2.1. Pre PlioMIP

Dowsett and Poore (1991) determined the spatial distribution of warming at ~3.0 Ma across the North Atlantic Ocean by generating microfaunal based seasonal SST estimates in closely spaced timeseries between 3.15 Ma and 2.85 Ma. In that work they determined the mean of short-term warm events delineated by samples with well-preserved fossil assemblages and high communalities (i.e. the fraction of raw faunal data in a sample accounted for by the 5-factor model of Dowsett and Poore (1990)). The seasonal means of warm events within the 3.15 Ma – 2.85 Ma time slab were then used as representative estimates of SST for each site. This ‘peak averaging’ technique preserved much of the high frequency information in the records, while avoiding problems associated with peak-to-peak correlations between cores.

Researchers using the NASA Goddard Institute for Space Studies (GISS) model and National Center for Atmospheric Research (NCAR) GENESIS model, required Pliocene global monthly SST fields to drive atmospheric models (Chandler et al., 1994; Sloan et al., 1996). To accomplish this, Pliocene winter and summer SST estimates at each site were differenced from modern February and August SST of Reynolds and Smith (1995) to develop warm and cold season anomalies. Anomalies were contoured and the resulting fields applied to modern SST fields to develop a Pliocene climatology. SST reconstructions began with the North Atlantic (Cronin and Dowsett, 1990; Cronin, 1991; Dowsett and Poore, 1991), grew to include the entire Northern Hemisphere, and eventually a series of global SST reconstructions (see Dowsett et al., 1996; Dowsett, 2007).

2.2. PlioMIP

The first PlioMIP included two experiments: one utilizing an atmospheric model (Haywood et al., 2010) and a second using a coupled ocean-atmosphere model (Haywood et al., 2011). PRISM3 SST fields (Dowsett et al., 2010) were used to force Experiment 1, but a verification data set was required for analysis of Experiment 2 output. The PRISM3 mean annual temperature verification data set had 100 localities ranging from ~80° S to ~80° N. The data were derived primarily from quantitative faunal or floral analyses, augmented where possible with Mg/Ca and alkenone SST estimates[REMOVED HYPERLINK FIELD] (Dowsett et al., 2012). PRISM3 faunal SST estimates represented a warm-peak-average, defined as the warm phase of climate from the interval between 3.264 Ma and 3.025 Ma at each locality. Variability about the estimated warm phase of SST was expressed as the standard deviation of warm peak estimates within the time slab, at each locality.

Different proxies measure different aspects of temperature by sampling the marine environment at various spatial and temporal scales. This is further complicated by effects unique to each signal carrier and method. Thus, simply averaging estimates from different proxies would have ignored the additional information recorded by the multiple proxies (Dowsett et al., 2013a). Instead, recognition of possible subtle changes in the timing of production of elements of the assemblages, along with taxa used for Mg/Ca paleothermometry living at different depths, allowed for an intricate though admittedly subjective reconstruction of surface and subsurface, providing seasonal and mean annual conditions during the Mid-Piacenzian at each location (Dowsett et al., 2013a; Lawrence and Woodard, 2017; Lam et al., 2021; Bova et al., 2021).

Data model comparison requires careful assessment of confidence in both proxy and model data. PRISM researchers assessed confidence of the PRISM3 SST verification data set following IPCC guidance (see Mastrandrea et al., 2011). A semiquantitative λ-confidence scheme was developed to provide a measure of confidence based upon chronology, sample density, sample quality, and type as well as performance of quantitative method used (Dowsett et al., 2012). Results of the initial PlioMIP DMC highlighted key regional (e.g. mid to high latitude North Atlantic) and dynamic (e.g. upwelling) situations of discord between the paleoenvironmental reconstruction and the climate model simulations (Dowsett et al., 2013b).

2.3. PlioMIP2

A primary finding of the initial PlioMIP DMC was realization that using paleoenvironmental estimates representing time-averaged conditions to verify simulated conditions representing a specific temporal horizon, was problematic (Dowsett et al., 2013b, Haywood et al., 2013). To reduce uncertainty, PlioMIP2 experimental protocols called for sub-orbital resolution verification data around a specific stratigraphic target of interest that exhibited orbital forcing close to that of present day (Haywood et al., 2016b). Marine Isotope Stage (MIS) KM5c, centered on 3.205 Ma, was adopted by PlioMIP2 and PRISM4 for DMC. The PRISM4 SST reconstruction was a compilation of published alkenone SST data that intersected the PRISM4 interval (3.210 – 3.200 Ma) (Foley and Dowsett, 2019). Only alkenone paleothermometry was utilized in the PRISM4 verification data to reduce uncertainty arising from incorporation of multiple temperature proxies which includes some degree of subjective interpretation (see section 2.2). The PRISM4 SST reconstruction consisted of 34 sites containing data falling within ±5 kyr of 3.205 Ma, and 37 sites, if a broader ± 15 kyr window was used. Localities ranged from 45° S to 79° N. The PRISM4 verification data were used for the PlioMIP2 DMC included in Haywood et al., (2020).

2.4. PlioMIP3

Over the past four decades the paleoceanographic community has developed a large amount of SST data for the Piacenzian (3.6 – 2.6 Ma). New data are continuously being generated and new techniques applied to use those data in innovative ways (e.g. Auderset et al., 2019; Tierney et al., 2019; de la Vega et al., 2020; Dvorak et al., 2025). PlioMIP3 includes for the first time an early Pliocene experiment as part of its experimental protocol (Haywood et al., 2024). Current PRISM5 research has pivoted to the early Pliocene, with specific emphasis on the western Atlantic region (Dowsett et al., 2023; Foley and Dowsett, 2024; Spivey, 2025; Foley and Dowsett, 2025; Dowsett and Spivey, 2025; Utsunomiya and Dowsett, 2025). Thus, the PRISM5 focus, and the focus of this paper, is on the early Pliocene, specifically the Zanclean, and the two early Zanclean time slices (PRISM5.1 and PRISM5.2) selected by PlioMIP3.

3.1. Chronology

Age models for data sets synthesized here are those of the original authors and have been used without alteration. Data discussed in this report fall into three discreet community-sourced reconstructions (Figure 1).

PRISM5x is a time slab covering part of the early Pliocene from 5.3 Ma to 4.2 Ma (Figure 1). As such, it falls within the Gilbert Polarity Chron, from just below the base of the Thvera normal subchron at 5.235 Ma to just below the top of the Cochiti normal subchron at 4.187 Ma. PRISM5x is located within planktic foraminifer Zones PL1 (Globorotalia tumida/ Globoturborotalita nepenthes Concurrent-range Zone) and PL2 (Globorotalia margaritae Highest occurrence zone) of King et al. (2020).

Situated within the PRISM5x interval are the two PlioMIP3 time slices (Figure 1), chosen due to their similarity to modern insolation distribution at the top of the atmosphere and coincidence with Marine Isotope Stages. The earlier time slice (PRISM5.2) at 4.870 Ma, coincides with Marine Isotope Stage (MIS) Si5 (Dolan et al., 2022; Dowsett et al., 2023), which extends from 4.883 Ma to 4.860 Ma (Lisiecki and Raymo, 2005). It is situated within the Gilbert Polarity Chron, within C3N.3n or the Sidufjall normal subchron (4.896–4.799 Ma) (Gradstein et al., 2020).

PRISM5.2 is located within foraminifer Zone PL1 (Globorotalia tumida/Globoturborotalita nepenthes Concurrent-range Zone) of King et al. (2020), between the last appearance of Globigerinoides seiglei (4.72 Ma) and the first appearance of Sphaeroidinella dehiscens s.l. (5.53 Ma). PRISM5.2 occurs within calcareous nannofossil Zone NN13 (Martini 1971) between the last appearances of Ceratolithus acutus (5.04 Ma) and Amaurolithus primus (4.50 Ma).

The younger time slice (PRISM5.1) coincides with MIS N1 which extends from 4.487–4.457 Ma (Lisiecki and Raymo, 2005). PRISM5.1 is situated within the Gilbert Polarity Chron, within C3N.1r, the base of which (top of the Nunivak normal subchron) occurs at 4.493 Ma (Gradstein et al., 2020). This interval occurs near the top of planktic foraminifera zone PL1 (Globorotalia tumida/Globoturborotalita nepenthes Concurrent- range Zone) of King et al. (2020), just above the last appearance of Sphaeroidinellopsis kochi (4.53 Ma) and the first appearance of Globorotalia exilis (4.45 Ma). PRISM5.1 occurs within calcareous nannofossil Zone NN13 (Martini, 1971) below the first common occurrence of Discoaster asymmetricus (4.04 Ma) and above the last appearance of Amaurolithus primus 4.50 Ma.

3.2. Sea Surface Temperature [SST]

SST reconstructions for the PRISM5x early Zanclean interval, which includes the PRISM5.1 and PRISM5.2 time slices sourced from previously published data sets, were compiled in Foley and Dowsett (2024, 2025). Sea surface temperature estimates were determined using the alkenone paleothermometer (Prahl and Wakeham, 1987). While other methods exist for Pliocene paleotemperature estimation (e.g., Mg/Ca, TEX86, faunal transfer functions) alkenone records are more plentiful in terms of spatial distribution. Several calibrations exist relating the ratio to SST (e.g. Prahl and Wakeham, 1987; Müller et al., 1998; Conte et al., 2006; Tierney and Tingley, 2018). Differences between these calibrations are minor and for consistency, the Müller et al. (1998) calibration is used throughout this report. Temperature estimates from studies covering sites Deep Sea Drilling Project (DSDP) 607, Ocean Drilling Program (ODP) 662, ODP 677, ODP 847, ODP 882, ODP 925, ODP 958, ODP 1014, ODP 1237 and International Ocean Discovery Program (IODP) 1425, were reported in the literature using other calibrations. In these cases, estimates were recalculated from unsaturation index values using the Müller et al. (1998) calibration. Note that ODP sites 642 and 907 are located outside the intended latitudinal range of the Müller calibration. The average temperatures calculated for ODP sites 925 and 1115 incorporate indices at or near the maximum for the technique (~28.97 °C). It is possible that samples from these sites represent higher temperatures than this method records.

Where multiple data sets covering the same core or site were found, only the data set with the greatest number of points within the Zanclean interval was selected. No data sets were blended based upon chronology or position of samples in core or outcrop. The mean of all data within each temporal range at each site, along with the number of data points, n, and when n is greater than 1, the standard deviation (σ) are provided in Appendix 1.

4.1. Early Zanclean mean data set (5.3 – 4.2 Ma)

The early Zanclean PRISM5x data synthesized in Foley and Dowsett (2024) contains 31 sites with 2055 individual estimates falling within the 5.3 Ma – 4.2 Ma interval (Figure 3). The spatial distribution of estimated SST anomalies for each site are shown in Figure 4. Most estimates (~80%) come from just 7 sites (see discussion; Figure 5). Unlike PRISM3 which used a warm peak averaging technique (see Dowsett, 2007), PRISM5x records a simple mean of all data falling between 5.3 Ma and 4.2 Ma at each site. Mean sample spacing within this time slab ranges from 2 kyr to 367 kyr.

4.2. Zanclean time slices (4.474 Ma and 4.870 Ma)

The spatial distribution of PRISM5.1 and PRISM5.2 SST anomalies are mapped in Figure 6. There are 17 sites with 42 estimates of SST available for the PRISM 5.1 time slice (4.474 Ma ±10 kyr) and 15 sites with a total of 47 data points for the PRISM5.2 time slice (4.870 Ma ±10 kyr). Sample spacing within the PRISM5.1 window is ≤ 20 kyr at 11 of 17 sites; for the older PRISM5.2 window sample spacing is ≤ 20 kyr at 9 of 15 sites. These sites have at least one estimate falling within the respective ± 10 kyr windows.

5.1. Time series resolution and other sources of uncertainty

A first order problem of deep time DMC is having sample spacing sufficient to prevent information loss or distortion in the SST signal. Sample spacing of one sample every ~10 kyr is sufficient to capture glacial-interglacial variability within Zanclean Age sequences dominated by the 41 kyr obliquity cycle. Since sediment accumulation rates are variable, and post depositional processes like bioturbation are not constant, in a practical sense, sample spacing should be ~1 sample every 5 kyr. It is rare that sample spacing of 5 to 10 kyr is maintained back through 5.3 million years of record. It is more common to find time series with widely spaced samples between more densely sampled intervals of paleoclimatic interest. At the time of this writing there are 7 time series openly available with sample spacing that allows for orbital scale resolution during the early Zanclean (Figure 5). The amplitudes of Zanclean records are greatly diminished in comparison to Piacenzian or Late Pleistocene records, making identification of specific orbitals challenging (Dowsett et al., 2023).

Even with sufficient sample spacing, sub-orbital scale chronologic correlations may be inaccurate due to spatially heterogeneous post depositional processes. Accumulation rates may differ, thereby integrating different amounts of time at different localities. For example, in the northeast Atlantic Ocean, Carvalho et al. (2011) found accumulation rates of 0.14 and 3.2 cm kyear⁻¹ from the Porcupine and Iberian Abyssal Plains, respectively. Hence, a typical 1 cm ODP sample could represent between ~300 and ~7000 years. With sediment mixing due to bioturbation reaching depths of ~10cm, a sample could represent as much as ~70,000 years. This makes it unlikely that comparison of samples from “within a time slice” provide a verifiable temporal comparison.

A common approach to insufficient sample spacing and variable accumulation rates is to smooth data with a low-pass filter. However, this obfuscates the PlioMIP3 choice of time slices with orbital parameters similar to present day and makes comparison of smoothed proxy temperature data to model output (representing an instant in time), somewhat dubious. Comparison of smoothed SST data to modeled temperatures adds another level of uncertainty for DMC.

There are other well-known complications to be addressed when comparing model and proxy data for DMC. Organisms responsible for producing alkenones may drift in terms of timing of production, on multiple time scales (e.g., diel to seasonal). Such changes are not regular and undoubtedly not synchronous on a global scale. Thus, while a calibration may be set to mean annual temperature, changing the timing of production may skew estimates.

Figure 7 shows the most resolved time series intersecting the upper (PRISM5.1) and lower (PRISM5.2) early Zanclean time slices. All records in Figure 7 were orbitally tuned. While the SST history at different sites should not necessarily show synchronous changes, especially when located in different hemispheres, in some instances sites from the same region show different phasing with respect to the LR04 oxygen isotope stack (Figure 7). Minor changes to age models would create more synchronous temperature responses, and it is tempting to assume the age models are at fault. The PRISM5.1 interval aligns with cooling at Site 907 but warming at Site 609. During the earlier PRISM5.2 time slice, Sites 642 and 907 show different temperature responses, possibly due to paleoceanographic differences (e.g. location of currents), despite being located in close proximity to each other. Thus, if a narrow time window is necessary for DMC, operators must implicitly rely on the soundness of age models.

Comparison of the individual time series shown in Figure 7 with the mapped anomalies shown in Figure 6 illustrates the problem of including different numbers of samples as well as the potential for minor variations in age models leading to different results.

The intention here is not to argue against DMC, but to remind users to consider and acknowledge a wide range of potential complicating factors when making comparisons.

5.2. North Atlantic latitudinal temperature gradients

In addition to focusing on individual time slices, long term trends in the North Atlantic data, when carefully interpreted, can provide useful information for DMC. Based upon proxy temperature data, open-ocean mPWP North Atlantic SSTs were warmer than pre-industrial conditions and warming generally increased toward higher latitudes (e.g. Dowsett et al., 1992; 2012; Herbert et al., 2016). The late Zanclean through early Gelasian (4 to 2.4 Ma) North Atlantic surface temperature gradient was variable, reduced relative to the modern gradient, and at times nearly absent (Naafs et al., 2020). Herbert et al. (2016) documented evolution of temperature in the North Atlantic to be decreasing over the entire Zanclean using alkenone data from Sites 907 and 982. Figure 8 shows those and additional data from Sites 609 (Auderset et al., 2019) and 642 (Bachem et al., 2017), plotted over the 5.3 Ma – 4.2 Ma (early Zanclean) interval. All four sites show the Zanclean to be warmer than the Piacenzian, which in turn was warmer than the pre-industrial, again documenting long term cooling over the entire Pliocene.

In low latitudes, SST differences across the Central American Isthmus (CAI), between Site 1241 (Pacific) and 999 (Caribbean), whether during the Zanclean, Piacenzian, or pre-industrial, are negligeable (Table 1). This contrasts with salinity changes on either side of the CAI that have been used to support a progressive closure of the CAS, punctuated by breaches of the CAI as late as 2.4 Ma (Cronin and Dowsett, 1996; Groeneveld et al., 2014; O’Dea et al., 2016; Auderset et al., 2019).

Closure of the CAS played an important role in altering oceanic circulation. An open CAS allowed exchanges of surface and bottom water while shoaling of the CAS promoted increased salinity in the Caribbean and concomitant warming in the Atlantic, especially along the southeastern coastline of North America (Berggren and Hollister, 1974; Keigwin, 1982; Coates et al., 1992; Cronin and Dowsett, 1996; Haug and Tiedemann, 1998; Haug et al., 2001; Steph et al., 2006; Sarnthein et al., 2009; De Schepper et al., 2013; Karas et al., 2017; Dowsett et al., 2019; Auderset et al., 2019; Dowsett et al., 2021).

5.3. Atlantic Coastal Plain

On the Atlantic Coastal Plain (ACP) of the United States, Zanclean aged units exist as the subsurface Wabasso beds in Florida and Georgia, and the Sunken Meadow Member of the Yorktown Formation in Virginia and North Carolina (Ward and Blackwelder, 1980; Huddlestun, 1988; Dowsett and Spivey, 2025) (Figure 9.). Paleontological data from shallow marine sequences of Central America and the Atlantic Coastal Plain of North America (e.g., Hazel, 1971; Cronin and Dowsett, 1990; Cronin, 1991; Dowsett et al., 1992; Cronin and Dowsett, 1996; Haug and Tiedemann, 1998; Johnson et al., 2017; Dowsett et al., 2019) indicate early Zanclean conditions were cooler and less saline (than during the pre-industrial). Benthic faunas do not show evidence of warm Gulf Stream waters moving north (Johnson et al., 2017), and planktic assemblages suggest periodic low salinity in the Caribbean during the Early Pliocene due to breaching of the shoaling CAS (Cronin and Dowsett, 1996; Dowsett et al., 2019). Farmer et al. (2025) suggest 4.1 Ma as the climatically significant closure of the CAS based upon foraminifera‐bound nitrogen isotopes (FB‐δ¹⁵N). Shoaling of the CAS to less than 100 m may have led to increased warmth (and increased salinity) in the Gulf Stream - North Atlantic Drift, which in turn could have increased thermohaline circulation (Karas et al., 2017; Auderset et al., 2019).

In southeastern Virginia, early Pliocene sediments of the Yorktown Formation show a 2-3°C cooling trend over part of the Zanclean Stage (Figure 8). This is followed by rapid warming at the base of the overlying mid Piacenzian unit, which has been correlated to the transition between MIS M2 and MIS M1, from Virginia to Georgia (Dowsett and Spivey, 2025). While this is evidence of cooling, available biochronology only places the unit shown in Figure 8B within a ~0.6 My interval which, given that pre Middle Pleistocene sea level variability corresponds to 41,000 year obliquity forcing, covers ~15 potential transgressive-regressive cycles. Rather than representing a long-term trend as seen in the open ocean sites further north, the sequence could simply represent cooling over part of a single depositional cycle. Molluscan faunas from this Zanclean unit are warm temperate, and the overlying unit, correlated to the mid Piacenzian, contains warmer elements. The influx of warmer molluscan taxa during the mPWP may be due to physical paleogeographic changes that allowed warmer water from the south to penetrate further north, or migration of the North Atlantic Gyre reducing cool southward flowing coastal currents from the north (Ward and Blackwelder, 1980; Johnson et al., 2017; Dowsett et al., 2021).

Mollusks and ostracods monitor bottom water temperatures and the alkenone paleothermometer is measuring mean annual SST at the surface, or possibly seasonal temperature during late spring or early summer algal blooms. Today, in analogous settings off the east coast of the US, surface and bottom water temperature differences can be as much as 10°C (Seidov et al., 2016).

Thus, while incorporation of coastal plain sequences in DMC schemes can provide additional insight, and a more nuanced understanding of the paleoenvironmental setting during the Zanclean, integration of those data into equator to pole temperature gradients needs to be done with caution, especially until age and temporal duration of land sections can be better constrained.

The North Atlantic latitudinal surface gradient is a robust feature of the Pliocene climate, present during both the Zanclean and Piacenzian. In terms of DMC, these proxy-based gradients for the Zanclean, Piacenzian and pre-industrial, compared with PlioMIP3 results, can provide a simple first-order check on performance of models relative to each other and the proxy data. Inclusion of potentially higher resolution coastal plain sequences promises a more in-depth understanding of the paleoenvironment, but at this time opportunities are limited by lack of sufficient chronological data.