Within the penultimate interglacial complex, the three pronounced steppe-forested intervals PAS Vc (113.7–109.1 mcblf; 242.5–227.4 ka), PAZ Va3 (104.2–101.3 mcblf; 216.3–207.6 ka), and PAZ Va1 (99.9–97.0 mcblf; 203.1–193.4 ka) can be broadly correlated with MIS 7e, 7c, and MIS 7a after Lisiecki and Raymo (2004), indicating high moisture availability and/or warmer temperatures (Figs. 2a, 3f).

The oldest terrestrial warm phase (242.5–227.4 ka; PAS Vc, MIS 7e) starts with the colonization of open habitats by pioneer trees, such as Betula, followed by deciduous Quercus and sclerophyllous Pistacia cf. atlantica. The occurrence of the frost-sensitive Pistacia, as a characteristic feature at the beginning of interglacials in the eastern Mediterranean region, indicates relatively mild winters, but also firmly points to the presence of summer aridity due to a higher temperature and evaporation regime (Litt et al., 2014, 2009; Pickarski et al., 2015a; Wick et al., 2003). Similar to the Holocene, the early interglacial spring and summer dryness might be responsible for the delay between the onset of climatic amelioration and the establishment of deciduous oak steppe-forest as the potential natural interglacial vegetation in eastern Anatolia. Here, the length of the delay depends on local conditions in keeping moisture availability below the tolerance threshold for tree growth in the more ecologically stressed areas. Indeed, a reduction in spring rainfall and the extension of summer-dry conditions favored the rapid development of a grass-dominated landscape (mainly Artemisia, Poaceae; Fig. 2b). Furthermore, the fire activity rose at the beginning of each warm phase when the global temperature increased and the vegetation communities changed from warm productive grasslands to more steppe-forested environments. Increased fire frequency is clearly visualized by a high charcoal concentration of up to 3000 particles cm-3 (Fig. 3e). After TIII at 243 ka, the vegetation change towards more steppe-forest environments correlates with depleted (negative) δ18Obulk values, which occur at the beginning of the early temperate stage (ca. 242–240 ka; Fig. 3c). As discussed earlier, depleted isotope values reflect intensified freshwater supply into the lake by melting Bitlis glaciers in summer months, favoring high detrital input into the basin (low Ca / K ratio; Fig. 3d) and/or enhanced precipitation during winter months (Kwiecien et al., 2014; Roberts et al., 2008).

The climate optimum of the first warm phase is characterized by the significant expansion of temperate summer-green taxa, mainly deciduous Quercus (above 20 % between ca. 240 and 237 ka), Pistacia cf. atlantica, Betula, and sporadic occurrence of Ulmus. The vegetation composition documents a warm temperate environment with enhanced precipitation during the growing season, which can be supported by depleted isotope values (δ18Obulk -2.17 ‰; Fig. 3c). Charcoal maxima (> 3000 particles cm-3) correlate, coeval with the delayed expansion of steppe-forest, with more fuel for burning. The gradual shift from depleted to enriched isotope values (δ18Obulk 5.15 ‰) indicates a change towards climate conditions with high evaporation rates and/or decreased moisture availability (Kwiecien et al., 2014; Roberts et al., 2008). Here, positive δ18Obulk values at Lake Van are attributed to the evaporative 18O enrichment of the lake water during the dry season. Furthermore, Kwiecien et al. (2014) described the relation between soil erosion processes and vegetation cover in the catchment area. They defined interglacial conditions related to increased precipitation indicated by a higher amount of arboreal pollen and lower detrital input. Our new high-resolution pollen record validates their hypothesis with a high authigenic carbonate concentration (high Ca / K ratio, low terrestrial input) along with increased terrestrial vegetation density (high AP percentages above 50 %) during the climate optimum (Fig. 3).

The ensuing ecological succession of the first warm stage is documented by a shift from deciduous oak steppe-forest towards the predominance of dry-tolerant and/or cold-adapted conifer taxa (e.g., Pinus and Juniperus; ca. 237–231 ka). High percentages of Pinus suggest a cooling and/or drying trend, which occurred during low seasonal contrasts (low summer insolation and high winter insolation; Fig. 3). Pinus (probably Pinus nigra) as a important arboreal component of the “Xero-Euxinian steppe-forest” occurs mainly in more continental western and central Anatolia and in the rain shadow of the coastal Pontic mountain range (van Zeist and Bottema, 1991; Zohary, 1973). Compared to the present distribution of Pinus nigra in Anatolia, the Lake Van region was probably more affected by an extended distribution area of pine during the penultimate interglacial as indicated by higher pollen percentages (Holocene below 5 %; PAZ Vc2 up to 26 %; PAZ Va3 up to 20 %; Fig. 4). Holocene pine pollen was mainly transported over several kilometers via wind into the Lake Van basin. Independent of environmental conditions around the lake, the presence of thermophilic algae (i.e., Pseudopediastrum boryanum) indicates warm and eutrophic conditions within the lake during the late temperate phase.

The presented regional vegetation composition can be described as an oak steppe-forest and marks one of the longest phases of the penultimate interglacial complex, lasting 15 000 years, with a climate optimum between 240 and 237 ka (Fig. 4c). However, this optimum does not appear to be of very high intensity as suggested by the lower development of temperate plants compared to the following warm phase.

The second terrestrial temperate interval (end of PAS Vb and PAZ Va3; 106.5–101.3 mcblf; ca. 221–207 ka; MIS 7c) starts with a shift from cold and arid desert-steppe vegetation (e.g., Chenopodiaceae) to less arid grassland vegetation (e.g., Poaceae, Artemisia; Fig. 2b). This was followed by an expansion of Betula and a high abundance of deciduous Quercus, and it continued with increased Pinus percentages. In this period, the occurrence of Pistacia cf. atlantica was not as pronounced as during the PAS Vc (MIS 7e), which can be explained by a lower winter insolation (cooler winters; Fig. 3b). Despite all this, the oxygen isotope signature displays similarly depleted values (δ18Obulk up to -3.8 ‰; Fig. 3c) at the beginning of the middle warm phase, right after TIIIA at 222 ka (Barker et al., 2011; Stockhecke et al., 2014a). In general, the second warm stage shows the highest amplitude of deciduous Quercus (peaked at 212.6 ka; Fig. 3f) of the entire sequence, which corresponds to the occurrence of the most floristically diverse and complete forest succession in southern European pollen diagrams at the same time (Follieri et al., 1988; Roucoux et al., 2008; Tzedakis et al., 2003b). In fact, deciduous Quercus percentages (ca. 56 %) reach the level of the last interglacial (MIS 5e) and the Holocene forested intervals, representing the most humid and temperate period during the penultimate interglacial complex at Lake Van (Fig. 4; Litt et al., 2014; Pickarski et al., 2015a).

Preliminary comparison with the pollen records of Tenaghi Philippon (Tzedakis et al., 2003b) and the Ioannina basin (Roucoux et al., 2008) suggests that the extent and diversity of vegetation development is clearly controlled by insolation forcing and associated climate regimes (high summer temperatures, high winter precipitation). At Lake Van, the interglacial forest expansion is closely associated with the timing of the mid-June insolation peak (Tzedakis, 2005). In general, Mediterranean sclerophylls and other summer-drought-resistant taxa expand during the period of maximum summer insolation, while thermophilous taxa are better suited to the less seasonal climates of the later part of the interglacial. Indeed, the highest expansion of deciduous Quercus occurs, coeval to Pinus, during the lowest seasonal contrasts (cooler summers and warmer winters). The different amplitudes in the deciduous tree development might have resulted from higher mid-June insolation at the beginning of PAZ Va3 (MIS 7c) relative to PAZ Vc4 (MIS 7e, similar to Holocene levels), despite lower atmospheric CO2 content (ca. 250 ppm, Fig. 5; Jouzel et al., 2007; Lang and Wolff, 2011; Petit et al., 1999; Tzedakis, 2005), thus mirroring the significant variability in regional effective moisture content and/or temperature.

After a short-term climatic deterioration between 207 and 203 ka, the spread of Pistacia cf. atlantica and Betula and the predominance of deciduous Quercus characterize the youngest warm phase PAZ Va1 (99.9–97.0 mcblf; 203.1–193.4 ka; MIS 7a) within the penultimate interglacial complex. Similar to the previous warm phases, the deciduous Quercus percentages (ca. 38 %) reach the level of the Holocene forested interval (deciduous Quercus ca. 40 %; Fig. 4). A possible explanation for the high thermophilous oak percentages within MIS 7a is the persistence of relatively large tree populations through the cold period equivalent to MIS 7b, which was also established in pollen records by Lac du Bouchet (Reille et al., 2000) and at the Ioannina basin (Roucoux et al., 2008).

All three forested stages of the penultimate interglacial complex are clearly recorded in other long terrestrial pollen sequences from Lebanon and southern Europe: (i) the Yammoûneh record (Gasse et al., 2015), (ii) the Tenaghi Philippon sequence (Tzedakis et al., 2003b), (iii) the Ioannina basin (Roucoux et al., 2008), and (iv) the Lake Ohrid sequence (Sadori et al., 2016). Figure 5 shows that the Lake Van pollen record generally agrees with the vegetation development of the Mediterranean region. However, we have to take into consideration that most southern European sequences, e.g., the Ioannina basin, are situated near refugial areas in which temperate trees persisted during cold stages (Bennett et al., 1991; Milner et al., 2013; Roucoux et al., 2008; Tzedakis et al., 2002). In these places where moisture availability was not limiting, the woodland expansion occurred near the glacial–interglacial boundary (Tzedakis, 2007). Despite this, high-resolution pollen records from the eastern Mediterranean region (e.g., Ioannina basin; Roucoux et al., 2008) suggest that the MIS 7 winter temperature during all three warm intervals seemed to be lower than during the Holocene and the last interglacial as indicated by smaller populations of sclerophyllous taxa. Reduced thermophilous components were also discussed for the Velay region (Reille et al., 2000), where the warm phases Bouchet 2 and 3, equivalent to MIS 7c and 7a, are described as interstadials rather than interglacials. This observation of a cooler MIS in southern Europe contradicts the vegetation development at Lake Van, where all warm intervals reach the level of the last interglacial and the Holocene. At Lake Van, there seems to be no reason to define MIS 7c and MIS 7a as interstadials separated from the MIS 7e interglacial.

The two periods between the three forested intervals, the first part of PAZ Vb (227–221 ka; 109.1–106.5 mcblf) and PAS Va2 (208–203 ka; 101.3–99.9 mcblf), are broadly equivalent to MIS 7d and MIS 7a (Lisiecki and Raymo, 2004). At Lake Van, cold periods are generally characterized by (i) extensive steppe vegetation when tree growth was inhibited either by dry and cold or low atmospheric CO2 conditions (Litt et al., 2014; Pickarski et al., 2015b), (ii) high dinoflagellate concentration (Spiniferites bentorii, which tolerates high water salinity conditions and suggests low aquatic bioproductivity; Fig. 2a), and (iii) high regional mineral input derived from the basin slopes (low Ca / K ratio; Kwiecien et al., 2014; Fig. 3d).

Due to the strongest development of extensive semidesert steppe plants (mainly Chenopodiaceae above 75 %) and a massive reduction in temperate trees (AP ca. 5 %; Fig. 2), the first cold phase suggests considerable climate deterioration and increased aridity. Furthermore, this period is marked by a large ice volume and extremely low global temperatures documented by low CO2 concentrations (∼ 210 ppm; Fig. 5) that are nearly as low as those of MIS 8 and 6 (McManus et al., 1999; Petit et al., 1999). Between 227 and 221 ka, the oxygen isotope record consistently displays δ18Obulk values above 0 ‰ that reflect dry climate conditions in the Lake Van catchment area (Fig. 3c). Such dry and/or cold periods within the entire penultimate interglacial complex can also be recognized in all pollen sequences from Lebanon and southern Europe (Fig. 5; e.g., Gasse et al., 2015; Roucoux et al., 2008; Tzedakis et al., 2003b). An exception is the Lake Ohrid record, which shows only a minor temperate tree decline (Sadori et al., 2016).

In contrast to conventional cold and dry periods at Lake Van, the second cold phase (PAS Va2) is recognized by only a slight and short-term steppe-forest contraction. Although the landscape was more open during the youngest phase, moderate values of Betula, deciduous Quercus (up to 16 %), and conifers (Pinus, Juniperus) formed steppe vegetation with still patchy pioneer and temperate trees. The significantly larger temperate AP percentages (ca. 20 %) during PAZ Va2 relative to PAZ Vb point to milder climate conditions. In addition, the continuously heavier oxygen isotope signature (δ18Obulk between 1.0–2.4 ‰) confirms the assumption of milder conditions with higher evaporation rates and more humid conditions. Based on these results, the Lake Van pollen record mirrored the trend seen in various paleoclimatic archives (Fig. 5). Indeed, several pollen sequences from the Mediterranean area and oxygen isotope records suggest that the North Atlantic and southern European region (e.g., Ioannina basin; Roucoux et al., 2008; Fig. 5d) did not experience severe climatic cooling during MIS 7b (e.g., Bar-Matthews et al., 2003; Barker et al., 2011; McManus et al., 1999; Petit et al., 1999). In addition, the global ice volume remains relatively low during MIS 7b in comparison with other stadial intervals with similarly low insolation values (e.g., Petit et al., 1999; Shackleton et al., 2000). The Vostok ice core sequence also records a relatively high CO2 content (ca. 230–240 ppm) during MIS 7d, supporting a slight decline in temperature compared with MIS 7d (CO2 content ca. 207–215 ppm; Fig. 5; McManus et al., 1999; Petit et al., 1999).

The direct comparison of the penultimate interglacial complex (MIS 7) with the last interglacial (Eemian, MIS 5e; Pickarski et al., 2015a) and the current interglacial (Holocene, MIS 1; Litt et al., 2009) provides the opportunity to assess how different successive climate cycles can be (Fig. 4).

In general, all interglacial climate optima were characterized by the development of an oak steppe-forest, all of which reached the level of the last interglacial and the Holocene, especially the extent of the temperate tree taxa. Such dense vegetation cover reduced the physical erosion of the surrounding soils in the lake basin. Furthermore, the dominance of steppe-forested landscapes and a productive steppe environment led to enhanced fire activity in the catchment area. In addition to these aspects, MIS 8–7e and MIS 7d–7c as well as the MIS 6–5e boundaries in the continental semiarid Lake Van region are recognized by a delayed expansion of deciduous oak steppe-forest of ca. 5000 to 2000 years. This is comparable to the pollen investigations in the marine sediment cores west of Portugal by Sánchez Goñi et al. (2002, 1999). As already shown in high-resolution pollen studies by Wick et al. (2003), Litt et al. (2009), and Pickarski et al. (2015a), a delay in temperate oak steppe-forest referring to the Pleistocene–Holocene boundary as defined in the Greenland ice core from NorthGRIP stratotype (for the Pleistocene–Holocene boundary; Walker et al., 2009) as well as from the speleothem-based synthetic Greenland record (GLT-syn; Barker et al., 2011; Stockhecke et al., 2014a) can be recognized. The length of the delay depends on the slow migration of deciduous trees from arboreal refugia (probably the Caucasus region) and/or changes in the seasonality of effective precipitation rates (Arranz-Otaegui et al., 2017; Pickarski et al., 2015a). In particular, oak species are strongly dependent on spring precipitation (El-Moslimany, 1986). A reduction in spring rainfall and the extension of summer-dry conditions favored the rapid development of a grass-dominated landscape (mainly Artemisia, Poaceae; considered as competitors for Quercus seedlings) and Pistacia shrubs in the very sparsely wooded slopes (Asouti and Kabukcu, 2014; Djamali et al., 2010). Furthermore, the high intensity of wildfires in late summer grasslands at the beginning of each warm period could be responsible for a delayed re-advance of steppe-forest in eastern Anatolia (Arranz-Otaegui et al., 2017; Pickarski et al., 2015a; Turner et al., 2010; Wick et al., 2003).

Despite the common vegetation succession from an early to late temperate stage, the three interglacial periods (MIS 7 complex, MIS 5e, and MIS 1) differ in their vegetation composition. One important difference in the last two interglacial vegetation assemblages is the absence of Carpinus betulus during MIS 7e, 7c, and 7a compared to a distinct Carpinus phase during MIS 5e (Pickarski et al., 2015a). In general, Carpinus betulus usually requires high amounts of annual rainfall (high atmospheric humidity) and a relatively high annual summer temperature, and it is intolerant of late frost (Desprat et al., 2006; Huntley and Birks, 1983). In oak–hornbeam communities, Carpinus betulus is replaced as the soils are relatively dry and warm or too wet (Eaton et al., 2016). Compared to the common hornbeam, deciduous Quercus species are “less” sensitive to summer droughts (even below 60 mm yr-1; Tzedakis, 2007), and therefore a decrease in soil moisture availability would favor the development of deciduous oaks (Huntley and Birks, 1983). The deep penetrating roots of Quercus petraea allow them to withstand moderate droughts by accessing deeper water (Eaton et al., 2016). However, a variation in temperature is difficult to assess because deciduous oaks at Lake Van include many species (e.g., Quercus brantii, Q. ithaburensis, Q. libani, Q. robur, Q. petraea) with different ecological requirements (e.g., San-Miguel-Ayanz et al., 2016). Finally, the absence of Carpinus betulus, the overall smaller abundances of temperate trees (e.g., Ulmus), and the generally low diversity within the temperate tree populations during the climate optimum of the first penultimate interglacial compared to the last interglacial indicates warm but drier climate conditions (similar to the Holocene). An exception is the second warm phase (MIS 7c), which reflects one of the largest oak steppe-forest developments (e.g., highest amplitude of deciduous Quercus) of the entire Lake Van pollen sequence and thus represents the most humid and temperate period within the penultimate interglacial complex (see discussion above).

Another important difference is the duration of each interglacial period. According to Tzedakis (2005), the beginning and duration of terrestrial temperate intervals in the eastern Mediterranean region is closely linked to the amplitude of summer insolation maxima and less influenced by the timing of deglaciation. Based on this assumption, the terrestrial temperate interval of all penultimate interglacial stages (max. 15.1 ka) is ∼ 4600 years shorter than the terrestrial temperate interval of the last interglacial at Lake Van (∼ 19.7 ka, Pickarski et al., 2015a; Fig. 4).

The occurrence of high-frequency climate changes within the Lake Van sediments provides an opportunity to compare the vegetation history of the last two glacial periods. Figure 6 illustrates that the first part of the penultimate glacial (ca. 193–157 ka) resembles MIS 3 regarding millennial-scale AP oscillations and the abruptness of the transitions in the pollen record. The series of interstadial–stadial intervals can be recognized in both glacial periods. This variability is mainly influenced by the impact of North Atlantic current oscillations and the extension of atmospheric patterns, in particular the northward shift of the polar front in eastern Anatolia (e.g., Cacho et al., 2000, 1999; Chapman and Shackleton, 1999; McManus et al., 1999; Rasmussen et al., 2014; Wolff et al., 2010).

The most distinct environmental variability occurred during MIS 6e (ca. 179–159 ka), which can be further divided into six interstadials based on rapid changes in the marine core MD01-2444 off Portugal (Margari et al., 2010; Roucoux et al., 2011; Fig. 6). They document abrupt climate oscillations below orbital cycles similar to the Dansgaard–Oeschger (DO) events or Greenland interstadials (GI) over the last glacial stage (e.g., Dansgaard et al., 1993; Rasmussen et al., 2014; Wolff et al., 2010). At Lake Van, MIS 6e reveals clear evidence of climate variability due to rapid alternation in abiotic and biotic proxies, such as oxygen isotopes, Ca / K ratio, and pollen data similar to the largest DO 17 to 12 during MIS 3 (ca. 60–44 ka; Pickarski et al., 2015b). Both intervals, MIS 6e and MIS 3, started at the point of summer insolation maxima. Here, the Northern Hemisphere insolation values reached the interglacial level at the beginning of MIS 6e comparable with MIS 7e (Fig. 5). In contrast, the interstadial–stadial pattern during late MIS 6 oscillated at a lower amplitude similar to the rates of change in the Dansgaard–Oeschger (DO) events during MIS 4 and 2, reflecting a general global climatic cooling.

Within MIS 6e, the subdued temperate tree pollen oscillations consist mainly of deciduous Quercus and Pinus, ranging between ∼ 1 and 15 %. In contrast, the identical AP composition oscillates between ∼ 1 and 10 % during the orbitally equivalent MIS 3 (ca. 61–28 ka; Pickarski et al., 2015b). The different amplitude in arboreal pollen percentages in both glacial stages and a generally dense temperate grass steppe during MIS 6e suggest more available moisture (Fig. 6). The depleted isotope signature may result from summer meltwater discharge from local glaciers (e.g., Taurus mountains, Bitlis Massif) or increased precipitation identified by climate modeling experiments over the eastern Mediterranean basin (e.g., Stockhecke et al., 2016). However, the presence of Artemisia and Poaceae makes it difficult to disentangle the effects of warming from changes in moisture availability in both glacials. Nevertheless, the abundance of Pinus, Ephedra distachya-type, and the cold-tolerant algae Pseudopediastrum kawraiskyi indicates colder and wetter climate conditions during MIS 6e compared to MIS 3.

Evidence of relatively humid but cold climate conditions during MIS 6e agrees with several other paleoclimate studies from the Mediterranean area. For example, the occurrence of open forest vegetation associated with a wetter climate is indicated at, e.g., Tenaghi Philippon (Tzedakis et al., 2006, 2003b) and Ioannina (Roucoux et al., 2011). In addition, isotopic evidence of the stalagmite record from Soreq Cave (Israel) shows enhanced rainfall (negative shift in the δ18O values) in the eastern Mediterranean at ∼ 177 ka and between 166 and 157 ka (Fig. 5; Ayalon et al., 2002; Bar-Matthews et al., 2003). Furthermore, a pluvial phase is also inferred from a prominent speleothem δ18O excursion in Argentarola Cave (Italy) between 180 and 170 ka based on U–Th dating (Bard et al., 2002). This phase coincides with maximum rainfall conditions during the MIS 6.5 event, coeval with the deposition of the “cold” sapropel layer S6 (ca. ∼ 176 ka) in the western and eastern Mediterranean basin (Ayalon et al., 2002; Bard et al., 2002). Finally, the progressive decline in effective moisture is a result of the combined effect of temperature, precipitation, and insolation changes in the Lake Van region.