Global climate simulations at 3 , 000-year intervals for the last 21 , 000 years 3 with the GENMOM coupled atmosphere-ocean model 4 5

35 We apply GENMOM, a coupled atmosphere-ocean climate model, to simulate 36 eight equilibrium time slices at 3000-yr intervals for the past 21,000 years forced by 37 changes in Earth-Sun geometry, atmospheric greenhouse gases (GHGs), continental ice 38 sheets and sea level. Simulated global cooling during the Last Glacial Maximum (LGM) 39 is 3.8 °C and the rate of post-glacial warming is in overall agreement with recently 40 published temperature reconstructions. The greatest rate of warming occurs between 15 41 and 12 ka (2.4 °C over land, 0.7 °C over oceans and 1.4 °C globally) in response to 42 changes in radiative forcing from the diminished extent of the Northern Hemisphere 43 (NH) ice sheets and increases in GHGs and NH summer insolation. The modeled LGM 44 and 6 ka temperature and precipitation climatologies are generally consistent with proxy 45 reconstructions, the PMIP2 and PMIP3 simulations, and other paleoclimate data-model 46 analyses. The model does not capture the mid-Holocene ‘thermal maximum’ and gradual 47 cooling to pre-industrial global temperature found in the data. Simulated monsoonal 48 precipitation in North Africa peaks between 12 and 9 ka at values ~50% greater than 49 those of the PI, and Indian monsoonal precipitation peaks at 12 and 9 ka at values ~45% 50 greater than the PI. GENMOM captures the reconstructed LGM extent of NH and 51 Southern Hemisphere (SH) sea ice. The simulated present-day Antarctica Circumpolar 52 Current (ACC) is ~48% weaker than the observed (62 versus 119 Sv). The simulated 53 present-day Atlantic Meridional Overturning Circulation (AMOC) of 19.3 ± 1.4 Sv on 54 the Bermuda Rise (33°N) is comparable with observed value of 18.7 ± 4.8 Sv. AMOC at 55 33°N is reduced by ~15% during the LGM, and the largest post-glacial increase (~11%) 56 occurs during the 15 ka time slice. 57


Introduction
The history of the climate system over the past 21 000 years reflects the combined changes in Earth-Sun orbital geometry, atmospheric greenhouse gas concentrations (GHGs; see Table A1 for list of abbreviations and acronyms), the extent of the Northern Hemisphere (NH) ice sheets, and sea level.GHG levels were lowest during the Last Glacial Maximum (LGM, ∼ 21 000 years ago, 21 ka) and increased thereafter to preindustrial (PI) levels (Brook et al., 2000;Monnin et al., 2001;Sowers et al., 2003).The LGM is further characterized by the large Laurentide (LIS), Cordilleran (CIS) and Fennoscandian (FIS) ice sheets.The height and extent of the ice sheets altered atmospheric circulation patterns, and the extent increased the NH albedo, thereby altering the global radiative balance.The effect of the ice sheets on climate progressively diminished from the LGM to the early Holocene as global warming driven by increasing GHGs combined with changes in NH summer insolation to accelerate ice sheet ablation.Abrupt departures from the comparatively smooth transition from the LGM through the Holocene, such as Heinrich and Dansgaard-Oeschger events, the Bølling-Allerød (BA), and the Younger Dryas (YD), are evident in geologic records, and these events likely influenced the overall trajectory of the deglaciation.
The climate of the past 21 000 years has been studied extensively, beginning with three international collaborative projects -Climate: Long range Investigation, Mapping, and Prediction (CLIMAP; CLIMAP Project Members, 1981) and the Cooperative Holocene Mapping Project (COHMAP; COHMAP Members, 1988), which evolved into the Testing Earth System Models with Paleoenvironmental Observations (TEMPO) project (Kutzbach et al., 1996a(Kutzbach et al., , 1998)).CLIMAP Published by Copernicus Publications on behalf of the European Geosciences Union.
focused on reconstructing the LGM climate; COHMAP focused on reconstructing the climate of seven time periods (18, 15, 12, 9, 6, and 3 ka); and TEMPO focused on reconstructing the climate of 21, 16, 14, 11, and 6 ka.These three projects pioneered data-model comparison through integrating climate model simulations and paleoclimatic data, which motivated the development of new techniques for analyzing geologic data and led to improvements in general circulation models.
More recently, the Paleoclimate Modelling Intercomparison Project (PMIP) is actively working to advance reconstruction of LGM and 6 ka climate through model-to-model evaluations and data-model comparisons.PMIP has now entered the third phase (PMIP3; Braconnot et al., 2012) and is a component of phase 5 of the Climate Model Intercomparison Project (CMIP5).In contrast to CLIMAP, COHMAP, TEMPO, and earlier PMIP model experiments that employed fixed sea surface temperatures (SST) and mixed-layer ocean models, some of the PMIP2 experiments and all of the PMIP3 experiments include fully coupled ocean and atmospheric models.Braconnot et al. (2012) review some of the highlights of the PMIP2 experiments and the design of the PMIP3 experiments and Harrison et al. (2013) evaluate the PMIP3 and PMIP2 simulations of LGM and 6 ka climates with data-model comparisons.In addition, continuous simulations of climate over the last 21 ka have been achieved with Earth system models of intermediate complexity (e.g., Timm and Timmermann, 2007), and the TraCE-21ka project at the National Center for Atmospheric Research (NCAR) conducted continuous, transient climate simulations from 22 to 6.5 ka with the coupled NCAR Community Climate System Model (Liu et al., 2009).Singarayer and Valdes (2010) simulated the climate of the last 120 000 years using model snapshots at 4 and 1 kyr intervals.
Here we explore past changes in late Pleistocene climate using the coupled atmosphere-ocean general circulation model (AOGCM) GENMOM.We simulated multicentury time slices that span the interval from LGM to preindustrial (PI) every 3000 years (21,18,15,12,9,6,and 3 ka and PI).The simulations were run with prescribed insolation, GHG concentrations, continental ice sheets, land extent, and sea level as boundary conditions.We analyze the within and between climatology of the time slices and compare the 21 and 6 ka results with terrestrial and marine climate reconstructions and results from the PMIP2 and PMIP3 simulations.The goal of our simulations is to adopt a methodological framework similar to that of PMIP to simulate time slices between the LGM and mid-Holocene.The simulations also serve as a baseline for applying GENMOM to more detailed and focused studies of late Pleistocene climate such as quantifying the effects of freshwater forcing and dynamic vegetation feedbacks.
In our simulations, we employ a coupled model with T31 spectral truncation, which corresponds to a grid of 96 longitudes (3.75 • ) by 48 Gaussian latitudes (∼ 3.71 • ).The atmosphere is represented by 18 vertical sigma levels with mid-layers ranging from 0.993 at the surface to 0.005 at the tropopause.GENESIS includes the Land Surface eXchange model, LSX (Pollard and Thompson, 1995), to simulate surface processes and to account for the exchange of energy, mass, and momentum between the land surface and the atmospheric boundary layer.MOM2 has 20 fixed-depth vertical levels and is implemented on essentially the same T31 horizontal grid as GENESIS through cosine-weighted distortion (Pacanowski, 1996).Sea ice is simulated by a three-layer model that accounts for local melting, freezing, and fractional cover (Harvey, 1988;Semtner, 1976) and includes the dynamics associated with wind and ocean current using the cavitating-fluid model of Flato and Hibler (1992).The atmospheric and ocean models interact every 6 h without flux corrections.
GENMOM reproduces observed global circulation patterns, such as the seasonal change in the position and strength of the jetstreams and the major semi-permanent sea level pressure centers (Alder et al., 2011).The simulated presentday (PD) 2 m air temperature climatology (Table 1) is 0.8 • C colder than observations globally, 0.7 • C colder over oceans, and 0.9 • C colder over land.Similar to other AOGCMs (e.g., Lee and Wang, 2014), GENMOM produces a split ITCZ over the equatorial Pacific Ocean.
The preindustrial Atlantic Meridional Overturning Circulation (AMOC) simulated by GENMOM is 19.3 ± 1.4 Sv, which is stronger than, but comparable to, the observed value of 17.4 Sv (Srokosz et al., 2012).Simulated SSTs display a  (Monnin et al., 2001), CH 4 (Brook et al., 2000), and N 2 O (Sowers et al., 2003) are estimated from ice core records by averaging the gas concentrations within a ± 300-year window centered at the time of interest.For comparison, the PMIP3 concentrations for 6 ka are 280 ppmV and 650 and 270 ppbV for CO 2 , CH 4 , and N 2 O, respectively, and 185 ppmV and 350 and 200 ppbV for 21 ka.In the table, e is eccentricity, ω-180 is precession, and ε is obliquity (Berger and Loutre, 1991).warm bias in some regions of the Southern Ocean, primarily south of 50 • S around Antarctica, and a warm bias exceeding ∼ 2 • C between 200 and 1000 m depth in parts of the tropics and midlatitudes.Alder et al. (2011) note that the warm bias in the Southern Ocean is associated with the relatively weak Antarctic Circumpolar Current (ACC) in GEN-MOM (62 Sv versus the observed value of 119 Sv) and Deacon cell upwelling, which allows excessive vertical mixing in the present-day GENMOM simulation, and that these together reduce sea ice around Antarctica, particularly during summer.Both of these features are present to some extent in our suite of simulations.We tested the Gent-McWilliams vertical ocean mixing scheme (Gent and Mcwilliams, 1990) in GENMOM, but it did not improve the Southern Ocean warm bias, so we did not implement it in our paleosimulations.
The climate sensitivity of GENMOM for a doubling of CO 2 from present day is 2.2 • C, which is in the lower range of other coupled AOGCMs (Meehl et al., 2007) and is consistent with recent estimates of 2.7 • C based on the PMIP3 LGM simulations (Harrison et al., 2013) and paleodatamodel estimates of 2.8 • C (Annan and Hargreaves, 2013) and 2.3 • C (Schmittner et al., 2011b).
The average NH 2 m temperature in our PI simulation is 0.79 • C cooler than our PD simulation, reflecting lower GHG concentrations, whereas the PD simulation is 1.97 • C cooler than observations and reflects the lower GHG concentrations specified in the PI simulation (Tables 1 and 2).The PI-to-PD warming in the NH is similar to the observed warming of ∼ 0.6-0.9• C (Brohan et al., 2006) and is in the range of response of other climate models (e.g., Otto-Bliesner et al., 2006b).The greatest regional warming between the PI and PD simulations (not shown) is ∼ 3 • C over the high Table 2. Annual average 2 m air temperatures and precipitation rates for the time slice simulations.NCEP is from the National Center for Environmental Predication NCEP/NCAR Reanalysis data set (Kalnay et al., 1996), PD2X is the 2× CO 2 simulation, PD is present day, and PI is preindustrial.Parenthetical values are the changes from the previous time slice, e.g., the global average temperature for the PD is 0.77 northern latitudes and northern polar regions during boreal autumn, winter, and spring, consistent with the observed polar amplification (Hassol, 2004).

Experimental design
We applied GENMOM to eight time periods for 21, 18, 15, 12, 9, 6, 3 ka, and preindustrial.We prescribed insolation at the top of atmosphere for each time slice (Fig. 1) by specifying appropriate orbital parameter values for precession, obliquity, and eccentricity (Table 1; Berger and Loutre, 1991).The solar constant was set to 1367 W m −2 for all time periods.We estimated GHG concentrations from ice-core records by applying a ± 300-year averaging window centered on the time period of interest (Table 1), and we specified the PMIP3 GHG concentrations for our PI simulation (Braconnot et al., 2007a).
To derive continental ice sheets for the time slices, we used the ICE-4G reconstructions (Peltier, 2002) for the Fennoscandian (FIS) and Cordilleran (CIS) ice sheets and the Oregon State University Laurentide Ice Sheet (OSU-LIS) reconstruction (Hostetler et al., 1999;Licciardi et al., 1998) (Fig. 2).The ICE-6G reconstruction was not available for our eight time slices at the time we began our simulations.However, the OSU-LIS reconstruction has similar ice sheet topography to that of ICE-6G (Ullman et al., 2014) and was available for our simulation periods.The combination of OSU-LIS and ICE-4G enables us to use a more realistic LIS topography than that of ICE-5G, particularly over the LIS during the deglacial, and facilitates adjusting sea level throughout our time slices.A similar ice sheet configuration (OSU-LIS and ICE-5G) was used as a boundary condition in the NASA GISS-E2-R LGM simulation submitted to the PMIP3 archive (Ullman et al., 2014).We specified the 10 ka OSU-LIS ice sheet to ensure that Hudson Bay remained covered by the LIS at 9 ka (Dyke and Prest, 1987).The ICE-4G reconstruction includes an ice sheet in eastern Siberia, which we removed because it did not exist (Felzer, 2001).
Topographic heights of the land masses were altered to reflect relative sea level change in ICE-4G.We created the topography and land mask for each time slice by applying orographic changes to the present-day Scripps global orography data set (Gates and Nelson, 1975).Orographic changes are based on ICE-4G exposed or flooded land grid cells associated with relative sea level (e.g., Indonesia, Papua New Guinea).We set the ocean bathymetry to modern depths for ocean grid cells.
We specified the modern distribution of vegetation (Dorman and Sellers, 1989) for all simulations because reconstructions of global vegetation for all time slices either do not exist or are not well constrained.We note that while setting vegetation to modern distribution for all simulations isolates the period-to-period climate response to other boundary conditions, we do not capture dynamic vegetation-climate feedbacks that may be important in some regions such as North Africa (Kutzbach et al., 1996b;Timm et al., 2010) and the high latitudes of the NH (Claussen, 2009;Renssen et al., 2004).The vegetation type on emergent land cells is set to be the same as neighboring existing land cells.The simulations do not include varying dust forcing across the time slices, which may account for up to 20 % of the radiative change (Köhler et al., 2010;Rohling et al., 2012).Freshwater flux from land-based precipitation is globally averaged and spread over the world ocean (Alder et al., 2011).
In accordance with the PMIP3 protocol, to conserve atmospheric mass we compensated for changes in global topography in each time slice by holding global average surface pressure constant.At T31 resolution the Bering Strait and the Strait of Gibraltar are closed in the default MOM2 bathymetry.We conducted sensitivity tests and adjusted the bathymetry to ensure that key passages (e.g., Drake Passage, Norwegian Sea, and Indonesian Throughflow) were adequately represented.Additional sensitivity testing revealed that the modeled AMOC and salinity of the Arctic are very sensitive to the bathymetry of the Norwegian Sea, particularly to the width of the passage between Scandinavia and Greenland as it narrowed by the growth of the FIS.We removed Iceland from the model to ensure that the passage remained sufficiently wide and deep to prevent unrealistic buildup of salinity in the Arctic.
Each time slice simulation was initialized from a cold start (isothermal atmosphere, latitudinally dependent ocean temperature profile, and uniform salinity of 35 ppt) and run for 1100 years.We exclude the first 1000 years from our analysis here to allow for spinup of ocean temperatures.The temperature drift in the last 300 years of our simulations (Fig. S1 in the Supplement) is acceptably small (Braconnot et al., 2007a;Singarayer and Valdes, 2010), with values of −0.05 • C per century for the LGM, 0.01 • C per century for 6 ka, and −0.02 • C per century for PI.Drift in the LGM and early deglacial simulation is attributed primarily to long-term cooling and the evolution of sea ice in the southern ocean.Simulated AMOC exhibits decadal-scale variability, but was free of drift over the last 300 years of the simulations.
In what follows, the monthly averages of the model output are based on the modern calendar as opposed to the angular calendar that changes with Earth-Sun geometry (Pollard and Reusch, 2002;Timm et al., 2008).The modern calendar is commonly used in data-model comparisons (e.g., Harrison, 2013).

Atmospheric circulation
The boreal winter (DFJ) 500 hPa heights in the PI simulation (Fig. 3a) display the observed high-and midlatitude ridgetrough-ridge-trough standing wave structure (wave number two) that arises from continent-ocean-continent-ocean geography of the NH (Peixoto and Oort, 1992).From 21 to 9 ka the LIS, the FIS, and Greenland Ice Sheet alter the NH standing wave structure, resulting in persistent, distinct troughs and cyclonic flow tendencies over northeastern Asia, the North Pacific, the continental interior of North America,  (Peltier, 2002) for the Fennoscandian, Cordilleran, and Antarctic, and OSU-LIS (Licciardi et al., 1998) for the Laurentide.the North Atlantic, and Europe (Fig. 3; maps of raw fields, Fig. S2).
Consistent with previous LGM studies using comparable (Braconnot et al., 2007a) and higher-resolution (Kim et al., 2007;Unterman et al., 2011) climate models, from 21 through 9 ka, the western edge of the Cordilleran Ice Sheet diverts the LGM winter polar jetstream, resulting in one branch that is weaker than PI over the Gulf of Alaska and the western and central regions of the ice sheet and a second branch to the south of the ice sheet that is stronger than the PI (Fig. 3a).The reorganization creates westward wind anomalies over the North American Pacific Northwest.The LIS effectively guides the convergence of the branches, and the meridional gradient of low and high 500 hPa height anomalies in the North Atlantic intensifies flow over North America, the North Atlantic, Europe, and North Africa (Figs. 3a), thereby altering the path of storm tracks.This flow pattern weakens progressively as the LIS recedes.
The influence of the NH ice sheets is also evident in summer (JJA), but to a lesser degree than in winter (Fig. 3b) due to continental heating and the absence of the strong, midlatitude storm tracks.Between 21 and 15 ka, the summer jetstream is constrained and therefore enhanced along and to the south of the southern margin of the LIS extending over the North Atlantic.At 18 ka, a trend toward positive JJA anomalies in 500 hPa heights emerges over the regions of the semi-permanent subtropical high pressure of the North Pacific and central Atlantic.The regions of positive height anomalies, and their associated anticyclonic wind anomalies, expand over central North America, peak from 12 through 9 ka, and diminish by 6 ka (Fig. 3).The DJF pattern of low-to-high height anomalies over the North Atlantic is replaced during JJA by a strengthened subtropical high.Anticyclonic flow around positive height anomalies on the western edge of the FIS alters regional flow patterns over and south of the ice sheet.The GENMOM responses to the NH ice sheets are similar to many previous modeling experiments that have established that changes in tropospheric pressuresurface heights and winds are primarily driven by changes in ice sheet height, and secondarily by temperature and albedo feedbacks (COHMAP Members, 1988;Felzer et al., 1996;1998;Otto-Bliesner et al., 2006a;Pausata et al., 2011;Pollard and Thompson, 1997;Rind, 1987).
From 21 to 12 ka, the largest changes in boreal winter sea level pressure (SLP) are associated with negative surface temperature anomalies over the continental ice sheets, the landmasses of the NH, and areas of expanded sea ice in the North Atlantic (Fig. 4a) where cooling increases subsidence and thus contributes to cold high surface pressure.From 21 to 15 ka, high pressure over the LIS produces anticyclonic flow across the northern Great Plains and over the Puget Lowland of the US.Similar anticyclonic tendencies are simulated along the margin of the FIS.Between 12 and 6 ka the winter SLP around the Aleutian low in the North Pacific and the Icelandic low in the North Atlantic is strengthened relative to PI.The Aleutian low is expanded southward, whereas the Icelandic low is confined on the northern edge by the FIS and is slightly displaced southeastward.
From 21 to 9 ka, the JJA SLP anomalies remain strongly positive over the ice sheets and sea ice, whereas from 12 to 6 ka the SLP anomalies over Northern Hemisphere landmasses are negative due to enhanced continental warming (Fig. 4b).The patterns of the JJA 500 hPa heights, SLP and the associated circulation over North America and adjacent oceans again illustrate similar responses to time-varying controls: changes from 21 to 15 ka are primarily driven by changes in the LIS, whereas from 12 to 6 ka the circulation changes are related to the changes in the seasonality of Holocene NH insolation (Fig. 2).

Near-surface air temperature
Our time slice simulations clearly display surface air temperature (SAT) changes attributed to radiative forcing from the presence of the continental ice sheets, GHGs (Clark et  and Indian monsoon regions, increased cloudiness associated with enhanced summer monsoonal precipitation leads to cooling from 15 to 6 ka. The relatively high rate of warming between 18 and 15 ka (1.5 • C land and 0.5 • C ocean, Fig. 7, Table 2) is commensurate with increased GHGs (Table 1).Periods of peak annual warming from 15 to 12 ka (2.4 • C land and 0.7 • C ocean) and   and 6a).The simulated rates of annual global warming between the LGM and the early Holocene (Fig. 5) are in agreement with data (Clark et al., 2012;Gasse, 2000) and the analyses by Shakun et al. (2012) and Marcott et al. (2013), who attribute a large component of the warming to rising GHG levels.
The DJF and JJA temperature differences in our 21 ka simulation are similar to those of the PMIP3, allowing for differences in between our prescribed NH ice sheets (ICE-4G+OSU-LIS in GENMOM) and the blended ice sheet of the PMIP3 simulations that essentially combines the height of the ICE-6G reconstruction with the extent of the Dyke and Prest (1987) reconstructions (Figs.S5-S10; Braconnot et al., 2012).In both seasons, GENMOM produces 0.5-1 • C less cooling in the tropical oceans and greater warming over Beringia.The positive JJA temperature anomaly south of the FIS in GENMOM persists through 15 ka.Summer warming in the presence of the ice sheet was identified in earlier versions of GENESIS (Pollard and Thompson, 1997) and is associated with subsidence over the ice (Rind, 1987).Similar JJA warming also occurs in some of the PMIP3 models, but is likely a model artifact (Pollard and Thompson, 1997;Ramstein and Joussaume, 1995;Rind, 1987).
The DJF and JJA temperature anomalies in our 6 ka simulation are also similar to those of the PMIP3 models (Figs.S7  and S8).Relative to PI, GENMOM produces slightly greater winter warming over Scandinavia than is evident in the average of the PMIP3 simulations, and is generally 0.5-1.0• C cooler over Asia, Africa, and South America.During boreal summer, GENMOM simulates warming over the NH landmasses and cooling over the North African and Indian monsoon regions, consistent with the PMIP3 models.Continental warming in GENMOM is ∼ 0.5-1.0• C weaker than most PMIP3 models, particularly in Europe and Asia.A portion of the weaker warming in GENMOM is attributed to the prescribed 6 ka GHG concentrations we derived from the icecore data, which differ slightly from those specified for the PMIP3 experiments (Table 1 caption).

Precipitation and monsoons
The simulated global precipitation anomalies display a progression from the drier and colder conditions of the LGM to the warmer and wetter conditions of the Holocene (Fig. 8, Table 2).The global mean annual precipitation change of −0.29 mm d −1 for the LGM is distributed as greater drying over land and ice sheets (−0.30mm d −1 ) than oceans (−0.22 mm d −1 ).Regionally coherent patterns of precipitation change (Figs. 8 and 9) are indicative of displacement and changes in the strength of storm tracks (Li and Battisti, 2008), the ITCZ, and the Hadley circulation, as well as the onset, amplification, and subsequent weakening of the global monsoons regions (Broccoli et al., 2006;Chiang, 2009;Chiang and Bitz, 2005).
Between the LGM and 15 ka, during DJF, areas over and adjacent to the NH ice sheets display predominately reduced precipitation arising from a combination of the desertification effect of the high and cold ice; lower-than-present atmospheric moisture; and cloudiness and the advection of cold, dry air off of the ice sheets (Figs. 3a,4a,6a,and 8a).The topographic and thermal effects of the LIS and the thermal effect of sea ice (Kageyama et al., 1999;Li and Battisti, 2008) alter 500 hPa geopotential heights along the southern margin of the ice sheet (Figs.3a and S2a), causing the development of positive precipitation anomalies extending from the eastern Pacific across the Gulf of Mexico and eastern North America and into the North Atlantic.Accompanying negative precipitation anomalies over the North Atlantic and positive anomalies over the Nordic Seas are related to changes in the location of storm tracks.The local effect of the ice sheets on precipitation diminishes during the early and mid-Holocene as their influence on circulation weakens and the atmosphere becomes warmer and moister (Fig. 9a).
The negative DJF anomalies that persist from 21 to 15 ka during austral summer along the equatorial and low-latitude areas of South and Central America, south-central Africa, Southeast Asia, northern Australia, the tropical Atlantic, the Indian Ocean, and the western Pacific warm pool are caused by changes in the location of the ITCZ and weakened southern monsoonal circulation.This particularly affects the winter monsoon in central South America (Cheng et al., 2012;Zhao and Harrison, 2012) and in Southeast Asia and Indonesia, where additional feedbacks in the energy and water balances over emergent land areas occur during low sea level stands (Figs. 1 and 8a) have been shown to alter the Walker Circulation (DiNezio and Tierney, 2013).
Precipitation for JJA also exhibits considerable change over time (Figs.8b and 9b).Similar to DJF, generally drier conditions are simulated over and adjacent to the NH ice sheets where anticyclonic flow tendencies suppress precipitation (Fig. 4b).Along portions of the southern margins of the LIS and FIS, however, orographic lifting enhances precipitation at 21 ka (Pollard and Thompson, 1997).Wetter conditions in the North American Southwest derive from enhanced westerly flow aloft and lower level southwesterly flow off the eastern Pacific that are associated with displacement of the jetstream by the ice sheets and the weakened Pacific subtropical high.Between 21 and 12 ka the LIS causes an increased pressure gradient from a strengthened Azores-Bermuda High and weakened subtropical high in the eastern Pacific (Figs. 3b and 4b), resulting in amplified and displaced westward winds, drying over Central America, and wetterthan-present conditions over northern South America.At the LGM, North Africa, Europe, and all but the western edge of Asia are drier than the PI, again reflecting the drier atmosphere of the full glacial.The magnitude, gradients, and spatial patterns of GEN-MOM 21 ka DJF precipitation anomalies are consistent with the PMIP3 experiments.Notable exceptions are greater drying than some models in the North Atlantic and the band of positive anomalies extending across the Gulf of Mexico and the southeastern USA.GENMOM produces positive precipitation anomalies over Australia, which is present in four of the PMIP3 models.The 21 ka JJA precipitation anomalies are also in agreement with PMIP3, but display weaker drying over eastern NA and slight drying over the North African monsoon region.
The time evolution from LGM to PI of the African and Indian monsoons reflects the interplay of changes in the location of the ITCZ and Hadley circulation that are linked to the receding NH ice sheets, GHG-driven global warming, enhanced NH JJA insolation, and changing land-SST temperature contrast.The North African and Indian monsoons are suppressed between 21 and 18 ka.After 18 ka, wetter-than-present conditions emerge in the monsoon regions of North Africa and India, where increased JJA insolation warms the continents, which amplifies the landsea temperature contrasts that drive monsoonal circulation (Braconnot et al., 2007b;Kutzbach and Otto-Bliesner, 1982;Zhao and Harrison, 2012).The simulated DJF air temperatures in North Africa cool from the LGM until 15 ka, and then warm monotonically through the rest of the deglaciation and Holocene (Fig. 10).Wintertime precipitation over the North African region is minimal.In contrast, JJA temperatures increase throughout the deglaciation, peak at 9 ka, decrease slightly at 6 ka, and increase thereafter.A commensurate increase in JJA precipitation over North Africa between 12 and 6 ka is associated with northward migration of the ITCZ (Braconnot et al., 2007a, b;Kutzbach and Liu, 1997), which enhances the transport of moisture into both the North African and Indian monsoon regions.Monsoonal precipitation peaks over both regions between 12 and 9 ka (Fig. 10).The change in precipitation between 9 and 6 ka over India (0.9 mm d −1 ) is nearly double the change over North Africa (0.5 mm d −1 ), consistent with the diagnoses of the mid-Holocene monsoon of Marzin and Braconnot ( 2009), who attribute the stronger ∼ 9 ka monsoon to insolation related to precession and snow cover on the Tibetan Plateau.The pattern of precipitation in the Indian monsoon region is similar to that of North Africa but exhibits a greater range between peak Holocene values and the PI.
The overall temporal progression and magnitude of precipitation changes in the time slice simulations are in agreement with the PMIP2 (Braconnot et al., 2007a, b) and PMIP3 simulations at 21 and 6 ka, and with other mid-Holocene modeling studies (Hély et al., 2009;Kutzbach and Liu, 1997;Kutzbach and Otto-Bliesner, 1982;Timm et al., 2010).More specifically, the June through September GENMOM precipitation anomaly of −0.6 mm d −1 over the North African monsoon region during the LGM is within the range (−0.9 to 0.1 mm d −1 ) of five PMIP2 AOGCMs (Braconnot et  The northward expansion and spatial pattern of precipitation anomalies of the 6 ka monsoons are in very good agreement with both the PMIP2 and PMIP3 experiments.Summer precipitation in the GENMOM simulation is enhanced by 0.9 mm d −1 relative to PI over North Africa, in agreement with the range (0.2-1.4 mm d −1 ) and mean (0.7 mm d −1 ) of 11 PMIP2 AOGCMs (Zhao and Harrison, 2012) and 12 PMIP3 models (range of 0.1-1.0 and average of 0.6 mm d −1 ).Over India, the 6 ka GENMOM precipitation anomaly of 1.1 mm d −1 exceeds the range (0.2-0.9 mmd −1 ) and mean (0.6 mm d −1 ) of the 11 PMIP2 models (Zhao and Harrison, 2012), but is within the range of the PMIP3 models (0.5-1.3 and average of 1.0 mm d −1 ).

Sea ice
DJF sea ice is present in the PI simulation over Hudson Bay and the Arctic Ocean; along the coast of eastern Canada; and around Greenland, the Nordic Seas, and the Baltic and North seas (Fig. 11), in agreement with observed present-day distributions (Jaccard et al., 2005).Ice fractions of up to 100 % are simulated over the Bering Sea and the Sea of Okhotsk.In the SH, sea ice persists through austral summer in the Weddell and Ross seas and a few scattered locations around Antarctica.While the locations of the ice around Antarctica are in agreement with observations (Gersonde et al., 2005), the model underestimates the ice extent over the Weddell Sea and between the Weddell and Ross seas.The lack of ice is partly attributable to a warm bias in the Southern Ocean associated with the previously mentioned weak ACC (discussed further below).During August and September, simulated sea ice is greatly reduced in the North Atlantic region (Fig. 11), with remnant ice persisting in the extreme north of Baffin Bay, and the east coast of Greenland, also in agreement with observations.In the SH, the corresponding winter sea ice grows substantially and the distribution is in generally good agreement with observations (Gersonde et al., 2005).The simulated annual average ice extents for the NH are 9.8 × 10 6 km 2 for the LGM, 15.8 × 10 6 km 2 for 6 ka, and 14.1 × 10 6 km 2 for PI (grid cells with fractional coverage > 15 %).Compounded with climate forcing, changes in both the distribution and areal coverage of the NH ice also reflect the change in ocean area due to the transition of land and ice sheets to ocean as sea level rises (Figs.11 and S13-S15).For the same time periods, the SH ice area extents, which are minimally affected by land-sea transitions with sea level rise, are 20.9 × 10 6 , 11.4 × 10 6 , and 11.1 × 10 6 km 2 , respectively.
During the 21 ka boreal winter, the Arctic Ocean and Baffin Bay are fully covered by ice and the ice around Greenland expands.The model displays increased sea ice in the western North Atlantic and decreased ice in the eastern North Atlantic and Nordic Seas, where the prescribed FIS margin advances into the water (Fig. 2).The limit of substantial coverage north of 55 • N is in agreement with reconstructions (de Vernal et al., 2006) and other LGM simulations (Otto-Bliesner et al., 2006a;Roche et al., 2007); however, slight fractional cover (pack ice) in the model likely extends too far south (to ∼ 45 • N) along the coast of North America.Fractional cover of up to 100 % is simulated in the far northwestern Pacific and the Sea of Okhotsk with a sharp, southward transition to reduced coverage.In boreal summer of the LGM, simulated sea ice retreats to 65 • N in the North Atlantic and persists along eastern Canada, Baffin Bay and south of Greenland and the extreme northern areas of the Nordic Seas.

Antarctic Circumpolar Current and Atlantic Meridional Overturning Circulation
The simulated ACC of 62 Sv is ∼ 48 % weaker than the observed value of 119 Sv through the Drake Passage (GECCO data; Köhl and Stammer, 2008).Although the T31 resolution of GENMOM is a factor in limiting flow through the Drake Passage, we attribute the underestimate of the ACC in part to insufficient wind stress at the latitude of the Drake Passage, which is caused by equatorward displacement of the core of the westerly winds, a shortcoming in common with other low-resolution AOGCMs (Alder et al., 2011;Russell et al., 2006;Schmittner et al., 2011a).Considerable uncertainty exists in the proxies that are used to infer past changes in AMOC strength (Delworth and Zeng, 2008;Lynch-Stieglitz et al., 2007).The 231 Pa/ 230 Th record from 33 • N on the Bermuda Rise (Lippold et al., 2009;Mc-Manus et al., 2004) indicates that, after the LGM, the strength of the AMOC began to diminish at ∼ 18 ka, was further reduced during Heinrich Event 1 (H1) at ∼ 17 ka, increased abruptly during the BA at 15 ka, and weakened again during the YD cold reversal at ∼ 12 ka.After the YD, the AMOC strengthened again and stabilized.In climate models, a variety of factors including the North Atlantic freshwater budget, model resolution, and parameterizations and the characteristics of simulated Antarctic Bottom Water (AABW) give rise to a considerable simulated range of AMOC (Weber et al., 2007).
The AMOC in our PI simulation (Fig. 12) is 19.3± 1.4 Sv at the core site of 33 • N, a value similar to the present-day estimate of 18.7± 4.8 Sv at 26.5 • N (Srokosz et al., 2012).The maximum AMOC simulated by GENMOM in the PI is 21.3 Sv at 41 • N, a value outside the range of 13.8-20.8Sv of five models in the PMIP2 experiments (Weber et al., 2007) but within the range of 3.8-31.7Sv of the IPCC AR4 models (Meehl et al., 2007;Schmittner et al., 2005).The newer CMIP5 models have a narrower range of AMOC of ∼ 14 to ∼ 30 Sv when sampled at 30 • N (Cheng et al., 2013); GEN-MOM simulates 16.0± 1.3 Sv at this location.
Our simulated LGM AMOC at the core site is 16.4 Sv, which is a ∼ 14.7 % reduction relative to the PI.The maximum LGM AMOC is 22.4 Sv at 40.8 • N, an increase of 1.1 Sv (5.1 %) relative to the PI maximum and within the considerable range of −6.2 to +7.3 Sv in five PMIP2 simulations (Weber et al., 2007).In the deglacial simulations (21 through 15 ka), the northward (positive) AMOC flow extends deeper than that of the PI (Fig. 12) and the southward flow or AABW consequently is somewhat weakened.The maximum AMOC in GENMOM is essentially constant at the 40.8 • N depth of 1.23 km for all time slices.Although the depth of the maximum is again comparable to the range of the PMIP2 models (1.24± 0.20), the invariance of the location and depth in GENMOM is likely a model-specific response.
Our time slice simulations display an increase in the strength of AMOC from the LGM to a maximum at 15 ka, decrease to a minimum at 9 ka, and remain more or less constant through the PI (Fig. 13), which is in apparent disagreement with the 231 Pa/ 230 Th records from which greater variability is inferred (Lippold et al., 2009;McManus et al., 2004).We do not expect to capture rapid and abrupt climate change events such as H1 (∼ 17 ka), the BA (∼ 15 ka), and the YD (∼ 12 ka) with only eight time slices, because we did not manipulate freshwater discharge to the North Atlantic in our experimental design.

21 and 6 ka data-model comparisons
We compare temperature and precipitation from our LGM and mid-Holocene simulations with paleoclimatic reconstructions and the PMIP3 simulations.For the LGM, we use the pollen-based reconstructions of mean annual mean temperature (MAT) and precipitation (MAP) from Bartlein et al. (2011) over land, as well as the Multiproxy Approach for the Reconstruction of the Glacial Ocean Surface (MARGO) project reconstructions over oceans (Waelbroeck et al., 2009).The gridded 2 • × 2 • pollen data include > 3000 terrestrial pollen records from Eurasia, Africa, and North America, and the global MARGO reconstruction comprises ∼ 700 analyses of planktonic foraminifera, diatom, dinoflagellate cyst and radiolarian abundances, alkenones, and planktonic foraminifera Mg / Ca from marine core sites.For 6 ka, we combine the pollen-based reconstructions of Bartlein et al. (2011) and the GHOST SST reconstructions (Leduc et al., 2010).The 6 ka GHOST data set contains ∼ 100 reconstructed temperature records based on analyses of alkenones and foraminifera Mg / Ca from marine sites located along continental margins and the Mediterranean Sea.

21 ka
Our simulated 21 ka anomalies of MAT and MAP are comparable with the pollen reconstructions (Fig. 14) and fall within the range of the PMIP3 models.GENMOM captures the mixed pattern of temperature and precipitation anomalies over Beringia that are present in the reconstructions (Fig. 14a, b) and in several of the PMIP3 simulations (Figs.S8, S9, and S16).The GENMOM SST anomalies indicate broad cooling of the global oceans (mean of −1.7 • C) but not as much cooling as is simulated in the PMIP3 models (mean of −2.9 • C); however, Harrison et al. (2013) found that the PMIP3 models tended to overestimate oceanic cooling.When sampled at the MARGO locations, GENMOM is generally warmer, but within the range of the PMIP3 models (Harrison et al., 2013).The overall agreement of the simulation with the MARGO data is good, but some features in the MARGO data are not reproduced by GENMOM.For example, similar to the PMIP3 simulations (Figs.S5, S6, and S16), the GENMOM simulation lacks the warming over the Greenland and Nordic Seas inferred from the data; however,  while the data indicate the Nordic Seas were ice-free at the LGM, the magnitude of the warming elsewhere, if it occurred, is somewhat unclear (de Vernal et al., 2006;Moller et al., 2013).The limited cooling along the western coast of North America and Mediterranean in GENMOM is attributed to the inability of the model to resolve the California Current and the Mediterranean circulation (Alder et al., 2011).The simulated LGM MAP anomalies are also comparable with the pollen-based reconstructions (Fig. 14c and d).The model simulates general drying of the NH and a mix of increased and decreased precipitation in Beringia, South America, southern Africa, Southeast Asia, and Australia.GEN-MOM produces strong drying over and around the NH ice sheets, wetter-than-present conditions in the southwestern United States, and drying in Central America.The simulation fails to reproduce the drying over eastern North America that is inferred from the pollen-based data.There is considerable variability in the PMIP3 simulations of MAP (Figs.S9  and S10).In common with the PMIP3 models, GENMOM simulates a general reduction of precipitation over the NH, the North African and Indian monsoon regions, and Southeast Asia, as well as increased precipitation south of the LIS, southern Africa, and much of Australia (Fig. S16).

6 ka
Relative to PI, the changes in 6 ka boundary conditions are predominantly in the seasonality of insolation (Table 1) as opposed to the stronger radiative forcing associated with changes in GHGs and continental ice sheets from the LGM through the early Holocene.The resulting changes in 6 ka climatology are thus more subtle than those of the deglaciation.The changes of 6 ka MAT simulated by GENMOM are generally within the range of ± 1 • C (Fig. 15b).Enhanced MAP and associated cooling are evident in the NH monsoonal regions (Fig. 15d).Elsewhere, MAP changes are within a range of ± 50 mm.
Pollen-based data reconstructions indicate highly heterogeneous changes in MAT during 6 ka; however, there are regions with spatially consistent changes in sign, such as warming south of Hudson Bay, areas of warming over Scandinavia and western Europe, and cooling in the Mediterranean region (Fig. 15a).Larger MAT changes at highelevation sites and regions with anomalies of mixed sign occur in the data over most continents.The GENMOM 6 ka MAT anomalies also display a mix of warming and cooling in a range of about ± 4 • C; however, where pollen-based records exist, the majority of the anomalies are within a narrower range of about ± 1.5 • C (Fig. 15b).GENMOM, and many of the PMIP3 models (Figs.S8, S9, and S16), produces a mixture of warm and cold 6 ka MAT anomalies that are generally in the range of ± 1 • C over the North Atlantic, Europe, and Scandinavia, which underestimates the proxybased anomalies by > 2 • C at some sites.
The Asian pollen-based reconstruction similarly displays a heterogeneous temperature pattern that is reproduced by GENMOM and the PMIP3 models.In all of the models, the sign of the anomalies does not vary abruptly in close proximity to the pollen sites.We note, however, that the smooth topography in GCMs limits the ability of the models to reproduce large and regionally spatially heterogeneous anomalies that are characteristic of the local climate at many highelevation pollen sites in western North America, the Alps, the central plateau of Africa, and Asia.
GENMOM displays cooling in the North African and Indian monsoon regions and warming over the high northern latitudes, consistent with the PMIP3 models (Fig. 15).In contrast, GENMOM simulates weak global cooling of 0.39 • C compared to no change in the PMIP3 model average, which is partially attributed to our lower prescribed GHG concentrations (Table 1

caption).
Precipitation anomalies inferred from the pollen-based data indicate that 6 ka was wetter than the PI in Europe, Africa, Asia, and some parts of western North America and drier than PI in much of eastern North America and Scandinavia (Fig. 15c).GENMOM simulates the gradients and coherent patterns of positive and negative MAP anomalies over North America and northern, central, and western Africa, in agreement with the data and the PMIP3 models.The data and GENMOM are also in agreement over the Asian monsoon region and northwestern Asia, where wetter conditions prevail, but anomalies of opposite sign are simulated over the Great Lowland Plain in north-central Eurasia and Southeast Asia.Bartlein et al. (2011) attribute cooling in Southeast Asia to a stronger winter monsoon at 6 ka.Our results (Figs. 6a and 8a), and many of the PMIP3 models, indicate cooler, drier winters (Figs.S7 and S11) and regionally variable changes in the summer (Figs.S8 and S12).
In Africa, the model captures the increase in precipitation in the northern and continental regions and drying along the southern coastal regions, as evident in the data.Strengthening of the African and Indian summer monsoons during the mid-Holocene corresponds well with the PMIP2 and PMIP3 models (Zheng and Braconnot, 2013).Both GENMOM and the data indicate drying over central Scandinavia, wetter conditions over east-central Europe, the Iberian Peninsula, and around the Mediterranean, but, over western Europe, the simulated decrease in MAP in GENMOM clearly disagrees with the data and some of the PMIP3 models (Figs. 15, S7, S8 and S16); however, the magnitude of the change in the models is very small and the sign of the change varies among models.Wetter conditions also prevail in Indonesia, and a southwest-to-northeast wet-dry gradient is simulated over Australia.

Summary
We have presented a suite of multi-century equilibrium climate simulations with GENMOM for the past 21 000 years at 3000-year intervals.Each 1100-year simulation was forced with fixed, time-appropriate global boundary conditions that included insolation, GHGs, continental ice sheets, and adjustment for sea level.The key drivers of climate change from the LGM through the Holocene are retreat of the NH ice sheets, increasing GHG concentrations, and latitudinal and seasonal variations in insolation.
GENMOM reasonably well reproduces the LGM to Holocene temperature trends inferred from the paleoclimate data syntheses of Shakun et al. (2012) and Marcott et al. (2013).The evolution of global temperature change simulated by GENMOM is consistent with three transient simulations, but is generally cooler during the deglacial time slices than the transient simulations when sampled at the proxy locations.The global LGM cooling of 3.8 • C simulated by GENMOM is within the range of 2.6-5.0 • C and average of 4.4 • C simulated by the PMIP3 models.Simulated LGM cooling of the tropical oceans is 1.6 • C, which is in good agreement with the MARGO reconstruction of 1.7± 1 • C. The weaker LGM global cooling is attributed to the sensitivity of GENMOM to CO 2 (2.2 • C for a 2× increase in the present-day value).
During the LGM, simulated precipitation is reduced globally by 8.2 % and gradually increases through the Holocene to present-day values in response to loss of the NH ice sheets, global warming, and related increases in atmospheric humidity.Between 15 and 6 ka seasonal changes in insolation altered the NH land-sea temperature contrasts, which, combined with shifts in global circulation, strengthened the summer monsoons in Africa and India.Monsoonal precipitation in both regions peaked between 12 and 9 ka, consistent with pollen-based reconstructions.The spatial patterns of mid-Holocene precipitation change simulated by GENMOM correspond well with the PMIP3 models, as do the 6 ka changes in monsoonal precipitation.In contrast to the pollen-based reconstructions, the GENMOM simulation shows western Europe to be slightly drier instead of slightly wetter than present.
The eight time slice simulations depict the glacialinterglacial transition that is in good agreement with other AOGCM simulations and compares reasonably well with data-based climate reconstructions.The data-model and model-model comparisons give us a measure of confidence that our paleo-GENMOM simulations are reasonable on broad spatial scales and add to the growing number of climate models that are capable of simulating key aspects of past climate change when constrained by a relatively small set of global boundary conditions.While our simulations are not continuous across the deglaciation and do not include events such as freshwater forcing, they do provide insights into between-period changes, such as altered NH storm tracks and strengthening of monsoons during the early to mid-Holocene and multi-century time series, that are useful, for example, to explore ecosystem responses to changes in mean climate and the related interannual variability in the model.Future work using the model output produced by this study will address how internal model variability and multidecadal variability influence comparison with proxy data, particularly North America, when using dynamical downscaling techniques.The Supplement related to this article is available online at doi:10.5194/cp-11-449-2015-supplement.

Figure 1 .
Figure 1.Boundary conditions for the time slice simulations.CO 2 concentrations are relative to the PI concentration of 280 ppmV.NH ice area is the total area covered by the continental ice sheets.June insolation anomalies are relative to PI at the indicated latitude.Midmonth insolation data from Berger and Loutre (1991).

Figure 3 .
Figure 3. Simulated seasonal 500 hPa geopotential height and wind anomalies relative to PI.(a) December, January, and February and (b) June, July, and August.Raw 500 hPa geopotential height and wind are shown in Fig. S2 in the Supplement.

Figure 4 .
Figure 4. Simulated seasonal average sea level pressure and 2 m wind anomalies relative to PI.(a) December, January, and February and (b) June, July, and August.Raw sea level pressure and wind are shown in Fig. S3.

Figure 5 .
Figure 5. Simulated and reconstructed changes in temperature from 21 ka to present.(a) Global mean surface air temperature from GEN-MOM compared to the PMIP3 ensemble average and three transient models (CCSM -Liu et al., 2009; LOVECLIM -Timm and Timmermann, 2007; and FAMOUS -Smith and Gregory, 2012).The transient model values are averages over a ±50-year window centered on the eight time slices.The symbols for the PMIP and transient models are the average of the ensembles and the bars represent the range of the ensembles.Data-model estimates of mean and range of LGM cooling by Annan and Hargreaves (2013) and Schmittner et al. (2011b) are offset from 21 ka for legibility.(b) Temperature change at the proxy sites used in the reconstructions by Shakun et al. (2012) and Marcott et al. (2013).The models were bilinearly interpolated and aggregated to the 5 • × 5 • boxes around the proxy sites as in Marcott et al. (2013).The 1σ uncertainty in the reconstructions is indicated by the shaded band.Marcott et al. (2013) is adjusted to a preindustrial (∼ 1850) base value rather than the original 1961-1990.Data younger than preindustrial are removed.The Shakun et al. (2012) and Marcott et al. (2013) time series are joined at their 11.5-6.5 ka means.

Figure 6 .
Figure 6.Simulated seasonal average 2 m air temperature anomalies relative to PI.(a) December, January, and February and (b) June, July, and August.

Figure 7 .
Figure 7. Simulated seasonal average changes in 2 m air between consecutive time slices.(a) December, January, and February and (b) June, July, and August.

Figure 8 .
Figure 8. Simulated seasonal average precipitation anomalies relative to PI.(a) December, January, and February and (b) June, July, and August.

Figure 9 .
Figure 9. Simulated seasonal average precipitation changes between consecutive time slices.(a) December, January, and February and (b) June, July, and August.

Figure 11 .
Figure 11.Simulated sea ice fraction for PI and 6 and 21 ka.Left two columns: February-March.Right two columns: August-September.Medium gray is continental land mass and dark gray is continental ice sheet.

Figure 12 .
Figure 12.Simulated annual average Atlantic Meridional Overturning Circulation (AMOC) for the eight time slices.

Figure 13 .
Figure 13.Simulated Atlantic Meridional Overturning Circulation (AMOC) compared to 231 Pa/ 230 Th proxy record at 33 • N and other AOGCMs.Observations are from 26.5 • N. GENMOM values are 100-year averages, with error bars representing standard deviations.The PMIP2 values represent the mean and standard deviation of the maximum AMOC from five models.The IPCC AR4 point represents the mean and standard deviation from a collection of IPCC AR4 models. 231Pa/ 230 Th data from McManus et al. (2004) and Lippold et al. (2009), observed value from Srokosz et al. (2012), PMIP2 data from Weber et al. (2007), and IPCC data from Schmittner et al. (2005).
Over the tropical ocean basins, the 21 ka GENMOM simulation is 1.6 • C colder than the PI, in good agreement with the inferred MARGO cooling of 1.7± 1 • C (Otto-Bliesner et al., 2009).Average simulated SST anomalies are also similar to MARGO over the Indian (−1.6 • C versus −1.4 ± 0.7 • C) and Pacific (−1.5 • C versus −1.2 ± 1.1 • C) oceans but are warmer than the data in the tropical Atlantic basin (−1.9 • C versus −2.9±1.3 • C).In each of these regions, the anomalies simulated by GENMOM fall within the range of six PMIP2 models analyzed by Otto-Bliesner et al. (2009).GENMOM captures the 2-4 • C cooling in the eastern coastal Atlantic evident in the MARGO data, and the SST anomalies are ∼ 2-4 • C colder over the Western Pacific Warm Pool.Neither GENMOM nor the PMIP3 models reproduce the warming over the central and eastern tropics, low latitudes, and the North Atlantic evident in the MARGO reconstruction.

Figure 14 .
Figure 14.Changes in 21 ka mean annual temperature (MAT) and precipitation (MAP) inferred from data and simulated by GENMOM.(a) Blended sea surface temperature from MARGO (Waelbroeck et al., 2009) and terrestrial temperature from Bartlein et al. (2011), (b) GENMOM temperature anomalies (blended sea surface temperature and 2 m air temperature over land), (c) precipitation from Bartlein et al. (2011), and (d) GENMOM precipitation anomalies.Grid cells with different land mask types in the 21 ka and PI simulation are shaded in gray to avoid comparing ocean temperature to land temperature in emergent cells.

Figure 15 .
Figure 15.Changes in 6 ka mean annual temperature (MAT) and precipitation (MAP) inferred from data and simulated by GENMOM.(a) Blended sea surface temperature from Leduc et al. (2010) and terrestrial temperature from Bartlein et al. (2011), (b) GENMOM temperature anomalies (blended sea surface temperature and 2 m air temperature over land), (c) precipitation from Bartlein et al. (2011), and (d) GENMOM precipitation anomalies.Grid cells with different land mask types in the 6 ka and PI simulation are shaded in gray to avoid comparing ocean temperature to land temperature in emergent cells.

Table 1 .
Atmospheric greenhouse gas concentrations for each time slice simulation.The 21 through 3 ka values for CO 2 • C warmer than the PI.