Late-glacial to late-Holocene shifts in global precipitation δ 18O

. Reconstructions of Quaternary climate are often based on the isotopic content of paleo-precipitation pre-served in proxy records. While many paleo-precipitation isotope records are available, few studies have synthesized these dispersed records to explore spatial patterns of late-glacial precipitation δ


Introduction
Isotopic compositions of late-glacial precipitation can be preserved in groundwaters, cave calcite, glacial ice, ground ice and lake sediments.These records have been used to better understand past climate changes for more than a half century (e.g., Münnich, 1957;Thatcher et al., 1961;Münnich et Published by Copernicus Publications on behalf of the European Geosciences Union. al., 1967;Pearson and White, 1967;Tamers, 1967;Gat et al., 1969).Each type of isotopic proxy record is distinguished by its temporal resolution, preservation of one or both 18 O / 16 O and 2 H / 1 H ratios, and frequency on land surface.For example, groundwater records contain both 18 O / 16 O and 2 H / 1 H ratios with widespread global occurrence, but have a coarser temporal resolution than other paleoclimate proxies (Rozanski, 1985;Edmunds and Milne, 2001;Edmunds, 2009;Corcho Alvarado et al., 2011;Jiráková et al., 2011).Speleothem records, by contrast, have high temporal resolution but usually only report calcite 18 O / 16 O ratios (without fluid inclusion 2 H / 1 H data) and are less common than groundwater records (e.g., Harmon et al., 1978Harmon et al., , 1979)).Late-glacial ice core and ground ice records have high temporal resolution, can be analysed for 18 O / 16 O and 2 H / 1 H ratios, but are rare on non-polar lands (Dansgaard et al., 1982;Thompson et al., 1989Thompson et al., , 1995Thompson et al., , 1997Thompson et al., , 1998)).Lake sediment records can have a high temporal resolution, can preserve 18 O / 16 O and 2 H / 1 H ratios and are available for a multitude of globally distributed locations (e.g., Edwards and McAndrews, 1989;Eawag et al., 1992;Menking et al., 1997;Wolfe et al., 2000;Anderson et al., 2001;Beuning et al., 2002;Sachse et al., 2004;Morley et al., 2005;Tierney et al., 2008).However, some lake water proxy isotope records may be impacted by paleo-lake evaporative isotope effects that obscure the primary meteoric water signal and mask paleo-precipitation isotope compositions (e.g., lake sediment calcite, diatom silica; Leng and Marshall, 2004).
This study examines speleothem, ice core and groundwater isotope records, focusing primarily on the groundwater isotope records due to their relative density in the published literature in comparison to the more limited number of published speleothem and ice core records (compilations by Pedro et al., 2011;Stenni et al., 2011;Clark et al., 2012;Shah et al., 2013;Caley et al., 2014a).There exist roughly twice as many groundwater reconstructions of late-glacial to late-Holocene precipitation δ 18 O shifts (n = 59) as the combined total of speleothem and ice core records (n = 27; where δ 18 O = ( 18 O / 16 O sample ) / ( 18 O / 16 O standard mean ocean water -1) × 1000).A recent global synthesis of paired precipitationgroundwater isotopic data demonstrated that modern annual precipitation and modern groundwater isotope compositions follow systematic relationships with some bias toward winter and wet-season precipitation (Jasechko et al., 2014).Systematic rainfall-recharge relationships shown by Jasechko et al. (2014) support our primary assumption in this study that groundwater isotope compositions closely reflect meteoric water.Because groundwater records can only identify climate change occurring over thousands of years due to hydrodynamic dispersion during multi-millennial residence times (e.g., Davison and Airey, 1982;Stute and Deak, 1989), we limit the focus of this study to meteoric water isotope composition changes from the latter half of the last glacial time period to the late-Holocene.The latter half of the last glacial period is defined as ∼ 20 000 to ∼ 50 000 years before present, using the end of the last glacial maximum as the more recent age limit (∼ 20 000 years before present; Clark et al., 2009) and the maximum age of groundwater that can be identified by 14 C dating as an approximate upper age limit (i.e., groundwater ages more recent than ∼ 50 000 years old).
For brevity, we refer herein to the time period representing the latter half of the last glacial period (∼ 20 000 to ∼ 50 000 years before present) as the late-glacial (e.g., δ 18 O late-glacial ).We adopt a definition of the late-Holocene as occurring within the last 5000 years following Thompson et al. (2006).Other work proposes the late-Holocene be defined as within the last 4200 years (Walker et al., 2012), which is consistent with the 5000 years before present definition (Thompson et al., 2006) within the practical uncertainty of 14 C-based groundwater ages (± ∼ 10 3 years).Further, although precipitation isotope compositions have varied over the late-Holocene, groundwater mixing integrates this variability, prohibiting paleoclimate interpretation at finer temporal resolutions.
Late-glacial to late-Holocene changes in precipitation isotope compositions provide important insights into conditions and processes of the past.Perhaps the two best-constrained global-in-scale differences between the late-glacial and the late-Holocene are changes to oceanic and atmospheric temperatures (MARGO Members, 2009;Shakun and Carlson, 2010;Annan and Hargreaves, 2013), and changes to seawater δ 18 O (Emiliani, 1955;Dansgaard and Tauber, 1969;Schrag et al., 1996Schrag et al., , 2002)).Atmospheric temperatures have increased by a global average of ∼ 4 • C since the last glacial maximum, with greatest warming at the poles and more modest warming at lower latitudes (Fig. 1; Shakun and Carlson, 2010;Annan and Hargreaves, 2013).Seawater δ 18 O during the last glacial maximum was 1.0 ± 0.1 ‰ higher than the modern ocean, as constrained by paleo-ocean water samples collected from pore waters trapped within sea floor sediments (Schrag et al., 2002).
The objective of this study is to analyse spatial patterns of measured late-glacial to late-Holocene precipitation δ 18 O changes from published groundwater, ground ice, glacial ice and cave calcite records, and to compare these measurements with output from five state-of-the-art isotope-enabled general circulation model simulations of last glacial maximum and pre-industrial or modern climate conditions.Synthesizing paleowater δ 18 O records provides an important constraint for isotope-enabled general circulation model simulations of atmospheric and hydrologic conditions during glacial climate states (Jouzel et al., 2000).We combine a new global compilation of late-glacial groundwater and ground ice isotope data (n = 59) with existing compilations for speleothems (n = 15; Shah et al., 2013) and ice cores (n = 12; Pedro et al., 2011;Stenni et al., 2011;Clark et al., 2012;Caley et al., 2014a).This compilation of late-glacial groundwater isotope compositions builds from earlier reviews of European and African paleowater isotope compositions (Rozanski, 1985;Edmunds and Milne, 2001;Darling, 2004;Edmunds, 2009;Négrel and Petelet-Giraud, 2011;Jiráková et al., 2011).

Data set and methods
In order to examine spatial patterns of change to meteoric water δ 18 O values we compiled δ 18 O, δ 2 H, δ 13 C and 14 C data from 1713 groundwater samples collected from 59 aquifer systems reported in 76 publications (data and primary references presented in the Supplement).δ 13 C, 3 H and 14 C data were used to estimate groundwater age (details within Supplement).Changes to precipitation δ 18 O values over time were determined by comparing groundwater isotope compositions of the late-Holocene (δ 18 O late-Holocene defined here as less than 5000 years before present; Thompson et al., 2006) and the latter half of the last glacial time period (δ 18 O late-glacial : 20 000 to ∼ 50 000 years before present).We acknowledge that these two relatively long time intervalsnecessarily long in order to examine groundwater isotope records -integrate precipitation δ 18 O variability over the course of each time interval.The late-Holocene time interval integrates known precipitation δ 18 O variability (e.g., Aichner et al., 2015), and the late-glacial time interval likely incorporates groundwater preceding the last glacial maximum, potentially during Marine Isotope Stage 3 or even older glacial time periods due to large uncertainties in 14 C-based groundwater ages (Supplement).
Proxy-based meteoric water δ 18 O changes from the latter half of the last glacial time period to the late-Holocene are described herein as measured 18 O late-glacial , where measured 18 O late-glacial = δ 18 O late-glacial − δ 18 O late-Holocene .A minimum groundwater age of 20 000 years before present was used to define the late-glacial to remain consistent with the timing of the last glacial maximum (∼ 20 000 years before present; Clark et al., 2009).Samples having a deuterium excess of less than zero (deuterium excess = δ 2 H − 8 × δ 18 O; Dansgaard, 1964) and falling along regionally characteristic evaporation δ 2 H/δ 18 O slopes (Gibson et www.clim-past.net/11/1375/2015/Clim. Past, 11, 1375-1393, 2015al., 2008) were removed from the analysis to avoid including groundwater samples impacted by partial evaporation.Further, studies reporting saltwater intrusion were avoided on the basis of groundwater δ 18 O and salinities showing evidence of seawater mixing (e.g., Schiavo et al., 2009;Yechieli et al., 2009;Hamouda et al., 2011;Han et al., 2011;Wang and Jiao, 2012;Currell et al., 2013) Speleothem and ice core isotope proxy records were also compiled.Lacustrine sediment δ 18 O records are not considered in this study because these records may preserve meteoric waters impacted by evaporative isotope effects (Leng and Marshall, 2004).Speleothem and ice core measured 18 O late-glacial values were calculated by subtracting average δ 18 O values for each of the two time intervals defined for the groundwater records: the late-Holocene (< 5000 years before present) and latter half of the last glacial time period (20 000 to 50 000 years before present).This step effectively lowered the temporal resolution of speleothem and ice core precipitation isotope records to be consistent with the temporal resolution of the groundwater records.A correction factor was applied to speleothem δ 18 O values to account for different H 2 O-CaCO 3 isotopic fractionation factors during the lateglacial and the late-Holocene because of differing land surface temperatures during each time period (details presented within Supplement).
Simulated 18 O late-glacial values were compiled from five isotope-enabled general circulation models (simulated 18 O late-glacial = δ 18 O last glacial maximum − δ 18 O pre-industrial ): CAM3iso (e.g., Noone and Sturm, 2010;Pausata et al., 2011a), ECHAM5-wiso (e.g., Werner et al., 2011), GISSE2-R (e.g., Schmidt et al., 2014;LeGrande andSchmidt, 2008, 2009), IsoGSM (e.g., Yoshimura et al., 2003) and LMDZ4 (e.g., Risi et al., 2010a).ECHAM5-wiso and IsoGSM outputs are for modern climate rather than pre-industrial conditions; however, the difference between the isotopic composition of pre-industrial and modern climate are expectedly small compared to late-glacial to late-Holocene δ 18 O shifts.An offset factor was applied to simulated mean seawater δ 18 O in all five models (Table S1 in the Supplement) to account for known glacial-interglacial changes to seawater δ 18 O (Emiliani, 1955;Dansgaard and Tauber, 1969;Schrag et al., 1996Schrag et al., , 2002)).Possible spatial differences in seawater δ 18 O changes from the last glacial maximum to the pre-industrial time period are not incorporated into simulations with prescribed sea surface temperatures (CAM3iso, ECHAM5-wiso, IsoGSM, LMDZ4) but are simulated by the coupled ocean-atmosphere simulation of GISSE2-R (Table S1).GISSE2-R was submitted to the CMIP5 archive and participated in PMIP3.LMDZ4 was submitted to the CMIP3 archive.ECHAM5 and CAM3iso did not participate in CMIP5, while IsoGSM uses different boundary conditions than proposed for CMIP5 (Yoshimura et al., 2008).The five models span a range of spatio-temporal resolutions and isotopic/atmospheric parameterizations described in detail in the above references.A selection of the inter-model similarities and differences are summarized in Table S1.
For clarity, empirical 18 O late-glacial values that are based on measured isotope contents of groundwater, speleothem, ground ice or ice core records are referred to herein as measured 18 O late-glacial ; simulated precipitation isotope compositions obtained from general circulation model results are referred to as simulated 18 O late-glacial .We acknowledge that the general circulation models explicitly analyse the last glacial maximum and the pre-industrial climate conditions (i.e., simulated 18 O late-glacial = δ 18 O last glacial maximum − δ 18 O pre-industrial ), whereas proxy record reconstructions of 18 O late-glacial integrate hydroclimatology over multi-millennial timescales that are different from the model simulations.

Measured ∆ 18 O late-glacial values
Measured groundwater (n = 59), speleothem (n = 15) and ice core (n = 12) 18 O late-glacial values are presented in Fig. 2 (references presented in the Supplement).Measured 18 O late-glacial values range from −7.1 ‰ (i.e., δ 18 O late-glacial < δ 18 O late-Holocene ) to +1.7 ‰ (i.e., δ 18 O late-glacial > δ 18 O late-Holocene ).Three-quarters of the compiled records have negative measured 18 O late-glacial values and one-quarter of compiled records have positive measured 18 O late-glacial values.Most groundwater-based late-glacial to late-Holocene shifts fall along δ 2 H/δ 18 O slopes of ∼ 8 (Fig. S58 in the Supplement), suggesting that most groundwaters record temporal shifts to precipitation isotope contents rather than to soil evaporation isotope effects (see Evaristo et al., 2015).More than 80 % of records with positive measured 18 O late-glacial values are located within 35 • of the equator and within 400 km of the nearest coastline (e.g., Bangladesh 18 O late-glacial of +1.5 ‰, less than 300 km from the coast; Figs.2-4).In comparison, negative measured 18 O late-glacial values are found in both coastal regions and farther inland.Negative measured 18 O late-glacial values of the greatest magnitude are located at high latitudes (e.g., northwestern Canada, latitude 64 • N: 18 O late-glacial of −5.5 ‰; northern Russia latitude 72   Bowen and Wilkinson, 2002).This broad spatial pattern is consistent with the non-linear isotopic distillation of air masses undergoing progressive rainout (i.e., Rayleigh distillation).Because seawater δ 18 O values were ∼ 1 ‰ higher-than-modern during the last glacial maximum (Schrag et al., 1996(Schrag et al., , 2002)), our finding that the majority of measured 18 O late-glacial values are negative suggests that isotopic distillation of air masses was greater during the late-glacial than under present climate.This finding is consistent with land surface temperature reconstructions that show larger glacial-to-modern changes to land temperatures at high latitude and continental settings (Fig. 1; Annan and Hargreaves, 2013).Tropical versus extratropical patterns of late-glacial/late-Holocene temperature change (Fig. 1a) are broadly similar to measured 18 O late-glacial values (Fig. 3), where both temperature and isotope shifts are greater at high latitudes relative to the equator.Therefore, it is possible that the larger late-glacial to late-Holocene temperature shifts at the poles relative to the equator may have served to amplify the non-linear, Rayleigh relationship describing the heavy isotope depletion of air masses undergoing progressive rainout during transport from lower 0 Figure 3. Latitudinal variations of 18 O late-glacial values of groundwater (circles, each circle is one aquifer), ice cores (diamonds) and cave calcite (i.e., triangles; where 18 O late-glacial = δ 18 O late-glacial − δ 18 O late-Holocene ).Dashed lines mark 10 • zonal mean simulated 18 O late-glacial values from five different general circulation models: CAM3iso, ECHAM5-wiso, GISSE2-R, IsoGSM and LMDZ4 (Yoshimura et al., 2003;Legrande andSchmidt, 2008, 2009;Risi et al., 2010a;Noone and Sturm, 2010;Pausata et al., 2011a;Werner et al., 2011).to higher latitudes.Further, the late-glacial was characterized by (i) lower-than-modern atmospheric temperatures with larger coastal-inland gradients, and (ii) lower-than-modern eustatic sea level leading to longer overland atmospheric transport distances.Each of these late-glacial/late-Holocene changes favours stronger-than-modern isotopic distillation of air masses transported inland from the coast during the late-glacial (Dansgaard, 1964;Rozanski, 1993;Winnick et al., 2014) (Yechieli et al., 2009), whereas groundwater of the Dead Sea Rift Valley has a 18 O late-glacial value of −1.8 ± 0.6 ‰ (Burg et al., 2013).Speleothem records have 18 O late-glacial values close to +1 ‰ (Frumkin et al., 1999;Bar-Matthews et al., 2003).In northern Turkey, speleothem and groundwater separated by ∼ 150 km have measured 18 O late-glacial values that differ by ∼ 3 ‰ (speleothem 18 O late-glacial −5.7 ± 0.4 ‰ versus groundwater 18 O late-glacial of −2.8 ± 1.0 ‰; Fleitmann et al., 2009;Arslan et al., 2013Arslan et al., , 2015)).While the locations of the groundwater and speleothem records differ, the compiled data suggest that groundwater and speleothem 18 O late-glacial values may capture different 18 O late-glacial values under similar climate conditions.
A number of potential processes could bias the preservation of precipitation isotope composition in ice core, speleothem or groundwater archives (Wang et al., 2001;Thompson et al., 2006;Edmunds, 2009).For example, groundwater and speleothem archives preserve only the isotope record of precipitation that traverses the vadose zone.Recent global analyses of paired precipitation-groundwater isotope compositions show that winter (extratropics) and wet season (tropics) precipitation contributes disproportionately to recharge (Jasechko et al., 2014), meaning that paleoclimate records may be more sensitive to changes to winter and wet seasons than summer or dry season (Vogel et al., 1963;Simpson et al., 1972;Grabczak et al., 1984;Harrington et al., 2002;Jones et al., 2000;Darling, 2004;Partin et al., 2012).Similarly, groundwater isotope records are unlikely to represent constant and continuous recharge fluxes during the late-Holocene or the late-glacial (McIntosh et al., 2012).Modern groundwater recharge fluxes are highest in humid climates (Wada et al., 2010).Groundwater δ 18 O records only represent precipitation that recharges aquifers, meaning that groundwater-based 18 O late-glacial values could be biased to subintervals (e.g., interstadials, pluvial periods) within the late-Holocene and late-glacial intervals when recharge fluxes were at local maxima.Speleothem records may be further complicated by processes impacting the timing of calcite precipitation.Recent modelling suggests that calcite precipitation in caves located outside of the tropics is greatest during the cool season and reduced during summer months due to changes in ventilation, meaning that higher latitude speleothems record oxygen isotope compositions biased to cool season climate change (James et al., 2015).Other recent work suggests that speleothem δ 18 O data may be impacted by disequilibrium isotope effects (Asrat et al., 2008;Daëron et al., 2011;Kluge and Affek, 2012;Kluge et al., 2013) or by partial evaporation of drip waters resulting in 18 Oenrichment (e.g., Cuthbert et al., 2014a) and greater fractionation due to evaporative cooling (Cuthbert et al., 2014b), potentially obscuring the preservation of primary precipitation isotope contents in the speleothem record.Compiled ice core records may have been influenced by post-depositional exchanges of ice with atmospheric vapour (Steen-Larsen et al., 2014).The impact of atmospheric vapour exchanges on ice core isotope records remains poorly understood.Potential biases in the preservation of precipitation δ 18 O differ among groundwater, glacial ice, and speleothem records, meaning that co-located records of differing record-type may preserve different 18 O late-glacial values under similar climate conditions.Finally, all proxy records may be impacted by past changes in the seasonality of precipitation, which can substantially impact annual precipitation δ 18 O values (e.g., Werner et al., 2000).
We cannot rule out the possibility that changes in seasonal biases of proxy record preservation occurred between the late-glacial and the late-Holocene and have impacted measured 18 O late-glacial values.Further, the chronologies of groundwaters and ice core records have uncertainties on the order of thousands of years, meaning that the time intervals used to calculate measured 18 O late-glacial values may be inaccurate.However, the plateauing of isotope content observed in most regional aquifers for 0-5000 years before present and for > 20 000 years before present supports our interpreting these data as records of late-glacial to late-Holocene isotopic shifts (see figures in the Supplement).Notwithstanding potential δ 18 O preservation biases and chronology uncertainties, the global data synthesized here show patterns consistent with the enhanced distillation of advected air masses originating as (sub)tropical ocean evaporate and undergoing progressive, poleward rainout under cooler-than-modern late-glacial temperatures.

Simulated ∆ 18 O late-glacial values
Simulated precipitation 18 O late-glacial values from five general circulation models are presented in Fig. 5.At least four of the five models agree on the sign of simulated 18 O late-glacial values -that is values consistently above or consistently below zero -for 68.8 % of grid cells covering Earth's surface (68.7 % of over-ocean areas and 68.9 % of over land areas; multi-model calculation completed using three of four models as a threshold at high-latitudes where IsoGSM data were unavailable).Simulated 18 O late-glacial values are consistently negative over the North Atlantic Ocean and the Fennoscandian and Laurentide ice sheets and consistently positive over most of the tropical oceans, whereas poorer agreement is found over tropical land surfaces.The negative simulated 18 O late-glacial values over the Northern Hemisphere ice sheets and North Atlantic are likely driven by the difference in ice sheet topography and sea ice cover, between the late-glacial and pre-industrial cli-  (Peltier, 1994) as advised for PMIP2 (Braconnot et al., 2007), whereas the GISSE2 replaces Ice 5G Laurentide ice with that of Licciardi et al. (1999) and ECHAM5-wiso uses ice topography from PMIP3 (Braconnot et al., 2007(Braconnot et al., , 2012; PMIP3 follows ice sheet topography blended from multiple ice sheet reconstructions: Argus and Peltier, 2010;Toscano et al., 2011).Ice sheet topography is an important driver of simulated tem-perature, precipitation and atmospheric circulation during the last glacial maximum (e.g., Justino et al., 2005;Pausata et al., 2011b;Ullman et al., 2014).Therefore, it is likely that inter-model differences in paleo-ice sheet topographies impacts atmospheric circulation and thus high latitude simulated 18 O late-glacial values reported in this study (Fig. 5).
Differences in the specification of initial seawater δ 18 O may also lead to inter-model differences in simulated 18 O late-glacial values.Seawater δ 18 O is set to be globally homogenous in CAM3Iso, IsoGSM and LMDZ4, and heterogeneous in ECHAM5-wiso (using modern gridded seawater δ 18 O heterogeneity of LeGrande and Schmidt, 2006) and GISSE2-R (coupled atmosphere-ocean model; seawater δ 18 O is calculated by the ocean model).Including surface ocean δ 18 O heterogeneities in model simulations impacts land precipitation δ 18 O by up to ∼ 1.5 ‰ relative to simulations with homogenous seawater δ 18 O ( LeGrande and Schmidt, 2006).However, different seawater δ 18 O specifications cannot account for all inter-model differences in simulated 18 O late-glacial values.
The models also show deficiencies in simulating measured 18 O late-glacial values in the tropics, particularly over tropical Africa.This finding could, in part, relate to the high sensitivity of precipitation δ 18 O to convective parameterizations (Lee et al., 2009;Field et al., 2014), although future research is required to test this.Another reason may be that the measured 18 O late-glacial integrates the hydroclimatological signal over multi-millennial timescales, whereas the simulated 18 O late-glacial values explicitly explore last glacial maximum and pre-industrial/present-day climate conditions.The smeared temporal resolution of groundwater-based measured 18 O late-glacial values due to storage and mixing in the aquifer precludes an ideal comparison of measured versus simulated 18 O late-glacial values.Further, as previously discussed in Sect.3.1, the measured 18 O late-glacial values are susceptible to a number of potential biases that may obscure the magnitude and direction of late-glacial to late-Holocene precipitation δ 18 O changes.Notwithstanding, models correctly simulate the sign of measured 18 O late-glacial values (i.e., positive or negative) in the extratropics more frequently than in the tropics.Better agreement in the sign of simulated versus measured 18 O late-glacial values in the extratropics compared to the tropics is likely linked to the substantial changes to extra-tropical ice-sheet topography and seaice cover between the two climate states in northern North America and Europe.Substantial changes to Northern Hemisphere ice volumes between the late-glacial and the late-Holocene likely enhanced upwind distillation of air masses leading to high-magnitude, negative  O last glacial maximumδ 18 O pre-industrial ) from five general circulation models: CAM3iso, ECHAM5-wiso, GISSE2-R, IsoGSM and LMDZ4 (Yoshimura et al., 2003;Legrande andSchmidt, 2008, 2009;Risi et al., 2010a;Noone and Sturm, 2010;Pausata et al., 2011a;Werner et al., 2011).Circles (groundwater), triangles (speleothems) and diamonds (ice cores) show measured 18 O late-glacial values from paleoclimate proxy records (Fig. 1, original data presented in Tables S2-S5).The panel entitled "Composite" shows the multi-model ensemble median simulated 18 O late-glacial value where at least four of the five models agree on the sign of simulated 18 O late-glacial values (i.e., positive or negative; all five model simulations of δ 18 O last glacial maximumδ 18 O pre-industrial were used to calculate multi-model median shown in "Composite").
3.3 Regional measured and simulated ∆ 18 O late-glacial values

Australia and Oceania
Measured 18 O late-glacial values from Australia and Oceania fall between −1 and +1 ‰ (Fig. 2).Australian climate during the last glacial time period was more arid (Nanson et al., 1992), dustier (Chen et al., 1993) and cooler (Miller et al., 1997) than present day.Simulated 18 O late-glacial values across Australia are variable among the five models.Measured 18 O late-glacial values across Oceania have been attributed to temporal changes in the strength of monsoons and convective rains (Aggarwal et al., 2004;Partin et al., 2007;Williams et al., 2010) potentially impacted by late-glacial to late-Holocene shifts in the position of the intertropical convergence zone (Lewis et al., 2010(Lewis et al., , 2011)).
North China Plain groundwaters have high-magnitude, negative 18 O late-glacial values (measured 18 O late-glacial of −2.3 ± 0.6 ‰; Chen et al., 2003) compared to coastal, more southerly counterparts.Combining the negative measured 18 O late-glacial in northern China (Chen et al., 2003;Ma et al., 2008;Currell et al., 2012;Li et al., 2015) with the positive measured 18 O late-glacial values in central and southeastern China (Wang et al., 2001;Yuan et al., 2004;Dykoski et al., 2005;Cai et al., 2010;Yang et al., 2010) reveals a south-tonorth decrease from positive (south) to negative (north) measured 18 O late-glacial values (Figs. 2 and 6).Previous studies of modern precipitation have identified increasing precipitation δ 18 O values from the coast to inland China during the wet season, sharply contrasting spatial patterns expected from Rayleigh distillation (Aragúas-Aragúas et al., 1998).A more recent work suggests that low wet-season precipitation δ 18 O values over southern China are controlled by the deflection of westerlies around the Tibetan Plateau, whereas precipitation δ 18 O values over northern China are controlled by local-scale rainfall and below-cloud raindrop evaporation (Lee et al., 2012).Therefore, measured 18 O late-glacial values from southern China may reflect changes to atmospheric circulation at broader spatial scales, whereas measured 18 O late-glacial values from northern China may indicate changes to more localized atmospheric conditions impacting processes such as raindrop evaporation in addition to meso-and synoptic-scale circulation changes.

Africa
Measured 18 O late-glacial values from Africa range from −2.9 to +0.1 ‰ (Figs. 2 and 6).Sixteen of 17 measured 18 O late-glacial values from Africa are negative.Near-zero measured 18 O late-glacial values are generally found near to coasts (e.g., Senegal 18 O late-glacial of +0.1 ± 0.8 ‰; Madioune et al., 2014), whereas higher magnitude, negative measured 18 O late-glacial values in Africa are found farther inland (e.g., Niger 18 O late-glacial values of −2.3 ± 2.0 and −2.9 ± 0.9 ‰: ∼ 800 km from the Atlantic coast).General circulation model 18 O late-glacial values show poor agreement with measured 18 O late-glacial over tropical Africa compared to model-measured comparisons for Europe and North America (Fig. 5), with positive simulated 18 O late-glacial values predicted over large parts of Africa where negative 18 O late-glacial values are measured.Figure 5 shows that Africa has the largest inter-model and model-measurement disagreements in the sign of 18 O late-glacial values of the continents.
Northern African hydrological processes are influenced by interlinked controls such as meridional shifts in the position of the intertropical convergence zone (Arbuszewski et al., 2013) and the strength of Atlantic meridional overturning circulation (Mulitza et al., 2008).Paleowater chemistry indicates that northern Africa was at least 2 • C cooler than today (Guendouz et al., 1998) and that westerly moisture transport was stronger than the present during the late-glacial (Sultan et al., 1997;Abouelmagd et al., 2012).
High magnitude, negative measured 18 O late-glacial values are located in Turkey and Georgia south and east of the Black Sea (−2.8 ± 1.0 to −5.7 ± 0.4 ‰; Fleitmann et al., 2009;Arslan et al., 2013;Melikadze et al., 2014).Westerly air mass trajectories distal to the Fennoscandian ice sheet topography may not have changed considerably since the lateglacial over western and central Europe (Rozanski, 1985;Loosli et al., 2001).Therefore, higher, near-zero measured 18 O late-glacial values in western Europe and lower, negative measured 18 O late-glacial values in eastern Europe indicate enhanced distillation of advected air masses during the lateglacial relative to the late-Holocene.
Changes to freeze-thaw conditions of the ground surface between the latter half of the last glacial time period and the modern climates may have impacted the seasonality of the fraction of precipitation recharging aquifers and thus 18 O late-glacial 2004, 2011;Jasechko et al., 2014).Geomorphic evidence suggests permafrost covered portions of Hungary at the last glacial maximum, suggesting that land temperatures may have been up to 15 • C cooler than present day (Fábián et al., 2014), a larger late-glacial to late-Holocene temperature shift than earlier, noble gasbased reconstructions (5-7 • C; Deák et al., 1987).European pollen records indicate that northern Europe was tundralike and that southern Europe was semi-arid during the last glacial maximum (Harrison and Prentice, 2003;Clark et al., 2012).The European late-glacial to late-Holocene transition from semi-arid deserts to temperate forests could have lowered 18 O late-glacial values as groundwater recharge ratios transitioned from more extreme winter-biased (e.g., semiarid lands during the late-glacial) to less extreme winter-biased groundwater recharge ratios (e.g., forests during late-Holocene; Jasechko et al., 2014).
The measured groundwater 18 O late-glacial value located in eastern Brazil is −2.7 ± 1.3 ‰ (Salati et al., 1974).Eastern Brazil was 5 • C cooler than today during the latter half of the last glacial period (Stute et al., 1995b).Four of the five general circulation models simulate positive 18 O late-glacial values across eastern Brazil (Fig. 5), highlighting a difference between simulated and measured 18 O late-glacial values in parts of the tropics.The negative measured 18 O late-glacial value in eastern Brazil has been previously interpreted to reflect higher-than-modern precipitation during the last glacial time period (Salati et al., 1974).Lewis et al. (2010) show that localized rainfall governs precipitation δ 18 O in eastern Brazil.Modern precipitation δ 18 O values are lowest in eastern Brazil when precipitation rates are at a maximum.Extending Lewis et al.'s interpretation linking local precipitation amount to precipitation δ 18 O would suggest that the negative measured 18 O late-glacial value found in eastern Brazil may indeed record wetter-than-modern conditions during the late-glacial as proposed by Salati et al. (1974).Further, disagreement between measured and simulated 18 O late-glacial in eastern Brazil highlights the need to critically evaluate climate model performance in regions where the precipitation amount is closely correlated with precipitation δ 18 O.

Measured
18 O late-glacial from North American proxy records range from −5.5 to +1.0 ‰.Canadian records of groundwater recharge that took place beneath the Laurentide ice sheet are not included in this synthesis ("subglacial recharge"; Grasby and Chen, 2005;Ferguson et al., 2007;McIntosh et al., 2012;Ferguson and Jasechko, 2015).These records were excluded because the subglacial meltwaters that recharged aquifers likely reflect precipitation that fell elsewhere on the paleo-ice sheet, potentially complicating the comparison of groundwater isotope compositions for the late-Holocene and last glacial time period.2002).Decreasing 18 O late-glacial values with increasing latitude along the USA east coast may be explained in part by the isotopic distillation of air masses advected northward from the subtropics under cooler-than-modern final atmospheric condensation temperatures.Indeed, paleoclimate records indicate that Maryland was more arid and as much as 9-12 • C cooler during the late-glacial relative to the Holocene (Purdy et al., 1996;Aeschbach-Hertig et al., 2002;Plummer et al., 2012).In addition to temperature change, late-glacial precipitation isotope compositions along eastern USA coastline were likely impacted by the lower-thanmodern late-glacial sea levels, which changed overland atmospheric transport distances between the late-glacial and late-Holocene (Clark et al., 1997;Aeschbach-Hertig et al., 2002;Tharammal et al., 2013).
Measured 18 O late-glacial values in the central and southwestern USA have the highest magnitude, negative measured 18 O late-glacial values of temperate North America, ranging from −1.0 to −3.4 ‰ .Central and southwestern USA measured 18 O late-glacial values contrast the positive measured 18 O late-glacial values found along the eastern USA coast at similar latitudes.Consistently negative 18 O late-glacial values in central and southwest USA suggest that advected moisture to the region underwent greater upstream air mass distillation during the late-glacial than under modern climate.Pollen, vadose zone and groundwater records show that late-glacial southwestern USA was ∼ 4 • C cooler, had greater groundwater recharge fluxes, and had more widespread forests than present day (Stute et al., 1992(Stute et al., , 1995a;;Scanlon et al., 2003;Williams, 2003).Negative measured 18 O late-glacial values found in the southwest USA have been ascribed to lowerthan-modern summer precipitation (New Mexico, Phillips et al., 1986), latitudinal shifts in the positions of the polar jet stream and the intertropical convergence zone (New Mexico, Asmerom et al., 2010) and changes to over-ocean humidity, temperature or moisture sources (Idaho, Schlegel et al., 2009).Wagner et al. (2010) interpret decreases to southwestern precipitation δ 18 O to reflect cooler and more-humid conditions.Extending this interpretation to negative measured 18 O late-glacial values found across the southwestern USA values supports earlier conclusions that the region was cooler and more humid than today during the late-glacial, possibly linked to changes in air mass trajectories and moisture sources (Asmerom et al., 2010;Wagner et al., 2010).Simulated 18 O late-glacial values across North America closely match spatial patterns of measured 18 O late-glacial synthesized in this study.Strong, multi-model agreement with measured 18 O late-glacial patterns supports continued application of isotope-enabled general circulation models when interpreting North American precipitation isotope proxy records.

Conclusions
While changes to the isotope content of precipitation between the last glacial time period and more recent times has been widely documented, few studies have synthesized these dispersed data to explore the global patterns of δ 18 O change driven by past shifts to regional climate.In this study we compile groundwater, speleothem, ice core and ground ice records of δ 18 O shifts between the late-glacial (20 to ∼ 50 thousand years ago) and the late-Holocene (within the past 5000 years).Late-glacial to late-Holocene δ 18 O shifts range from −7.1 ‰ (i.e., δ 18 O late-glacial < δ 18 O late-Holocene ) to +1.7 (i.e., δ 18 O late-glacial > δ 18 O late-Holocene ).Aquifers with positive measured 18 O late-glacial values (23 % of records) are most common along the subtropical coasts.The majority (77 %) of measured 18 O late-glacial values are negative, with the highest magnitude differences between δ 18 O late-glacial and δ 18 O late-Holocene observed at high latitudes and far from coasts.This spatial pattern suggests that isotopic distillation of advected air masses was greater during the late-glacial than under present climate, likely due to the non-linear nature of Rayleigh distillation, accentuated by larger glacialinterglacial atmospheric temperature changes at the poles relative to lower latitudes.Regionally divergent precipitation δ 18 O responses to the ∼ 4 • C of global warming occurring between the late-glacial and the late-Holocene suggest that continued monitoring of modern precipitation isotope contents may prove useful for detecting hydrologic changes due to ongoing, human-induced climate change.Future paleoprecipitation proxy record δ 18 O research can use these new global maps of 18 O late-glacial records to target and prioritize field sites.In the near term, a global compilation of large lake sediment isotope records that accounts for paleoevaporative isotope effects could enhance spatial coverage of interglacial-glacial δ 18 O shifts.
General circulation models agree on the sign and magnitude of terrestrial precipitation 18 O late-glacial values better in the extra-tropics than in the tropics.Differences in simulated precipitation isotope composition changes amongst the models might be linked to different parameterizations of seawater δ 18 O, glacial topography and convective rainfall, however, these hypotheses require further testing.

Figure 1 .
Figure 1.The change in surface air temperatures from the last glacial maximum to the preindustrial era (gridded data from Annan and Hargreaves, 2013).(a) Percentile ranges of temperature changes since the last glacial maximum for 10 degree latitudinal bands.Blue shading marks the 25th-75th percentile range; thin horizontal lines mark the 10th-90th percentile range.The grey band shows the globally averaged estimate of temperature change since the last glacial maximum of −4.0 ± 0.8 • C. (b) Gridded surface air temperature anomaly from the last glacial maximum to the preindustrial era (data from Annan and Hargreaves, 2013).

Figure 6 .
Figure 6.Regional proxy record 18 O late-glacial values for (a) southeastern Asia, (b) Africa, (c) Europe, and (d) the contiguous United States of America (where 18 O late-glacial = δ 18 O late-glacial − δ 18 O late-Holocene ).The multi-model ensemble median simulated 18 O late-glacial value is shown as a grid (0.5 degree smoothing).Groundwater records are represented by circles, speleothems by triangles, and ice cores by diamonds, labels show measured 18 O late-glacial values for each individual record.
Future model research should focus on quantifying the relative roles of inter-model spread in the simulated climate versus the isotopic response to climate change on resulting simulated precipitation δ 18 O.This would provide guidelines to interpret model-data isotopic differences and to identify what aspects climate models have greatest difficulties capturing.The Supplement related to this article is available online at doi:10.5194/cp-11-1375-2015-supplement.www.clim-past.net/11/1375/2015/Clim.Past, 11, 1375-1393, 2015 water δ 18 O change from the late-glacial (20 000 to ∼ 50 000 years ago) to the late-Holocene (within past ∼ 5000 years; average 18 O late-glacial values shown, where 18 O late-glacial = δ 18 O late-glacial − δ 18 O late-Holocene ).The low temporal resolution of groundwater records means that δ 18 O variations within each time period are smoothed and likely represent unequal temporal weighting.References for measured meteoric water δ 18 O changes for ice cores, groundwater and cave calcite are presented in the Supplement.Our synthesis shows that measured 18 O late-glacial values in the tropics are closer to 0 ‰ (i.e., no change) than 18 O late-glacial values at high latitudes and continental interiors that generally have high magnitude, negative 18 O late-glacial values.High magnitude, negative measured 18 O late-glacial values are most common where present-day precipitation δ 18 O values are at a minimum (e.g., have negative measured 18 O late-glacial values that are of greater magnitude than non-polar measured 18 O late-glacial values (Antarctic and Greenland 18 O late-glacial values range from −3.6 to −7.1 ‰; Fig.3).
18O late-glacial values that are well captured by the climate simulations.However, simulated 18 O late-glacial values over Antarctica and Greenland show large inter-model spread, suggesting that modelbased interpretations of polar ice core records may vary widely among different atmospheric models.