δ18Oc is dependent on δ18Olw and the temperature of lake water at the time of mineral precipitation, assuming equilibrium conditions (Leng and Marshall, 2004). Sediment trap data from Lake Ohrid show that calcite precipitation is seasonally induced, with up to 3 times more TIC formed during summer months in comparison to winter months (Matzinger et al., 2007). The precipitation of calcite is thought to be associated with increased temperatures and the photosynthetic removal of CO2 within the epilimnion, providing there is sufficient supply of calcium (Ca2+) and bicarbonate (HCO3-) ions, which are mainly sourced from spring inflows (Matzinger et al., 2006a). The production of phytoplankton in the lake reaches a maximum between June and August, during which the temperature of the main productivity zone ranges between approximately 12 and 22 ∘C (Stankovic, 1960). If the average temperature of the photic zone during summer months is approximately 18 ∘C, and we take the average modern δ18Olw of -3.5 ‰ (Fig. 4), the calculated δ18Oc of contemporary calcite precipitation should be approximately -4.0 ‰, using the equation of O'Neil et al. (1969), or –4.4 ‰, using the Leng and Marshall (2004) expression of Kim and O'Neil (1997). This is similar to the average δ18Oc through the Holocene (-4.9 ‰) and to the most recent measurement of δ18Oc (-4.5 ‰) from core Co1262 (Lacey et al., 2015), suggesting that δ18Oc most likely corresponds to summer lake water conditions.

Calcite may comprise up to 80 % of the total sediment composition (assuming TIC mainly represents CaCO3; Wagner et al., 2008), with only a minor biogenic component (< 0.1 %) and limited terrigenous contribution (Lézine et al., 2010). As Lake Ohrid is located within a karst catchment, a proportion of the carbonate could be of detrital origin, which commonly has a different isotope composition to the endogenic fraction (Leng et al., 2010b). The catchment geology has variable δ18O (-9.7 to -2.6 ‰; Leng et al., 2010a); however, previous investigations on sediment trap and core material have shown that the calcite crystals have morphological characteristics (for example size and shape) typical of an endogenic origin (Lézine et al., 2010; Matter et al., 2010).

The modern isotope data from Lake Ohrid indicate a clear evaporative disparity between the isotope composition of lake water and that from meteoric and groundwater sources (Fig. 4). The calcite precipitated from a hypothetical, exclusively meteoric, water source would have δ18Oc= -10.6 ‰ (using the mean summer lake water temperature of +18 ∘C and inflow δ18O of -10.1 ‰). Similarly low δ18Oc is not observed in any isotope data from Lake Ohrid to date, including core catcher data covering the entire 1.2 Ma sediment sequence (Wagner et al., 2014), which indicates that lake water has always been subject to a varying extent of evaporative fractionation. This suggests that δ18Olw is primarily influenced by long-term changes in the precipitation/evaporation ratio (P / E). Although temperature changes will influence δ18Olw values, the effect is reported to be roughly +0.2 ‰ ∘C-1 for the central Mediterranean region (Bard et al., 2002), which is quantitatively compensated for by the equilibrium isotope fractionation between carbonate and water (-0.2 ‰ ∘C-1; Leng and Marshall, 2004).

The δ18Oc data from calcite in Lake Ohrid are largely restricted to the interglacial (or interstadial) periods. Interglacial sediments are characterised by concomitant increases in both TIC and TOC (Francke et al., 2016), suggested to be the result of enhanced primary productivity associated with a warmer climate (Wagner et al., 2010). Calcite precipitation is favoured by elevated temperatures during interglacials, which drives higher evaporation rates, thereby concentrating Ca2+ and HCO3- ions. Further, elevated catchment soil activity and temperature will enhance the dissolution of carbonate rocks leading to a greater concentration of dissolved ions in karst spring water. Warmer surface waters also lower the calcite saturation threshold (Lézine et al., 2010). Conversely, glacial sediments typically have low TIC and TOC that are inversely correlated with K and Ti concentrations, indicating low productivity and increased clastic input (Vogel et al., 2010). Lower temperatures during glacial periods would lead to a more oxygenated water column through increased vertical mixing and more frequent complete deep convective overturn. Enhanced levels of mixing breaks down water column stratification and associated oxygenation of the water column increases the rate of aerobic decomposition of organic matter, releasing CO2 that reduces pH levels and increases calcite dissolution (Vogel et al., 2010). Following extensive organic matter degradation the C / N ratio of sediments may be significantly reduced, as observed in the DEEP site cores (Francke et al., 2016) and in previous cores from Lake Ohrid where during the Last Glacial C / N values were typically very low (4–5) compared to higher values (8–12) in both the Holocene and MIS 5 (Wagner et al., 2009; Leng et al., 2010a). Catchment permafrost may have also been prevalent in glacial periods, limiting the supply of Ca2+ and HCO3- ions to the lake by reducing the volume of karstic spring inflow (Belmecheri et al., 2009), which is supported by pollen-inferred mean annual temperatures during the last glacial period of between -3 and +1 ∘C (Bordon et al., 2009). Although there is no (or limited) calcite in the glacials, previous work on Lake Ohrid has shown spikes in TIC during MIS 2–3 (Wagner et al., 2010), and similar increases in glacial TIC are observed throughout the 5045-1 composite profile (Francke et al., 2016). These TIC spikes are most likely analogous to those found in Lake Prespa glacial sediments during MIS 4–2, which comprise siderite (Leng et al., 2013).

Thin sections from discrete higher-TIC glacial intervals reveal individual siderite crystals (< 5 µm) and siderite crystal clusters (50–100 µm) nucleating within an uncompacted clay matrix (Fig. 6). The distribution of siderite within each thin section is variable; a higher concentration of siderite crystals is contained within burrow-like structures that impart a mottled texture to the sediment. Occasional dolomite grains, large (> 20 µm) and distinct from the fine clay matrix, are fringed by 5 µm grains of siderite. The dolomite crystals are thought to be detrital as they are larger than the individual siderite grains and have irregular margins. Siderite comprises the principal carbonate component in these horizons, and apart from the occasional dolomite crystals, no other type of carbonate was observed. Individual siderite crystals appear to predominantly form within the open framework of the clay matrix, which suggests they precipitated in situ within the available pore space. The siderite is therefore most likely to be early diagenetic and formed before compaction within the sediment. Discrete horizons enriched in Fe have been previously observed in Lake Ohrid (Vogel et al., 2010), neighbouring Lake Prespa (Wagner et al., 2010; Leng et al., 2013), and in other ancient lakes, such as Lake Baikal (Granina et al., 2004), where the formation of Fe-enriched layers up to approximately 25 cm below the sediment–water interface is thought to be related to bottom water redox conditions and significant changes in sedimentation regime. Assuming the siderite is formed in superficial sediments during the initial stages of diagenesis, like calcite, its isotope composition can be used as an indicator of depositional environment (Mozley and Wersin, 1992).

To enable comparison between δ18Oc and δ18Os, we convert both to δ18Olw using specific mineral fractionation equations and different estimates of temperature. For calcite data we use the equation of O'Neil et al. (1969) and a precipitation temperature of +18 ∘C (±3 ∘C) to represent average summer conditions within the photic zone during the period of maximum phytoplankton activity. For siderite data we use the equation of Zhang et al. (2001), which is considered robust for defining equilibrium precipitation at lower temperatures (Ludvigson et al., 2013), and assume a bottom water temperature of +6 ∘C (±2 ∘C; Stankovic, 1960). The calculated δ18Olw is given in Fig. 3 and the averages for each MIS are compared to those of δ13Cc and δ13Cs in Fig. 7a.

The calculated δ18Olw from glacial siderite is generally lower compared that from calcite (higher δ18Olw) during warmer interglacial periods (Fig. 7). Although siderite horizons probably represent distinct rapid and recurrent events in Lake Ohrid (Vogel et al., 2010), overall lower δ18Olw may nevertheless be expected through glacial periods due to reduced lake water evaporation as a result of decreased temperatures. Jones et al. (2007) calculated evaporation rates at Eski Acıgöl in central Turkey and showed that glacial evaporation was around 3 times lower compared to that of the Late Holocene (0.4 m yr-1 vs. 1.1 m yr-1). If a similar calculation is conducted for Lake Ohrid, glacial evaporation may have been over 4 times lower than during the present interglacial (0.4 m yr-1 vs. 1.8 m yr-1), after Jones et al. (2007) using the equation of Linacre (1992). Higher evaporation rates are typically associated with closed lake basins and covariance between δ18O and δ13C (Talbot, 1990; Li and Ku, 1997), which is observed in the Lake Ohrid data as interglacial δ18Olw from calcite has a moderate covariance (r= 0.30; Fig. 7b), corroborating that interglacial periods were characterised by higher evaporation. As rates of evaporation reduce during colder intervals, the influence of other controlling factors, such as δ18Op, may have had a greater importance in determining δ18Olw during glacial periods.

During colder intervals, δ18Op would have been lower as a direct correlation exists between annual precipitation and temperature of +0.6 ‰ ∘C-1 at mid–high latitudes (Dansgaard, 1964) and +0.2 ‰ ∘C-1 in the central Mediterranean (Bard et al., 2002). If a mean annual temperature difference of up to 9 ∘C is assumed between interglacial and glacial periods, based on pollen-inferred temperature data from nearby Lake Maliq (Bordon et al., 2009), δ18Op may have decreased by between -5.4 and -1.8 ‰ in glacial periods. When considering interglacial–glacial timescales, changes to the oxygen isotope composition of seawater may also influence δ18Op. The isotope composition of mean global seawater is reported to be 1.0 ‰ higher during the last glacial due to the expansion of global ice volume (Schrag et al., 2002) and up to 1.2 ‰ in the Mediterranean due to local evaporative enrichment (Paul et al., 2001). Therefore, the net effect of temperature and source changes during glacial periods results in lower δ18Op.

In addition to more regional effects on δ18Olw, local influences may also contribute to lower isotope values through glacial periods. Today, a significant proportion of winter precipitation occurs as snowfall at higher altitudes in the Ohrid-Prespa catchment, which is ultimately transferred to the lakes during spring when temperatures remain high enough for the snow to melt (Hollis and Stevenson, 1997; Popovska and Bonacci, 2007). Average winter temperatures at present are around 2 ∘C (Stankovic, 1960); however, winter temperatures would have been considerably reduced during glacial periods and temperatures during summer months may also have been lower (Bordon et al., 2009). If lower temperatures persisted throughout much of the year, a higher proportion of annual precipitation may have fallen in winter as snow. Snow is typically characterised as having much lower δ18O than rainfall, which reflects in-cloud equilibrium conditions and cooler condensation temperatures (Darling et al., 2006), and so would provide a further potential source for low δ18O (Dean et al., 2013).

Temperatures may have been sufficiently reduced during glacials to also allow (at least discontinuous) permafrost to form in the Ohrid catchment, thereby decreasing input from karst waters and perhaps restricting the inflow of water from Lake Prespa (Belmecheri et al., 2009). Lake Prespa provides a large proportion of water input to Ohrid through the underground network of karst channels, which has higher δ18Olw when compared to measured precipitation (Fig. 4; Leng et al., 2010a). This infers that during periods where glacial conditions were prevalent in the catchment, the inflow of water comprising high δ18O from Lake Prespa may have reduced, and input would have instead been principally sourced from a combination of direct precipitation and surface run-off, both of which would result in lower δ18Olw.

When a carbonate mineral precipitates under equilibrium conditions it captures the δ13C of the total dissolved inorganic carbon (TDIC) of lake water. TDIC in most lakes (at neutral pH) can be approximated to dissolved HCO3-, which is principally derived from the dissolution of carbonate catchment rocks, soils, and atmospheric CO2 (Cohen, 2003). Consequently there are several carbon reservoirs that may influence δ13CTDIC, in addition to two major fractionation effects: (1) the chemical exchange between atmospheric CO2 and dissolved HCO3- and (2) kinetic processes during the formation of organic matter (Hoefs, 1980; McKenzie, 1985). In Lake Ohrid, endogenic calcite precipitated within the epilimnion is thought to form in equilibrium with surface waters; thus δ13Cc can provide information on past variations in δ13CTDIC and associated carbon-cycle transitions (Leng and Marshall, 2004). The DEEP site record (Fig. 3) shows overall high and consistent δ13Cc throughout the core (+0.4 ± 0.6 ‰, 1σ, n= 924), which is most likely driven in part by the carbon isotope composition of inflow (δ13CTDIC). Within karst catchments the local and groundwater chemistry will be dominated by Ca2+ and HCO3- ions (Cohen, 2003), and δ13CTDIC will be high as catchment limestones usually comprise ancient marine carbonates (average δ13C = 0 ‰; Hudson, 1977; Jin et al., 2009) and typically range between -3 and +3 ‰ (Andrews et al., 1993; Hammarlund et al., 1997). Although δ13CTDIC has not been measured for Lake Ohrid, analysis conducted on several of the main geological units in the catchment provides average δ13C of +1 ‰ (Leng et al., 2010a), which confirms that δ13CTDIC will most probably be high as over 50 % of water input to Ohrid is derived from springs fed by karst aquifers (Matzinger et al., 2006b). Over glacial–interglacial timescales, the extent to which geological sources of carbon contribute to TDIC will primarily be determined by hydrological balance and associated changes in the residence time of water passing through the karst system; lower δ13CTDIC may occur during wetter periods with a lower residence and higher δ13CTDIC in more arid periods with a higher residence time.

In opposition to geological sources of high δ13C, a major source of carbon in ground- and river water typically derives from CO2 liberated during the decay of terrestrial organic matter (low δ13C). Low δ13CTDIC is measured for Lake Prespa water inputs (average δ13CTDIC= -11.5 ‰; Leng et al., 2013) and may be similar to the inflow to Lake Ohrid. Over glacial–interglacial timescales, variations in the proportion of soil-derived CO2 incorporated into catchment waters will likely influence Lake Ohrid δ13CTDIC. In colder periods, δ13CTDIC would be higher due to poor soil development (Panagiotopoulos et al., 2014; Sadori et al., 2016); conversely, during wetter and warmer intervals the development of dense forests would promote well-developed soils (lower δ13CTDIC) and encourage the delivery of Ca2+ and HCO3- through the dissolution of carbonate catchment rocks.

Equilibrium exchange between atmospheric CO2 and lake water will result in δ13CTDIC of approximately +3 ‰ (fractionation factor +10 ‰ when in equilibrium with CO2 gas). As δ13Cc is thought to reflect changes in δ13CTDIC, higher δ13Cc in the Ohrid record may also reflect variable degrees of equilibration between atmospheric CO2 and dissolved HCO3-. This process is also observed in isotope data from Lake Prespa, where low δ13CTDIC entering the lake (-11.5 ‰) is modified by within-lake processes to increase lake water δ13CTDIC to give the average value of -5.2 ‰ (Leng et al., 2013). In addition to evaporation, it is also likely that biogenic productivity drives higher δ13CTDIC in Prespa (and Ohrid) due to the preferential incorporation of 12C during photosynthesis, assuming the organic carbon is exported to the lake floor and buried (Meyers and Teranes, 2001).

Evaporative drawdown will also affect δ18O as the preferential loss of 16O during evaporation can drive higher δ18O resulting in covariance between δ18O and δ13C. The signal of covariance will be recorded in primary lacustrine carbonates and can potentially be used to determine the degree of past hydrological closure (Talbot, 1990). The extent to which isotope measurements covary can depend on several factors, including hydrological balance, stability of the lake volume, vapour exchange, and evaporation (Li and Ku, 1997). Covariance may therefore not simply be a function of residence time or hydrological closure (Leng et al., 2010a) and has been shown to be spatially inconsistent between Mediterranean lake sediment records (Roberts et al., 2008). Nevertheless, lake level fluctuations are thought to have occurred in Lake Ohrid, at least during MIS 6, as evidenced by the presence of subaquatic terraces on the north-east shore of the lake (Lindhorst et al., 2010). Reductions in lake level, most probably coincident with periods of regional aridity and generally lower P / E, may limit surface outflow, solely met at present by the river Crim Drim (66 %; Matzinger et al., 2006b), which in turn would extend lake water residence time and increase the possibility of evaporation and isotope exchange, resulting in higher δ18O and δ13C. However, periods of higher covariance are generally restricted to certain intervals, for example MIS 5 (r= 0.53, p= < 0.001, n= 104), as throughout the whole core correlation between δ18O and δ13C is generally weak (r= 0.30, p= < 0.001, n= 924; Fig. 7). In those areas where δ18O and δ13C are decoupled, local in situ process are likely to dominate the evolution of δ13CTDIC and act to buffer any climate signal (Regattieri et al., 2015).

In contrast to δ13Cc, δ13Cs is higher in Lake Ohrid sediments by > 8 ‰ and has a mean value of +12.3 ± 0.5 ‰ (1σ, n= 22). Higher δ13C is characteristic of siderite formed in non-marine sediments and is most probably associated with the incorporation of 13C-enriched bicarbonate derived from methanogenesis. The metabolic pathway utilised by bacteria during the reduction of organic matter strongly fractionates in favour of 12C, which, for isotopic mass balance, produces methane (low δ13C) and proportionally enriches the bicarbonate ion in 13C. The methane is subsequently removed by ebullition or by emission through the stems of aquatic macrophytes, and the enriched bicarbonate is incorporated into TDIC (Curry et al., 1997).

At the transition between MIS 16 and 15, δ13Cc shows a prolonged trend to lower values between ca. 625 and 615 ka, whereas a δ18Oc minima occurs at ca. 622 ka, following which values increase (Fig. 8) concomitant with increasing TIC and biogenic silica (BSi; Francke et al., 2016). This suggests an initial phase of higher precipitation (elevated P / E) may have been associated with a mediated catchment soil development and gradual climate warming following the MIS 16 glacial, and after ca. 622 ka lake waters became more evaporated. Excursions to lower δ18Oc and δ13Cc within MIS 15, centred on ca. 615, 600, and 577 ka, are most likely related to lower δ18O in LR04 (Fig. 8; Lisiecki and Raymo, 2005), the Mediterranean Sea (Kroon et al., 1998), and the Ionian Sea (Rossignol-Strick and Paterne, 1999), representing an influx of glacial meltwater into the oceans and warmer conditions during MIS 15e, 15c, and 15a respectively (Fig. 8). In addition, this is mirrored by high arboreal pollen (AP) at Tenaghi Philippon (Tzedakis et al., 2006) and warmer temperatures during these times also promoted vegetation growth in the catchment at Lake Ohrid (Sadori et al., 2016), leading to enhanced soil development and lower δ13Cc.

MIS 14 is the only glacial of the record to contain a higher proportion of TIC and TOC throughout the majority of the stage (Francke et al., 2016), indicating temperatures at Lake Ohrid did not decrease significantly to abate productivity and calcite production (therefore calcite isotope data are available for the majority of MIS 14). Hydroclimate conditions are assumed to have remained fairly similar to MIS 15 as average δ18Oc remains consistent between the two stages (Fig. 8), suggesting that MIS 14 was a particularly weak glacial. Sustained warmth through MIS 14 is in agreement with a range of global records, including in sea surface temperature (SST) estimates from the Iberian margin (Fig. 8; Rodrigues et al., 2011) and South China Sea (Yu and Chen, 2011), BSi and magnetic susceptibility from Lake Baikal (Prokopenko et al., 2002, 2006), and several proxies from Antarctic ice cores (Jouzel et al., 2007; Masson-Delmotte et al., 2010). Although overall warmer conditions may have prevailed during MIS 14, the Ohrid record suggests that colder glacial-like conditions occurred between ca. 540 and 531 ka, which is supported by the presence of siderite (Fig. 3). δ13Cc increases throughout MIS 14 imply a progressive decline in catchment soil development. A reduction in the amplitude of δ13Cc oscillations, for example when compared to MIS 15, may be due to low orbital eccentricity reducing the influence of precession and insolation variability (Fig. 8).

MIS 13 in Lake Ohrid δ18Oc and δ13Cc represents a relatively stable period, experiencing only minor oscillations through the stage. The average isotope composition is comparable to that observed during MIS 14 and 15 (Fig. 7); however MIS 13 is generally considered to be one of the weakest interglacials of the last 800 ka (Lang and Wolff, 2011). Similar to the onset of MIS 15, MIS 13c is characterised by relatively rapid transition to lower δ18Oc at ca. 529 ka followed by increasing values through to ca. 521 ka, associated with a gradual concomitant trend to lower δ13Cc (Fig. 8). An excursion to higher δ13Cc centred around ca. 510 ka is probably linked to cooler conditions and the onset of MIS 13b, and a transition to higher δ18Oc suggests lower P / E, which is similar to observations from Tenaghi Philippon and Antarctic ice cores that indicate reduced AP and lower reconstructed temperatures (Tzedakis et al., 2006; Jouzel et al., 2007). The LR04 record also exhibits higher δ18O at this time, signifying an expansion of global ice volume (Fig. 8; Lisiecki and Raymo, 2005). A shift to lower δ18Oc and δ13Cc at ca.  502 ka most likely represents the onset of MIS 13a, where δ13Cc is the lowest of the stage (ca. 530 to 474 ka), suggesting MIS 13a most probably experienced warmer conditions than MIS 13c. The timing of interglacial conditions is unique within the Lake Ohrid record, and correspondingly in global sequences, as minimum ice volume (Lisiecki and Raymo, 2005) and maximum warmth (Jouzel et al., 2007; Loulergue et al., 2008) occurred in the final substage of the MIS rather than directly following the glacial termination (Fig. 8). This was most probably due to MIS 14 experiencing only weak glacial conditions resulting in a low-amplitude glacial termination (Voelker et al., 2010). Peaks to higher δ18Oc and δ13Cc toward the end of MIS 13a are coincident with an increased proportion of siderite in sediments (Fig. 3), and so fluctuations may be the result of a mixed carbonate composition rather than due to environmental change.

MIS 11 is thought to have a characteristic orbital and climate configuration potentially analogous to that of the Holocene (Loutre and Berger, 2003) and follows a strong and relatively wet glacial MIS 12 at Lake Ohrid (Sadori et al., 2016). Following the glacial termination δ18Oc is low (around -6 ‰) and increases over the next 15 ka to a maximum (-4 ‰) at ca. 410 ka, whereas δ13Cc transitions to lower values between ca. 425 and 410 ka, suggesting a prolonged period of warm and relatively stable conditions. This period most likely corresponds to MIS 11c, where warmer, wetter conditions are supported by high TIC and AP at Lake Ohrid (Francke et al., 2016; Sadori et al., 2016). The overall progression of δ18Oc and δ13Cc through MIS 11c corresponds to high SST at the Iberian Margin (Fig. 8; Rodrigues et al., 2011) and compares well with the development of CO2 and CH4 measured from the EDC Antarctic ice core record (Loulergue et al., 2008; Lüthi et al., 2008). A distinct δ18Oc and δ13Cc maxima centred around ca. 405 ka is likely associated with colder conditions and a drier environment (lower P / E) through stadial phase MIS 11b, which is supported by higher K intensity, lower TIC, and a substantial decrease in AP (Francke et al., 2016; Sadori et al., 2016). An overall trend to lower δ18Oc and δ13Cc through MIS 11a (ca. 402 to 385 ka) shows greater variability in comparison to MIS 11c, suggesting that the stability of hydroclimate conditions changed after peak interglacial conditions. Greater variability after ca. 400 ka is also seen in other proxies from Lake Ohrid (TIC, BSi; Francke et al., 2016), as well as in reconstructed SST at the Iberian Margin (Rodrigues et al., 2011) and atmospheric CH4 and temperature profiles from Antarctic ice cores (Fig. 8; Petit et al., 1999; Jouzel et al., 2007). Following the climatic optimum of MIS 11c, δ18Oc transitions to the lowest value of the record at ca. 384 ka (-7.6 ‰), over which time there is an increase in global ice volume and decrease in atmospheric greenhouse gas concentrations observed through MIS 11b and 11a (Lisiecki and Raymo, 2005; Loulergue et al., 2008). A trend to lower δ18Oc at Ohrid also traces reducing SST in the North Atlantic and on the Iberian Margin (Stein et al., 2009; Rodrigues et al., 2011). The progression to lower δ18Oc during a period of overall cooling is expected, given that the reconstructed δ18Olw from glacial siderite is typically lower than interglacial δ18Olw, as calculated from calcite (Fig. 3).

The initial ca. 4 ka of MIS 9 is marked by a transition to higher δ18Oc and lower δ13Cc, similar to previous interglacial stages. After ca. 330 ka variations in δ18Oc and δ13Cc are coupled and a minimum in both values is observed around ca. 329 ka (Fig. 8), which is assumed to correspond to peak interglacial conditions during MIS 9e. The onset of full interglacial conditions during MIS 9e is indicated to be relatively rapid, given that a maximum of ca. 6 ka elapses between the onset of calcite precipitation and peak interglacial conditions, which is consistent with warming at the start of MIS 9 observed in AP from Tenaghi Philippon (Tzedakis et al., 2006) and SST from the Iberian Margin (Fig. 8; Martrat et al., 2007). An abrupt transition to the highest δ18Oc and δ13Cc of MIS 9 between ca. 321 and 318 ka is coincident with lower TIC, BSi, and AP at Lake Ohrid (Francke et al., 2016; Sadori et al., 2016). A reduction in Pinus populations in the Ohrid catchment alongside lower P / E indicates an overall drier climate, which is also documented at Tenaghi Philippon (Tzedakis et al., 2006) and most likely associated with stadial conditions during MIS 9d. The subsequent transition to lower δ18Oc and δ13Cc towards ca. 310 ka is probably associated with a warmer climate, higher precipitation and P / E during interstadial MIS 9c, reflected at Tenaghi Philippon by higher AP (Tzedakis et al., 2006) and moderately lower δ18O in the LR04 stack (Fig. 8; Lisiecki and Raymo, 2005). A subsequent break in TIC preservation occurs between ca. 308 and 293 ka, which is preceded by a transition to higher δ18Oc and δ13Cc, suggesting a drier climate prevailed, and is most likely associated with stadial MIS 9b. This interval is also indicated to be an extended period of cold and dry conditions at Lake Ohrid by low TIC and BSi (Francke et al., 2016) and low AP (Sadori et al. 2016), where the presence of siderite advocates that a glacial-like climate state persisted throughout much of the stadial phase (Fig. 3). A trend to lower δ13Cc between ca. 293 and 286 ka closely corresponds to an increase in TIC, BSi, and AP, and a rapid shift to lower δ18Oc after ca. 293 ka indicates the onset of warmer and wetter conditions during interstadial MIS 9a. Low and relatively stable δ18Oc values between ca. 292 and 288 ka imply a fresh lake system and higher P / E, which may be driven by increased precipitation in association with the deposition of sapropel S′ in the Mediterranean Sea (Ziegler et al., 2010) and summer insolation maxima (Fig.  8; Laskar et al., 2004). Overall, low δ18Oc through MIS 9 and its calculated δ18Olw are coincident with similar δ18Olw values from siderite during MIS 10 and MIS 8 (Fig. 3), indicating glacial and interglacial lake water broadly converge through this interval.

MIS 7 at Lake Ohrid is characterised by three distinct phases of calcite preservation (Fig. 3; Francke et al., 2016), which most likely correspond to MIS sub-stages 7e, 7c, and 7a (Railsback et al., 2015). The first peak in TIC between ca. 245 and 238 ka is associated with initially low and increasing δ18Oc and high but decreasing δ13Cc, which both show a higher amplitude of variability after ca. 242 ka. Increasing δ18Oc suggests a transition from wetter to drier climate through to ca. 238 ka; however, warm and wet conditions are indicated by overall low δ18Oc coincident with high TIC, BSi, and AP (Francke et al., 2016; Sadori et al., 2016). This period most probably corresponds to interglacial MIS 7e and is coincident with warming indicated by arboreal expansion at the Ioannina (Roucoux et al., 2008) and Tenaghi Philippon (Fig. 8; Tzedakis et al., 2006) basins in Greece and at Lake Van in Turkey (Litt et al., 2014). Overall warming is supported by higher reconstructed SST for the Adriatic Sea (Piva et al., 2008) and on the Iberian Margin (Fig. 8; Martrat et al., 2007). A cessation of TIC production between ca. 238 and 221 ka indicates a colder, extended glacial-like climate state during stadial MIS 7d, which is also suggested by the presence of siderite (Fig. 3). A relatively abrupt decrease in δ13Cc after ca. 221 ka marks the onset of warmer conditions at Lake Ohrid and the transition to interstadial MIS 7c. An interval of low δ13Cc between ca. 221 and 212 ka represents the lowermost values of the core, which suggests MIS 7c may have experienced warmer conditions than during the full interglacial (MIS 7e) and is supported by a higher TIC plateau (Francke et al., 2016). In addition, a greater diversity of AP is observed at Tenaghi Philippon alongside a greater abundance of thermophilous taxa (Tzedakis et al., 2003b) and higher BSi at Lake Baikal (Prokopenko et al., 2006). Although CO2 and CH4 concentrations were lower in MIS 7c (Loulergue et al., 2008; Lüthi et al., 2008), higher summer insolation during MIS 7c most likely promoted warmer climate conditions (Fig. 8). A change to stadial conditions is suggested around ca. 211 ka by a shift to higher δ18Oc and δ13Cc, as well as the presence of siderite (Fig. 3), which likely corresponds to MIS 7b. Decreasing δ13Cc and an increasing δ18Oc trend between ca. 210 and 202 ka most probably correspond to interstadial MIS 7a; however, unlike substages 7e and 7c, the interval is characterised by large amplitude coupled oscillations in both δ18Oc and δ13Cc. A change in climate stability is also reflected in AP at Tenaghi Philippon (Tzedakis et al., 2003b) and in speleothem δ18O from northern Israel (Bar-Matthews et al., 2003). After ca. 202 ka siderite is observed to be abundant through MIS 6.

MIS 5 has been characterised as one of the strongest warm periods of the last 800 ka (Lisiecki and Raymo, 2005; Jouzel et al., 2007; Lang and Wolff, 2011). At Lake Ohrid, the highest δ18Oc and δ13Cc of the record are observed during MIS 5 (Fig. 8), suggesting low P / E and high rates of evaporation, which may be related to severe lake level changes inferred during MIS 5 (Lindhorst et al., 2010). A rapid transition to lower δ18Oc and δ13Cc after ca. 129 ka is most likely associated with the onset of full interglacial conditions during MIS 5e. A climatic optimum is probably associated with the point of lowest isotope values at ca. 124 ka; however, δ18Oc decreases at a faster rate in comparison to δ13Cc (Fig. 8). This may be due to terrestrial and lacustrine proxies decoupling due to local effects of ice cap and snowfield meltwater entering the lake, as is observed at Lake Ioannina (Wilson et al., 2015). Local ice caps on the mountains surrounding Lake Ohrid are indicated for the Last Glacial by catchment moraine deposits (Ribolini et al., 2011) and are likely to have been present during early glacials, such as MIS 6 (Francke et al., 2016). Overall, the interval of lower δ18Oc and δ13Cc is coincident with higher Mediterranean SST (Piva et al., 2008; Martrat et al., 2014), the deposition of sapropel S5 (Ziegler et al., 2010), and greater regional precipitation (Bar-Matthews et al., 2003; Drysdale et al., 2005). An increase in δ18Oc and δ13Cc values after ca.  114 ka signals the onset of stadial conditions (MIS 5d) and infers that interglacial conditions persisted for approximately 15 ka, which is in broad agreement with reported durations from other regional sequences (Tzedakis et al., 2003a; Brauer et al., 2007; Pickarski et al., 2015). High δ18Oc and δ13Cc between ca. 114 and 108 ka suggest drier conditions during MIS 5d, which is supported by the presence of siderite and a reduction in AP (Sadori et al., 2016). An excursion to lower δ13Cc between ca. 108 and 91 ka is coincident with higher AP at both Lake Ohrid and Tenaghi Philippon (Tzedakis et al., 2003b; Sadori et al., 2016) and higher SST at the Iberian Margin (Fig. 8; Martrat et al., 2007), which suggests warmer temperatures during MIS 5c. δ18Oc is observed to be lower in MIS 5c than during full interglacial conditions through MIS 5e (Fig. 8), which could be due to refilling after a lake water lowstand (Lindhorst et al., 2010). Alternatively, given higher insolation and atmospheric CO2 and CH4 concentrations during MIS 5e (Petit et al., 1999; Laskar et al., 2004; Loulergue et al., 2008), evaporation may have been stronger and driven higher δ18Oc, even under a regime of elevated regional precipitation (Bar-Matthews et al., 2003; Drysdale et al., 2005). A transition to higher δ18Oc and δ13Cc after ca. 91 ka corresponds to reduced TIC, BSi, and AP (Francke et al., 2016; Sadori et al., 2016) and high siderite abundance, which are assumed to be associated with stadial conditions during MIS 5b. The transition to higher isotope values occurs over a shorter time interval in comparison to MIS 5d, suggesting MIS 5b probably experienced a more produced change to cold and dry climate conditions. This is supported by higher speleothem δ18O at Soreq Cave (Bar-Matthews et al., 2003), lower reconstructed SST from the Adriatic and Alboran seas (Martrat et al. 2004; Piva et al., 2008), and lower atmospheric CO2 (Fig. 8; Petit et al., 1999). A decrease in δ18Oc and δ13Cc after ca. 86 ka is thought to be related to the onset of interstadial conditions during MIS 5a. A minimum in both isotope values centred around ca. 82 ka infers climate conditions may have been wetter and warmer toward the end of the stage. A transition to lower δ18O is also observed at Soreq Cave (Bar-Matthews et al., 2003), suggesting regional rainfall may have been enhanced toward the end of MIS 5a, which may be related to a coincident peal in summer insolation (Fig. 8; Laskar et al., 2004).