The DEEP core chronology down to 247.8 mcd (Francke et al., 2016) is based on radiometric ages of 11 tephra layers (first-order tie points), and on tuning of biogeochemical proxy data to orbital parameters (second-order tie points; Laskar et al., 2004). The second-order tie points were obtained by tuning minima in total organic carbon (TOC) and TOC / TN against increasing summer insolation and winter season length. The timing of increasing summer insolation and winter season length caused cold and dry conditions in the Balkan Peninsula (Tzedakis et al., 2006; Francke et al., 2016), which may have led in Lake Ohrid to restricted primary productivity during summer and prolonged mixing and better decomposition of organic matter during winter. This likely resulted in low TOC and a low TOC / TN ratio (Francke et al., 2016). Finally, the age model for the sediment cores was refined by a comparison with the age model of the downhole logging data by Baumgarten et al. (2015). Correlation of the tephra layers with well-known eruptions of Italian volcanoes and a re-calibration of radiometric ages from the literature have been carried out by Leicher et al. (2015).

Sample processing and pollen microscope analysis are the fruit of strict cooperative work by several investigators across many European laboratories. Prior to the pollen analysis, considerable time was invested in assessing and standardizing the treatment protocol and pollen identification issues. More specifically, (1) we joined previous lists of taxa that were derived from older studies in Lake Ohrid and the western Balkans and produced a final list that has been accepted by all the analysts; (2) we thoroughly elaborated on systematic issues like synonyms and different degrees of pollen determination, particularly focusing on the identification of problematic taxa; (3) we shared pollen pictures of key taxa (e.g. oak types) and of dubious ones; (4) we also performed analyses of samples from the same core depth in different laboratories. Samples were mostly distributed in batches of consecutive samples; and (5) finally, close checks were performed at the intervals where two different analysts' samples met in order to avoid any potential identification bias.

A total of 306 sediment samples at 64 cm intervals down to the depth of 197.55 m taken from the DEEP core have been chemically processed for palynology in order to establish an overview diagram (named the skeleton diagram hereafter) spanning the past ∼ 500 ka. According to the age model by Francke et al. (2016), the mean resolution between two samples is ∼ 1600 years.

For each sample, 1/1.5 g of dry sediment was treated with cold HCl (37 %), cold HF (40 %) and hot NaOH (10 %). In order to estimate the pollen concentration, two tablets containing a known number of Lycopodium spores (Stockmarr, 1971) were added to each sample. To draw pollen percentage diagrams, different pollen basis sums (PS) have been used, following the criteria listed by Berglund and Ralska-Jasiewiczowa (1986). Terrestrial pollen percentages have been calculated excluding Pinus from the PS due to its high overrepresentation in a large number of samples. The Pinus percentage was calculated on a different pollen sum which includes pines.

Oak pollen has been divided into three types according to morphological features following Smit (1973): Quercus robur type, which includes deciduous oaks, Quercus ilex type including the evergreen oaks minus Q. suber, and Quercus cerris type, including semi-deciduous oaks and Q. suber. Further identifications follow Beug (2004), Chester and Raine (2001) and Reille (1992, 1995, 1998). Juniperus type includes pollen grains of Cupressus, Juniperus and Taxus. Pollen curves/diagrams (Fig. 2, 3 and 4) were drawn using the C2 program (Juggins, 2003). Ages are expressed in thousands of years BP (ka BP). Pollen zone boundaries were established with the help of CONISS (Grimm, 1987). Given the millennial temporal resolution of the skeleton diagram and considering the ongoing and planned high-resolution studies, we assigned 13 (i.e. OD-1 to OD-13) Pollen Assemblage SuperZones (PASZ, sensu Tzedakis, 1994a) that correspond to major shifts in glacial–interglacial vegetation. This approach allows for the definition of new pollen zones and subzones within these superzones as high-resolution (centennial) data from the Lake Ohrid archive will emerge.

Climate variability paces the pronounced intra-interglacial vegetational shifts inferred from the pollen record, while different patterns of ecological succession emerge during interglacials (Fig. 3).

Long-term vegetation dynamics correspond accurately to the glacial and interglacial periods, even if admittedly the established chronology for the Lake Ohrid DEEP record could be further improved with tuning to higher-resolution proxy data (see Zanchetta et al., 2015), with the detection of other tephra layers and the general improving of analyses obtained for the record.

In addition, most interstadials and some higher-order variability have been previously reported from south-eastern Europe, i.e. Ioannina (MIS6: Roucoux et al., 2011) and Tenaghi Philippon (MIS8: Fletcher et al., 2013). Ongoing high-resolution studies will help define dynamics of specific taxa, revealing extinctions and detecting possible new refuge areas.

A close look at the Lake Ohrid pollen record reveals distinct characteristics for glacial and interglacial phases during the investigated past 500 ka. Glacial periods are generally characterized by dominance of NAP (e.g. Poaceae, Chenopodiaceae and Artemisia). An exception to this behaviour is found during older glacial phases (OD-12, OD-11 and OD-9; Table 1) when Pinus pollen show high percentages and medium/high concentrations that appear reduced only at the end of OD-11 (Figs. 3, 4). Interglacial/interstadial periods are characterized by expansions of woodland organized in vegetation belts (e.g. forests with Abies, Picea, Quercus robur type, Q. cerris type) and by increases in AP–Pinus pollen concentration. This general pattern of glacial–interglacial alternations is at times punctuated by minor expansions of AP during glacials and accordingly by forest opening (stadials) during interglacial complexes. This is in agreement with previous studies from Greece, e.g. Ioannina (Tzedakis, 1994b; Tzedakis et al., 2002; Roucoux et al., 2008, 2011) and Tenaghi Philippon (e.g. Milner et al., 2012; Fletcher et al., 2013; Pross et al., 2015), and from central Italy (Follieri et al., 1998), suggesting a sensitive response of vegetation to climate change on a regional scale in south-eastern Europe. At Lake Ohrid, most tree taxa show a rather continuous presence, even during glacial phases, suggesting that the Ohrid region has been a plant refugium. The investigation of dynamics of specific taxa and time of extinctions and the detection of possible refuge areas are among the issues that must be refined by ongoing high-resolution studies.

A clear correspondence between the climate signals provided by our terrestrial pollen record and marine oxygen isotope records (Fig. 4) is apparent, even if the limits between pollen zones and marine isotope stages are often not identical (Figs. 2, 3).

Glacial periods (PASZ OD-12, 11, 9, 7, 5, 4, 2, Table 1) are generally characterized by dominance of Poaceae, Artemisia, and Chenopodiaceae that are indicative of open environments around the lake. Poaceae probably include aquatic macrophytes from the lacustrine belt and herbs from grassland formations in the catchment of Lake Ohrid. Artemisia and Chenopodiaceae, which are typically components of steppe–desert environments, consist of shrub and sub-shrub species. In OD-12/11 and OD-9, high percentages of Pinus can either point to the local presence of widespread thickets like those currently growing at very high elevations in the surroundings of the lake, or to transport from a long distance in a barren land. Another aspect to consider is that a large lake such as Ohrid could partially resemble the marine realm, leading to over-representation of pollen grains that float easily. But this should be a constant factor in the analysed records, unless big changes in the lake surface occurred. The available seismic data, not completely processed yet, suggest anyway (K. Lindhorst and S. Krastel, personal comments, 2015) that the lake size was not significantly different prior to 330 ka.

In contrast, interglacial complexes (PASZ OD-13, 10, 8, 6, 3 and 1, Table 1) are marked by expansions of woods dominated by Abies, Picea, the Quercus robur type and the Q. cerris type. This pattern is at times punctuated by minor expansions of AP during glacial periods and by forest opening during interglacial ones.

The pollen diagram shows that, in the past 285 ka (PASZ OD-7 to OD-1), non-forested periods (herb-dominated) prevailed and that their duration was longer than between 500 and 285 ka. Forest phases show wetter and cooler conditions in the lower part of the diagram (PASZ OD-13 to OD-8, 502–288 ka) as indicated by the dominance of conifers, while in the upper part (PASZ OD-3 and OD-1, 129 ka–present) there was a “general” increasing trend in temperature indicated by the presence of mesophilous broadleaved trees. In OD-6 (245–190 ka) a balanced alternation of the two vegetation “types” can be observed.

This general trend is visible in the reduction of montane trees present in OD-10 and 12 (roughly corresponding to MIS11 and 13) and the expansion of mesophilous and Mediterranean taxa in the present and penultimate interglacials (Fig. 3). The pre-penultimate interglacial (OD-8, 333–288 ka, cf. MIS9) shows increased mesophilous trees. The penultimate interglacial (OD-6, 245–190 ka, cf. MIS7) shows intermediate features, with balanced presence of montane and mesophilous taxa. This trend seems to be confirmed also by herbs: Poaceae and Cyperaceae decrease, while Artemisia and Chenopodiaceae increase towards the top of the diagram. Steppes and steppe forests seem to characterize the last two glacial periods.

OD-12 (488–459 ka) shows a dominance of AP and the overwhelming presence of pine pollen. This suggests that this period, corresponding to the first part of the MIS12 glacial phase, could have been cold but not very dry, so that conifer montane taxa such as Pinus, Picea and Abies were growing in the lake basin. In the following zone OD-11 (459–428 ka), stronger glacial conditions are evidenced by decreased AP and increased herbs. The curve of Hippophaë, the only arboreal plant with increasing percentages (Fig. 2), confirms this interpretation. The climate of this glacial phase was anyway wetter than the following ones, as evidenced by the permanence of both trees and the expansion of Cyperaceae. The relative humidity recorded at Lake Ohrid during the second part of MIS12 (OD-11) is consistent with the high endemism and biodiversity of the site. The buffering capacity of the lake has to be considered together with the possibility that a part of pine pollen could be from Pinus peuce, a species with high ecological plasticity, which currently has only a relict distribution and is adapted to cold and moist conditions (Aleksandrov and Andonovski, 2011). The surrounding area of the lake could have acted as a refugium for conifers such as Macedonian pines. The relatively low abundance of the xerophytic Mediterranean “ecogroup” also supports this view.

If we do not consider pine, the passage to the following interglacial (OD-10, 428–366 ka) is marked by an important and multi-millennial-long expansion of Abies (accompanied by the Quercus robur type) followed by a ∼ 10 ka-long expansion of Picea (accompanied by the Quercus cerris type). This vegetation pattern indicates that the first part of this interglacial was warmer and wetter than the second one. Moreover, this long-term succession, which has also been documented in Praclaux (de Beaulieu et al., 2001) and in the central European lowlands (Koutsodendris et al., 2010), is not represented in the rest of the diagram, pointing to the unique character of MIS 11. Both fir (Abies) and spruce (Picea) could have occupied the montane belt (with pines at higher elevations or in poor soils), while deciduous oaks (Quercus robur type) first, and subsequently semi-deciduous oaks (Quercus cerris type), were most likely growing at lower elevations.

Glacial conditions prevailed during zone OD-9, 366–333 ka (cf. MIS10), even if oscillations of mesophilous trees occurred and alternated with herb expansions. Cichorioideae, together with Asteraceae undiff., characterized the herbaceous vegetation, although their values may be increased in the pollen profile because of taphonomic issues that still need to be further investigated.

The following interglacial OD-8, 333–288 ka (cf. MIS9), shows a three-phase widespread mesophilous arboreal expansion. The Quercus robur type prevailed in the first and longer phase, while the Q. cerris type at the end of the zone indicated a successive change from warmer and wetter to cooler and drier conditions interrupted by short cool events (NAP increases).

OD-7, 288–245 ka (cf. MIS8) shows low AP percentages (pioneer vegetation mainly consisting of the Juniperus type is rather abundant) and increased values of Poaceae. Even if Poaceae pollen could originate from the Phragmites lacustrine vegetation belt, such high values are mainly ascribed to the presence of regional grasslands that are typical for glacial periods in south-eastern Europe (e.g. Tzedakis et al., 2001; Pross et al., 2015).

OD-6 (245–190 ka) shows a very high forest variability, with three expansions of trees interrupted by two herb expansions. This interglacial/interstadial complex, possibly corresponding to MIS7, has a vegetation behaviour quite different from that of MIS9 and MIS11. MIS7 at Lake Ohrid is marked by warmer and wetter conditions as suggested by decreasing Abies and increasing Picea percentages. The first NAP increase is characterized by many taxa with similar values (Poaceae, Chenopodiaceae, Artemisia and other Asteroideae): the second one by Poaceae and the first strong increase in the Artemisia percentage in the diagram.

A long glacial phase is represented in OD-5 (190–160 ka) and OD-4 (160–129 ka). The limit between the two open formations is marked by a change from a grassland-dominated environment (Poaceae and Cyperaceae) to a steppe-dominated (Artemisia) one. Dry conditions are also indicated by a decreasing Quercus robur type and an increasing Q. cerris type together with Juniperus type and Hippophaë percentages. The second part of MIS6 (OD-4) appears to be the driest phase of the diagram. This is in good agreement with hydro-acoustic data and sediment core analyses from the north-eastern corner of Lake Ohrid, which revealed that during MIS6 the water surface of the lake was 60 m lower than today (Lindhorst et al., 2010). Similarly, sedimentological data from the DEEP core (Francke et al., 2016) show that an accumulation of thin mass movement deposits (MMD) occurred during the second part of MIS6, which might be also indicative of low lake levels.

Forests of OD-3, 129–70 ka (cf. MIS5) are characterized by less variability than the previous OD-6 interglacial/interstadial complex. Mesophilous communities prevailed on the montane vegetation. Quercus robur type and Q. cerris type values are rather similar. Picea is very rare and Fagus shows the highest values of the entire record. Similarly to all previous interglacials, the vegetation seems to be organized in altitudinal belts. Periods with open vegetation are featured by expansions of Artemisia, Chenopodiaceae and Poaceae.

The last glacial period, i.e. MIS4-2, is represented in PASZ OD-2 (70–14 ka). It has a rather high variability, evidenced, already at this step of analysis, by important oscillations of most trees.

The present interglacial is characterized by the strong and prominent expansion of the Quercus robur type accompanied by the Q. cerris type and relatively low montane taxa such as Abies and Fagus. The uppermost samples show opening of the landscape by humans, with evidence of crops and spreading of fruit trees such as Juglans (included in Juglandaceae in Fig. 2). The reduced presence of Picea matches both the palynological data from Lake Prespa for the last glacial (Panagiotopoulos et al., 2014) and the present-day vegetation features of FYROM, where spruce is represented by relic populations in few forested areas. During the penultimate glacial (MIS6), Picea populations were probably too near to their tolerance limit to survive. The importance of ecological thresholds for temperate trees was carefully investigated in three Greek records located in contrasting bioclimatic areas (Ioannina, Kopais, Tenaghi Philippon; Tzedakis et al., 2004a). This turned out to be crucial to understand the importance of local factors in modulating the biological response to climatic stress that occurred in the last glacial and to comprehend the present-day distribution of arboreal species in the Balkans.

In Fig. 4 alignment of the TOC, TIC, AP percentages and AP + NAP concentrations from Lake Ohrid (and “ecogroup” curves of Fig. 3) with both Tenaghi Philippon AP % (Tzedakis et al., 2006) and marine isotope curves shows a very good general agreement between the different records. TOC and AP + NAP (pollen of terrestrial plants) concentration as well as AP % show the same main changes, indicating that there is a tight coupling between the plant biomass and the organic carbon deposited in the lake. TIC increases are mostly in phase with vegetation changes too. The main discrepancies between both TIC / TOC and pollen data are found during glacial phases OD-12 (488–459 ka) and OD-9 (368–333 ka).

The similarity between Lake Ohrid and Tenaghi Philippon curves is striking. All the main changes in forest cover match, and they both correspond to marine records too. There are some differences in the timing of the onset of interglacial phases. DEEP core chronology benefited in fact from the presence of several tephra layers (see Fig. 2, Leicher et al., 2015). The main difference with Tenaghi Philippon is in the fact that arboreal taxa show a continuous presence at Lake Ohrid, even during the glacials, while at Tenaghi Philippon they often disappear to spread again during the interglacials, often with a certain delay. This behaviour could anyway have been expected considering the differences in water availability at the two sites. In Greece, not only Tenaghi Philippon, but also the Kopais (Okuda et al., 2001) areas, resulted in not being ideal refugia for mesophilous trees (Tzedakis et al., 2004a). A quite different situation is found at Ioannina (western Greece), a refugial site for temperate trees featuring sub-Mediterranean climate and vegetation in the last ∼ 480 ka (Tzedakis, 1994b; Tzedakis et al., 2002, 2004a).

Besides a close correspondence to the Tenaghi Philippon AP % curve, Fig. 4 also shows a close correspondence between our pollen data and the Mediterranean benthic and planktic composite curves (Wang et al., 2010; Konijnendijk et al., 2015). Compared to the global isotope stack (Lisiecki and Raymo, 2005; Railsback et al., 2015), additional detail in the pollen diagram is clearly representative of regional Mediterranean conditions and of the influence of moisture availability on the expansion of plants. Both marine deep and surface water features show additional warm phases during interglacials that are also observed in the pollen data. For example, the tripartite forests during MIS7 are well reflected in the pollen data but likely overprinted by the effect of ice volume in the global benthic isotope stack. Completion of the downcore analysis of the DEEP core from Lake Ohrid will allow for a more accurate correlation of the entire sequence with the orbitally tuned Mediterranean isotope records, and provide a finer tuning of the present age model (Francke et al., 2016) to independently dated records in the Mediterranean region where available.