Volume 32, Issue 6 p. 600-621
Research Article
Free Access

Closing an early Miocene astronomical gap with Southern Ocean δ18O and δ13C records: Implications for sea level change

Kenneth G. Miller

Corresponding Author

Kenneth G. Miller

Department of Earth and Planetary Sciences, Rutgers University, Piscataway, New Jersey, USA

Institute of Earth, Ocean, and Atmospheric Sciences, New Brunswick, New Jersey, USA

Correspondence to: K. G. Miller,

[email protected]

Search for more papers by this author
Ronidell Baluyot

Ronidell Baluyot

Department of Earth and Planetary Sciences, Rutgers University, Piscataway, New Jersey, USA

Search for more papers by this author
James D. Wright

James D. Wright

Department of Earth and Planetary Sciences, Rutgers University, Piscataway, New Jersey, USA

Institute of Earth, Ocean, and Atmospheric Sciences, New Brunswick, New Jersey, USA

Search for more papers by this author
Robert E. Kopp

Robert E. Kopp

Department of Earth and Planetary Sciences, Rutgers University, Piscataway, New Jersey, USA

Institute of Earth, Ocean, and Atmospheric Sciences, New Brunswick, New Jersey, USA

Rutgers Energy Institute, New Brunswick, New Jersey, USA

Search for more papers by this author
James V. Browning

James V. Browning

Department of Earth and Planetary Sciences, Rutgers University, Piscataway, New Jersey, USA

Search for more papers by this author
First published: 19 May 2017
Citations: 17

Abstract

We present orbital-scale resolution (~10 kyr) benthic foraminiferal δ18O and δ13C records from the Kerguelen Plateau (Ocean Drilling Program Sites 751 and 747) from 14.5 to 20.0 Ma spanning the Miocene climate optimum (15–17 Ma). Our records fill a critical gap from ~17 to 18 Ma, a time when many other deep-sea records are affected by dissolution. We tested the fidelity of published magnetobiostratigraphic age models for these sites by astronomically tuning to the 405 kyr eccentricity cycle. A comparison of spectral estimates between the untuned and tuned records, as well as coherency with Laskar's (2004) eccentricity solution, revealed quasi-100 kyr cyclicity in δ18O and δ13C. There is only a weak signal associated with the 41 kyr obliquity cycle, likely due to the 10 kyr sampling limiting resolution. The δ18O variations point to persistent 405 and quasi-100 kyr modulations of temperature and sea level changes through the early to middle Miocene as predicted by astronomical solutions, with changing dominance of the 100 and 41 kyr beat. Comparison of δ18O records with early to middle Miocene sequences from the New Jersey shelf, northeast Australian margin, Bahamas, and Maldives suggests that the dominant sea level period preserved is the 1.2 Myr obliquity cycle, with sequence boundaries associated with δ18O increases or maxima. On the New Jersey margin, higher-order sequences reflect the quasi-100 kyr eccentricity cycles as modulated by 405 kyr cycles. We suggest that “nesting” of stratigraphic cycles is a function of the following: (1) pervasive (though changing) Milankovitch forcing of global mean sea level change and (2) preservation that depends on sufficient sediment supply and accommodation.

Key Points

  • We provide new astronomically tuned benthic foraminiferal δ18O and δ13C data to fill a critical gap from ~17 to 18 Ma
  • We delineate 100 and 41 kyr cyclicity to understand the significance of Mi events during the early to middle Miocene and the Miocene climate optimum
  • On the New Jersey margin, higher-order sequences reflect the quasi-100 kyr eccentricity cycles as modulated by 405 kyr cycles

Plain Language Summary

We present orbital-scale resolution benthic foraminiferal δ18O and δ13C records from 14.5 to 20.0 Ma spanning the Miocene climate optimum (15–17 Ma). Our records fill a critical gap (~17–18 Ma) during a time period when other deep-sea records are affected by dissolution. A comparison between the untuned and tuned records, as well as coherency with Laskar's (2004) eccentricity solution, revealed a roughly 100 kyr cyclicity in both δ18O and δ13C. The δ18O variations point to persistent 405 and quasi-100 kyr modulations of temperature and sea level changes through the early to middle Miocene, with changing dominance of the 100 and 41 kyr beat. Comparison of δ18O records with early to middle Miocene sequences on the New Jersey shelf, northeast Australia, Bahamas, and Maldives suggests that the dominant period preserved in the sediment record is the 1.2 Myr obliquity cycle. On the New Jersey margin, higher-order sequences reflect the quasi-100 kyr eccentricity cycles. We suggest that such “nesting” of stratigraphic cycles is a function of both: (1) pervasive Milankovitch forcing of global average sea level change and (2) preservation that is largely a function of sufficient sediment supply and accommodation.

1 Introduction

The early to early middle Miocene was an interval of relative global warmth, with peak temperatures occurring during the Miocene climate optimum (MCO) at ~17–15 Ma [Holbourn et al., 2007, 2015; Foster et al., 2012]. This interval takes on added importance if estimates of atmospheric pCO2 are correct, showing levels ranging from preindustrial Holocene values (280 ppm) to nearly double Holocene (250–500 ppm) based on leaf stomata [Kürschner et al., 2008], boron isotopes [Foster et al., 2012; Greenop et al., 2014], and updated alkenone reconstructions [Zhang et al., 2013]. Before ~14 Ma, the East Antarctic ice sheet was not permanently developed [Kennett, 1977; Marchant et al., 1993; John et al., 2004] and experienced periods of large-scale (~40 m sea level equivalent) growth and decay on the Myr scale [e.g., Miller et al., 1991, 2005]. It is unclear whether significant Northern Hemisphere ice sheets had developed by this time period [Foster et al., 2012; Pekar and DeConto, 2006], though ice-rafted debris in the Fram Strait testifies to at least intermittent glaciation of Greenland [Wolf-Welling et al., 1996]. Ice volume and deep-ocean temperatures varied on the Myr scale, as evidenced by benthic foraminiferal δ18O variations of 0.5 to 1.0‰, designated as “Mi events” [Miller et al., 1991; Wright and Miller, 1992], with a periodicity following the long (1.2 Myr) obliquity cycle [Boulila et al., 2011]. The MCO ended with development of a quasi-permanent East Antarctic ice sheet during the Middle Miocene Climate Transition (MMCT), associated with at least two major stepwise δ18O increases (Mi3, 13.8 Ma and Mi4, 13.0 Ma; Figure 2) [Kennett, 1977; Miller et al., 1991; Westerhold et al., 2005; Foster et al., 2012; Holbourn et al., 2015].

High-resolution Miocene benthic foraminiferal isotope records have shown that global ice volume and ocean carbon reservoir changes are correlated with astronomical cycles, particularly the 405 kyr long eccentricity, quasi-100 kyr eccentricity, and 41 kyr obliquity cycles [Woodruff and Savin, 1991; Billups et al., 2004; Pälike et al., 2006; Holbourn et al., 2007]. Though a minor direct contributor to insolation, temperature, and ice volume changes [e.g., Shackleton et al., 2000], eccentricity forcing is pervasive on the Cenozoic carbon cycles evidenced by δ13C records [e.g., Cramer et al., 2003]. Wade and Pälike [2004] suggested that the relationships among orbital parameters and climatic excursions were complicated; e.g., during the early Oligocene, glacial conditions were most frequent when the climate system was preconditioned by low-obliquity amplitude variations and were triggered by individual quasi-100 and 405 kyr eccentricity extremes. This configuration was also associated with the middle Miocene Mi3 event [Holbourn et al., 2005]. Although the direct causes of early Miocene climate changes are still uncertain, geologic records show that the primary pacemakers for the changes are the 405 and 100 kyr eccentricity and 41 kyr obliquity cycles in combination with shifts in ocean circulation and/or long-term global carbon budgets [Holbourn et al., 2007; Pälike et al., 2006].

Beginning with the Quaternary [Imbrie et al., 1984], astronomical time scales relying first on δ18O records and later on cyclical lithologic variations reflected in sediment type, physical properties, and geochemical properties have provided superior resolution for parsing geological time than biostratigraphy alone. In the best cases, astronomical time scales with reliable correlations to precessional variations yield correlations with an uncertainty of <10 kyr [Imbrie et al., 1984]. In reality, most astronomical time scales, even in the late Neogene (past 10 Myr), are calibrated by the long eccentricity cycle (405 kyr) [e.g., Hilgen, 1991], with many individual precessional variation not represented in the record. Thus, uncertainties in Cenozoic astronomical calibrations are as long as one long eccentricity cycle (405 kyr) [e.g., Kuiper et al., 2008].

Rapid progress has been made in extending astronomical time scales prior to the late Neogene using stable isotopes. Shackleton et al. [1999] developed the first astronomically tuned time scale for the latest Oligocene to Miocene (20 to 25.5 Ma) using lithologic cycles from Sites 926 and 929 (Figures 1 and 2) on the Ceara Rise. Shackleton et al. [2000] used this astrochronology to suggest that the Oligocene/Miocene boundary was ~0.9 Myr younger (22.9 ± 0.1 Ma; now 23.03 Ma after astronomical retuning) than previously thought (23.8 Ma) [Cande and Kent, 1995]. Recent astronomically tuned high-resolution stable isotopic records with good paleomagnetic data from ODP Sites 1146 (South China Sea) and 1237 (Nazca Ridge, eastern Pacific; Figures 1 and 2) provide a record from the middle Miocene to the latest early Miocene (12.4–17.2 Ma) [Holbourn et al., 2007, 2015] based primarily on Cibicidoides mundulus or Planulina wuellerstorfi. Pälike et al. [2006] extended the Site 926–929 record from 17.8 to 26.5 Ma using Laskar et al.'s [2004] improved orbital solution, though this record shows greater variability in the interval younger than 20 Ma likely due to increased dissolution (Figure 2). In addition, these sites lack magnetostratigraphy and astronomical time scales should be ideally calibrated at sites with first-order magnetochronology. High-resolution isotopic data from Site 1090 in the South Atlantic Ocean (Figures 1 and 2) [Billups et al., 2002, 2004] have been calibrated to paleomagnetic data of Channell et al. [2003], allowing for a direct correlation of astronomically tuned time to a portion of the geomagnetic polarity time scale (GPTS), Chron C5n (16 Ma) through C6AAr (21.5 Ma). However, a high-resolution record at Site 1090 was only possible by combining the benthic species Cibicidoides mundulus with other taxa (e.g., Oridorsalis umbonatus) that are not typically used for stable isotopic studies. Also, there are several gaps in this record due to the scarcity of benthic foraminifera and the time series degrades between 16 and 18 Ma (hatched on Figure 2) due to dissolution, exhibiting numerous single point peaks [Billups et al., 2004].

Details are in the caption following the image
(a) Location of IODP sites with high-resolution early to early Miocene δ18O data and Sites 747 and 751 on Kerguelen Plateau; (b) site map for Leg 120, Kerguelen Plateau from Schlich et al. [1989]; and (c) paleogeographic map for 16 Ma generated using http://www.odsn.de/odsn/services/paleomap/paleomap.html.
Details are in the caption following the image
High-resolution benthic foraminiferal δ18O time series spanning the interval from Chronozone C5AB to C6A (13.8 to 20.5 Ma). Isotopic scales (bottom) have been offset to maximize overlap. The benthic foraminiferal δ18O records from sites (purple) 926 + 929, (blue) 1090, and (red) 1237 are orbitally tuned to Laskar et al.'s [2004] astronomical solutions and are referenced to the Astronomically Tuned Neogene Time Scale of Gradstein et al. [2004, 2012].

Previous work has thus left a high-resolution δ18O time series gap spanning ~17 to 18 Ma, with uncertain correlations from 18 to 20 Ma (Figure 2). Many sedimentary successions of this age are either incomplete (in part due to a strong erosional pulse in the Atlantic from ca. 16–18 Ma) [Miller and Katz, 1987; Wright et al., 1992] or affected by burial diagenesis and/or carbonate dissolution. Lourens et al. [2004] noted that astronomical calibrations from 14 to 18 Ma are too problematic to allow for astronomically tuned time scales and basin-wide correlations due to either a lack of well-preserved benthic foraminifera or poor magnetostratigraphic data, though recent studies by Holbourn et al. [2014, 2015] have provided data from 14 to 20 Ma. Those records and ours fill the astronomical gap that not only provides potential for astronomical scale correlations, but also provides a record of deepwater and sea level changes.

Ocean drilling has addressed sea level changes not only by drilling deep-sea sections for oxygen isotopes that provide a proxy for ice volume changes, but also by drilling the early to middle Miocene on the northeast Australian margin (Queensland and Marion Plateaus) (see summary in John et al. [2011]), the New Jersey onshore and offshore [e.g., Mountain et al., 2010], the Bahamas [Eberli et al., 1997], and the Maldives [Betzler et al., 2016]. These studies provide a chronology of sequence boundaries apparently formed during lowerings of global average sea level [e.g., Miller et al., 2005; John et al., 2011; Browning et al., 2013]. Studies on the New Jersey passive margin targeted lower to middle Miocene sequences to evaluate the role of global mean sea level (GMSL), tectonic, and sedimentation processes on development of sequences (unconformity bounded units) [Miller et al., 2003; Mountain et al., 2010]. Integrated Ocean Drilling Program (IODP) Expedition 313 was designed to test sequence stratigraphic relationships across a series of lower to middle Miocene clinothems on the New Jersey shallow shelf (Figure 3) [Mountain et al., 2010]. Fifteen lower to middle Miocene (ca. 23–13 Ma) seismic sequence boundaries were recognized using criteria of onlap, downlap, erosional truncation, and toplap [Monteverde et al., 2008; Monteverde, 2008; Mountain et al., 2010]. Mountain et al. [2010] and Miller et al. [2013a] recognized sequence boundaries in cores and logs based on integrated study of key core surfaces, lithostratigraphy (grain size, mineralogy, facies, and paleoenvironments), facies successions, benthic foraminiferal water depths, downhole logs, core gamma logs, and chronostratigraphic ages. Ages of sequences and hiatuses are derived by integrating Sr-isotope stratigraphy and biostratigraphy (diatoms, nannofossils, and dinocysts) on age-depth diagrams with a resolution of ±0.25 to ±0.5 Myr [Browning et al., 2013]. Ten to 11 of these sequences correlate with deep-sea δ18O increases and can be linked to the 1.2 Myr long tilt cycle [Browning et al., 2013]. Three middle to lower middle Miocene sequences were sampled across of a full range of topset, foreset, and bottomset: m5.8 (19.2–20.1 Ma), m5.4 (16.6–17.6 Ma), and m5.2 (14.6–15.6 Ma) (ages from Browning et al. [2013]). Notably, the 1.2 Myr-scale sequence m5.4 sampled three higher-order (100 kyr scale) sequences, yet detailed deep-sea benthic foraminiferal δ18O records were previously lacking (e.g., the gap between Site 1237 and 929, Figure 2) for comparison to the New Jersey margin record.

Details are in the caption following the image
Benthic foraminiferal δ13C, δ18O, and coarse fraction data plotted against meters composite depth (mcd). Dashed line represents the conversion of the mcd at Site 747 to meters composite depth (mcd) using the base of Chronozone C5E as a tie point to make initial comparisons with Site 751. Arrow indicates stratotype for oxygen isotope Zone Mi2a defined here at a level 116.341 mcd where the maximum δ18O value occurs.

In this study, we generated a 7–10 cm scale resolution stable isotopic record for the early Miocene using benthic foraminifera from ODP Kerguelen Plateau Sites 747 and 751 from Chron C5AD to C6A (~14–20 Ma) [Wright and Miller, 1992]. Both sites have magnetostratigraphic records, apparently continuous deposition, and relatively shallow burial depths providing good preservation and high abundance of the benthic foraminiferal Cibicidoides mundulus. This high-resolution record provides a history for the upper deep waters of the Indian Ocean and improves the quality of the δ18O data and Mi events used for stratigraphic correlations (Figure 2). We place the time series for Sites 747 and 751 on a time scale against the Astronomically Tuned Neogene Time Scales 2004 used in Gradstein et al. [2004, 2012] and use the benthic foraminiferal δ18O and Mg/Ca records [Cramer et al., 2011] to derive a sea level record. We compare this sea level record with sequence stratigraphic records from New Jersey, northeast Australia, the Bahamas, and the Maldives to evaluate the response of margin sedimentation to ice volume.

2 Materials and Methods

2.1 Sampling

Sites 747A and 751A are located in water depths of 1985 m and 1634 m, respectively. Both were located above the regional Calcite Compensation Depth through the Neogene. Together with the shallow burial (<250 m), this allows for excellent preservation of calcareous microfossils.

Hole 751A is located in the central part of the Raggart Basin on the southern portion of the Kerguelen Plateau (Figure 1). An advance hydraulic piston corer (APC) was used at Site 751, providing excellent recovery (98%) to 166.2 meters composite depth (mcd) until the drill string failed. The sediment is a nearly uniform white to light-gray mix of biosiliceous and calcareous ooze. Paleomagnetic data are of mixed quality, but key polarity reversals (Chrons C3, C3a, C4, C5, C5C through C6) were thought to be identifiable (Heider in Schlich et al. [1989]; Heider et al. [1992]). At Site 751, Chron C5E is represented from approximately 152 to 158 mcd [Schlich et al., 1989]. We initially followed the polarity interpretation of Heider et al. [1992], but our tuning prompted us to digitize and reinterpret the inclination data for Site 751, yielding a more conservative interpretation (Figure 4). The IODP Kochi Core Repository provided 528 samples from Site 751 at 10 cm intervals from 114 mcd to 165 mcd. The estimated paleolatitude for Site 751 is approximately 58–60°S with a paleodepth between 1400 and 1600 m [Wright et al., 1992].

Details are in the caption following the image
Age versus depth plot comparing magnetochronologic age models of Berggren [1992] using planktonic foraminifera zones and Harwood et al. [1992] using integrated magnetobiostratigraphic interpretations with astrochronologic age models using initial tuning (Figure 5) and revised tuning (Figure 6). Inclination data are after Heider et al. [1992]. Two polarity interpretations are shown, one from Heider et al. [1992] and a more conservative interpretation (this study).

Because drilling at Site 751 ended at 162.9 mbsf in of upper lower Miocene sediments (Chron C5En), we sampled Site 747 to reach the target Chron C6A (~20 Ma). Site 747 is located on a broad terrace at a transition zone from the northern and southern parts of the Kerguelen Plateau (Figure 1). Advanced piston coring (APC) recovered core to a depth of 151.5 mcd before drilling operations switched to an extended core barrel to 256 mcd, collectively obtaining sediments from the upper Pleistocene to the upper Oligocene. Paleomagnetic data extended the record into the upper Oligocene (Chron C10) (Heider in Schlich et al. [1989]). Samples for this study were obtained from 104 mcd to 115 mcd, which had 100% core recovery and only minor to moderate drilling disturbance. The paleolatitude for Site 747 during this interval was ~55°S and paleodepths were 1500–1600 m [Wright et al., 1992]. The sediment is mainly uniform nannofossil ooze and chalk containing foraminifera throughout, with minor components of radiolarians and volcanic fragments [Schlich et al., 1989]. The top of core 12 and the upper portion of core 13 (104 mcd to 115 mcd) represent Chrons C5E to C6C. A modified mcd scale was constructed for Site 747 to make it easier to compare the data with Site 751. Using the base of Chron C5E as baseline, the sample from 106.4 mcd was converted to 157.9 mcd (Figure 3) to match the position of Chron C5E at Site 751. A 7 cm sampling interval (versus 10 cm at Site 751) yields 156 samples to provide a resolution of ~14 kyr per sample at a location with a sedimentation rate of 0.5 cm kyr−1.

2.2 Oxygen and Carbon Stable Isotopes

Samples were oven-dried at 40°C overnight and weighed before washing with tap water over a 63 μm sieve to separate the sand fraction. Sediments >63 μm were oven-dried at 40°C on filter paper then weighed to generate percent coarse fraction data (Figure 3); the coarse fraction is dominated by foraminifera and radiolarians. The 686 δ18O and δ13C values (Figure 3 and Table S1 in the supporting information) were generated from benthic foraminifers >250 μm (primarily Cibicidoides mundulus, except for eight samples where Planulina wuellerstorfi or mixed Cibicidoides assemblages were used in the absence of Cibicidoides mundulus; supporting information Table S1). Three well to moderately preserved foraminifera (2 specimens for 35 of the samples where Cibicidoides mundulus were rare, with no observable offset from adjacent samples when 2 specimens versus 3 were used) were chosen (supporting information Table 1) and cleaned with distilled H2O in an ultrasonic bath for 15–30 s then oven-dried at 40°C. Preservation is noted in the supporting information Table S1 and documented using light photograph and Scanning Electron Micrographs (supporting information Plate 1), with the vast majority of the specimens analyzed classified as well preserved (supporting information Table S1).

Samples were loaded into reaction vials for the automated stable isotope analysis using a Multiprep device attached to a Micromass Optima mass spectrometer. Samples were reacted for 800 s in phosphoric acid at 90°C, and the evolved CO2 was collected in a liquid nitrogen cold finger. Stable isotope values are reported relative to Vienna Pee Dee Belemnite (V-PDB) through the analysis of an in-house laboratory reference material (RGF1). The 1 sigma standard deviation of RGF1 made during these analyses (typically 8 RGF1 analyses for every 24 samples) was 0.05 and 0.09‰ for δ13C and δ18O, respectively. RGF1 is routinely calibrated to NBS-19 to ensure consistency, using 1.95 and −2.20‰ for δ13C and δ18O, respectively, as reported by Coplen [1994]. The internal lab reference material differs from NBS-19 by +0.10 and +0.04‰ for δ13C and δ18O, respectively. The lab analyzes NBS-18 to monitor changes in source linearity for δ18O values. The average δ18O value of NBS-18 analyzed during the period in which these samples were analyzed is −23.07‰, similar to the value of −23.01‰ reported by Coplen [1994]. Therefore, no correction for linearity was made. Stable isotopic data, as well as coarse fraction data, are plotted versus depth with the interpreted locations of Mi events (Figure 3).

2.3 Spectral Analysis

Spectral Analysis was carried out using the SSA-MTM tool kit [Ghil et al., 2002]. The time series was analyzed using three methods: Maximum Entropy Method, Blackman-Tukey, and the multitaper method (MTM). The spectral estimates for each method were compared to ensure that peaks remained at similar frequencies or wavelengths without regard to the method used to analyze the time series. After determining that certain peaks corresponding to similar frequencies were persistent among methods, the MTM was chosen as the primary method to analyze the time series for Sites 747 and 751. This method is widely used for analyses of climate data and uses Mann and Lees's [1996] robust AR (1) process to estimate background noise and determine significance levels.

Spectra of the unturned time series have the largest amplitude peaks near 0.1 m−1 using the Analyseries software package [Paillard et al., 1996]. Sites 747 and 751 data were interpolated to 0.0701 m and 0.0981 m intervals, respectively, to provide evenly spaced data and linearly detrended prior to spectral analysis. Gaussian notch filters were applied and centered at 0.0071 m−1 with a bandwidth of 0.1 m−1 for Site 751 and centered at 0.0021 m−1 with a bandwidth of 0.21 m−1 to remove signals with periods greater than those of interest. Gaussian notch filters centered at 10−6 y−1 with a bandwidth of 0.001 were used to remove frequencies corresponding to signals with periods ≥0.999 Myr. Comparison of spectral estimates with and without the notch filters shows no major changes in the overall pattern of the spectral estimate. However, the power at the lower frequencies (<0.001 kyr−1 or 0.1 m−1) is suppressed and makes the detection of desired frequencies that may correspond to the 405 and 100 kyr eccentricity cycles much easier. Additionally, the Nyquist frequency for Site 751 and 747 is 5.09 m−1 (0.196 m period) and 7.13 m−1 (0.140 m period), respectively, and correspond to 0.048 kyr−1 (21 kyr period) for Site 751 and 0.0357 kyr−1 (28 kyr period) for Site 747. Thus, neither site is sampled sufficiently to resolve precessional scale variations.

3 Results

3.1 Evaluation of Previous Chronologies

Three main interpretations of the paleomagnetic data from Sites 747 and 751 have been proposed (Figure 4): (1) based on preliminary pattern matching to the GPTS [Heider et al., 1992], (2) using planktonic foraminiferal biostratigraphy to correlate magnetochrons to the GPTS [Berggren, 1992], and (3) by integrating of radiolarian, diatom, planktonic foraminifera, calcareous nannofossil, and silicoflagellates [Harwood et al., 1992]. For Site 747, the interpretations agree for the interval considered here: Chronozones C5E, C6, and the upper portion of C6A (104 to 114 mcd) and with correlation of the Mi1b in Chron C6n [Wright and Miller, 1992]. However, for Site 751, where the major portion of the record is obtained (113 to 165 mcd), the previous interpretations agree with the placement of the base of Chron C5E, but there are significant disagreements in the interpretations of other chrons. Both age models place the top of Chron C5En at a depth of 152.1 mcd; however, Berggren [1992] placed the top of C5Bn1n at 114.1 mcd, while Harwood et al. [1992] interpreted that same normal anomaly as part of Chron C5AD (Figure 5). This provides a magnetobiostratigraphic sedimentation rate of 0.93 cm kyr−1 based on Harwood et al.'s [1992] age model or 1.09 cm kyr−1 from Berggren's [1992] age model for Site 751. Harwood et al.'s [1992] age model predicts that the section represents about 12 long eccentricity cycles (14.25 to 19.25 Ma), whereas Berggren [1992] predicts about 11 long eccentricity cycles (14.8 to 19.24 Ma) for Site 751. As we show below, initial analysis suggested that the top of the section studied is 14.1 Ma, but our preferred tuning suggests that indicates that the top is better correlated to 14.5 Ma.

Details are in the caption following the image
Proxies at Site 747 and 751 plotted against age after tuning to the 405 kyr eccentricity target. (a) The interpretation of magnetochrons using the polarity data and correlation to Wade et al.'s [2011] carbon isotope-magnetochronology (37-Mi-C5ADn, etc.). (b) Filtered δ13C data from Site 747 and 751 compared with the eccentricity target filtered with a Gaussian filter centered at 0.0025 kyr−1 with a bandwidth of 0.0005 kyr−1, filtering frequencies from 333 to 500 kyr centered at 400 kyr. (c–e) The δ13C, δ18O, and coarse fraction data, respectively, resampled every 10.4 kyr after the assignment of tie points to provide evenly spaced data for spectral analysis. Phasing of the cycles is in good agreement apart from the interval from 17 to 18 Ma. Note time offset of Sites 751 and 747 stable isotopic data, suggesting miscorrelation.

3.2 Spectral Analysis in the Depth Domain

Prior to tuning, we conducted spectral analysis on the δ13C and δ18O data versus depth to determine if there are quasiperiodic signals that might correspond to Milankovitch cycles (Figures S1 and S2). For the δ13C data at Site 751, there are peaks above the 90% significance level with a periodicity of roughly 4 m (0.25 m−1) and 1.3 m (0.76 m−1), with similar high power in the δ18O spectrum (Figure S1). Using magnetobiostratigraphic sedimentation rates of 0.93 cm kyr−1 based on Harwood et al.'s [1992] age model or 1.09 cm kyr−1 from Berggren's [1992] predicts that the ~4 m cycle is 384–440 kyr, encompassing the 405 kyr long eccentricity cycle. Significant peaks in the δ13C and δ18O data with a periodicity of 1.3 m may represent the short eccentricity cycle, though magnetobiostratigraphic sedimentation rates would suggest the periods were 125–140 kyr rather than the average 100 kyr. The set of peaks above the 99% significance level with periodicities between 0.39 and 0.5 m can be found in both δ13C and δ18O and are consistent with 41 kyr obliquity cycles. At Site 747 (Figure S2), the signal thought to be produced by the 100 kyr signal is well represented in both δ13C and δ18O, though the 405 kyr cycle is not as strong due to the shorter length of the record.

3.3 Astronomically Tuned Chronology

Wade et al. [2011] and Wade and Pälike [2004] proposed a naming scheme that ties the 405 kyr eccentricity cycles calculated from Laskar et al.'s [2004] astronomical solution to magnetochronology. The cycles are numbered starting at 1, representing the most recent long eccentricity minimum and proceeding back through the Cenozoic to 167. The numbers are accompanied with a subscripted code for the geological epoch, and the magnetochron closest to a 405 kyr eccentricity minimum, for example, 46 Mi-C5En, represents a 405 kyr eccentricity minimum during the Miocene within Chron C5En [Wade et al., 2011]. Rather than converting from depth to time using all of the tie points of Harwood et al. [1992] or Berggren [1992], tuning was done with respect to depth. We started with mean sedimentation rate of 0.93 cm kyr−1 for Site 751 and 0.47 cm kyr−1 for Site 747 obtained using the paleomagnetic tie points at a depth of 113.8 mcd (14.178 Ma) and 152.1 mcd (18.281 Ma) for Site 751 [Harwood et al., 1992] and 157.77 mcd (18.781 Ma) and 164.224 mcd (20.131 Ma) for Site 747.

The δ13C records from Sites 747 and 751 were tuned using the long eccentricity cycle and nomenclature of Wade et al. [2011], assuming that benthic foraminiferal δ13C records were dominated by the long eccentricity cycle as noted in previous studies [Woodruff and Savin, 1991; Holbourn et al., 2007; Pälike et al., 2006; Billups et al., 2002]. δ13C minima were tied to eccentricity minima, resulting in an uncertainty in age approximately equal to half of a long eccentricity cycle (~200 kyr) assuming that the 405 kyr δ13C are all represented and properly correlated. This conservative procedure relies on the robustness of the 405 kyr beat and differs from other tuned records for the Miocene that implemented tuning δ18O data at the obliquity (41 kyr) frequency to a target curve containing some mix of eccentricity, obliquity, and precession from Laskar et al.'s [2004] orbital solutions [Holbourn et al., 2007; Pälike et al., 2006; Shackleton et al., 1999].

Using a sedimentation rate of 0.93 cm kyr−1 for Site 751, the 405 kyr eccentricity forcing yields a cycle every 3.7 m with a wavelength of 0.268 m−1. Using a constant sedimentation rate of 0.47 cm kyr−1 for Site 747, the expected occurrence of the long eccentricity cycle should be every 1.88 m or a signal with a wavelength of 0.532 m−1. These estimated values are similar to significant peaks observed in the depth domain (Figures S1 and S2).

The record at Site 747 examined here appears to be too short for tuning to the 405 kyr cycle. If there is a signal expected every 1.88 m and the length of the time series at Site 747 is 10.87 m, there should be approximately six cycles within the time series. This just meets the criteria suggest by Weedon [2003] of six cycles needed to detect a cycle. However, the time series is not long enough to generate a significant peak (Figure S2), and it is possible that filtering the 405 kyr cycle at Site 747 could isolate frequencies of noise rather than signal. Nevertheless, we tuned Site 747 in the same manner as Site 751 for consistency and comparison. Because of this uncertainty in the Site 747 astrochronology, we focus on interpretation of data from Site 751.

3.3.1 Initial Age Model

Based on our initial tuning strategy, proxies from Site 747 and 751 are plotted against an astronomical chronology (Figure 5). We used the polarity interpretation of Heider et al. [1992] and Wade et al.'s [2011] carbon isotope-magnetochron relationships to reidentify magnetochrons (Figures 4 and 5). Even though tuning was done in the depth domain, the phasing of the long eccentricity cycles at Sites 751 and 747 is in good agreement when compared to the filtered 405 kyr eccentricity solution of Laskar et al. [2004] (Figure 5b). Between 16.8 and 17.6 Ma (corresponding to a depth of 135 to 155 mcd), the phase of the cycles do not match as well. This time interval is associated with a decrease in coarse fraction percent (Figure 5e), suggesting a change in sedimentation rate occurred likely from dissolution.

Even though the phasing of the eccentricity minima generally aligns well, the alignment of the Mi2 δ18O event versus magnetochrons in the initial age model does not agree with previous first-order isotopic-magnetostratigraphic correlations [Wright and Miller, 1992]. Mi2 falls within Chron C5Br in the initial astronomical age model (Figure 5); however, Mi2 has previously been placed in the upper normal of Chron C5C at three sites with first-order magnetochron-isotopic correlations (North Atlantic Site 608 and 563 and Site 747) [Wright and Miller, 1992; Wright et al., 1992; Miller et al., 1991], similar to that of Site 1090 (Figure 2).

We evaluate if the miscorrelation is due to incorrect identification of magnetochrons at Site 751. Examination of the inclination data from Site 751 suggested that several intervals interpreted as normal polarity are not interpretable due to numerous reversed points (depicted by the diagonal stripes on Figure 4, “column Polarity (this study)”). The physical properties of the cores from 136 to 148 mcd show slight to moderate drilling disturbance and poor paleomagnetic data, although the integrity of stable isotope data does not seem to degrade within this interval. Despite the uncertainties in some polarities and magnetochrons, the identification of C5Br is robust even with our conservative reinterpretation of the inclination data (Figures 4 and S4). The lack of agreement with the first-order correlation of the Mi2 increase prompted us to revise the astronomical tuning. Reinterpretation of the magnetochrons at Site 751 is also supported by the offset between a δ18O decrease following Mi1ab that is assigned an age of 18.8 Ma at Site 751 and 19.2 Ma at Site 747 (Figure 5).

3.3.2 Revised age Model for Site 751

The δ18O data and location of Mi2 provide an alternative age model (using the same tuning procedure) that is still within biostratigraphic constraints. Using the location of Mi2 at 130 mcd at Site 751 (Figures 3 and 4) and the same tuning method to Wade et al.'s [2011] 405 kyr eccentricity minima, we propose a preferred age model that integrates stable isotope stratigraphy.

The revised age model using the same tuning procedure described above was computed, but rather pairing the first δ13C minima with 36 Mi-C5ACn, it was tied to 37 Mi-C5ADn. This moves the cycles one 405 kyr cycle older (i.e., the time series starts at ca. 14.5 not 14.1 Ma; cf. Figures 5 and 6). The sedimentation rate varies between 0.81 and 1.15 cm kyr−1, similar to the initial tuning conducted, 0.84 to 1.14 cm kyr−1.

Details are in the caption following the image
Version of Figure 5 using a revised age model for Site 751. Rather than matching the first δ13C eccentricity minimum to 36 Mi-C5ACn, we correlated to 37 Mi-C5ADn, roughly shifting the time series at Site 751 405 kyr older. The filtered δ13C data (b) show a much better agreement with the target in the interval from 17 to 18 Ma. Note the excellent correlation of Sites 751 and 747 stable isotopic data. Slanted lines in Magnetochron column indicate differences in age between the GPTS (left vertical line) and ours (right vertical line).

Similar to the initial model, the phases of the filtered long eccentricity cycle are in good agreement with one another (Figure 6b). However, unlike the initial age model, the filtered long eccentricity data at Site 751 maintain strong coherency with the target within the interval where there is an implied change in sedimentation rate. Additionally, with the reinterpretation of paleomagnetic data based on shifting the eccentricity cycles, there is a better agreement for the location of Mi events with published first-order magnetochron correlations. Mi2 falls within the eccentricity minima 41 Mi-C5Cn correlating with the late part of Chron C5C rather than 42 Mi-C5Cr that correlates with the reverse portion of Chron C5C. This revised tuning results in excellent alignment between δ18O records between Sites 751 and 747, with Mi1ab and subsequent decrease correlating precisely from site to site (compare tuning in Figure 6 against the tuning in Figure 5).

Slight offsets of our astromagnetochrology with the GPTS require consideration (Figures 6 and S4). Reinterpretation of the inclination data (Figure 4) suggests that Chron C6/C5Er is the most robust of the magnetochronzonal boundaries. The GPTS places this boundary at 18.784 Ma, whereas our chronology places it at 18.62 Ma, a difference of 160 kyr (slanted line Figure 6). There are at least two possibilities to explain this: (1) the GPTS is slightly miscorrelated; or (2) our assumption that δ13C minima are tied to eccentricity minima is incorrect. Most other chron boundaries are ambiguous due to the increased amount of uncertain polarities or consistent (C5ADr/C5B), though slight differences in the placement of the base of C5Br (Figure 6; GPTS, 15.974; ours, 16.05; 76 kyr) may be due to uncertainties in the identification of the magnetochronozones.

Tuning to a target curve provides a more linear sedimentation rate because tuning to an astronomical target allows for the correction of random variations in accumulation rate (Figure 4). Also, both tuned age models lie within the boundaries of the previous age models for Site 751 based on biostratigraphy, implying that tuning to Laskar et al.'s [2004] eccentricity solution allowed for the refinement of the age models. The age versus depth plot of the two tuned records are parallel with a 405 kyr shift from the change in the initial correlation from 36 Mi-C5ACn to 37 Mi-C5ADn.

4 Discussion

4.1 Astronomical Tuning

The benthic foraminiferal δ13C data were directly tuned to the eccentricity target, so it is not surprising that the signal at 405 kyr period is high and that the coherence between target and data is close to 1 (Figures 7a and 7c). For the Site 751 records (Figure 7a), the 100 kyr period is not well represented in the δ13C data, and the coherence barely rises above the 95% significance level. However, there is a peak with a period of approximately 125 kyr, which may be a representation of the short eccentricity cycle, one of the major periods predicted at 95 and 124 kyr [Laskar et al., 2004]. There is also a significant peak with a period of 41 kyr at Site 751 that corresponds to obliquity.

Details are in the caption following the image
Log power spectrum for stable isotope data was generated using the multitaper method with 4 tapers and significance levels determined using a Robust AR1 noise background. Prior to analysis frequencies with periods greater than 1 Myr were removed using a Gaussian notch filter. Coherency plots comparing the data and eccentricity were generated using the Blackman-Tukey Method with a lag equivalent to 33% of the number of data points. Values below the solid line at 0.057 represent the 95% confidence level testing nonzero coherence. (a, b) Spectral estimates for the time series at Site 751 that was resampled to be evenly spaced at 10.4 kyr resulting in a Nyquist frequency of 0.048 kyr−1 (20.8 kyr). (c, d) Spectral estimates for the time series at Site 747 that was resampled to be evenly spaced at 13.98 kyr resulting a Nyquist frequency of 0.035 kyr−1. For the δ13C data the signal at 400 kyr frequency is high and that the coherence is close to 1, this was directly tuned to the target and is partly caused by the tuning procedure (Figure 7a).

For the Site 751 benthic foraminiferal δ18O data (not directly tuned to the target, Figure 7b), the 405 kyr eccentricity cycle is weaker and short eccentricity cycle is stronger. Similar to the δ13C data, the signal with a period of 405 kyr has high coherence, but the spectral power of the peak is lower. For the short eccentricity cycle, the δ18O data show much better coherence at approximately 125 and 100 kyr, with equally strong spectral power peaks that correspond to the predicted short eccentricity cycle (95 and 124 kyr) [Laskar et al., 2004]. Similar to δ13C, there is some power at the frequency that could correspond to the obliquity cycle with a period of 41 kyr.

There are similar results regarding the long and short eccentricity cycles for the δ13C data at Site 747, though the 405 kyr cycle is weaker (Figures 7c and 7d). The obliquity cycle at Site 747 is not present, likely due to sampling (approximately 2 data points per obliquity cycle).

There are several additional peaks that rise above the 99% significance level (Figure 7). Several possibilities are the following: (1) no cycles are present at the site and we tuned to noise generating peaks at random frequencies with some happening to correspond to Milankovitch cycles; (2) Milankovitch cycles are present, but the tuning procedure increased the amplitudes of spurious peaks to above the 99% significance level; (3) non-Milankovitch frequencies above the 99% significance levels represent harmonics; or (4) the noise background generated for tuned records could be underrepresented [Proistosescu et al., 2012]. Overall, those signals that do not correspond to Milankovitch cycles are found at similar frequencies in the δ18O and δ13C data. This suggests that δ18O and δ13C record may record harmonics at frequencies that do not correspond to Milankovitch cycles, similar to the middle Miocene records of Tian et al. [2013].

Prior to the assignment of ages, spectral estimates of the data sets were studied for cyclicity versus depth (Figures S1 and S2). The position of several of the peaks considered significant in the depth domain corresponded well with expected Milankovitch cycles given the mean sedimentation rates. If these Milankovitch cycles are present, it is possible that tuning exaggerated the presence of harmonic peaks. For example, there are significant peaks in δ18O data around the 100 kyr frequency (Figures 7b and 7d) that could create harmonic peaks at the 50 kyr and the 33 kyr frequency, both considered significant. If this were applied to the 405 kyr cycle, harmonics could be located at 405, 200, 133, 100, 80, 66 kyr, and so forth, reducing the power of the fundamental (405 kyr) peak and spreading it through the harmonic peaks. This could be addressed using a comb filter that allows the frequencies from the harmonics to be concentrated into the fundamental peak but that is not within the scope of this paper and a minimal approach to filtering the data was taken.

We suggest that the number of significant peaks relative to forcing may be an artifact due to the change in sedimentation rate. Weedon [2003] noted that abrupt changes in accumulation rates (e.g., the abrupt change in coarse fraction from a dissolution pulse; Figure 3) could generate twice the number of spectral peaks relative to a record without a “step.” Such a step is indicated using wavelet analysis (Figure 8). Based on the revised age model, a small change in accumulation rates occurred at approximately 17.2 Ma (Figure 4). Wavelet analysis of the δ13C data shows that there is a clear change in spectral characteristics between the interval from 17.2 to 20.0 Ma versus 14.5–17.2 Ma, with greater power at frequencies corresponding to the 405 kyr and 41 kyr cycles in the younger record (Figure 8). This change is not observed in the forcing as inferred from Laskar et al.'s [2004] solution (Figure 8). The short eccentricity cycle (95–125 kyr band) appears to be a persistent signal in the δ13C data in the wavelet analysis, with a slight decreasing trend in the power after 17.5 Ma. The δ18O data show an increase in the 405 kyr band after 17.5 Ma, with a further increase at ~16.5 Ma. There are periods where the power near the short eccentricity bands are relatively weak while the obliquity signal is stronger and vice versa, e.g., between 16 and 17 Ma. This response reflects relatively weak 100 kyr forcing from 16 to 17.5 Ma of the Laskar et al. [2004] solution (Figure 8). Overall, the relative power of δ18O and δ13C spectra shows that there is a change in the spectral densities coinciding with the change in the coarse fraction data.

Details are in the caption following the image
Time frequency analysis using wavelet analysis [Torrence and Compo, 1998]. Filtered and interpolated stable isotopic data from Site 751 were used with sampling interval of 10.4 kyr. Wavelet spectrums for (a) δ18O and (b) δ13C respectively were generated using Morlet waves along with the (c) eccentricity solution of Laskar et al. [2004]. Dashed lines represent the location of the Milankovitch frequencies of interest.

Tuning the 405 kyr cycle improved delineation of other Milankovitch periodicities. Comparison of the spectral estimates of the untuned records from their tuned counterparts (supporting information Figure 3) shows that largest difference between the records is the amplitude of the 405 kyr peak, which could be a combination of a real signal and/or an artifact of the tuning procedure. The peaks corresponding to a signal with approximately 125 kyr also increased after tuning. The 100 kyr eccentric cycle is not well represented in the δ13C data. The two largest peaks for the δ18O data are those belonging to a period with 405 kyr and 125 kyr; the latter is likely attributable to the quasi-100 kyr cycle (supporting information Figure S3). Tuning resulted in a clear differentiation of the 100 kyr cycle (supporting information Figure S3). The amplitudes of these peaks increased and narrowed after tuning (supporting information Figure S3). Also, the peak corresponding to the 100 kyr frequency was virtually absent in the untuned record, but after tuning, there is a significant increase in its amplitude (supporting information Figure S3). The 41 kyr cycle is significant when plotted logarithmically, though it does not show well when plotted against relative power. Overall, increased power and coherency at frequency bands that were not directly tuned to the target curve provides some measure of the success in this tuning procedure [Imbrie et al., 1984].

4.2 Astronomical Correlations From 17 to 20 Ma

Our revised, tuned benthic foraminiferal δ18O record generally compares remarkably well with Sites 1237 and 929/926 (Figure 9a). The Site 751 δ18O record does not display the numerous (>10) spikes of ~1‰ that are found in the Site 926 record between 18 and 20 Ma and lacks three similar spikes in the Site 1237 record (Figure 9). Otherwise, the Site 751 and 929/926 and 1237 records are essentially the same in the interval of overlaps, at least on 405 kyr and Myr scales as shown by smoothed records (Figure 9). Many higher-order, ~100 kyr δ18O variations are reproduced in two records (e.g., δ18O increases at 14.8–14.9, 16.9, and 18.2 Ma), though further tuning at this scale should align cycles that appear misaligned (e.g., a ~100 kyr cycle at 14.7 Ma appears antiphased between Sites 751 and 1237).

Details are in the caption following the image
(a) As in Figure 2 showing δ18O records from sites 926 and 929 (purple) and 1237 (red) but with Site 1090 deleted and our record from Site 751 (green) and the identification of Mi2a added. Also shown on right panel is our smoothing for data from Site 751 (heavy green line), 926 and 928 (magenta), and 1237 (red) interpolated to 10.4 kyr and smoothed with a 41 point Gaussian convolution filter removing periods shorter than 213 kyr. Arrow indicates stratotype for oxygen isotope Zone Mi2a defined here at a level 116.341 mcd (Figure 3) and an age of 14.82 Ma where the maximum δ18O value occurs. (b) As in Figure 9a showing δ18O records from Sites 1337 (blue) [Kochhann et al., 2016] and 1338 (red) [Holbourn et al., 2014]. We spliced the records at 15.5 Ma (see text). Site 1338 provides the older record from 15.5 to 20 Ma. Also shown on right panel is our smoothing for data from Site 751 (heavy green line), 1337 (blue), and 1338 (red) interpolated to 10.4 kyr and smoothed with a 41 point Gaussian convolution filter removing periods shorter than 213 kyr.

We also compare our data with recently published records from eastern Pacific Sites 1337 and 1338 (4463 and 4200 m water depth, respectively; Figure 9b) [Holbourn et al., 2014; Kochhann et al., 2016]. Site 1338 provides the older record from 15 to 20 Ma and thus also begins to fill the astronomical gap, though it displays greater variability than Site 1337 which provides a record younger than 16 Ma. We thus spliced the two records at 15.5 Ma for a clearer comparison (Figure 9b). Our data from Site 751 compare very well with the Site 1337–1338 record (Figure 9b) except for a mismatch between 17 and 18 Ma, with the Site 1338 record lagging Site 751 by ~400 kyr. Given that the correlations of Site 751 with Site 1237 are excellent from 17.5 and younger, this implies that Site 1338 record may be off by one 405 kyr cycle (or that both Sites 1237 and 751 are miscorrelated).

Our comparisons further clarify the significance of the Mi-isotopes events. Mi1ab, Mi1b, Mi2, and Mi3 stand out in our comparison (Figures 9a and 9b), though it is sometimes difficult to identify Mi1ab, Mi1b, and Mi2 in δ18O Pacific records [Holbourn et al., 2014, 2015, personal communication], suggesting that larger deepwater temperature changes in the Atlantic (Site 929/926) and Indian (Site 751) Oceans enhance the events. Like Mi1 (23.1 Ma) [Flower et al., 1995; Zachos et al., 1997; Boulila et al., 2011], Mi1b and Mi2 appear to be 1.2 Myr modulations of the obliquity cycle, with the δ18O maxima (= definition of the events) following several progressively increasing and then peak δ18O values associated with 41 kyr cycles. As shown by the low-pass filter (Figures 5 and 6) and the wavelet analysis (Figure 8), the 405 kyr cycle is also apparent in the Site 751 δ18O record. Wavelet analysis shows that the 405 kyr cycle is strong from 14.5 to 17.3 Ma but is weaker from 17.3 to 20.0 Ma (Figure 8).

Several studies have noted significant interregional benthic foraminiferal δ18O increases that occurred between Mi2 and Mi3 (Mi3a, 14.6 Ma [Browning et al., 2009; Kulpecz et al., 2009]; “Mi2b,” 15.4 Ma; Mi3a, 14.8 Ma [John et al., 2011]). We date Mi2 as 15.95–16.1 Ma on our time scale (Figure 9), with the best placement at the maximum δ18O value at 15.95 Ma and Mi3 at 13.75–13.9 Ma (Figure 9) [Holbourn et al., 2014], with the best placement at the maximum δ18O value at 13.75 Ma. Boulila et al. [2011] suggested that the informal (i.e., versus most other Mi events, no formal stratotype has been designated) Mi2a δ18O maximum at ~14.8–15.0 Ma corresponded to a node in forcing at 14.99 Ma [Laskar et al., 2004]. Previous studies have attempted to relate δ18O increases between Mi2 and Mi3 to continental margin sequence boundaries in Virginia (Mi3a, 14.6 Ma [Browning et al., 2009; Kulpecz et al., 2009]), the northeast Australian margin (“Mi2b”, 15.4 Ma; Mi3a, 14.8 Ma [John et al., 2011]), and offshore New Jersey (Mi2a, 14.6 Ma, and two higher-order events at 15.8 and 15.1 Ma [Browning et al., 2013]). The large (>1‰) Mi2a event illustrated by Browning et al. [2013] may partly be an artifact because of a splice between isotopic records from equatorial Pacific Sites 574 (10.64 to 14.00 Ma and 14.6–16.0 Ma [Pisias et al., 1985]) and southwest Pacific Site 588 (14.029 to 14.642 Ma [Kennett, 1986]). Nevertheless, a large (0.8‰) δ18O increase at Site 751 occurred at 14.9–15.0 Ma that also is apparently at Site 1237 (Figure 9) and equatorial Pacific Site U1338 [Holbourn et al., 2015].

Here we formally designate Mi2a δ18O zone at Site 751 with peak δ18O values at 116.341 mcd (Figure 3) and 14.82 Ma (Figure 6) to stabilize nomenclature (e.g., this event has been termed Mi3a and Mi2a and has been previously assigned ages of 14.6–15.0 Ma) (Mi3a, 14.6 Ma [Browning et al., 2009; Kulpecz et al., 2009]; Mi3a, 14.8 Ma [John et al., 2011]). Though the 1.2 Myr cyclicity controlling Mi2a is clear in the forcing at 14.99 Ma [Laskar et al., 2004], it has a muted δ18O expression because it occurs near the heart of the MCO (peak warmth at 15.0 Ma at both sites, Figure 9).

Previous studies of early to middle Miocene benthic foraminiferal δ18O records have noted several changes in the dominant pacemaker from eccentricity dominated (~100 and 405 kyr) to obliquity dominated (41 kyr) [Pälike et al., 2006; Holbourn et al., 2007]. The interval from 20.8 to 23.0 Ma was dominated by the 41 kyr obliquity cycle [Pälike et al., 2006]. A 100 kyr dominated world has been identified in δ18O records from 19.2 to 20.8 Ma [Pälike et al., 2006] and from 16.2 to 14.5 Ma [Holbourn et al., 2007]. Until this study, δ18O data were not sufficient to document the dominant pacing for 16.2–19.2 Ma. Here we show that the pacing of benthic foraminiferal δ18O records varied in the gap from eccentricity dominated (18.7–19.2 Ma) to likely obliquity dominated (18.7–17.7 Ma), to eccentricity dominated 17.7–14.5 Ma. This change in pacing may be responsible for differential preservation of sequences on passive continental margins, with 41 kyr dominated intervals resulting in poor preservation of sequences [Browning et al., 2013]. The latter noted this empirical relationship that can be explained by either insufficient sedimentation rates or insufficient sea level amplitudes to resolve 41 scale variations.

4.3 Ice Volume, Sea Level, and Margin Sequences

Climate variability recorded by the Site 751 δ18O record (Figure 10a) can be compared with the early to middle Miocene sequence stratigraphic record from the New Jersey passive margin (Figure 10b), Marion and Queensland Plateaus (northeast Australian margin) [John et al., 2011], the Bahamas [Eberli et al., 1997], and the Maldives (Figure 10d) [Betzler et al., 2016]. Mi1b is correlated with Myr-scale sequences m5.4 in New Jersey (Figure 10b), MSB1.2 on the Marion Plateau, sequence O in the Bahamas, and PS6 in the Maldives. Mi2 is correlated with Myr-scale sequences m5.3 in New Jersey offshore (Figure 10b), QU2 on the Queensland Plateau, MSB2.1 on the Marion Plateau, sequence N in the Bahamas, and PS7 in the Maldives (Figure 10d). Slight differences in ages assigned to these sequence boundaries (<0.5 Myr) may be due to uncertainties in age correlations or more likely to dating of different higher-order sequences as can be shown in New Jersey.

Details are in the caption following the image
(a) Benthic foraminiferal δ18O record from Site 751 correlated with ages of sequences on the New Jersey margin [Browning et al., 2013; Miller et al., 2013a, 2013b]. The 40, 41, etc. are eccentricity cycles named by Wade et al. [2011] and Wade and Pälike [2004]. Raw benthic foraminiferal δ18O data (thin blue line) were filtered with a Gaussian convolution filter removing periods shorter than 213 kyr (thick blue line). (b) Solid red boxes are best age estimate of Browning et al. [2013] and crosshatch indicates age errors. The m5.4-1, m5.34, and m5.33 are higher-order sequences within m5.4 [Miller et al., 2013b]. (c) Benthic foraminiferal δ18O record scaled to sea level using a smoothed record (>2 Myr) of Cenozoic Mg/Ca variations [Cramer et al., 2011], the paleotemperature equation of Cramer et al. [2011], a calibration of −0.25‰/°C, assuming that short-term (Milankovitch scale, 104–105 year) temperature changes comprise ~33% of the benthic foraminiferal δ18O changes, and scaling of δ18Oseawater to global mean sea level (GMSL) changes (Figure 1) using a calibration of 0.13 ± 0.02‰ δ18Oseawater/10 m [Winnick and Caves, 2015]. Preferred interpretations of the ages of sequences are given next to the benthic foraminiferal δ18O in black. Yellow squares are the backstripped sea level estimates of John et al. [2004, 2011]. (d) The Maldives sequence boundaries after Betzler et al. [2016].

Further detailed comparisons are possible for the New Jersey shallow shelf where both 1.2 Myr-scale and higher-order sequences were cored and dated by IODP Expedition 313 [Mountain et al., 2010]. Drilling targeted the physically most complete sequences on foresets, including the m5.4 Myr sequence (17.7–16.6 Ma) [Browning et al., 2013]. Though sequence m5.4 is essentially bracketed by the Mi1b and Mi2 δ18O increases, examination of seismic profiles, age constraints, and lithologic and log stacking patterns revealed that sequence m5.4 can be broken into three higher-order sequences (m5.4-1, m5.34, and m5.33; Figure 10b) with a tempo of ~100 kyr [Miller et al., 2013b].

Oligocene-Miocene sequences on the New Jersey margin have a typical duration of 1.0–1.2 Myr, though they are typically bracketed by hiatuses and the amount of time represented by a sequence is generally less than 1 Myr [Miller et al., 1996; Browning et al., 2013]. Comparison with global benthic foraminiferal δ18O records with the New Jersey record indicates a dominant pacing by the 1.2 Myr long obliquity cycle [Boulila et al., 2011]. As discussed above, the δ18O records from Site 751 reflect variations at the 405, 100/125, and ~41 kyr scale. Age estimates for the three higher-order sequences within sequence m5.4 are 0.1–0.2 Myr in duration, with sedimentation rates of 33–44 cm/kyr; this suggests that only a portion of the overall sequence (300–400 kyr of 1.2 Myr) is represented by sediment and that the three higher-order sequences are 100 kyr scale. Though sedimentation rates on the margin were apparently sufficiently high to potentially record 41 kyr sequences (with a predicted thickness of ~4 m), there is no obvious expression of the 41 kyr sea level signature.

Examining the amplitudes of the ice volume and attendant sea level signal based on the Site 751 benthic foraminiferal δ18O records provides clarity as to the margin response to changes in ice volume. We scale our benthic foraminiferal δ18O values to sea level using a smoothed record (>2 Myr) of Cenozoic Mg/Ca variations [Cramer et al., 2011] to account for the effects of long-term temperature changes (Figure 10). We used the paleotemperature equation of Cramer et al. [2011, equation (1)] and a calibration of −0.25‰/°C appropriate for the deep sea [Shackleton, 1974], assuming that short-term (Milankovitch scale, 104–105 year) temperature changes comprise ~33% of the benthic foraminiferal δ18O changes, as in the Pleistocene [Fairbanks, 1989]. The resultant δ18Oseawater estimate was scaled to GMSL change (Figure 10) using a calibration of 0.13 ± 0.02‰ δ18Oseawater/10 m [Winnick and Caves, 2015]. The scaled sea level estimates (Figure 10) provide an estimate of ice volume and attendant GMSL changes with errors of approximately ±10 m (see error analysis for a similar Pliocene estimate of GMSL using Mg/Ca) [Miller et al., 2012] but are not true GMSL estimates because they do not account for effects that change the volume of the ocean basin, other tectonic effects, or changes due to sediment input.

Examination of our ice volume and sea level estimates show several interesting patterns:
  1. Individual 41 kyr sea level amplitudes were generally small (~10 m) and did not cause observable changes in sediments examined to date [Mountain et al., 2010; Miller et al., 2013b].
  2. The quasi-100 kyr cycles are associated with δ18O decreases and attendant sea level changes of ~15–20 m (Figure 10).
  3. The 405 kyr cycles (e.g., 44, 43, 42, 41, and 40; Figure 9) are associated with rapid δ18O decreases (e.g., at the bases of cycle 43 and 42 shifts occur in <20 kyr) in the mean of the 41 kyr variations, with sea level falls of ~20 m; it is these shifts in the mean that appear to be reflected in the higher-order sequence stratigraphic record, though the duration of the preserved sequences appears to be on the order of the 100 kyr scale.
  4. A 1.2 Myr-scale change is clear from Mi1b to Mi2 and represents a general rise of 35 m and fall of ~20 m. This would be equivalent to growth and decay of the modern East Antarctic Ice Sheet (52 m) [Fretwell et al., 2013] and appears to be typical of the early Miocene Mi events.

Our records are consistent with other estimates of the amplitude of sea level change for the Miocene. Though there is minimal temporal overlap with the late early to middle Miocene backstripped eustatic estimates derived from the Marion and Queensland Plateaus [John et al., 2004, 2011], it is important to note that they obtained similar amplitudes (~20–50 m) for Myr-scale Miocene sea level changes. Backstripped sea level estimates from the Marion-Queensland, onshore New Jersey [Miller et al., 2005; 2011], and offshore New Jersey [Kominz et al., 2016] are similar to our δ18O-Mg/Ca estimates (Figure 10). These contrast sharply with previously reported estimates of over 100 m of sea level change for this time interval (e.g., Haq et al., 1987).

We explore the impact of the inferred changes in ice volume and sea level on New Jersey sequences (Figure 10). The m5.4-1 higher-order sequence would classically be interpreted as a lowstand systems tract within the Myr m5.4 sequence, the m5.34 sequence as a transgressive systems tract (with an expansive maximum flooding surface within it), and the m5.33 sequence as a highstand systems tract [Miller et al., 2013b]. Sequence m5.4-1 is clearly associated with peak δ18O values (Figure 10A) and an estimated ice volume maximum/sea level minimum (Figure 10C); we suggest that the sequence actually best correlates with the lowest lowstand of an ~100 and 405 kyr (“reinterpreted ages,” Figure 10c). Sequence m5.34 was also correlated with the lowstand at 17.6–17.4 Ma) but has the most widespread MFL surface (reflector m5.32) of the three higher-order sequences; this suggests that it best correlates with the major sea level rise at 17.2 Ma. This change of 0.2–0.3 Myr (“Reinterpreted ages”, Figure 10c) for sequence m5.34 is our largest change from the age estimate of Browning et al. [2013] but is well within the age errors (±0.25 Myr). The m5.33 sequence corresponds with the highest GMSL estimate and the beginning of the next fall (Figure 10). We caution that though the age errors of Browning et al. [2013] are relatively well constrained (±0.25 Myr) [Browning et al., 2013] compared to studies of most shallow-water sequences [e.g., Greenlee and Moore, 1988 and Greenlee et al., 1992 reported ±1 Myr or worse resolution], our detailed comparison (Figure 10c) is not unequivocal. Nevertheless, our suggestion that the higher-order sequences are paced by interaction of the quasi-100 and 405 kyr eccentricity cycles and that they fit within distinct parts of a sea level cycle is consistent with sequence stratigraphic models [e.g., Posamentier and Vail, 1988] that have asserted this position lacking any firm age data. The preservation of only short intervals (three 100 kyr scale higher-order sequences out of the1.2 Myr m5.4 sequence) is likely due to sedimentation rates which are ~10 cm/kyr during m5.4 time [Browning et al., 2013]. After 14 Ma, sedimentation rates more than doubled and numerous 100 kyr scale sequences can be resolved [Browning et al., 2013].

Sequences deposited during 41 kyr worlds tend to be poorly represented. Sequences m5.45 and m5.47 are highly truncated, poorly preserved sequences [Monteverde, 2008; Monteverde et al., 2008], and sequence m5.6 is also thin and dissected. All three of these sequences fall into interval of 41 kyr domination of the ice volume signal (Figure 8), whereas we show that the very well preserved m5.4 composite sequence (including m5.4–1, m5.34, and m5.3; Figure 10b) and m5.3 sequence were deposited during an interval of dominant eccentricity signal (Figure 8). This is consistent with the hypothesis of Browning et al. [2013] that sequences are more poorly preserved in 41 kyr worlds versus eccentricity-dominated worlds.

We conclude that the stratigraphic record of sea level changes largely reflects composites of sea level cycles nested together. The bundling together of higher-order sea level cycles (in our case primarily quasi-100 and 405 kyr) yield a predictable packaging of sequences on the Myr scale. This bundling also supports the notion that the sequence stratigraphic record is not fractal [Schlager, 2004], but as noted by Miller et al. [2013a], the sequences are fact hierarchical and controlled by Milankovitch forcing.

We suggest that such “nesting” of stratigraphic cycles is a function of the following: (1) pervasive climate forcing of GMSL change, though this forcing continually changes through times as the strength the eccentricity and obliquity components vary (e.g., Figure 8); (2) interactions of the eccentricity and obliquity cycles (Figures 9 and 10a); and (3) preservation that is largely a function of sufficient sediment supply and accommodation. The lack of a 41 kyr cycle can be attributed to small forcing (<10 m sea level changes) and insufficient sedimentation rates. Though the composite record may yield sequences that may appear to be continuous or at least representative of large chunks of time, the record is actually a remarkably discontinuous patchwork of sequences of various scales bundled together by long-period forcing (e.g., 1.2 Myr obliquity in the Icehouse world and perhaps 2.4 Myr eccentricity in the Greenhouse world) [Boulila et al., 2011]. This empirical observation of snippets of time being preserved in the most apparently physically complete sections confirms that adage that for shallow marine sequences, there is, in fact, “more gap than record” [Ager, 1981].

5 Conclusion

Site 751 provides a robust record to examine orbital cycles in the early to early middle Miocene. Compared with existing records, the data from Site 751 improve the resolution of deep water stable isotopic data from 16 to 20 Ma (Figure 13). Two tuned age models for Site 751 are presented with a difference of 405 kyr. The preferred tuning (1) requires reinterpretation of magnetochrons at Site 751 that is allowed by uncertain magnetobiostratigraphic data, (2) provides an improved match of the phasing of cycles with respect to the Laskar et al. [2004] solution especially from 16.5 to 17.5 Ma, and (3) displays excellent agreement with data from Sites 926, 929, 1237, 1337, and 1338 and provides better resolution for the Myr-scale Mi events for the early Miocene.

Spectral analysis clarified the fundamental Milankovitch signals captured at Kerguelen Plateau showing the importance of the 405 kyr eccentricity cycle on both benthic foraminiferal δ18O and δ13C records. The long and short eccentricity cycles are captured, and a high-frequency signal that corresponds to the obliquity signal is present. Although the resolution of this study (10–11 kyr) may not be as high (3–4 kyr) as other high-resolution records due to the low sedimentation rate, it is not heavily affected by dissolution.

We use wavelet analysis to examine the time frequency characteristic of stable isotopes at Site 751 showing changes in the dominant Milankovitch cycles from quasi-100 kyr to 41 kyr forcing as suggested for 17–12 Ma by Holbourn et al. [2007]. We clarify the significance of the Mi isotopic zones as composites of 41 kyr cycles bundled together by eccentricity and long tilt modulations and formally name Mi2a (14.82–15.0 Ma δ18O increase and 14.82 Ma maximum).

Comparison of records from the New Jersey shallow shelf, northeast Australian, Bahamas, and Maldives show that Myr-scale sequence boundaries are associated with the Mi1b and Mi2 δ18O events. Scaling the Site 751 δ18O record to ice volume and sea level using Mg/Ca provides insight to the response of passive margin sequences. Amplitudes obtained by this method agree with backstripped estimates from NE Australia and New Jersey. It again shows that the 1.2 Myr scale was the dominant sea level cycle in the early Miocene and reflects ~50 m sea level changes associated with growth and decay of the Antarctic Ice Sheet. Changes in ice volume on the 405 kyr scale caused rapid shifts in the mean of the 41 kyr sea level variations by ~20 m that are reflected in the higher-order sequence stratigraphic record. Preservation of sequences on the 100 kyr scale may be due to modulation by the 405 kyr cycle. Changes in ice volume and sea level on the 41 kyr scale were general small (~10 m) and did not cause observable changes in sedimentation. The bundling together of higher-order sea level cycles controlled by Milankovitch yields a predictable packaging of sequences on the Myr scale.

Acknowledgments

The data generated and spectral analyses were part of R.D. Baluyot's master degree research at Rutgers University supported by the Department of Earth and Planetary Sciences. Baluyot generated the data in the Rutgers Stable Isotope Lab under the supervision of Wright, computed spectral analyses under the supervision of Kopp, and developed the age models under the supervision of Miller. Miller was responsible for global correlations, and Miller and Browning were responsible for all comparisons with the New Jersey margin. Samples were provided by the International Ocean Discovery Program Koichi Repository. We thank C. John and an anonymous reviewer for comments and A. Holbourn for suggesting comparison with Sites 1337 and 1338. Stable isotope and coarse fraction data are archived at the NGDG and Pangea databases in compliance with AGU policy. We thank R.A. Mortlock and N. Abdul for help with stable isotope analyses and W. Si for taking the SEM micrographs and light photographs. Supported by NSF grant OCE14-63759 (to K. Miller).