Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

# Rapid northern hemisphere ice sheet melting during the penultimate deglaciation

## Abstract

The rate and consequences of future high latitude ice sheet retreat remain a major concern given ongoing anthropogenic warming. Here, new precisely dated stalagmite data from NW Iberia provide the first direct, high-resolution records of periods of rapid melting of Northern Hemisphere ice sheets during the penultimate deglaciation. These records reveal the penultimate deglaciation initiated with rapid century-scale meltwater pulses which subsequently trigger abrupt coolings of air temperature in NW Iberia consistent with freshwater-induced AMOC slowdowns. The first of these AMOC slowdowns, 600-year duration, was shorter than Heinrich 1 of the last deglaciation. Although similar insolation forcing initiated the last two deglaciations, the more rapid and sustained rate of freshening in the eastern North Atlantic penultimate deglaciation likely reflects a larger volume of ice stored in the marine-based Eurasian Ice sheet during the penultimate glacial in contrast to the land-based ice sheet on North America as during the last glacial.

## Introduction

During glacial terminations, retreating ice sheets release large meltwater fluxes into the ocean. This leads to rising sea level and can initiate strong climate feedbacks when sufficiently large meltwater fluxes reach regions of deepwater formation in the North Atlantic. These ocean-atmosphere interactions, together with associated ocean CO2 release, operate as strong and rapid amplifiers of the original orbital induced insolation change. The rapid retreat of ice sheets which yields high meltwater fluxes, can be caused by marine ice sheet instability, ice sheet saddle collapse and marine ice cliff failure1,2. While the timing and associated deglacial feedbacks have been extensively studied for the last deglaciation (Termination I, TI)3, knowledge of the insolation thresholds, rate of ice sheet melting, and feedback sequence of previous terminations is more limited due to the lack of direct absolute chronology for ice retreat and deglacial warming. Some studies suggest that the sequence of millennial feedbacks may be different in previous terminations4,5. The penultimate deglaciation, Termination II (TII), is of particular interest because orbital boundary conditions were different, and the termination was followed by an interglacial with a + 1.2 to 5.3 m sea level highstand - suggesting both Greenland and Antarctic ice sheets retreated further than during TI6, despite similar highs in atmospheric CO27. To fully understand the last interglacial highstand, clear knowledge of the ice sheets melting time and retreat rates during TII is needed8,9.

Water stored in high latitude ice sheets has a δ18O lower than the mean ocean, meaning the deglacial melting of the Northern Hemisphere (NH) ice sheets lowers the δ18O of the surface ocean (δ18Osw). For regions proximal to release of glacial meltwater, such as the North Atlantic, the rate of δ18Osw depletion and freshening can exceed the global average10,11, which has been useful in diagnosing meltwater routes during TI12,13,14. We propose that the evolution of the North Atlantic δ18Osw can be recorded in coastal European speleothems through the transfer of the δ18Osw signal of the ocean moisture source to the δ18O of rainfall, and thereby the δ18O of dripwater15 from which the speleothem is formed. We report on the North Iberian Speleothem Archive (NISA) from caves within 10 km of the Atlantic coast (Fig. 1) at coastal elevation (<70 m above sea level), directly adjacent to the main moisture source region in the eastern North Atlantic. We employ speleothems spanning the last 25 ky to evaluate the δ18O and δ13C proxy relationships against independent records of regional temperature and δ18Osw. The analysis of speleothems covering TI and TII from the same caves and same proxy indicators provides, by the first time, the opportunity to unambiguously compare the timing and rate of the last two deglaciations on an absolute 230Th chronology, with rates further refined by annual layer counting in TII. We show that the NISA record provides a unique opportunity to directly link the impact of rapid freshening of the North Atlantic on regional atmospheric temperatures, using the same archive, in order to test mechanisms for deglacial feedbacks.

## Results and discussion

### Controls on the δ18O and δ13C of NISA speleothems

Using six stalagmites spanning the last 25 ky to generate a composite splice speleothem δ18O record (Methods; Supplementary Figs. 1 and 2), we test the relationship between the speleothem δ18O and independent δ18Osw estimates from foraminifera in North Atlantic marine sediment cores (Supplementary Fig. 3).

We find that throughout the last 25 ka, where both marine and speleothem records rely on independent absolute chronologies, δ18ONISA exhibits an unusually close correlation with the δ18Osw in the eastern North Atlantic Ocean. The δ18ONISA is most strongly correlated (r2 of 0.91) with the δ18Osw on the Irish Margin16 and a low standard error indicates a consistent relationship throughout the deglaciation despite the potential for variation in atmospheric and surface ocean boundary conditions (Fig. 2, Supplementary Table 1). Strong correlation is also attained for the δ18Osw of the S. Iberian Margin17 and W. Iberian Margin18 (Fig. 2, Supplementary Table 1), as expected given the southward flowing eastern boundary current from the Irish margin along the Iberian margin and via the Atlantic Jet surface water flow through the Straits of Gibraltar19. The latter regions δ18Osw are also closely correlated with the Irish margin δ18Osw (Supplementary Table 2) and show no systematic changes in the δ18Osw gradients (Supplementary Fig. 3). A HadCM3 model of the early phases of the TI deglaciation, simulating meltwater based on ICE6G ice sheet evolution and drainage routing (meltwater ~45% from EIS and 55% from the NAIS), yields similar spatial pattern of meltwater- induced salinity anomaly. Low salinity is concentrated in the North Atlantic north of 40°N, including the Irish Margin10 and extending further southward due to eastern boundary current20 along the Iberian Peninsula, NW coast of Africa, and into the Alboran Sea (Fig. 1). The strong relationship between δ18ONISA and the proximal surface ocean δ18Osw is not observed in Mediterranean region speleothems over TI (Supplementary Table 3), likely due to additional hydrological effects21,22,23.

Several factors may contribute to the coherence between δ18ONISA and the δ18Osw of the proximal ocean. Our analysis of all precipitation events in Northern Spain at a station <100 km west of the cave site in 2015–2016, shows that the majority of rainfall events in this region derive from oceanic moisture uptake in the easternmost North Atlantic ocean proximal to the Iberian Peninsula, especially north of Iberia (Fig. 2d). The high east-west topographic barrier of the Pyrenees and Cantabrian mountains of Northern Spain likely contributes both to the modern dominance of this northern proximal ocean moisture source and the stability of this moisture source over time. The location of caves at coastal elevation, directly adjacent to the main Atlantic moisture source region, minimizes the isotopic distillation between the moisture source area and the cave site. The imprint of such distillation on speleothem δ18Ofrom European caves increases with their increasing inland distance and elevation24; and temperature modulates this distillation so that temperature rather than the δ18Osw of moisture source, becomes the dominant influence on δ18ONISA in central Europe25,26. In our NW Iberian cave locations, rainfall monitoring shows that the slight decrease in rainfall δ18O with decreasing temperature27 appears of similar magnitude but opposite sign as the temperature-dependent fractionation between dripwater and calcite in the cave, leaving the δ18Osw of the Atlantic moisture source as the principal variable expressed in the stalagmites. We propose that the δ18Osw18ONISA relationship identified in TI remained stable over TII.

Additionally, we confirm that over TI, the δ13CNISA in stalagmites from our caves varies inversely with regional temperature records from marine archives (Methods, Fig. 3; Supplementary Fig. 4), matching millennial variations in SST despite potential deviations in marine 14C chronology due to variation in surface ocean reservoir ages28 and the potential for alkenones of differing production ages to be deposited together29. Similar temporal correlation between temperature and δ13C during marine isotope stages 3 and 4 has been observed in speleothems from the Atlantic coastal region of France30. We conclude that the main trends are attributable to the carbon isotope signature acquired through rain equilibration with soil gas and bedrock dissolution, rather than in-cave processes such as prior calcite precipitation (Methods). Multiproxy process modeling suggests that this correlation arises because higher temperature strongly increases vegetation productivity in this biogeographic regime, enhancing soil CO2 production and oversupply of CO2 to karst waters, both factors which produce more negative δ13C in speleothems (δ13Cspeleo)31. We infer that the specific slope of the temperature vs δ13CNISA relationships may be specific to a given cave systems and host lithology, and may be sensitive to the degree of smoothing of millennial-scale changes which is affected by the speleothem growth rate and the years aggregated in each drilling increment.

### NISA speleothem records spanning TII

In the same coastal cave systems, we document the evolution of δ18ONISA and δ13CNISA over TII as indicators of δ18Osw and temperature, respectively. Three 230Th–dated NISA stalagmites replicate the main features of the TII deglaciation between 135 and 129 ka (Fig. 4a). Annual countable fluorescent banding in stalagmite Garth (Supplementary Figs. 5 and 6) additionally provides a precise estimation of the rate of the main δ18ONISA transitions. The record from stalagmite Garth extends to 112 ka BP and provides further context for the deglaciation, albeit with slowed growth and lower resolution between 127.5 and 122.5 ka. We assess the potential effect of in-cave processes such as PCP on the isotope records to focus on the most robust proxy trends (Methods, Supplementary Figs. 710) Given the correlation between δ18ONISA and marine microfossil records of δ18Osw from west and south Iberia in TI (Fig. 1), we tune marine sediment age models to speleothem chronology by synchronizing the major freshening associated with deglacial ice melting and some key temperature events (Methods, Supplementary Figs. 1115).

δ13CNISA records the stepped deglacial warming from 134 to 128 ka, punctuated by millennial cooling events, while δ18ONISA records the deglacial freshening of the eastern North Atlantic beginning 135.7 ka (Fig. 4a). The subsequent descent into the last glacial cycle is also reflected in the increase in δ18ONISA after 122 ka and persistent cooling in δ13CNISA after 118 ka which suggest temperatures similar to the PGM by 112 ka. The annual layer counted chronology in many sections of stalagmite Garth resolves significant centennial to millennial scale variations in both the rate of eastern North Atlantic freshening (δ18ONISA) and its relationship with regional warming (δ13CNISA) (Fig. 4b).

Two large freshening pulses characterize the TII onset (MWPTII-A and MWPTII-B, Fig. 4b). Following previous studies32, we infer that the addition of meltwater is the main driver of rapid deglacial freshening and declines in the δ18Osw in the eastern North Atlantic. Both abrupt freshening pulses began during a period of rapid warming, then, after several decades of rapid freshening, temperatures cooled rapidly, consistent with freshening-induced AMOC slowdown (Fig. 4b). Yet despite cooling, the freshening rate remained high for nearly a century during both events. Following the first period of rapid meltwater release (MWPTII-A), a local freshwater δ18O anomaly was maintained in the eastern North Atlantic for a duration of ~ 600 yrs before the rate of warming increased and over several centuries the surface ocean δ18O anomaly was reduced (Fig. 4b). We infer that slowed meltwater flux and re-invigorated AMOC dissipated the δ18O anomaly through dilution and mixing of meltwater throughout the global ocean.

Following the second and most intense negative δ18O shift (MWPTII-B) and AMOC slowdown, a North Atlantic freshwater δ18O anomaly persisted. Unusually, for ~ 3000 years, the δ18ONISA remained lower than the final interglacial state. The maintenance of a freshwater δ18O anomaly for so long in the surface eastern North Atlantic requires addition of meltwater at rates greater than it can be distributed through mixing with the global ocean. Thereafter, starting around 131 ka, the North Atlantic freshwater δ18O anomaly diminishes rapidly, which suggests that the rate of addition of meltwater into the eastern North Atlantic had slowed sufficiently so that ocean circulation homogenized the deglacial meltwater anomaly to the global ocean average. This decline coincides with a rapid decrease in IRD delivery in the Northeast Atlantic and Labrador Sea (Fig. 4b, Supplementary Fig. 12). The rate of decrease in δ18O of benthic foraminifera (δ18Obenthic) on the Iberian margin also slowed significantly at 131 ka (Figs. 4 and 5), consistent with slowed melt rate.

We detect late phases of melting between 130 and 129 ka. A rapid negative shift in δ18ONISA around 129.7 ka, during regional warming, likely reflects acceleration of meltwater addition. Notably, the continued meltwater flux through 129.3 ka appears to coincide with the deposition of a distinctive sediment layer on the Labrador margin33 which occurred at ~129.3 ka on our chronology (Supplementary Fig. 13), a depositional event attributed to a final North American Ice Sheet (NAIS) flood outburst, analogous to the 8.2 ka event following TI. The 8.2 ka event initiated when ~90% of North American ice had melted34, so if the sediment layer at 129.3 ka is analogous33, it suggests that a similarly high fraction of melting of PGM NAIS ice was likewise complete by this time.

On the W Iberian margin, our speleothem based chronology for TII indicates a much faster rate of depletion in δ18Obenthic during TII than during TI, suggesting that deep waters in the North Atlantic underwent freshening much more rapidly in TII than during TI (Fig. 5). The chronology of the global marine δ18Obenthic stack35, used as index of glacial ice volume to estimate rates of deglaciation and as inputs for model calculations36 is based on tuning to NH summer insolation, assuming a comparable phase relationship to TI. δ18Obenthic curves have also previously been used to estimate sea level37 and the rate of sea level rise9 during TII compared to TI. However, our chronology suggests that the rapid TII deglaciation precedes the peak in high latitude NH summer insolation and the existing chronology35 probably yields minimum estimates of rates of deglaciation. The speleothem chronology for the δ18Obenthic on the W Iberian margin provides an alternate estimate of the rate of sea level change and deglaciation (Fig. 5). This likely represents the upper estimate of the rate of sea level rise, duration, and amplitude of freshwater forcing compared to those employed in recent models using original chronology9. Our independent tuned chronology for δ18Obenthic-based sea level estimates are closest to the oldest chronology (95% upper CI) for the Red Sea curve38,39, and our chronology implies the potential for an earlier and even more concentrated meltwater pulse than estimated from the derivative of the Red Sea sea level curve (Fig. 5)40.

### Similarities and contrasts between the meltwater anomaly during TI and TII deglaciation

Our speleothem absolute chronology confirms that both TI and TII deglaciations initiated at a similar threshold of caloric summer insolation (5.8 GJ m−2) (Fig. 6). Yet, the speleothem chronology for δ18Obenthic reveals attainment of interglacial values within ~5 ky of the TII deglacial onset, compared to nearly 9 ky in the TI deglaciation. Notably, the development of the eastern North Atlantic δ18Osw anomaly during TII is markedly different from that of TI (Fig. 6). During TII, δ18Osw remained more negative than the final interglacial state for nearly 3000 years, a situation which never occurred in TI. During TI, the regional surface ocean δ18Osw anomaly, calculated relative to the benthic δ18Ocalcite on the Iberian margin ((δ18ONISA-BC; Methods), remains below −1‰ only between 16.4 and 15.2 ka (1.2 ky duration), whereas during TII it remains below −1‰ from 134.7 to 131.2 (3.5 ky duration; Fig. 6).

During TI, this period of greatest regional δ18Osw anomaly coincides with the interval of fast melt rate of the Eurasian ice sheet according to recent sea level reconstructions41 (Fig. 7; Supplementary Fig. 16). One interpretation is that EIS meltwater leads to a greater negative δ18Osw anomaly in the eastern north Atlantic compared to NAIS meltwater. Southward advection of surface waters from the Nordic seas along the European continental margin is simulated during periods of retreat of the northern and western EIS20. Intermediate complexity models comparing identical freshwater forcing in different outlets infer a greater salinity anomaly in the eastern North Atlantic due to Eurasian meltwater compared to North American meltwater routed through Hudson Bay, Gulf of Mexico, or Gulf of St. Lawrence. This is because for these NAIS routes, the salinity (and δ18Osw anomaly) would be diluted during advection across the Atlantic by the North Atlantic Drift42. These model simulations, albeit limited, suggest the potential for higher amplitude δ18Osw anomaly in the eastern North Atlantic during EIS melt than NAIS melt, given similar δ18Oice as assumed in previous studies43. Therefore, one explanation for the longer duration surface ocean δ18Osw anomaly during TII is a larger proportion of EIS-derived meltwater in TII compared to TI.

In support of this explanation, Eurasian PGM deposits are found well beyond the limit of ice extent from the LGM, with the largest expansion in the Barents-Kara sectors of northern Russia (Fig. 1). However, the sparse direct chronology of ice positions does not distinguish if these advances were synchronous across all sectors to yield at a single time during the penultimate glacial a significantly larger EIS, or if, as in the last glacial cycle, sectors were out of phase44,45. The potential for a larger PGM EIS ice sheet has been explored in dynamical ice sheet models with total EIS ice volumes of up to 70 m sea-level equivalent, nearly triple that of the LGM EIS46. Because of similar estimated PGM and LGM sea level and benthic δ18O, this simulated large PGM EIS likely coexisted with a smaller than LGM NAIS46. The ice sheet reconstruction featuring a large TII EIS46 encompasses a 2.5–3.5x greater volume of marine-grounded (grounded below paleo sea level) EIS ice at the PGM, compared to the LGM (Methods) with thickest ice in the Fennoscandian sector. Because of isostatic loading and higher relative sea level caused by self-gravitation, the larger PGM EIS would have been even more sensitive to ocean forcing and marine ice sheet instability which could have contributed to a rapid retreat. Although the onset of modeled TII EIS retreat differs slightly from our record8, its duration is broadly consistent with the here documented duration of eastern North Atlantic δ18O anomaly.

An alternative explanation for the greater freshening in the eastern N Atlantic during TII is that LGM-like ice sheet configuration melted faster in TII than in TI. However, we consider this scenario less likely because it would require faster melting of the portions of NAIS which were longest lived during TI, and we have no evidence of a stronger forcing during TII. An Arctic ice shelf, neutrally buoyant but potentially containing a freshwater equivalent to 10 m s.l.e. is proposed to have existed at some time during the penultimate glacial cycle but the timing of formation and collapse remains controversial47,48,49. Based on our records, we cannot confirm nor rule out disintegration of such an ice shelf as a contributor to freshening at the onset of TII, leading the retreat of Northern sectors of the EIS. However, the modest magnitude of Arctic shelf ice volume suggests that this cannot be the only factor responsible for differences in the meltwater anomaly and rate of change in δ18Obenthic and sea level rise over TII (Figs. 5 and 6).

In addition, models suggest that the intensity and duration of the regional δ18Osw meltwater anomaly are also increased by a slowing of the rate of dilution of the meltwater signal, primarily because slowed overturning circulation reduces the vertical dissipation rate11. During TI, the greatest regional freshening anomaly (16.4 to 15.2 ka) also coincides with a period of AMOC reduction inferred from proxies such as Pa/Th50 (Fig. 7, Supplementary Fig. 16). For this time interval, the relative importance of the EIS meltwater source vs AMOC amplification of meltwater anomaly cannot yet be directly deconvolved because existing Pa/Th records suggest only a 30% AMOC reduction during HE151 while the high resolution isotope-enabled OA-GCM experiments simulate a very extreme AMOC shutdown (from 16 to 2.5 Sv). Nonetheless, the temporal coincidence of reduced AMOC and the phase of most rapid EIS retreat in TI may not be coincidence. Simulations suggest that meltwater release from the Northern and western margin of the EIS (and the Mackenzie River outlet of the North Americans), is more effective at decreasing AMOC compared to North American melt routed through the Gulf of Mexico, Hudson Bay, or St. Lawrence outflows42,52, because the former meltwaters reach the locations of northern North Atlantic convection with lesser dilution by other surface waters53. Therefore, the strong surface meltwater δ18Osw between 16.4 and 15.2 ka in the eastern north Atlantic may be both a direct consequence of high EIS melt rates releasing water in the eastern North Atlantic, and an indirect consequence of the high EIS melt rates on AMOC stability and concomitant amplification of the surface meltwater δ18Osw. In and of itself, reduced AMOC intensity, such as that inferred to have occurred during the Younger Dryas according to Pa/Th (Fig. 7, Supplementary Fig. 16), does not coincide with a large freshwater δ18Osw anomaly in the eastern North Atlantic. This suggests that the EIS source effect may be the dominant factor leading to the surface meltwater δ18Osw anomaly early in TI. This hypothesis should be further explored in experiments in high resolution coupled OA-GCM simulations with meltwater release in realistic regions, also exploring the effect of a range of AMOC reduction intensities on the rate of dissipation and therefore the amplitude and persistence of meltwater δ18Osw anomaly.

Further evidence for continual meltwater flux during the long TII meltwater anomaly comes from marine IRD records. On our speleothem chronology, marine records show that peak IRD deposition in both the northeastern Atlantic ODP sites 983 and 984 as well as in sites around the Iberian Margin coincides with the long δ18ONISA anomaly. Because large IRD fluxes are commonly interpreted to reflect a high flux of ice calving from instability in marine-terminating ice sheets54, the IRD peak suggests that the long surface ocean δ18O anomaly was not the result of slowed ocean circulation alone, but that continued meltwater addition from decaying ice sheets either contributed to or was the dominant driver of the signal (Fig. 4b). Large North Atlantic IRD and Heinrich Events (HE) have been linked to the large meltwater release during multiple glacial terminations55. We suggest this period corresponds to HE11, and distinguish the first freshening-induced AMOC slowdown of MWPTII-A as HE11’. While the amplitude and duration of the eastern North Atlantic δ18O anomaly may potentially be enhanced if the release of meltwater shuts down AMOC11, we consider it unlikely that a more intense or longer duration of AMOC weakening during TII is the only cause for the long period of extreme eastern North Atlantic δ18O fresh anomaly. Although there are no direct quantitative proxies for AMOC intensity in TI or TII51, some mid-latitude proxy records are likely to be sensitive to the strong winter season climatic impacts of AMOC slowdown such as the relative abundance of Neogloboquadrina pachyderma sinistral56 or relative abundance of C37:4 alkenone57,58,59 (Fig. 6). These climatic indicators suggest a less extreme AMOC slowdown and sea ice response to stratification during HE11 in TII than in HE1 during TI (Fig. 6, Supplementary Fig. 9). The duration of HE11 and HE1 are comparable. Thus, available evidence would not support a difference in physical circulation as the only cause for the contrasting surface ocean eastern North Atlantic δ18O anomaly in TI vs TII.

Based on these considerations, we suggest that our record of freshening is most consistent with a PGM EIS much larger than its LGM counterpart, and is compatible with the larger EIS volume scenarios of 60 m to 71 m sea level equivalent60,61 based on glacial-isostatic modelling. Further ice sheet modeling and assimilation of near field geophysical data are required to evaluate this interpretation.

### Causes and feedbacks from meltwater-induced AMOC disruption

Our new, highly resolved record of TII reveals that, as in TI, the deglacial sequence was characterized by a series of millennial-scale variations not previously resolved in records of sea level rise35,37,38,39,40. We find evidence for multiple phases of AMOC reduction and reinvigoration in TII which provide new clues to the mechanisms of AMOC instability during glacial terminations. It has been debated, whether meltwater release causes AMOC reduction, or if it is the subsurface ocean warming during AMOC shutdown that causes accelerated ice sheet collapse and meltwater release53. Here, our new annually laminated records show that enhanced freshwater addition during brief periods of rapid warming led, and therefore likely caused, AMOC reductions early in TII.

AMOC recovery has most often been attributed to a slowing of the rate of meltwater addition. Alternatively, recent model experiments suggest that AMOC can also recover, despite sustained meltwater flux, if there is a change in the boundary conditions which set the threshold for weakened AMOC. For example, AMOC may have recovered after HE1 in TI during the Bølling-Allerød because the rapid CO2 rise during HE1 raised the forcing threshold required to maintain weak AMOC5,62. TII was previously proposed to lack such a mid-termination AMOC recovery5. Yet, we identify a recovery after the first AMOC reduction of TII (Fig. 4b). This event occurred earlier in the deglaciation (e.g. δ18Obenthic on the Iberian margin was 4.45 ‰ during the TII recovery compared to 4.25‰ at the start of the Bølling-Allerød) (Fig. 6). One possibility is that AMOC recoveries at TII may have been driven not by evolving thresholds in AMOC sensitivity but rather by strong temporal variations in meltwater forcing. This short duration of the first TII AMOC reduction (600 years vs ~3000 years for HE1 in TI) may reflect the operation of a negative feedback of AMOC slowdown on TII melting rate, if the main location of early TII melting occurred in an area sensitive to AMOC-induced cooling.

On the other hand, if a change in boundary conditions is required for AMOC recovery, then our results in TII present a paradox, because according to the current available ice core chronology, the onset of the first AMOC recovery occurred when CO2 had risen only to 207 ppmv7, much less than the 245 ppmv attained by the onset of the Bølling-Allerød (Fig. 8)7. Even by the end of HE11, according to AICC2012 chronology, pCO2 had only reached 245 ppmv, much less than the 260 ppm reached by the end of the Younger Dryas. This ice core CO2 chronology would suggest a very different sensitivity of AMOC to CO2 between TI and TII. This circumstance could reflect other initial boundary conditions, such as a lower glacial height of the North American ice sheet or different initial PGM thermohaline regime63, which conditioned the AMOC threshold.

It is also possible that ice core records may underestimate the rate of TII CO2 rise due to the differing precision of age models based on annual layer counting for TII64, and age models based on age interpolation between two age control points at 121 and 135 ka tuning ice core O2/N2 ratios to local summer insolation65. The chronology of the main TII CO2 rise can be further explored from an available marine CO2 proxy record based on planktic foraminifera δ11B from the western Caribbean66. This record indicates a large initial CO2 rise synchronous with a surface ocean freshening recorded by the same planktic foraminifera67. Although tuning this site to our speleothem chronology has some uncertainty (Methods, Supplementary Fig. 17), and proxy CO2 estimates are less precise than direct measurements in ice cores, a conservative correlation suggests that a larger portion of the deglacial CO2 rise may have occurred rapidly, coincident with the onset of rapid EIS melting and the second AMOC slowdown (Fig. 7). An early and rapid CO2 rise, introducing greenhouse-forced warming, is also consistent with the progressive warming above glacial background levels occurring after MWP-TIIB in our air temperature record, which would have maintained the melting of NH ice sheets during the HE11 AMOC slowdown (Fig. 8). Thus, diagnosing the links between rate of ice melting, AMOC recovery, and deep ocean carbon storage during TII may require reassessment of the rate of CO2 rise.

Overall, the first direct absolute age constraints on the timing and rate of TII freshwater release to the North Atlantic are consistent with a large EIS, which was prone to rapid retreat due to its enlarged marine-based boundaries. These observations highlight the control that glacial boundary conditions exert, by shaping ice sheet anatomy, over the rates of ice melting and also the nature of the subsequent interglacial9,68. The new records also provide the first evidence for discrete centennial scale meltwater pulses indicative of phases of accelerated ice sheet failure during TII and a detailed view of their coupling with cooling consistent with AMOC reductions. The role of internal ice processes and climate feedbacks associated with the retreat of large marine-based northern hemisphere ice sheets should be further investigated with coupled ocean-climate-ice sheet models.

## Methods

### Identification of final moisture source exchange via backtracking

Moisture uptake regions of 104 rain events which were sampled <100 km west of the cave during 2015 and 2016, were calculated using backward-trajectory analysis with 1 h temporal resolution, performed using the Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) Model (Version 4.8)69 and following a similar methodology as previous studies70. Moisture uptake regions along 850 hPa, 700 hPa and 500 hPa were identified using a specific humidity threshold of 0.5 g/kg increment in at least 6 h, using meteorological data provided by backward-trajectories. Results are shown in Fig. 2d.

### Stalagmite samples

Stalagmite samples were analyzed from the following previously described71 NW Iberian caves: Pindal Cave (stalagmites Candela, Laura), La Vallina Cave (Garth, Gael, Galia, Gloria), and Cueva Rosa (Neith, Alicia). These stalagmites cover TI, Greenland Stadial 22 (GS22), and TII27,71,72. Additional growth phases have been studied in the stalagmite Candela since earlier publication72, and we report additional dates and higher resolution geochemical data. In Candela, A ~10 cm long basal growth phase (CANB) has been sampled along its main growth axis and redated, contrasting with strongly off-axis and condensed sampling of this phase in earlier work. Additionally, within the Bølling-Allerød, a transient 1.5 cm shift in growth axis has been resampled along its expanded central growth axis for 2.5 cm, expanding by 5-fold the record previously sampled in a lateral portion at only 0.5 cm thickness.

### Measurement of seasonal fluorescent banding for growth rate evaluation in key intervals

To account for variation in growth rates73 between 230Th dates, for intervals of rapid freshening in TI and TII, we refine age models by estimating stalagmite growth rates from the width of growth bands defined by fluorescent layers74 which have been described to result from seasonal delivery and/or incorporation of fluorescent organic substances in dripwater75.

Stalagmite slabs were mounted in low-viscosity Laromin C260 epoxy (cycloaliphatic diamine polymer), ground flat with SiC paper, and then polished with a diamond suspension to 0.25 µm in preparation for Confocal laser scanning microscopy (CLSM) and laser ablation ICP-MS. Imaging by CLSM was performed at the Scientific Center for Optical and Electron Microscopy (ScopeM) at ETH Hönggerberg using an OIympus Fluoview 3000. Here, a series of overlapping images were obtained under 100–200x magnification using an incident wavelength of 488 nm and measuring fluorescence in a 490–590 nm window. Images were obtained at 1024 × 1024 or 2048 × 2048 resolution, corresponding to 1.2 to 0.6 µm/pixel (respectively) under 100x magnification. Images were processed using Fiji/ImageJ, where the exact distances between growth layers were recorded on an absolute length scale and later cross-referenced to laser ablation and drilling (isotope, trace element) data. Confocal images were counted in stalagmite Garth section from 177.5 to 269.5 mm from the tip. Growth layer thickness was used to determine rates of isotopic change in sections which could be layer counted, and to develop accurate interpolation of ages between U/Th dates. In areas where annual layers were not immediately visible, typically when layer thickness was less than 3 µm, an average growth rate for that section was assigned in order to be consistent with the sequence of U/Th ages.

### 230Th and 14C absolute dating and age model development

Absolute chronology is based on subsamples drilled for 230Th dating. New dating was completed using standard methods and analysis on MC-ICP-MS (Thermo-Finnigan Neptune)76 at Xi’an Jiaotong University and the University of Minnesota; some previously published aged used in this study also include determinations on SF-ICP-MS (Thermo-Finnigan ELEMENT77); results are given in Supplement Tables 4 and 6. To evaluate growth rate in stalagmite Laura, 14C ages were measured at ETH Zurich with a gas ion source in a Mini Carbon Dating System at the Laboratory of Ion Beam Physics, ETH, via on-line acid digestion78 and calibrated with INTCAL 202079 and results are given in Supplement Table 5. No new age models are elaborated for GS22 and these are shown in27. Because agricultural and pasture land uses became widespread in Northern Spain after 5.5 ka BP80 which may decouple the natural relationships between climate and vegetation over cave sites, we do not present or discuss speleothem records younger than 5 ka BP.

Age models for stalagmites younger than 30 ky are illustrated in Supplementary Fig. 1. Sequences of multiple 230Th ages are interpolated using the Bchron software81,82 (Maria, Candela, Galia, and Alicia A) whereas linear interpolation was employed for short growth phases bracketed only by basal and top U/Th dates (Alicia B, C). Two approaches are used to refine the age model within the 18 to 14 ka interval in Candela. Where visible growth laminae are present, their thickness is used to constrain growth rates (green line in growth rate curve in Supplementary Fig. 1a) and to identify the point at which growth becomes most strongly condensed. For example, the age model is pinned at a new MC age at 18.0 ka ± 78 year 230Th date (at reference level 4.5 mm), and advanced upwards ~90 year using width of fluorescent growth layers to constrain growth rate. Growth layers are not countable in the following 7.7 mm of stalagmite (brown line in Supplementary Fig. 1a). Consequently, in this section two tie points with the geochemical variations in stalagmite Laura (red points in Supplementary Fig. 1a) were used to constrain the Candela age model, and growth rate was varied from 10 to 1 μm/year inversely with Candela δ13C, with a 630 year negligible deposition (<0.2 μm/year) interval at the end of the δ13C maximum. The precision of the age model is lowest between 17.85 and 16.1 ka, and improves by 16 ka due to a 15.4 ka 230Th date and the 16.1 and 14.5 ka tie points with Laura. Following these tie points, the age model until the hiatus at 11.6 ka is established from 230Th tie points, with interpolation between dated points further constrained by counting of 600 annual layers (14.5 to 13.9 ka) during the Bølling-Allerød. This set of speleothem records chronicles the changes across TI in higher detail than previous records from the region, with the following average absolute 95% CI age uncertainties 20–18 ka (Candela, +/−300 years); 18–16 ka (Candela, +/−500 years), 16–14 ka (Laura, +/−150 years); 14–12.5 ka (Candela +/− 260 years).

Age models for stalagmites spanning TII and the Bchron uncertainty windows, are illustrated in Supplementary Figs. 5 and 6. For TII, Garth provides anchor chronology due to its high U content and growth rate and presence of annual fluorescent growth banding; Gael also provides an independent chronology albeit of lesser resolution. For Neith, whose very low U content complicates a precise age model, we develop an age model by tuning main features of isotope record to that of Garth within the uncertainties of the Bchron model. Neith is therefore presented to indicate the reproducibility of the main features of the isotopic records shown in Garth, not as independent verification of precise event chronology. The age model in Garth was developed with the aid of laminae counting (Supplementary Fig. 5).

### Stable isotope analysis and TI stable isotope splice

New stable isotope determinations were measured at Scientific and Technologic Centers from the University of Barcelona with a Thermo-Finnegan MAT-252 coupled to a CarboKiel-II and at ETH Zürich with a Thermo-Finnegan Delta V Plus coupled to Gas Bench II83 and are reported in ‰ relative to the Vienna PDB Standard. Analytical uncertainties for calibration with two in-house carbonate standards and NBS-19 and NBS-18 international standard and were 0.08 ‰ for both isotopes. Samples for isotopes were micromilled at 0.1 to 0.5 mm increments, or drilled at 1 to 5 mm increments, from the central axis of stalagmites. Stalagmite δ18O and δ13C were screened for the coupled effects of CO2 degassing and prior calcite precipitation, as detailed below. A spliced record of δ18ONISA and δ13CNISA is compiled from stalagmites spanning 26–5 ka BP. Where multiple stalagmites grow synchronously, we employ for the splice, the stalagmite with best constrained chronology and the stalagmite in which there is the least variation in Mg/Caindex over the included time interval. The goal of this approach is to limit the potential influence of PCP on the inferred trends in δ18ONISA and δ13CNISA.

### Speleothem Mg/Ca analysis, Mg/Ca as PCP index, and screening for PCP influence on isotope records

Mg/Ca was measured on splits of the isotope samples, at the University of Oviedo (Thermo ICAP DUO 6300)72, and with similar standardization approaches at ETH Zürich (Agilent QQQ 8800); all ratios are reported in mmol/mol. Mg/Ca is widely used indicator of variation in PCP84 and we use speleothem Mg/Ca and the Mg/Caindex as a qualitative indicator of the influence of the coupled process of CO2 degassing and prior calcite precipitation (PCP) on isotope trends, hereafter referred to as PCP. To facilitate comparison across stalagmites from different portions of the cave, which sample stratigraphically distinct portions of the host limestone, we normalize the measured Mg/Ca in each sample (Mg/Casample) to the minimum Mg/Ca of that stalagmite (Mg/Camin), where high index suggests little loss of Ca due to PCP and a low index suggests significant loss of Ca due to PCP.

$${{{{{\rm{Mg}}}}}}/{{{{{{\rm{Ca}}}}}}}_{{{{{{\rm{index}}}}}}}=({{{{{\rm{Mg}}}}}}/{{{{{{\rm{Ca}}}}}}}_{{{\min }}})/({{{{{\rm{Mg}}}}}}/{{{{{{\rm{Ca}}}}}}}_{{{{{{\rm{sample}}}}}}})$$
(1)

Since the δ13C of dripwater acquired through rain equilibration with soil gas and bedrock dissolution is the primary signal of interest, we plot stalagmite δ13C coded by the Mg/Caindex (Supplementary Figs. 4, 8, 10) to identify when there may be an additional component of variation in δ13CNISA due to in-cave modification of dripwater δ13C due to CO2 degassing and PCP. Monitoring suggests that secondary factors such as changes in Mg partitioning, non-bedrock Mg sources, and variations in the congruency of bedrock dissolution due to variable water-rock contact times exert a limited influence on speleothem Mg/Ca in most of our studied stalagmites. We refrain from use of the Mg/Ca index, and interpretation of isotope results, for the <9 ka portions of Pindal Cave stalagmites in which the cave position on the sea cliff generates a strong marine aerosol Mg/Ca influence. Based on our finding that in stalagmites spanning TI, the main glacial/interglacial δ13CNISA contrasts and main transitions do not coincide with unidirectional changes in Mg/Caindex we conclude these are not artefacts of varying PCP.

In TII, the Mg/Caindex is used to ascertain trends which are not influenced by PCP. In stalagmite Garth, the principal transition to strongly negative δ13CNISA136–130 ka occurs within a very narrow range in Mg/Ca (4.7 ± 0.37 (1 s) mmol/mol) (Supplementary Figs. 8, 10), confirming this δ13CNISAtransition is not driven by variable PCP. However, Mg/Ca in Garth rises abruptly after 129 ka and between 127 and 122 ka, the coincidence of maximum Mg/Ca ratios and strongly condensed growth indicated by 230Th ages is most likely explained by enhanced PCP (Supplementary Fig. 10). Between 127 and 122 ka, δ13CNISA may be shifted to more positive values than would be attained if PCP values were comparable to other sections of the stalagmite and therefore, PCP may be attenuating the magnitude of change in δ13CNISA during peak MIS5e warmth in this stalagmite. Consequently, we rely on other stalagmite record for this interval. Stalagmite Gael, shows overall more constant Mg/Caindex than Garth (Supplementary Fig. 8h), suggesting that relative trends in δ13CNISA between early termination (134–130 ka) and peak MIS 5e (127–123 ka) are less influenced by PCP in Gael. At the same time, the trend in δ13CNISA at the termination onset may be amplified in Gael by decrease in PCP between 135 and 134 ka. In stalagmite Neith, Mg/Ca and δ13CNISA correlate strongly within glacial and termination time windows (Supplementary Fig. 8i) and although a shift in mean δ13CNISA between glacial and interglacial is evident, PCP may be the dominant source of temporal variation in δ13CNISA within each of these time windows, so we abstain from interpreting the time series of δ13CNISA in Neith. We propose stalagmite Garth, with high and relatively stable Mg/Caindex and most precise chronology, to present the most reliable δ13CNISA record, except during its condensed interval between 128 and 122 ka; for this latter period, we splice in the temporal variation observed in stalagmite Gael so that long term trends can be accurately assessed.

We also use the Mg/Ca index to evaluate the potential effect of varying PCP on δ18O, since stalagmite δ18O may be isotopically enriched when forming from dripwaters which have experienced significant PCP at rates exceeding those of oxygen isotope equilibration between dissolved inorganic carbon and water, as observed in laboratory studies85,86. The magnitude of this effect in cave settings could be buffered by the large water reservoir exchanging with DIC especially if carbonic anhydrase from soil and cave microbial87,88 communities catalyzes exchange. Using previously published δ18ONISA records from 100 to 80 ka27, we find that a lower Mg/Caindex coincides with stalagmite sections offset to higher δ18ONISA relative to coeval stalagmites, suggesting this index is useful to identify PCP-related influences on δ18ONISA in these cave settings (Supplementary Fig. 7).

In stalagmites spanning TI, the major temporal trends in our δ18ONISA records do not correlate with Mg/Caindex in the intervals of each record selected for the splice (Supplementary Fig. 2). In the spliced record overall, there is no systematic correlation between Mg/Caindex and the decrease in δ18O. Therefore there is no evidence the δ18O trends are artefacts of temporally variable PCP.

During TII, we evaluate whether deglacial δ18ONISA changes coincide with changes in cave-influences (such as PCP) using crossplots of isotopic records with Mg/Ca (Supplementary Fig. 8). We color code data by age. This analysis confirms that in Garth, there is no correlation between Mg/Ca and δ18ONISA; the full transition of δ18ONISAfrom 136 to 132 ka occurs when the speleothem maintains a very narrow range in Mg/Ca (4.8 ± 0.35 (1 s) mmol/mol), suggesting that variation in PCP is not a significant influence on the trend in δ18ONISA in this time interval. Over the termination there is also no correlation between growth rate and δ18ONISA in Garth, the stalagmite for which growth rate is most precisely resolved due to annual layer counting. (Supplementary Fig. 8). Although the timing and features of the δ18ONISA are similar in stalagmite Gael, its δ18ONISA is on average 1% higher compared to stalagmite Garth or Neith. Gael has a lower average Mg/Caindex than the other two stalagmites, suggestive of greater PCP which could lead to offset of the absolute δ18ONISA for stalagmite Gael. Additionally in Gael, the shift to slightly lower mean value of Mg/Ca midway through the termination onset coincides with a reduced offset in δ18ONISA between Gael and Garth (Supplementary Fig. 9), amplifying the δ18ONISA transition in Gael between 135 and 134 ka by about 0.5%. In Neith, there is no correlation between δ18ONISA and Mg/Ca, and for Gael and Neith, growth rates are less precisely constrained but there is no consistent correlation between average growth rate and δ18ONISA. Therefore, we employ stalagmite Garth, with high and relatively stable Mg/Caindex and most precise chronology, as our core δ18ONISA record, except during its condensed interval between 128 and 122 ka, when the period of most significant PCP may increase the δ18ONISA compared to other time intervals in that stalagmite. During the interval of condensed growth in Garth, we splice in the temporal variation observed in stalagmite Gael, plotting them on a separate y-axis so that long term trends can be accurately assessed. We employ the record from stalagmite Garth to estimate the eastern North Atlantic surface ocean δ18Osw anomaly relative to the Iberian margin δ18Obenthic18ONISA-BC; Methods).

### Estimation of δ18Osw relationship with δ18ONISA over TI and calculation of surface ocean anomaly

The δ18O of surface seawater (δ18Osw) is commonly estimated from coupled Mg/Ca and δ18O of planktic foraminifera89. We compare published marine δ18Osw with δ18ONISA over the time interval from 25 to 5 ka BP in fixed 500 year time bins (Fig. 2, Supplementary Fig. 3). The marine records are based on original 14C chronology which for the S Iberian margin assumes constant reservoir age17. For the W Iberian margin we apply recent age model assuming variable reservoir age90. For the Irish margin, the original chronology is based on tuning the % N. pachyderma sinistral relative abundance to the GISPII δ18O variation since 14C dates suggested variable reservoir age32. Superimposed multicentennial- scale variability is evident in both δ18Osw and δ18ONISA, but at this time cannot be confidently independently correlated due to the uncertainty of marine 14C-based and correlation-based age models.

The calculation of regional surface ocean δ18Osw anomaly as a surface ocean δ18Osw minus a global δ18O “ice volume effect” is subject to uncertainties in the age correlation between the δ18Osw and the eustatic sea level curve used to calculate the ice volume effect. To reduce this source of uncertainty and to use a similar metric for TI and TII, we instead describe the local surface ocean δ18Osw18ONISA) relative to the benthic δ18Ocalcite on the Iberian margin, which can be precisely and independently tuned to speleothem chronology for TII. This metric is denominated (δ18ONISA-BC) or (δ18OSW-BC). Although the decrease in benthic δ18Ocalcite is driven by both deep ocean warming and decreasing deep ocean water δ18O components, high resolution OA GCM model simulations show that in the depth range of our Iberian Margin North Atlantic records, decreasing deep ocean δ18O and warming are closely temporally coupled when meltwater release decreases AMOC intensity, because shallowed North Atlantic winter mixed layer depth lead to mid-depth warming11.

### Tuning of marine age models for Iberian Margin to speleothem chronology

We compile published δ18Osw records for TII18,91 based on Mg/Ca SST and δ18Ocalcite and align the major depletion in both δ18Osw of ODP 97791 and in δ18ONISA in Garth, as illustrated in Supplementary Fig. 11. Because NW Iberian caves also lie in the center of a broad region affected by pronounced cooling during AMOC weakening events10, we extend the NW Iberian speleothem chronology by the additional alignment of three key cooling events in the δ13CNISA record with events identified in the alkenone-SST record also from ODP 97792. ODP 977 is the primary tuning target because is the only available δ18Osw record that matches exactly the amplitude of the initial speleothem freshening in both TI and TII (Fig. 6; Supplementary Fig. 13), likely because by the strong density gradient between the surface melt waters and the underlying Mediterranean waters maintains strong surface stratification, but also because higher sedimentation rates in ODP 977 may better preserve the original δ18Osw amplitude (Supplementary Fig. 12). The tuning is based on tie points between 135 and 125 ka leading to highest precision in the marine chronology during this tuned interval and we do not discuss the details of marine records prior to MWPTII-B.

The ODP 977 marine chronology is transferred to ODP 97693 on the S Iberian margin and MD01-244418,59 in the W Iberian margin by alignment of five tie points that represent main trends changes in the G. bulloides δ18Oplank curves (Supplementary Fig. 12) as this species and record has previously been shown to be useful stratigraphical tool in this region to synchronize their signal to Greenland climate evolution94,95,96. The resulting marine chronologies yield coherent main structures in alkenone-SST records92,97 and δ18Osw records18,59 of each site (Supplementary Figs. 13 and 14), but 18Osw on the W Iberian Margin (MD01-2444) is somewhat attenuated, potentially due to low biogenic productivity and foraminiferal growth periods which are biased against extreme conditions96,98.

### Tuning of subpolar-polar North Atlantic marine age models for Iberian Margin to speleothem chronology

For ODP 98399 and 984100 we align the major reduction in %N. pachyderma sinistral with the deglacial temperature rise in the Alkenone SST record from ODP 977, a clear structure that correlates well all along the Iberian margin cores, as well as a subsequent local maximum in %N. pachyderma sinistral evident as a local minimum in alkenone SST (Supplementary Fig. 14). The strategy is based on the synchronicity of millennial scale cooling events resulting from AMOC reorganization, widely used to apply an Antarctic or synthetic Greenland chronology to marine records99. Our tuning is consistent with δ18Oplank and IRD records (Supplementary Fig. 14). We align the Eirik Drift core MD03-2664101,102 by correlating two major structures in the %N. pachyderma sinistral with ODP 984 (Supplementary Fig. 14), which results in end of IRD deposition nearly synchronous with the other N Atlantic sites. The δ18Osw at this location should have a strong regional signature of the NAIS melting through the Labrador Sea, and the major melting is observed after the main NE Atlantic freshening as recorded in the NW Iberian δ18ONISA (Supplementary Fig. 13), but coincident with one of the minor freshening structures of the NISA record at the onset of the MIS 5e and with comparable amplitude. The benthic record δ18Obenthic at these shallow N Atlantic sites (e.g., 984 1650 m, 983 1985 m) is known to lead that of deep N Atlantic and S Atlantic and Pacific over the TI103 and the applied tuning suggests a similar pattern in the TII (Supplementary Fig. 15).

Within H11, we have not relied on temperature records for tuning, because the alkenone temperatures exhibit a slower rate of warming than implied by δ13CNISA. Additionally, compared to alkenones, Mg/Ca SST records from the same cores (not shown)18,91,93 exhibit divergent millennial structure and amplitude of SST change across H11 likely reflecting a seasonality control in the H11 temperature signal. More intense winter cooling but summer warming104 during cold reversals may lead to differences in rate of winter vs annual or summer warming, which may enhance discrepancies among proxies which may have differing influence of summer and winter seasons. An additional factor which has been documented to slow the manifestation of deglacial SST rise in alkenone records on the Iberian Margin is the transport of glacial-aged alkenones into sediments of deglacial and early interglacial age29.

### Tuning of Western Caribbean marine age models for Iberian Margin to speleothem chronology

We tune ODP 999 using its δ18Osw record during the termination and using its δ18Oplanktic record during the interglacial. δ18Osw tuning of TII is based on observed lags during TI. During TI, the δ18Osw of ODP 99967 is based on 14 C age model, updated to INTCAL2013 with constant 400 year reservoir age. During TI the W Iberian margin δ18Obenthic105 is on chronology tuned to GRIP106 and Hulu90. We observe that depletion of δ18Osw at ODP 999 lags that of W Iberian margin δ18Obenthic by an average of 1100 years (Supplementary Fig. 17a). This lag is consistent with timescales of ocean mixing required to distribute the ice volume signal107. Therefore the same lag is used on TII to align the δ18Osw and W Iberian margin δ18Obenthic105 and assign chronology to ODP 999. We do not use the 999 δ18Oplanktic for tuning during the termination because during the phase of meltwater addition, δ18Oplanktic would be expected to have contrasting temperature components on the Iberian margin (strong cooling linked to AMOC shutdown) and western Caribbean (no cooling from AMOC shutdown). However, during the interglacial, these two subtropical locations may be expected to have similar evolution of SST and δ18Osw.Therefore we use the combined signal of δ18Oplanktic to serve directly for correlation during MIS 5e (Supplementary Fig. 17b).

### Estimation of terrestrial and marine components of the PGM EIS

We approximate the ‘marine’ proportion of the T1 and T2 Eurasian ice sheet using the ice sheet reconstructions of46,61 and an ice history that is the same as ICE6G for T2108. The sea level history and glacial isostatic adjustment modelling is from8, using the same ice sheet histories. ‘Marine’ ice is defined here simply as the volume of ice that is grounded below the reconstructed sea level.

### Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.

## Data availability

The absolute U-Series and Radiocarbon dates are provided in the Supplementary Information. The Speleothem stable isotope data and interpolated ages are submitted to the SISAL database, and additionally are available from the ETH Research data archive at doi: 10.3929/ethz-b-000545925.

## References

1. DeConto, R. M. & Pollard, D. Contribution of Antarctica to past and future sea-level rise. Nature 531, 591–597 (2016).

2. Gregoire, L. J., Payne, A. J. & Valdes, P. J. Deglacial rapid sea level rises caused by ice-sheet saddle collapses. Nature 487, 219–222 (2012).

3. Shakun, J. D. et al. Global warming preceded by increasing carbon dioxide concentrations during the last deglaciation. Nature 484, 49 (2012).

4. Cheng, H. et al. Ice age terminations. Science 326, 248–252 (2009).

5. Barker, S. & Knorr, G. Millennial scale feedbacks determine the shape and rapidity of glacial termination. Nat. Commun. 12, 1–12 (2021).

6. Dyer, B. et al. Sea-level trends across The Bahamas constrain peak last interglacial ice melt. Proc. Natl. Acad. Sci. 118, 1–11 (2021).

7. Köhler, P., Nehrbass-Ahles, C., Schmitt, J., Stocker, T. F. & Fischer, H. A 156 kyr smoothed history of the atmospheric greenhouse gases CO2, CH4, and N2O and their radiative forcing. Earth Syst. Sci. Data 9, 363–387 (2017).

8. Dendy, S., Austermann, J., Creveling, J. & Mitrovica, J. Sensitivity of Last Interglacial sea-level high stands to ice sheet configuration during Marine Isotope Stage 6. Quat. Sci. Rev. 171, 234–244 (2017).

9. Clark, P. U. et al. Oceanic forcing of penultimate deglacial and last interglacial sea-level rise. Nature 577, 660–664 (2020).

10. Ivanovic, R. et al. Acceleration of northern ice sheet melt induces AMOC slowdown and northern cooling in simulations of the early last deglaciation. Paleoceanogr. Paleoclimatol. 33, 807–824 (2018).

11. Zhang, J. et al. Asynchronous warming and δ18O evolution of deep Atlantic water masses during the last deglaciation. Proc. Natl. Acad. Sci. 114, 11075–11080 (2017).

12. Wickert, A. D., Mitrovica, J. X., Williams, C. & Anderson, R. S. Gradual demise of a thin southern Laurentide ice sheet recorded by Mississippi drainage. Nature 502, 668 (2013).

13. Keigwin, L. D. et al. Deglacial floods in the beaufort sea preceded younger dryas cooling. Nat. Geosci. 11, 599 (2018).

14. Eynaud, F. et al. New constraints on European glacial freshwater releases to the North Atlantic Ocean. Geophys. Res. Lett. 39, 1–6 (2012).

15. Baker, A. et al. Global analysis reveals climatic controls on the oxygen isotope composition of cave drip water. Nat. Commun. 10, 1–7 (2019).

16. Peck, V. L., Hall, I. R., Zahn, R. & Elderfield, H. Millennial‐scale surface and subsurface paleothermometry from the northeast Atlantic, 55–8 ka BP. Paleoceanography 23, 1–11 (2008).

17. Català, A., Cacho, I., Frigola, J., Pena, L. D. & Lirer, F. Holocene hydrography evolution in the Alboran Sea: a multi-record and multi-proxy comparison. Climate 15, 927–942 (2019).

18. Skinner, L. & Shackleton, N. Deconstructing Terminations I and II: Revisiting the glacioeustatic paradigm based on deep-water temperature estimates. Quat. Sci. Rev. 25, 3312–3321 (2006).

19. Vargas-Yáñez, M. et al. About the seasonal variability of the Alboran Sea circulation. J. Mar. Syst. 35, 229–248 (2002).

20. Bigg, G. et al. Sensitivity of the North Atlantic circulation to break-up of the marine sectors of the NW European ice sheets during the last glacial: A synthesis of modelling and palaeoceanography. Glob. Planet. Change 98, 153–165 (2012).

21. Zanchetta, G. et al. Coeval dry events in the central and eastern Mediterranean basin at 5.2 and 5.6 ka recorded in Corchia (Italy) and Soreq caves (Israel) speleothems. Glob. Planet. Change 122, 130–139 (2014).

22. Bajo, P. et al. Persistent influence of obliquity on ice age terminations since the Middle Pleistocene transition. Science 367, 1235–1239 (2020).

23. Drysdale, R. et al. Evidence for obliquity forcing of glacial termination II. Science 325, 1527–1531 (2009).

24. McDermott, F., Atkinson, T., Fairchild, I. J., Baldini, L. M. & Mattey, D. P. A first evaluation of the spatial gradients in δ18O recorded by European Holocene speleothems. Glob. Planet. Change 79, 275–287 (2011).

25. Moseley, G. E. et al. Multi-speleothem record reveals tightly coupled climate between central Europe and Greenland during Marine Isotope Stage 3. Geology 42, 1043–1046 (2014).

26. Koltai, G. et al. A penultimate glacial climate record from southern Hungary. J. Quat. Sci. 32, 946–956 (2017).

27. Stoll, H. et al. Interpretation of orbital scale variability in mid-latitude speleothem δ 18 O: Significance of growth rate controlled kinetic fractionation effects. Quat. Sci. Rev. 127, 215–228 (2015).

28. Skinner, L. C., Waelbroeck, C., Scrivner, A. E. & Fallon, S. J. Radiocarbon evidence for alternating northern and southern sources of ventilation of the deep Atlantic carbon pool during the last deglaciation. Proc. Natl Acad. Sci. 111, 5480–5484 (2014).

29. Ausín, B. et al. (In) coherent multiproxy signals in marine sediments: Implications for high-resolution paleoclimate reconstruction. Earth Planet. Sci. Lett. 515, 38–46 (2019).

30. Genty, D. et al. Precise dating of Dansgaard–Oeschger climate oscillations in western Europe from stalagmite data. Nature 421, 833–837 (2003).

31. Lechleitner, F. A. et al. Stalagmite carbon isotopes suggest deglacial increase in soil respiration in western Europe driven by temperature change. Clim. Past 17, 1903–1918 (2021).

32. Peck, V. L. et al. High resolution evidence for linkages between NW European ice sheet instability and Atlantic Meridional Overturning Circulation. Earth Planet. Sci. Lett. 243, 476–488 (2006).

33. Nicholl, J. A. et al. A Laurentide outburst flooding event during the last interglacial period. Nat. Geosci. 5, 901 (2012).

34. Tarasov, L., Dyke, A. S., Neal, R. M. & Peltier, W. R. A data-calibrated distribution of deglacial chronologies for the North American ice complex from glaciological modeling. Earth Planet. Sci. Lett. 315, 30–40 (2012).

35. Lisiecki, L. E. & Raymo, M. E. A Pliocene‐Pleistocene stack of 57 globally distributed benthic δ18O records. Paleoceanography 20, 1–17 (2005).

36. Düsterhus, A., Tamisiea, M. E. & Jevrejeva, S. Estimating the sea level highstand during the last interglacial: a probabilistic massive ensemble approach. Geophys. J. Int. 206, 900–920 (2016).

37. Waelbroeck, C. et al. Sea-level and deep water temperature changes derived from benthic foraminifera isotopic records. Quat. Sci. Rev. 21, 295–305 (2002).

38. Grant, K. et al. Rapid coupling between ice volume and polar temperature over the past 150,000 years. Nature 491, 744 (2012).

39. Grant, K. et al. Sea-level variability over five glacial cycles. Nat. Commun. 5, 5076 (2014).

40. Marino, G. et al. Bipolar seesaw control on last interglacial sea level. Nature 522, 197 (2015).

41. Yokoyama, Y. et al. Rapid glaciation and a two-step sea level plunge into the Last Glacial Maximum. Nature 559, 603–607 (2018).

42. Roche, D. M., Wiersma, A. P. & Renssen, H. A systematic study of the impact of freshwater pulses with respect to different geographical locations. Clim. Dyn. 34, 997–1013 (2010).

43. Roche, D. M., Paillard, D., Caley, T. & Waelbroeck, C. LGM hosing approach to Heinrich Event 1: results and perspectives from data–model integration using water isotopes. Quat. Sci. Rev. 106, 247–261 (2014).

44. Svendsen, J. I. et al. Late Quaternary ice sheet history of northern Eurasia. Quat. Sci. Rev. 23, 1229–1271 (2004).

45. Rohling, E. J. et al. Differences between the last two glacial maxima and implications for ice-sheet, δ 18 O, and sea-level reconstructions. Quat. Sci. Rev. 176, 1–28 (2017).

46. Colleoni, F., Wekerle, C., Näslund, J.-O., Brandefelt, J. & Masina, S. Constraint on the penultimate glacial maximum Northern Hemisphere ice topography (≈ 140 kyrs BP). Quat. Sci. Rev. 137, 97–112 (2016).

47. Stein, R., Fahl, K., Gierz, P., Niessen, F. & Lohmann, G. Arctic Ocean sea ice cover during the penultimate glacial and the last interglacial. Nat. Commun. 8, 1–13 (2017).

48. Jakobsson, M. et al. Evidence for an ice shelf covering the central Arctic Ocean during the penultimate glaciation. Nat. Commun. 7, 1–10 (2016).

49. Geibert, W., Matthiessen, J., Stimac, I., Wollenburg, J. & Stein, R. Glacial episodes of a freshwater Arctic Ocean covered by a thick ice shelf. Nature 590, 97–102 (2021).

50. Gherardi, J.-M. et al. Evidence from the Northeastern Atlantic basin for variability in the rate of the meridional overturning circulation through the last deglaciation. Earth Planet. Sci. Lett. 240, 710–723 (2005).

51. Bradtmiller, L. I., McManus, J. F. & Robinson, L. F. 231 Pa/230 Th evidence for a weakened but persistent Atlantic meridional overturning circulation during Heinrich Stadial 1. Nat. Commun. 5, 1–8 (2014).

52. Condron, A. & Winsor, P. Meltwater routing and the Younger Dryas. Proc. Natl Acad. Sci. 109, 19928–19933 (2012).

53. He, C. et al. North Atlantic subsurface temperature response controlled by effective freshwater input in “Heinrich” events. Earth Planet. Sci. Lett. 539, 116247 (2020).

54. Alvarez-Solas, J., Robinson, A., Montoya, M. & Ritz, C. Iceberg discharges of the last glacial period driven by oceanic circulation changes. Proc. Natl Acad. Sci. 110, 16350–16354 (2013).

55. McManus, J. F., Oppo, D. W. & Cullen, J. L. A 0.5-million-year record of millennial-scale climate variability in the North Atlantic. Science 283, 971–975 (1999).

56. Barker, S. et al. Early interglacial legacy of deglacial climate instability. Paleoceanogr. Paleoclimatol. 34, 1455–1475 (2019).

57. Wang, K. J. et al. Group 2i Isochrysidales produce characteristic alkenones reflecting sea ice distribution. Nat. Commun. 12, 1–10 (2021).

58. Martrat, B. et al. Four climate cycles of recurring deep and surface water destabilizations on the Iberian margin. Science 317, 502–507 (2007).

59. Tzedakis, P. et al. Enhanced climate instability in the North Atlantic and southern Europe during the Last Interglacial. Nat. Commun. 9, 4235 (2018).

60. Peyaud, V. Rôle de la dynamique des calottes glaciaires dans les grands changements climatiques des périodes glaciaires-interglaciaires, (2006).

61. Lambeck, K. et al. Constraints on the Late Saalian to early Middle Weichselian ice sheet of Eurasia from field data and rebound modelling. Boreas 35, 539–575 (2006).

62. Obase, T. & Abe‐Ouchi, A. Abrupt Bølling‐Allerød warming simulated under gradual forcing of the last deglaciation. Geophys. Res. Lett. 46, 11397–11405 (2019).

63. Liakka, J., Löfverström, M. & Colleoni, F. The impact of the North American glacial topography on the evolution of the Eurasian ice sheet over the last glacial cycle. Climate 12, 1225–1241 (2016).

64. Marcott, S. A. et al. Centennial-scale changes in the global carbon cycle during the last deglaciation. Nature 514, 616–619 (2014).

65. Suwa, M. & Bender, M. L. Chronology of the Vostok ice core constrained by O2/N2 ratios of occluded air, and its implication for the Vostok climate records. Quat. Sci. Rev. 27, 1093–1106 (2008).

66. Chalk, T. B. et al. Causes of ice age intensification across the Mid-Pleistocene Transition. Proc. Natl Acad. Sci. 114, 13114–13119 (2017).

67. Schmidt, M. W., Spero, H. J. & Lea, D. W. Links between salinity variation in the Caribbean and North Atlantic thermohaline circulation. Nature 428, 160 (2004).

68. Gomez, N., Weber, M. E., Clark, P. U., Mitrovica, J. X. & Han, H. K. Antarctic ice dynamics amplified by Northern Hemisphere sea-level forcing. Nature 587, 600–604 (2020).

69. Draxler, R. & Rolph, G. HYSPLIT (HYbrid Single-Particle Lagrangian Integrated Trajectory) model access via NOAA ARL READY website (http://ready.arl.noaa.gov/HYSPLIT.php), NOAA Air Resources Laboratory. Silver Spring, MD 8, 26 (2010).

70. Baldini, L. M., McDermott, F., Baldini, J. U., Fischer, M. J. & Möllhoff, M. An investigation of the controls on Irish precipitation δ 18 O values on monthly and event timescales. Clim. Dyn. 35, 977–993 (2010).

71. Stoll, H. M. et al. Paleoclimate and growth rates of speleothems in the northwestern Iberian Peninsula over the last two glacial cycles. Quat. Res. 80, 284–290 (2013).

72. Moreno, A. et al. A speleothem record of glacial (25–11.6 kyr BP) rapid climatic changes from northern Iberian Peninsula. Glob. Planet. Change 71, 218–231 (2010).

73. Romanov, D., Kaufmann, G. & Dreybrodt, W. Modeling stalagmite growth by first principles of chemistry and physics of calcite precipitation. Geochimica et. Cosmochimica Acta. 72, 423–437 (2008).

74. Sliwinski, J. & Stoll, H. Combined fluorescence imaging and LA-ICP-MS trace element mapping of stalagmites: Microfabric identification and interpretation. Chem. Geol. 120397, 1–15 (2021).

75. McGarry, S. F. & Baker, A. Organic acid fluorescence: Applications to speleothem palaeoenvironmental reconstruction. Quat. Sci. Rev. 19, 1087–1101 (2000).

76. Cheng, H. et al. Improvements in 230Th dating, 230Th and 234U half-life values, and U–Th isotopic measurements by multi-collector inductively coupled plasma mass spectrometry. Earth Planet. Sci. Lett. 371, 82–91 (2013).

77. Shen, C.-C. et al. Uranium and thorium isotopic and concentration measurements by magnetic sector inductively coupled plasma mass spectrometry. Chem. Geol. 185, 165–178 (2002).

78. Wacker, L., Lippold, J., Molnár, M. & Schulz, H. Towards radiocarbon dating of single foraminifera with a gas ion source. Nucl. Instrum. Methods Phys. Res. Sect. B: Beam Interact. Mater. At. 294, 307–310 (2013).

79. Reimer, P. J. et al. The IntCal20 Northern Hemisphere radiocarbon age calibration curve (0–55 cal kBP). Radiocarbon 62, 725–757 (2020).

80. Kaal, J. et al. Long-term deforestation in NW Spain: Linking the Holocene fire history to vegetation change and human activities. Quat. Sci. Rev. 30, 161–175 (2011).

81. Parnell, A. & Parnell, M. A. Package ‘Bchron’. (2018).

82. Parnell, A. C., Haslett, J., Allen, J. R., Buck, C. E. & Huntley, B. A flexible approach to assessing synchroneity of past events using Bayesian reconstructions of sedimentation history. Quat. Sci. Rev. 27, 1872–1885 (2008).

83. Breitenbach, S. F. & Bernasconi, S. M. Carbon and oxygen isotope analysis of small carbonate samples (20 to 100 µg) with a GasBench II preparation device. Rapid Commun. Mass Spectrom. 25, 1910–1914 (2011).

84. Fairchild, I. J. & Treble, P. C. Trace elements in speleothems as recorders of environmental change. Quat. Sci. Rev. 28, 449–468 (2009).

85. Polag, D. et al. Stable isotope fractionation in speleothems: Laboratory experiments. Chem. Geol. 279, 31–39 (2010).

86. Riechelmann, D. F. et al. Disequilibrium carbon and oxygen isotope fractionation in recent cave calcite: Comparison of cave precipitates and model data. Geochimica et. Cosmochimica Acta. 103, 232–244 (2013).

87. Sauze, J., Jones, S. P., Wingate, L., Wohl, S. & Ogée, J. The role of soil pH on soil carbonic anhydrase activity. Biogeosciences 15, 597–612 (2018).

88. Wingate, L. et al. The impact of soil microorganisms on the global budget of δ18O in atmospheric CO2. Proc. Natl Acad. Sci. 106, 22411–22415 (2009).

89. Lea, D. W. In The Oceans and Marine Geochemistry. (ed H. Elderfield) 365–390 (Elsevier-Pergamon, 2003).

90. Freeman, E., Skinner, L., Waelbroeck, C. & Hodell, D. Radiocarbon evidence for enhanced respired carbon storage in the Atlantic at the Last Glacial Maximum. Nat. Commun. 7, 11998 (2016).

91. Torner, J. et al. Ocean-atmosphere interconnections from the last interglacial to the early glacial: An integration of marine and cave records in the Iberian region. Quat. Sci. Rev. 226, 106037 (2019).

92. Martrat, B. et al. Abrupt temperature changes in the Western Mediterranean over the past 250,000 years. Science 306, 1762–1765 (2004).

93. Jiménez‐Amat, P. & Zahn, R. Offset timing of climate oscillations during the last two glacial‐interglacial transitions connected with large‐scale freshwater perturbation. Paleoceanography 30, 768–788 (2015).

94. Shackleton, N. J., Hall, M. A. & Vincent, E. Phase relationships between millennial‐scale events 64,000–24,000 years ago. Paleoceanography 15, 565–569 (2000).

95. Cacho, I. et al. Dansgaard‐Oeschger and Heinrich event imprints in Alboran Sea paleotemperatures. Paleoceanography 14, 698–705 (1999).

96. Salgueiro, E. et al. Temperature and productivity changes off the western Iberian margin during the last 150 ky. Quat. Sci. Rev. 29, 680–695 (2010).

97. Martrat, B., Jimenez-Amat, P., Zahn, R. & Grimalt, J. O. Similarities and dissimilarities between the last two deglaciations and interglaciations in the North Atlantic region. Quat. Sci. Rev. 99, 122–134 (2014).

98. Ausín, B., Hodell, D. A., Cutmore, A. & Eglinton, T. I. The impact of abrupt deglacial climate variability on productivity and upwelling on the southwestern Iberian margin. Quat. Sci. Rev. 230, 106139 (2020).

99. Barker, S. et al. 800,000 years of abrupt climate variability. Science 334, 347–351 (2011).

100. Mokeddem, Z., McManus, J. F. & Oppo, D. W. Oceanographic dynamics and the end of the last interglacial in the subpolar North Atlantic. Proc. Natl Acad. Sci. 111, 11263–11268 (2014).

101. Irvalı, N. et al. Rapid switches in subpolar North Atlantic hydrography and climate during the Last Interglacial (MIS 5e). Paleoceanography 27, 1–16 (2012).

102. Irvalı, N. et al. Evidence for regional cooling, frontal advances, and East Greenland Ice Sheet changes during the demise of the last interglacial. Quat. Sci. Rev. 150, 184–199 (2016).

103. Lisiecki, L. E. & Stern, J. V. Regional and global benthic δ18O stacks for the last glacial cycle. Paleoceanography 31, 1368–1394 (2016).

104. Schenk, F. et al. Warm summers during the Younger Dryas cold reversal. Nat. Commun. 9, 1–13 (2018).

105. Skinner, L., Elderfield, H. & Hall, M. Phasing of millennial climate events and Northeast Atlantic deep-water temperature change since 50 ka BP. Wash. DC Am. Geophys. Union Geophys. Monogr. Ser. 173, 197–208 (2007).

106. Hodell, D. et al. Response of Iberian Margin sediments to orbital and suborbital forcing over the past 420 ka. Paleoceanography 28, 185–199 (2013).

107. Stern, J. V. & Lisiecki, L. E. Termination 1 timing in radiocarbon‐dated regional benthic δ18O stacks. Paleoceanography 29, 1127–1142 (2014).

108. Peltier, W. R., Argus, D. & Drummond, R. Space geodesy constrains ice age terminal deglaciation: The global ICE‐6G_C (VM5a) model. J. Geophys. Res.: Solid Earth 120, 450–487 (2015).

109. Barker, S. et al. Icebergs not the trigger for North Atlantic cold events. Nature 520, 333 (2015).

110. Skinner, L. & Shackleton, N. Rapid transient changes in northeast Atlantic deep water ventilation age across Termination I. Paleoceanography 19, 1–11 (2004).

111. Carlson, A. E., Clark, P. U., Haley, B. A. & Klinkhammer, G. P. Routing of western Canadian Plains runoff during the 8.2 ka cold event. Geophys. Res. Lett. 36, 1–5 (2009).

## Acknowledgements

We thank student assistants Raquel Rodriguez-Francos, Tim Loeffel, Romain Alosius, and Pien Antjewierden for assistance with drilling, and Joaquin Perona (CCiT-UB) and Madalina Jaggi (ETH) for assistance with stable isotope and trace element analyses. We thank staff at Scientific Center for Optical and Electron Microscopy (ScopeM) at ETH Zürich for support for Confocal imaging and Hongrui Zhang for support in processing of climate model output. We thank J. Zhang for provision of the TRACE-21 POP2 model output, and Andrea Burke and James Rae for discussions. We thank Jacky Austermann and Sarah Dendy for providing the relative sea level data.

## Funding

The research was funded by HIDROPAST (CGL2010-16376) OPERA (CTM2013-48639-C2-1-R) and CHIMERA (CTM2016-75411-R) all to HMS, IC, and AM, and ETH core funding to HS. JT and MI were funded by doctoral FPI fellowships of CTM2013-48639-C2-1-R, and OK by doctoral Fellowship ETH-13 18-1. CP and HC acknowledge NSFC Grant 41888101, AM acknowledges CGL2016-77479-R, and R.L.E acknowledges NSF 1702816 and the 111 program of China grant D19002. IC and JT acknowledge Generalitat de Catalunya Grups de Recerca Consolidats grant 2017 SGR 315 to GRC Geociències Marines, and IC acknowledges the ICREA-Academia programme from the Generalitat de Catalunya and PID2019-105523RB-I00. EG is funded by a Royal Society University Research Fellowship (R1/180317). Samples used in this study were collected in accordance with governing regulations.

## Author information

Authors

### Contributions

Interpretations and manuscript writing was carried out by HS, IC and EG with input from all authors. HS initiated the study and led the speleothem interpretations, IC led the interpretation of δ18sw records and marine tuning, and EG led interpretation of ice sheet history. U-Th chronology was conducted by HS, AM, CP, and MI with support from HC and RLE. Laminae counting and its integration in TII age model development was led by JS. 14 C chronology was completed by NH and OK; OK led TI chronology revision and increased data resolution. Modern rainfall monitoring and HYSPLITT was conducted by MI and new δ18Osw records were produced by JT.

### Corresponding author

Correspondence to Heather M. Stoll.

## Ethics declarations

### Competing interests

The authors declare no competing interests.

## Peer review

### Peer review information

Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work.

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

## Rights and permissions

Reprints and Permissions

Stoll, H.M., Cacho, I., Gasson, E. et al. Rapid northern hemisphere ice sheet melting during the penultimate deglaciation. Nat Commun 13, 3819 (2022). https://doi.org/10.1038/s41467-022-31619-3

• Accepted:

• Published:

• DOI: https://doi.org/10.1038/s41467-022-31619-3