Palaeo-seasonality of the last two millennia reconstructed from the oxygen isotope composition of carbonates and diatom silica from Nar Gölü, central Turkey

are deposited together in in for the composition ( investigated simultaneously from both Here, d 18 O carbonate are compared to d 18 O diatom the sediments of a monitoring carbonate is probably precipitated d 18 O carbonate is a proxy for regional water balance. Diatom activity is mainly weighted towards the At times between w 301 and 561 while d 18 O carbonate values are the highest for the entire 1710 period, suggesting drought, d 18 O corrected-diatom values are among the lowest. d 18 O lakewater values estimated for the times of diatom growth and carbonate precipitation show large differences. We suggest this could be explained by increased snowmelt that formed a freshwater lid on the lake at the time of peak diatom growth. Increased snowmelt is also inferred w 561 e 801 AD. From 801 AD to the present, precipitation is less winter-dominated, although increased snowmelt is inferred 921 e 1071 AD and in the latter part of the Little Ice Age (i.e. the mid to late 1800s AD). By combining oxygen isotope data from hosts that form in lakes at different times of the year, we show that such analyses can provide insights into palaeo-seasonality.

Palaeo-seasonality of the last two millennia reconstructed from the oxygen isotope composition of carbonates and diatom silica from Nar Gölü, central Turkey

Introduction
Water in the Near East is a politically sensitive resource (Issar and Adar, 2010) and water stress in the region is projected to increase during the 21st century (Cruz et al., 2007), so an improved understanding of hydrological variability over long timescales is required. It has been proposed that changes in water availability have influenced the rise and fall of civilisations over the past 2000 years in the region (Issar and Zohar, 2007). While studies such as Jones et al. (2006) have provided high resolution and well dated proxy records of shifts in water balance, there are few reconstructions of seasonal variation in precipitation, despite its importance for human societies (Rosen, 2007). Precipitation in the region today is markedly seasonal, with dry summers and wetter winters and springs (Türkeş , 2003), and shifts in seasonality are anticipated globally in a warming world (Meehl et al., 2007). There is consequently a requirement for the development of methods that allow past seasonality changes to be reconstructed at an adequate temporal resolution for understanding change on human timescales.
Seasonality change analysis requires climate proxies that are sensitive to different seasons. Several reconstructions of seasonality using isotope analysis of lake sediments have recently been published (Henderson et al., 2010;Anderson, 2011;Barker et al., 2011). In the Near East, Stevens et al. (2001Stevens et al. ( , 2006 suggested that Holocene changes in isotope records from Iranian lake sediments were driven by changes in the seasonality of precipitation and Orland et al. (2009Orland et al. ( , 2012 and Rowe et al. (2012) investigated past seasonality using the oxygen isotope composition of speleothems from the Soreq Cave, Israel and a cave in northeast Turkey respectively. Here we consider how to best compare the oxygen isotopic composition of carbonate and diatom components in lacustrine sediments from central Turkey to give insights into changing seasonality.
Oxygen isotope analysis (d 18 O) of lake sediments is most commonly carried out on carbonates. If precipitated in equilibrium, d 18 O carbonate is dependent upon the temperature and isotopic composition of the lake water in which it is precipitated, with the degree of fractionation at À0.24& C À1 . In hydrologically open lakes, temperature and the d 18 O of precipitation entering the lake will be the most important drivers of final d 18 O values. In closed lakes with long residence times, the effect of evaporation will usually be far more important (Li and Ku, 1997;Leng and Marshall, 2004;. d 18 O analysis can also be carried out on diatom silica. The controls on d 18 O diatom are very similar to those on d 18 O carbonate (Leng and Barker, 2006) and the fractionation factor between diatom silica and the water from which it is precipitated has been shown as being similar (ca. À0.2& C À1 ) to calcite-water fractionation (Brandriss et al., 1998;Moschen et al., 2005;Crespin et al., 2010). Under some circumstances, if the pH of lake waters is between 7 and 8, and if silica is not limited, carbonates and diatoms may be preserved in sufficient quantities for d 18 O analysis to be carried out on both hosts. Leng et al. (2001) first suggested that if carbonates and diatom silica are precipitated at different times of the year, comparing d 18 O from the two hosts would provide information on seasonality.
Working on a Late Pleistocene sediment sequence from Lake Pınarbaş ı (37 28 0 N, 33 07 0 E) in Turkey and assuming carbonates were precipitated in the summer and diatoms grew throughout the year but especially in the spring and autumn, they suggested d 18 O carbonate was a proxy for mean summer temperature and d 18 O diatom a proxy for spring snowmelt, accounting for the substantial differences between d 18 O trends from the two hosts. Large differences in d 18 O were also seen in the record from Lake Go scią _ z (52 35 0 N, 19 21 0 E) in Poland (Rozanski et al., 2010), again thought to be because of differences in the time of deposition of diatoms and carbonates. In contrast, a study by Lamb et al. (2005) using the sediments of Lake Tilo in the Ethiopian Rift Valley (7 03 0 N, 38 05 0 E), for the period 9.0e5.7 ka, found that while d 18 O diatom was more variable (ascribed by the authors to be largely the result of tephra contamination) and did not pick up two arid events seen in the d 18 O carbonate record, the general trends were the same because diatom growth and carbonate precipitation occurred at similar times of the year. However, these studies lack the detailed limnological monitoring required for a full understanding of the lake isotope system. Also, they did not rigorously quantify contamination of diatom d 18 O samples. As such, it is unclear whether differences in trends between carbonate and diatom d 18 O are due to shifts in lake conditions or changes in the amount of contamination in the samples, although it should be acknowledged that in these studies the authors thought they were analysing relatively pure diatom material.
Here  (Morley et al., 2005;Brewer et al., 2008;Mackay et al., 2011). We apply this to the study of sediments from Nar Gölü, central Turkey, from where the carbonate (a mixture of endogenic calcite and aragonite) d 18 O (Jones et al., 2005 and the diatom species Roberts, 2010, 2011; records have already been published.

Site description
Nar Gölü (38 20 0 24 00 N, 34 27 0 23 00 E; 1363 m a.s.l.; Fig. 1) is a small (w0.7 km 2 ) but relatively deep (>20 m) maar lake in Cappadocia, central Turkey. The east and west sides of the crater are rimmed by basalt, with the southern side dominated by exposures of ignimbrite (Gevrek and Kazancı, 2000). Consequently, there is no potential for problems associated with detrital carbonate contamination (cf. Leng et al., 2010). The climate of the region is continental Mediterranean (Kutiel and Türkeş , 2005) with annual precipitation at Ni gde, 45 km from Nar, averaging 339 mm between 1935 and 2010. July, August and September are very dry, receiving only 6% of the total precipitation, while April and May are the wettest months, accounting for 27% of the total (Fig. 1). The hottest months are July and August, when temperatures average þ23 C, while from December to February temperatures average þ0.7 C. Measured evaporation at Ni gde between 1935 and 1970 was 1547.6 mm per year (Meteoroloji-Bulteni, 1974).
The modern limnology has been investigated over the past decade. The lake has no surface outflow and lake waters are evaporatively enriched in 18 O compared to spring waters (Table 1), plotting off the Ankara Meteoric Water Line (as discussed in detail in Jones et al., 2005;. This evaporative enrichment leads to the concentration of sodium, chloride and bicarbonate ions, with high pH values w8, which aids the preservation of both diatoms and carbonates. Using sediment traps, and comparing these data to thin section analysis of the sediment cores, it is estimated that carbonate is presently precipitated in MayeJune and diatom activity is weighted towards the spring, with a secondary growth peak in the autumn (Jones et al., 2005;. Data loggers have been used to record temperature through the water column (Eastwood et al., unpublished data) and suggest that through 2009e2011 lake waters were stratified from early March to late November, and mixed during December, January and February. Based on these data, we suggest a likely epilimnion temperature range of þ15eþ20 C at the time of carbonate precipitation and þ5eþ15 C at the time the majority of diatom growth occurred.

Fieldwork and chronology
A 376 cm core sequence was obtained in 2001/2 (NAR01/02) using Glew (Glew et al., 2001) and Livingstone (Livingstone, 1955) corers from the deepest part of the lake (25 m water depth at the time). The cores were laminated throughout and 210 Pb and 137 Cs dating on the top 50 cm of the core and the analysis of sediment trap material indicates the couplets are annual (Jones et al., 2005;. A maximum age uncertainty of 2.5% for the varve counting was calculated from counting replicate cores, although true dating precision is likely to be better than this (Jones et al., 2005). Additionally, a 36 cm core (NAR06, as described in ) and a 44 cm core (NAR10), both collected using a Glew corer, were used to provide samples for the 20th and 21st centuries, including recent sediments deposited since the collection of NAR01/02 core. The three core sequences were stratigraphically tied through counting varve couplets and identifying matching sediment patterns between cores. Water samples were collected from springs and the centre or edge of the lake and maximum lake depths were estimated using a Garmin Ò Fish Finder and a weighted tape during field visits between 1997 and 2012.

d 18 O analysis of waters and carbonates
Water samples were analysed for d 18 O using an equilibration method on a VG SIRA mass spectrometer at the NERC Isotope Geosciences Laboratory (NIGL), with data presented as & deviations from VSMOW. Analytical reproducibility was 0.05&. On the NAR01/02 core sequence each of the uppermost 900 carbonate layers were individually analysed for d 18 O carbonate at NIGL using classic vacuum techniques and an Optima dual-inlet mass spectrometer, with the following 825 analysed from contiguous bulk samples at a 5 year resolution . Every carbonate layer from the top of the NAR10 core was analysed in the same way. The data are presented here as & deviations from VPDB and analytical reproducibility was 0.1&.

Diatom isotope sample preparation
For diatom oxygen isotope analysis, bulk samples consisting of 3 varve years were taken at 10 varve year intervals from 301 AD to 1921 AD on the NAR01/02 core sequence, with d 18 O diatom analysis on samples from the same varve years from different cores in the sequence confirming the core overlaps. The NAR06 and NAR10 cores were used to provide additional sample material for the past 100 years, with 4 and 5 varve year bulk samples taken respectively.
Samples need to be as free as possible of contamination since the method for analysing d 18 O diatom will liberate oxygen from minerogenic material. The cleaning process was similar to that of Morley et al. (2004), with the use of hydrogen peroxide, nitric acid, hydrochloric acid, differential settling and sieving stages. Although sieving at 10 mm meant that small diatom frustules, including those of the endemic species Clipeoparvus anatolicus , were lost from the sample, SEM-EDS (Scanning Electron Microscopy-Energy-Dispersive X-ray Spectroscopy) showed that unsieved samples were significantly more contaminated than sieved samples because of the presence of clays. Samples from the last 100 years from NAR06 and NAR10 (n ¼ 16) underwent a final density separation stage using sodium polytungstate (SPT) to help remove minerogenic material. This stage can be problematic as SPT needs to be removed from the samples otherwise it will itself become a contaminant. SPT was therefore flushed away from samples by filtering them with distilled water at 0.45 mm.

EDS estimations of contamination, d 18 O analysis of diatom silica and mass balance corrections
The process described above can sometimes yield samples of a sufficient purity that contamination is not an issue (e.g. Swann et al., 2010). However, even when SPT is used, diatom samples can remain significantly contaminated, especially when minerogenic material is attached to diatoms by electrostatic charges or trapped within diatoms (Fig. 2) and where there is no significant density contrast between diatoms and the contaminants (Brewer et al., 2008). Even the samples from NAR06 and NAR10 on which SPT was used contain an average of 10% non-diatom silicate material. Contamination in samples from Lake Baikal, initially analysed by Morley et al. (2005) and reanalysed by Mackay et al. (2011), averaged 29.2% despite the use of SPT. In these difficult sediments, mass balancing is the only way to remove the effects of contamination on d 18 O diatom . Morley et al. (2005) and Mackay et al.
(2008) estimated percentage contamination using light microscopy on randomly selected areas. However, point counting gives a surface area, not volumetric, and only a semi-quantitative assessment of contamination. Lamb et al. (2007) and Brewer et al. (2008) used XRF, Chapligin et al. (2012) used Inductively Coupled Plasma-Optical Emission Spectroscopy (ICP-OES) and Swann and Patwardhan (2011) used Fourier Transform Infrared Spectroscopy (FTIR) to provide more accurate contamination estimates. The abundance of elements such as aluminium is typically used to represent the level of contamination (Brewer et al., 2008). However, conventional XRF and ICP-OES require fairly large amounts of material. Therefore, in this study an EDS probe in a SEM, which can make measurements on very small amounts (<0.5 mg) of material, was used instead.
To test the accuracy of the EDS probe in measuring contamination, a series of samples with known mineralogy were analysed following the methodology described below. The Al content of 8 samples from Lake Baikal, previously analysed for d 18 O diatom and XRF (Mackay et al., 2011), and 10 samples with known mixtures of mica, montmorillonite, kaolinite and chlorite, were compared to Al measured on the EDS. This exercise gave an error for the Al reading from the EDS of AE0.5%. Chapligin et al. (2012) found EDS underestimated the amount of contamination in diatom samples. Our analyses show a similar underestimation and we therefore divide our measured contamination by 0.64 (identical to Chapligin et al.'s correction). This led to some samples reporting over 100% contamination, which is clearly impossible, so the values were again adjusted, dividing by the overall range after the 0.64 correction, to restrict the contamination to a range from 0 to 100%.
Prepared samples were viewed at 100Â magnification under SEM, and EDS was used to detect oxygen, sodium, magnesium, aluminium, silicon, phosphorus, sulphur, potassium, calcium, titanium, manganese and iron. Since most clay-sized minerogenic material should have been removed by sieving at 10 mm and carbonates and organics by chemical processes, most of the contamination in 'clean' samples was assumed to be from minerogenic material presumably washed in from the catchment and similar in size, shape and specific gravity to the diatoms. Oxygen isotope analysis of the 'cleaned' diatom samples was carried out at NIGL using the stepwise fluorination technique as described in Leng and Sloane (2008), a method that has been verified through an inter-laboratory comparision exercise (Chapligin et al., 2011). The data are presented as & deviations from VSMOW and analytical reproducibility was 0.3&.
Following Brewer et al. (2008) and Mackay et al. (2011), the amount of contamination in samples as a percentage of overall content was calculated by: diatom sample, % contamination and % diatom (i.e. 100 e % contamination ) are estimated by EDS and d 18 O contamination is the isotope value of contamination. The latter can be estimated in a number of ways.
Measuring d 18 O of the contamination directly is difficult, as much of the minerogenic material is removed during the processing of samples, such that the end member contaminant left is the minerogenic material that has been through chemical and physical separation processes. We kept this effect to a minimum by preparing silt samples in the same way as the diatom samples prior to d 18 O analysis, but to ensure this had no effect we use a modified version of the linear regression method of Chapligin et al. (2012). By plotting measured d 18 O diatom against % diatom (the inverse of % contamination ), an estimated end member contamination d 18 O value of þ16.5& (Fig. 3) is calculated, which we use in Eq.
(2). The uncertainty associated with mass balancing was calculated by combining the errors associated with the various components of Eq.
(2), as discussed above. It should be noted that two factors will likely have led to an overestimation of % contamination . Firstly, some minerogenic material will be removed by the first fluorination stage before d 18 O measurements are made (Swann and Leng, 2009). Secondly, diatom frustules can incorporate Al into their skeletons and therefore Al% in the samples will not just reflect minerogenic contamination (Beck et al., 2002;Koning et al., 2007;Swann, 2010). However, because of the difficulties and uncertainties of correcting for these components, these two factors were not accounted for in this study.

Calculation of d 18 O lakewater
To allow for direct comparison of the d 18 O data from carbonate and diatoms, removing the effects of temperature fractionation, we estimate d 18 O lakewater at the time of diatom growth and carbonate precipitation. We took into account changes in carbonate mineralogy between calcite and aragonite (Jones, 2004), which has an effect on d 18 O carbonate (Grossman, 1984;Abell and Williams, 1989), and re-express the calcite (Leng and Marshall, 2004), aragonite (Grossman and Ku, 1986) and diatom (Crespin et al., 2010)

Water, core top and sediment trap d 18 O data
Nar Gölü has been monitored during field seasons for over a decade so the modern isotope limnology is fairly well understood. There is a strong relationship between lake depth and d 18 O lakewater over the past decade (Fig. 4 and þ37.0& in the last decade) are similar to those from core top sediments.

Contamination
The mean percentage of contamination assessed by EDS across all diatom isotope samples was 40% (n ¼ 149). Samples with >60% contamination were removed from subsequent analyses (n ¼ 22); while this is a fairly arbitrary point, it is around this level that errors increase to >10&. Fig. 5 shows that while some periods on the whole have more contaminated samples than others, high and low % contamination values are found throughout the record.
4.3. Comparing the diatom and carbonate oxygen isotope records 301e2010 AD d 18 O corrected-diatom values for the sediment core samples follow the same general trend as d 18 O diatom but are on average increased by 6.9& (Fig. 5) the variables that could influence it, and in particular which variables have the potential to account for the large magnitude (>15&) shifts seen in Fig. 5.

Contamination of samples
Minerogenic contamination has an effect on d 18 O diatom , and while contamination is high in parts of zone 1 where d 18 O diatom is low, it is also high in parts of the record where d 18 O diatom is high (Fig. 5). This suggests other factors are more important drivers of d 18 O diatom than contamination. In any case, mass balance corrections have been applied to remove the effect of contamination to produce d 18 O corrected-diatom . While there are uncertainties associated with mass balance corrections, the shifts in d 18 O corrected-diatom in the Nar record are far greater than these (Fig. 5). In less sensitive lakes, this issue will be more of a problem as isotopic shifts may not   be outside of error, leading Chapligin et al. (2012) to suggest that the mass balance method is only valid up to 15% contamination.

Changes in species assemblage composition
While species vital effects in diatoms have been shown to be of limited importance (Brandriss et al., 1998;Schmidt et al., 2001;Moschen et al., 2005;Swann et al., 2006;Schiff et al., 2009), downcore changes in the time of year that diatoms are growing in Nar would change the lake conditions (i.e. temperature and d 18 O lakewater ) that d 18 O corrected-diatom is recording. Uncertainties in diatom ecology make it difficult to reconstruct how the seasonality of diatom growth changed through the record. Freshwater species Nitzschia paleacea, Synedra acus and Stephanodiscus parvus were identified as bloom species from thin section analysis between 1100 AD and present Roberts, 2010, 2011). Based on sediment trap data and thin section analysis, N. paleacea and S. parvus probably bloom immediately prior to carbonate formation, while S. acus blooms in the autumn or early spring .
In contrast, Cyclotella meneghiniana dominates biovolume calculations before 800 AD during the periods of significantly lower d 18 O corrected-diatom (Fig. 6) and it may also have been a bloom species (thin sections are not available for this period so this cannot be confirmed). While C. meneghiniana has a high conductivity optimum in the European diatom database (6600 mS cm À1 ; Juggins, 2012), it tolerates a wide range of alkalinity (Gasse, 1986), with a conductivity optimum of 912 mS cm À1 and a tolerance of 218e4168 mS cm À1 in a Ugandan crater lake training set (Mills and Ryves, 2012) and an optimum of 2000 mS cm À1 in the Turkish training set (Reed et al., 2012). In the Ugandan lakes, nutrient input, not salinity, is seen as the most important driver of changes in the abundance of this species (Ryves et al., 2011;Mills and Ryves, 2012). With species such as C. meneghiniana, the importance of salinity relative to nutrients can be difficult to unravel (Fritz et al., 2010).
From temperature monitoring data 2009e2011 we know epilimnion temperature changes in Nar through the spring from wþ5 C in March to wþ15 C in May. d 18 O lakewater values also increase through the year (e.g. 1 May 2009 epilimnion d 18 O lakewater was À1.17& but it had risen to À0.56& by 16 July 2009). However, it is possible that C. meneghiniana grew at times when there was a greater intra-annual range of d 18 O lakewater values, leading to lower d 18 O corrected-diatom values than recorded by the modern bloom species. Whatever the driver of C. meneghiniana, it is interesting to note that its appearance is closely linked to Dd 18 O lakewater in Nar. As well as dominating zone 1, it returns to the diatom record in zone 3 ( Fig. 6), suggesting the lake conditions that lead to low d 18 O correcteddiatom (negative d 18 O lakewater ) values in Nar are conditions that are favoured by C. meneghiniana.

Temperature effect on d 18 O corrected-diatom
If the temperature at the times of diatom growth and/or carbonate precipitation changed significantly, this would shift calculated d 18 O lakewater values. However, assuming a temperature coefficient of we0.2& C À1 (Brandriss et al., 1998;Moschen et al., 2005;Crespin et al., 2010), it would take an unrealistic temperature change of 75 C to explain the w15& increase in d 18 O correc-tedÀdiatom w800 AD if d 18 O lakewater was kept constant. While temperature will have some effect on isotope values, it alone cannot account for the size of the shifts seen in d 18 O corrected-diatom .
Intra-annual temperature differences are taken into account to produce estimates of d 18 O lakewater at the times of diatom growth and carbonate precipitation (Fig. 6) to try to remove this variability.

Changes in lake water d 18 O and the freshwater lid hypothesis
If the large shifts in d 18 O lakewater calculated from d 18 O correcteddiatom are not due to contamination, temperature or species effects, what is driving them? The other potentially significant variable is the d 18 O of the lake water itself. Changes in d 18 O lakewater can occur due to shifts in the source of precipitation, the d 18 O of the source water, the water balance of the lake and the type of precipitation. The effect of the first two is not considered large enough through this time period to have caused the large shifts in d 18 O correcteddiatom seen in this record (Roberts et al., 2008). d 18 O carbonate has already been shown to be a strong proxy for water balance at Nar (Jones et al., 2005;also Fig. 4) (Fig. 6) from the two hosts only show an offset of a few per mil due to differences in the time of year they grow/precipitate and error from mass balancing, it would be reasonable to assume that d 18 O corrected-diatom is also responding to changes in water balance. While the temperature range and minimum and maximum errors from the mass balance correction are plotted on  Fig. 6. However, here we propose one mechanism that would allow intra-annual d 18 O lakewater shifts at times in Nar to be greater than would be expected for a closed lake with a long residence time.
Snowfall d 18 O is significantly lower than rain d 18 O because snow reflects equilibrium conditions in the cloud rather than being in isotopic equilibrium with near-ground water vapour (Darling et al., 2006;IAEA/WMO, 2012). Normally in closed lakes, it would be expected that the effects of changes in d 18 O precipitation would be far outweighed by evaporative effects. However, an input of low d 18 O water may not immediately mix with the bulk of the lake water. Large amounts of snowmelt in the spring could form a freshwater lid on the lake surface because of the density contrast with the saline waters left over from the previous summer. If the majority of diatom growth occurred in this low d 18 O water and if by the time of carbonate formation this freshwater lid had mixed with the rest of the lake water, large intra-annual differences in epilimnion d 18 O lakewater could occur. In Canada and Greenland, freshwater lids form after large inputs of snow into lakes that have underlying saline waters (McGowan et al., 2003;Willemse et al., 2004;McGowan et al., 2008;Pieters and Lawrence, 2009). Fig. 6 shows that the lowest d However, there is some support from other proxies. As discussed in Section 5.3, the appearance of C. meneghiniana in the record appears to roughly coincide with low d 18 O corrected-diatom values (Fig. 6), suggesting there were indeed noteworthy changes in lake conditions at this time. We suggest that the difference between lake conditions at times when Dd 18 O lakewater is more negative and C. meneghiniana dominates biovolume calculations and times such as the present, when Dd 18 O lakewater is closer to 0 and N. paleacea and S. acus are the dominant bloom species, could be much larger inputs of low d 18 O snowmelt in the former that were sufficient to form freshwater lids, possibly with associated increases in the inwash of nutrients. While we acknowledge the limitations of the available documentary sources, it does appear there may have been significantly more snowfall in the region at times of increased Dd 18 O lakewater , with ancient texts from the first half of the first millennium AD, for example, describing the people of Cappadocia as "reeking of snow" and roads impassable until Easter (Van Dam, 2002). Significant snowfalls in other parts of Anatolia were also reported at this time (Stathakopoulos, 2004). Ni gde, 45 km from Nar, saw on average from 1935 to 2010 AD only 33 snowy days per year, perhaps explaining why a significant freshwater lid does not seem to have formed over the past few decades, with Dd 18 O closer to zero and no C. meneghiniana blooms. Therefore, since the other factors that could influence d 18 O corrected-diatom (namely contamination and temperature) would not be able to cause the magnitude of the shifts seen, and based on our present understanding of Nar, we propose that increased snowmelt that formed a freshwater lid is the best explanation for low d 18 O corrected-diatom values.

Implications for Near East palaeoclimatology
Previous studies have shown that summers in the Near East were dry in Late Antiquity, i.e. 300e560 AD at Nar  and 100e700 AD in Israel (Orland et al., 2009). Increased climate variability and a decline in precipitation were also inferred for central Europe around this time (Buntgen et al., 2011). However, reconstructions of the severity of winters were lacking. Here, d 18 O corrected-diatom is interpreted as indicating significantly increased snowfall for much of the period 301e801 AD (zone 1). The period 301e561 AD (zone 1a) coincided with a time of lower water balance, suggesting increased seasonality with winterdominated precipitation and dry summers. In zone 1b (561e801 AD), there seems to be less significant summer droughts with lower d 18 O carbonate but d 18 O corrected-diatom is still very low suggesting significant winter snow at this time too. This period of unfavourable climate for people w300e700 AD (the 'Dark Ages') is seen across Europe. Bond event 1 is dated to around this time (Bond et al., 1997) and a decline in agriculture in NW Europe has been linked to the cooler conditions (Berglund, 2003). The fact climate shifts seem to have been broadly synchronous from the North Atlantic to the Near East suggests a common climate driver and indeed Jones et al. (2006) showed how the winter climate of Nar is influenced by the North Atlantic Oscillation and the North Sea-Caspian Pattern Index.
Links between the climates of the North Atlantic and central Turkey are also seen in later parts of the Nar record. As described in Roberts et al. (2012) aridity w1400e1900 AD during the LIA, with drought independently recorded in tree ring and historical sources from Anatolia 1580e1610 AD (Kuniholm, 1990;Touchan et al., 2007;White, 2011). Dd 18 O lakewater suggests a less seasonal climate at this time (zone 4). However, towards the end of the LIA, a period of increased Dd 18 O lakewater is seen (zone 5), which coincides, within the dating error, with the reported freezing of the Black Sea 1823 AD (Yavuz et al., 2007) and the Dalton Sunspot Minimum 1790e1830 AD (Wagner and Zorita, 2005).

Implications for the stable isotope community
Using d 18 O corrected-diatom alongside d 18 O carbonate data we are therefore able to provide insights into the seasonality of Near East precipitation over the past 1710 years. This was possible because, as in the studies of Leng et al. (2001) and Rozanski et al. (2010), the climate of the study site is seasonal and the two hosts apparently record conditions in the lake at different times of the year. At Pınarbaş ı (Leng et al., 2001), d 18 O carbonate is seen as a proxy for summer temperature because of the short residence time of the lake, but Nar has a long residence time meaning that d 18 O carbonate here responds to water balance. Therefore, the insights provided by comparing d 18 O from carbonates and diatoms are likely to depend on the hydrological residence time of the lake and what time of the year each host grows/precipitates, emphasising the requirement for detailed limnological monitoring and acknowledgement that precipitation/growth times could vary down core. We also highlight how d 18 O carbonate is easier to interpret than d 18 O corrected-diatom because of the many factors that can influence the latter, not least changes in species assemblages shifting the time of year to which d 18 O corrected-diatom is weighted and the error associated with mass balance correc-  Leng et al. (2001), rather than being direct substitutes for each other, comparing d 18 O records from hosts that form at different times of the year can provide additional information to aid palaeohydrological reconstructions.

Conclusions
Building on previous carbonate-diatom d 18 O comparison studies, we highlight the necessity for an understanding of the modern limnology of the lake, diatom species work and a thorough assessment of error to support interpretations of oxygen isotope records. In particular, we demonstrate the potential of EDS to provide quantitative assessments of contamination in diatom isotope samples. While the d 18 O carbonate record is seen as a proxy for water balance, we suggest d 18 O corrected-diatom may be driven by changes in the amount of spring snowmelt. Thus, we tentatively propose increased snowmelt (more severe winters) for much of the periods 301e801 AD, 921e1071 AD and 1821e1898 AD.
Combining this record with the d 18 O carbonate record, we are therefore able to highlight the potential for oxygen isotope analysis from different hosts in lake sediments to provide insights into palaeo-seasonality.