The age model for the study interval uses three previously published tie-points between ODP 976 Mg / Ca-derived SST (Jiménez-Amat and Zahn, 2015) and the speleothem temperature records from Dongge cave in China (Kelly et al., 2006; see Fig. S1 in the Supplement). Several other marine records covering MIS 6 in the region are chronologically tuned to speleothems (e.g. Sierro and Andersen, 2022; Tzedakis et al., 2018). For the lower interval, the low resolution of planktonic isotopic data available for ODP 976 did not allow direct correlation to global temperature stacks or orbital configuration (von Grafenstein et al., 1999). Instead, we chose to align the higher-resolution pollen record produced in this study with the one from MD01-2444 core on the Portuguese margin (Margari et al., 2010, 2014; Tzedakis et al., 2018). Previous studies highlighted the strong similarities between pollen records on the Atlantic margin and the Alboran Sea during the last glacial period (Fletcher et al., 2010; Fletcher and Sánchez Goñi, 2008; Sánchez Goñi et al., 2002), supporting this approach. MD01-2444 chronology is based on the alignment of benthic isotopic events with the Antarctic temperature record, on AICC2012 timescale (Jouzel et al., 2007; Margari et al., 2010; Shin et al., 2020). The eleven tie-points between MD01-2444 core and EPICA Dome C can be found in the Supplement (Table S1). Five peaks of temperate forest in MD01-2444 were used as control-points for ODP 976 (Table 1). The main assumptions of these tuning approaches to the Dongge cave speleothem and the Antarctic record are discussed in detail in Jiménez-Amat and Zahn (2015) and Margari et al. (2010) respectively. The obtained chronology for ODP 976 core for MIS 6 prevents any assessment of the southwestern Mediterranean vegetation response to global climatic events.

A linear regression was applied to obtain a continuous age for the study interval, spanning from 126.4 to 196.6 ka BP, with a mean resolution of about 350 years for the record (Fig. 2).

Two hundred samples have been analysed in this study, between 40 and 54 m (mcd) depth. The sample processing followed the traditional steps used for pollen extraction (Faegri and Iversen, 1964) and previously applied to the ODP 976 core (Combourieu-Nebout et al., 2002; 2009; Sassoon et al., 2023, Charton et al., 2025). It included sample weighing between 5 and 10 g of sediments, a 150 µm sieving for retrieving macrofossils and macroparticles, followed by 10 % HCl, 40 % HF, 20 % HCl and a final 10 µm sieving.

A minimum of 150 pollen grains were counted for each sample, excluding Pinus as it is usually overrepresented in marine sequences (e.g. Combourieu-Nebout et al., 2002; Fletcher et al., 2010; Mudie, 2011, and references therein), and represents often more than 50 % of the total pollen sum in the study interval (see Fig. 3).

Ecological groups of pollen taxa were defined following previous studies of the ODP 976 record (Charton et al., 2025; Combourieu Nebout et al., 2009; Sassoon et al., 2023). The percentage pollen diagram was constructed using the rioja R package (Juggins, 2024). Constrained Incremental Sum-of-Squares (CONISS) cluster analysis was applied for pollen zonation, using the vegan package on R (Oksanen et al., 2026).

Four methods were applied to the ODP 976 record to reconstruct past climate changes during MIS 6. This is the first time this approach is used for the entire MIS 6 interval (Sinopoli et al., 2019). The multi-method approach allows for a more accurate climate reconstruction (trends and rapid events) compared to the traditional single-method approach (Chevalier et al., 2020; Peyron et al., 2011, 2013; Salonen et al., 2019; Sassoon et al., 2025). It also allows us to compare the reliability and biases of the different methods, which are based on different ecological principles and mathematical algorithms. Four methods were used in this study.

The Modern Analogue Technique (MAT) is the first “assemblage” method ever developed to estimate climate parameters based on pollen assemblages, and is still the most widely used (Guiot, 1990). It is based on the calculation of a dissimilarity index between the fossil samples and samples from a modern pollen dataset. The values of the closest modern analogues selected (here, 4) are averaged to reconstruct the climate parameters for each fossil sample. Weighted-Averaging Partial Least Squares (WA-PLS) (ter Braak and Juggins, 1993) is the second most widely used method, and is based on a different mathematical approach using non-linear regression. Assuming that taxa are most abundant where they find their optimum climatic conditions, WA-PLS models the plant/climate relationships from the modern calibration dataset, weighing the climatic values based on the pollen taxa percentage. These plant pollen abundance/climate transfer functions are then used to calculate the climate parameters of the fossil samples. The last two methods, Random Forest (RF) and Boosted Regression Trees (BRT), rely on a completely different approach using machine learning: they generate a large set of regression trees based on a randomised pollen dataset by bootstrapping (with pollen taxa selected randomly). Contrary to RF (Prasad et al., 2006), BRT (Salonen et al., 2012) assigns a higher probability to select samples that have not been selected before (boosting), increasing the performance of the model for elements that are less well predicted (Chevalier et al., 2020). The application of these machine learning methodologies in paleoclimatology is very promising, especially for BRT, and they have already been validated through different European and Mediterranean pollen records, for different time periods (Charton et al, 2025; d'Oliveira et al., 2023; Robles et al., 2023; Robles, 2022; Salonen et al., 2019; Sassoon et al., 2025).

The four methods were run on R using the Rioja package for MAT and WA-PLS (Juggins, 2024), dismo for BRT (Hijmans et al., 2024) and randomForest for RF (Liaw and Wiener, 2022).

We used the modern pollen dataset compiled by Peyron et al. (2013, 2017) and updated by Dugerdil et al. (2021) and Robles et al. (2023). Samples belonging to non-relevant biomes for this study were excluded (Taiga, Tundra, Pioneer Forest, warm steppe and hot desert), resulting in 2373 samples for calibration dataset spanning Eurasia and NW Africa (see Supplement, Fig. S2). A total of 103 harmonized pollen taxa are included in the dataset, excluding Pinus and aquatic taxa.

Six climate variables were reconstructed: PANN (annual precipitation), MAAT (annual temperatures), SUMMERPR (summer precipitation), WINTERPR (winter precipitation), MTWA (mean temperature of the warmest month) and MTCO (mean temperature of the coldest month). The climatic tolerance spectra of the ten most abundant pollen taxa in the ODP 976 record have been reconstructed based on the modern dataset (Fig. S3). They show that steppe and semi-desert taxa display the highest tolerance to low winter temperature and precipitation, while Mediterranean taxa (Olea and Quercus ilex-type) are the most tolerant taxa to high annual and summer temperature, and Cedrus and Ericaceae to higher annual and seasonal precipitation. SUMMERPR was poorly reconstructed according to the accuracy indicators (Table 2). This is in agreement with previous studies showing the poor reliability of summer parameters (e.g. Camuera et al., 2022). Therefore, we chose to represent seasonal parameters as contrast values for a better visualization: TCON (temperature contrast) = MTCO - MTWA, and PCON (precipitation contrast) = WINTERPR - SUMMERPR. The MTWA and SUMMERPR results can be found in the Supplement (Fig. S4).

For comparison with the present-day climate, and the calculation of anomalies, the modern values were extracted from ERA 5 reanalysis of the ECMWF (European Centre for Medium-Range Weather Forecasts), based on data assimilation into meteorological modelling from 1960 to 2022 (Hersbach et al., 2020). Pollen dispersal is reduced beyond a radius of 175 km (Rojo et al., 2016). However, other studies suggest that pollen can be transported up to distances of 200–300 km (Fernández-Rodríguez et al., 2014) and even over distances greater than 500 km (Bayr et al., 2023; Damialis et al., 2017). Sediments from marine cores such as ODP 976 can therefore include close and long-distance pollen. To account for these observations, an averaged value of the climate parameters on a 400 km radius around the ODP 976 site was extracted (Fig. S5), giving MAP = 478 mm, MAAT = 16.78 °C, MTCO = 13.78 °C, WINTERPR = 172 mm, TCON = -9.59, PCON = 141.5 mm. These values for modern climate, averaged temporally and spatially, provide a better basis for understanding the nature of the climate signal extracted from a marine palynological sequence at a regional pluri-annual scale.

The reliability of the different methods and climate parameters reconstructed is evaluated with bootstrapping cross-validation through two indicators: the correlation coefficient between the variables (R2) and the root mean square error (RMSE).

The pollen diagram shows the vegetation dynamics between 196.6 and 127.5 ka BP, spanning late MIS 7 to early MIS 5 (Fig. 3). Five pollen zones were separated by CONISS cluster analysis, with zone 4 being divided in three subzones (Table 2).

Zone 1 most represented taxa are deciduous Quercus and Cedrus, with important abundance variability showing an unstable phase at the transition from MIS 7 to MIS 6e and 6d. Zone 2 displays the maximum expansion of steppe and semi-desert vegetation together with other open vegetation taxa and Cupressoideae during the MIS 6b and 6c. An additional noteworthy observation within this interval is the presence of gastropod shells identified as Limacina retroversa (Jeanne Rampal, personal communication, 2023), recovered during sieving of a sample at 46.4 m, corresponding to around 155.5 ka BP (Fig. 4). This species is usually most abundant in temperate to subpolar waters in the North-Atlantic (Thabet et al., 2015). Zone 3 is mainly characterized by the abundance of Ericaceae at the final stage of MIS 6 (6b and 6a). Zone 4, at the end of MIS 6a, is divided into 3 subzones which display fast vegetation changes during the transition from MIS 6 to MIS 5 (Termination II), and Heinrich Stadial 11. The fast expansion of steppe and semi-desert taxa (Artemisia, Amaranthaceae, Ephedra) occurs simultaneously with the first increase of deciduous temperate and Mediterranean forest indicative of the initialization of interglacial conditions (zone 4a). This episode of arid vegetation dominance is interrupted in zone 4b by the fast expansion of montane vegetation mainly represented by Cedrus. A new steppe and semi-desert vegetation increase is observed in zone 4c, while the deciduous temperate forest and the Mediterranean vegetation continue to expand. Finally, zone 5 is characterized by the maximum abundance of mesophilous and thermophilous elements, mainly represented by deciduous Quercus and Quercus ilex, typical of the MIS 5 interglacial.

Results show significant temperatures and precipitation variations in connection with the glacial/interglacial cyclicity and shorter-term variability (Fig. 5, Table 2). The most reliable methods according to the two R2 and RMSE indicators are MAT and BRT, and the most accurately reconstructed parameters are MAAT and MTCO (Table 3). The four methods are in agreement for the general trends, although MAT shows the widest amplitude of variations, and RF has the smoothest curve.

Temperatures are lower than the present during the complete MIS 6 interval, except at the onset of MIS 5. A cooling trend is reconstructed during the final stage of MIS 7 (pollen zone 1, MIS 6e and 6d), while MTCO and the seasonal temperature contrast (TCON) are stable. The methods do not agree on the precipitation patterns during this phase, with MAT showing a trend toward aridity, decreasing WINTERPR and seasonal precipitation contrast (PCON), while the three other methods display a slight precipitation increase and stable PCON. From 166 ka BP onward (pollen zone 2, MIS 6c and 6b), both temperatures and precipitation decrease, and the seasonal contrast between winter and summer climate conditions reduced progressively. The MIS 6 minimum temperatures and precipitation are reconstructed in pollen zone 2 between around 150 and 160 ka BP, corresponding to the transition between MIS 6c and 6b. Subsequently, both temperatures and precipitation increase progressively in the late pollen zone 2 and early pollen zone 3 (MIS 6b to 6a). Between 140–135 ka BP (early MIS 6a), the four methods reconstruct temperatures similar to late MIS 7, a seasonal contrast close to the present, and precipitation higher than the present (except for BRT). This climate optimum is abruptly interrupted by the HS11 extreme arid event between ∼ 134 and ∼ 129 ka BP (pollen zone 4, late MIS 6a), during which temperatures, precipitation and seasonal contrasts are significantly reduced, reaching climate conditions similar to the MIS 6 glacial maxima ∼ 155 ka BP. A short climate amelioration is evidenced at ∼ 132 ka BP (pollen zone 4b), where precipitation and seasonal contrasts increase abruptly. Finally, after 130 ka BP the climate amelioration toward the MIS 5 interglacial conditions happens very fast (pollen zone 5).

Important changes are recorded during MIS 6, that are consistent with orbital-scale variability during the different glacial substages.

Three phases can be discerned. The early phase (pollen zone 1) spans late MIS 7, MIS 6e and 6d (∼ 196–166 ka) and is characterized by high percentages of deciduous Quercus, Cedrus and Cichorioideae, with marked variability and several abrupt semi-desert and steppe increases under cool and humid climate conditions, with seasonal contrast similar to present-day. The middle phase (pollen zone 2) extends from MIS 6c to late 6b (∼ 165–143 ka), and displays the maximum expansion of steppe and semi-desert taxa together with very low temperatures and precipitation between ∼ 160 and ∼ 150 ka, a chronology compatible with the maximum Drenthe ice advance (Ehlers et al., 2011). Finally, the late phase (pollen zone 3), spanning late MIS 6b and 6a (∼ 143–133 ka), is marked by a major expansion of Ericaceae vegetation associated with higher reconstructed precipitation and winter temperatures, as well as enhanced seasonal contrast during this phase.

Previous studies are consistent with a subdivision of MIS 6 into three phases characterized by different general trends and amplitude of millennial-scale oscillations (Margari et al., 2014; Nehme et al., 2020). Margari et al. (2014) described an early phase between 185 and 160 ka BP, with warmer and wetter conditions and important rapid climate variability, a middle transitional phase between 160 and 150 ka BP, and a late phase with stable glacial conditions between 150 and 135 ka BP. This three-phasing for MIS 6 glaciation matches our interpretation of ODP 976 pollen zones 1, 2 and 3.

The final phase of MIS 6 is characterized by important changes in vegetation and climate, indicating a rapid modification of vegetation communities and atmospheric configuration at the transition between MIS 6 and MIS 5. Termination II (TII), defined as the period of fast reorganization of the climate system from full glacial (MIS 6) to full interglacial (MIS 5) conditions, is indeed characterized by extreme and fast internal dynamics including a major Heinrich Stadial, HS11 (Broecker and Henderson, 1998; Gouzy et al., 2004; Martrat et al., 2014; Moseley et al., 2015; Ovsepyan and Murdmaa, 2017). The timing for TII has been estimated based on the initialisation and termination of Weak Asian Monsoon evidenced in the Dongge cave speleothems, lasting from ∼ 136 to 129 ka BP (Bajo et al., 2020; Kelly et al., 2006; Menviel et al., 2019). These boundaries for TII give a total duration of about 7 ka. The timing of TII in the ODP 976 marine record is directly dependent on the Dongge Cave chronology (see methods, Sect. 3.1). Approximately the same duration is observed in the vegetation response to TII, but with ∼ 1 ka delay: the imprint of HS11 on the vegetation in the Western Mediterranean region is indeed recorded here between 133.3–128.8 ka BP (pollen zone 4). This delay in marine and terrestrial proxies may reflect the vegetation response to the first cold pulse of HS11. ODP 976 provides for the first time a very detailed record of vegetation successions during this arid event (pollen zones 4a, 4b and 4c), in agreement with other SSTs and speleothem records that depict a three-phases or “W” pattern for the event (see Sect. 5.4). After the first rapid increase of steppe and semi-desert taxa (pollen zone 4a), the middle phase shows an abrupt decrease of steppe and semi-desert vegetation, and a fast increase of montane trees (mainly Cedrus) percentages (pollen zone 4b). Climate reconstructions reflect this event through a fast increase of both precipitation and temperatures. This pattern is fully compatible with the ODP 976 SSTs trend (Jiménez-Amat and Zahn, 2015; Martrat et al., 2014), although a delay of about 1 kyr is observed between the abrupt drop in alkenone-based SSTs at the onset of HS11 and the expansion of steppe and semi-desert vegetation. In the same way, the abrupt sea surface warming in the middle of HS11 (∼ 133 ka BP) is shifted in the pollen record, to around 131.5 ka BP (pollen zone 4b).

Pollen analyses help us to characterize the processes behind Organic Rich Layers (ORLs) deposition in this western Mediterranean region. Like sapropels, ORLs reveal important changes in the water stratification and circulation, with reduced bottom water ventilation and enhanced organic productivity in direct connection with (i) increase in the freshwaters Atlantic inflow at times of deglaciation and (ii) enhanced rivers runoff regionally linked with increased precipitation (Murat, 1999; Pérez-Asensio et al., 2020; Rogerson et al., 2008). Although they are often considered as “ghost sapropels”, their timing and the mechanism behind them may differ from those of Eastern Mediterranean sapropels (Rogerson et al., 2008).

ORL bed 607 coincides with a period of enhanced precipitation around 176 ka reconstructed through our pollen-based approach (Fig. 7). Its basis appears almost synchronous with the onset of Sapropel S6 layer deposition in the Eastern Mediterranean (Emeis et al., 2003; Rohling et al., 2015; Savannah et al., 2024). Its duration also appears shorter than Sapropel S6, possibly indicating an interruption of favourable climate conditions in the Western Mediterranean region by a stadial event occurring around 172 ka BP and marked by an abrupt decrease in precipitation (Fig. 7).

ORL bed 606 was deposited during the second half of Termination II, at a time of deglaciation and directly following the first aridity pulse of HS11. Pollen-based reconstructions show enhanced precipitation and seasonal contrast during this time, suggesting intense precipitation together with deglacial freshwater input as combined causes for ORL deposition in the Western Mediterranean, which finds no counterpart in the Eastern Mediterranean. The implications of such organic layer deposition occurring at times of enhanced precipitation or deglaciation will be further discussed in Sect. 5.4.

The ODP 976 pollen record documents MIS 6 vegetation changes in the Western Mediterranean with a temporal resolution comparable to the most detailed terrestrial palynological sequences from the Eastern Mediterranean (Tenaghi Philippon and Ioannina). A W-E transect of Mediterranean palynological records offers valuable insights on the spatial pattern of vegetation changes during the penultimate glaciation (Fig. 6).

At the MIS 7–6 transition, no abrupt decline of temperate forest is recorded in the ODP 976 and Padul records, in contrast with central and Eastern pollen records such as Tenaghi Philippon, Ioannina, Ohrid and Castiglione where the transition is very abrupt (Follieri et al., 1988; Koutsodendris et al., 2023; Roucoux et al., 2011; Sadori et al., 2016). At these sites, a higher contrast has been described between interglacial periods with very high percentages of temperate deciduous forest taxa and glacial periods with very reduced tree cover (Tzedakis, 1993; Tzedakis et al., 2006). The western Mediterranean region, at the contrary, was generally characterized by a high proportion of herbaceous taxa, even during interglacial periods, attenuating the vegetation contrasts during transitions to glacial periods.

The first half of MIS 6 (∼ 185–165 ka BP) is marked by relatively high percentages of arboreal pollen across all records, especially deciduous forest (Roucoux et al., 2011; Margari et al., 2010, 2014). The abundance of montane taxa (mainly Cedrus) is characteristic of ODP 976 record, and reflects the development of altitudinal trees on the Moroccan mountains. Montane elements also increase during early MIS 6 at Valle di Castiglione, mainly represented by Fagus and Abies (Follieri et al., 1988), and at Ioannina, mainly represented by Pinus (Roucoux et al., 2011). No equivalent pattern is recorded in Padul where herbaceous vegetation is largely dominant. The scarcity of palynological data from the western Mediterranean, especially from North Africa, limits our understanding of the spatio-temporal significance of Cedrus expansions during MIS 6, and past glaciations in general.

The main phase of steppe and semi-arid vegetation is recorded between ∼ 165–145 ka BP in the Alboran Sea record, consistent with Ohrid and Ioannina pollen sequences (Fig. 6). In Padul, however, the maximum expansion of steppe and semi-desert taxa occurs later, between ∼ 155 and ∼ 137 ka BP. This time window corresponds to the glacial maximum recorded at Azzano X (northern Italy) between 148 and 135 ka BP (Pini et al., 2009), and to a period of low pollen concentration at Valle di Castiglione, likely reflecting full glacial conditions. The ∼ 10 ka lag in the vegetation glacial maxima between ODP 976 and Padul is probably the result of the low temporal resolution of the latter, in addition to differences in the age models used. However, the presence of pioneer vegetation and to a lesser extent of montane elements in Padul during the maximum glacial phase matches the ODP 976 pattern in pollen zone 2, where Cupressoideae and Cedrus display high abundance.

A distinctive feature of the Western Mediterranean vegetation recorded during the final stage of MIS 6 (pollen zone 3) is the marked increase in Ericaceae observed in the ODP 976 record, whereas Ericaceae are almost absent in the rest of the Mediterranean region. A similar expansion is observed in the Atlantic margin, where core MD01-2444 recorded patterns of Ericaceae expansion matching the three insolation minima during MIS 6 (Margari et al., 2014). Ericaceae development during minimal summer insolation is particularly favoured by reduced summer evaporation at times of low seasonal contrast in precipitation, as already noted in the Alboran Sea during the last glacial (Fletcher and Sánchez Goñi, 2008). Following the same interpretation, the expansion of heathland vegetation in the Iberian Peninsula as recorded in ODP 976 during the final stage of MIS 6 may therefore mark the renewed influence of westerlies and Atlantic moisture preceding the onset of the transition to MIS 5 interglacial (Margari et al., 2014). Supporting this state, an increase in temperate deciduous forest at the end of MIS 6 is also seen in Padul, Ohrid and Ioannina records before the transition to MIS 5.

Despite some discrepancies due to the differences in age models and temporal resolutions, all records display comparable variations in temperate pollen percentages during the penultimate glacial. These variations support the persistent sensitivity of Mediterranean plant ecosystems to global-scale millennial climate variability during the penultimate glaciation, with modulation of the vegetation response depending on the local geography. Strong similarities can be observed between ODP 976 and Padul, despite the lower temporal resolution of Padul record. In both sequences, temperate pollen percentages reached ∼ 30 % of total pollen as a maximum during the rapid forest expansions events in the first half of MIS 6. This similarity supports the validity of the ODP 976 marine record to reconstruct the SW Iberian Peninsula temperate forest history. However, ODP 976 sequence provides a more regional image of the vegetation, including higher percentages of Ericaceae pollen coming from the Atlantic coast, and Cedrus pollen from the Moroccan mountains (Jiménez-Moreno et al., 2020), compared to Padul where percentages of Mediterranean taxa and hygrophyte herbaceous are higher due to the local nature of the signal (Camuera et al., 2019). Looking further east, Valle di Castiglione recorded various temperate trees expansions and contractions during the lower MIS 6, before the full glacial conditions. In the Italian Peninsula, various interstadials have also been identified further north at Azzano X (Pini et al., 2009). In the Balkans, Tenaghi Philippon shows the highest percentages of semi-desert and herbaceous vegetation throughout MIS 6, with more abrupt changes than all the other Mediterranean records, and more amplitude of the trees' contractions. This pattern was already described during the last climatic cycle and reflects the exacerbated vegetation dynamics locally, with episodes of rapid and enhanced colonization by tree vegetation (Koutsodendris et al., 2023; Tzedakis, 2005; Tzedakis et al., 2004). This has been mainly explained by the location of the site in a low altitudinal plain characterized by a more continental climate with lower winter precipitation: tree populations at lower altitudinal location are closer to their ecological threshold in term of precipitation, and are likely to be very affected even by minimal changes in the amount of rainfall. On the contrary, the Ioannina record shows the highest deciduous forest percentages of all the records presented here, supporting its character as a local trees refugium (Roucoux et al., 2011).

Termination II displays a particular pattern in vegetation records from the Mediterranean region: while the expansion of trees and temperate vegetation is fast and continuous, HS11 represents at the same time a remarkable episode of abrupt steppe and semi-desert expansion. Although this event is visible in almost all the records, it is particularly prominent in the ODP 976 record (pollen zone 4), and appears less pronounced in the eastern Mediterranean sequences. This observation is compatible with previous observations that Heinrich stadials during the last glacial had a minor impact on the eastern Mediterranean vegetation compared to the western Mediterranean, likely due to the already limited presence of tree vegetation in the eastern records during glacials (Tzedakis, 2005). A similar interpretation can be proposed for the differential response of eastern and western Mediterranean vegetation to HS11, supporting the major sensitivity of the southwestern Mediterranean vegetation to North Atlantic cold events. In Padul, no major expansion of xerophyte vegetation is detected, but a small decrease of temperate deciduous taxa was interpreted as the HS11 imprint (Camuera et al., 2019), and the signal might be hindered by the low resolution of the record. A pattern of fast arid vegetation increase contemporaneous to the temperate forest expansion is also found in central Italy at Lago Grande di Monticchio, which was not presented in Fig. 6 as its record does not extend beyond 132 ka BP (Allen and Huntley, 2009; Brauer et al., 2007).

Finally, all palynological sequences reveal high-frequency oscillations of temperate and semi-desert pollen, compatible at first look with DO-like variability based on their duration and intensity. They represent a particularly distinctive feature of the lower part of MIS 6.

In order to investigate the character of rapid climate variability during MIS 6, a comparison with regional and global climatic archives is essential. The events of arboreal pollen increase observed in the ODP 976 record show percentages and timing comparable to those from the Portuguese margin core MD01-2444 (Margari et al., 2010, 2014; Tzedakis et al., 2018) (Fig. 7m). However, the ODP 976 record generally presents lower values of temperate deciduous pollen percentages compared to the Atlantic record, due to its more semiarid Mediterranean influence as previously evidenced for the last glacial period (Charton et al., 2025; Fletcher et al., 2010). To better capture temperate vegetation dynamics, we added Ericaceae, a clear marker of Atlantic influence in the ODP 976 record, to the deciduous temperate forest, to obtain a “total temperate pollen sum” which enhances the main warming peaks and strengthens the correlation between the two marine cores on both sides of Gibraltar Strait (Fig. 7m and l). The ODP 976 pollen-inferred climate reconstructions show generally consistent patterns with the SSTs trends based on alkenones (Martrat et al., 2004, 2007), and the southern Iberian humidity recorded in the speleothem from Cueva Gitana (Hodge et al., 2008) (Fig. 7e and g). The ODP 976 pollen and climate record therefore appears to reflect well regional variations in both temperature and humidity across MIS 6.

Warm events in the northern hemisphere are generally well-correlated to peaks in the ODP 976 temperate pollen curve (Fig. 7c and l). An active bipolar seesaw dynamics was described during the penultimate glacial (Davtian and Bard, 2023; EPICA Community Members, 2006; Stocker, 1998), and the Antarctic record was used to elaborate the Greenland GL_T-syn (Greenland temperature synthetic) curve showing predicted δ18O millennial-scale events for the past glacial eight climatic cycles, which are not directly recorded in Greenland ice (Barker et al., 2011; Bazin et al., 2013; Jouzel et al., 2007). Six Antarctic Isotopic maxima (AIM) events were recognized on the Deuterium curve during MIS 6 (6i to 6vi), correlated with increases in CO₂ concentrations and benthic isotope minima in the North Atlantic (Barker et al., 2011; Hodell et al., 2023; Margari et al., 2010, 2014; Shin et al., 2020) (Fig. 7a–c). These AIM and benthic minima in the Atlantic are not easily correlated with steppe expansions in the ODP 976 record, indicating a limited response of vegetation in the Western Mediterranean to the Antarctic warm events.

Barker et al. (2011) predicted the occurrence of eleven millennial-scale warming events during MIS 6 (Fig. 7c), while nine interstadials were recognized in the Alboran Sea from the alkenone record (Fig. 7e), (Martrat et al., 2004, 2007). In the loess record of Harletz in central Europe, ten interstadials were described (Rousseau et al., 2020), strongly matching the Chinese speleothems records of stadial and interstadial events related to the Asian Monsoon dynamics (Cheng et al., 2006; Li et al., 2014; Wang et al., 2018, 2001; Xue et al., 2019). The global nature of fast climate oscillations in the northern hemisphere thus appears controlled by the coupled influence of Atlantic cold events, and tropical monsoon variations as evidenced by the eastern Mediterranean speleothems records from Sofular, Soreq and Kanaan caves (Ayalon et al., 2002; Held et al., 2024; Matthews et al., 2021; Nehme et al., 2018).

The three phases identified during MIS 6 based on the ODP 976 vegetation and climate record can be compared with regional and global records to be interpreted in a broader context, based on general climatic trends and the expression of climatic instability.

Various studies have pointed out strong similarities between the millennial-scale oscillations of the last glacial period and the penultimate glacial period, with the division between MIS 3 and MIS 2 being analogous to the early and mid-late phase of MIS 6 respectively (Held et al., 2024; Margari et al., 2010, 2014; Roucoux et al., 2011; Rousseau et al., 2020; Shin et al., 2020; Sierro et al., 2020). The same studies argued in favour of pervasive impact of stadial events on the continental climate and vegetation in the Mediterranean region, even in absence of typical Heinrich layers (Roucoux et al., 2011). The ODP 976 record shows a cooling and aridification trend during the first half of MIS 6 (Fig. 9), with decreasing intensity of interstadials events, that recalls the pattern of MIS 3 D–O cycles (Bond et al., 1993).

However, the absence of clear successions of stadial events and especially Heinrich stadials, together with the more subdued expression of interstadials in the vegetation record, limits the resemblance between the two glacial periods. The pacing of interstadial peaks also seems to be reduced compared to the last glacial period high-frequency oscillations, as previously highlighted from the high-resolution speleothem record from Sofular cave in Turkey (Held et al., 2024).

A comparison of millennial-scale changes during the past two glacial periods based on the ODP 976 and MD95-2040 records, on either side of the Gibraltar Strait, supports our view (Fig. 9). The last glacial period (encompassing MIS 4 to MIS 2) was characterized in the Alboran Sea by high-intensity oscillations in both temperate and semi-desert vegetation correlated with D–O cycles and intense ice-rafting events HE1 to HE7 in MD95-2040. During interstadial events, temperate and Mediterranean vegetation (deciduous forest + Mediterranean + Ericaceae) could reach values above 60 % of total pollen; during stadial events, the semi-desert pollen values reached values as high as 70 % of total pollen (during HS3, HS4 and HS5). In comparison, the penultimate glacial (MIS 6) displays much lower intensity events, with interstadials characterized by 45 % as a maximum value for temperate vegetation, and stadials with 65 % for the steppe and semi-desert vegetation (during HS11). High-intensity cold episodes during MIS 6 are limited to the HS11, and the ∼ 172 ka BP event. This is consistent with the multiproxy record of core MD95-2040 on the Portuguese margin, which evidenced reduced variability in the N. pachyderma abundance and IRD deposition during the penultimate glacial compared to the last glacial (de Abreu et al., 2003; Voelker and de Abreu, 2011). The ice rafting episodes appear to be of different nature during MIS 6 (Hodell et al., 2008; Liu et al., 2018; McCarron et al., 2021), with the main iceberg discharges originating from the European ice sheet, contrary to the typical Hudson Strait origin of the last glacial Heinrich layers. SST reconstructions in the western Mediterranean also show less intense cooling during MIS 6 than during MIS 3 (Martrat et al., 2004, 2007), supporting limited incursions of polar waters in the Mediterranean during MIS 6 compared to MIS 3 coldest stadials, and especially Heinrich stadials (Cacho et al., 1999).

Like ODP 976, Ioannina records lower intensity arboreal pollen oscillations during early MIS 6 compared to the last glacial (Roucoux et al., 2011). In comparison, the Atlantic pollen record from MD01-2444 core displays similar amplitude of tree percentages during the last and the penultimate glacial (Margari et al., 2010). This difference can be explained by the different climate conditions, and the higher sensitivity to cold and aridity of sclerophyllous and deciduous forest vegetation on the Mediterranean side, as recorded in the ODP 976 and Ioannina palynological sequence. It appears that temperate vegetation in SW Mediterranean responded to millennial-scale climatic oscillations with higher intensity during the last glacial compared to the penultimate, probably because the climate in Europe was colder during MIS 6 compared to MIS 2. This is supported by larger European ice-sheet extension during the penultimate glacial (Ehlers et al., 2011; Ehlers and Gibbard, 2007; Shackleton, 1987), favouring the long-term establishment of open landscapes mainly composed by steppe and semi-desert plants. The differences in humidity might not be as easily interpretable, with an early MIS 6 more humid, and a MIS 6 glacial maximum more arid, compared to MIS 3 and MIS 2 as also suggested by the Ioannina record (Roucoux et al., 2011). Future climate reconstructions applied to the complete last glacial cycle in ODP 976 and other Mediterranean long pollen sequences will help understanding the different climate configurations between the last two glacial periods.

Only a limited number of sites in South-Western Europe has yielded archaeological layers attributed to MIS 6, and even fewer of them have been radiometrically dated allowing for a robust comparison with the environmental changes (Fig. 10 and Table S2). The environmental proxies available in the archaeological layers (pollen, charcoal, macro and microfauna) can help the chronological attribution, but are often insufficient to establish a precise correlation with the high-resolution chrono-environmental framework of marine and glacial archives. Even when absolute dates are available, their large uncertainty range and the poor resolution of the archaeological record represent a major limitation and make it difficult to correlate the human occupation phases with a specific substage of MIS 6.

It is generally accepted that the northern part of Europe was almost completely depopulated during MIS 6, with very few sites identified compared to the southern European fringes, indicating discontinuous occupation during more favourable climatic episodes (Hérisson et al., 2016) or total abandonment like in the British lands (Scott, 2011; Shaw et al., 2016; White and Pettitt, 2011). Southern France, Italy and the Iberian Peninsula could have represented climate refugia during the most extreme ice-cap advances (Bicho and Carvalho, 2022). Notably, Italy is particularly deprived of sites well-dated to MIS 6 (Fig. 10) including the isolated Neanderthal of Altamura, the short episode of elephant scavenging at Poggetti Vecchi, and the long sequence of San Bernardino cave which chronological range extends up to ∼ 154 ka BP, a period marked by the most extensive glacial conditions of MIS 6. Some other few archaeological layers have been attributed to MIS 6, but they lack a robust chronological attribution (Aureli and Ronchitelli, 2018; Fontana et al., 2010; Fig. 10). One can hypothesise that regional climate conditions in the peninsula were particularly harsh after 150 ka, and that potential refugia sites remain to be identified in Italy. Palaeoecological reconstructions at the Poggetti Vecchi site indicated cold and dry open environment (Aranguren et al., 2019; Benvenuti et al., 2017). Interestingly, the chronological range for the site could coincide with a major stadial event at 171 ka BP identified in the ODP 976 core, and particularly well expressed in the Valle di Castiglione record (Fig. 6).

In Southern France and the Iberian Peninsula, according to available radiometric dates, human occupation appears to have been continuous across MIS 6, even during the glacial maximum, with both cave and open-air sites. France provides a comparable number of cave and open-air sites mainly concentrated in the southwestern region. The Portuguese record is mainly constituted by open-air sites in fluvial terrace systems of the lower Tagus, offering important insights into short-term occupations during the full-glacial stage, but with complex chronological attribution (Cunha et al., 2012, 2017; Pereira et al., 2019). The Spanish record includes various open-air settlements in the upper Tagus valley (Panera et al., 2011, 2014; Yravedra et al., 2019), as well as the Duero (Diez-Martín, 2010) and the Guadalquivir (Caro Gómez et al., 2011) valleys. Cave sites are fewer and are mainly located closer to the coast (Cueva del Bolomor, Cueva del Angel, Lezetxiki, Arlanpe, Ventalaperra), with the two exceptions of Cueva de Maltravieso and Villacastín. Key sites like Lazaret Cave (Late Acheulean, France) and Cueva del Bolomor (Middle Palaeolithic, Spain) evidence the persistence of human groups in possible climate refugia (Ochando et al., 2019; Valensi et al., 2005).

MIS 6 in Europe saw the final stage of the cultural transition from the Lower to the Middle Palaeolithic industries (MIS 8–5), mainly characterized by the emergence of more complex core technologies such as Levallois debitage and changes in subsistence strategies. No rupture is observed between the technocomplexes, as cultural diversity and the permanence of Acheulean bifacial tools associated to technological innovation mark these Early Middle Palaeolithic industries in Southern Europe (Santonja et al., 2016; Terradillos-Bernal et al., 2023). The distribution of archaeological sites and timing of human occupation in South-Western Mediterranean at that time reflect this pattern (Fig. 10). A mosaic of traditional and innovative behavioural traits can be observed, with late Acheulean and Early Middle Palaeolithic coexisting continuously (Cueto et al., 2016; de Lumley, 2018; Mathias et al., 2020; Moncel et al., 2025; Santonja et al., 2022; Torres et al., 2024; Valensi et al., 2013). Acheulean technocomplexes are progressively abandoned during MIS 6 in Europe (Álvarez-Alonso, 2014; Key et al., 2021), with the latest chronologies found possibly in the Manzanares basin in central Iberia at Arriaga sites (Panera et al., 2014; Rubio-Jara et al., 2016; Rubio-Jara and Panera, 2019; Silva et al., 2013), or at Lazaret cave (Michel et al., 2022), and dated to the beginning of MIS 5. No clear explanation is accepted for the emergence and generalization of the Levallois debitage, and while cognition might not be the only factor, some authors suggested that MIS 6 glaciation could have played a role in the final abandonment of Acheulean industries (Moncel et al., 2020; Valensi et al., 2005). It is hard to claim that specific environmental pressures favoured Levallois technology over bifacial production, as these lithic technologies seem to have co-existed in Western Europe since MIS 12–11 over several glacial/interglacial cycles (Baena et al., 2017; Moncel et al., 2020), including extremely cold stages (like MIS 12 and 10). This “mosaic” pattern for the Lower to Middle Palaeolithic transition, although at least partly imputable to the large dating uncertainties, points toward more complex processes leading to the generalization of Middle Palaeolithic industries from MIS 5. Interestingly, the end of the Lower to Middle Palaeolithic transition is also associated with a shift in the morphology of human remains, from “Early Neanderthals” (MIS 7–5) to “Classical Neanderthals” (MIS 5–3) (Di Vincenzo and Manzi, 2023). Sites like La Chaise (Abri Suard), Lazaret and Altamura provide fossil evidence for these “Early Neanderthals” which share characteristics with earlier Middle Pleistocene populations, and with later Neanderthals (Buzi et al., 2025; Couture-Veschambre et al., 2021; de Lumley, 2018). Genetic data also support an important population shift in western Europe sometimes around the transition from MIS 6 to MIS 5 (Peyrégne et al., 2019). Therefore, an important population reorganization seems to have occurred at the time of the final Acheulean industries, leading to the onset of the so-called “Classical neanderthal world” in western Eurasia, with generalized Middle Palaeolithic industries and established Neanderthal morphological features. The role of environmental changes occurring during MIS 6 in this population reorganization remains poorly understood.

Changes in land use and mobility pattern have been evidenced in north-central Iberia, and can be viewed as adaptations to the severe climatic conditions of MIS 6 evidenced in the ODP 976 sequence during pollen zone/phase 2: increasing mobility, more short-term occupations and reliance on more local resources for subsistence strategies (Diez-Martín, 2010; Diez-Martín et al., 2008; Rios-Garaizar, 2016; Sánchez-Yustos, 2009). However, identifying cultural phases in the archaeological sequences linked with specific climatic episodes is generally hindered by the poor resolution of the archaeological record and chronological data. Among the sites identified in this synthesis, Lazaret and Bolomor caves probably present the most informative and well-dated sequences with several archaeological layers dated to MIS 6.

Cueva del Bolomor stands out in the Iberian Peninsula record as it provided an exceptionally long and continuous record of human presence, and the oldest evidence of fire use in Spain during the Middle Palaeolithic (Vidal-Matutano et al., 2019). Climate changes during MIS 6 documented in Bolomor's sediments through multiple proxies are consistent with the different phases identified in the ODP 976 record, with a more humid and cool phase at the beginning of MIS 6, and the most arid phase taking place at the middle of Phase III (layers X–VIII) (Arsuaga et al., 2012; Fernández Peris et al., 2008). The site is described as a climatic refugium where Mediterranean vegetation persisted during the colder phase of MIS 6 thanks to the coastal reservoir character of the site (Ochando et al., 2019). No clear change in lithic production has been identified in the MIS 6 layers of Bolomor: according to Fernández Peris et al. (2008), archaeological layers XII to VII (Phase III, MIS 6) are all dominated by limestone flakes with few retouches, few recycling, and the presence non-Acheulean macro-lithic elements. These layers are characterized by expeditive flaking (including Levallois débitage) relying on local raw-material, pointing toward a high degree of mobility and search for immediate effectiveness. It is thus hard to distinguish different techno-cultural tendencies during this phase. The most visible change in the archaeological sequence occurs in layer VI (MIS 5) which shows an intensification of lithic production dominated by flint, the production of more specialized tools including microlithic elements and associated to more intense and stable occupation in the cave (Fernández Peris et al., 2008). Therefore, a clear behavioural change in the technological and economical exploitation of raw materials is identified at the beginning of MIS 5. Faunal remains show a large and constant diversity, including abundant micro supporting both short-term and long-term not-specialized occupation, with no clear change in the site's function across the sequence (Blasco et al., 2013).

Lazaret Cave, in south-eastern France, shows a very distinct scheme: the Lower to Middle Palaeolithic transition is well documented in the archaeological sequence (Unit CII and CIII), with the progressive replacement of large bifacial tools production by more standardized and smaller flakes (Cauche, 2012). Levallois débitage is already present and well-mastered in the lower MIS 6 levels, although rare, and becomes dominant in the upper levels (early MIS 5). Therefore, a subdivision of unit CII can be made with a lower interval rich in handaxes, and an upper layer characterized as “final Acheulean” with rare handaxes and more abundant flakes. Faunal assemblages are very constant throughout MIS 6, with the large dominance of red deers and rare presence of cold species (Rangifer tarandus, Coelodonta antiquitatis). Thus, the climatic oscillations of MIS 6 do not seem to have influenced different hunting strategies or prey selection by human populations at Lazaret Cave, as large game hunting of red deer prevailed (Valensi et al., 2013).

Therefore, Lazaret and Bolomor caves are examples of different strategies of site exploitation and technological evolution during MIS 6: Lazaret Cave represents a specialized red-deer hunting camp evidencing a progressive change from Lower to Middle Palaeolithic tools, while Cueva del Bolomor can be characterized as a short-term camp with more generalized hunting and stable expeditive Middle Palaeolithic industry. Despite these differences both sites provide evidence for a change in lithic assemblages occurring at the end of MIS 6/beginning of MIS 5: the disappearance of handaxes and generalization of Levallois flakes in Lazaret, and the intensification and standardization of flint flakes in Bolomor.

The fast climate dynamics during Termination II as evidenced in the ODP 976 paleoenvironmental record could have represented a critical period for human population. At the end of MIS 6, more sites have been identified in the Iberian Peninsula than Southern France, showing the latter could have represented a climate refugium at the time of maximum glacial expansion, with more intense human occupation regionally. Many of these late MIS 6 sites present a chronological boundary at the top of the sequence compatible with the onset of Termination II around 136 ka BP, and with HS11, within the dating uncertainty: Matalascañas, Tarrazona, Los Aljezares, Bolomor Unit III, Almonda, Pegos do Tejo, Puente Pino, Arlanpe Unit SQ2, Lazaret Unit CII, Payre layer D, Vauffrey unit IV, and Pech de l'Azé II layer 5. The extreme character of this event in the South-Western Mediterranean as expressed in the ODP 976 sequence could have put further environmental pressure on hominin groups already diminished. A niche modelling approach based on 41 sites of Western Eurasia since 145 ka BP has shown that the projected potential niche space for Neanderthals at the end of MIS 6 (∼ 145 ka BP) was very reduced and concentrated in Western Europe (Yaworsky et al., 2024). This coincides with the end of pollen zone/phase 2 in the ODP 976 record, the most cold and arid phase of the penultimate Glacial before HS11. Then, the authors reconstruct a progressive expansion of Neanderthal potential niche space between 145 and 130 ka, compatible with the climatic warming and moistening during pollen zone/phase 3 in the ODP 976 record (Yaworsky et al., 2024). The temporal resolution of the model (1000 years) does not allow to detect the impact of HS11 on the niche projection, and only a small slowdown and decline of the projected niche is visible at ∼ 130 ka BP, before the MIS 5e optimum (Yaworsky et al., 2024; Fig. 5). A regional study focused on North-Western Spain argued in favour of a demographic vacuum at the end of MIS 6, compatible with HS11 and leading to a population reorganization implying contraction or micro-extinction, before the generalization of Middle Palaeolithic industries during MIS 5 (Sánchez-Yustos and Diez-Martín, 2015). According to the same authors, following this crisis, Neanderthal population entered a “reorganisation phase” leading to demographical stability and more technological standardization, visible in the explosion of the number of sites in Europe in general, especially after the MIS 5e climatic optimum (Bringmans, 2007; Lewis et al., 2011; Wenzel, 2007). This statement is supported by niche modelling which shows a peak in projected potential niche space of Neanderthal during MIS 5e (Yaworsky et al., 2024), and by recent genetic data which provided evidence for at least two radiation events linked with the environmental conditions of the last interglacial (Vernot et al., 2021), and Neanderthal population continuity since ∼ 120 ka BP (Peyrégne et al., 2019). Thus, HS11 did not lead to complete extinction of hominin groups but might have induced deep demographical and technological reorganization, representing the first and one of the most intense abrupt changes that Neanderthal population had to face in South-Western Europe before the Last Glacial largest oscillations (HS6–4). According to this view, the emergence of the “classical Neanderthal” world in Europe after the MIS 6/5 transition corresponds to the initialization of dynamics of repeated population contraction and expansion in response to the Upper Pleistocene instability, especially during MIS 4–2 (Sánchez-Yustos, 2009). This view is increasingly supported by genetic data (Fotiadou et al., 2026). In that sense, the subdued environmental instability during MIS 6 evidenced in the ODP site 976 record compared to the last glacial period (Sect. 5.5 and Fig. 9) could also have implications for human populations, with less fragmented (although harsh) habitats and more stable (although reduced) population during MIS 6. This hypothesis remains however hard to test based on the very different nature and quality of preservation of the archaeological record during MIS 6 compared to MIS 4–2 (e.g. Charton et al., 2025).