The PRISM4 (mid-Piacenzian) paleoenvironmental reconstruction

. The mid-Piacenzian is known as a period of relative warmth when compared to the present day. A comprehensive understanding of conditions during the Piacenzian serves as both a conceptual model and a source for boundary conditions as well as means of veriﬁcation of global climate model experiments. In this paper we present the PRISM4 reconstruction, a paleoenvironmental reconstruction of the mid-Piacenzian ( ∼ 3 Ma) containing data for paleogeography, land and sea ice, sea-surface temperature, vegetation, soils, and lakes. Our retrodicted paleogeography takes into account glacial isostatic adjustments and changes in dynamic topography. Soils and lakes, both signiﬁcant as land surface features, are introduced to the PRISM reconstruction for the ﬁrst time. Sea-surface temperature and vegetation reconstructions are unchanged but now have conﬁdence assessments. The PRISM4 reconstruction is being used as boundary condition data for the Pliocene Model Intercomparison Project Phase 2 (PlioMIP2) experiments.


Introduction
The Pliocene, specifically the mid-Piacenzian (3.264 to 3.025 Ma), has been a focus of synoptic paleoclimate research for the past 25 years.The mid-Piacenzian is the most recent time in Earth's past to have exhibited climates not unlike those projected for the end of the 21st century (Dowsett, 2007a;IPCC, 2013).With widespread recognition by most experts that anthropogenic drivers have been the dominant cause of observed warming since the mid-20th century (Verheggen et al., 2014), and surface temperatures projected to rise during the 21st century under all emission scenarios (IPCC, 2013), understanding the Pliocene climate has taken on new importance.While not a temporal analog to future climate conditions, there is much to learn about the magnitude and spatial distribution of processes from this, in essence, natural climate laboratory (Crowley, 1990).
Borrowing heavily from methodology used by the CLIMAP (Climate/Long-Range Investigation, Mapping and Predictions) for the reconstruction of the last glacial maximum (LGM), the US Geological Survey (USGS) and collaborators initiated a large-scale data collection and interpretation project: Pliocene Research, Interpretation and Synop-tic Mapping (PRISM) (see Dowsett, 2007a;CLIMAP, 1976).The first PRISM reconstructions consisted of data sets for sea-surface temperature (SST), vegetation, land ice distribution and volume, sea-ice cover, land elevation, and sea level (Dowsett et al., 1994(Dowsett et al., , 1999)).Since its inception in the late 1980s, PRISM has grown in both size and scope, evolving through four global-scale reconstructions, each one adding a new component of the Piacenzian paleoenvironment or improving upon methods to increase confidence in the reconstructions (see Dowsett, 2007;Dowsett et al., 1994Dowsett et al., , 1996Dowsett et al., , 1999Dowsett et al., , 2010Dowsett et al., , 2013)).These reconstructions serve two purposes: to assemble the best information possible to provide a conceptual model of the Piacenzian paleoenvironment and to provide the data as quantitative, gridded arrays to the paleoclimate modeling community for global climate model simulations.
PRISM3 was the basis for the Pliocene Model Intercomparison Project (PlioMIP), the first Pliocene multi-model comparison using fully coupled atmosphere-ocean general circulation models (AOGCMs) (Haywood et al., 2010(Haywood et al., , 2011)).Much effort was put into improving the land surface cover and ocean components for this penultimate reconstruction (Dowsett et al., 2010).PRISM3 introduced a multiproxy SST reconstruction, a deep-ocean temperature reconstruction, and for the first time a biome classification of the Pliocene land surface.
The PRISM3 reconstruction also addressed uncertainty in both terrestrial and marine paleoclimate estimates (Dowsett et al., 2012(Dowsett et al., , 2013a;;Salzmann et al., 2013).Marine and terrestrial data-model comparisons (DMCs) presented in IPCC AR5 documented robust large-scale features of the Piacenzian climate but at the same time identified areas of disagreement between data and models, highlighting the need for additional assessments of confidence for both reconstructed and simulated environments (Haywood et al., 2013;Salzmann et al., 2013;Dowsett et al., 2012Dowsett et al., , 2013a)).
PRISM data have also been used to study diversity and ecological niche changes in planktic foraminifers, mollusks, and fish in the face of profound global warming (e.g., Yasuhara et al., 2012;Saupe et al., 2014;Jacobs, 2015).
In this paper we document and summarize our most recent reconstruction, PRISM4.PRISM4 is a conceptual model of mid-Piacenzian conditions for which major efforts have been focused on improving the paleogeographic and cryospheric components.New topography/bathymetry, Greenland ice sheet (GIS), soil, and lake distributions are discussed, including those PRISM4 features being used as boundary condi-tions for climate model simulations.Confidence in paleoenvironmental reconstruction is receiving much interest in the paleoclimate modeling community, and changes necessary to address uncertainty are discussed.We also discuss both terrestrial and marine high-resolution data being developed as part of PRISM4 to better understand Piacenzian paleoclimate variability.

PRISM chronology
Previous PRISM paleoclimate reconstructions were based upon a time slab concept developed early in the project to overcome the inability to make long distance high-resolution correlations (Dowsett and Poore, 1991).This was achieved for SST data and land surface cover reconstruction by averaging the warm phase of climate or selecting the vegetation representing the wettest and warmest period within the defined temporal slab at each locality (Dowsett et al., 2009(Dowsett et al., , 2013;;Salzmann et al., 2008Salzmann et al., , 2013)).The PRISM reconstruction therefore approximates the average "interglacial" condition at each site.The initial slab was a ∼ 300 ky interval ranging from within the Mammoth to just above the Kaena reversed-polarity subchrons of the Gauss normal-polarity chron (Fig. 1).This and a number of biochronologic events allowed approximate identification of the slab position in both marine and terrestrial settings (see Dowsett, 2007a, and references cited therein).
The PRISM3 time slab (Dowsett et al., 2010) was refined to the interval between 3.264 and 3.025 Ma, easily identified in marine sequences by its position between oxygen isotope enrichments M2 and G20 (Fig. 1) in the LR04 chronology of Lisiecki and Raymo (2005).The 3.264 to 3.025 Ma range is still appropriate for most components of the PRISM4 reconstruction (see discussion).
New PRISM4 time-series data use a nested chronology since various components of the paleoenvironmental reconstruction can achieve different maximum resolution.PRISM4 marine time series are generated from localities possessing the characteristics needed for precise orbitalscale correlation.The challenges of terrestrial age dating and stratigraphic control of many terrestrial archives currently limit expansion of the terrestrial reconstruction of PRISM4 time series in all but a few locations.
The PRISM time slab or PRISM "interval", as defined above, occurs within the Piacenzian Age.The Piacenzian is roughly equivalent to the Gauss normal-polarity chron (∼ 3.6 to 2.6 Ma).Prior to 2010, the Pliocene Epoch included the Zanclean, Piacenzian, and Gelasian ages.Thus, it was common practice to refer to the PRISM interval as the mid-Pliocene.Changes enacted by the International Commission on Stratigraphy revised the placement of the Pliocene-Pleistocene boundary from the base of the Calabrian Stage (1.801 Ma) to the base of the Gelasian Stage (2.588 Ma) (Gibbard et al., 2010).This change makes it awkward to refer  et al. (1996).
to the mid-Piacenzian PRISM interval as mid-Pliocene (see also Dowsett and Caballero-Gill, 2010).Previous publications referring to the PRISM interval, PRISM time slab, mid-Pliocene warm period (mPWP), or mid-Piacenzian all refer to the same interval of time originally defined by Dowsett and Poore (1991) and revised by Dowsett et al. (2010) as discussed above (Fig. 1).We propose the term mid-Piacenzian be used hereafter.

PRISM4 Piacenzian paleoclimate reconstruction
PRISM4 contains internally consistent and integrated data sets representing our best synoptic understanding of paleogeography, sea level, ocean temperature, terrestrial vegetation, soils, lakes, and land and sea-ice distribution (Table 1).Some PRISM4 data sets received only minor alterations from PRISM3, while others represent fundamental changes in methodology or add additional facets of the paleoenvironment not explicitly included in earlier versions of our reconstruction.
The single largest change accompanying the PRISM4 reconstruction is the addition of our paleogeography (topo-graphic and bathymetric data) (see Rowley et al., 2013).These data are integrated with Greenland and Antarctic ice sheets (see Sect. 3.1).Our land cover analysis has been enhanced with new data sets depicting the spatial distribution of lakes and soil types (Pound et al., 2014).SST and terrestrial vegetation cover are essentially unchanged from PRISM3, but confidence schemes have been created to assess individual sites used for the reconstruction (Salzmann et al., 2013;Dowsett et al., 2013).In addition, for the first time, new PRISM4 high-resolution time series are being developed to specifically address variability within the PRISM time interval.
The interrelationships and dependencies between the PRISM4 and PRISM3 reconstructions are illustrated in Fig. 2. Major components of the reconstruction are briefly discussed in the following sections, and data are available here: http://geology.er.usgs.gov/egpsc/prism/4_data.html.Readers are referred to the original citation of each data set for more detailed background and descriptions of methodologies (Table 1).

Ice sheets
Direct evidence for Piacenzian ice sheets (Greenland and Antarctica) is sparse.Therefore, the size, placement, and volume of ice in polar regions must be remotely assessed by a number of other techniques (e.g., proximal marine records of ice-rafted detritus, marine oxygen isotope records combined with bottom water temperature estimates and indicators of sea ice).These techniques constrain global and possibly regional ice volume but not its three-dimensional expression.Ice sheet models can be used to generate potentially realistic configurations based upon all available evidence.
The PRISM4 GIS is based upon results from the Pliocene Ice Sheet Modeling Project (PLISMIP) presented in Koenig et al. (2015) and Dolan et al. (2015a).This GIS configuration (Fig. 3) represents our highest confidence for the possibility of ice sheet location during the warmest parts of the mid-Piacenzian and is constrained by available proxy data.The PRISM3 GIS was centrally located and covered approximately 50 % of the area of the present-day ice sheet (Dowsett et al., 2010).In contrast, the PRISM4 GIS is confined to high elevations in the Eastern Greenland Mountains, covering an area approximately 25 % of the present-day GIS.
The dynamic nature of the Antarctic cryosphere during the Pliocene has been debated in the literature for decades and, as with the GIS, direct evidence for ice conditions is limited to the Dry Valleys and proximal marine records (e.g., Shackleton and Kennet, 1975;Webb et al., 1984;Dowsett and Cronin, 1990;Prentice et al., 1993;Shackleton et al., 1995;Kennett and Hodell, 1995;Naish et al., 2009;McKay et al., 2012) PRISM3 and PRISM4,respectively). Starting points in the generation of these data sets are indicated by teal rounded rectangles.These are the basic geochemical, faunal, floral, soil, cryospheric, topographic, bathymetric, sedimentologic, and stratigraphic data.Marine temperature estimates (SST and DOT (deep-ocean temperature)) are based upon multiple proxies (including faunal, floral, geochemical, and biomarker analyses).Terrestrial vegetation data are translated into a BIOME4 scheme and compared to an independent BIOME4 landscape produced by the HadAM3/TRIFFID model (red circle) initialized with PRISM2 boundary conditions.Integration of data and model output results in the hybrid PRISM3 BIOME reconstruction (Salzmann et al., 2008).Antarctic and Greenland ice sheets are based upon PRISM3 and PLISMIP results, respectively.Paleogeography is based upon an initial ETOPO1 digital elevation model incorporating PRISM4 ice sheets, GIA, and adjustments due to mantle convection (red circle).Soils are determined through comparison of sedimentological and stratigraphic data with the PRISM3 BIOME reconstruction.Lakes are determined from stratigraphic and sedimentological data (Pound et al., 2014).Sea-level is estimated from the PRISM4 ice sheets and is consistent with the paleogeography presented here.Dowsett et al. (2009Dowsett et al. ( , 2013) ) [PRISM3_SST_v1.0.nc]Deep-ocean temperature Dowsett et al. (2009) [Global_dot_v2.0.nc] cate both may have experienced a dynamic Pliocene ice history (Young et al., 2011;Austermann et al., 2015).The PRISM3 Antarctic ice reconstruction was generated using the British Antarctic Survey Ice Sheet Model (BA-SISM), driven by an AGCM experiment using prescribed PRISM2 boundary conditions (Hill et al., 2007;Hill, 2009;Dowsett et al., 2010).In the PRISM2 reconstruction, the Wilkes and Aurora subglacial basins had reduced areal extent of ice sheets relative to the present day.PRISM3, however, removed the West Antarctic Ice Sheet based upon available information that suggested there was no ice present during the warmest parts of the Pliocene (Naish et al., 2009;Pollard and DeConto, 2009;Hill et al., 2010).Since the East Antarctic Ice Sheet configuration used in PRISM3 is still supported by recent work, we have retained the PRISM3 Antarctic ice configuration for the PRISM4 reconstruction (Fig. 3).

Paleogeography
Modeling the dynamic nature of the Earth's surface over time is an area of increasing interest and complexity.It has become apparent that a physically based methodology for deriving Piacenzian global topography is necessary and that it needs to account for first-order effects of mantle convection (dynamic topography) and glacial isostatic adjustment (GIA) (Raymo et al., 2011;Rowley et al., 2013).
PRISM3/GISS topography was based upon the paleogeographic maps of Markwick (2007) (Sohl et al., 2009).For PRISM4 we have developed a new procedure to retrodict the paleogeography of the Piacenzian taking into account major processes that affect surface topography and sea level as a function of time.The main processes considered are (i) changes in dynamic topography associated with mantle flow and (ii) GIA due to ice loading.We also calculate a global land-sea distribution (sea level) prediction based on the reconstructed Piacenzian topography.Essentially, the retrodicted PRISM4 Piacenzian paleogeography is the result of a number of steps detailed below (and summarized in Fig. 4).The paleogeography is constrained to agree with global integral properties of the Earth, such as the conservation of the Earth's mean radius and the preservation of the mass of ice plus water at the surface of the Earth.

Initial base topography
Our starting conditions are based on the ETOPO1 (1 arcmin) global relief model of the Earth's surface that integrates land topography and ocean bathymetry (Amante and Eakins, 2009).Figure 4a shows the "bedrock" (base of the ice sheets) variation of ETOPO1 sampled at the working resolution of a quarter degree.As no GIA correction has been applied to this version of ETOPO1, initially we apply a GIA correction to account for both the remaining disequilibrium due to the ongoing relaxation of topography from Last Glacial Maximum (LGM) ice loads (Raymo et al., 2011;Fig. 4b) and all present-day ice and GIA associated with melting of this ice.This results in a nominal equilibrium topography that would pertain in several hundred thousand years, assuming all ice and ice-related loading has been accounted for.There are, however, features that are prominent "fingerprints" of Pleistocene glaciations, such as the Great Lakes and the Canadian Archipelago, that were not features of the landscape during the Piacenzian (Fig. 4c).Therefore, the rebounded topography reconstruction shown in Fig. 4b was contoured at 50 m intervals and then the Pleistocene landscape features were filled in to achieve an approximately uniform average elevation in each region (Fig. 4d).In reality the sediments eroded from areas such as the Great Lakes will have been redeposited elsewhere, for example in the Gulf of Mexico (Galloway, 2008).A more consistent method of applying a Pleistocene fill would be to take sediment from known depositional environments and redistribute it in the erosional settings.However, we have chosen to implement a more simplistic representation as there are currently no detailed isopachs of preserved sediment thickness or reconstructions of denudation to insure that our model agrees with observations.

Dynamic topography
Dynamic topography is defined as the response of the Earth's surface to the radial component of stresses originating from buoyancy residing in the mantle and the lithosphere (Hager et al., 1984;Forte et al., 1993).It is the component of topography, supported by buoyancy forces in the mantle, which is not explained by variations in crustal thickness and/or density and can be considered as the difference between present-day topography and the crustal isostatic topography (Forte et al., 1993).
In order to compute an estimate of dynamic topography change since the Piacenzian, we use a calculated global mantle flow field (Moucha et al., 2009;Rowley et al., 2013).This calculation is based on a simultaneous inversion of seismic travel-time data with various geophysical observables (Forte et al., 2010;Simmons et al., 2006Simmons et al., , 2009)).The latter include global free-air gravity anomalies, crust-corrected estimates of dynamic surface topography, plate divergence, and the dy-namic ellipticity of the core-mantle boundary (Forte et al., 2010;Simmons et al., 2006Simmons et al., , 2009)).The mantle flow calculations (e.g.Forte, 2007Forte, , 2011;;Forte et al., 1993Forte et al., , 2010;;Moucha et al., 2008aMoucha et al., , b, 2009) ) are formulated for a fully compressible, self-gravitating mantle with a mean density profile given by PREM (Dziewonski and Anderson, 1981) and geodynamic inferences of the depth variation of the mantle viscosity based on Mitrovica and Forte (2004).
For the dynamic topography correction shown in Fig. 4e, we use the mantle viscosity model V2 (Moucha et al., 2008) and the seismic tomography model TX2008 (Simmons et al., 2009) that represent a sensible realization of the dynamic topography change contributions to Piacenzian topography.Our choice of seismic tomography-derived density distribution and the mantle viscosity model will affect both the amplitude and location of our dynamic topography estimate and its rate of change (Rowley et al., 2013), but not significantly on the scale of the quarter-degree resolution of the reconstruction.Application of the dynamic topography correction results in the paleogeography shown in Fig. 4f.

Glacial isostatic adjustment
Glacial Isostatic adjustment (GIA) refers to the deformational, gravitational, and rotational adjustment of the Earth as a consequence of changes in ice sheet loading (Farrell and Clark, 1976;Mitrovica and Milne, 2002).The present topography of the Earth is still significantly impacted by the loading and unloading of ice during the Late Pleistocene glacial cycles.The current grounded ice sheets of Greenland and Antarctica also affect the global surface topography.Finally, the potential ice sheet loading based on our defined mid-Piacenzian ice sheets (Fig. 4g; see above) will also have an impact on the reconstructed topography.Following methods employed in Raymo et al. (2011) and Rowley et al. (2013) and based on the seminal paper of Farrell and Clark (1976), we incorporate a full GIA correction into our PRISM4 paleogeography (Fig. 4h), thus taking into account the largest factors that influence topography over time.
Figure 5 shows the final land-sea distribution of the PRISM4 paleogeography.The major differences from the PRISM3 paleogeography of Sohl et al. (2009), PRISM2 topography of Dowsett et al. (1999), and the present-day geography are a closed Bering Strait and Canadian Arctic Archipelago, effectively shutting off the Atlantic connection to the Arctic through the Labrador Sea and Baffin Bay.The PRISM4 mid-Piacenzian Indonesian region is elevated compared to present.Shoaling in this region would have created a more restricted seaway between the Pacific and Indian oceans with potential effects on circulation and heat exchange with the atmosphere (Godfrey, 1996).See the Discussion section for a complete analysis of the paleogeographic differences between PRISM4 and previous PRISM topographies.

Sea level
Piacenzian sea-level estimates in the literature range from no change to almost +50 m above present day based upon a variety of techniques (e.g., Dowsett and Cronin, 1990;Wardlaw and Quinn, 1991;Brigham Grette and Carter, 1992;Kennett and Hodell, 1995;Miller et al., 2005Miller et al., , 2012;;Sosdian and Rosenthal, 2009;Naish and Wilson, 2009;Dwyer and Chandler, 2009;Winnick and Caves, 2015).Early PRISM sea-level estimates were primarily glacioeustatic and dealt with tectonic subsidence and uplift in a relatively simplistic fashion (Dowsett and Cronin, 1990).Sea-level changes must also take into account an evolving topography, the result of GIA combined with changes due to mantle convection (Rowley et al., 2013).PRISM3 qualitatively compared ranges of sea-level estimates from a number of sources and found a sea-level rise of +25 m relative to present day fit all available data (Dowsett et al., 2010).
The reconstructed PRISM4 ice sheets have a total volume of 20.1 × 10 6 km 3 .Relative to the modern observed volume of the Greenland (Bamber et al., 2013) and Antarctic ice sheets (Fretwell et al., 2013), we use the method described in de Boer et al. (2015) to estimate a sea-level-equivalent change for the mid-Piacenzian.Our ice sheet configurations equate to a sea-level increase of less than ∼ 24 m, without taking into account changes to the size of the global ocean in the Pliocene (as shown by changes in the land-sea mask shown in Figs. 3 and 5).The PRISM4 sea-level estimate differs slightly from the PRISM3 sea-level estimate (based upon the PRISM3 ice sheets) due to the reduction in the size of the Greenland ice sheet relative to PRISM3 (Dowsett et al., 2010).

Sea-surface temperature and sea ice
Global fields of SST and sea ice (Fig. S1 in the Supplement) were described in Dowsett et al. (2009Dowsett et al. ( , 2010) )  ing a warm peak averaging (WPA) technique and time slab concept.While useful for a general synoptic view of mid-Piacenzian interglacial ocean conditions, the increasing use of the PRISM reconstruction for data-model comparison requires a more refined temporal window.A reduced need to drive atmospheric GCMs with monthly temperature fields has led to only small enhancements to this portion of the PRISM data set (Fig. 2).Instead, emphasis has been placed on new locality-based, confidence-assessed, high-resolution time series (Dowsett et al., 2012(Dowsett et al., , 2013a;;Haywood et al., 2016a; see Sect. 4).For completeness, monthly sea-ice distributions are included in the Fig. S1.

Terrestrial biomes
The PRISM4 biome reconstruction (Figs. 3 and 6a) has been updated from Salzmann et al. (2008Salzmann et al. ( , 2013) ) and Dowsett et al. (2010).Biomes are based upon data from 208 sites compiled from the literature (Salzmann et al., 2008).Surface temperature and precipitation anomalies have been derived from literature using the coexistence approach (Mosbrugger and Utescher, 1997) and multi-proxy temperature reconstructions.The relative confidence one can place on estimates from a particular locality was determined by Salzmann et al. (2013).In order to address climate variability, our biome map provides, for selected paleobotanical localities with sufficient data resolution, a biome reconstruction for a colder/drier and warmer/wetter interval within the Piacenzian stage (Fig. 6a).Future PRISM terrestrial analysis, like the marine work, will shift focus to localities that allow for high-resolution chronology required to address magnitude and variability of climate within a much (temporally) reduced time slab.

Soils and lakes
PRISM4 soil and lake distributions have been generated from a compilation of published literature (Pound et al., 2014).Nine soil types were identified from 54 paleosol occurrences and recorded.Pound et al. ( 2014) related soils at these 54 localities (Fig. 6b) to the PRISM3 biome reconstruction of Salzmann et al. (2008) and derived soil types for all land areas (Fig. 3).Based upon color and texture, each soil can be assigned an albedo value (Pound et al., 2014;Haywood et al., 2016b; see Table S1 in the Supplement).Lake data have been compiled from sedimentological evidence, dynamic elevation models, topographic studies, fauna, flora, or a combination of these (e.g., Adam et al., 1990;Muller et al., 2001;Drake et al., 2008;Otero et al., 2009).Piacenzian lake surface area was taken from published estimates or from reconstructed lake extents.These surface areas were then translated into a percentage of a PRISM4 grid cell (2 • latitude by 2 • longitude), and where megalakes occupied more than one grid cell, the geographic distribution (See Fig. 3) was based upon the published distal-most latitude-longitude points and reconstructed shape (for full methodology see Pound et al., 2014).

Discussion
4.1 Areas of paleogeographic change

Marine
The Isthmus of Panama is above sea level in the PRISM4 reconstruction, effectively eliminating ocean circulation between the Atlantic and Pacific via the Central American Seaway (CAS).While there is evidence for occasional brief breaches of the Isthmus subsequent to the mid-Piacenzian based upon Caribbean salinity records (e.g., Coates et al., 1992;Cronin and Dowsett, 1996), the overall restriction of Caribbean and Pacific waters by the Isthmus was fully established by ∼ 4.2 Ma (Haug et al., 2001;Montes et al., 2015).This CAS closure set up the present-day salinity contrast between the Atlantic and Pacific, shunting warm salty water northward along the east coast of North America as the Gulf Stream Current.The Gulf Stream diverges from North America and becomes the North Atlantic Current, directing warm salty water toward the northeastern North Atlantic.De Schepper et al. ( 2009) related Marine Isotope Stage (MIS) M2 to a brief reopening of the CAS that weakened the North Atlantic Current and reduced northward ocean heat transport.Larger (compared to present day) ice sheets during MIS M2 (∼ 3.3 Ma) are supported by a coupled oceanatmosphere climate simulation targeted at this time interval (Dolan et al., 2015b); however, this study did not test the hypothesis of an open CAS as a mechanism for ice sheet initiation.New PRISM4 surface air and SST time series covering the MIS M2 through MIS KM5 interval should provide further information on marine and terrestrial conditions during and after MIS M2.
The Bering Strait initially opened in the late Miocene, possibly earlier, based upon the presence of the Atlantic bivalve Astarte occurring in sediments of the Alaskan Peninsula correlated to subzone b of the Neodenticula kamtchatica Diatom Zone (5.5-4.8Ma) (Marincovich andGladenkov, 1999, 2001).Faunal exchange of thermophilic ostracodes between the Atlantic and Pacific via the Arctic also supports an open Bering Strait during at least parts of the Pliocene and early Pleistocene (Cronin et al., 1993;Matthiessen et al., 2009).This was the rationale for an open seaway in the PRISM3 reconstruction.Our independent PRISM4 paleogeographic analysis (described above) indicates a closed Bering Strait.The open or closed state of the Bering Strait, since the mid-Piacenzian, is dictated in large part by global eustatic and regional tectonic processes.Our decision to follow the paleogeographic model is based upon the shallow depth of the seaway and evidence for repeated episodes of subaerial exposure in both the early Pliocene and during the Pleistocene (Hopkins, 1959(Hopkins, , 1967;;Nelson et al., 1974).Circles show dominant Piacenzian biomes for each paleobotanical site.For a full explanation of methods and list of literature for each site, see Salzmann et al. (2008Salzmann et al. ( , 2013)).(b) Geographic distribution of Piacenzian soil localities used in the global reconstructions.Circles show the dominant soil type reconstructed from paleosol occurrences.For a full explanation of the method and list of literature for each site, see Pound et al. (2014).
The effect of a closed Bering Strait has been shown to increase Atlantic Meridional Overturning Circulation (AMOC), which provides for an increase in ocean heat transport to the Arctic (Hu et al., 2010(Hu et al., , 2015)).Climate model simulations to date have been unable to achieve the high-latitude warming estimated for both terrestrial and marine environ-ments of the mid-Piacenzian (Salzmann et al., 2013;Dowsett et al., 2012Dowsett et al., , 2013a)).Hu et al. (2015) show that, regardless of climate state, a closed Bering Strait produces a warmer North Atlantic.Thus, a closed Bering Strait may improve the fit between existing Piacenzian marine and terrestrial data and model simulations.Matthiessen et al. (2009) suggest the possibility of a closed Canadian Arctic Archipelago with Arctic-North Atlantic communication restricted to the Fram Strait.Evidence for this configuration was presented by Harrison et al. (1999), who postulated establishment of an Early Pleistocene drainage system on underlying Pliocene deposits that drape the Canadian Arctic (Harrison et al., 1999) and the presence of fluvio-deltaic sedimentation around the margins of the Arctic continental shelf (Yorath et al., 1975;Jones et al., 1992).This fits well with the PRISM4 paleogeographic model that shows most of the Canadian Arctic Archipelago to be subaerially exposed (Figs. 3, 5).
The depths of the present-day Gulf of Carpenteria, Arafura Sea, and Java Sea, as well as sills in the region, are shallow (Smith and Sandwell, 1997).Our paleogeographic model suggests regional tectonics produced even shallower conditions during the mid-Piacenzian, relative to present-day, from Southeast Asia to the Gulf of Carpenteria (Figs. 3 and 7).A shallower than present Indonesian Throughflow (ITF) would alter regional sea-surface topography and restrict circulation, thus reducing heat transport from the Pacific to the Indian Ocean.Model simulations have shown that the result is an increase in SST and precipitation in the western Pacific with an accompanying decrease in both SST and rainfall in the eastern Indian Ocean.Such changes in a dynamically sensitive region of the tropics can lead to altered atmospheric pressure and wind stresses (Schneider, 1998;Lee et al., 2002;Sprintall et al., 2015) with complex teleconnections capable of distributing the effects globally (Shukla et al., 2011).

Terrestrial
In addition to changes to important ocean gateways mentioned above, general differences exist between the new PRISM4 mid-Piacenzian and present-day paleogeographies.Our reconstruction shows most continental areas at slightly higher elevations during the Piacenzian than they are today (Fig. 7).In our mid-Piacenzian reconstruction, decreased ice loading of the continents plus increased water loading of the oceans, relative to present day, results in mantle displacement from beneath the oceans to beneath the continents, raising most continental elevations (Raymo et al., 2011).Exceptions are areas immediately adjacent to the rift valleys in Africa, the Sierra Madres of Mexico, the Andes north of Lima and south of Santiago, the Chugach in northwestern North America, the Tibetan Plateau, Sumatra, and the northern part of the Great Dividing Range in eastern Australia.
Most of Greenland was at a lower elevation during the mid-Piacenzian, relative to present day, due to the PRISM4 GIS being restricted to eastern Greenland (Koenig et al., 2015;Dolan et al., 2015a).Portions of Antarctica also show large elevation changes due to cryospheric changes implemented in the PRISM4 paleogeography.These changes would have affected atmospheric circulation, precipitation patterns, and distribution of terrestrial vegetation.

Uncertainties in the ice sheet reconstruction
Both the Greenland and Antarctic ice sheets are reduced in terms of their volume and extent in the PRISM4 reconstruction relative to present day.As described previously there is little geological evidence that directly corroborates the exact extent and volume of the major ice sheets during the mid-Piacenzian.Therefore, the ice sheet configurations used in the PRISM4 reconstructions are plausible representations of potential mid-Piacenzian ice sheets.Nevertheless, they have inherent uncertainties associated with their reconstruction which should be noted.Recent studies by de Boer et al. (2015), Dolan et al. (2015a), and Koenig et al. (2015) have demonstrated unequivocally that ice sheet model reconstructions for Antarctica and Greenland are dependent upon both the climate forcing employed and the choice of ice sheet model.As both of the reconstructions are based on model output, there will be some inherent model dependency in the PRISM4 reconstruction (e.g. if we had used a different model/climate forcing, the configuration could be somewhat different).This is less significant over Greenland as the PRISM4 Greenland ice sheet is based upon the results in Koenig et al. (2015) and supported by simulations in Dolan et al. (2015a), which take into account a variety of model differences for simulations over Greenland.The reconstructed Greenland ice sheet is also consistent with modeling results presented in Contoux et al. (2015) and Yan et al. (2014) and a number of different proxy records for the mid-Piacenzian.For example, Bierman et al. (2014) have shown a preservation of a preglacial landscape which suggests potential subaerial exposure in central Greenland for up to 1 million years prior to major ice build-up at 2.7 Ma. Evidence of vegetation suggesting ice-free conditions can be found in northern Greenland (Funder et al., 2001), at Île-de-France (Bennike et al., 2002), on Ellesmere Island, and the Canadian Archipelago (De Vernal and Mudie, 1989;Thompson and Fleming, 1996;Ballantyne et al., 2006;Csank Clim. Past, 12, 1519-1538, 2016et al., 2011), and these offer limited constraints on a mid-Piacenzian Greenland reconstruction.A recent reassessment of pollen derived from ODP Hole 646B off southwestern Greenland (de Vernal and Mudie, 1989) confirms that southern Greenland may have been vegetated (boreal and cooltemperate conditions) during warm parts of the Pliocene (A. de Vernal, personal communication, 2014).Analysis of sea ice at Site 910 (∼ 80 • N, 6 • E) suggest conditions probably similar to the modern sea-ice summer minimum at this time.
A more thorough analysis of model dependency over Antarctica is currently underway; however, de Boer et al. (2015) show that the choice of ice sheet model has a smaller effect than the climate forcing on the reconstructed ice sheet.As the PRISM4 Antarctic ice sheet is not updated from PRISM3 it is necessary to assess whether the reconstruction remains consistent with more recent proxy records and modeling studies.
A recent study by Yamane et al. (2015) using cosmogenic nuclide exposure ages to investigate past ice sheet thickness showed that different areas of the East Antarctic ice sheet were hundreds of meters thicker during the Pliocene.Ice sheet modeling presented within Yamane et al. (2015) suggested that the continental interior could have been up to 600 m higher than today, which is consistent with the increased elevation in our PRISM4 reconstruction.
The reduction in ice at the Wilkes Land margin of East Antarctica is supported by proxy evidence suggesting a dynamic ice margin during the Pliocene (e.g.Young et al., 2014) and ice modeling incorporating a retroacted mid-Piacenzian Antarctic bedrock topography (Austermann et al., 2015).
Moreover, independent simulations presented in Pollard et al. (2015), which are broadly representative of a warm Pliocene climate, are consistent with the areas of ice sheet retreat in the PRISM4 reconstruction (albeit to a lesser extent).Pollard et al. (2015) detail new mechanisms for ice-cliff failure and melt-driven hydrofracture at the grounding line of the Antarctic ice sheet that lead to significant retreat into the Aurora and Wilkes subglacial basins as well as over areas of West Antarctica (producing around 17 m of global sea-level rise).

Ocean surface characteristics and implications
The most pronounced large-scale feature of Piacenzian climate reflected in the marine (and terrestrial) realm is a reduced pole-to-Equator temperature gradient, enhanced by polar amplification and the relative stability of low-latitude SST (Salzmann et al., 2013;Dowsett et al., 2013a;Haywood et al., 2013).Microfossil assemblages from Piacenzian tropical oceans exhibit a high degree of similarity to extant tropical assemblages.Conversely, mid-to high-latitude Piacenzian assemblages appear to be displaced toward the poles relative to present-day distributions.These assemblages pre-sumably migrated in response to changes in climate (Saupe et al., 2014).These displaced Piacenzian faunas suggest polar amplification of surface temperature, in agreement with other independent estimates of Piacenzian SST based upon biomarkers, faunal and floral assemblages, and Mg / Ca paleothermometry (e.g., Lawrence et al., 2009;Bartoli et al., 2006;Cronin, 1988;De Schepper et al., 2009;Barron, 1992;Knies et al., 2014).
Piacenzian marine records from the Arctic are rare; thus, our understanding of environmental conditions in this important region is inadequate.The polar amplification exhibited in our SST reconstruction is supported by analyses from a large number of mid-to high-latitude localities (Dowsett et al., 2012;Fig. 8a).Sites 907, 909, and 911 are the highest latitude sites in the North Atlantic used in our reconstruction and have a large influence on the SST anomaly field for the Piacenzian because of the strong warming indicated at those locations based upon Mg / Ca and alkenone methods (Robinson, 2009).Mattingsdal et al. (2014) revised the age model at Site 911 and it appears that the interval studied by Robinson (2009) now falls outside the PRISM interval.Sites 907 and 909 are located at 69 and 78 • N, respectively.They record SST of ∼ 12 • C.However, Lawrence et al. (2013) noted a sharp decline in productivity in the northern Atlantic starting as far back as 3.3 Ma, with earliest signs of decline occurring at 69 • N and progressing transiently to lower latitudes.In addition, Robinson (2009) proposed that the discontinuous temperature data could represent brief pulses of warmth advecting into the Arctic in a cooler background climate state, similar to what has been recorded by Bauch (1999) and Bjørklund (2012) in younger age material.However, the warm SST data are in agreement with available terrestrial estimates (Ballantyne et al., 2010;Salzmann et al., 2011).
If temperatures in the far northern Atlantic did begin to cool earlier than the Piacenzian, the very warm SSTs in the PRISM data at these sites could be linked to signal carriers brought in by currents or could be the result of reworked older material.There is not sufficient evidence at this point in time of any such reworking, so for now PRISM4 maintains high-latitude warm values at Sites 907 and 909, but community scrutiny of these sites will continue and the PRISM SST reconstruction north of the Arctic Circle should be considered tentative until well-dated, higher-confidence marine and terrestrial data are available.
The routinely used SST proxies in tropical regions all have limitations that may obscure low-latitude warming above preindustrial levels (Dowsett and Robinson, 2009;Dowsett et al., 2013).Faunal assemblage techniques are limited by a calibration to the modern ocean which does not exceed ∼ 30 • C. Similarly, the U K 37 alkenone unsaturation index becomes saturated at 28 • C, precluding its use in warm pool regions.While the Mg / Ca paleothermometer is able to estimate temperatures above 30  fossil tests composed of CaCO 3 were secreted.In addition, syn-and post-depositional processes may alter the geochemical signal and resulting SST estimates.In a recent study, O'Brien et al. ( 2014) applied another biomarker (TEX 86 ) and concluded that previous estimates of Pliocene tropical warm pool temperatures based upon Mg / Ca paleothermometry may have been underestimated.Dowsett (2007b) suggested the possibility of western equatorial Pacific conditions warmer than present day based upon structural changes in planktonic foraminifer assemblages.Thus, where the lowlatitude PRISM SST reconstruction is based upon these proxies, it may be obscuring tropical warming.

Land surface characteristics and implications
Parallel to changes in ocean surface temperatures, the zonation of terrestrial biomes during the Piacenzian also indicates a diminished Equator-to-pole thermal gradient enhanced by polar amplification (Salzmann et al., 2013).The most prominent changes in Northern Hemisphere Piacenzian biome distribution compared to present day include a northward shift of temperate and boreal vegetation zones in response to a warmer and wetter climate as well as expanded tropical savannas and forests at the expense of deserts (Fig. 6a).Climate estimates based on paleobotanical and multiproxy evidence indicate that mean annual temperatures in the northern high Arctic were, during warm Piacenzian intervals, between 14 and 20 • C higher than today (Fig. 6a; e.g., Ballantyne et al., 2010;Andreev et al., 2012).There is evidence that greater warming occurred in the winter than during summer (e.g., Elias and Matthews, 2002;Wolfe, 1994;Pound et al., 2015).
A globally wetter climate during the Piacenzian supported the formation of megalakes, including Lake Zaire and Lake Chad in Africa and Lake Eyre in Australia (e.g., Drake et al., 2008;Otero et al., 2009).Model sensitivity experiments show that the addition of soils to the new BIOME4 recon-struction notably increased surface air temperatures in Australia, southern North Africa, and Asia, whereas megalakes generated an increase in precipitation in central Africa and western North America (Pound et al., 2014).
The PRISM4 terrestrial surface reconstruction has a number of uncertainties, particularly in regions of low data coverage such as central North America, South America, and northwestern Africa, including the Sahara (Fig. 6a).Additional uncertainties are caused by insufficient age control and resolution of individual paleorecords (Salzmann et al., 2012).The global reconstruction shown in Figs. 3 and 6a, like the SST reconstruction, present "snapshots" of terrestrial environments which did not necessarily co-occur in one orbitally defined "interglacial" or "glacial" interval.Highresolution records available for the Piacenzian (e.g.Andreev et al., 2014;Panitz et al., 2015) indicate large shifts of vegetation zones during warm/wet and cold/dry periods within the PRISM interval (Fig. 6a).Although these shifts clearly indicate that the Piacenzian warm climate was not stable and uniform, it should be noted that the Piacenzian vegetation shifts were minor in comparison with Pleistocene interglacial-glacial fluctuations.

Chronology
The single most important element of any paleoenvironmental reconstruction is the confidence that can be placed in the chronologic framework within which it is created.In a general sense, Pliocene geochronology is stable compared to other parts of deep time, and the combination of radiometric dating, biochronology, paleomagnetic stratigraphy, and LR04 marine oxygen isotope stratigraphy provides the potential to place all events on the same temporal scale with high confidence.
Ideally, all points in a synoptic reconstruction should represent the same instant in time; however, no paleoreconstruction can actually achieve synchrony.The PRISM time slab covers six "interglacials" in a ∼ 260 ky interval (Fig. 1).The warm peak averaging technique used for SST estimation strives to develop mean interglacial conditions within the time slab, at each site.There is no a priori reason to assume the values obtained are synchronous across multiple sites.Analysis of more recent glacial terminations shows diachronous δ 18 O responses between ocean basins of up to 4 ky, making correlations independent of climate signals even more important (Skinner and Shackleton, 2005;Lisiecki and Raymo, 2009).In both the marine and terrestrial realms, synand post-depositional processes result in time-averaged signals where most of the variance is concentrated at low frequencies, further complicating the synoptic nature of reconstructions (Bradley, 1999).
For the purposes of a conceptual understanding of the mid-Piacenzian, however, this equilibrium climate reconstruction has proven most useful.By providing uncertainty ranges and paleoenvironmental estimates in terms of cold/dry and warm/wet (terrestrial) and maximum, mean, and minimum warming (marine), the data sets provide guidelines and realistic limits for a robust Piacenzian global paleoenvironmental reconstruction.For the purposes of verifying paleoclimate model simulations, the time slab would ideally need to collapse (temporally) as much as possible toward a true time slice, like the 3.205 Ma datum proposed by Dowsett et al. (2013b) and Haywood et al. (2013).At that time, correlative with MIS KM5c, orbital forcing was close to presentday forcing.In addition, simulations suggest minimal surface temperature changes for 20 ky before and 20 ky after isotope peak KM5c (Prescott et al., 2014).Even then, due to the time averaging discussed above, the practice of comparing model output and paleoenvironmental estimates derived from geological archives still has a number of methodological limitations and uncertainties (Haywood et al., 2016a).

Paleoenvironmental estimates
The confidence we have in the PRISM4 reconstruction and in each of its components speaks directly to its value in advancing our understanding of past warm climates as well as our ability to verify model simulations.Salzmann et al. (2013) assessed the confidence placed in terrestrial paleoenvironmental estimates (surface air temperature and precipitation) for terrestrial localities included in PRISM3.Dowsett et al. (2012) attempted to place a semiquantitative estimate of confidence (λ) on the marine localities and materials they yielded for paleoenvironmental estimation, as well as on the performance of the techniques used to form temperature estimates.These λ values allow others to know how confident, in a relative sense, those most familiar with the data are of the process.While a step in the right direction, λ does not address calibration errors or impacts of processes that can-not be quantified.For example, under a certain level of postdepositional carbonate dissolution, fragile thermophilic taxa may be preferentially removed from a planktic foraminifer assemblage, making the resulting assemblage appear to come from a colder environment.Quantifying the degree of dissolution, or in some cases whether dissolution has taken place, is difficult at best.Also, regardless of fossil group or technique, all methods are in effect analog methods and require an assumption of stationarity.While it is possible to measure the impact that drift in environmental preferences of a particular taxon has on the resulting environmental estimate, it is not necessarily possible to quantify the degree to which drift took place.Therefore, these confidence schemes are hybrids between those parts of the environmental estimate (e.g., temperature) that can be quantified and those based upon expert judgment.

PlioMIP2 data-model comparisons
A first-order outcome of the initial PlioMIP data-model comparisons (Dowsett et al., 2012(Dowsett et al., , 2013a;;Salzmann et al., 2013) was to show that using paleoenvironmental estimates that represent a time-averaged equilibrium climate to verify simulated conditions representing a specific temporal horizon is problematic.In an effort to reduce uncertainty, PlioMIP2 experimental protocols now call for sub-orbital resolution verification data around a specific stratigraphic target of interest that exhibits orbital forcing close to that of present day (Dowsett et al., 2013b;Haywood et al., 2013).An initial time slice centered at 3.205 Ma (Fig. 1) has been adopted by the community for the generation of short time series to be used for comparison with PlioMIP2 experiments (Haywood et al., 2012(Haywood et al., , 2016b)).
PRISM4 is generating short high-resolution time series between MIS M2 and MIS KM3 (Fig. 8; see Figs.S2  and S3).Where independent means of correlation are available through the oxygen isotope record, some sequences will achieve suborbital chronologic resolution.This allows a more direct comparison of change at multiple sites from approximately the same time.
Time series with orbital-scale chronological resolution sufficient to provide estimates within the time slice window are not abundant on a global scale and therefore data-model comparisons will need to have a regional focus.Alternatively, transient experiments provide a method for comparison of model-derived paleoclimate variables and time series developed from geological archives.Comparing transient simulations to time series will provide much needed information on the variability and dynamics of Piacenzian climate change and may have less of a requirement for synchrony.Figure 8b shows the location of some of the initial PRISM4 time series for the North Atlantic region.Two of these sequences also carry high-resolution pollen records which allow marine-terrestrial correlation and comparison with simulations (Panitz et al., 2015).

Paleoenvironmental data integration
Localities suitable for high-resolution marine-terrestrial correlation (Fig. 8b) also lend themselves to a different approach to paleoenvironmental reconstruction, which in turn provides an alternative for data-model comparison.Dowsett et al. (2013b) outlined challenges to conventional paleoclimate reconstruction citing as rationale information loss due to underutilization of multivariate data sets capable of a more holistic reconstruction.They called for new methods focused on regional and/or process-oriented reconstructions similar to those being developed for PRISM4.This paleoenvironmental data integration (PDI) moves away from single variables like temperature and uses the full potential of paleontological and geochemical data to provide a more nuanced holistic paleoenvironmental reconstruction (Figs.s2 and S3).
These regional and process-oriented reconstructions are useful both in terms of our conceptual understanding of Piacenzian conditions and as the basis for a more robust verification of models.Rather than compare a single variable (e.g., temperature) in isolation, Earth system models can use a variety of output to simulate paleoenvironments and compare those with Piacenzian reconstructions that are strengthened by multiple lines of independent geological evidence.Placing error bars on individual paleoenvironmental estimates (e.g., surface temperature) generally only takes into account a small number of potential contributions to uncertainty.The overall agreement between model and data, using multiple complimentary environmental parameters, would be a strong indication of robust model performance.It is otherwise possible to have univariate agreement for the wrong reason.Stated differently, it is conceivable for both model and data to produce an identical surface temperature, but under completely different environmental conditions.Thus, the PDI approach would provide an alternative indication of the models predictive abilities.

Summary and conclusions
We have assembled an integrated series of data sets that together reflect our current best understanding of the features of the Pliocene (mid-Piacenzian) world.The PRISM4 reconstruction indicates reduced pole-to-Equator temperature gradients with polar amplification due in large part to reduced sea-ice extent.The current reconstruction contains, for the first time, an independent elevation (and bathymetry) data set that takes into account GIA and dynamic topography.This reconstruction suggests shoaling in the Indo-Pacific throughflow region, a closed Bering Strait, and a much-reduced connection between the Atlantic and Arctic oceans.The inclusion of soils and lakes increases the diversity of information provided for the terrestrial realm.These are the result of, and forcing agents for, a warmer and wetter Piacenzian.
There is an inverse relationship between the acuity with which we understand the paleoenvironment and the degree of confidence we have in that understanding.Future PRISM4 work is focusing on high-resolution sequences including but not limited to paleontological data preserved in marine and terrestrial sediments with attributes that allow correlation to the geomagnetic polarity timescale and/or marine isotopic stages, and unique settings with excellent age control.The new regional and process-oriented approach will make use of the full potential of paleontological and geochemical data without over-interpreting temperature or precipitation in order to provide a more confident, nuanced holistic paleoenvironmental reconstruction.
Piacenzian data, including PRISM data, will be used to verify the next set of PlioMIP2 paleoclimate simulations using several new developments.Conventional datamodel comparison will be realized using samples from high resolution time series that intersect the PlioMIP2 time slice (3.205 Ma) and by comparison of time series to transient model experiments between 3.3 and 3.2 Ma.An everincreasing number of model simulations will enable us to cross-check our interpretations and develop paleoenvironment settings.This type of locality and region-based reconstruction, while not readily conforming to conventional methods of data-model comparison, does provide constraints on simulated climates both in time and space.
Given the high priority of understanding the mid-Piacenzian within the paleoclimate community, we anticipate a combination of conventional and holistic approaches will provide enhanced understanding of the time period and a better and independent assessment of paleoclimate model performance.

Data availability
All elements of the PRISM4 reconstruction are available from the data section of the USGS PRISM web page: http: //geology.er.usgs.gov/egpsc/prism/4_data.html.
The Supplement related to this article is available online at doi:10.5194/cp-12-1519-2016-supplement.Edited by: W.-L. Chan Reviewed by: two anonymous referees

Figure 1 .
Figure 1.PRISM Pliocene magnetobiostratigraphic and stable isotopic framework.Benthic δ 18 O record from Lisiecki and Raymo (2005).Position of PRISM3 time slab (yellow band) and extent of PRISM4 time series (green highlighted section of LR04) between MIS M2 and MIS KM5.PlioMIP2 time slice (3.205 Ma) shown by horizontal red line.Magnetic polarity ages from Lourens et al. (1996).

Figure 2 .
Figure 2. Interrelationships between PRISM3 and different parts of the PRISM4 reconstruction.Twelve data sets representing components of our Piacenzian reconstruction are shown in yellow and orange rectangles (PRISM3 and PRISM4, respectively).Starting points in the generation of these data sets are indicated by teal rounded rectangles.These are the basic geochemical, faunal, floral, soil, cryospheric, topographic, bathymetric, sedimentologic, and stratigraphic data.Marine temperature estimates (SST and DOT (deep-ocean temperature)) are based upon multiple proxies (including faunal, floral, geochemical, and biomarker analyses).Terrestrial vegetation data are translated into a BIOME4 scheme and compared to an independent BIOME4 landscape produced by the HadAM3/TRIFFID model (red circle) initialized with PRISM2 boundary conditions.Integration of data and model output results in the hybrid PRISM3 BIOME reconstruction(Salzmann et al., 2008).Antarctic and Greenland ice sheets are based upon PRISM3 and PLISMIP results, respectively.Paleogeography is based upon an initial ETOPO1 digital elevation model incorporating PRISM4 ice sheets, GIA, and adjustments due to mantle convection (red circle).Soils are determined through comparison of sedimentological and stratigraphic data with the PRISM3 BIOME reconstruction.Lakes are determined from stratigraphic and sedimentological data(Pound et al., 2014).Sea-level is estimated from the PRISM4 ice sheets and is consistent with the paleogeography presented here.

Figure 4 .
Figure 4. Generation of the PRISM4 paleogeography.Panel (a) shows "bedrock" (base of the ice sheets) version of ETOPO1 sampled at the working resolution of a quarter degree, (b) GIA correction, (c) Pleistocene landscape fill, (d) Pleistocene landscape features filled in, (e) dynamic topography correction, (f) dynamic topography correction applied to (d), (g) Piacenzian ice sheets, and (h) retrodicted PRISM4 paleogeography (g) applied to (f).

Figure 6 .
Figure 6.(a) Vegetation reconstruction for a cold/dry (upper square) and warm/wet (lower square) period of the Piacenzian (3.6-2.6 Ma).Circles show dominant Piacenzian biomes for each paleobotanical site.For a full explanation of methods and list of literature for each site, seeSalzmann et al. (2008Salzmann et al. ( , 2013)).(b) Geographic distribution of Piacenzian soil localities used in the global reconstructions.Circles show the dominant soil type reconstructed from paleosol occurrences.For a full explanation of the method and list of literature for each site, seePound et al. (2014).

Figure 8 .
Figure 8. PRISM North Atlantic region data.(a) Map of PRISM3 North Atlantic region SST anomalies (from Dowsett et al., 2013a).(b) Location of PRISM4 sites having integrated paleoenvironmental interpretations and time series.
. McKay et al. (2012) document a dynamic ice sheet prior to 3.3 Ma followed by expansion of the ice sheet and concomitant SST cooling.Recent investigations of the Wilkes and Aurora subglacial basins (East Antarctica) indi-

Table 1 .
Components of the PRISM4 paleoenvironmental reconstruction.
and remain unchanged from the PRISM3 reconstruction.These fields are based upon multiple proxy analyses derived us- www.clim-past.net/12/1519/2016/Clim.Past, 12, 1519-1538, 2016 • C, successful application requires knowledge of the composition of seawater at the time Acknowledgements.Harry Dowsett, Marci Robinson, and  KevinFoley are supported by the US Geological Survey Climate and Land Use Change Research and Development Program.Aisling Dolan and Alan Haywood acknowledge that this research was completed in receipt of funding from the European Research Council under the European Union's Seventh Framework Programme (FP7/2007-2013)/ERC grant agreement no.278636.Ulrich Salzmann, Alan Haywood, and Matthew Pound acknowledge funding received from the Natural Environment Research Council (NERC grant NE/I016287/1).David Rowley, Alessandro M. Forte, Jerry X. Mitrovica, and Robert Moucha acknowledge support from the Canadian Institute for Advanced Research's Earth Clim.Past, 12, 1519-1538, 2016 www.clim-past.net/12/1519/2016/System Evolution Program.Alessandro M. Forte also thanks the Natural Sciences and Engineering Research Council of Canada.Mark Chandler is supported by the NASA Modeling, Analysis, and Prediction program (NASA grant NNX14AB99A) and the NASA High-End Computing (HEC) Program through the NASA Center for Climate Simulation (NCCS) at Goddard Space Flight Center.We thank Daniel Hill, Stephen Hunter, Linda Sohl, and Adam Bloemers for helpful input and Robert Schmunk for the Panoply visualization software.Harry Dowsett, Aisling Dolan, Alan Haywood, Ulrich Salzmann, and Matthew Pound also thank the EPSRC-supported Past Earth Network.This research used samples and/or data provided by the International Ocean Discovery Program (IODP), Ocean Drilling Program (ODP), and Deep Sea Drilling Project (DSDP).