The use of planktonic foraminifera transfer function and Hill sigmoidal fit to reconstruct upper ocean thermal stratification

Abstract

Planktonic foraminifera assemblages have been extensively used to reconstruct paleotemperatures along the Quaternary. Most of reconstructions focused on surface temperature or in a specific water depth. However, assemblages preserved in sediments represent a pluriannual deposition of species inhabiting the upper 1,000 m. Based on those assumptions, fossil assemblages should reflect better the thermal structure of the water column than a determined water depth. Considering this, we applied transfer functions based on planktonic foraminifera and the Hill sigmoidal function on two sediment cores of the Western South Atlantic in order to simulate and reconstruct past upper ocean thermal structure. These sediment cores were retrieved from the equatorial and subtropical continental slope and cover the last 185 kiloyears (kyr), which allowed us to make inferences about the glacial and interglacial heat storage and release. Eleven paleotemperature reconstructions along the upper 1,000 m were calculated by modern analog technique (MAT) followed by Hill’s sigmoidal function fitting to simulate the past thermal structures. Hill’s coefficients were used to estimate physical parameters in order to improve the paleoceanographic diagnostic. The double-stepped Hill function performed the best simulations of thermal structures. MAT-derived paleotemperatures for 11 depths and their respective errors were within the range of best analogs, indicating that those results are reliable to be applied in reconstructions. The reconstructions indicated that key depths to investigate the glacial–interglacial thermal variation were different in our two study sites. Important variations of the heat storage occurred in the upper 80–120 m in the equatorial margin and the lower thermocline layer in the subtropical margin. Based on this, four main scenarios of heat distribution were suggested for the western border of the tropical South Atlantic, which were associated with events linked to variations of the Earth’s orbit, trade wind intensity, and the South Atlantic large-scale circulation.

1 Introduction

Changes of environmental variables influence the relationship between species and the environment, changing the hierarchical structure in the community. In the case of planktonic foraminifera, one of the most used microfossils for paleoceanographic reconstructions, the sea temperature and the food availability are the main environmental variables that influence the community hierarchy and species distribution (Giamali et al., 2021; Giamali et al., 2020; Kucera, 2007). Studying that close relationship, Bé (1977) classified five different oceanic provinces (tropical, subtropical, transitional, subpolar, and polar), which also can be subdivided depending on the hierarchical composition of communities (Bé, 1977; Bé and Hamlin, 1967; Boltovskoy, 1962; Boltovskoy, Boltovskoy, and Brandini, 2000). Assuming that the environment experienced significant variations in the past, the planktonic foraminifera fossil community, that is, the assemblage, should have responded to such variations. Based on this, several methods using the fossil assemblage for paleotemperature (Imbrie et al., 1973; Hutson, 1980; Pflaumann et al., 1996; Waelbroeck et al., 1998; Malmgren and Nordlund, 1997), paleoecology, and bio-ecostratigraphy (Lessa et al., 2014; Drinia et al., 2016; Antonarakou et al., 2019; Kontakiotis et al., 2023) reconstructions were developed in order to better understand the past environment and to forecast the impact of future environmental changes.

Specifically for temperature reconstructions, one of the most extensively used methods using fossil assemblages is the transfer function (Malmgren et al., 2001; Kucera et al., 2005). Transfer functions estimate an environmental parameter through microfossil assemblages, which have a close relationship with this parameter when living. Planktonic foraminifera have been the main microfossil group used to reconstruct paleotemperature in the ocean. Different transfer function methods are available, which can be based on mathematical regression fitting (weighted averaging), neural networks (Malmgren and Nordlund, 1997), or similarity between fossil and recent assemblages (Hutson, 1980; Kucera et al., 2005; Pflaumann et al., 1996; Waelbroeck et al., 1998). This last method, called the modern analog technique (MAT), is the most used. The MAT method needs a databank of core top planktonic foraminifera assemblages, measured temperatures for each core top site (here called a training set) and fossil assemblages for which we wish to estimate the temperature. A (dis) similarity coefficient is calculated between the fossil and all training set, and the estimated temperature for the fossil assemblage will be the (weighted) average of the n chosen (less) most (dis) similar assemblages of the training set.

Previously, most studies using transfer functions focused on reconstructing sea surface temperature (SST) (Imbrie et al., 1973; Hutson, 1980; Pflaumann et al., 1996; Waelbroeck et al., 1998; Malmgren and Nordlund, 1997; Toledo et al., 2007). However, sediment settled assemblages represent a pluriannual compilation of species that dwell at different depth layers and different seasons and are not necessarily in contact with the surface (Kucera et al., 2005; Lončarić et al., 2006; Rebotim et al., 2017). In recent publications, changes in the upper ocean stratification were reported during the Late Quaternary (Sánchez and Carriquiry, 2024; Santos et al., 2020; Ballalai et al., 2019; Portilho-Ramos et al., 2018; Venancio et al., 2018; Kontakiotis, 2016; Scussolini et al., 2015). It encouraged the use of transfer functions to reconstruct Late Quaternary paleotemperature at other water depths of the upper ocean (Ramos et al., 2015; Lessa et al., 2017; Lessa et al., 2019; Ramos et al., 2019) or selecting species by their vertical distribution (Becquey and Gersonde, 2002). Reconstructing temperatures in one or multiple water depths may not reflect reality if the training set is correlated either with temperature at a determined depth or with other environmental variables, or if the thermal structure of the water column changed in the past (Telford et al., 2013; Telford and Birks, 2005; Sánchez et al., 2022).

The upper ocean structure in the Western (Sub) Tropical Atlantic oscillated from a high water stratification during interglacial periods and a warm, less stratified thermocline during glacial maxima in response to the South Atlantic heat retention or exportation to the North Atlantic, known as the bipolar see-saw (Santos et al., 2020; Pahnke and Sachs, 2006). However, orbital and regional factors may dampen such responses depending on the sector. In the tropical sector, the stratification and thermocline depth respond to the precession that modulates the migration of the Intertropical Convergence Zone and SE trade winds (Venancio et al., 2018; Nascimento et al., 2021). In the subtropical sector, the stratification and thermocline depth are influenced by the dynamics of the South Atlantic high-pressure system that promotes upwelling in the continental shelf that spreads offshore in favorable orbital settings (Lessa et al., 2019). Such variations in the upper ocean modify the planktonic foraminifera faunal composition that settles on the sea floor, which can be used as an indicator of those structures.

Considering the assumptions that planktonic foraminifera assemblages in sediments represent a pluriannual imprint of the upper 1,000 m, that the community composition varies vertically, and that species do not perform significant vertical migration (Meilland et al., 2019; Bé, 1977; Lessa et al., 2020; Be, 1960), the planktonic foraminifera community would be better connected to an upper ocean thermal structure than a single temperature value at a determined water depth. Based on this, this study aimed to reconstruct the past thermal structure of the upper water column through the transfer function of planktonic foraminifera assemblages and the use of the Hill sigmoidal function (Brown and Rothery, 1993) as a fit. Reconstructing the upper thermal structure also allowed us to obtain physical parameters related to the thermal structure (e.g, mixed layer depth, stratification by thermocline abruptness, and depth of a given temperature value), which, when applied to paleoceanographic reconstructions, would yield new insights into the dynamics of heat transportation in the past. As a case study, we applied the method and discussed paleoceanographic implications for the last 185,000 years (kyr) in two cores located in two target sites linked to Atlantic Meridional Overturning Circulation (AMOC) heat transport: the equatorial and subtropical margins of the Western South Atlantic, responsible for the transport and retention of heat, respectively.

2 Materials and methods

2.1 Atlantic annual thermal structures and application of the Hill curve

Simulating the upper ocean thermal structure followed two basic patterns, according to World Ocean Atlas (WOA) (2018) (Locarnini et al., 2018). In tropical to temperate areas, the vertical temperature gradient shows high variability with single and multi-stepped structure forming the mixed, upper, and lower thermocline layers. In subpolar and polar areas, the vertical temperature variation is usually absent or has a slight reversed thermocline. Figure 1 shows examples of those types of thermal structures in different South Atlantic latitudes. The equatorial-like thermal structure comprises a clear two-step thermocline due to the shoaling of cold waters (upwelling) modulated by the equatorial convergence. The tropical-like structure comprises a deep and warm mixed layer and a single slope thermocline. The subtropical structure has a shallow mixed layer and a soft slope thermocline, which is also observed in gyre areas where warm surface waters are piled up and downwelled. Subpolar and polar thermal structures have no or very limited variation along the entire water column. Polar water columns can have a slightly cold surface with a temperature-increasing thermocline slope, likely due to seasonal sea ice formation.

FIGURE 1

In order to simulate the upper ocean thermal structure, we applied the Hill sigmoidal fit curve (Brown and Rothery, 1993) using the PAST 4.03 software (Hammer, Harper, and Ryan, 2001). The Hill fit curve is based on the equation:

where a (d) is the maximum (minimum) y values when . c is the Hill’s slope level, which can be smoother (abrupt) if c is a low (high) value. Positive c values generate an upward slope, whereas negative values generate a downward slope. b is the IC_50% of the response of the subject, that is, the middle slope x value. The structure of the Hill sigmoidal curve is shown in Figure 2. Considering x as the water depth (D) and y the temperature (T), Equation 1 can be rewritten as:

FIGURE 2

In order to calculate the depth where a determined temperature value occurs, Equation 2 can be rearranged to isolate D:

In addition, properties of the simulated thermal structure based on Hill coefficients can be calculated and applied to reconstruct parameters of the upper ocean. For that, we focused only on the mixed layer and upper thermocline, where the most pronounced thermal variations occur. Equation 3 was used to estimate the past mixed layer depth (MLD, Equation 4). Assuming the mixed layer has short temperature variability, we can assume a threshold drop as the boundary between the mixed layer and the upper thermocline. Here, we used a difference of 0.5 °C from the SST as the threshold (Sprintall and Matthias, 1992), which also allows estimating the mixed layer when the Hill curve mismatches with the observed shallow mixed layer. Thus, by subtracting 0.5 °C from the estimated SST, we can estimate the MLD using Equation 3 as a reference:

where SST is calculated from Equation 2, assuming the water depth D is near, but not equal to zero (here we used D = 1 m). Considering that the equatorial thermal structure is generally two stepped in the South Atlantic (Locarnini et al., 2018) and the boundary between steps is located at approximately 200 m in most regions, we conducted a double-stepped simulation to improve the fitting. The default boundary between surface and deep curves was set at 200 m, but depending on the estimated upper thermocline base, the boundary depth was adjusted to improve the fitting. The reconstruction of complex structures (e.g., multi-stepped structures) was not explored in this study.

In addition, the upper Hill curve allows us to estimate the water column stratification, which could be directly related to vertical mixing. The stratification index was set to be dependent on two parameters: the amplitude of the thermal variation and the slope of the (upper) thermocline. In order to prevent bias between upper and lower thermoclines, we considered only the first half of the surface curve to estimate the thermal amplitude variation. Then, the thermal amplitude variation was considered to be the difference between SST and the temperature at depth b multiplied by two. Thus, the physical stratification index can be measured by the following equation:

where c is the Hill slope coefficient, and T_b is the temperature at the depth b, which is midway between a and d. In general, the best fit is reached with a 0–200 m depth range. However, some reconstructions or areas require shoaling or deepening of the range to reach the best fit. For data series where different depth ranges are employed, Equation 5 can be depth normalized (Equation 6), allowing a direct comparison:

where D is the depth range used to build the Hill upper curve. Except for polar latitudes, temperature decreases with the water depth, so the c values must be negative. Low negative c values will produce a soft slope, and the stratification index will be low. High negative c values will produce an abrupt slope, suggesting a stronger thermal barrier, and, therefore, higher stratification values. The double of the SST and T_b difference represents the approached temperature range from the mixed layer to the bottom of the (upper) thermocline, where the temperature rapidly decreases. We opted to use SST instead of a or d values to avoid overestimation in thermal structures without a defined mixed layer.

2.2 Using transfer functions to simulate the upper ocean thermal structure in the past

We used the similarity-based transfer function known as the modern analog technique (MAT) (Hutson, 1980) to simulate the past upper ocean thermal structure. We used a training set comprising a constrained squared grid (10°N – 60°S; 20°E − 90°W) with 639 Atlantic and Southern Oceans core tops from the ForCens database (Siccha and Kucera, 2017). Oceanic temperature values for each top core site were retrieved from WOA (2018) raw data values (Locarnini et al., 2018) using the Ocean Data View software (Schlitzer, 2015). This temperature databank comprises each 1° of latitude and longitude. We coupled ForCens core tops with the temperature values of the nearest measured temperature value from WOA (2018). We constrained the training set to avoid non-existent thermal structures from outside the South Atlantic, which could lead to no analog thermal structures. It means that our reconstructions can exhibit some distinctions from previous studies, such as those of Lessa et al. (2017) and Lessa et al. (2019). Fossil assemblages were retrieved from piston cores GL-1248 and GL-1090 sampled in the equatorial and subtropical margin of the Western South Atlantic, respectively (see Section 2.3). The MAT model was built on software C2 (Juggins, 2003) using square chord distance, and the model performance was cross-validated using the leave-one-out method. Temperature reconstructions for fossil assemblages were based on the weighted average of the ten least dissimilar (best analog) core tops.

To analyze if MAT-derived temperatures match simulated thermal structures of the best analogs, we performed some calculations before the final reconstructions. To obtain the best analogs for the thermal structure, we calculated the dissimilarity between fossil and training set assemblages and extracted the upper 1,000 m thermal structure from the ten best analog coordinates. Next, we developed models and calculated paleotemperatures at 1 m, 10 m, 30 m, 50 m, 75 m, 100 m, 150 m, 200 m, 300 m, 500 m, and 1,000 m. Although planktonic foraminifera can be found in the upper 1,000 m (Be, 1960), concentrations of individuals are strongly reduced in the lower thermocline, so we decided to make interpolations to the upper 700 m, where concentrations are still significant (Rebotim et al., 2017; Meilland et al., 2019; Lessa et al., 2020). Paleotemperatures for determined subsamples of the core GL-1248 were compared to evaluate the adherence of estimated paleotemperatures to the best analogs of thermal structures. The chosen subsamples were based on the difference in their assemblage composition. Distinct assemblages are seen at samples dated from 0.7 kyr (core top), 36 kyr (MIS 3 interstadial), 38.5 kyr (MIS 3 stadial), and 120 kyr (the last interglacial). To determine whether the estimated paleotemperatures are robust, we plotted the estimated paleotemperatures together with the best analogs. Performing temperature estimations at a water depth range is important to quantify uncertainties, which cannot easily be calculated for an entire thermal structure. After estimating paleotemperatures for those depths, we simulated thermal structures using the Hill sigmoidal function and plotted these reconstructions and their respective standard deviations together with best analogs. We manually extracted the four coefficients (a, b, c, and d) from both surface and deep curves. Using Hill curve estimated structures, we can calculate the estimated temperature at any depth and vice versa in the range using the four coefficients.

The performance of interactions was evaluated through the explanation coefficient (r²) and the standard error of the regression (SER). Two reconstruction methods were followed. The first method was to estimate the interpolated variation of the past 0–700 m thermal structure. For that, the boundary between upper and lower thermocline curves was set to 200 m as the default for all entries because both curves had irrelevant mismatches. The second method was to focus on the physical properties of the mixed layer and upper thermocline. In this case, the depth range of the surface curve was adjusted so that the whole curve structure could be obtained without decreased adherence or extrapolation. As previously mentioned, the surface curve (mixed layer and upper thermocline) comprises the highest temperature variation and takes the leading role over the photic zone. The extraction of the physical properties of thermal structures was separated from the interpolated reconstruction of past upper ocean thermal structure.

2.3 Regional settings and sediment core sampling

The equatorial and subtropical margins of the Western South Atlantic (Figure 3) are key areas to understand the dynamics of the heat content in the AMOC upper limb and the Subtropical Gyre. The tropical water (TW) is formed at the mixed layers of the South Equatorial Current (SEC). Reaching South America at 10°S, the SEC flows equatorward, giving rise to the North Brazil Current (NBC) or poleward, giving rise to the BC, which recirculates in the South Atlantic Subtropical Gyre (Peterson and Stramma, 1991). The lower thermocline layer is filled by the South Atlantic Central Water (SACW). This water mass originates from the interaction of the BC with the northward flow of the Malvinas Current. The SACW also receives a contribution of the Indian Central Water through the Agulhas Leakage in the southeastern Atlantic (e.g., Poole and Tomczak, 1999). The South Atlantic Subtropical Gyre is spatially narrower at the SACW layer, and its northern portion encounters the South American continental border at 20°S, bifurcating into the northward branch of the North Brazil Undercurrent and the BC as the southward branch (Stramma and England, 1999). From the bottom of the thermocline to approximately 1800 m, the Antarctic Intermediate Water (AAIW; sourced in the Subantarctic Front and Malvinas Current) flows. As SACW, the AAIW penetrates and circulates in an even narrower subtropical gyre, encountering the South American border at 28°S and bifurcating into the Intermediate Western Boundary Current equatorward and the BC southward (Stramma and England, 1999).

FIGURE 3

The water mass geometry has some discrepancies between the two studied sites in the upper 1,000 m. In the subtropical margin (25°S), the upper 200 m are composed of TW, followed by the SACW between 200 m and 800 m, and the AAIW occupying the lowermost 200 m. In the equatorial margin (1°S), the upper 150 m is composed of TW, the SACW is located between 150 m and 500 m, and the AAIW occupies the lower half. Consequently, modern upper ocean thermal structures at our sites also reveal discrepancies (Figure 3). In the equatorial margin, the structure is typically a thick (approximately 70 m) mixed layer, a sharp upper thermocline, and a smooth lower thermocline. In the subtropical margin, the annual thermal structure shows a smooth two-structure: the mixed layer is thin or absent, followed by a smooth sloped upper thermocline down to 150 m, followed by the still smoother lower thermocline until 1,000 m.

Each water mass contains a different planktonic foraminifera community. Spinose species from the genera Globigerinoides, Trilobatus, Globoturborotalita, and Globigerina preferentially dwell in the upper photic zone in the TW (0–50 m) where light is abundant, and the productivity is low (Lessa et al., 2020; Venancio et al., 2016; Sousa et al., 2014). Other species, such as Globigerinella (spinose), Pulleniatina, and Globorotalia menardii (non-spinose), dwell in the upper thermocline, especially in oligotrophic areas, where it is located below 100 m (Lessa et al., 2020; Venancio et al., 2016; Sousa et al., 2014). In shallow upper thermocline regions (such as upwelling areas), Neogloboquadrina (non-spinose) and Globigerina bulloides (spinose) species are present, with Trilobatus sacculifer abundant in the surface (Lessa et al., 2020). In the lower thermocline (SACW layer), a distinct fauna is observed. In tropical regions, Globorotalia truncatulinoides, Globorotalia crassaformis, Globorotalia tumida, and Hastigerina pelagica dwell in the upper layer (200–400 m), and Globorotalia scitula dwells in the lower layer (below 400 m). In subtropical regions, the lower thermocline is entirely inhabited by G. scitula, Globoconella inflata, and a low contribution of G. truncatulinoides (Lessa et al., 2020).

Piston cores GL-1248 (0°55.20′ S, 43°24.1′ W, 2,264 m of water column and 19.29 m length) and GL-1090 (24.92°S, 42.51°W in a water column of 2,225 m with a 19.14 m length) (Figure 3) were collected by Petrobras in equatorial and subtropical margins, respectively. The chronology of the sediment core GL-1248 was based on 13 radiocarbon and visual alignments of the Ti/Ca ratio with the NGRIP ice core stable oxygen isotope (δ¹⁸O) (Venancio et al., 2018), revealing the occurrence of a hiatus between 29 kyr BP and 14 kyr BP. The species Globigerinoides ruber and Neogloboquadrina dutertrei of the core GL-1248 were published in Piacsek et al. (2021). In this study, we provide the complete planktonic foraminifera assemblage at a 10 cm resolution. The chronology of sediment core GL-1090 was based on nine radiocarbon dating points and visual alignment between benthic foraminifera δ¹⁸O of the core and reference curves (Santos et al., 2017). Later, Ballalai et al. (2019) and Santos et al. (2020) updated the age model with improvements in the early last glacial section. The planktonic foraminifera assemblages for the last 20 kyr and the 140–70 kyr interval were previously published by Lessa et al. (2017). In this study, we present the complete foraminifera assemblage record at a 4 cm resolution for 30–55 kyr and at a 10 cm resolution for the remaining core.

According to radiocarbon dating, the top of cores GL-1248 and GL-1090 date from 0.6 kyr BP to 6.2 kyr BP, respectively. The micropaleontological analysis followed Lessa et al. (2017). Ten cubic centimeters of humid sediment were washed on a 63 μm meshed sieve and dried at 50 °C for 24 h. The dry sediment was sieved on a 150 μm mesh and split until 300 specimens remained, which were handpicked and identified at the species level, according to Schiebel and Hemleben (2017) and the Mikrotax database (Mikrotax, 2025).

3 Results

3.1 Planktonic foraminifera assemblage of cores GL-1248 and GL-1090

We identified 34 species in the core GL-1248 (equatorial margin). The most abundant species are shown in Figure 4. Most were from warm and oligotrophic areas with high relative abundances during MIS 5 and MIS 1 (Figure 4). Species linked to productivity and upper thermocline dwellings increased during MIS 3 and MIS 4. G. truncatulinoides was the only abundant lower thermocline-dwelling species in this core, reaching approximately 10% of the assemblage at 75 kyr BP and between 45 kyr BP and 40 kyr BP. Cold water species were more abundant during MIS 3, aligning with low Ti/Ca values (related to MIS 3 interstadials), whereas oligotrophic species Globigerinoides ruber pink and Trilobatus sacculifer increased abundance when the Ti/Ca ratio was high in MIS 3. Thermocline-dwelling species were scarcely abundant and had no significant effect on the assemblage composition. Thirty-two species were identified in the core GL-1090 (subtropical margin), which the sections refer to MIS 5, and the last 20 kyr were previously published in Lessa et al. (2017). Tropical oligotrophic species were more abundant during MIS 5e to 5b, MIS 4, and MIS 2–1 (Figure 5). Cold water and productivity-related species were more abundant during the Early MIS 6, MIS 5a, and MIS 3. Lower thermocline dwelling species G. truncatulinoides, G. inflata, and G. crassaformis had significant abundance increases with separated predominance during MIS 6 and between MIS 5a and MIS 2, reaching up to 20% of the total assemblage, which can be considered a significant contribution.

FIGURE 4

FIGURE 5

3.2 Performance of generic simulations of the thermal structure

Except for 0–1,000 m simulations at the equator with an r² up to 0.97 and SER up to 1.08 °C, simulations had r² values greater than 0.99 and SER less than 0.6 °C (Table 1). Double-stepped simulations had even better results, with r² near 0.999, and SER less than 0.13 °C. The 0–1,000 m simulation did not successfully simulate the mixed layer (stable temperature surface layer), whereas 0–200 m and 200–1,000 m simulations closely estimated the observed (Locarnini et al., 2018) vertical thermal variation (Figure 6). Based on this, we concentrated our efforts on double-step simulations for the reconstruction of past thermal structures.

FIGURE 6

3.3 Modern analog technique (MAT) models and past reconstructions of thermal structures

Based on the r², the model’s performance tended to decrease with the water depth (Table 2; Figure 7). Model uncertainties (RMSEP) were higher between 50 m and 150 m, matching the average upper thermocline layer, which is spatially much variable.

FIGURE 7

Figure 7 shows examples of how the estimated MAT paleotemperatures were coherent with the range of the thermal structure analogs. Most standard deviation bars are in the analog range, with high values reflecting broad temperature variation among analogs, especially in the upper thermocline layer. The double-stepped Hill function successfully simulates past thermal structures because the simulation was never out of the paleotemperature standard deviation range. In addition, the Hill sigmoidal function followed the average area and standard deviation of analog thermal structures (Figure 7). Hill coefficients are summarized in Table 3. Observed offsets between the 0–200 m and 200–1,000 m curves are within the standard deviation range of the estimated paleotemperature at 200 m, which maintains valid simulations. The double-stepped simulation with a break at 200 m demonstrated a high consistency with paleotemperature estimations.

3.4 Reconstructions in cores GL-1248 and GL-1090

Thermal structure reconstructions of core GL-1248 (equatorial margin) showed a thermally stable surface layer in most of the core (Figure 8A) with a highly stratified upper thermocline layer and a cold and less variable lower thermocline. Compared with the core top (Figure 8C), reconstructions showed a colder upper thermocline with the maximum at approximately 100 m prior to the Holocene. The lower thermocline tended to have little variation with a slight reduction of the slope (Figure 8B), which did not cause a significant temperature increase compared to the core top (Figure 8C). These reduced lower thermocline slope events were numerous and correspond to the MIS 3 stadial setup in Figure 7, with a less warm mixed layer and pronounced downward spread of the thermocline. The temporal evolution of the thermal structure along the last 130 kyr showed a tropical-like shape during MIS 5 with a warm mixed layer and a narrow thermocline. Events of MIS 3 stadial setup started to occur from MIS 5b (82 kyr), followed by the return of the default MIS 5 setup. From MIS 4 (70 kyr) until 30 kyr, the setup of the thermal structure changed to a near to equatorial-like shape with a shallow, abrupt upper thermocline but with a colder mixed layer and common occurrences of MIS 3 stadial setup. The last 14 kyr were newly marked by a stable tropical-like thermal structure with much higher temperature values for the mixed layer and upper thermocline than during MIS 5.

FIGURE 8

Regarding reconstructions for the core GL-1090 (subtropical margin) in a general overview, the thermal structure showed a typical subtropical structure with a narrow mixed layer and smooth upper and lower thermoclines in most of the core (Figures 8D–F). Many temperature variations can be seen in the upper 150 m with an apparent shoaling of the lower thermocline during the penultimate and last glacial section (Figure 8E) and a deepening of the warm mixed layer in MIS 1. Compared with the core top (Figure 8F), reconstructions indicated a colder upper thermocline, as in the equatorial margin, but the lower thermocline was variable during this time. The layer between 300 m and 500 m was cooler during early MIS 6 and most of MIS 5 (except for the MIS 5b–5a transition), whereas the layer between 500 m and 700 m had five warming events (Figures 8D,E), where the three oldest warming periods comprised the whole lower thermocline, interrupting the observed cold anomalies. The temporal evolution of the thermal structure of the core GL-1090 showed a stable subtropical setup during most of MIS 6 (185–150 kyr). In Late MIS 6 (150–130 kyr), the mixed layer was colder, and the thermocline became smoother with higher temperatures in the lower thermocline. During MIS 5e–5c (130–85 kyr), the thermal structure returned to the subtropical setup, but with a relatively colder mixed layer and smoother upper thermocline during MIS 5e and 5c. From MIS 5b (85 kyr) until early MIS 3 (55 kyr), the subtropical structure was interrupted by the short-term events of sharp smoothing of the thermocline at 87 kyr and 65 kyr, where the mixed layer was cold, and the lower thermocline had significant warming. From MIS 3 (55 kyr) to the Last Glacial Maximum (LGM, 20 kyr), the thermal structure evolved to a cold glacial setup (cold mixed layer and abrupt, less stratified upper thermocline) interrupted by a subtropical shape in approximately 45 kyr. From LGM to MIS 1 (core top), a strong warming in the mixed and upper thermocline layers modified the thermal structure to a tropical setup. The estimated temperature values in these upper layers were the highest in the record.

4 Discussion

Applying MAT to estimate paleotemperatures using gridded temperature values is a tempting option to obtain temperature values for the training set core top, with exact coordinates, because temperature databanks (e.g., Locarnini et al., 2018) and core top locations are usually unmatched. However, values obtained from a gridding are not recommended because these temperature values can be autocorrelated and do not reflect the real temperature for the sites (Siccha et al., 2009; Telford and Birks, 2005; Telford and Birks, 2011; Telford and Birks, 2009). Furthermore, different vertical distribution of planktonic foraminifera species can lead to unrealistic results for water depths in which the most abundant species do not dwell (Rebotim et al., 2017; Lessa et al., 2020; Telford et al., 2013). This study uses the MAT in multiple water depths with the nearest location raw data because quality control of reconstructions and model performance can be easily accessed, and autocorrelation is avoided. Reconstructions can also be obtained by a (weighted) average of the temperatures at depth D of analogs. However, our MAT-derived paleotemperatures were within the highest density in the range of analog values (Figure 7), which means this method is suitable for obtaining the best values for thermal structure reconstructions. Regarding the assemblage assumptions, estimating an entire 700 m thermal structure includes the expected vertical distribution of species, as seen in situ (Lessa et al., 2020; Rebotim et al., 2017; Meilland et al., 2019). However, we cannot be certain whether assemblages along the core would reflect continued annual or be biased by a seasonal signal in a section. Even though G. ruber, the most abundant species in our study sites, is annually present in the western tropical Atlantic (Jonkers and Kučera, 2015), some species may increase in abundance during the austral winter and spring, such as G. inflata, G. truncatulinoides, and G. scitula (Lončarić et al., 2006; Jonkers and Kučera, 2015). These species significantly increased in abundance during the glacial period in the subtropical margin (Figure 5). Despite this, temperature and productivity thresholds in our Western South Atlantic sites are not variable enough to trigger significant seasonal blooms.

Our reconstructions of upper thermal structures indicated variations in both areas during glacial and interglacial cycles and abrupt events; some are linked to AMOC variations, such as Heinrich and Dansgaard–Oeschger events. However, the thermal variability was stamped in different depths depending on the basin, which could be treated as key depths of heat storage. In the equatorial margin, the most noticeable temperature variation occurred in the upper 100 m (Figure 9C), whereas marked temperature variations in the subtropical margin occurred below 200 m (Figure 8F). This difference demonstrates that the method is sensible and able to reveal the temporal variation of the heat storage dynamic in different thermal layers and compartments of the South Atlantic.

FIGURE 9

In the equatorial margin, the mixed layer and upper thermocline had strong vertical oscillations along the studied time. During MIS 1, MIS 5e, MIS 4, and some MIS 5 and MIS 3 short events with high Ti/Ca values (increased continental input), the mixed layer was deep, and the upper thermocline was less abrupt, similar to the North Brazil Current (NBC) front. These implications are supported by high abundances of warm and surface species, indicating low productivity (Figure 4). Increased abundance of G. ruber pink matching with MIS 3 peaks of Ti/Ca suggested that shelf waters influenced the equatorial margin (Schmuker and Schiebel, 2002), reinforcing that the continental input was a strong contributor to sustaining the biota (Piacsek et al., 2021). During most of MIS 3 and MIS 5, reconstructions showed a cold shallow mixed layer and a more abrupt upper thermocline, which is more similar to the ocean waters near the equator, such as the NBC retroflection (Locarnini et al., 2018; Figures 9C,D). This setup was supported by high abundances of cold water Globigerina bulloides, Globigerina falconensis, and Neogloboquadrina incompta, especially during MIS 3 interstadials (Figure 4), indicating increased productivity associated with the equatorial upwelling.

Regarding the water stratification in the equatorial margin, we made a comparison between the previously published stratification index based on planktonic foraminifera Δδ¹⁸O_dut-rub of the core GL-1248 (Venancio et al., 2018) and our stratification index extracted from Hill’s function to evaluate and compare their responses. For most of the record, both methods were in phase. However, independent trends were observed during specific time periods, such as interglacials (MIS 5e and MIS 1) and MIS 3 from 50 kyr BP to 30 kyr BP (Figure 10). These divergences could be explained through the different sources on which the indices were based. Our index extracts stratification through the thermal properties of the upper water column estimated by the Hill fit curve. The stratification index from Venancio et al. (2018) uses G. ruber and Neogloboquadrina dutertrei δ¹⁸O, two different depth-dwelling planktonic foraminifera species, whose δ¹⁸O difference (Δδ¹⁸O_dut-rub) can be linked to the temperature difference between the upper and lower photic zones. In a direct comparison between our parameters and the Δδ¹⁸O_dut-rub from Venancio et al. (2018), one observes that Δδ¹⁸O_dut-rub values are more variable than our stratification index during MIS 3, and they had an opposite trend during interglacials, especially MIS 1. Because N. dutertrei δ¹⁸O determines the Δδ¹⁸O_dut-rub variation (Venancio et al., 2018), significant vertical migrations of the upper thermocline depth can influence the temperature in the N. dutertrei habitat depth, and therefore, the imprint on δ¹⁸O. N. dutertrei could have recorded mixed-layer δ¹⁸O values in its depth habitat due to downward migration of the upper thermocline, suggested by high b values, during MIS 1 and MIS 5e, making the Δδ¹⁸O_dut-rub and our stratification index diverge. On the other hand, it is not recommended to link our thermal-based stratification index with biological productivity. A high thermal stratification may not necessarily indicate oligotrophy because a shallow upper thermocline, as suggested for MIS 3 interstadials, has little impact on our stratification index, but it can strongly affect the upper layer, which would directly affect Δδ¹⁸O_dut-rub.

FIGURE 10

Different from the equatorial margin, the upper thermal structure variation in the subtropical margin indicated that variations in the thermocline layer were linked to glacial stages (beginning of MIS 6, MIS 4, and MIS 2) as a response to heat storage (downwelling) changes in the Subtropical Gyre thermocline. The Subtropical Gyre thermal structure is characterized by a low stratification comprising warm surface temperature, a narrow or absent mixed layer, and a thick, soft slope upper thermocline that can extend down to 500 m. This thermal structure is similar to the subtropical branch of the BC (south of 23°S) that differs from the Subtropical Gyre by higher stratification, colder surface temperatures, and a shallower (down to 200 m), less soft slope upper thermocline. These areas mismatch the tropical BC (north of 23°S) structure, which has a well-defined imprint of a tropical thermal structure with a deep mixed layer and high stratification from the piled warm surface waters. Based on this huge variation in the upper thermocline slope, the b depth (migration of the upper thermocline) should not be applied to the subtropical margin, as is done in the equatorial margin. Then, we calculated the upper thermocline length (UTL) through the formula UTL = 2×(b-MLD) together with the stratification index order, to track the differences between Gyre, subtropical BC, and tropical BC structures (Figure 11D). Our data suggest that the tropical BC structure observed in the core top was not mandatory during the last 185 kyr (Figure 11). In fact, this structure was only observed during MIS 1, whereas a subtropical BC structure prevailed for most of the studied time. Outside MIS 1, tropical BC and Gyre matched with deep mixed layer events in the equatorial margin (Figures 9, 11).

FIGURE 11

Based on the data shown on Figures 4, 5, 9–11, we suggested four main settings of the upper thermal structure in the Western South Atlantic, which are illustrated in Figure 12: two interglacial scenarios for MIS 1 (Holocene) and MIS 5e (Last Interglacial) and two glacial scenarios with South Atlantic warming (glacial maxima and Dansgaard–Oeschger stadial events) and cooling (MIS 3).

FIGURE 12

The first scenario comprises the early and mid MIS 1 (Figure 12A), where a deep mixed layer and high stratification suggested a tropical structure for both equatorial and subtropical margins. NBC and BC are expressed as thick lines in Figure 12A to suggest enhanced piling of warm waters along the Brazilian border. The warm mixed layer suggested vigorous surface heat transport from the Western Equatorial Atlantic to both the Northern Hemisphere and the Brazil Current, expanding the tropical BC setup into the subtropical margin. This setup suggests that the SEC transported heat vigorously, as part of the intensified AMOC, to its branches, increasing the piling of warm waters into both BC and NBC mixed and upper thermocline layers. Such settings promoted a warming of the North Atlantic, displacing the Intertropical Convergence Zone (ITCZ) northward and weakening the SE trade winds, which reduced the downwelling capacity of the Subtropical Gyre, making it retract to the open Atlantic.

The second scenario comprises the last interglacial (MIS 5e–5b), whose thermal structure is also a tropical-like structure with a deep mixed layer and high stratification. Although the inferred setups of NBC and BC fronts were similar to MIS 1, the absolute temperature values in the upper 200 m were different, and this difference can be attributed to the different orbital settings between MIS 5 (Figure 12A) and MIS 1 (Figure 12B). Indeed, differences in both areas can be attributed to varying intensities of the seasonal cycles in response to the eccentricity. However, processes that acted on both regions were distinct. In the equatorial margin, the increased seasonality during MIS 5e increased the amount of precipitation of ITCZ and spread southward during the austral summer (Fischer and Jungclaus, 2010; Nikolova et al., 2013). Consequently, it could have made our upper thermal structure reconstructions closer to an equatorial-like structure during MIS 5e, contrasting with the strong tropical-like structure estimated for MIS 1. In the subtropical margin, low estimated surface temperatures, a shallower upper thermocline compared to MIS 1, and high abundances of G. bulloides could be linked to the offshore spreading of shelf upwelling spots. The upwelling offshore spreading was a response to the strengthening and closer position of the South Atlantic High Pressure System to the South American continent. This setup increased the influence and strength of NE winds that modulate the Southwestern Atlantic shelf upwelling, promoting its offshore spreading (Lessa et al., 2017; Fischer and Jungclaus, 2010).

Glacial scenarios exhibited not only a cold thermal structure mode but also warm thermocline episodes, which were smooth during MIS 6 and pronounced during the last glacial stage, MIS 2–4. The third scenario corresponds to the occurrence of a warm South Atlantic (Figure 12C), which represents the regional oceanic and atmospheric responses to the most intense glacial inceptions, such as Late MIS 6, MIS 4, and MIS 2. The most prominent features of that scenario were relatively warm SSTs and deepening of the upper thermocline in the equatorial margin and warm and low-stratified upper thermocline in the subtropical margin (gyre-like). This scenario shows that during the orbital-driven glacial ice sheet expansions (Late MIS 6, MIS 4, and MIS 2), the Subtropical Gyre played an important role in retaining the heat through downwelling because the AMOC was weaker during those time periods (Böhm et al., 2015). The absence of vigorous transport northward from AMOC made the South Atlantic Subtropical Gyre expand landward and influenced the thermocline layer of the BC front, impacting the biota with higher temperatures and decreased productivity. Because of this, very few thermocline-dwelling species are observed, and G. ruber white had its highest dominance (Figures 4, 5). The retained heat in the South Atlantic also affected the equatorial margin not only during ice sheet expansion periods but also in Heinrich and Dansgaard–Oeschger stadial events by displacing the ITCZ southward. The monsoonal precipitation increased in Northeast Brazil (Wang et al., 2004) and consequently increased the discharge of the Parnaiba River over our site in the equatorial margin, which became the main contributor to the productivity (Piacsek et al., 2021) at a more intense level than during ice sheet maxima expansions.

The fourth scenario corresponds to the occurrence of a cooled South Atlantic comprising most of MIS 3. It was marked by low temperatures and decreased stratification in both regions (Figure 12D). The thermal structure showed lower temperatures at a mixed layer (Figures 8B,E), decreased stratification, and a strong cold imprint on upper thermocline anomalies (Figures 8C,F). This estimated glacial thermal structure differed from recent patterns for both equatorial and subtropical margins, although the visual pattern looks like the equatorial outline (Figures 1, 7, 8). This cold imprint has a background of the increasing G. inflata and Neogloboquadrina incompta (transitional species) abundances (Figures 4, 5), and it is corroborated by other temperature proxies (Pahnke and Sachs, 2006; Santos et al., 2017; Venancio et al., 2020), especially between 50 kyr and 30 kyr when reduced eccentricity and obliquity triggered the cold MIS 3 glacial South Atlantic thermal reverse (Santos et al., 2017). In addition, the closed gateway of warm waters from the Agulhas Current and the north displaced Southern Ocean fronts could have reinforced southwesterly trade winds affecting the Subtropical Gyre circulation and contributing to cold water injection in the upper 100 m in our sites (Little et al., 1997; Marino et al., 2013; Peeters et al., 2004).

Based on those scenarios, we observe that the thermal structure of the western border of the South Atlantic is complex with responses to several oceanographic, atmospheric, and orbital forcing. Variations of the upper branch of the AMOC in response to orbital variation seem to take the main role. Such forcing encompasses the properties of heat storage in the Tropical and Subtropical Atlantic areas during the glacial and exporting heat to the Northern Hemisphere during terminations and interglacials. The second forcing is the intensity and zonality of trade winds and oceanic fronts that took the main role in control of the upper ocean thermal structure when the first forcing was decreased. Finally, regional scale forcing that took the third main role in our two study sites, but separately: the Atlantic ITCZ migration in response to millennial events in the equatorial margin and the influence of shelf upwelling expansions in the subtropical margin.

5 Conclusion

We used the Hill sigmoidal function to simulate the tropical upper ocean thermal structure and to reconstruct variation along the last 185 kyr in two sites: the western equatorial South Atlantic (equatorial margin) and western subtropical South Atlantic (subtropical margin). This method successfully fits with the upper ocean thermal structure in the two study sites using a two-step fitting with the boundary set to 200 m. The close proximity of the estimated values by MAT and the average of the 10 best analogs, as well as high r² and low standard error of the estimate (SEE), demonstrated that the method could be a reliable tool for reconstructions of ocean thermal structure. The application of the method in equatorial and subtropical margins showed different key depths to investigate heat-storing and -releasing properties during the last 185 kyr. In the equatorial margin, most temperature variations were concentrated in the upper 80–120 m, while in the subtropical margin, most temperature variations occurred throughout the thermocline layer. The different variation pattern for both interglacial and glacial periods indicated how complex the oceanography of this area is, where factors linked to Earth’s orbit, atmosphere, and changing water masses could have influenced the upper 700 m thermal structure, differing from observed recent thermal structures. Finally, physical water column parameters such as mixed layer depth and stratification estimated by Hill function coefficients could contribute to inferences related to ocean circulation changes, improving climatic and oceanographic modeling in future research.

References

• 1

  Antonarakou A. Karageorgis A. P. Besiou E. Zarkogiannis S. Drinia H. Mortyn G. P. et al (2019). Eco-biostratigraphic advances on late Quaternary geochronology and palaeoclimate: the marginal Gulf of Mexico analogue. Geol. Q.63 (1). 10.7306/gq.1457

  • CrossRef
  • Google Scholar

• 2

  Ballalai J. M. Santos T. P. Lessa D. O. Venancio I. M. Chiessi C. M. Johnstone H. J. H. et al (2019). Tracking spread of the agulhas leakage into the Western South Atlantic and its northward transmission during the last interglacial. Paleoceanogr. Paleoclimatology34, 2019PA003653. 10.1029/2019PA003653

  • CrossRef
  • Google Scholar

• 3

  Be A. W. H. (1960). Ecology of recent planktonic Foraminifera--Part 2, bathymetric and seasonal distributions in the Sargasso Sea off nermuda. Micropaleontology6 (4), 373–392. 10.2307/1484218

  • CrossRef
  • Google Scholar

• 4

  Bé A. W. H. (1977). “An ecological, zoogeographic and taxonomic review of recent planktonic foraminifera,” in Oceanic micropalaeontology. Editor RamsayA. T. S. (London: Academic Press), 1–100.

  • Google Scholar

• 5

  Bé A. W. H. Hamlin W. H. (1967). Ecology of recent planktonic foraminifera: part 3: distribution in the north Atlantic during the summer of 1962. Micropaleontology13 (1), 87–106. 10.2307/1484808

  • CrossRef
  • Google Scholar

• 6

  Becquey S. Gersonde R. (2002). “Past hydrographic and climatic changes in the subantarctic zone of the South Atlantic – the Pleistocene record from ODP site 1090.” Palaeogeogr. Palaeoclimatol. Palaeoecol.182 (3): 221–239. 10.1016/S0031-0182(01)00497-7

  • CrossRef
  • Google Scholar

• 7

  Berger A. Loutre M. F. (1991). “Insolation values for the climate of the last 10 million years.” Quat. Sci. Rev.10 (4): 297–317. 10.1016/0277-3791(91)90033-Q

  • CrossRef
  • Google Scholar

• 8

  Böhm E. Lippold J. Gutjahr M. Frank M. Blaser P. Antz B. et al (2015). Strong and deep Atlantic meridional overturning circulation during the last glacial cycle. Nature517, 73–76. 10.1038/nature14059

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 9

  Boltovskoy E. (1962). Planktonic foraminifera as indicators of different water masses in the south Atlantic. Micropaleontology8 (3), 403. 10.2307/1484531

  • CrossRef
  • Google Scholar

• 10

  Boltovskoy E. Boltovskoy D. Brandini F. (2000). Planktonic foraminifera from south-western Atlantic epipelagic waters: Abundance, distribution and year-to-year variations. J. Mar. Biol. Assoc. U. K.80 (2), 203–213. 10.1017/S0025315499001794

  • CrossRef
  • Google Scholar

• 11

  Brown D. Rothery P. (1993). Models in biology: mathematics, statistics and computing. Chichester: John Wiley and Sons Ltd.

  • Google Scholar

• 12

  Drinia H. Antonarakou A. Tsourou T. Kontakiotis G. Psychogiou M. Anastasakis G. (2016). Foraminifera eco-biostratigraphy of the southern Evoikos outer shelf, central Aegean Sea, during MIS 5 to present. Cont. Shelf Res.126, 36–49. 10.1016/j.csr.2016.07.009

  • CrossRef
  • Google Scholar

• 13

  Fischer N. Jungclaus J. H. (2010). Effects of orbital forcing on atmosphere and ocean heat transports in Holocene and eemian climate simulations with a comprehensive Earth system model. Clim. Past6 (2), 155–168. 10.5194/cp-6-155-2010

  • CrossRef
  • Google Scholar

• 14

  Giamali C. Kontakiotis G. Koskeridou E. Ioakim C. Antonarakou A. (2020). Key environmental factors controlling planktonic foraminiferal and pteropod community’s response to late Quaternary hydroclimate changes in the South Aegean sea (eastern mediterranean). J. Mar. Sci. Eng.8, 709. 10.3390/jmse8090709

  • CrossRef
  • Google Scholar

• 15

  Giamali C. Kontakiotis G. Antonarakou A. Koskeridou E. (2021). Ecological constraints of plankton bio-indicators for water column stratification and productivity: a case study of the Holocene North Aegean sedimentary record. J. Mar. Sci. Eng.9, 1249. 10.3390/jmse9111249

  • CrossRef
  • Google Scholar

• 16

  Hammer Ø. D A. T. H. Ryan P. D. (2001). Paleontological statistics software: package for education and data analysis. Palaeontol. Electron.4. Available online at: http://palaeo-electronica.org/2001_1/past/issue1_01.htm

  • Google Scholar

• 17

  Hutson W. H. (1980). The agulhas current during the late Pleistocene: analysis of modern faunal analogs. Science207 (4426), 64 LP–66. 10.1126/science.207.4426.64

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 18

  Imbrie J. van Donk J. Kipp N. G. (1973). Paleoclimatic investigation of a late Pleistocene Caribbean deep-sea core: comparison of isotopic and faunal methods. Quat. Res.3 (1), 10–38. 10.1016/0033-5894(73)90051-3

  • CrossRef
  • Google Scholar

• 19

  Jonkers L. Kučera M. (2015). Global analysis of seasonality in the shell flux of extant planktonic foraminifera. Biogeosciences12 (7), 2207–2226. 10.5194/bg-12-2207-2015

  • CrossRef
  • Google Scholar

• 20

  Juggins S. (2003). C2: software for ecological and palaeoecological data analysis and visualisation (user guide version 1.5).

  • Google Scholar

• 21

  Kontakiotis G. (2016). Late Quaternary paleoenvironmental reconstruction and paleoclimatic implications of the Aegean sea (eastern mediterranean) based on paleoceanographic indexes and stable isotopes. Quat. Int.401, 28–42. 10.1016/j.quaint.2015.07.039

  • CrossRef
  • Google Scholar

• 22

  Kucera M. (2007). Chapter six planktonic foraminifera as tracers of past Oceanic environments. Dev. Mar. Geol. De Vernal1213–262. 10.1016/S1572-5480(07)01011-1

  • CrossRef
  • Google Scholar

• 23

  Kucera M. Weinelt M. Kiefer T. Pflaumann U. Hayes A. Weinelt M. et al (2005). Reconstruction of sea-surface temperatures from assemblages of planktonic foraminifera: multi-technique approach based on geographically constrained calibration data sets and its application to glacial Atlantic and Pacific oceans. Quat. Sci. Rev.24 (7), 951–998. 10.1016/j.quascirev.2004.07.014

  • CrossRef
  • Google Scholar

• 24

  Lessa D. V. Ramos R. P. Barbosa C. F. Silva A. R Belem A. Bruno T. et al (2014). Planktonic foraminifera in the sediment of a Western boundary upwelling System off cabo Frio, Brazil. Mar. Micropaleontol.106, 55–68. 10.1016/j.marmicro.2013.12.003

  • CrossRef
  • Google Scholar

• 25

  Lessa D. V. O. Santos T. P. Venancio I. M. Albuquerque A. L. S. (2017). Offshore expansion of the Brazilian coastal upwelling zones during marine isotope stage 5. Glob. Planet. Change158 (August), 13–20. 10.1016/j.gloplacha.2017.09.006

  • CrossRef
  • Google Scholar

• 26

  Lessa D. V. O. Santos T. P. Venancio I. M. Santarosa A. C. A. dos Santos Junior E. C. Toledo F. A. L. et al (2019). Eccentricity-induced expansions of Brazilian coastal upwelling zones. Glob. Planet. Change179 (October 2018), 33–42. 10.1016/j.gloplacha.2019.05.002

  • CrossRef
  • Google Scholar

• 27

  Lessa D. Morard R. Jonkers L. Venancio I. M. Reuter R. Baumeister A. et al (2020). Distribution of planktonic foraminifera in the subtropical south Atlantic: depth hierarchy of controlling factors. Biogeosciences17 (16), 4313–4342. 10.5194/bg-17-4313-2020

  • CrossRef
  • Google Scholar

• 28

  Little M. G. Schneider R. R. Kroon D. Price B. Summerhayes C. P. Segl M. (1997). Trade wind forcing of upwelling, seasonality, and heinrich events as a response to sub-milankovitch climate variability. Paleoceanography12 (4), 568–576. 10.1029/97PA00823

  • CrossRef
  • Google Scholar

• 29

  Locarnini M. Mishonov A. Ok B. Boyer T. Zweng M. Garcia H. et al (2018). World ocean atlas 2018. Available online at: https://archimer.ifremer.fr/doc/00651/76338/.

  • Google Scholar

• 30

  Lončarić N. Peeters F. J. C. Kroon D. Brummer G. J. A. (2006). Oxygen isotope ecology of recent planktic foraminifera at the central walvis ridge (SE Atlantic). Paleoceanography21 (3), 1–18. 10.1029/2005PA001207

  • CrossRef
  • Google Scholar

• 31

  Malmgren B. A. Nordlund U. (1997). Application of artificial neural networks to paleoceanographic data. Palaeogeogr. Palaeoclimatol. Palaeoecol.136 (1): 359–373. 10.1016/S0031-0182(97)00031-X

  • CrossRef
  • Google Scholar

• 32

  Malmgren B. A. Kucera M. Nyberg J. Waelbrocck C. (2001). Comparison of statistical and artificial neural network techniques for estimating past sea surface temperatures from planktonic foraminifer census data. Paleoceanography16 (5), 520–530. 10.1029/2000PA000562

  • CrossRef
  • Google Scholar

• 33

  Marino G. Zahn R. Ziegler M. Purcell C. Knorr G. Hall I. R. et al (2013). Agulhas salt-leakage oscillations during abrupt climate changes of the late Pleistocene. Paleoceanography28 (3), 599–606. 10.1002/palo.20038

  • CrossRef
  • Google Scholar

• 34

  Meilland J. Siccha M. Weinkauf M. F. G. Jonkers L. Morard R. Baranowski U. et al (2019). Highly replicated sampling reveals no diurnal vertical migration but stable species-specific vertical habitats in planktonic foraminifera. J. Plankton Res.41 (2), 127–141. 10.1093/plankt/fbz002

  • CrossRef
  • Google Scholar

• 35

  Mikrotax (2025). Mikrotax. Available online at: https://www.mikrotax.org/pforams/(Accessed December 02, 2025).

  • Google Scholar

• 36

  Nikolova I. Yin Q. Berger A. Singh U. K. Karami M. P. (2013). The last interglacial (Eemian) climate simulated by LOVECLIM and CCSM3. Clim. Past.9 (4), 1789–1806. 10.5194/cp-9-1789-2013

  • CrossRef
  • Google Scholar

• 37

  Nascimento R. A. Venancio I. M. Chiessi C. M. Ballalai J. M. Kuhnert H. Johnstone H. et al (2021). Tropical atlantic stratification response to late quaternary precessional forcing. Earth Planet. Sci. Lett.568, 117030. 10.1016/j.epsl.2021.117030

  • CrossRef
  • Google Scholar

• 38

  Pahnke K. Sachs J. P. (2006). Sea surface temperatures of southern midlatitudes 0-160 Kyr B.P. Paleoceanography21 (2), 1–17. 10.1029/2005PA001191

  • CrossRef
  • Google Scholar

• 39

  Peeters F. J. C. Acheson R. Brummer G. J. A. De Ruijter W. P. M. Schneider R. R. Ganssen G. M. et al (2004). Vigorous exchange between the Indian and Atlantic oceans at the end of the past five glacial periods. Nature430 (7000), 661–665. 10.1038/nature02785

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 40

  Peterson R. G. Stramma S. (1991). Upper-level circulation in the south Atlantic Ocean. Prog. Oceanogr.26 (1), 1–73. 10.1016/0079-6611(91)90006-8

  • CrossRef
  • Google Scholar

• 41

  Pflaumann U. Duprat J. Pujol C. Labeyrie L. D. (1996). SIMMAX: a modern analog technique to deduce Atlantic sea surface temperatures from planktonic foraminifera in deep-sea sediments. Paleoceanography11 (1), 15–35. 10.1029/95PA01743

  • CrossRef
  • Google Scholar

• 42

  Piacsek P. Behling H. Gu F. Venancio I. M. Lessa D. V. O. Belem A. et al (2021). Changes in sea surface hydrography and productivity in the Western equatorial Atlantic since the last interglacial. Palaeogeogr. Palaeoclimatol. Palaeoecol.562:109952. 10.1016/j.palaeo.2020.109952

  • CrossRef
  • Google Scholar

• 43

  Poole R. Tomczak M. (1999). Optimum multiparameter analysis of the water mass structure in the Atlantic Ocean thermocline. Deep-Sea Res. Part I Oceanogr. Res. Pap.46 (11), 1895–1921. 10.1016/S0967-0637(99)00025-4

  • CrossRef
  • Google Scholar

• 44

  Portilho-Ramos R. Pinho T. M. L. Chiessi C. M. Barbosa C. F. (2018). Interactive comment on Understanding the mechanisms behind high glacial productivity in the southern Brazilian margin. Clim. Past Discuss., 1–19. 10.5194/cp-2018-98

  • CrossRef
  • Google Scholar

• 45

  Ramos P. da Costa R. Ferreira F. Calado L. Frontalini F. Toledo M. B. (2015). Variability of the upwelling system in the southeastern Brazilian margin for the last 110,000 years. Glob. Planet. Change135, 179–189. 10.1016/j.gloplacha.2015.11.003

  • CrossRef
  • Google Scholar

• 46

  Ramos P. da Costa R. Marcos Lima Pinho T. Chiessi C. M. Barbosa C. F. (2019). Understanding the mechanisms behind high glacial productivity in the southern Brazilian margin. Clim. Past15 (3), 943–955. 10.5194/cp-15-943-2019

  • CrossRef
  • Google Scholar

• 47

  Rebotim A. Voelker A. H. L. Jonkers L. Waniek J. J. Meggers H. Schiebel R. et al (2017). Factors controlling the depth habitat of planktonic foraminifera in the subtropical eastern north Atlantic. Biogeosciences14 (4), 827–859. 10.5194/bg-14-827-2017

  • CrossRef
  • Google Scholar

• 48

  Sánchez A. Carriquiry J. (2024). Thermocline-level connectivity between the tropical equatorial Pacific and tropical northeast Pacific during deglaciation. Quat. Int.705, 86–93. 10.1016/j.quaint.2024.03.010

  • CrossRef
  • Google Scholar

• 49

  Sánchez A. Sanchez-Vargas L. Balart E. Dominguez-Samalea Y. (2022). Stable oxygen isotopes in planktonic foraminifera from surface sediments in the California current system. Mar. Micropaleontol.173, 102127. 10.1016/j.marmicro.2022.102127

  • CrossRef
  • Google Scholar

• 50

  Santos T. P. Douglas O. L. Venancio I. M. Chiessi C. M. Mulitza S. Kuhnert H. et al (2017). Prolonged warming of the Brazil current precedes deglaciations. Earth Planet. Sci. Lett.463 (April):1–12. 10.1016/j.epsl.2017.01.014

  • CrossRef
  • Google Scholar

• 51

  Santos T. P. Ballalai J. M. Franco D. R. Oliveira R. R. Lessa D. O. Venancio I. M. et al (2020). Asymmetric response of the subtropical Western south Atlantic thermocline to the dansgaard-oeschger events of marine isotope stages 5 and 3. Quat. Sci. Rev.237:106307. 10.1016/j.quascirev.2020.106307

  • CrossRef
  • Google Scholar

• 52

  Schiebel R. Hemleben C. (2017). Planktic foraminifers in the modern ocean. Springer.

  • Google Scholar

• 53

  Schlitzer R. (2015). Data analysis and visualization with ocean data view. CMOS Bull. SCMO43 (1), 9–13. Available online at: http://www.cmos.ca/site/cmos%5C_bulletin.

  • Google Scholar

• 54

  Schmid C. (2014). Mean vertical and horizontal structure of the subtropical circulation in the south Atlantic from three-dimensional observed velocity fields. Deep-Sea Res. Part I Oceanogr. Res. Pap.91, 50–71. 10.1016/j.dsr.2014.04.015

  • CrossRef
  • Google Scholar

• 55

  Schmuker B. Schiebel R. (2002). Planktic foraminifers and hydrography of the eastern and northern Caribbean Sea. Mar. Micropaleontol.46 (3-4), 387–403. 10.1016/s0377-8398(02)00082-8

  • CrossRef
  • Google Scholar

• 56

  Scussolini P. Marino G. Brummer G. J. A. Peeters F. J. C. (2015). Saline Indian Ocean waters invaded the south Atlantic thermocline during glacial termination II. Geology43 (2), 139–142. 10.1130/G36238.1

  • CrossRef
  • Google Scholar

• 57

  Siccha M. Kucera M. (2017). ForCenS, a curated database of planktonic foraminifera census counts in marine surface sediment samples. Sci. Data4 (1), 170109. 10.1038/sdata.2017.109

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 58

  Siccha M. Trommer G. Schulz H. Hemleben C. Kucera M. (2009). Factors controlling the distribution of planktonic foraminifera in the Red Sea and implications for the development of transfer functions. Mar. Micropaleontol.72 (3–4), 146–156. 10.1016/j.marmicro.2009.04.002

  • CrossRef
  • Google Scholar

• 59

  Sousa S. H. M. de Godoi S. S. Amaral P. G. C. Vicente T. M. Martins M. V. A. Sorano M. R. G. S. et al (2014). Distribution of living planktonic foraminifera in relation to oceanic processes on the southeastern continental brazilian margin (23°S-25°S and 40°W-44°W). Continental Shelf Res.89, 76–87. 10.1016/j.csr.2013.11.027

  • CrossRef
  • Google Scholar

• 60

  Sprintall J. Matthias T. (1992). Evidence of the barrier layer in the surface layer of the tropics. J. Geophys. Res. Oceans97 (C5): 7305–7316. 10.1029/92JC00407

  • CrossRef
  • Google Scholar

• 61

  Stramma L. England M. (1999). On the water masses and mean circulation of the south Atlantic Ocean. J. Geophys. Res. Oceans104 (C9), 20863–20883. 10.1029/1999JC900139

  • CrossRef
  • Google Scholar

• 62

  Telford R. J. Birks H. J. B. (2005). The secret assumption of transfer functions: problems with spatial autocorrelation in evaluating model performance. Quat. Sci. Rev.24 (20): 2173–2179. 10.1016/j.quascirev.2005.05.001

  • CrossRef
  • Google Scholar

• 63

  Telford R. J. Birks H. J. B. (2009). Evaluation of transfer functions in spatially structured environments. Quat. Sci. Rev.28 (13): 1309–1316. 10.1016/j.quascirev.2008.12.020

  • CrossRef
  • Google Scholar

• 64

  Telford R. J. Birks H. J. B. (2011). A novel method for assessing the statistical significance of quantitative reconstructions inferred from biotic assemblages. Quat. Sci. Rev.30 (9–10), 1272–1278. 10.1016/j.quascirev.2011.03.002

  • CrossRef
  • Google Scholar

• 65

  Telford R. J. Li C. Kucera M. (2013). Mismatch between the depth habitat of planktonic foraminifera and the calibration depth of SST transfer functions may bias reconstructions. Clim. Past9 (2), 859–870. 10.5194/cp-9-859-2013

  • CrossRef
  • Google Scholar

• 66

  Toledo F. A. L. Costa K. B. Pivel M. A. G. (2007). Salinity changes in the Western tropical south Atlantic during the last 30 Kyr. Glob. Planet. Change57 (3–4), 383–395. 10.1016/j.gloplacha.2007.01.001

  • CrossRef
  • Google Scholar

• 67

  Venancio I. M. Franco D. Belem A. L. Mulitza S. Siccha M. Albuquerque A. L. S. et al (2016). Planktonic foraminifera shell fluxes from a weekly resolved sediment trap record in the southwestern atlantic: evidence for synchronized reproduction. Marine Micropaleont.125, 25–35. 10.1016/j.marmicro.2016.03.003

  • CrossRef
  • Google Scholar

• 68

  Venancio I. M. Mulitza S. Govin A. Santos T. P. Lessa D. O. Albuquerque A. L. S. et al (2018). Millennial-to orbital-scale responses of Western equatorial Atlantic thermocline depth to changes in the trade wind system since the last interglacial. Paleoceanogr. Paleoclimatology33 (12), 1490–1507. 10.1029/2018PA003437

  • CrossRef
  • Google Scholar

• 69

  Venancio I. M. Shimizu M. H. Santos T. P. Lessa D. O. Portilho-Ramos R. C. Chiessi C. M. et al (2020). Changes in surface hydrography at the Western tropical Atlantic during the younger dryas. Glob. Planet. Change184 (April 2019), 103047. 10.1016/j.gloplacha.2019.103047

  • CrossRef
  • Google Scholar

• 70

  Waelbroeck C. Labeyrie L. Duplessy J.-C. Guiot J. Labracherie M. Leclaire H. et al (1998). Improving past sea surface temperature estimates based on planktonic fossil faunas. Paleoceanography13 (3), 272–283. 10.1029/98PA00071

  • CrossRef
  • Google Scholar

• 71

  Wang X. Auler A. S. Lawrence Edwards R. Cheng H. Cristalli P. S. Smart P. L. et al (2004). Wet periods in northeastern Brazil over the past 210 Kyr linked to distant climate anomalies. Nature432 (7018), 740–743. 10.1038/nature03067

  • Pubmed Abstract
  • CrossRef
  • Google Scholar