Introduction

Comprehending past rapid climate shifts is pivotal to evaluate the dynamics of current and future climate changes. As occurring today, during the last deglaciation (Termination I, T-I) global temperature1,2 and greenhouse gas concentrations3 rose rapidly, the sea level increased4 as ice sheets retreated5,6 and atmospheric and oceanic circulation patterns reorganised accordingly7,8,9. The final stages of T-I in Greenland, characterised not only by temperature increases but also by cooling events, lasted ~3.0 kyrs according to ice-core records10. After Greenland interstadial (GI) and stadial (GS) 2, a centennial11,12, or even only decadal-scale13,14 temperature increase (GI1) of around 10 °C starting at 14.64 ± 0.18 ka (ka = 103 years before 1950 C.E.)15 was followed by a comparable temperature drop and a consequent return towards glacial-like conditions at 12.85 ± 0.14 ka (GS1). A centennial-scale ~10 °C warming finally led to the Holocene (11.65 ± 0.10 ka)16. These periods of Greenland climate instability preceding the Holocene, namely GI1 and GS1, occurred within the Bølling-Allerød (BA) and Younger Dryas (YD) chronozones respectively15. Uncovering how these abrupt northern high-latitude climate shifts were propagated to lower latitudes, and their pace and tempo at specific sites17, is crucial to refine the understanding of the functioning of the intra-hemispherical climate dynamics on scales similar to the current climate change; thus, it helps to better assess the effect of the current warming of the Arctic on highly populated mid-latitude areas such as the Mediterranean region. The latter is a climate hotspot18, and consequences on the environment and economy of this region will be particularly pronounced. Extreme storms19,20 have already resulted in flooding in the last decade, where Mediterranean cities such as Bari (2005), Vicenza (2010), Genoa (2011, 2014), Nice (2015), Mallorca (2018), Palermo (2020), Como (2021) and Catania (2021) had numerous casualties. Summer heatwaves coupled with dryness are even more often impacting local economies21.

Recent studies have sensibly improved the T-I continental chronology in the Mediterranean17, but precisely dated terrestrial archives are largely confined to the western and eastern sides of the basin17,22,23,24, while they are scarce and/or fragmentary in the central portion. Here, speleothem-based reconstructions of T-I are virtually absent25, and paleo-data are largely derived from lacustrine records only26,27. As a result, multi-archive information about the impact of T-I on important components of in-land climate, as for example rainfall, are unavailable for large parts of the Mediterranean.

One such area is Italy, which is strategically located for the study of the central Mediterranean climate. Italy’s present-day climate is governed by the westerlies, which transport moisture from the Atlantic and modulate temperature28, thus having a major impact on agriculture and ecosystems29. The availability of Atlantic moisture and the strength of the westerlies in delivering moisture across the Mediterranean change on glacial-interglacial timescales, in relation to the ice-sheet extent in the northern high latitudes30. The Atlantic Meridional Overturning Circulation (AMOC) is weaker and the northern ice front moves southward during glacials31; lower temperatures reduce evaporation from the surface of the Atlantic and limit the advection of moisture towards the Mediterranean32, leading to a drier climate. The westerlies and the AMOC, besides controlling hydroclimate dynamics in Mediterranean regions close to the Atlantic coast24, also have a far-reaching impact on the climate further to the east, by for example influencing the hydroclimate and monsoonal activity in the Middle East and Asia, respectively22,33. Deglaciations reverse the above-mentioned situation and, during T-I, the greater availability of Atlantic moisture and stronger westerlies resulted in increased rainfall in the Mediterranean region34, as also suggested for the preceding deglaciation32. Cyclogenesis also occurs in the Mediterranean realm35, and this might predominate when the advection of Atlantic moisture is limited36. Stronger Mediterranean cyclogenesis is expected during times of reduced westerly air flow. These processes have been documented for the western side of the basin for the last ~35 ka24,37,38 as well as for some older periods39,40. For the central Mediterranean, however, rainfall dynamics across T-I remain largely underexplored. Studies of lacustrine sediments did not specifically examine T-I26,27,30 and speleothem-based reconstructions are only available for older periods36. Tightly constraining the timing and progression of T-I in the central Mediterranean and its associated rainfall dynamics is the goal of this paper. This is accomplished by using a high-resolution palaeoclimate reconstruction based on a stalagmite (SA1) from southern Italy (Sant’Angelo Cave, Apulia). Oxygen stable isotope (δ18O) time series is used to 1) accurately constrain the timing of the hydroclimate across T-I, and (2) to evaluate the impact of GI1, GS1 and the early Holocene, as well as other associated events especially in terms of rainfall variability. By doing so, this work ascertains the role of the westerlies in controlling central Mediterranean rainfall dynamics during the largest climate change in recent Earth’s history, with implications for a better understanding of regional hydroclimate changes in this region related to current global warming.

Results and discussion

SA1 stalagmite

SA1 is a ~30 cm-long calcite stalagmite sampled in Sant’Angelo Cave (40.73° N, 17.57° E, Fig. 1, Supplementary Fig. 1 and methods). The age-depth model41 based on high-resolution U-Th dating (total n = 36, period of interest n = 24, methods) attests its deposition between 47.8+0.41/−0.39 and 6.63+0.62/−0.57 ka (Fig. 1). δ18O analyses (total n = 1045; period of interest n = 546, methods) result in an average resolution of 28 years during the period of interest (from ~25 to ~10 ka), and average uncertainties of ±0.30 kyrs (Fig. 1 and Supplementary Fig. 2).

Fig. 1: Sample, analyses and study site.
figure 1

a SA1 stalagmite and analyses accomplished in this study. Coloured bars refer to U-Th dating at Xi’an Jiaotong (blue), National Taiwan (red) and Melbourne (black) Universities, while the blue line marks δ18O analyses (methods) carried out at Innsbruck University. The age model is reported with dots and 2σ error bars (indicating the U-Th ages) and their relative interpolation (green line = average model; dotted lines = propagation of ±2σ uncertainties). b, c Study site and other records discussed in the text.

Full size image

Rainfall variability during T-I

Several lines of evidence point to the primary role of the amount and source effects42 in controlling SA1-δ18O (methods). SA1-δ18O is thus here taken as a rainfall proxy, with more negative values representing wetter periods with a major contribution of Atlantic-derived rainfall. Less negative values are instead suggesting drier periods with a higher contribution of Mediterranean-derived moisture. The same pattern is found in past24,37 and modern43 rainfall dynamics on the western side of the Mediterranean basin. From the late glacial to the Holocene, SA1-δ18O and Greenland ice cores10 show overall similar trends (Fig. 2). This high degree of similarity implies that the central Mediterranean climate changed in concert with temperature and ice-sheet extent variations at high latitudes, as also testified by other studies on the Atlantic fringe of northern Iberia44. In the study area, these variations triggered changes in rainfall amount and source. Similar to older periods32,36,45, low temperatures of the North Atlantic and northern ice-sheet advances caused relative aridity in the central Mediterranean, while glacial retreats and higher temperatures led to an increase in precipitation. Concurrently, wetter periods in the study area entail a higher proportion of Atlantic-sourced humidity, whereas drier intervals feature a higher percentage of moisture originated within the Mediterranean basin.

Fig. 2: SA1 hydroclimate proxies compared to Greenland ice-core data and the Iberian precipitation index.
figure 2

a δ18O Greenland ice-core time series (grey line10) with highlighted GI1 (BA-chronozone), GS1 (YD-chronozone) and Holocene, together with the precipitation index (P-Index) from southern Iberia (blue line24). b SA1 δ18O time series with age uncertainty error bars. c SA1 initial 234U/238U activity ratio and d) growth rate. Peaks of GI2.1 and GS2.1a are marked with a green and pink circle, respectively. The red arrow indicates the gradual warming after peak GS2.1a conditions in Greenland. The black arrows highlight the rapid rainfall variation in the study site and northern Iberia after the same event. Steps 1 and 2 indicate the progression toward GI1 captured in SA1 (see text). The blue double-arrow shows an intra-Holocene dry interval recorded by both SA1 and the P-Index.

Full size image

Ice-core data show that GI2 was composed of two rapid warmings (GI2.2 and GI2.1) starting at 23.34 ± 0.60 and 23.02 ± 0.58 ka respectively, while GS2.1a started at 17.48 ± 0.33 and peaked at 16.05 ± 0.20 ka15. SA1 registered an increase in rainfall (and a higher ratio of Atlantic vs. Mediterranean-sourced humidity) at the onset of the Mediterranean equivalent of GI2.1 at 23.16+0.19/−0.19 ka, as shown by the shift towards more negative values of δ18O (Fig. 2); instead, from 17.74 +0.11/−0.12 ka SA1-δ18O demonstrates a period of less negative values reflecting a decrease in rainfall amount (and a reduced ratio of Atlantic vs. Mediterranean-sourced humidity - Fig. 2). This culminates with a very dry interval between 16.53+0.10/−0.11 and 16.04+0.11/−0.12 ka, synchronous with the peak of GS2.1a in Greenland (Fig. 2). The latter driest interval, lasting ca. 500 years, is associated with Heinrich Stadial 1 (HS1)46, which marked a drop in temperature in northern Iberia44 (Fig. 3) and resulted in record-dry conditions in southern Iberia24. The timing of the transition into GI1 in the NGRIP record and its equivalent in SA1 is consistent, although showing a different pattern. NGRIP records a first gradual increase in temperature (red arrow in Fig. 2) starting at 16.05 ± 0.2 ka, followed by a rapid shift at 14.64 ± 0.18 ka leading to GI1. SA1 instead captured a step-like increase in rainfall (and an increase in the ratio of Atlantic/Mediterranean-sourced humidity) starting at 16.04+0.11/−0.12 ka, with a first rapid pulse leading to a more stable period (step 1 in Fig. 2), from 15.53+0.16/−0.14 to 14.04+0.46/−0.42 ka. This centennial-scale increase coincided with increased rainfall in southern Iberia24 (black arrow in Fig. 1). Step 1 was followed by a second increase in rainfall (step 2 in Fig. 2) starting at 14.04+0.46/−0.42 ka and marking the inception of the equivalent of GI1; the latter terminated at 13.02+0.31/−0.30, consistent within uncertainties with the Greenland ice-core records (12.85 ± 0.14 ka). The end of the GI1 marks the beginning of GS1, a return to near-glacial conditions in Greenland until 11.65 ± 0.1 ka. The equivalent of GS1 in the study area was characterised by a dry climate (and a reduced ratio of Atlantic/Mediterranean-sourced humidity) in the central Mediterranean according to SA1, terminating at 11.95 +0.15/−0.19 ka. The southern Iberian rainfall index24 records a drier climate during this period (Fig. 2). The start of the Holocene entailed a rapid warming in Greenland and an associated increase in rainfall in southern Iberia and in the study area (Fig. 2). Accordingly, SA1 suggests a first precise constraint for the lower boundary of the Holocene in the central Mediterranean (11.95 +0.15/−0.19 ka), which needs to be confirmed by future studies targeting terrestrial proxies in the same region. The link between western and central Mediterranean rainfall variability is further constrained at intra-Holocene timescales (blue double-arrow in Fig. 2). Indeed, from 10.51 +0.41/−0.43 to 9.60 +0.16/−0.14 ka SA1 records a relatively dry period that has a counterpart in southern Iberia24.

Fig. 3: SA1 δ18O time series compared to circum-Mediterranean records.
figure 3

a Ice rafted debris (IRD) from the Gulf of Biscay related to the HS1 event (black fill47), Portugal Margin SST (red line50) and Greenland ice-core δ18O (grey line10). b Ostolo Cave stalagmite δ18O, a proxy for northern Iberian temperature (yellow line44). c SA1 δ18O (light blue line, this study) taken as palaeo-rainfall proxy and Adriatic SST (red line51). d Planktonic foraminifera from the Aegean Sea (brown line49), the Tyrrhenian Sea (purple line48) and the Iberian Margin (dotted orange line46), proxies for Mediterranean planktonic-derived temperatures. e Pollen from Monticchio (green and brown fill26) and Tenaghi Philippon lakes (yellow dotted line27), representing vegetation changes. For locations see Fig. 1.

Full size image

Controls on Mediterranean climate across T-I

Growth of SA1 between 47.82+0.41/−0.39 and 6.63+0.62/−0.57 ka, and of the PC stalagmite36 (from Pozzo Cucù cave, located 40 km northwest of SA) from 106.02+2.8/−2.7 to 26.6+0.8/−0.9 ka, demonstrates that (1) rainfall in southern Italy efficiently recharged the karst aquifers even during the peak of the last glacial; and (2) coupled with the availability of moisture, temperatures were never too low to impede vegetation and soil bioproductivity, hence allowing speleothem deposition in these caves. Precipitation minus evaporation (P-E) was thus positive during the SA1 and PC growth periods, and the widespread aridity and low temperatures affecting central Europe, leading to growth stops in speleothems progressing towards the last glacial acme25, were apparently less severe in the central Mediterranean region. This agrees with observations from the eastern side of the basin22,23, where speleothems grew continuously across the last glacial period. However, as registered by SA1-δ18O, rainfall fluctuated across T-I (Figs. 2 and 3 and methods), but recharge of karst aquifers and related speleothem deposition did not cease even during relatively dry periods. These variations overall agree with Greenland GI/GS2, GI/GS1, and the Holocene inception (Fig. 2), and are also comparable with equivalent rapid climate shifts occurring in continental southern Iberia24. Along a west-to-east transect across the Mediterranean, this pattern of T-I climate variability is also seen in planktonic foraminifera δ18O data of core ODP977 from the Spanish Margin47, core GDEC-4-2 from the Tyrrhenian Sea48, as well as from the Aegean Sea49 (LC21 core) (Fig. 3). This shows that SA1 captured the regional climate variability rather than local cave-related processes. Sea-surface temperatures (SST) changed accordingly on the Portugal Margin50 (MD95-2042 core) and in the Adriatic Sea51, as well as in continental western (Ostolo Cave44) and central (Milandre Cave52) Europe (Fig. 3 and Supplementary Fig. 3). For example, air temperatures increased ~6 °C at the GS1-Holocene transition in Switzerland52 (Supplementary Fig. 3), synchronous with the equivalent transition captured is SA1. Lacustrine ostracod δ18O data from Ammersee in southern Germany agree with this temperature shift53 (Supplementary Fig. 3), while pollen from southern Italy (Monticchio Lake26) and Greece (Tenaghi Philippon Lake27), although of low resolution, testify the expansion of trees and Mediterranean taxa during the equivalent of GI1 and the GS1-Holocene transition (Fig. 3). Especially during the peak of GS2.1a and to a smaller extent during GS1, the reduction in the arboreal pollen was accompanied by the expansion of steppe environments in the Monticchio Lake area26. The latter lake is located in the southern Apennine chain (656 m a.s.l.) ca. 170 km west of the study site, with a maximum catchment elevation of 956 m a.s.l. This implies that the local ecosystem at Monticchio responded to the same rainfall variability across T-I, likely coupled with temperature changes, also captured by SA1. As recently demonstrated, the reorganisation of the northern hemisphere (hydro)climate during T-I occurred within a few decades in response to temperature and ice-sheet extent changes in the Arctic54. It impacted not only the climate in central Europe and the Mediterranean, but also the hydroclimate in the Middle East23 and the monsoon activity in Asia33 (Supplementary Fig. 3).

Our data suggest a rapid variability of the westerly moisture advection across T-I in the central Mediterranean area. A similar reorganisation of the atmospheric circulation has been proposed for the western side of the basin24, suggesting that the entire western and central Mediterranean responded in a coherent manner to Atlantic and northern latitude climate forcings. Indeed, the westerly moisture advection toward the Mediterranean reached its lowest intensity during the equivalent of GS2.1a (Fig. 4). The period of lowest westerlies intensity lasted for ca. 500 years between 16.53+0.10/−0.11 and 16.04+0.11/−0.12 ka, when HS1 induced minimum temperatures as testified by records from the Bay of Biscay (MD95-2002 core46), Northern Iberia (Ostolo Cave44) and the Adriatic (MD90917 core51) (Figs. 1 and 3). The related weakened AMOC triggered a period of record-low rainfall in the central Mediterranean. Because of the reduced availability of Atlantic-sourced moisture in colder-than-today periods, likely partially sourced from the subtropical moisture reservoir55, karst aquifer recharge at SA1 was possibly sustained by Mediterranean-sourced rainfall (methods). At this time, the Atlantic/Mediterranean-sourced rainfall ratio reached its lowest value within the period of interest. Additionally, low air temperatures in the study area51 played a role in maintaining P-E > 0.

Fig. 4: Controls on central Mediterranean rainfall during T-I.
figure 4

Extension of Holocene and GI1 northern ice (green shading) compared to GS2 and GS1 (orange shading). The extent of the Greenland, Laurentide (L) and Fennoscandian (F) ice sheets is shown by dotted lines. The two main moisture sources for the study area (red dot) are marked by open triangles and spirals during the Early Holocene and GI1 (green) and GS1 and GS2 (orange), with their size proportional to their importance. During GS1 and GS2, northern ice sheets expanded, global temperatures decreased and the AMOC weakened. This reduced the westerly moisture advection into the central Mediterranean, where rainfall was mainly derived from Mediterranean cyclogenesis (orange spirals). In the central Mediterranean (study area), the rainfall amount dropped accordingly. During GI1 and the Early Holocene, the northern ice sheets retreated, regional temperatures increased and the AMOC became stronger. The higher availability of Atlantic moisture and increased advection across the Mediterranean Sea led to higher rainfall in the central part of the basin, while Mediterranean cyclogenesis played only a minor role in the study site, similar to modern conditions (see Supplementary Fig. 4).

Full size image

The transition toward the equivalent of GI1 in SA1 shows a step-like pattern (Fig. 2). A similar pattern was recorded by Adriatic SST51 and, albeit of lower resolution, by Iberian Margin47 planktonic foraminifera δ18O (Fig. 3). We propose that rainfall in central-south Italy rapidly increased at 15.53+0.16/−0.14 in concert with the western side of the basin24 (black arrow in Fig. 2), while Greenland was gradually warming (red arrow in Fig. 2) and the AMOC was strengthening with respect to GS2.1a (Fig. 4). However, during the phase called step 1 (Fig. 2), Mediterranean cyclogenesis in the study area was possibly still influencing the rainfall, as the same step-like pattern was not found in Iberia24. The latter, situated between the Atlantic and southern Italy, recorded a higher sensibility to the Atlantic-dominated hydroclimate dynamics than the study site, during a phase of gradual temperature increase in Greenland and a reorganisation of the atmospheric circulation. It is thus possible that the westerlies were delivering moisture to southern Iberia, but their strength was not sufficient to overcome the Apennine mountain belt, giving rise to a rain shadow effect in the study site (Supplementary Fig. 1), However, the rapid GI1-related warming in Greenland at 14.64 ± 0.18 ka was associated with a strengthening of the westerlies delivering moisture to the study site (Fig. 4). Accordingly, this warming resulted in a net increase in rainfall in the study site from 14.04+0.46/−0.42 to 13.02+0.31/−0.30 ka during the equivalent of GI1, testifying the northern latitude control on central Mediterranean rainfall over this (and the following) period. Indeed, GI1 ended at 12.85 ± 0.14 ka in Greenland and temperatures in the Adriatic, Portugal and on the Iberian Margin, as well as in northern Iberia, dropped accordingly (Figs. 2 and 3) because of the establishment of the GS1, which in Greenland terminated at 11.65 ± 0.1 ka. SA1 records a consistent period of low rainfall amount from 13.02+0.31/−0.30 to 11.95 +0.15/−0.19 ka, which was associated with reduced Atlantic evaporation and weak westerly moisture transport (and GI1, Fig. 4) that also led to a synchronous rainfall reduction in southern Iberia24 (Fig. 2). Because temperatures in the Adriatic dropped by at least 5 °C during the equivalent of the GI/GS1 transition (Fig. 3), with values even lower than those before GI1(51), Mediterranean-sourced moisture was not able to counterbalance the Atlantic-derived moisture deficit. Thus, during GS1, the Atlantic/Mediterranean-sourced rainfall ratio was intermediate between the lowest values reached during the peak of GS2.1a and the highest values reached during the Early Holocene (see below). Finally, the onset of the Holocene in Greenland (11.65 ± 0.1 ka) was characterised by a rapid warming and a decrease in ice-sheet extent. The AMOC rapidly returned to its interglacial mode, resulting in a high availability of Atlantic moisture and its efficient westerly advection across the Mediterranean. In the study site, the Holocene started at 11.95 +0.15/−0.19 ka testified by an increase in rainfall amount (Figs. 2 and 3), also reported from southern Iberia24 (Fig. 2). Today the majority of rainfall in the central Mediterranean is related to the westerlies28,35 (Fig. 4 and Supplementary Fig. 4), and Mediterranean-sourced moisture plays only a minor role. The Atlantic/Mediterranean-derived rainfall ratio was possibly even higher during the Early Holocene, as suggested by studies on the western side of the Mediterranean basin24,37,38. Therefore, the beginning of the Holocene in the study area led to the resumption of strong westerlies, transporting moisture towards the central Mediterranean, with the Atlantic/Mediterranean-sourced rainfall ratio reaching maximum values. Analogous to quantitative rainfall reconstructions in southern Iberia56, SA1 attests that the westerly-driven rainfall variability found across T-I was mostly related to changes in winter rainfall amount (Supplementary Fig. 5). This should be taken into account in models aiming to assess future rainfall patterns in the central Mediterranean in response to the ongoing warming of the Arctic.

Conclusions

This study provides, for the first time, a robust speleothem-based chronology for the T-I related events (i.e., the GI/GS variability) and Holocene inception in the central Mediterranean. Atmospheric and oceanographic oscillations in the high northern latitudes during T-I led to changes in moisture sources and rainfall amount. These changes occurred in response to a reduced availability of Atlantic moisture and a weakened westerly moisture transport across the Mediterranean during periods of weak AMOC (GS1 and 2). In the central Mediterranean, this led to a pronounced decrease in rainfall amount as well as of the Atlantic/Mediterranean-sourced rainfall ratio. Conversely, Atlantic moisture availability and westerly advection increased at times of strong AMOC (GI1 and Early Holocene), resulting in an increase in rainfall in the central Mediterranean climate as well as of the Atlantic/Mediterranean-sourced rainfall ratio. In the Mediterranean, modern rainfall mostly occurs in winter and autumn (methods), while spring and especially summer are typically dry. Heavy storms19,20 are increasingly procuring flooding in highly populated areas of the Italian Peninsula, mostly occurring during winter and autumn. Together with summer heat waves, current climate change is increasingly causing economic losses (i.e., 40 and 25% reduction in fodder and maize respectively because of the summer 2003 heatwave21). Such weather extremes will possibly intensify according to future climate scenarios57. Our research documents that rainfall variability in the central Mediterranean during T-I can be linked to events occurring at high northern latitudes. This study reaffirms the role of Atlantic moisture in procuring high rainfall in the central Mediterranean during high-magnitude warming events that characterised the end of the last glacial period.

Methods

Area of study and stalagmite

Apulia hosts one of the largest karst areas in Italy, with more than 2000 caves. The study cave belongs to the Apulian carbonate foreland platform, which is a poorly deformed low-altitude plateau stretching NW-SE for over 100 km, between the southern Apennine foothills and the coastline. Sant’Angelo cave (40.73°N, 17.57°E) is located in the municipality of Ostuni, ~6 km from the Adriatic coastline (Fig. 1 and Supplementary Fig. 1). It is carved in Cretaceous limestone of the Calcare di Altamura Formation, with an entrance at 148 m a.s.l. and comprising a subhorizontal tunnel of 250 m length. The thickness of overlying rock is between 20 and 40 m. The study area is characterised by a meso-thermo Mediterranean sub-humid climate. We refer to meteorological data from the city of Ostuni (last 60 years), the closest available station (<5 km from the cave entrance) of the Italian Protezione Civile service. Average annual precipitation is 699 mm, with November and July being the wettest (93 mm/y) and driest (16 mm/y) months, respectively. January (min: 5.8 °C, max: 11.0 °C, average: 8.4 °C) and August (min: 21 °C, max: 29.1 °C, average: 25.0 °C) are the coldest and warmest months, respectively (Fig. Sup. 4). Annual average temperature is 16.4 °C (min: 13.0 °C, max: 19.9 °C) The majority of precipitation is due to westerly-driven air masses, whose sources are located in the Atlantic, and wind direction is from north-west. Secondary sources are the Mediterranean basin itself and the Euroasian landmass (Supplementary Fig. 4). SA1 stalagmite was found broken close to its original growth position. It is ~30 cm tall, yellowish/translucent in most of its parts and shows macroscopic layering over most of its length. It widens ~9 cm from the bottom and then shows a regular candle-shape morphology with a diameter of ~8 cm. The bottom 9 cm show evidence of diagenesis (dissolution and reprecipitation) and have therefore been excluded from the analyses (Fig. 1).

U-Th dating, age model and δ18O analyses

Powdered subsamples (n = 36) weighing ~50 to ~100 mg were obtained from SA1 by using a manual drill with a 1 mm diameter drill bit (Fig. 1 for locations). These subsamples were extracted perpendicular with respect to the growth axis and forming narrow and elongated trenches following as much as possible the speleothem lamination. U-Th dating analyses were performed in the High-Precision Mass Spectrometry and Environment Change Laboratory of the National Taiwan University (Taiwan) (n = 18), in the School of Earth Sciences of the University of Melbourne (Australia) (n = 14), and at the Institute of Global Environmental Change of Xi’an Jiaotong University (China) (n = 5). For the period considered in this study, from ~25 to ~10 ka, a total of 24 U-Th ages were produced. Briefly, the preparation of the samples involved dissolving calcite in HNO3, separating U and Th compounds through decontaminated resin ion-exchange columns and spiking by a solution of known 229Th/233U/ 236U ratio. The obtained solutions were then analysed using a multi-collector inductively coupled plasma mass spectrometer (MC-ICP-MS). The standard chemical and instrumental procedures applied in the three institutions, as well as the methods used for the correction of the raw data, are detailed elsewhere: Xi’an Jiaotong54, Melbourne58 and Taiwan59. Half-lives of nuclides used for age calculation are given elsewhere54. The age-depth model was constructed using the COPRA algorithm41, which provides ages and propagated uncertainties.

For δ18O, SA1 was milled at 0.2 mm spacing along the central growth axis (drill bit diameter 0.2 mm), and ~0.1 mg of the resulting powder was prepared in a continuous-flow system (GasBench II) and were then measured in a ThermoFisher Delta V plus mass spectrometer at the Institute of Geology, University of Innsbruck (Austria). NBS18, NBS19, CO1, and CO8 standards were used as references. The results are expressed in per mil (‰) units relative to the Vienna Pee Dee Belemnite (VPDB) standard. The 1σ analytical reproducibility was ±0.08‰ for δ18O. SA1 δ18O results (total n = 1045; period of interest n = 546) were anchored to the age-depth model to create the time series interpreted in this paper.

Transition detection algorithm

The stadial-interstadial transitions of stalagmites PC36 and SA1 (this study) during Dansgaard/Oeschger (DO) cycles were detected via an automated approach (Supplementary Fig. 6). Two neighbouring running windows (window length = 200 years) on the annually interpolated stalagmite δ18O data were used. The maximum difference of the average of both sliding windows indicates periods of fast δ18O changes in the data. As stadials show less depleted values compared to interstadials in both speleothems, local minima are the best candidates for a stadial-interstadial transition. In order to find corresponding interstadials in the speleothem data and the Greenland ice cores the stadial-interstadial transitions as established in the NGRIP δ18O record10 on its GICC05modelext time scale15 are used as a reference. The method analyses only those local maxima of the differences of the average of both windows which are in a narrow temporal vicinity with respect to the start of interstadials as defined by NGRIP15. For an individual NGRIP defined stadial-interstadial transition, the time period in which the transition of the stalagmite shall be attributed, is defined to be as large as the half-length of the duration between the stadial-interstadial transition and the next nearest one. We widened this range by the 2σ uncertainty of the U-Th chronology of the respective time period for individual stadial-interstadial transitions. If the age difference between two corresponding stadial-interstadial transitions in both archives was too far off, no DO event was detected. Another limiting factor was the maximum difference between both window averages, which must be large enough to be accounted for. Here, the standard deviation of the differences of both sliding windows was applied as a threshold; local maxima must be larger than the standard deviation of the difference curve of both sliding windows to be considered as an event.

Results

All U-Th ages are in stratigraphic order within uncertainties (Supplementary Data 1). Concentration of 238U is on average 303 ppb (min: 130 ppb; max: 600 ppb), while the detrital thorium is minimal, considering that the average 230Th/232Th activity ratio is well above 1000 (min: 10; max: 11030). For these reasons, the resulting age uncertainties are small (min: 0.3%; average: 2%; max: 17%). According to the COPRA age model, SA1 grew from 47.8+0.41/−0.39 to 6.63+0.62/−0.57 ka; the period of interest for this study spans from 24.80+0.23/−0.25 to 9.57+0.16/−0.15 ka, with average chronological uncertainties of ±0.30 kyrs. SA1 [234U/238U]i varies through time (Fig. 2); from values of around 1.025 at around the peak of GI2.1, it then shifts toward the maximum values (above 1.04) at around the peak of GS2.1a (16.04+0.11/−0.12 ka); afterwards the values monotonously decrease toward the Holocene and reaching the minimal values (1.01) at 9.57+0.16/−0.15 ka. The average growth rate is 15.4 mm/ka during the period of interest (Fig. 2). Between ~25 to 16 ka values are always below 10 mm/ka (Fig. 2), reaching minimal values at 16.04+0.11/−0.12 ka (~0.8 mm/ka). Just after 16.04+0.11/−0.12 ka, growth rate demonstrates the first peak, rapidly reaching (and exceeding) ~50 mm/ka. Then, a period of decreased growth rate with values around ~10 mm/ka coinciding with the step 1 of the progression toward the equivalent of GI1 (Fig. 2) is followed by a second peak of high growth rates (~40 mm/ka) during the equivalent of GI1 at 13.36+0.35/−0.36 ka. Afterwards, values decrease again reaching ~26 mm/ka during the equivalent of GS1 at 12.05+0.14/−0.13 ka. During the first part of the Holocene and until ~10 ka, growth rate remained around 20 mm/ka, while lower values occur after ~10 ka towards the top of the stalagmite.

During the period of interest the resolution for SA1-δ18O is 28 years. The values oscillate between −3.4 and −6.3 ‰, while its pattern resembles the NGRIP ice-core record10 but showing an inverse correlation: more negative δ18O values of SA1 are associated with less negative NGRIP δ18O and vice versa. These similarities, together with the palaeoclimate significance of SA1-δ18O (see below), allowed to identify the main climate periods considered within this work: GI/GS 2, GI1/GS 1 and the early Holocene inception (Fig. 2). As these periods in SA1-δ18O are characterised by abrupt shifts, we took the ages of the δ18O values at the midpoint of the transitions to assign base and top ages for two contiguous events. For example, at around ~12 ka SA1-δ18O shifts from −4.5 to −5.7 ‰ in less than 300 years; the midpoint of this transition at 11.95 +0.15/−0.19 ka is considered as the onset of the Holocene. The same approach was applied to the equivalents of GI1 and GS1. The peaks of equivalents of GI2.1 and GS2.1a were identified by minima and maxima in SA1-δ18O values an at 23.00+0.18/−0.20 ka and 16.04+0.11/−0.12 ka, respectively.

Finally, the transition detection algorithm was used to identify the equivalents of DOs 12, 11, 10, 9, 7, 6, 3 and 2 in SA1-δ18O. This also allowed to identify that the transitions into DO 12, 11, 10, 9, 7 6, 3 occurred synchronously, within uncertainties, in PC and SA1.

Proxy interpretation

The resemblance of SA1 time series with the NGRIP curve and other records presented in the text is already a strong indication that SA1-δ18O reflects surface-related events rather than karst processes. Indeed, if calcite was deposited under non-equilibrium conditions60, kinetic processes would have dominated δ18O fractionation, concealing the original isotopic signature. This is well known61 and different methods are usually applied to test whether or not stalagmites were deposited under equilibrium conditions. Here, SA1-δ18O is compared to PC-δ18O, a stalagmite from a cave of the same area of SA136, thus procuring a replication test62 for the period between ~48 and ~27 ka. Indeed, if the δ18O of two stalagmites from the same cave/area replicate, this means they were responding to the same (external) climate signal trigger rather than to local karst-related factors. SA1 and PC overlap between ~48 and ~27 ka (Supplementary Fig. 6), and their δ18O time series show a high degree of reproducibility. In terms of absolute values they both oscillate between −4‰ and −6‰. Within age uncertainties their δ18O pattern mostly agrees. Taking in consideration DO cycles found in Columbu et al. (2020)36, SA1 and PC replicate DOs 12, 11, 10, 9, 7, 6 and 3 well. This correlation was rigidly constrained by using the transition detection method, applied to both SA1 and PC δ18O time series (Supplementary Fig. 6). At the same time, δ18O in SA1 and PC show divergences during the overlapping period (Supplementary Fig. 6). The two cave sites are dominated by similar climate today (in terms of temperature and rainfall, see Supplementary Fig. 4), and the topographic/geographic setting is also comparable (Supplementary Fig. 1). Thus, the observed divergences might be related to 1) the different resolution of the two records, and 2) heterogeneities of the two karst systems, influencing the δ18O signature of the stalagmites. Interestingly, SA1-PC [234/238U]i and growth rate (Supplementary Fig. 6) are also similar within the overlapping section, suggesting that the two karst systems were also responding to the same climate drivers in terms of (i) residence time; and (ii) cave dripping and CaCO3 supersaturation (see later). Therefore, if the replication test attests that SA1 was deposited under equilibrium conditions from ~48 to ~27 ka, it is reasonable to think that the younger portion (the target of this study) has the same characteristics. However, it is known that carbonates63, including stalagmites64 never form under full equilibrium conditions, although caves are conservative environments where processes favouring kinetics (i.e., evaporation, rapid CO2 degassing, etc.) are commonly limited65. For this reason, a small influence of kinetic fractionation in SA1-δ18O cannot be excluded and, similar to our previous studies36,45,66,67, SA1 is considered as having been deposited under quasi-equilibrium conditions. The replication test also rules out peculiar in-karst processes as main factors driving SA1-δ18O. Accordingly, SA1-δ18O variations primarily reflect palaeoclimate/palaeo-environmental changes. Finally, post-depositional diagenetic processes such as calcite dissolution/redeposition can compromise the original geochemical composition of the stalagmite68. However, SA1 does not show any visual evidence of recrystallization in the studied section (i.e., above the dotted line in Fig. 1). A further indication that the original geochemical system is pristine is that all U–Th dates are in stratigraphic order (Fig. 1, Supplementary Fig. 2 and Supplementary Data 1).

If stalagmites are deposited under quasi-equilibrium conditions and diagenetic processes did not alter the original geochemical status, speleothem δ18O (δ18Osp) is principally inherited by rainfall δ18O (δ18Orw). Considering that on glacial/interglacial time scales global ice-volume changes modify seawater δ18O (and thus δ18Orw), SA1-δ18O has been checked to evaluate this contribution according to other works69. However, the peculiar SA1-δ18O pattern is still well recognisable after the correction for global ice-volume (Supplementary Fig. 7), meaning that other factors influenced δ18Orw, and thus δ18Osp. We have reasons to believe that SA1-δ18O was controlled by the amount effect42 combined with the uptake of moisture from two endmember-like sources, the Atlantic and the Mediterranean. Indeed, the primary role of the amount effect controlling δ18Osp in speleothems of the central and western Mediterranean realm has been indicated by other Mediterranean studies, for example along the Spanish Mediterranean borderlands39 and islands70, Croatia71 and Macedonia72. In Italy, the same effect dominates δ18Osp in the northern and central portion of the Peninsula32,67,73,74,75,76,77 and Sardinia45,66. Rainfall amount was also driving δ18Osp in the above-mentioned PC stalagmite36 in a cave ~40 km NW with respect to Sant’Angelo (Supplementary Fig. 1). This is because air temperature and rainfall amount compete in modulating δ18Orw at central and western Mediterranean latitudes, but while rainfall amount and δ18Orw show a strongly negative gradient of ~−1.6‰ × 100 mm–1 rain, temperature and δ18Orw show a weakly positive gradient78 of ~0.22% °C–1. Additionally, laboratory79 and in-situ80 measurements reveal that during speleothem deposition the equilibrium water-calcite δ18O fractionation ranges between −0.24‰ °C–1 and −0.18‰ °C–1; this makes the temperature-dependent isotope fractionation of atmospheric precipitation above the cave inefficient in driving δ18Osp with respect to the amount effect. Therefore, SA1-δ18O primarily reflects palaeorainfall amount, with more negative δ18Osp indicating increased rainfall and vice versa.

There are other lines of evidence supporting that rainfall controlled SA1-δ18O. Rainfall variations should also affect the karst hydrology and thus the interaction between infiltrating water and bedrock. When rainfall decreases the groundwater residence time increases, leading to more efficient leaching of 234U due to the alpha recoil effect81 and resulting in higher [234/238U]i ratios in speleothems. The opposite effect is expected during wet periods. At the same time, during wet periods carbonate deposition in caves is high and speleothem growth rate potentially increases. Besides the higher rainfall, increased temperature associated with humid periods promotes faster CaCO3 deposition in the cave. The opposite is expected during relatively drier (and colder) periods, thus resulting in slower growth rates. A clear tendency towards lower (higher) [234/238U]i when δ18O values are shifted toward more (less) negative ones is visible in SA1 (Fig. 2 and Supplementary Fig. 6). Furthermore, growth rate maxima are mostly correlated with rapid SA1-δ18O shifts toward more negative values. This ultimately highlights the same control for [234/238U]i, growth rate, and δ18O in SA1, which is ultimately attributed to variation in rainfall amount.

For the reasons outlined above, we are confident that variations in rainfall amount controlled at least in part SA1-δ18O. According to Corchia Cave δ18O32 in northern Italy, which reflects rainfall amount variability across T-II (Supplementary Fig. 8), rainfall changes in southern Italy during T-I were in turn driven by the strength of the westerlies delivering Atlantic moisture across the Mediterranean. The westerlies transport more moisture to the central Mediterranean area when the northern high-latitude ice sheets are small, and the AMOC is strong (as it is today), favouring the advection of moisture across Europe and the Mediterranean (Fig. 432) In contrast, rainfall decreases when advection of moisture is reduced32. This occurs when the northern ice sheets expand and the AMOC weakens. Orbital parameters were similar during T-I and T-II with respect to obliquity and winter insolation82 (Supplementary Fig. 8). Indeed, in the study area, obliquity has a major control on glacial terminations83, while summer insolation has a minor effect. During T-I and T-II SA1-δ18O and Corchia-δ18O were coupled to δ18O changes in planktonic foraminifera (G. bulloides) on the Iberian Margin47 (Supplementary Fig. 8), further indicating that they share the same regional climate trigger during deglaciations.

Besides rainfall amount, another process is responsible in influencing SA1-δ18O. In agreement with Columbu et al. (2020)36, the differential moisture uptake from the Atlantic and the Mediterranean was likely modulating SA1-δ18O. Today the Atlantic Ocean provides most of the rainfall to Apulia (Supplementary Fig. 4), while a smaller contribution is given by the Mediterranean Sea84. Atlantic rainfall δ18Orw is more negative than Mediterranean rainfall δ18Orw, and the primary role of the Atlantic-sourced rainfall occurs if: (1) moisture produced in the Atlantic is abundant; and (2) westerlies efficiently delivers this moisture to the Mediterranean area. This signifies that when polar ice sheets expand, during colder-than-Holocene climate, the role of Atlantic-related moisture decreases in favour of Mediterranean-related moisture (Fig. 4). The opposite is expected if polar ice sheets shrink. The Atlantic vs Mediterranean source effect42 generally mimics the Greenland T-I cyclicity if other effects are of minor importance. In central Mediterranean speleothems, this produces more negative δ18O values during GI1 and the Holocene and less negative values during GS2 and GS1, as also found in the western side of the Mediterranean basin24.