MODELING HIGH-RESOLUTION VARIATION IN CENOZOIC DEEP-SEA PALEOTEMPERATURE AND BARYSTATIC SEA LEVEL FROM BENTHIC FORAMINIFERAL δ18O AND MG/CA DATA

Determining high-resolution variations in Cenozoic δ¹⁸O of seawater (δ¹⁸O _sw) and barystatic sea level (BSL) from benthic foraminiferal δ¹⁸O (δ¹⁸O_b) represents a challenge considering complexities in separating the isotopic signatures of deep-sea temperature changes and ice volume fluctuations. Comprehensively considering recent advances in methods and new data that can be applied to identify these signals, we have created a numerical model that facilitates the estimation of paleotemperature and BSL variations through the Cenozoic. The model primarily relies on Pacific δ¹⁸O_b measurements and global deep-sea foraminiferal Mg/Ca data. The global Mg/Ca data is transformed to a Pacific basin deep-sea paleotemperature estimate using Bayesian hierarchical modeling with Gaussian process priors. A modeled calculation of BSL begins with a given initial temperature and proceeds using a set of equations that parameterize the proportion of δ¹⁸O_b changes that reflect alterations in paleotemperature over time. A benthic foraminiferal paleotemperature equation can then be used to generate estimates of δ¹⁸O _sw and paleotemperature at each timestep. BSL is then determined from the modeled δ¹⁸O _sw values using an ice-sheet model that accounts for temporal variability in the relationship between δ¹⁸O _sw and BSL variations. A Markov chain Monte Carlo algorithm is used to identify model parameters that produce: 1) an estimate of BSL change that correlates closely with stable-isotope independent estimates of BSL variation derived from the stratigraphic record; and 2) paleotemperature variation implied by differences in measured Pacific δ¹⁸O_b and modeled δ¹⁸O _sw that corresponds with Pacific deep-sea paleotemperature estimates derived from the Gaussian process regression applied to the Mg/Ca data. The model resolution is limited only by data sampling rate of δ¹⁸O_b, and uncertainties are quantified using the Bayesian optimization framework. The model reveals Milankovitch-scale sea-level variations increased from ~5-10m in the Eocene, to ~25m through the Oligocene and Early Miocene, and to 30-40m in the Late Miocene, with a 50m sea level fall during event Oi-1. The ice-sheet model suggests early northern hemisphere glaciation post-Oi-1, and a consistent Greenland ice sheet presence in the time since, excluding the mid-Miocene climate optimum.