The MIS 5, as the last warmest and longest interglacial of the past 500 ka (Howard, 1997), with an abrupt glacial–interglacial transition (Howard, 1997; Müller and Kukla, 2004; Sirocko et al., 2005) is considered to be a good analogue for the actual warm Holocene (Howard, 1997; Kukla, 1997) and even a partial analogue for +1–2 ∘C scenarios because of polar temperatures 3 to 5 ∘C warmer than today and a sea level about 6.6 m higher than today (Clark and Huybers, 2009; Kopp et al., 2009). During this interglacial, Chalk et al. (2019) observed a change in the current circulation in the Atlantic Ocean, with an enhanced Antarctic Bottom Water (AABW) below 3400 m b.s.l. and well-ventilated, high-pH, and high [CO32-] water mass around 2200 m b.s.l. They also highlighted a correlation between the [CO32-] and the pCO2 in the West Atlantic during cold intervals, with an increase of the volume of the high dissolved inorganic carbon (DIC), low [CO32-] deep-water masses in the North Atlantic.

The MIS 9 in the equatorial Atlantic presents well-preserved sediment at a period known to be under high obliquity with a unique insolation signal. Stable oxygen isotope values are low during this period (low ice volume). It is one of the interglacials showing the highest pCO2 (around 300 ppm) and pCH4 (around 25 ppb) conditions in the late Pleistocene. This period is also one of the warmest, stablest and shortest interglacials, with a weak surface water ventilation (Past Interglacials Working Group of PAGES, 2016; Marino et al., 2014; Voelker et al., 2010).

The Pliocene warm period (PWP) MIS KM5 corresponds to a period with a similar orbital forcing to present day and an insolation distribution close to the modern one (Haywood et al., 2013). This interval (3.264–3.025 ka) is also described as a negative oxygen isotope slope and a sea level 21–23 m above the present-day one (Lunt et al., 2008, 2010; Naish et al., 2009; Pollard and DeConto, 2009) with a well-ventilated deep Atlantic Ocean (Bell et al., 2015). The temperature is 3 ∘C higher than pre-industrial values (Haywood et al., 2000; Lunt et al., 2010) and the CO2 concentration is close to the present value, i.e. 330–425 ppmv during the warm interglacials (Pagani et al., 2010; Seki et al., 2010), making it a good analogue for future climate (Ravelo and Wara, 2004) and an important period to understand the climate system (Lunt et al., 2010). Furthermore, this period is also described as being wetter than today (Leroy and Dupont, 1994; Dodson and Macphail, 2004), but the latitudinal distribution of the rainforest was close to the present day distribution (Salzmann et al., 2011).

The MCO corresponds to a period with an eccentricity-modulated precession δ18O signal and low global ice volume, featuring a Northern Hemisphere that is free of continental ice sheets, important 100 and 400 kyr orbital variability, and an Antarctic ice sheet that is smaller but more dynamic than today (De Vleeschouwer et al., 2017; Holbourn et al., 2007). Haq et al. (1987) highlighted the large sea level amplitude from 16 to 14 Ma, and the annual global temperature was 3 to 8 ∘C higher than today (Pound et al., 2012; You et al., 2009). The climate during the MCO is known to be correlated with atmospheric CO2 concentration changes (Foster et al., 2012), with CO2 concentration being generally lower than at present (Foster et al., 2012; Ruddiman, 2010; Zachos et al., 2008; Zachos, 2001a, b) but peaking at 16 Ma between 460 and 564 ppmv (Kürschner et al., 2008). Even if a decline in δ18O and δ13C at 16.9 Ma was suspected to be linked to increase of carbonate dissolution, a sign of strong changes in the carbon cycle (Holbourn et al., 2015), carbonate production appears to have been the main control of the CaCO3 record (Liebrand et al., 2016).

Ceara Rise, located in the equatorial Atlantic Ocean, represents an ideal location to quantify the variability in tropical Atlantic pelagic carbonate production since the Miocene. This aseismic ridge rises several kilometres above the surrounding abyssal plain, well above the modern regional lysocline, located between 4100 and 4200 m b.s.l. (Frenz et al., 2006; Gröger et al., 2003a, b; Curry et al., 1995; Cullen and Curry, 1997; Bickert et al., 1997). The ridge is bathed by the shallower North Atlantic deep water (NADW) and the deeper Antarctic bottom water (AABW) (Rühlemann et al., 2001; Gröger et al., 2003b; Herrford et al., 2017), and the interface of the two water masses corresponds to the regional lysocline depth. Around the ridge, the average depth of the seafloor is at 4500 m b.s.l., but the Ceara Rise ridge rises by as much as 1900 m above the surrounding abyssal plain, with its top reaching the depth of 2600 m b.s.l. (Curry et al., 1995). This provides an opportunity to sample pelagic sediments that are largely unaffected by dissolution, and their accumulation therefore mainly reflects changes in pelagic carbonate production as suggested by Brummer and van Eijden (1992). The Ceara Rise (Fig. 1) has been visited by Ocean Drilling Program (ODP) Leg 154 (Curry et al., 1995), recovering a transect of sediment sequences ranging into the Eocene that are rich in carbonate and show prominent cycles due to variable input of clastic material from the Amazon fan (Shackleton et al., 1999; Bickert et al., 1997; Shackleton and Crowhurst, 1997). The cycles are reflected in sediment physical properties, such as colour or magnetic susceptibility, and because of the very good recovery and repeated coring at the same sites, continuous spliced records could be produced that facilitated the development of orbitally tuned age models (Shackleton et al., 1999; Zeeden et al., 2013; Wilkens et al., 2017; Shackleton and Crowhurst, 1997), a prerequisite for the quantification of carbonate accumulation. Since all high-resolution Neogene records of carbonate accumulation (Drury et al., 2021; Lyle et al., 2019), including those from the Ceara Rise (Curry et al., 1995; King et al., 1997) show a large orbital-scale variability, hinting at prominent orbital-scale variability in pelagic carbonate production, next to a compilation and re-evaluation of existing carbonate records, the selected time slices had to be newly sampled and analysed at higher resolution.

The combination of the availability of high-resolution age models and good carbonate preservation make the Ceara Rise a model region to study pelagic carbonate production and preservation. We compiled existing data on carbonate content (CaCO3 %) at all the Leg 154 sites since the Miocene (Curry et al., 1995; Frenz et al., 2006; King et al., 1997) and used those to calculate carbonate accumulation rates (CaCO3 AR). The few other existing datasets on carbonate content from the Ceara Rise sites (e.g. Tiedemann and Franz, 1997) could not be used because some of the information needed to calculate accumulation rates or the original samples ID and depths was not available.

The carbonate content data were combined with dry bulk density (DBD) and sedimentation rate (SR) to calculate the CaCO3 AR as follows. CaCO3AR=(CaCO3%/100)×DBD×SR Following the approach by Lyle (2003), we first derived for each site a calibration between the gamma ray attenuation (GRA) bulk density and DBD using data from Curry et al. (1995). The resulting site-specific calibrations (Fig. S1) were then applied on GRA bulk density values, which were extracted from Curry et al. (1995), and interpolated to the position of the analysed samples using linear interpolation. This yielded DBD values between 0.40 and 1.64 g cm-3. For two samples, the calibration returned negative DBD (at 129.62 and 135.47 mcd) due to two anomalous GRA values below 1. In these cases, we used the DBD of the nearest point instead.

We sampled the record at Site 927 at high resolution for the four periods of interest (Fig. 2), making sure that for each interval both the interglacial and the flanking glacial in the Quaternary and at least two full eccentricity cycles during the Pliocene and Miocene have been covered. These four intervals cover a large range of global temperature and CO2 values (Fig. 2). We selected Site 927 because it is one of the two shallow sites of Leg 154, located well above the lysocline at present (Frenz et al., 2005; Curry et al., 1995; Bickert et al., 1997), and because numerous palaeoceanographic datasets and carbonate measurements exist for this site (e.g. Pälike et al., 2006a; Bickert et al., 1997; Frenz et al., 2006; Gröger et al., 2003b; King et al., 1997; Curry and Cullen, 1997). The site appears generally less affected by slumps or turbidites than the four others, which were not observed in the four studied intervals (sampled out of the slumps and turbidites reported lithostratigraphic units) (Curry et al., 1995). The sampling was guided by the Wilkens et al. (2017) age model for the samples from 0 to 14 Ma and by the Shackleton et al. (1999) age model for the samples from 14 to 16.5 Ma. Considering the typical mixing depth of -10 cm in deep-sea sediments, we sampled at 5 cm in the Quaternary and 10 cm in the Neogene, which in both cases provides sub-orbital resolution. The resolution was higher in the Quaternary because the peak interglacial warmth periods are short (<10 kyr for MIS 5e; Stolz and Baumann, 2010; Müller and Kukla, 2004; Sirocko et al., 2005), and we wanted to cover these using multiple samples. In total, we collected and analysed 139 samples for the two Quaternary intervals, 72 samples for the Pliocene and 50 samples for the Miocene.

We performed stable isotopes analyses (δ18O and δ13C) at Bremen University, using a Thermo Fisher Scientific MAT 253plus gas isotope ratio mass spectrometer with Kiel IV automated carbonate preparation device. This gives δ18O values with a standard deviation of house standard (Solnhofen limestone) over a measurement period of 0.07 ‰ and δ13C values with standard deviation of house standard (Solnhofen limestone) over measurement period of 0.03 ‰. The sediment samples were washed and sieved at 63 µm using tap water and dried overnight in the oven at 50 ∘C. Then, they have been dry-sieved at 150 µm for benthic foraminifera picking. All the Miocene samples have been picked, and only three samples did not have enough material to run the stable isotopes analyses. For some of the samples we had enough material to analyse two or three replicates using different species known to be relevant markers for δ18O seawater: Cibicidoides mundulus, Cibicidoides wuellerstorfi and Oridorsalis umbonatus (Katz et al., 2003; Rathmann and Kuhnert, 2008). We did not mix the species in one single measurement. For the species-specific δ18O and δ13C values correction, we used the calibration given in the Supplement Table S3 from Westerhold et al. (2020).

To determine the CaCO3 AR for the newly sampled intervals, we performed carbonate content analyses on the bulk sediment using a LECO CS744 elemental analyser at Bremen University. The analysis was performed by heating 0.1 g of homogenized material in a ceramic dish and measuring the resulting CO2 in IR cells. The carbonate content has been calculated as the difference between the total carbon content and the organic carbon content, measured in a second sample that was pre-treated with hydrochloric acid to remove carbonates. Both measurements have an accuracy of 0.001 mg (1 ppm) or 0.5 % relative standard deviation (RSD). The inorganic carbon was then converted to carbonate content using the molecular mass of calcium carbonate. Dry bulk density for all the newly analysed samples at Site 927 was determined from GRA bulk density as described above (Sect. 2.2.) and combined with the carbonate content and SR from the modified age models to calculate the CaCO3 AR.

Using existing carbonate content data for all Leg 154 sites (Curry et al., 1995; Frenz et al., 2006; King et al., 1997), combined with new age models (Wilkens et al., 2017), for each site, records of CaCO3 AR since the mid-Miocene were calculated (Fig. 3). Curry et al. (1995) noted the occasional presence of slumps or hiatuses in the sediment sequences, especially at Site 928 and Site 929. Here we used the age models for the entire sediment package, ignoring the presence of these events. This is because the slumps only represent a small fraction of the sediment sequence and therefore are unlikely to affect the overall trends.

The mean CaCO3 AR varies considerably among the sites, reflecting their depth and therefore likely the amount of dissolution. ODP Sites 925 and 927 (present depth 3041 and 3315 m b.s.l.) show consistently higher CaCO3 AR (between 1.5 and 3 g cm-2 ka-1) than the three remaining sites, located below 3400 m b.s.l. (around 1 g cm-2 kyr-1). Curry and Cullen (1997) show an effect of distance from the Amazon Fan on sediment composition on Ceara Rise for the late Quaternary, but this change is only manifested as differences in the AR of terrigenous (non-carbonate) sediments. This is seen in patterns of carbonate content of the sediment (their Fig. 2) but not in changes in carbonate accumulation. Also, there is little evidence that the Amazon discharge plume reaches far enough offshore to induce changes in productivity over the plateau. At present, the discharge is strongly deflected northwards and stimulates productivity mainly along a narrow coastal strip (Gouveia et al., 2019). The same authors note that some of the Amazon discharge may be deflected into the North Brazil Current, but this affects productivity only little and mainly north off the Ceara Rise. To visualize long-term trends, we subtracted at each site the mean values of CaCO3 AR and SR (Fig. 4). All sites show a prominent trend of increasing SR, beginning in the late Miocene (8 Ma ago) (Fig. 4), which is known to reflect increasing amount of clastic material transported from the Amazon Fan (Curry et al., 1995; Pälike et al., 2006b; Bickert et al., 1997; Harris et al., 1997; Shackleton and Crowhurst, 1997). The CaCO3 AR, on the contrary, shows a less obvious temporal trend on a long timescale (Fig. 4), indicating that the increase in SR is compensated by decreased carbonate content in the sediment. Instead, the CaCO3 AR record at all Ceara Rise sites show a pervasive short-term (likely orbital) variability, with substantial magnitude (Curry et al., 1995).

During the Pleistocene, the CaCO3 AR at Site 927 was driven only by the carbonate content, indicating that the signal is affected by dissolution. This is confirmed by the presence of very low values of carbonate content and CaCO3 AR during the cold intervals in the Pleistocene, in phase with the δ18O and insolation signal, indicating a relationship to changes in deep-water circulation, confirming the conclusions by Bickert et al. (1997). As expected from the overall stratigraphy and palaeoceanography of the Ceara Rise sites (Curry et al., 1995; Frenz et al., 2006; King et al., 1997), the new carbonate content (Fig. 8e) and CaCO3 AR (Fig. 8c) records from Site 927 show strong minima during cold intervals of the Quaternary (indicated by δ18O record), which is consistent with the shoaling of the corrosive AABW (Miller et al., 2012; Harris et al., 1997), causing dissolution at shallower depth (Gröger et al., 2003a, b). In contrast, the maxima in CaCO3 AR and carbonate content in the sediment during the Quaternary do not appear to be affected by dissolution. Therefore, whilst we cannot use the Quaternary variability in the CaCO3 AR to estimate the orbital-scale variability in pelagic carbonate production, we can use the interglacial maxima (Fig. 10) to estimate pelagic carbonate production during the Quaternary, assuming that during the studied Pleistocene interglacials Site 927 was positioned above the lysocline, as is the case during the Holocene.

Because the Ceara Rise sites became periodically affected by the more corrosive Antarctic bottom water only after the initiation of the North Hemisphere glaciation (Liebrand et al., 2016; Harris et al., 1997; Pälike et al., 2006a), the studied Pliocene and Miocene intervals should not be affected by dissolution. Paul et al. (2000) note that the exact subsidence history of the Ceara Rise is unknown but assume minimal subsidence since the early Miocene. Similarly, sea level differences among Quaternary interglacials and the Pliocene and Miocene were likely on the order of tens of metres. Therefore, the largest changes in palaeodepth would have been due to sediment cover, which would make the studied mid-Miocene interval about 300 m deeper compared to the present one (this depth is still above the present day lysocline depth of 4200 m b.s.l.). Throughout the entire studied interval since the Miocene (Fig. 3), the shallowest cores (925 and 927) record higher CaCO3 AR values than the deeper ones. This also indicates that these sites likely remained above the lysocline (Curry et al., 1995; Bickert et al., 1997; Frenz et al., 2006; Gröger et al., 2003a, b) and that the CaCO3 AR signals recorded at these sites primarily record changes in pelagic carbonate production. To provide further support for the lack of dissolution control on the pre-Quaternary variation in CaCO3 AR, for the Pliocene and Miocene interval we generated new data on the degree of fragmentation of planktonic foraminifera shells, a commonly accepted proxy for the extent of carbonate dissolution (Berger et al., 1982; Preiss-Daimler et al., 2013). The fragmentation data (Fig. S4) reveal good preservation (see also Fig. S7) of foraminiferal shells throughout the Pliocene and Miocene intervals, showing no correlation with CaCO3 AR (Fig. S5), confirming that the CaCO3 AR was not driven by dissolution at that time and therefore must reflect pelagic carbonate production changes.

Assuming dissolution did not play a significant role in the observed variations in CaCO3 AR in Pleistocene interglacials and prior to the Quaternary and that pelagic carbonate is the main component of the carbonate fraction of the sediment (Curry et al., 1995), we here observe the changes in the export flux of pelagic biogenic carbonate. Under the same assumption, the new record from Site 927 reveals that pelagic carbonate production (assessed by the pelagic CaCO3 AR) in the equatorial ocean (avoiding large-amplitude temperature changes) has changed on a geological timescale by a factor of 2 and on orbital timescales by up to 50 %. The presence of orbital-scale variability in pelagic carbonate production is an interesting phenomenon that requires further analysis. First, we tested whether or not this variability is periodic, i.e. whether the underlying changes in pelagic carbonate production responded to orbital forcing. Such analysis is possible because the studied intervals have been tuned to the orbital target using parameters other than carbonate content (Figs. 6 and 7). Multitaper method (MTM) spectra derived with the astrochron package in R (Meyers, 2014, using R 4.1.2., R Core Team, 2021) (Fig. 11) highlight significant periodicity close to the precession band for the MIS KM5 and periodicities in the obliquity and 100 ka eccentricity bands for the MCO. This implies that during both intervals, the pelagic production likely varied in response to orbitally driven environmental factors, such as insolation (light intensity for phytoplankton, Cavaleiro et al., 2018) or nutrient availability due to changes in upwelling (Cavaleiro et al., 2020). Interestingly, the dominant periodicities appear different between the Pliocene and Miocene. Next, we asked whether or not the observed periodicities in CaCO3 AR are coherent with the actual insolation, obliquity and eccentricity signals. This is possible because the underlying age models have been tuned such that they should preserve the correct phase relationship with the orbital forcing (Figs. 6 and 7). To this end, we carried out cross Blackman–Tukey (BT) analyses using the Analyseries software 2.0 (Paillard et al., 1996). The results (Fig. 12) indicate a coherence with insolation in the precession band and with the 41 ka obliquity for the MIS KM5. In both cases, the coherence occurs in phase. In contrast, for the MCO we observe a coherence at 41 ka with the obliquity periodicity and at 100 ka with eccentricity, but in both cases the coherence is anti-phased. That the pelagic carbonate production is responding to an eccentricity-paced periodicity (Figs. 11 and 12) is interesting, as eccentricity was not the main driver of the Earth climate signal (Westerhold et al., 2020; De Vleeschouwer et al., 2020). The carbon cycle in the Miocene appears to show eccentricity pacing (Holbourn et al., 2007, 2018; De Vleeschouwer et al., 2020; Raitzsch et al., 2021), and our results indicate that pelagic carbonate productivity may play a role in modulation of this cyclicity. We also note that the discovery of eccentricity forcing pelagic carbonate production in the Miocene and a shift towards obliquity and precession forcing in the Pliocene is consistent with the observations from mid-latitudes by Drury et al. (2021), and the modelling study by Vervoort et al. (2021) provides potential mechanisms on how the eccentricity and obliquity frequencies in carbonate production may arise despite the dominance of the precession frequencies in the forcing.

Finally, we consider the apparent shift in the phase in the relationship between orbital forcing and the CaCO3 AR record between the Pliocene and Miocene. This relationship implied by the cross-spectral analysis is clearly visible in the raw data (Fig. 13), and we consider it unlikely that it is due to tuning artefacts. We note that the Miocene record ends with a strong and distinct minimum in the oxygen isotope record, which provides a strong constraint on the phase relationship between the youngest CaCO3 AR and obliquity cycle. These show an opposite phase relationship to that observed during the Pliocene. This could be explained by a change in the carbonate production response to the insolation changes between the Pliocene and the Miocene. Indeed, the production of different pelagic calcifiers could be promoted by a decreased mean annual insolation at equatorial latitude (with high E+T-P and high obliquity) during the Pliocene compared to the Miocene, when the pelagic carbonate calcifiers appears to be promoted by a higher mean annual insolation at equatorial latitude (with low E+T-P and low obliquity). We can then expect a higher weight of the foraminifera (non-photosynthetic) in the carbonate production balance during the Pliocene and a higher weight of the coccolithophores (doing photosynthesis) in the carbonate production balance during the Miocene. This is coherent with the climate–carbon cycle changes occurring between the Miocene and the Pliocene, highlighted by De Vleeschouwer et al. (2020), who found changes in the phase relationship of δ18O and δ13C before and after 6 Ma.

Because of the observed changes in what appears to be carbonate production among the studied intervals and especially within the studied intervals, we conclude that tropical pelagic calcifiers responded to environmental or biotic forcing on orbital cycles, as well as to long-term shifts in climate and/or ocean chemistry. In other words, either the production, the community composition or the biomineralization of the tropical pelagic calcifiers may respond to local changes in light, temperature and nutrients delivered by upwelling, which followed orbital cycles, as well as to long-term shifts in climate and/or ocean chemistry. The inferred changes in pelagic carbonate production on both timescales are sufficiently large that when extrapolated on a global scale, they could have played a role in the regulation of the carbon cycle. For example, Boudreau et al. (2018) estimated that changes in global pelagic carbonate production on the order of 10 % would be sufficient to affect the marine carbon cycle on timescales from years to millions of years. Whereas the drivers of the orbital-scale variability could be plausibly attributed to changes in local oceanic parameters affecting primary production, the causes of the long-term shifts require another explanation.

There are two studies presenting long continuous CaCO3 AR records from the Miocene to the present (Fig. 14). Both records show an increasing CaCO3 AR from the early Miocene to Pliocene and a decreasing CaCO3 AR from the Pliocene to Quaternary and both records indicate the presence of high-amplitude variability on orbital scales throughout the last 16 Myr (Drury et al., 2021; Lyle et al., 2019). The observed CaCO3 AR at the Ceara Rise appears coherent with both records (Fig. 14), as well as with the recent results by Sutherland et al. (2022) from the South Pacific. Our record shows similar absolute values to Lyle et al. (2019) and Drury et al. (2021) (a CaCO3 AR between 0 and 5 g cm-2 ka-1) and a similar overall trend, with its highest values being in the late Miocene–early Pliocene. Clearly, the overall trend of CaCO3 AR at the Ceara Rise supports the existence of a late Miocene carbonate maximum also under tropical conditions. Interestingly, our observations from the Ceara Rise also support the conclusion from Sutherland et al. (2022) that there does not appear to be any strong relationship between pelagic carbonate production and global CO2, other than the fact that the lowest CaCO3 AR in both their and our records are observed during the MCO with presumably highest CO2.