Heterogenous Late Holocene Climate in the Eastern Mediterranean—The Kocain Cave Record From SW Turkey

Palaeoclimate variability must be constrained to predict the nature and impacts of future climate change in the Eastern Mediterranean. Here, we present a late Holocene high‐resolution multiproxy data set from Kocain Cave, the first of its kind from SW Turkey. Regional fluctuations in effective‐moisture are recorded by variations in magnesium, strontium, phosphorous and carbon isotopes, with oxygen isotopes reacting to changes in precipitation and effective‐moisture. The new record shows a double‐peak of arid conditions at 1150 and 800 BCE, a wet period 330–460 CE followed by a rapid shift to dry conditions 460–830 CE, and a dry/wet Medieval Climate Anomaly/Little Ice Age pattern. Large discrepancies exist between Turkish records and the Kocain record, which shares more similarities with other Eastern Mediterranean coastal records. Heterogeneity of regional climate and palaeoclimate proxy records are emphasized.

Extensive archeological and pollen investigations (e.g., Vandam et al., 2019;Woodbridge et al., 2019) make SW Turkey a suitable testbed for examining human-climate-environment interactions. However, high-resolution palaeoclimate datasets from the region only extend back ∼1,000 (tree-rings) and ∼1,400 (Lake Salda) years (Danladi & Akçer-Ön, 2018;Heinrich et al., 2013), or do not cover the late Holocene (Dim Cave; Rowe et al., 2020;Ünal-İmer et al., 2015. Stable-isotopes from Lake Gölhisar (Eastwood et al., 2007) reveal low-resolution (∼80 years) changes in lake water balance (LWB) throughout the Holocene, albeit with significant dating uncertainties of ±165 years. This record and tree-rings are seasonally biased towards spring/summer, whereas precipitation mainly occurs in winter (Peterson & Vose, 1997). High-resolution palaeoclimate archives are available from other regions of Turkey (Lake Nar, Sofular Cave; Dean et al., 2018;Göktürk et al., 2011); however, these are not local and experience wholly different climatic conditions (Section 5.1). Here, we provide a new speleothem record (Ko-1) from Kocain Cave, SW Turkey, to fill the late Holocene gap. We present highly resolved trace-element (T-E) data starting ∼950 BCE, and a stable-isotope record that extends from the present to ∼1350 BCE. An age-model is constructed from uranium-series dates ( 230 Th), with supporting evidence from the impact of historically attested earthquakes on Ko-1. This enables us to establish high-resolution climate variability in SW Turkey for >3,000 years during the late Holocene.
Precipitation (1929-2018Peterson & Vose, 1997) at Antalya exhibits a marked winter-peak, 90% occurring November-March, and high inter-annual variability, ranging from 207 mm (2008) to 1,914 mm (1969). Alike the entire EM (Lionello, 2012;Xoplaki et al., 2018), SW Turkey experiences spatial heterogeneity of climate across short distances ( Figure S2 in Supporting Information S1). Moisture is brought by westerly storm tracks (Ulbrich et al., 2012) and mountains promote orographic precipitation caused by rising moist air and associated rainout effects (Evans et al., 2004). Weather station data ( Figure 2b) reveals that despite similar seasonal patterns, coastal stations (e.g., Antalya) are significantly warmer and wetter than inland stations (e.g., Isparta). Precipitation and temperature are enhanced during negative phases of the North-Sea Caspian Pattern (NCP), Arctic Oscillation (AO), and North Atlantic Oscillation (NAO), likely linked to increased cyclonic activity and circulation over the warm Mediterranean; however, these patterns are not the same across Turkey (Sariş et al., 2010;Unal et al., 2012; Section 5.1).

Materials, Methodology and Chronology
The actively growing stalagmite (Ko-1) from Kocain Cave, was collected ∼450 m from the cave entrance in August 2005. Bedrock thickness above Ko-1 is ∼80 m. A total of 31,503 measurements of T-Es (Ca, Mg, Sr, and P) were performed on the top 156 mm at a resolution of ∼5 µm using Laser Ablation-Inductively Coupled Plasma-Mass Spectrometry (Tanner et al., 2002). For oxygen (δ 18 O) and carbon (δ 13 C) isotope measurements, the first 174.5 mm was sampled at intervals of 0.5 mm or less, providing a total of 370 measurements. Further methodological description and sample extraction locations can be found in Text S2 and Figure S3 in Supporting Information S1.
For the chronology of Ko-1, 25 230 Th ages were produced (following the analytical protocol of Cheng et al., 2013) ranging from 61 ± 51 to 3387 ± 80 BP (years before 1950 CE). Eight ages affected by significant detrital contamination ( 230 Th/ 232 Th ratios <30) were not included in the age model ( Figure 3b). The 17 remaining 230 Th ages have uncertainties varying from ±38-133 years (M = ±67) and only one, at 43 mm  (Koç et al., 2020), and dust-layer (335-485 CE) are displayed. (b) Average monthly precipitation (solid lines) and temperature (dashed lines) from weather stations in SW Turkey (Peterson & Vose, 1997). (c) Map of SW Turkey with late Holocene palaeoenvironmental archives (triangles) and weather stations (squares); colors correspond to stations in panel (b).

Interpretation of the Ko-1 Multi-Proxy Record
δ 13 C and δ 18 O values from Ko-1 were previously interpreted as reflecting changes in winter temperature and associated snow melt (Göktürk, 2011). New T-E measurements ( Figure 3) disprove this interpretation, indicating variations in the multi-proxy record can be used to characterize regional fluctuations in effective-moisture (Mg/Ca, Sr/Ca), effective-moisture/biological activity (P/Ca, δ 13 C), and effective-moisture/ precipitation amount (δ 18 O).
All Ko-1 proxy records correlate and are visually similar as all are influenced, to various extents, by changes in effective-moisture ( Figure 3a and Figure S6 in Supporting Information S1). Prior calcite precipitation (PCP) occurs when cave drip-waters reach a gas phase above the cave with lower partial pressure of carbon dioxide (pCO 2 ) than the soil gas CO 2 with which they were previously in equilibrium (McDonald et al., 2004). This enhances Mg/Ca, Sr/Ca, and δ 13 C ratios, as Ca 2+ and 12 C are preferentially deposited (Fohlmeister et al., 2020;McDermott et al., 2006). Additional PCP occurs in periods of low effective-moisture as there are more aerated spaces above the cave and longer aquifer interaction times (Fairchild & Treble, 2009;Treble et al., 2003;Tremaine & Froelich, 2013). A positive correlation between Mg/Ca and
Furthermore, agreement between high magnitude changes in the Ko-1 proxies, and other regional proxies, suggest they reflect effective-moisture (Figures 2 and 4; Section 6). Most notable is a distinct phase of high effective-moisture (330-460 CE), near-contemporaneous with a distinct brown/orange dust-layer on Ko-1 (87.3-98 mm; 335-485 CE), containing a soot layer (Koç et al., 2020), and cistern construction in Kocain Cave ( Figure 2). A prominent labarum/Chi-Rho symbol (☧) gives this cistern (∼250 m 3 capacity) a 312 CE terminus post quem (earliest possible construction date), as that is when the symbol was incorporated as a shield emblem by Emperor Constantine (Cameron & Hall, 1999), its use remained extensive until the sixth century CE (Hörandner & Carr, 2005). During numerous visits to the cave by the authors between August 2005 and April 2019, this cistern was 0%-10% full (0-25 m 3 ) and spring flow was occurring but minimal. We suggest it was built during a period of greater spring flow and this, combined with the caves large opening width, made it suitable for use by herders. This assumption is further supported by a regional increase in grazing during the Late Roman Period (300-450 CE), specifically a shift towards goat herding in "marginal" mountainside areas (De Cupere et al., 2017;Fuller et al., 2012;Izdebski, 2012;Poblome, 2015). Animal herds' use of the cistern would have mobilized fine dust from the cave floor, which was then incorporated into the stalagmite. High Fe/Ca ratios are detected in this layer, suggesting dust particles were trapped as it was precipitated (Fairchild & Treble, 2009; Figure S4 in Supporting Information S1). This mechanism could explain the dust-layer; which would usually suggest drier conditions if corresponding to increases in Mg/Ca (see Carolin et al., 2019).
High effective-moisture in the fourth and early fifth centuries CE is also evidenced by wild weeds that require high moisture availability growing in the territory of Sagalassos (Bakker et al., 2012;Kaniewski et al., 2007); wetland conditions and spring reactivation in the Bereket Valley and Gravgaz Marsh Kaptijn et al., 2013;Van Geel et al., 1989;Vermoere et al., 2002); and deep-water conditions at Lake Burdur (Tudryn et al., 2013). Similar changes are evidenced in EM proxies, suggesting this wet phase was a regional phenomenon (see below). The above interpretations, and corroborating evidence, strengthen our claim that decreases in Ko-1 Mg/Ca, Sr/Ca, δ 13 C and δ 18 O, with increases in P/Ca, are indicative of wetter climatic conditions in SW Turkey.

Ko-1 Record and EM Palaeoclimate
Key palaeohydrological changes for SW Turkey are reflected in geochemical proxies in Ko-1 (Figure 3, Figure S7 in Supporting Information S1). First, distinct phases of low effective-moisture are centered at ∼1150 and ∼800 BCE, with intervening wetter conditions between ∼1000 and 900 BCE. Second, high effective-moisture occurred between ∼330 and 460 CE, followed by a rapid shift to drier conditions that lasted until ∼830 CE. Finally, there was a dry/wet Medieval Climate Anomaly (MCA; 850-1300 CE)/Little Ice Age (LIA; 1400-1700 CE) pattern, with high variability during 1450-1550 CE.
Wet conditions between ∼330 and 460 CE rapidly shift to an arid phase between ∼460 and 830 CE in the Ko-1 record, roughly coincident with the Dark Ages Cold Period (450-800 CE; Helama et al., 2017;Figure 4). An effective-moisture peak in SW Turkey is supported by local palaeoenvironmental evidence and archeological evidence in Kocain Cave (see above). Similar wet peaks are observed across the EM at ∼300-500 CE. Speleothem δ 18 O data from Mavri Trypa and Skala Marion caves demonstrate wet conditions at ∼300-350 CE (Finné et al., 2017;Psomiadis et al., 2018). Effective-moisture proxies from Lake Trichonida show an apparently delayed response, with the records wettest phase between ∼420 and 500 CE (Seguin et al., 2020). Reconstructed precipitation based on Dead Sea data suggests ∼350-490 CE may be the wettest interval in the late Holocene for the southern Levant, whereas a depletion of isotopes from Jeita Cave suggests a break from arid conditions between ∼320 and 400 CE (Cheng et al., 2015;Morin et al., 2019).
Enhanced variation in effective-moisture is evidenced in the Ko-1 record from ∼800 until 1850 CE (Figures 3 and 4). From ∼900 CE until ∼1460 CE, drier conditions prevailed, with more-humid intervals every ∼120-150 years (∼1030, ∼1180, ∼1300 CE), encompassing the MCA. Hydroclimate was highly variable between ∼1450 and 1550 CE, experiencing an extreme dry-wet-dry-wet pattern. The driest conditions in the entire Ko-1 record occur between 1510-1530 CE, indicated by the highest δ 18 O value and 15-year Mg/ Ca and P/Ca averages. Subsequently, effective-moisture was still highly variable but elevated until ∼1840 CE, a period roughly coincident with the LIA (1400-1850 CE). Reconstructed winter-spring temperatures from tree-rings in Jsibeli suggest cooling after ∼1500 CE, with the coldest conditions at ∼1750 CE (Heinrich et al., 2013), when there was a break from high effective-moisture at Kocain (Figure 4).

Heterogeneity of Eastern Mediterranean Climate and Proxies
Large discrepancies exist between the Ko-1 record of effective-moisture and other hydrological proxies from the EM, most likely caused by: (a) spatial climate variations and challenges in palaeoclimate analysis, related to (b) interpretation of different types of proxies with varied sensitivity to hydroclimatic change, and (c) chronological uncertainties. The greatest differences between records discussed here are observed between Ko-1 and other records from Turkey. Climatic heterogeneity in SW Turkey is more extreme across the large country (780,000 km 2 ), which has complex and diverse topography, and numerous moisture sources (Lionello, 2012;Xoplaki et al., 2018). These factors lead to varied temperatures (Aydın et al., 2019), seasonal patterns (Sariş et al., 2010), and impacts from teleconnections (Ünal-İmer et al., 2015;Unal et al., 2012).
The two other high-resolution Turkish records that contrast with Ko-1, Lake Nar (central Anatolian plateau: CAP) and Sofular Cave (NW Turkey; Black Sea coast), are in completely different climatic regions. The high elevation CAP region experiences low precipitation (m = 455 mm/yr −1 ), with two peaks (April-May/ October-December), and cold semi-arid and dry continental climates (Öztürk et al., 2017;Peel et al., 2007). The Black Sea coast is temperate, with precipitation of a similar magnitude to SW Turkey (m = 915 mm/ yr −1 ), but there is no dry season and precipitation is high throughout the year (Göktürk et al., 2008Karaca et al., 2000). The impact of large-scale atmospheric teleconnections (NCP, AO, NAO) also differs in these regions, compared to SW Turkey which has enhanced precipitation and temperature during negative phases Sezen & Partal, 2019). Negative phases cause higher temperatures across Turkey, particularly in winter. However, the CAP experiences significantly greater increases (Kutiel & Türkeş, 2005;Türkeş & Erlat, 2009). Impacts on precipitation are more varied. The Black Sea weakens the impacts of teleconnections on precipitation in NW Turkey Türkeş & Erlat, 2003). AO-and NCPphases cause their most significant impact on precipitation in SW Turkey (Kutiel & Benaroch, 2002), with CAP only impacted by AO-phases in winter (Sezen & Partal, 2019) and the transition between enhancements/reductions in precipitation from NCP-phases located <50 km from Nar Kutiel & Türkeş, 2005). NAO influence is weaker and focused on the western and central regions (Unal et al., 2012). These differences lead to spatial variations in droughts, which impact each record differently (Figure 1 and Figure S10 in Supporting Information S1; Vicente-Serrano et al., 2010). Lake Nar records LWB, with higher δ 18 O corresponding to hydrological droughts (lake-water deficits) . Speleothems record fluctuations in effective-moisture, which are more akin to agricultural droughts (soil-moisture availability) (Fleitmann et al., 2009;Göktürk et al., 2011). However, none of these records are simple, being influenced by multiple climatic and geological/geographical factors, the importance of which changes over time. Additionally, proxies represent different seasons. The carbonate δ 18 O record from Lake Nar is primarily deposited in early summer in response to evaporation and aridity (Dean et al., 2015). Speleothem records are winter-season biased due to the lighter-isotopic signature of winter precipitation and seasonality of precipitation (e.g., in SW Turkey).
The impact of temperature change on precipitation and proxy records is also poorly understood and variable. Antalya GNIP data shows a negative correlation between precipitation and temperature (r = −0.53, p < 0.0001; Figure S8 in Supporting Information S1). However, proxy records show both increases and decreases in effective-moisture during periods with lower temperatures: during the CYCE effective-moisture is low, but during the LIA effective-moisture is high at Kocain Cave (Figure 4).
Comparison between records is further complicated by chronological uncertainties of decadal-centurial length in lake and speleothem records. Multiple and varied lags are present between climatic changes in different regions, and between climatic shifts and their signal in records. Different resolutions hinder comparison and the specifics of resolutions, that is, whether a sample is an average across a large period or a specific point in time, are rarely addressed.

Conclusion
Stalagmite Ko-1, from Kocain Cave, provides the first highly resolved, well-dated palaeohydrological proxy record covering the late Holocene for SW Turkey. Key periods of palaeoclimatic change are revealed, notably: (a) a double-peak of arid conditions (1150 and 800 BCE), (b) a distinct period of high effective-moisture in the fourth and fifth centuries CE (∼330-460 CE), followed by (c) a rapid shift to low effective-moisture (460 CE) that persisted until ∼830 CE, and finally (d) a dry/wet MCA/LIA pattern. Changes were often in contrast to palaeoclimate records from northern and central Turkey, and sometimes locally, more frequently correlating with changes in coastal records from the Aegean and Levant regions. Considering the heterogeneity of climate and the multitude of impacts on records, palaeoclimatic interpretations are complex and care must be taken, especially when they are utilized for discussions of societal impacts.

Data Availability Statement
The new Kocain speleothem (Ko-1) uranium-series, trace-element and stable-isotope data used in this study are available from the NOAA palaeoclimate archive via https://www.ncdc.noaa.gov/paleo/study/33854 and in the Supporting Information S1. The isotope and trace-element data were provided by the National Environmental Isotope Facility and the Department of Earth Science, University of Cambridge, respectively.