4.1. Stalagmite LC-2 and Devils Hole
 Why do the DH and LC-2 records appear to disagree on the timing of T-II in the Great Basin? The 10-kyr offset between them cannot be attributed to dating uncertainties as the offset is over an order of magnitude larger than the uncertainties. Given the integrity of the dating, it is unlikely that the penultimate deglaciation occurred at such different times within the state of Nevada, suggesting that the climate signal in one or both of these records has been compromised by other factors and/or misinterpreted.
 LC-2 is a much simpler recorder of climate than Devils Hole. The 60-cm long T-II record in LC-2 is from a cave stalagmite that communicated rapidly with the surface and was fed by local rainfall. Furthermore, stalagmites from this region appear to faithfully record Northern Hemisphere warming events, as suggested by the presence of Dansgaard/Oeschger events in speleothems from nearby Goshute Cave, Nevada [Denniston et al., 2007]. In contrast, the DH T-II record is from 1 cm of vein calcite deposited from groundwater that traveled 80 to >160 km through a regional aquifer over several millennia. On the other hand, LC-2 is limited by its short duration. For instance, as LC-2 does not record the onset of warming, it is possible that LC-2 captures only the latter part of the termination and thus underestimates the age of T-II. We emphasize that this is an important uncertainty and cannot be discounted without more speleothem records from this region. Nonetheless, the DH record levels off at interglacial values at or before (when accounting for groundwater transit time) the onset of the 2‰ shift in LC-2 at 133.5 ka, pointing to what are perhaps irreconcilable differences between the records if both are interpreted to record the timing of changes in the isotopic composition of precipitation. With the data at hand, we cannot resolve the apparent discrepancy between LC-2 and DH and leave this issue to future research.
4.2. Devils Hole Interpretation
 We turn now to the interpretation of the DH record. Much of our discussion focuses on T-I given the large number of well-dated records from this interval. DH also shows an early timing for this termination relative to insolation and sea level (Figures 3j and 3k), with a ∼1‰ increase from 27–17 ka (Figure 3h) [Winograd et al., 2006], suggesting that insights gained from T-I may apply to T-II as well.
 While DH was originally interpreted as a record of global glacial cycles [Winograd et al., 1988, 1992], it is generally now thought to be linked to Pacific SSTs via an atmospheric teleconnection [Herbert et al., 2001; Lea et al., 2000; Winograd et al., 1997, 2006]. Compelling evidence for this interpretation was first presented by Herbert et al. . They developed alkenone-based SST records along the western margin of North America spanning the past several glacial cycles and found that warming at more northerly sites preceded global deglaciation, as recorded by benthic δ18O, by ∼10–15 kyr during terminations. They suggested that DH records this anomalous early warming of the northeast Pacific and thus provides a record of regional, but not global, significance. While attractive, this interpretation has some drawbacks. Perhaps most important, California margin alkenone and foram-based SST records diverge during T-I, raising some questions over the reliability of these proxies during terminations [Hendy, 2010]. There are also some notable disagreements between the alkenone records and DH. First, during T-II the SST records begin warming at ∼145 ka on their orbitally tuned benthic δ18O chronologies, which lags behind initial warming in the DH record by 5–10 kyr. This lag is likely even longer when considering that the DH record must be shifted older to account for groundwater transit time and the SST records should perhaps be shifted younger as terminations in Pacific benthic δ18O may be delayed by several kyr [Lisiecki and Raymo, 2009; Skinner and Shackleton, 2005]. Second, SST records show a two-step warming pattern over T-II (ODP 1012 and 1020 [Herbert et al., 2001]; ODP 1014 [Yamamoto et al., 2004]), in contrast to the smooth termination in the DH record. Third, the DH record plateaus after 17 ka during T-I despite the ∼1‰ decrease in seawater δ18O associated with ice-sheet melting [Adkins et al., 2002]. This ice-volume effect is apparently masked in the DH record by warming of equal and opposite magnitude. No such warming is evident in the alkenone records, however. We return to this issue below.
 Winograd et al.  extended Herbert et al.'s  interpretation, suggesting that DH is associated with early warming of Pacific SSTs not just in the northeast part of the basin, but perhaps also in the tropics [Lea et al., 2000] or even in both hemispheres of the Pacific. They point to high correlations (r) of 0.66–0.83 over the last glacial cycle between DH and several SST records from 2–41°N in the eastern Pacific as evidence of this linkage. We suggest that these correlations are largely a function of the strong 100-kyr signal present in most paleoclimate records of the last glacial cycle and may not necessarily imply that the early warming in DH during terminations is related to Pacific SSTs. For example, we obtain similar correlations between DH and the following records that do not display early terminations using the same approach as Winograd et al.  (i.e., 160–4.5 ka window, a common 1.7 kyr time step): LR04 benthic δ18O, −0.50 [Lisiecki and Raymo, 2005]; Vostok CO2, 0.62 [Petit et al., 1999]; Dome C δD, 0.72 [EPICA Community Members, 2004]; Aghulas core MD962077 SSTs, 0.85 [Bard and Rickaby, 2009]. We also use the leading principal components of radiocarbon-dated, high-resolution SST records of T-I from several regions of the Pacific (P. U. Clark et al., Global climate evolution during the last deglaciation, submitted to Nature Geoscience, 2011) to determine if and/or how DH is related to large-scale modes of SST variability. The PC1s account for 53, 44, 84, and 81% of the variance in the North (n = 8), eastern tropical (n = 9), western tropical (n = 11), and South (n = 8) Pacific, while the PC2s explain 19, 27, 5, and 8% of the variance in these regions. All of the PC1s are dominated by the deglacial warming signal from 20–11 ka (Figure 3b), and the PC2s display millennial-scale oscillations similar to the Heinrich 1-Bolling/Allerod-Younger Dryas sequence in the North Atlantic. DH, on the other hand, exhibits a steep rise from 19–17 ka followed by a plateau extending to the mid-Holocene (Figure 3h). Thus, DH appears largely unrelated to the two leading modes of SST variability during T-I around the Pacific, which together account for 71 to 89% of the variance in these areas.
 Regardless of its driver or spatial significance, the DH record is fundamentally interpreted as a record of isotopes in precipitation in the Great Basin. If it provides an at least regionally representative climate record, we would expect other continental records, particularly those also recording isotopes in precipitation, to display similar signals. On the contrary, various proxy records from around the western U.S., including lake sediments [Benson, 2003] and glaciers [Clark et al., 2009], disagree with the early timing for T-I and instead place the regional termination in phase with the global deglaciation. More significantly, two speleothem records from Arizona and New Mexico bear little resemblance to DH on the timing and structure of T-I (Figure 3g). The speleothems are in phase with the global deglaciation and are apparently more strongly related to Atlantic than Pacific forcing, with both exhibiting pronounced millennial variability associated with the Bølling-Allerød and Younger Dryas events in the North Atlantic [Asmerom et al., 2010; Wagner et al., 2010]. These southwestern speleothems are in fact better correlated with distant speleothem and ice core records of T-I than with the nearby DH record, as seems to also be the case with stalagmite LC-2 during T-II (Figure 3). Lastly, oxygen isotopes in radiocarbon-dated groundwaters from southern Nevada record T-I from roughly 20–11 ka [Benson and Klieforth, 1989], well after the termination in DH. Thus, DH disagrees with various land-based climate records from the region during T-I, much as it seems to with LC-2 during T-II.
4.3. Devils Hole Ice-Volume Correction
 DH, as any isotopic record, is subject to overprinting by changes in the oxygen isotopic composition of the ocean associated with variations in global ice volume. At times when DH diverges from the marine δ18O record during terminations, this effect may be particularly pronounced. We use an inverse ice-sheet model reconstruction of seawater δ18O to correct the DH record for changes in ice volume during the last two terminations (Figure 3h) [Bintanja and van de Wal, 2008] We use the originally published age model for this reconstruction, and do not account for the groundwater transit time of the ice-volume signal through the Devils Hole aquifer or for possible local seawater δ18O changes in the Devils Hole vapor source region. During T-II, this ice-volume correction slightly dampens the initial rise in DH δ18O, and more significantly, extends the termination by ∼7 kyr, thus delaying the onset of peak interglacial conditions until ∼127 ka. This lengthened termination may help bring DH into better agreement with our LC-2 stalagmite record, although the δ18O shift in LC-2 would still be many times larger than in DH during their period of overlap. T-I is similarly affected. The gradual early warming in DH becomes more muted and most of the deglacial rise now occurs from ∼19–11 ka, or in phase with the global sea-level rise and SST warming (Figures 3b and 3k). If this admittedly simple ice-volume correction is reasonably accurate, therefore, DH's depiction of terminations and interglacials may require some reconsideration [Winograd et al., 1997]. We acknowledge, however, that while this correction brings DH into general synchrony with the global deglaciation during T-I, the much-discussed onset of T-II in DH still begins much earlier than would be expected from classic Milankovitch forcing.