changing environments has been, and continues to be, tested in mostly single-driver laboratory experiments, producing short-term, species-specific, and mainly physiological information [e.g., Kroeker et al., 2010;Pörtner et al., 2014]. Such experiments are valuable but reflect neither the complexity of the natural environment nor the adaptability of organisms on long time scales. Records of periods of past climate change, can, however, provide a detailed, quantifiable account of biotic response [e.g., Hönisch et al., 2012;Speijer et al., 2012]. A series of global warming and carbon release events ("hyperthermals") of variable intensity, occurring superimposed upon gradually rising global temperatures during the early to mid-Palaeogene [Thomas and Zachos, 2000;Cramer et al., 2003;[Winguth et al., 2012], with bottom water deoxygenation common along continental margins [Thomas, 1998;Nicolo et al., 2010] and the inferred occurrence of a broad expansion of oxygen minimum zones in the open ocean [Zhou et al., 2014]. Bottom water deoxygenation may have occurred at some open ocean southeast Atlantic sites [Chun et al., 2010;Post et al., 2015] but not in the Pacific [Pälike et al., 2014]. Nutrient availability and productivity may have increased in marginal basins but decreased in pelagic settings, although discussion is still ongoing due to regional difference in nutrient availability and productivity [Gibbs et al., 2006;Thomas, 2007;Winguth et al., 2012;Schneider et al., 2013;Sluijs et al., 2014;Stassen et al., 2015]. Knowledge of these changes is important because it allows exploration of the relationships between ecological sensitivity and environmental change.
In this paper, we assess the biotic response of benthic ecosystems to ETM2 environmental changes at Walvis Ridge. We analyze a series of coupled climate and conceptual Earth system modeling experiments in order to explore the potential causes and consequences of benthic ecological change.
We took a subset of samples for sediment analysis, at 1.0 cm resolution across ETM2 and 10-15 cm preevent and postevent, as defined by Zachos et al. [2004aZachos et al. [ , 2004b (Tables S1 and S2 in the supporting information).

10.1002/2015PA002821
Samples were washed through a 63 μm sieve using Reverse Osmosis deionized water, dried, and split into 63-150 μm and >150 μm size fractions. For benthic foraminiferal analysis a subset of the samples from Stap et al. [2009] was used, with a sample spacing of 2.0 cm across ETM2 and 10 cm above and below (Tables S3 and S4).

Age Model
In the ETM2 age model for Walvis Ridge, Stap and coworkers [Stap et al., 2009] adjusted the terrigenous flux using Gaussian fitting techniques to optimally align the carbon isotope and calcium carbonate weight percent records. The result is an inferred fluctuating terrigenous flux at Site 1262 (higher during peak ETM2 conditions) and Site 1265 (lower during peak ETM2 conditions), with stable rates of terrigenous input at Sites 1263 and 1267. Disparity in the sign of terrigenous flux change across the event is somewhat unlikely, given the relative geographic proximity of the sites. Forcing an exact alignment of the primary features of the records is also potentially problematic because the apparent timing of events depends on bulk sediment rate and extent of bioturbation [Ridgwell, 2007], as well as differences in carbonate preservation [Kirtland Turner and Ridgwell, 2013], both of which can be expected to differ between sites and may vary in time.
We hence constructed an alternative age model, assuming a stable, site-specific terrigenous flux across the ETM2. There is evidence for generally elevated rates of chemical terrestrial weathering across the PETM [e.g., Kelly et al., 2005;Ravizza et al., 2001] and thus presumably also ETM2, but the total supply rate of particulate terrigenous material to Walvis Ridge may not necessarily have increased. In contrast, if the terrigenous input were dominated by airborne dust, a decrease under global warming would be expected [Mahowald et al., 2006].
We calculated relative sediment age based on reported CaCO 3 wt % and dry bulk density [Zachos et al., 2004a[Zachos et al., , 2004b and using a terrigenous mass accumulation rate (TMAR) derived from an interval of sedimentation characterized by relatively stable climatic conditions immediately prior to ETM2 onset, and between precession cycle tie points of Westerhold et al. [2007], at 298.52-301.52 mcd at the shallow Site 1263 (~6 × 21 kyr cycles) and 118.5-121.83 mcd at the deep Site 1262 (~12 × 21 kyr cycles). The resulting TMARs were 0.154 g/cm 2 /kyr for Site 1262 and 0.191 g/cm 2 /kyr at Site 1263. In order to facilitate comparison with previous studies, we calculated terrigenous sedimentation rates for both sites: 0.13 cm/kyr at Site 1263 and 0.12 cm/kyr at Site 1262, and we adopt the zero relative age point defined by Stap et al. [2009]. Our age model is compared with that in Stap et al. [2009] in Figure 2. A full list of equations can be found in Text S1 in the supporting information.

Sedimentology and Benthic Foraminiferal Analysis
Site-specific bulk and carbonate mass accumulation rates (MARs) were calculated based on our age model. The CaCO 3 fine fraction (FF) (<63 μm) consists predominantly of calcareous nannofossils, the coarse fraction (CF) of planktic foraminifera. We hence used the coarse fraction MAR (>63 μm) to approximate the  foraminiferal mass accumulation rate (FAR). The foraminifera to nannofossil AR was then approximated by dividing the CF AR by the FF AR (>63 μm/<63 μm) (see Text S1 for a full equation list). Planktic foraminiferal accumulation rates, in terms of number of specimens (foraminifera #/cm 2 /kyr), were calculated from planktic foraminiferal counts in the >150 μm size fraction. The effect of dissolution was assessed using fragmentation data [Le and Shackleton, 1992], based on counts of 500 specimens per sample: fragmentation ratio (%) = 100% × (number fragments/8)/(number fragments/8 + number whole).
We determined the relative abundances of benthic foraminiferal taxa and used these to infer changes in carbonate saturation state, oxygenation, and food supply [Jorissen et al., 1995[Jorissen et al., , 2007Thomas, 1998Thomas, , 2007Gooday, 2003;Gooday and Jorissen, 2012;Foster et al., 2013]. Comparisons between past and recent benthic environments need careful evaluation, because Eocene deep-sea benthic foraminiferal assemblages were structured very differently from todays'. For instance, taxa reflecting highly seasonal deposition of organic matter were generally absent or rare, and cylindrically shaped taxa with complex apertures, which are now extinct, were common [e.g., Thomas and Gooday, 1996;Thomas, 2007;Hayward et al., 2012]. The distribution of these extinct taxa resembles that of buliminids [Hayward et al., 2012], and they were probably infaunal, as confirmed by their δ 13 C values [Mancin et al., 2013]. The living species Nuttallides umbonifera [Bremer and Lohmann, 1982] reaches high relative abundances between lysocline and carbonate compensation depth, and we infer that increases in relative abundance of its ancestral species N. truempyi similarly correlate with poorly saturated waters, as confirmed by its bathymetric occurrences [Thomas, 1998]. Benthic foraminiferal accumulation rates (BFARs) are a proxy for delivery of food to the seafloor and generally are higher at shallower depths [Herguera and Berger, 1991;Jorissen et al., 2007]. Benthic foraminiferal accumulation rates were calculated as BFAR = benthic foraminifera (# g À1 ) × bulk MAR. (A full list of sedimentological and derived accumulation rate definitions and calculations is given in Text S1.)

Earth System Modeling
We explore some of the possible influences on the sediment record of ETM2, including changes in benthic foraminiferal abundance and bioturbation, using the GENIE Earth system model. GENIE comprises a 3-D ocean circulation model coupled to a 2-D sea ice and atmospheric energy-moisture-balance model plus representation of ocean-sediment-weathering carbon cycling, as summarized by Archer et al. [2009]. Continental configuration and climatology, initial ocean chemistry, atmospheric CO 2 , and total global weathering flux, are as described by Ridgwell and Schmidt [2010]. The model is spun-up for a total of 200,000 years to fully balance marine CaCO 3 sedimentation versus weathering and create a sufficient sediment column thickness to support any subsequent CaCO 3 "burn-down" [Ridgwell, 2007]. The model grid and initial distribution of sedimentary wt % CaCO 3 is illustrated in Figure 1a.
To perturb bulk carbonate content and the recording of the δ 13 C signal, the model was run with a prescribed time history of atmospheric composition. We assumed a gradual doubling of pCO 2 over 45 kyr from 834 ppm  to 1668 ppm at the peak of ETM2 [Stap et al., 2009], followed by a decline.  Atmospheric CO 2 δ 13 C is mirror-imaged and assumes an excursion magnitude of À1.5‰ [Stap et al., 2009]. We do not aim to reconstruct the history of CO 2 emissions (unlike, e.g., Kirtland Turner and Ridgwell [2013]) but instead create and apply to the model a deliberately conceptual time history of atmospheric pCO 2 . Doubling of CO 2 in our idealized carbon forcing drives an~2.9°C warming in mean annual average ocean surface temperatures (~3.0°C in the deep sea)-consistent with available ETM2 temperature proxies [Stap et al., 2010a;d'Haenens et al., 2014]. The form of prescribed pCO 2 and δ 13 C are also chosen such that together, the decline and recovery of δ 13 C can be replicated at Site 1262. Note that we do not attempt to explicitly model δ 18 O, which requires a detailed simulation of atmospheric moisture transport and hence a coupled climate model [e.g., Tindall et al., 2010]. Given the similarity between δ 13 C and δ 18 O ETM2 horizon anomalies, it is unlikely that simulating δ 18 O in the model would provide additional constraints.
To explore what factors might help explain the different sedimentological and isotopic (δ 13 C) observations at Site 1263, we ran permutations of (i) bioturbational mixing occurring continuously throughout the experiment versus discontinuous bioturbational mixing, with bioturbation ceasing during the peak of the event; (ii) "interface" dissolution of carbonate (the default setting in GENIE) versus "homogeneous" dissolution [Ridgwell, 2001]; and (iii) no significant ocean circulation change versus reduced bottom water saturation at intermediate water depths (which we crudely simulate by increasing the pressure used in calculating carbonate stability at 1263 by the equivalent of 2000 m water depth), summarized in Table 1. All experiments were run for 100 kyr and sediment cores "extracted" from the model grid [Kirtland Turner and Ridgwell, 2013;Ridgwell, 2007] at locations corresponding to the Walvis Ridge area ( Figure 1a)-one at 1500 m model water depth (the model Site 1263 analogue) and one at 3600 m (analogue to 1262). The chronology for the model cores is created analogous to the observations and assumes a constant terrigenous flux to the sediments, which assumes a fixed globally uniform value of 0.180 g cm À2 kyr À1 following Panchuk et al. [2008]. We also ran a 100 kyr long control experiment ("CTRL"), in which no atmospheric forcing (or modification of bioturbation or local carbonate saturation) was applied.

Results
The reconstructed sedimentation rate for both sites is shown in Figure 3, plotted with sediment lightness, alongside core photos [Zachos et al., 2004a[Zachos et al., , 2004b. Sedimentation rate at the shallow site was approximately twice the rate at the deep site, with preevent (<20 ka) deposition at Site 1263 averaging 1.96 cm/kyr, compared with 1.08 cm/kyr at Site 1262. From preevent to peak-event values, the former sees an~12-fold drop in sedimentation rate, the latter a 10-fold decrease.
As noted but not explained by Stap et al. [2009], bulk stable isotope records for the two sites show clear differences (Figures 2, 4a, and 4b). At the deeper site, bulk δ 13 C values exhibit a gradual decline and then recovery across ETM2. The shallow site, however, also shows a gradual decline/recovery for the start/end of the event but exhibits an additional excursion during the peak phase (ETM2 horizon;~38-56 ka). The bulk δ 18 O record similarly shows a greatly enhanced difference within the ETM2 horizon, with much more Shown are the combination of assumptions of (i) interface dissolution model (otherwise homogeneous); (ii) bioturbation of upper sediment layers (otherwise no vertical mixing); and (iii) reduced carbonate saturation, simulated by increasing the pressure used in calculation carbonate saturation by the equivalent of 2000 m, that are applied to each of three separate phases of the total 100 kyr of model simulation.

10.1002/2015PA002821
negative values at the shallow site, although surface dwelling Acarinina records do not follow this trend [Stap et al., 2010b]. Intermediate sites along the depth transect, 1265 and 1267 (not shown), are similar to the deep site [Stap et al., 2009].
The records of sedimentary CaCO 3 content ( Figure 4c) share some features of the bulk stable isotope records [Stap et al., 2009], in as much as the minimum in CaCO 3 wt % at 1263 occurs over a much shorter interval than at 1262, although the minimum CaCO 3 wt % values at both sites are similar. Fragmentation (Figure 4h and Table 2) increased at both sites during ETM2, with the pattern largely mirroring that of CaCO 3 wt % but noisier, with the increase in fragmentation more gradual and longer lasting at the deep site. Similarly, CaCO 3 MAR patterns broadly follow the trend of CaCO 3 content, but the CaCO 3 MAR was higher at the shallow site by a factor of 1.5-2.0 before and after ETM2. Patterns in coarse fraction (i.e., planktic foraminiferal) MAR, susceptible to dissolution and thus indicative of corrosiveness, resemble CaCO 3 MARs, and thus FF MAR, despite its minor contribution to the sediment (Figure 4e and Table 2). Planktic foraminiferal accumulation rates (PFARs- Figure 4g) at both sites were identical prior to the events but differed during the interval when CaCO 3 wt % remained even at Site 1263 while declining at 1262. The differential changes between the foraminiferal and the coccolithophore response result in relative increases in the foraminiferal contribution to the bulk carbonate at the shallow site between about 16 and 38 ka (Figure 4f and Table 2), the interval just before the peak of ETM2.
Benthic foraminiferal parameters generally resemble sedimentary records. Benthic foraminiferal accumulation rates (BFARs- Figure 5a) at each site were similar before and after the event, with overall slightly higher values at the shallower site. At both sites, BFAR started to decline gradually at the start of ETM2 but more pronouncedly at the deeper site. During start and recovery, the difference in BFAR between the two sites was significantly higher than during background conditions, but during peak ETM2  Figure 3. Photographs of sections across the ETM2 event in two cores, from Site 1262 and Site 1263, from Walvis Ridge ODP Leg 208, and the respective approximate sedimentation rates for the two sites, as well as sediment lightness (color [Zachos et al., 2004a, 200b]).  Table 4). Samples were essentially barren of benthic foraminifera [Stap et al., 2010b]. All species-specific ARs declined ( Figure S1 in the supporting information and Table 5). No species bloomed during ETM2: all declined in abundance, though some more than others. The diversity (rarefied number of species) declined parallel to BFAR, with largest differences between the sites during the start (and recovery) phase when values at the shallower site remained relatively high while those at the deep site had started to decline (and had not yet recovered) (Table 4).
Benthic foraminiferal assemblages during background conditions were diverse, with 133 taxa recognized, 117 at the deep site (18 not present at 1263), and 116 at Site 1263 (17 not present at Site 1262) (Data Set S1 in the supporting information). The number of species (rarefied to 100 specimens) was higher at the shallow site (background values of 41 species) than at the deep site (33 species) (Figure 5b), largely due to the presence of diverse species of Lenticulina and other lagenid species. Nuttallides truempyi was the most common species at both sites, with Oridorsalis umbonatus, Quadrimorphina profunda, Bulimina kugleri, and Bulimina simplex (Table 3). Species present at the shallow site only include Cibicidoides alleni and C. laurisae  During ETM2, Nuttallides truempyi and N. umbonifera increased in relative abundance (Figure 5c) at the deep site, as did Abyssamina poagi, Globocassidulina subglobosa, and Cibicidoides species, whereas Tappanina selmensis and Siphogenerinoides brevispinosa, probably opportunistic infaunal taxa [Steineck and Thomas, 1996;Thomas, 1998Thomas, , 2003, decreased in relative abundance (Figures 5d, 5f, and 5g). Epifaunal species thus overall increased in relative abundance at the deep site, and infaunal species decreased (Figure 5e) during the full duration of ETM2. In contrast, at the shallow site, infaunal taxa as a whole and the generally infaunal buliminid and cylindrical species remained equal or increased somewhat in relative abundance during the start of ETM2, so that the difference in relative abundances at the two sites increased (Figures 5e, 5h, and 5i). A similar difference developed during the recovery phase. In addition, the shallow infaunal Oridorsalis umbonatus [Thomas and Shackleton, 1996] increased in relative abundance at the shallow site just after the peak event (Figure 5j). During the peak event benthic foraminifera were essentially absent at Site 1263. After the peak event and extending after the recovery phase, agglutinant taxa at the deep site remained less abundant, as did Siphogenerinoides brevispinosa. Epistominella exigua became more common after ETM2, and Quadrimorphina profunda did so at the shallow site (e.g., Figures 5g and 5k).

Discussion
The biotic and sedimentary records across ETM2 at Walvis Ridge are striking in their similarity. Both record more gradual change at the deeper site, with a generally more extreme and much shorter superimposed change during the peak of the event at the shallower site only. If CO 2 addition and associated decline in carbonate saturation alone were driving the sedimentary observations, we would have expected a sharper wt % CaCO 3 response at 1262 compared to 1263 because of the lower initial saturation and hence lower fractional carbonate preservation at greater depth [Stap et al., 2009]. Assuming a similar ocean acidification (carbonate ion decline) at all depths, the nonlinear nature of the wt % CaCO 3 scale means that at lower initial wt % CaCO 3 , only a relatively small decline in carbonate preservation is needed to produce a large change in wt %. Instead, we observe the opposite, i.e., a sharper response at the shallow site, which starts at higher wt % CaCO 3 . There is no indication of unconformities (Figure 3) [Stap et al., 2009] bracketing the ETM2 horizon and hence no indication of a removal of most of the onset and recovery at Site 1263 to explain the sharp transitions. We also discard the possibility of sampling biases. Stap et al. [2009] sampled 1262 more closely spaced (at 0.5 cm) within the ETM2 horizon. Hence, wt % CaCO 3 , δ 13 C, and δ 18 O (Figures 4a-4c) measurements are more closely spaced in time at the deep site, implying that the smoother bulk carbonate composition trends at 1262 cannot be due to a sampling artifact. If anything, the less frequent sampling in depth (thus time) across the ETM2 horizon at 1263 could have underestimated the abruptness of the transition into and out of peak ETM2 conditions. We also rule out sampling differences as an explanation for the absence of benthic foraminifera at 1263. Unlike the bulk carbonate records, sampling for benthic foraminiferal analysis was regular (2 cm). The difference in sedimentation rates, which prior to ETM2 averaged 1.08 cm/kyr at the deep site compared with 1.96 cm/kyr at the shallow site, hence leads to a higher frequency in time of sampling at 1263 versus 1262. It is extremely unlikely that the presence of foraminifera during the ETM2 horizon was missed in the more frequently sampled (in time) core. In general, the most diverse benthic assemblages, with cooccurring epifaunal and infaunal dwellers, are indicative of intermediate food availability. When little particulate organic carbon arrives at the seafloor, there is insufficient food to sustain infaunal populations, and at extreme food abundance, oxygen levels in pore waters (and finally in bottom waters) become too low to sustain infaunal populations [Jorissen et al., 2007]. The relative abundance of infaunal taxa thus is a proxy for increased food supply and/or declining oxygen levels.
At the deep site, BFAR as well as relative abundance of infaunal taxa (buliminids and cylindrical taxa) declined gradually to reach the lowest levels for that site during the peak event, before increasing again ( Figure 5 and Tables 4 and 5). Relative abundances of N. truempyi and N. umbonifera, indicative of undersaturated bottom waters and/or oligotrophic conditions [Bremer and Lohmann, 1982;Thomas, 1998], increased, as did that of the abyssaminids. The latter are extinct but were generally more abundant at greater depths [e.g., Thomas, 1998], thus probably indicative of oligotrophic conditions. All benthic foraminiferal indicators point to a declining food supply to the seafloor during ETM2 at Site 1262. In contrast, calcareous nannofossil evidence for nearby Site 1265 does not indicate significant changes in productivity in the region [Dedert et al., 2012].
Can indicators of relatively unchanging surface productivity be reconciled with an interpretation of declining benthic food supply? It is unlikely that the strong decrease in BFAR and diversity is driven by taphonomic dissolution only, because the proportion of Abyssamina poagi, a small, smooth, dissolution-prone taxon, increased during peak ETM2, i.e., maximum dissolution (Figure 5c), whereas dissolution would have led to a relative increase in relatively large, heavily calcified taxa [e.g., Nguyen et al., 2009;Nguyen and Speijer, 2014]. Instead, we suggest that temperature changes associated with ETM2 are key. Higher temperatures influence biological processes [Pörtner et al., 2014] due to their effect on enzyme reactions, diffusion, and membrane transport [Hochachka and Somero, 2002], increasing metabolic rates  e.g., demonstrated for the Eocene of offshore Tanzania on the basis of reconstructed water column δ 13 C gradients [John et al., 2014]. Increased metabolic rates combined with increased remineralization of organic matter in the water column lead to a lesser arrival of food at the seafloor despite constant productivity [Ma et al., 2014] and could, coupled with the highly food-limited nature of benthic foraminifera in today's oceans [Linke, 1992], explain the strongly reduced BFARs.
Faunal changes were more complex at the shallower site, despite the fact that the sites are relatively close to each other and hence under waters with similar primary productivity [Zachos et al., 2004a[Zachos et al., , 2004b. Whereas BFAR, species diversity, and buliminid taxa all decreased simultaneously during the early and recovery phases of ETM2 at the deep site, the relative abundance of buliminid taxa at the shallow site increased, despite decreasing BFAR and species diversity ( Figure 5). Several infaunal taxa decreased in relative abundance at Site 1263 during the peak ETM2 (e.g., S. brevispinosa and T. Selmensis; Figures 5f and 5g), suggesting that these taxa were less able to survive the lowered food supply at this site, indicated by the more severe drop in BFAR, than other infaunal taxa such as buliminids. In contrast, these species decline similarly to buliminids at Site 1262. During the recovery phase, the relative abundance of the shallow infaunal O. umbonatus increased; this increase was likely not caused by an increase in food supply, because the BFAR remained low relative to preevent values. Buliminid taxa and O. umbonatus % increased in relative abundance just prior to the peak event (20-40 kyr). Both calcify in the less saturated pore waters rather than in bottom waters, so the increase might have been caused by increasing undersaturation [Foster et al., 2013], but this does not agree with the observation that at the shallow site, the carbonate parameters (CaCO 3 wt %, fragmentation, and PFAR) remained constant during the interval with increased abundance of buliminid taxa. This increased abundance of buliminid taxa and O. umbonatus during declining food levels and invariant carbonate corrosiveness thus indicates that oxygenation was declining in bottom and/or pore waters at the shallow site, possibly due to rising temperatures, increased remineralization of organic matter, or changes in preformed oxygen levels due to changes in ocean circulation pattern.
During the peak phase of ETM2, benthic foraminifera were absent at Site 1263, indicating that bottom and pore water conditions could not support them, and were less favorable than at the deeper Site 1262 where benthic foraminifera remained present. Deoxygenation was more severe and persisted longer at the  [Chun et al., 2010;Pälike et al., 2014] and mineralogical data [Post et al., 2015]. A similar occurrence during ETM2 would help explain the differential benthic assemblage changes between Sites 1262 and 1263. Benthic foraminiferal records during the PETM cannot be compared between the sites because of the severe carbonate dissolution during the peak PETM, with CaCO 3 fully dissolved for part of the PETM at all sites, longer at the deeper sites .

Ocean Circulation as a Driver of Depth-Specific Ecological Change
We interpret our observations in terms of a change in the source of intermediate waters bathing Site 1263, driving a much larger warming and decrease in dissolved oxygen compared to 1262. Support for this comes from the results of Paleocene/early Eocene fully coupled atmosphere-ocean climate general circulation model experiments [Lunt et al., 2010]. These experiments demonstrate that an atmospheric CO 2 and surface warming threshold could exist, beyond which any further CO 2 rise and surface warming lead to a disproportionately larger increase in temperature increase in the intermediate waters than in the deep ocean. For instance, in the simulations of Lunt et al. [2010], going from 2 × PAL to 6 × PAL CO 2 , where PAL is 280 ppmv, produced a warming of 1.7°C at 1500 m compared to 0.2°C at~3500 m ( Figure 6). All other things being equal, a change in water mass source and/or mixing that leads to higher local temperatures will be associated with lower dissolved O 2 , although the specific pathway and hence integrated remineralization of organic matter along that pathway will also affect the local value of [O 2 ].
A change in circulation during ETM2 has also been suggested by d' Haenens et al. [2014], inferred from a shortlived reversal of meridional δ 13 C gradients of 0.50-1.00‰ between the North and South Atlantic (Deep Sea Drilling Project Sites 401 and 550, NE Atlantic and the Walvis Ridge sites, respectively). Similarly, ocean circulation change has been inferred at Site 1263 during the PETM as implied by the largest CIE (À3.5‰) in deep-sea benthics, although comparison with the other sites is not possible due to the severe dissolution [McCarren et al., 2008]. Direct evidence for a circulation change-driven warming does not yet exist however. Although a 3°C warming during ETM2 was estimated from benthic foraminifera at 1262 [Stap et al., 2010a], the relative temperature change at the shallower site is not recorded due to the absence of benthic foraminifera during the critical interval.

Origins of the "Anomalous" Bulk Sediment Response During Peak ETM2 Conditions
We suggest that the Site 1263 phenomena: (1) a sharp excursion in wt % CaCO 3 together with bulk carbonate δ 13 C and δ 18 O that constitutes the ETM2 horizon and (2) temporary exclusion of benthic foraminifera are causally linked, via the impact of changes in the benthic foraminiferal contribution to bioturbation [Grosse, 2002]. We infer that sediment mixing by benthic foraminifera would have effectively ceased at the shallow site during the peak of ETM2. Changes in bottom water conditions would have also affected other benthic biota (including burrowers) because animals are more severely affected by deoxygenation than protists such as foraminifera [Gooday et al., 2010]. Surface sediment mixing thus may have ceased during the peak of ETM2 at the shallow site, but not at the deep site, as may be seen in the core photographs, and in the larger and more abrupt change in sediment color (lightness) (Figure 3). This is important, as mixing reduces the recorded magnitude and increases the apparent duration of a signal [Ridgwell, 2007;Kirtland Turner and Ridgwell, 2013]. Indeed, numerical modeling of hyperthermal events illustrates that a sharper onset to low carbonate content sediments is observed in the absence of bioturbation [Ridgwell, 2007;Kirtland Turner and Ridgwell, 2013]. An enhanced degree of carbonate dissolution in the ETM2 horizon at 1263 might also have played a role, as the temporary emplacement of a less well ventilated intermediate water mass would be expected to have higher respired dissolved CO 2 concentrations and hence lower saturation. We turn to the Earth system modeling experiments (Table 1) to explore this further.
We first test whether the assumed atmospheric perturbation (Figure 7a) can produce a sediment record consistent with observations from the deep site, where we expect a relatively straightforward and predictable response to ETM2 ocean acidification. In experiment "STD" (Table 1), we simulate a reduction in carbonate content to around 50 wt % in response to increasing atmospheric CO 2 followed by an initially more rapid recovery (Figure 5b), qualitatively consistent with trends observed at Site 1262. Toward the end of the simulation, the modeled sediment record displays an "overshoot" in carbonate content which is also expected [Dickens et al., 1997 . It is assumed that the temperature is that of the 2 × PAL simulation between 0 and~40 kyr, then (following a circulation switch) the temperature can be characterized by that of the 6 × PAL simulation for a period of~40 kyr, then the temperature reverts to the preswitch state. The warming following the circulation switch is greater at intermediate depths than in the deep ocean.  Table 1). Paleoceanography 10.1002/2015PA002821 example, it occurs due to a forced removal of CO 2 from the atmosphere (Figure 7a) rather than via an explicit calculation of silicate weathering feedback [Colbourn et al., 2013]. Carbonate δ 13 C (Figure 5b) exhibits an excursion size slightly less than the applied À1.5‰ magnitude of the forcing (Figure 5a), also as expected [Kirtland Turner and Ridgwell, 2013]. However, the δ 13 C minimum lags that of wt % CaCO 3 by about 10 kyr, whereas in the Site 1262 observations (Figures 4a and 4c), they are approximately synchronous. In experiment "ALT," we hence substitute a homogeneous carbonate dissolution model for the default interface assumption [Ridgwell, 2001], so that newly deposited carbonate is mixed into the surface sediment layer before carbonate is removed through dissolution. This brings the δ 13 C and wt % CaCO 3 minima into alignment (Figure 7c), producing a better match to the Site 1262 observations. (In the interface model of carbonate dissolution, a δ 13 C signal from the surface cannot be imprinted on the sediments once the total dissolution flux exceeds the rain flux.) However, little change in wt % CaCO 3 is recorded at the analogue location to Site 1263 (Figure 7d). In addition, the simulated δ 13 C record at 1263 is too regular and exhibits none of the abrupt transitions characterizing the observed transition into and out of the ETM2 horizon (Figures 3 and 4). In experiments "ALT_bio," "ALT_sat," and "ALT_satbio," we hence explore the possible impact of reduced carbonate saturation, reduced bioturbation, and both together ( Table 1).
The temporary cessation of sediment mixing on its own (ALT_bio) at 1263 does little more than introduce small step-like features in the simulated evolution of bulk carbonate δ 13 C (Figure 7d), with little noticeable impact on wt % CaCO 3 . In contrast, temporarily decreasing carbonate saturation on its own (ALT_sat) reduces wt % CaCO 3 toward observed Site 1262 values (Figures 7c and 7f). The transitions in bulk composition occur relatively rapidly, to create a simulated feature more reminiscent of the ETM2 horizon ( Figure 4c). Combining both temporary saturation decline and cessation of bioturbation (ALT_satbio) leads to a more sharply defined wt % CaCO 3 anomaly, particularly with respect to the transition into the peak of the event (Figure 7g). However, only small steps occur in δ 13 C.
Although not successful in reproducing all the observations, these simple experiments reveal the potential processes associated with specific sedimentary features. First, we find that a change in water mass saturation appears to be key to reproducing the magnitude of anomalous decline and recovery in wt % CaCO 3 at Site 1263. That said, we cannot rule out the possibility that the GENIE model does not exhibit an appropriate sensitivity of carbonate preservation to CO 2 addition, particularly as a function of ocean depth. Although outside the scope of this particular paper, the model response to ETM2 could be assessed by contrasting the changes in CaCO 3 across the event (Figures 1a and 1b) at multiple sites spanning different ocean basins (e.g., as in Panchuk et al. [2008]) and the applied forcing refined, perhaps by means of formal inversion [Kirtland Turner and Ridgwell, 2013]. In contrast to reduced saturation, the importance of bioturbation is apparent in dictating the details of the recorded shape of the signal. Only by stopping mixing (bioturbation) between model sediment layers can a sharp decline at the onset of the ETM2 horizon be reproduced. In our model (experiments ALT_bio and ALT_satbio) bioturbation is switched fully back on at 65 ka, and the consequential transition in wt % CaCO 3 is comparatively gradual. The BFAR record ( Figure 5a) suggests a more drawn-out recovery of the benthos and attendant gradual increase in the intensity of bioturbation. If implemented in the model, we would expect a sharper transition at the end of the ETM2 horizon, closer to observations. If the above analysis is correct and the attributes of the ETM2 horizon at Site 1263 are primarily driven by a local circulation change and its associated benthic ecological impact, this creates a challenge for understanding when these additional effects occurred relative to the primarily CO 2 -driven carbonate dissolution at greater depth. In our age model, we adopt the same tie point as Stap et al. [2009] to define the start of ETM2 (0 ka in, e.g., Figure 2). This places the required circulation change at Site 1263 approximately coincident with peak ETM2 conditions at 1262. If we shifted the record for 1263 older by one precession cycle instead, the circulation change would occur close to the ETM2 onset at 1262. This would be a plausible alternative alignment, particularly if the carbon release was rapid.
Finally, our failure to explain the full magnitude of observed δ 13 C changes at Site 1263 (and not explored here in the model-also of δ 18 O) is more difficult to account for. We thus do not rule out that differential dissolution or diagenetic alteration might explain some of the observed disparity in bulk carbonate proxy responses between sites. However, the carbon isotope signals in marine carbonate are generally thought Paleoceanography 10.1002/2015PA002821 to not be significantly affected by diagenesis [Sexton et al., 2006] and burial depth [Schrag et al., 1995]. Furthermore, differences between the bulk δ 13 C values at the two sites are not likely caused by differences in the nannoplankton assemblage composition, because vital effects are minor [Ziveri et al., 2003] and assemblages at the sites similar Raffi and De Bernardi, 2008;Dedert et al., 2012]. Although there is some variation in the relative proportions of the CF (foraminifera) and FF (nannoplankton) during the ETM2, the ratio continues to be dominated by calcareous nannofossils (Figure 4d). For δ 18 O, recrystallization [Schrag et al., 1995] could potentially imprint a component of bottom water temperature at Site 1263, but this would imply that the 1262 record reflects extensive recrystallization because of the lower bottom water temperatures at that site. This seems rather unlikely as burial depths were on the order of 100 m only at Site 1262, i.e., much less than the >300 m at Site 1263, and recrystallization should have been much less pronounced at the deeper site [Zachos et al., 2004a]. None of these diagenetic-based explanations are thus particularly compelling.

Conclusions
During the ETM2, Walvis Ridge Sites 1263 and 1262 both record a δ 13 C excursion, warming, and evidence of ocean acidification. The benthic foraminiferal ecosystem was perturbed in response to environmental change during the ETM2, with a decrease in abundance, diversity, and assemblage change at both sites. However, a more severe benthic response occurred at the shallow site, resulting in the temporary absence of benthic foraminifers. We infer that this was caused by more pronounced intermediate water warming, leading to effective decline in food supply and deoxygenation driven by a circulation change. This in turn led to a cessation of bioturbation and a possible accentuation of the sedimentological record of the event at 1263. We used a simple conceptual carbon forcing model for a temporary cessation of sediment mixing plus a decrease in carbonate saturation associated with changing intermediate water mass properties. Using this model, we can qualitatively account for the bulk sediment and carbon isotopic observations at both sites. However, a full explanation for the greater magnitude of recorded isotopic excursion at Site 1263 remains to be identified. Our study illustrates that the biotic response to a global change event can be highly spatially heterogeneous and not necessarily scale simply with the magnitude of the event. Instead, the effects of increased atmospheric CO 2 can lead to ocean circulation change and other feedback that create a far more complex picture of the influence of climate change on biota. In turn, changes in biota can distort the sedimentary proxy record.