Evolution of Pliocene-Pleistocene tropical terrestrial Andean temperature amplification

Significance

The tropics are home to 40% of the world’s population, and countries in the tropics are ill-equipped to adapt to future climate change. While the Pliocene epoch is one of the best climatic analogues to explore present and future change, it remains understudied in the tropics. We present a temperature record from the northern tropical Andes, at 4°N of the equator, and show that land temperature changes are similar to, but larger than ocean temperatures over ~4 My. The “excess” in temperature response on tropical land points to important regional feedbacks that may amplify climate change in the future.

Abstract

The Pliocene is the most recent epoch in which the Earth warmed under atmospheric CO₂ levels similar to today (>400 ppm). The Pliocene then transitioned to the colder Pleistocene epoch, with the initiation of large-scale Northern Hemisphere glaciations. Although ocean temperature changes across these epochs are relatively well-known, quantitative estimates of the magnitude of land temperature change in the tropics are scarce. We provide a Plio-Quaternary quantitative air temperature record based on the distribution of bacterial branched glycerol dialkyl glycerol tetraethers (brGDGTs) preserved in sediments of the Funza-II core in the Sabana de Bogotá, Colombia (~4°N). Using a refined age model based on new U-Pb zircon dates from ash layers, and a novel mixed-source model that disentangles contributions from lake- and soil-derived brGDGTs, we show that warm Pliocene (3.8 to 2.58 Ma) temperatures were $4.8_{-1.4}^{+1.8}$ °C warmer than the last ~800,000 y of the colder Pleistocene. The evolution of Pliocene-Pleistocene temperature in our record largely mirrors long-term tropical sea surface temperature (SST) cooling, highlighting the linkages between sea and land temperatures in the low latitudes via greenhouse-gas forcing. The median amplitude of Pliocene-Pleistocene cooling in the northern tropical Andes exceeds that predicted by theory, highlighting the importance of regional feedbacks including lapse rate adjustments and/or changes in Pacific SST gradients to the long-term evolution of Andean temperature. This first quantitative terrestrial temperature reconstruction within 5° of the equator over the past 3.8 My highlights that both regional and global processes must be considered when constraining uncertainties for future warming scenarios.

In recent decades, land temperatures in regions between 40°N and 40°S have warmed 42% faster and ~1.5 °C more than ocean temperatures (1, 2). Model simulations predict that both the rate and amplitude of land warming will continue to exceed that of the ocean under different scenarios of global warming (1, 3). In the geologic past, analyses of terrestrial amplification, the land-to-sea surface warming ratio, in the tropics have focused on the last glacial maximum (LGM) (4), a period with CO₂ concentrations around 180 ppm (5). Although estimates of terrestrial warming from the LGM to pre-industrial appear consistent with terrestrial amplification theory, it is not certain whether the land-to-sea ratio calculated for the LGM can be linearly extrapolated to warmer climate states with atmospheric CO₂ levels higher than pre-industrial.

The Pliocene, from ~5.3 to 2.58 Ma, is the most recent epoch when Earth’s mean temperature was ~2.5 °C to 4 °C warmer than today, CO₂ concentrations were higher than pre-industrial levels and the northern hemisphere was largely unglaciated (6–10). The warm climate state of the Pliocene makes it one of the best recent analogues to explore forcings and feedback mechanisms governing future climates given its similarity in CO₂ concentrations to today (11–13). Although the tropics play an important role in regulating global climate, most inferences of Pliocene surface temperatures are derived from ocean sediments or high-latitude continental records, whereas existing quantitative land temperature records include no sites within 10° of the equator (9). The lack of tropical land-surface temperature records during this time limits our understanding of terrestrial temperature evolution under high-CO₂, warmer climates.

We constrain the magnitude of surface temperature changes since Pliocene time in the northern tropical Andes using a novel statistical approach to determine the probability distribution of soil- and lake-derived brGDGT sources in a unique, long (~580 m) sediment core from the Bogotá basin. We use the inferred mixtures to obtain brGDGTs-based temperature estimates from the Pliocene to the late Pleistocene and compare those to long-term changes in tropical sea surface temperatures [SSTs; (14)]. Our new Pliocene-Pleistocene quantitative land temperature reconstruction provides unprecedented insight into tropical terrestrial temperature amplification and climate feedbacks in a globally warm climate state.

The Funza-II Core Record

The Eastern Cordillera of Colombia preserves upper Miocene-Pleistocene sediment successions with paleosols, fluvial, and lacustrine deposits (15). These sedimentary archives include a drill-core retrieved in 1988 (Funza-II; 4°59’N, 74°22’W, 2,550 m.a.s.l., Fig. 1) from the deepest part of the Bogotá basin, from which ~580 m of sediment were retrieved (16, 17). Fossil pollen assemblages and sedimentological studies show that the Funza-II core was deposited after the high plain of Bogotá reached its present altitude and with a flat-lying geometry, indicative of a slight basin subsidence [see SI Appendix for details; (18, 19)].

Fig. 1.

Study area and chronology of the Funa-II core. (A) Topographic map of the northern tropical Andes showing the location of the Funza-II core. (B) Study area (black square in the Inset) in reference to North, Central, and South America. (C) Age model of the Funza-II core derived from new zircon U-Pb chronology from this study.

Previous terrestrial Pliocene temperature estimates derived from paleobotanical proxies at low latitudes have shown colder to no difference in temperatures relative to pre-industrial times (9). However, these estimates are usually based on a single comparison point with pre-industrial or late Holocene values rather than on a continuous reconstruction of temperature since the Pliocene. Because the Sabana de Bogotá has remained at its present-day elevation during the deposition of the sequences sampled by the Funza-II core (19), our quantitative temperature reconstruction from this almost continuous Pliocene-Pleistocene record is critical for our understanding of long-term tropical climate dynamics and terrestrial amplification during this time without the influence of tectonic processes.

Results and Discussion

Zircon U-Pb Geochronology for the Funza-II Core.

We report a new age model derived from chemical abrasion thermal ionization mass spectrometry (CA-TIMS) absolute dates from ash layers in the Funza-II record. The samples used for the age model (Fig. 1C) have depths and CA-TIMS dates of 154.96 m (1.118 ± 0.026 Ma), 264.95 m (1.555 ± 0.013 Ma), 306.85 m (1.860 ± 0.025 Ma), 438.67 m (2.657 ± 0.043 Ma), and 507.11 m (3.371 ± 0.094 Ma) (see SI Appendix for details). This new age model indicates relatively constant sedimentation rates and predicts that sediment in the bottom ~150 m of the core accumulated during the Pliocene epoch (~3.8 to 2.58 Ma), which is ~1 to 1.5 Ma older than a previous model based on orbital tuning (17, 20). Our age model (Fig. 1C) indicates that the bottom of the Funza-II core preserves the Piacenzian (from ~3.6 to 2.58 Ma), while the top ~250 m of the Funza-II are assigned a mid to late Pleistocene age. Because these deposits may contain hiatuses that could obscure orbital features and compromise previous age models (SI Appendix for details), we based our climatic interpretations of the Funza-II core on our new age model based on absolute dates from ash layers.

Probability Distribution of brGDGTs Sources for Temperature Reconstruction.

Because the depositional environment of the Funza-II core fluctuated between swamp, lacustrine, fluvio-lacustrine, and fluvial deposits (17), the sediment preserved in the Funza-II core is made up of organic matter from varying sources and environments from the Sabana de Bogotá during the last ~3.8 My. Thus, brGDGTs preserved in this sedimentary sequence represent a mixture of sources from soil organic matter transported by rivers as well as in-situ, aquatic bacterial biomass. Therefore, a brGDGT-derived temperature reconstruction from this archive must not only take this mixing into account, but also the application of both soil-based and lacustrine-based temperature calibrations. This approach yields more accurate temperature values than those derived from a single-source calibration, as observed in other studies (21, 22). For instance, when using single calibrations based on MBT’_5me in our record for both lacustrine (23) and peat (24) samples, the largest change in temperature at ~350 m (~2.12 Ma; SI Appendix, Fig. S5) coincides with changes in the MBT’_5me index and brGDGT sources, which indicates a transition in the depositional environment, not a change in temperature (SI Appendix, Figs. S2 and S4).

To evaluate the sources and mixing of brGDGTs in this sedimentary archive, we compared the distribution of the summed pentamethylated, tetramethylated, and hexamethylated brGDGTs against those from tropical soils (25) and tropical lakes (23) (SI Appendix, Fig. S1A). We used modern samples only from the tropics to avoid biases in temperature estimates from regions with strong seasonality. Since brGDGTs in the Funza-II core fall between those of tropical soils and tropical lakes (SI Appendix, Fig. S1A), we used a logistic regression model (Eq. 1) that calculates the probability that brGDGTs in each core sample belong in these two categories, or end-members, and consequently are a mixture of both. For example, if the probability that a core sample is lake-derived is P_lake = 0.3 (or 30%), then the probability it is soil-derived is 0.7 (or 70%).

$$P\left(x\right)=\frac{e^{-47.54+45.05\ast Ia+27.13\ast Ib+74.2\ast Ic+45.23\ast IIa+130\ast IIb+52.77\ast II{a}^{\prime }+61.32\ast II{b}^{\prime }+55.86\ast IIIa+45.82\ast III{a}^{\prime }}}{1+e^{-47.54+45.05\ast Ia+27.13\ast Ib+74.2\ast Ic+45.23\ast IIa+130\ast IIb+52.77\ast II{a}^{\prime }+61.32\ast II{b}^{\prime }+55.86\ast IIIa+45.82\ast III{a}^{\prime }}}.$$ [1]

This approach allowed us to calculate the probabilities of lake-derived and soil-derived brGDGT end-members of the modern lake and soil samples using Eq. 1 (SI Appendix, Fig. S1B). We tested the efficacy of our logistic regression model by predicting the probability distributions for the original modern tropical samples used to build the model (SI Appendix, Fig. S1 B and C). The model predicted that 84% of modern lake samples have a probability distribution higher than 0.5 (SI Appendix, Fig. S1 B and C), closer to the lacustrine brGDGT end-member. For the modern soils, the model predicted that 95% of the samples have a probability distribution between 0 and 0.5, corresponding to the soil end-member (SI Appendix, Fig. S1 B and C).

We applied the mixing model to the brGDGTs from the Funza-II core samples to estimate brGDGT sources and temperature. The bottom ~300 m of the Funza-II core (~295 to 585 m depth) have been interpreted to represent a mixture of swamp and wetland deposits with fluvial intervals, and the interval between 280 and 340 m constitutes a transition towards a more lacustrine environment [SI Appendix, Fig. S2, (17)]. The probability distribution of brGDGT sources shows that 89% of samples between 340 and 585 m, i.e., before the lake expansion, are more than 50% soil-derived (or yield a probability density distribution between 0 and 0.5), with the remaining 11% of the samples representing lake-derived brGDGTs. In contrast, samples between 210 and 280 m depth consist of 70 to 100% lake-derived brGDGTs (SI Appendix, Fig. S2A). This depth interval of the Funza-II core has been interpreted to reflect a fully developed lake environment based on the high spore abundance of the aquatic Isoëtes [or quillworts; (17)], which validates the performance of our mixed-source model (26) with independent proxies (Fig. 2A and SI Appendix, Fig. S2). In addition to our mixed-source approach, we tested the BIGMaC algorithm (22) in the Funza-II core (SI Appendix, Fig. S6A) by using peat-based and lacustrine-based brGDGT calibrations (SI Appendix, Fig. S5A). Temperature estimates from the BIGMaC results yield unrealistic estimates for our record (SI Appendix, Figs. S5A and S6B); Mean Annual Air Temperature (MAAT) values from BIGMaC are as low as −2.8 °C and have a high root-mean-squared-error (RMSE) of 4.7 °C (SI Appendix, Figs. S5A and S6B). Therefore, we argue that the soil-brGDGT regional calibration for the northern Andes (25) yields the most accurate temperature estimates for intervals dominated by non-lacustrine brGDGTs in the Funza-II core. As a result, to evaluate the evolution of surface air temperature since ~4 Ma in the northern tropical Andes, we use the regional calibration (25) for soil-derived brGDGTs and the tropical lacustrine calibration (23) for lake-derived brGDGTs, and their mixture (SI Appendix).

Fig. 2.

Br-GDGT-derived temperature reconstruction from the Funza-II core compared to the tropical SST stack from ref. 14 and atmospheric CO₂ estimates from the last ~4 My (10). (A) Predicted probability of the percentage of lake-derived brGDGTs present in samples of the Funza-II core versus Age (Ma) using our mixed-source approach (binomial logistic regression). (B) Tropical Land Surface Temperature changes (ΔLSTs) from the Funza-II core using combined calibrations for soil-derived (25) and lake-derived brGDGTs (23) after the binomial logistic regression. (C) Tropical sea surface temperature change (ΔSST) stack from Clark et al. (14). (D) Atmospheric CO₂ proxy estimates (gray circles) and 100,000 y smoothed average (magenta line) from Cenozoic CO₂ Proxy Integration Project (CenCO₂PIP) Consortium (2023) along with the EPICA CO₂ record [blue line; (27)]. Gray intervals represent the MPT from 0.8 to 1.5 Ma and the Pliocene (>2.58 Ma). The red horizontal line in (B) and (C) represents the late Holocene temperature estimates for each record.

A Quantitative Record of Terrestrial Tropical Temperature over the Past 3.8 My.

Land surface temperature (LST) estimates from the northern tropical Andes show long-term cooling since the Pliocene, similar to the tropical SST stack (Figs. 2 B and C and 3A). In this region of the tropical Andes, late Holocene brGDGT-based temperature estimates average 14.9 °C ± 1.5 °C (see Materials and Methods for details). The reconstructed temperatures from Pliocene samples in the region exhibit a median value that is $3.7_{-1.4}^{+1.8}$ °C warmer than our late Holocene estimate. Temperature estimates colder than the late Holocene mostly occur during the last 800,000 y of the Pleistocene with a median temperature of $-1.1_{-1.7}^{+1.8}$ °C (Fig. 2B). Similar to the tropical SST stack, our brGDGT-based temperature reconstruction shows no rapid cooling during the intensification of northern hemisphere glaciation (iNHG) ~2.6 Ma, but rather documents gradual cooling across and after this transition.

Fig. 3.

Comparison of tropical surface temperature changes between ocean and land. (A) Boxplots show the proxy-based estimated distribution of tropical [SST; (14)] and LST (this study) anomalies normalized to the median of the last 800,000 y for three intervals; Last 800,000 y of the Pleistocene (0.03 to 0.8 Ma), Pleistocene before 800,000 y (0.8 to 2.58 Ma), and Pliocene (2.58 to 4 Ma for SST and 2.58 to 3.8 Ma for LST). The medians of the 0.03 to 0.8 Ma boxplots in SST and LST are normalized to zero (0). Overlaid black circles are the bootstrap-derived median, with vertical error bars showing ± 1-σ. Small dark-red triangles represent the first (Q1) and third (Q3) quartiles of the bootstrap distribution. (B) Land-sea warming temperature comparison between our brGDGT-based temperature estimates (dark blue) and calculations based on (4) (orange) and the mean values of the 4 × CO₂ (yellow) and SSP5-8.5 (purple) scenarios of the CMIP6 models. The ΔSST is based on the difference between the median temperature value of the last 800,000 y [left-most boxplot in (A)] and the median SST Pliocene [2.58 to 4 Ma, third boxplot from left to right in (A)] value (circles), and the difference between the late Holocene estimate and the median SST Pliocene value (triangles). The ΔLST error bars in our study are the first and third quartile of the bootstrap distribution in Fig. 3A.

During the last 800,000 y of the Pleistocene, our median LST estimates were $4.8_{-1.4}^{+1.8}$ °C colder than the Pliocene (Fig. 3A). Though our core does not preserve the LGM, and our temperature estimates lack orbital (high) resolution, the observed LST range over the last ~800,000 y (top 160 m, 5.0 °C ± 3.1 °C, SI Appendix, Fig. S7) is similar in magnitude to previous higher resolution studies of glacial-interglacial records in Andean tropical areas (28–31) as well as in tropical African mountains (32, 33). Previous work in the Eastern Cordillera of Colombia using pollen assemblages in the Fúquene lake (31) suggested that the LGM was 8 °C cooler than today, and that a 4 °C change in temperature occurred from MIS 5 to MIS 4. Subsequent ultra-high resolution pollen analyses at the same lake estimated a warming of 7 °C to 10 °C from MIS 6 to MIS 5 and from MIS 8 to MIS 7, and a similar magnitude of temperature variability during MIS 3 (29). More recently, analyses of fossil pollen in the Funza09 record (34) suggested that temperatures at the upper forest line (~2,550 m.a.s.l.) fluctuated up to 8 °C between glacial and interglacials in the top 250 m of the core (last ~1 My based on their age model). The variability in brGDGT-derived temperature over the last 800,000 y of the Pleistocene in the Funza-II core is similar to that inferred from pollen assemblages in the northern tropical Andes (29, 31, 34) and other terrestrial temperature reconstructions from South America (see SI Appendix for details), lending credence to our reconstruction.

Tropical Land-to-Sea Surface Temperature Comparison and Amplification.

To assess changes in the land-to-sea temperature ratio between the warm climate state of the Pliocene and the mid to late Pleistocene (i.e., the most recent ~800,000 y), when global temperature variations were dominated by glacial cycles and CO₂ reconstructions are more reliable (10), and between the Pliocene and the late Holocene, we compared the LST estimates in our record to tropical SSTs from Clark et al. (14) (Fig. 2 B and C). This tropical SST record is based on alkenones, Mg/Ca, and faunal proxies from all ocean basins. In addition to the late Holocene, we also use the last 800,000 y of the Pleistocene as an interval of comparison because, a) tropical SSTs stabilized at this time (14) and, b) our record has a higher number of samples to derive a more robust median temperature estimate than the late Holocene. We exclude the last 30,000 y of the tropical SST stack as the Funza-II core does not preserve this time interval (Fig. 3A). However, this does not impact our results and interpretation.

The brGDGT-based temperature estimates in the northern tropical Andes exhibit a similar long-term cooling as seen in the tropical SST stack since 4 Ma [Fig. 3A; (14)]. In the tropical and global SST compilation (14), temperatures started to cool at 4 Ma and stabilized at ~800,000 y [or after the Middle Pleistocene Transition (MPT)], when cooling coincided with a shift in the frequency of glacial-interglacial variability from 41,000 to 100,000 y. Similarly, the coldest interval in our record is the last ~800,000 y of the Pleistocene, when 69% of our temperature estimates are below the late Holocene (Fig. 2B and SI Appendix, Fig. S7). For Pliocene LSTs, all estimates are above our late Holocene estimate (Fig. 2B). There is no abrupt temperature change at the iNHG, which is consistent with other low latitude SST records (14, 35).

The main driver of changes in SST in low latitudes during the late Pliocene and since 4 Ma is thought to be atmospheric pCO₂ [Fig. 2D, (14, 36, 37)]. In particular, results from climate model simulations of the Pliocene (37) indicate that changes in CO₂ concentrations drive 36 to 61% of the global temperature changes, and are particularly important for temperature change in the low latitudes due to the greater importance of feedbacks including ice cover, vegetation, and orography in the mid and high latitudes. Additionally, an energy balance analysis of low-latitude temperature variations showed that CO₂ forcing contributed more to low latitude Pliocene warming than cloud feedbacks (37). The overall similarities in long-term cooling between the temperature record from the northern tropical Andes and the tropical SST stack over the past 4 My (14), and the 100,000 y pCO₂ smoothed average record (10) support the hypothesis that tropical terrestrial temperatures also respond to changes in greenhouse gas forcing during the last ~4 My.

Our terrestrial record shows a larger magnitude of cooling over the last ~4 My compared to the SST stack (Fig. 3). This is due at least in part to the low heat capacity of land relative to the ocean and other physical processes that cause terrestrial amplification (4, 38). A theoretical basis for terrestrial temperature amplification at low latitudes during the LGM, referred to as the BO18 theory (4), was developed using recent climatic observations of amplified warming (1). Relative to the ocean, land surfaces have lower specific heat capacity and lower evaporation rates, both of which are linked to differences in the moist static energy (MSE) of air above ocean and land (1, 3, 39). Changes in MSE near the surface are uniform across the tropics, and therefore equal over land and ocean. However, the moisture availability over land is lower than over the ocean, so a larger temperature increase over land is needed to maintain uniform MSE. This theory is most applicable at low latitudes where convection homogenizes MSE under moist adiabatic conditions and where the coriolis effect is weak enough to promote the spatial homogenization of temperature. This theory is also most applicable at elevations below 1,000 m.a.s.l. Because possible adjustments in lapse rate could amplify the temperature variations at higher elevations, this is particularly important for cold and dry climates where shifts from moist to dry adiabatic lapse rate could exist (e.g., LGM). However, the extent of the magnitude of lapse rate adjustment in a warmer climate like the Pliocene remains understudied.

We use the BO18 theoretical framework for the LGM ($1.{37}_{-0.23}^{+0.27}$) to evaluate if this land-to-sea ratio of temperature amplification, which is derived from a cold climate state, persists in low latitudes during the warm climate state of the Pliocene with CO₂ concentrations around 400 ppm. We use median temperature estimates from the warm Pliocene (3.8 to 2.58 Ma) and the last 800,000 y of the cold Pleistocene, as well as comparisons to the late Holocene. The tropical SST stack (14) yields a median cooling of ${2.95}_{-0.52}^{+0.49}$ °C from the Pliocene to the median temperature value of the last 800,000 y of the Pleistocene, whereas the temperature record in the northern tropical Andes indicates a cooling of $4.8_{-1.4}^{+1.8}$ °C (Fig. 3A). Pliocene tropical median SSTs are ${1.91}_{-0.52}^{+0.49}$ °C warmer than late Holocene estimates, compared to the ~3.7 °C of the LSTs in our study. Both of these median estimates in land warming yield values that are higher than predictions from the land-to-sea ratio determined for the LGM (4) and mean values of the land-sea warming ratios of the 4 × CO₂ and SSP5-8.5 scenarios from CMIP6 models (Fig. 3B). The median temperature amplification between Pliocene and the last 800,000 y of the Pleistocene, and between Pliocene and late Holocene estimates, in the northern tropical Andes is 1.63 ± 0.62 and 1.94 ± 1.26 times that of ocean temperatures in low latitudes, respectively (Fig. 3B). Our median temperatures suggest that the land-to-sea warming ratio in general is larger in the northern tropical Andes during the warm climate state of the Pliocene relative to that of the colder LGM estimates (4). However, given that our record does not preserve the LGM, we cannot conclude if this magnitude of temperature amplification is higher at low latitudes during periods with high CO₂ concentrations, such as the Pliocene, or if the increase in the land-to-sea warming ratio prevailed in this region across time intervals with low CO₂ concentrations. To investigate regional feedbacks that could play an additional role in the land temperature amplification seen in the northern tropical Andes, we explore the tropical lapse-rate adjustment during the Pliocene and temperature responses from Pacific SST gradients.

During the LGM, tropical mountains exhibit amplified cooling with elevation due to a steeper lapse rate than today, which results from cooling, decreased atmospheric water vapor concentrations, and potentially a shift from a dry (LGM) toward a moist (today) adiabatic lapse rate in the low troposphere (32, 40). However, studies focusing on lapse rate adjustments during the Pliocene are not as well-developed as the LGM, and the magnitude of elevation-dependent warmth during a warm climate state remains to be investigated. To quantify possible lapse rate adjustments during the Pliocene compared to the pre-industrial, we use the mean tropical (20°N to 20°S) air temperature output from HadCM3 model at five pressure gradients [1,000, 925, 850, 700, and 600 hPa, (41)]. The model projects that the Pliocene tropical lapse rate shallows only by 0.2 °C/km (SI Appendix, Fig. S8), indicating a small difference between Pliocene and pre-industrial lapse rate values. We assume that the pre-industrial lapse rate values of HadCM3 are comparable to our late Holocene scenario. Given that a small adjustment in lapse rate cannot fully explain the large amplification of temperature change seen in our record (SI Appendix, Fig. S8), other mechanisms are needed.

Amplified terrestrial temperature could also be related to dynamic mechanisms such as responses to Pacific SST gradients since the Pliocene. In particular, during the Pliocene the equatorial SST gradient was reduced, similar to a modern El Niño event, and which has been referred to as a “permanent” El Niño-like state, or El Padre (7, 42–48). In the northern tropical Andes, modern El Niño events cause droughts and anomalous warming (49–51). For example, the 1997/1998 El Niño event had similar patterns of SST to those reconstructed during the Pliocene, when the surface temperatures of the eastern tropical Pacific warmed by as much as ~4 °C (45, 52). Pérez-Angel and Molnar (51) showed that the largest positive temperature anomaly between 1961 and 2016 in the northern tropical Andes was during the 1997/1998 El Niño event, when the Eastern Cordillera of Colombia warmed by ~2 °C (51). We propose that a portion of the warm Pliocene temperature estimates in the Funza-II core, particularly between ~3.8 and 3.3 Ma, could be explained as a response of the northern tropical Andes to an SST distribution in the eastern tropical Pacific different from that observed today, and associated with an El Padre state during the Pliocene. Additional studies based on terrestrial proxy data in northwestern South America (53) support the evidence of shifts in atmospheric circulation that are consistent with a low zonal SST gradient in the tropical Pacific and an El Padre state during the Pliocene.

Our results suggest that the terrestrial amplification theory derived from past cold climate states (4) does not fully explain the magnitude of Pliocene warmth seen in the northern tropical Andes. We suggest that regional feedbacks during long-term warm climate states, which could amplify - as shown in this study -or mitigate the relatively rapid physics of terrestrial amplification, have not yet been observed on Earth today and will become a key feature of regional future climate change. Our study shows that large-scale temperature changes should incorporate dynamic processes such as responses to changes in Pacific SST gradients. Accurate simulations of the long-term adjustments in both zonal and meridional Pacific SST gradients will be critical for climate projections by the end of the 21st century, both for the large-scale mean and regional variability in terrestrial amplification. This will allow us to assess uncertainties of future warming scenarios at a regional scale, particularly in the tropics, where individuals, communities, and ecosystems will be most directly affected.

Materials and Methods

Sample Collection and Lipid Analysis.

We collected and analyzed 124 samples from the Funza-II core for biomarkers and 10 samples for U-Pb geochronology at the Servicio Geológico Colombiano in Bogotá and Bucaramanga. Since almost the entire ash layer sections were previously collected for fission-track analysis (54), we obtained samples and zircons for U-Pb dating from the uppermost and lowermost level of the ash layers reported and dated in previous studies (17, 54). Rock samples for biomarker analysis, with an overall thickness of 2.5 cm, were freeze dried, cleaned, and crushed before extraction. We extracted 2 to 6 g of rock sample using an ASE 200 DIONEX (20 samples) and an ASE 350 DIONEX (104 samples) at 100 °C and 2,000 psi with dichloromethane:methanol (DCM:MeOH 9:1 v:v). Sample processing and analysis for brGDGTs was identical to that reported in refs. 25 and 26. The analytical uncertainty of brGDGTs, determined as the area of each compound measured in triplicate, was less than 0.08%.

Zircon U-Pb Geochronology.

Samples were screened and analyzed at the Isotope Geology Laboratory at Boise State University. Zircons from samples that were large enough were mounted in epoxy and polished until their centers were exposed to obtain cathodoluminescence (CL) images (SI Appendix, Fig. S9). Five samples were analyzed by laser ablation inductively coupled plasma mass spectrometry (LA-ICPMS). U-Pb dates were obtained from single zircon grains from six samples using the chemical abrasion isotope dilution thermal ionization mass spectrometry (CA-TIMS) method (55). The specific methods for LA-ICPMS and CA-TIMS analysis can be found in SI Appendix. Zircons from two samples with CL images and LA-ICPMS data were removed from the epoxy mounts for CA-TIMS analysis, whereas zircons from the other four samples were too small for mounting, so they were selected based on the presence of sharp facets for CA-TIMS analysis (SI Appendix, Fig. S9). Only samples with CA-TIMS dates were used for interpretation of depositional ages in the Funza-II core [Fig. 1C and SI Appendix, Fig. S4 and Table S1; (26)]. We obtained our age-depth model using a version of the Bchron R package (56), modified for use in deep time (57), and the depositional ages of our samples. All U-Pb ages have a Gaussian distribution assigned to them, while the uppermost age of the Funza-II core, 0.03 Ma (17), had a uniform distribution in the model (Fig. 1C).

Statistical Analyses of brGDGT Sources.

Because the relative abundance of brGDGTs varies among different depositional environments, several environment-specific calibrations (e.g., Lake-specific, soil-specific, or peat-specific) have been developed to accurately estimate past temperatures. However, sedimentary archives that are composed of a mixture of lithologies can preserve a combination of different brGDGT sources that can yield misleading temperature estimates associated with variations in depositional environments rather than climate. Because the depositional environment of the Funza-II core fluctuated between swamp, lacustrine, fluvio-lacustrine, and fluvial deposits (17), we aim to disentangle the brGDGTs sources in each core sample for more accurate temperature reconstructions since the Pliocene. We carried out a binomial logistic regression (58) to estimate the probability that brGDGTs from samples of the Funza-II core are lake- or soil-derived (26) (see SI Appendix for details). We used brGDGT abundances from both tropical soils (25) and tropical lakes deeper than 8 m (23) as representative brGDGT end-member distributions and used Akaike Information Criterion (AIC) to select the best parsimonious model, a detail script for this model is available on GiTHub (https://github.com/LinaPerez/BrGDGT-BinomialSourceModel) (see SI Appendix for details). Then, using the best model we calculated the probability of lake-derived brGDGTs (P_lake), which we assigned to be 1 or 100%, soil-derived (1−P_lake) with a value of 0, and/or a mixing of both for each Funza-II sample.

Temperature Estimates.

We combine the probability estimates of the Funza-II core from the logistic regression with the temperature equations for Colombian soils (25) and tropical lakes (23) to obtain a combined temperature distribution for each sample (details of calibrations and estimates in SI Appendix, Materials and Methods and Dataset S1). For what we defined as “late Holocene” temperature estimates in the Sabana de Bogotá, we averaged the estimated temperature from the youngest sample (1.9 m) in the Funza-II core, which represents a late Holocene-aged deposit (59), and the 25 to 30 cm and the 45 to 50 cm soil samples (SI Appendix, Dataset S2) in the Bogotá basin reported by Pérez-Angel et al. (25). Because of the nature of the sampling, and the uncertainties in the age model and temperature reconstructions, our record lacks orbital (high) resolution and we rely on long-term trends and averages. To avoid relying on outliers and the skewed nature of the data in the brGDGT-based temperature reconstruction of the Funza-II core, and appropriately propagating the error of calibrations within each time interval, we used a bootstrapping method (60) with 1,000 resamplings when estimating the median temperature values, as a more robust and conservative measure of our Pliocene-Pleistocene temperature interpretations. Uncertainties reported for LST and SST as “$N_{-}^{+}°\mathrm{C}$ values represent the first (−) and third (+) quartiles of the bootstrap distribution.

Supplementary Material

Appendix 01 (PDF)

Dataset S01 (XLSX)

Dataset S02 (XLSX)

Data, Materials, and Software Availability

Study data are included in the article and/or supporting information.