A Deep Eastern Equatorial Pacific Thermocline During the Last Glacial Maximum

The mean state and variability of the tropical Pacific is influenced by the depth of the thermocline. During the Last Glacial Maximum (~21,000 years ago), the zonal sea surface temperature gradient across the equatorial Pacific was reduced and productivity was generally lower than modern. To understand the thermocline depth's role in determining the Last Glacial Maximum tropical mean state, we reconstruct the upper ocean δ18O profile from multiple species of planktic foraminifera. We synthesize existing records of surface and subsurface dwelling foraminifera to reconstruct the vertical δ18O gradient throughout the eastern equatorial Pacific. We find the thermocline was deeper during the Last Glacial Maximum than the Holocene throughout the eastern equatorial Pacific region. The thermocline depth's role in the dynamic forcing of the cold tongue contributed to the reduced zonal SST gradient across the equatorial Pacific, decreased productivity, and presumably impacted El Niño‐Southern Oscillation variability relative to the Holocene.


Introduction
The thermocline is a vital component of the coupled ocean-atmospheric system that determines the mean state and variability of the tropical Pacific (Fiedler & Talley, 2006). The eastern equatorial Pacific (EEP; Figure 1) is characterized by a cold tongue and an Eastern Pacific Warm Pool (EPWP). The thermocline is shallow in the east; in the cold tongue, upwelling favorable winds bring cold, nutrient rich water to the surface, whereas in the EPWP the surface is warm and well stratified. The western equatorial Pacific has warm sea surface temperatures (SSTs) with a deep thermocline (Rebert et al., 1985). The cold eastern and the warm western equatorial Pacific create a strong zonal SST gradient that generates strong atmospheric circulation, which reinforces the zonal SST gradient and thermocline tilt (deep in the west, shallow in the east). Perturbations in the thermocline depth, zonal SST gradient, and atmospheric circulation generate El Niño-Southern Oscillation (ENSO) events (Guilyardi et al., 2009).
During the Last Glacial Maximum (LGM; 19,000 to 24,000 years ago), pCO 2 concentrations were lower than today and an ice sheet covered North America (Clark et al., 2009). SST reconstructions show the equatorial Pacific was cooler than modern and that the western and eastern warm pools responded primarily to pCO 2 radiative forcing, while the cold tongue responded to pCO 2 radiative forcing and dynamic forcing (Ford et al., 2015;Koutavas et al., 2002;Lea et al., 2000). As a consequence of this variable cooling pattern, the zonal SST gradient was likely reduced during the LGM relative to today. Proxy reconstructions suggest that productivity was likely reduced during the LGM (Costa et al., 2017) despite an upwelling strength that was similar to today (Loubere, 2000). Given the thermocline depth's role in determining the mean state of the tropical Pacific, how did the thermocline contribute to the SST and productivity patterns during the LGM?
Limited faunal (Andreasen & Ravelo, 1997) and stable isotope (Loubere, 2001;Patrick & Thunell, 1997;Xu et al., 2010) records suggest the thermocline was deep during the LGM; however, there are few records that capture the spatial heterogeneity of the EEP (i.e., within the EPWP and cold tongue). Here we use stable isotopes of multiple species of planktic foraminifera to reconstruct upper water column profiles to infer changes in thermocline structure at Ocean Drilling Program (ODP) Site 849, located in the western extension of the cold tongue. We also synthesize existing and generate new δ 18 O records of surface and subsurface dwelling foraminifera to (1) examine the surface and subsurface spatial δ 18 O variability and (2) reconstruct the vertical δ 18 O gradient as a thermocline depth proxy. We find the thermocline was deeper throughout the EEP region during the LGM relative to the Holocene.
For comparison to our Holocene and LGM δ 18 O depth profiles, we used modern hydrographic data ( Figure 2a) (Figure 2b), we used average annual temperature (Locarnini et al., 2010), global gridded δ 18 O seawater data (Legrande & Schmidt, 2006), and the low-light paleotemperature equation of (Bemis et al., 1998). In the EEP, temperature determines the density structure of the upper ocean as salinity and δ 18 O seawater are largely invariant ( Figure 2a). Thus, we argue the δ 18 O carbonate values of vertically distributed planktic foraminifera are a good proxy for thermocline structure and depth.
Additionally, to constrain planktic foraminifera calcification depth variability in our Holocene and LGM δ 18 O depth profiles, we summarized modern planktic foraminifera preferred calcification depths for the EEP (Figure 2b). Planktic foraminifera abundance and vertical distribution are influenced by mixed layer depth, temperature, chlorophyll concentrations, and other factors that change seasonally to decadally (Ravelo & Fairbanks, 1992;Rebotim et al., 2016;Rincón-Martínez et al., 2011). These factors may also influence the export of foraminifera to depth and their incorporation into the marine sediment archive (Kawahata et al., 2002;Thunell et al., 1983;Venancio et al., 2017). To consider this variability, we
We then used a Monte Carlo simulation to estimate uncertainty in our planktic foraminifera δ 18 O depth profiles related to calcification depth and δ 18 O values (Table 1, Figure 2b). Using the planktic foraminifera (1) depth averages and standard deviation and (2) δ 18 O averages and pooled standard deviation, we performed 10,000 Monte Carlo simulations (Figures 2c and 2d).

Thermocline Reconstruction Throughout the EEP
To reconstruct the surface and subsurface conditions across the EEP, we generated additional surface and subsurface dwelling foraminifera stable isotope data and used previously published records (Figure 1, Tables 1 and 2, and Data Set S1; Benway et al., 2006;Faul et al., 2000;Koutavas & Lynch-Stieglitz, 2003 Sadekov et al., 2013;Spero et al., 2003;Spero & Lea, 2002). Additional surface and subsurface dwelling foraminifera (Table 1,  Approximately 10-20 specimens of G. ruber (250-355 μm), and N. dutertrei (400-450 μm) were picked and sonicated in methanol prior to analysis. Samples were run at the Stable Isotope Geosciences Facility at Texas A&M University on a Thermo Scientific Kiel IV Automated Carbonate device coupled to a Thermo Scientific MAT 253 dual inlet isotope ratio mass spectrometer. The analytical uncertainty is 0.05‰ for δ 18 O (NBS-19 standard).
Surface dwelling G. ruber and T. sacculifer and subsurface dwelling N. dutertrei and G. tumida δ 18 O values were used to reconstruct mean Holocene (0-7,000 years) and LGM (19,000 to 24,000 years) conditions (Data Set S1). The available δ 18 O values for a species were averaged within a time window (i.e., 0-7,000 years). If two species were available within a time window, they were then averaged together for a surface or subsurface mean. The mean Holocene value was then subtracted from the mean LGM value to calculate an anomaly. Note that 1.2‰ was subtracted from LGM values to correct for ice volume (Fairbanks, 1989). A LGM to Holocene subsurface to surface gradient was calculated by subtracting the surface mean from the subsurface mean. Here we use the subsurface to surface δ 18 O gradient as a proxy for thermocline depth; this method has had wide applications outside the eastern Pacific (Fan et al., 2018;Nathan & Leckie, 2009;Sagawa et al., 2011), within the EEP (Jakob et al., 2017;Matsui et al., 2017;Patrick & Thunell, 1997;Ravelo & Shackleton, 1995;Spero et al., 2003), and is based on modern core-tops (Hollstein et al., 2017;Ravelo & Fairbanks, 1992;Venancio et al., 2017).

A Deep Thermocline at Site 849 During the LGM
The multispecies reconstruction at Site 849 indicates the thermocline strength was weaker and the thermocline depth was deeper during the LGM relative to the Holocene (Figure 2). During the Holocene, the vertical gradient between the surface and subsurface dwellers was~2.0‰ whereas it was~1.3‰ during the LGM. This reduced δ 18 O vertical gradient during the LGM is consistent with a deeper thermocline. Ice volumecorrected LGM surface δ 18 O values are lower and the subsurface δ 18 O values (at the depth of N. dutertrei and G. tumida) are higher than the Holocene. Colder surface temperatures during the LGM compared to the Holocene are well documented in proxy reconstructions and modeling results (Otto-Bliesner et al., 2009). We predict isotopically lower than Holocene ice volume-corrected δ 18 O values in the deeper thermocline during the LGM, but unfortunately our profile does not include deep-dwelling species such as Globorotalia truncatulinoides (absent from our samples) to verify this extrapolation. Accordingly, we suggest the LGM thermocline was thicker than the Holocene, resulting in an overall deepening of the thermocline.

Regional Patterns in Surface and Subsurface δ 18 O
Surface δ 18 O values (δ 18 O surface ) from G. ruber and T. sacculifer show the EEP was generally cooler and/or characterized by heavy δ 18 O seawater values during the LGM in comparison to the Holocene (Figure 3a). The EPWP has the largest anomaly. The pattern within the cold tongue is more complex with heavier, equal to, or lighter δ 18 O surface values during the LGM in comparison to the Holocene. Overall, the meridional δ 18 O surface gradient across the EPWP and cold tongue was reduced (Koutavas & Lynch-Stieglitz, 2003). This is due in part to stronger cooling in the EPWP in comparison to the cold tongue during the LGM (Ford et al., 2015;Koutavas et al., 2002; although see Dubois et al., 2009, for a discussion of proxy choice). During the LGM, the EPWP primarily responds to pCO 2 radiative heating and its associated feedbacks (Ford et al., 2015). In contrast, the cold tongue responds to a variety of dynamic feedbacks during the LGM including a deep thermocline, weak winds, and a more southern position of the Intertropical Convergence Zone (Ford et al., 2015;Koutavas & Lynch-Stieglitz, 2003).
Subsurface δ 18 O values (δ 18 O subsurface ) from N. dutertrei and G. tumida show the EPWP and cold tongue have different regional patterns (Figure 3b). From the LGM to the Holocene, the EPWP δ 18 O subsurface values decreased (an increase in temperatures and/or decrease in δ 18 O seawater values), whereas the cold tongue δ 18 O subsurface values increased (a decrease in temperatures and/or an increase in δ 18 O seawater values). The limited subsurface records available do not show a consistent spatial pattern and suggest temperatures were 1-2°C cooler or 1-2°C warmer in the EEP (Ford et al., 2015;Hertzberg et al., 2016;Pena et al., 2008;Sadekov et al., 2013). The spatial subsurface temperature and δ 18 O seawater pattern within the EEP is currently poorly constrained.

A Deep Thermocline Throughout the EEP
Our synthesis of surface and subsurface δ 18 O values shows, on average, the LGM subsurface to surface δ 18 O gradient (Figure 3c) was reduced by 0.37‰ compared with the Holocene, suggesting the EEP thermocline was deeper during the LGM than the Holocene. This result is not sensitive to time interval choice within the Holocene (e.g., mid-Holocene [5,500 to 6,500 ka] vs. the last 7,000 years as used in this study). Planktic foraminifera calcification depth preferences are dependent on the coupled thermocline and nutricline in the tropical Pacific (Bé et al., 1985;Fairbanks et al., 1982;Faul et al., 2000;Patrick & Thunell, 1997;Rincón-Martínez et al., 2011;Watkins et al., 1996). If there was a change in calcification depth of the subsurface foraminifera used to calculate the subsurface to surface δ 18 O gradient, this is likely a response to a change in thermocline/nutricline depth itself and thus integrated into the subsurface to surface δ 18 O gradient calculation. The deep LGM EEP thermocline is corroborated by our multispecies δ 18 O vertical depth Monte Carlo reconstruction at Site 849, the previously published δ 18 O vertical profile at Site TR163-31B (Patrick & Thunell, 1997) and faunal reconstructions (Andreasen & Ravelo, 1997). As ocean and atmospheric dynamics are coupled in the EEP, a deep thermocline during the LGM has broad implications for productivity, upwelling, zonal SST reconstructions, and ENSO. Costa et al. (2017) show the majority of productivity records ( 231 Pa/ 230 Th, excess Ba, opal, and organic carbon fluxes) indicate there was reduced productivity during the LGM, and we argue that the deep thermocline may have contributed to this low-productivity state. While the balance of productivity reconstructions suggest reduced productivity (Bradtmiller et al., 2006;Costa et al., 2016Costa et al., , 2017Pichat et al., 2004), some show enhanced productivity (Lyle et al., 1988;Murray et al., 2012) during the LGM relative to today, likely related to the dynamic nature of the EEP and proxy interpretation. Although productivity is tightly linked to upwelling in the EEP today, upwelling has been relatively constant over the last 30,000 years (Costa et al., 2016;Loubere, 2000). This suggests changes in productivity are unrelated to changes in upwelling intensity and instead are related to thermocline depth, biological activity (Pichevin et al., 2009), nutrient availability (Rafter et al., 2012;Sarmiento et al., 2004), and/or dust fertilization (Costa et al., 2016;Loveley et al., 2017;Murray et al., 2012).
During the LGM, a deep thermocline may have contributed to the reduced productivity by limiting nutrient delivery to the surface via upwelling. In the modern ocean, upwelling-favorable winds exist across the 10.1029/2018GL079710 equatorial Pacific (Wyrtki, 1981); however, the western equatorial Pacific has low productivity because the thermocline is deep and the nutrients are beyond the vertical reach of Ekman pumping (Richards & Pollard, 1991). During the LGM, a deep thermocline within the cold tongue of the EEP may have limited nutrient delivery to the surface in a similar manner. Additionally, thermocline waters may have had lower nutrient concentrations and/or different sources during the LGM relative to modern (Costa et al., 2016;Loubere, 2001). Today, equatorial thermocline waters are sourced in the extratropics, mostly south Pacific midlatitude regions (Fiedler & Talley, 2006). An increase in the delivery of Southern Ocean intermediate waters (Pena et al., 2013), which were nutrient poor due to an increase in high-latitude nutrient utilization (Sarmiento et al., 2004), may have also limited productivity during the LGM.
The tropical SST pattern and a deep EEP thermocline suggest the LGM equatorial Pacific was in an El Niño-like mean state. Overall, the zonal SST gradient between the EEP cold tongue and western equatorial Pacific warm pool was reduced during the LGM (Ford et al., 2015;Koutavas et al., 2002). The EEP regional surface δ 18 O (this study; Koutavas et al., 2002) and the balance of SST records indicate there was spatial heterogeneity in cooling during the LGM relative to modern. Surface temperatures in the EPWP and western equatorial Pacific warm pool were~2.5°C cooler during the LGM, whereas the EEP cold tongue was~1.5°C cooler (Benway et al., 2006;de Garidel-Thoron et al., 2007;Hertzberg et al., 2016;Kienast et al., 2001;Koutavas et al., 2002;Koutavas & Sachs, 2008;Lea et al., 2000Lea et al., , 2006Leduc et al., 2007;Rosenthal et al., 2003;Stott et al., 2007). This suggests the EPWP and western equatorial Pacific warm pool primarily respond to pCO 2 -radiative forcing (and associated feedbacks), while the cold tongue responds to pCO 2 -radiative and dynamic forcing. This dynamic forcing includes a deep thermocline, which limited the ability of Ekman pumping to bring cold water to the surface and further cool the cold tongue.
A deep EEP thermocline could have reduced ENSO variability during the LGM by weakening the thermocline feedback that initiates and propagates ENSO (Ford et al., 2015;Manucharyan & Fedorov, 2014;Zhu et al., 2017). Reconstructions based on single foraminifera analyses show either reduced (Ford et al., 2015;Leduc et al., 2009) or enhanced (Koutavas & Joanides, 2012;Sadekov et al., 2013) ENSO during the LGM. Additional ENSO variability reconstructions are needed to reconcile these records (Lu et al., 2018). Much of the disagreement may depend on site location and choice of foraminifera species (Ford et al., 2015;Thirumalai et al., 2013). Studies aiming to reconstruct ENSO variability should carefully select sites that are sensitive to changes in ENSO (and not the seasonal cycle; Thirumalai et al., 2013) and preferably use surface-dwelling foraminifera (Ford et al., 2015). Using thermocline dwelling foraminifera to reconstruct ENSO variability is difficult due to changes in mean thermocline depth and possible foraminifera calcification depth preferences (this study; Ford et al., 2015;Zhu et al., 2017). If thermocline dwelling foraminifera are used, reconstructing the δ 18 O depth profile using our Monte Carlo approach may deconvolve variability changes related to thermocline depth and ENSO variability. Understanding changes in thermocline depth is important to isolating the ENSO feedback mechanisms important to different mean climate states (Manucharyan & Fedorov, 2014).

Conclusions
During the LGM, the EEP thermocline was weaker and deeper in the EPWP and cold tongue region than during the Holocene. We use the δ 18 O values of multiple species of foraminifera at Site 849 along with new and existing records of surface and subsurface δ 18 O to show that the subsurface-to-surface δ 18 O gradient was reduced during the LGM. This deep thermocline contributed to reduced productivity throughout the region and unequal temperature change between the EPWP and cold tongue region during the LGM relative to the Holocene. The LGM was characterized by an El Niño-like mean state that included a reduced zonal SST gradient and a deep thermocline, which may have impacted ENSO variability.