Constraining Holocene hydrological changes in the Carpathian–Balkan region using speleothem δ 18 O and pollen-based temperature reconstructions

Abstract. Here we present a speleothem isotope record (POM2) from Ascunsă Cave (Romania) that provides new data on past climate changes in the Carpathian–Balkan region from 8.2 ka until the present. This paper describes an approach to constrain the effect of temperature changes on calcite δ18O values in stalagmite POM2 over the course of the middle Holocene (6–4 ka), and across the 8.2 and 3.2 ka rapid climate change events. Independent pollen temperature reconstructions are used to this purpose. The approach combines the temperature-dependent isotope fractionation of rain water during condensation and fractionation resulting from calcite precipitation at the given cave temperature. The only prior assumptions are that pollen-derived average annual temperature reflects average cave temperature, and that pollen-derived coldest and warmest month temperatures reflect the range of condensation temperatures of rain above the cave site. This approach constrains a range of values between which speleothem δ18O changes should be found if controlled only by surface temperature variations at the cave site. Deviations of the change in δ18Ocspel values from the calculated temperature-constrained range of change are interpreted towards large-scale variability of climate–hydrology. Following this approach, we show that an additional ∼0.6‰ enrichment of δ18Oc in the POM2 stalagmite was caused by changing hydrological patterns in SW Romania across the middle Holocene, most likely comprising local evaporation from the soil and an increase in Mediterranean moisture δ18O. Further, by extending the calculations to other speleothem records from around the entire Mediterranean basin, it appears that all eastern Mediterranean speleothems recorded a similar isotopic enrichment due to changing hydrology, whereas all changes recorded in speleothems from the western Mediterranean are fully explained by temperature variation alone. This highlights a different hydrological evolution between the two sides of the Mediterranean. Our results also demonstrate that during the 8.2 ka event, POM2 stable isotope data essentially fit the temperature-constrained isotopic variability. In the case of the 3.2 ka event, an additional climate-related hydrological factor is more evident. This implies a different rainfall pattern in the Southern Carpathian region during this event at the end of the Bronze Age.


Introduction
In the region surrounding the eastern Mediterranean, proxy records suggest that conditions were more humid during the early Holocene compared to present-day moisture budgets (Rossignol-Strick, 1999;Rohling et al., 2002). Enhanced freshwater flux roughly between 10 000 years before present (10 ka) and 6 ka led to stratification and sapropel formation in the eastern Mediterranean. At the same time, depleted δ 18 O values in lacustrine calcareous microfossils and endogenic carbonate deposits suggest less evaporation across the region during that time compared to the present day. A stacked oxygen isotope record generated from lake proxies around the eastern Mediterranean shows a general drying trend between 6 and 4 ka (Roberts et al., 2008). A pattern of increasing δ 18 O values with time from the early to mid-late Holocene is documented in speleothem records from south-central Europe and the eastern Mediterranean (McDermott et al., 2011, and references therein). This has been suggested by McDermott et al. (2011) to reflect more efficient rainout from an Atlantic source of moisture and less of it reaching the eastern Mediterranean region during the late Holocene.
Due to the topographic complexity and rather sparse data distribution, reports from across the Carpathian-Balkan region do not provide yet a common view on past environmental change, but rather point to possibly contrasting Holocene hydroclimatic evolution at the regional scale (Feurdean et al., 2008;Magyari et al., 2013). Lake records from the southern Balkans, such as Ioannina, Greece (Frogley et al., 2001), and Prespa and Ohrid, Macedonia (Leng et al., 2010) indicate high humidity throughout the early Holocene, whereas palaeolimnological records from Steregoiu, NW Romania (Feurdean et al., 2007) and Sfânta Ana, central Romania (Magyari et al., 2009), suggest that lower humidity persisted in the area. At Sfânta Ana, a volcanic crater lake with no outflow, water levels began to rise only after 7.4 ka (Magyari et al., 2009).
Several Holocene speleothem records are available from the Romanian Carpathians (Onac et al., 2002;Tȃmaş et al., 2005;Constantin et al., 2007). Trends towards higher values seen in these time series throughout the Holocene were interpreted as reflecting rising temperatures. McDermott et al. (2011) expanded the interpretation of individual European speleothem records and suggested a decreasing rainout gradient across the Holocene along a longitudinal transect. However, a more specific distinction between hydrology and temperature-driven changes is not easily achieved because interpreting stable oxygen isotope records from speleothems in terms of palaeoclimate is not generally straightforward (McDermott, 2004;Lachniet, 2009;Tremaine et al., 2011). For instance, the effects of temperature and climate-hydrology changes may cancel each other out, for example if there is a change in rainfall seasonality. Changes in seasonality of both rainfall and calcite precipitation are difficult to detect (Baker et al., 2011). Furthermore, moisture sources and transport trajectories, which generally affect the stable isotopic composition of meteoric water in Europe (Rozanski et al., 1982), may respond to regionalscale climate changes in contrast to local ones. Consequently, specific temperature or hydrological information is rarely directly quantifiable from speleothem stable isotope records. An example is the muted or even absent signal around the 8.2 ka cold event in speleothem δ 18 O records throughout Romania (Tȃmaş et al., 2005;Constantin et al., 2007), despite this event being clearly identifiable in peat bog pollen records (Feurdean et al., 2007), in Balkan lake records (Pross et al., 2009;Panagiotopoulos et al., 2013), as well as in the Aegean Sea (Marino et al., 2009). This ambiguity of speleothem δ 18 O records with respect to climatic events and transitions raises the question of how more specific information on the nature of climate change can be extracted from this proxy.
In this study we present a new speleothem isotopic record from Ascunsȃ Cave located on the eastern slopes of the Carpathian Mountains in Southern Romania (Fig. 1), in an area under periodical Mediterranean hydroclimate influences (Bojariu and Paliu, 2001;Apostol, 2008). We combine information from regionally averaged pollen-based temperature reconstructions from Europe across the Holocene (Davis et al., 2003) with the new oxygen isotope data from Ascunsȃ Cave, alongside a detailed comparison with published speleothem records from Romania (Onac et al., 2002;Tȃmaş et al., 2005;Constantin et al., 2007) and the Mediterranean (McDermott et al., 1999;Bar-Matthews et al., 2003;Drysdale et al., 2006;Vollweiler et al., 2006;Verheyden et al., 2008;Fleitmann et al., 2009). We also attempt to constrain the regional-scale hydrologic information inherent by speleothem δ 18 O change across the Holocene, focusing on Mediterranean climate trends observable after 6 ka  Dragotȃ and Baciu, 2008). (Mayewski et al., 2004;Roberts et al., 2008Roberts et al., , 2011McDermott et al., 2011). Finally, local pollen data sets (Feurdean et al., 2008;Bordon et al., 2009) are used to constrain selected rapid climate shifts associated with the 8.2 ka and the 3.2 ka events.

Cave setting and stalagmite characteristics
Ascunsȃ Cave is located on the eastern slopes of Mehedinţi Mountains, Southern Carpathians (45.0 • N, 22.6 • E, 1050 m alt.) in southwestern Romania (Fig. 1). It is a 400 m long and over 200 m deep contact cave developed by river erosion of Turonian-Senonian wildflysch (mélange) below an Upper Jurassic-Aptian limestone cover (Codarcea et al., 1964).
The cave is well decorated with speleothems and throughout its course there is a chaotic mixture of collapsed blocks and speleothem fragments reflecting the undermining of the wildflysch walls by fluvial erosion or their failure to support massive flowstone formations.
The analysed stalagmite (POM2) is 77.4 cm long and composed of well-laminated and densely compacted white calcite (Figs. 3 and 4). Topographic survey at the cave site revealed that limestone thickness above the stalagmite sampling site is ∼ 100 m.

Present day climatology of the study area and cave monitoring
The regional climate of the Romanian Carpathians is temperate-continental, characterised by a predominantly Atlantic origin of air masses (Baltȃ and Geicu, 2008). It is also influenced (in the southwestern part) by Mediterranean cyclonic activity that is responsible for milder temperatures and increased winter rainfall in the area of the study site compared to northern or eastern Carpathians (Bojariu and Paliu, 2001;Apostol, 2008). Most of the cyclones affecting the study area originate in the central Mediterranean (around the Gulf of Genoa), but cyclones from the Aegean Sea also reach this region periodically (Apostol, 2008). Seasonal variability is observed in the formation of these cyclones, as shifts of the polar jet stream in winter affect Mediterranean cyclogenesis (Trigo et al., 2002). Figure 2 illustrates the seasonal differences in precipitation recorded between 1961 and 2000 at two meteorological stations relevant for this study, Drobeta (SW Romania) and Stâna de Vale (W Romania) (data from Dragotȃ and Baciu, 2008). There is a clear difference in rainfall seasonality between the two regions, with Stâna de Vale having one rainfall peak in the summer, whereas at Drobeta two main rainfall periods are peaking in spring and early winter (Fig. 2).
Ascunsȃ Cave was monitored between July 2012 and November 2013 for atmospheric physical parameters and drip water isotopic composition. Temperature (T), relative humidity (RH) and CO 2 concentrations were measured at three sampling locations (POM A, POM2, and POM B) within Ascunsȃ Cave, using two Vaisala probes, GMP222 for CO 2 and HMP75 for T and RH. Drip water collected from stalactite tips at the sampling sites was analysed for δ 18 O and δD on a Picarro L2130-i Cavity Ring-Down Spectroscope at Babeş-Bolyai University (Cluj-Napoca, Romania) following the method described by Brand et al. (2009). The analytical precision is better than ±0.03 ‰ for δ 18 O and ±0.07 ‰ for δD. For data normalisation, two laboratory reference waters (VEEN and HTAMP) that were calibrated directly against VSMOW were measured repeatedly in each run. Results are expressed in ‰ on the VSMOW scale.

U-series dating and stable isotope analysis of speleothem samples
For U-Th dating, calcite samples were analysed on a THERMO Neptune MC-ICP-MS following procedures outlined in Hoffmann et al. (2007) and Hoffmann (2008). In total, 14 U-Th samples were measured, covering the entire length of the stalagmite. Three pairs of samples were drilled immediately underneath and above visible changes in growth axis at 43.4, 54.4 and 63.9 cm. A total of 150 stable isotope samples were hand drilled at 5 mm resolution using a 0.5 mm drill bit. All samples were analysed at the University of Oxford on a Thermo Delta V Advantage mass spectrometer equipped with a Kiel IV Carbonate Device. Results are reported relative to the Vienna Pee Dee Belemnite (VPDB) standard, and external precision on replicate samples (NBS 18,NBS 19, and a local carbonate standard) run daily on this system was 0.06 ‰ for δ 18 O and 0.03 ‰ for δ 13 C.

U-series dating results and growth model
The U-Th ages suggest that the stalagmite started growing around 17.2 ka, but most of the growth occurred between 8.2 ka and the present. The age model for the Holocene part of the stalagmite (Fig. 5) is based on 11 U-Th ages with typical dating uncertainties ranging between 1 and 6 % (2 σ ) ( Table 1). The stalagmite was active at the time of sampling, thus the age at the top (77.4 cm) is assumed to be 0 (relative to 2009, the year of sampling) and is used as an additional tie point in the growth model calculation. The 238 U concentration varies between 18 and 50 ng g −1 , and 232 Th concentration ranges between 0.1 and 12.2 ng g −1 . Dating uncertainties are therefore mainly resulting from small U concentration and a significant correction for initial Th, combined with the young age of the stalagmite which yields low 230 Th / 232 Th activity ratios (< 10) for six of the age determinations. Two samples (POM 09-2 / III and POM 09-2 / VI) were entirely dominated by detrital Th, with 230 Th / 232 Th activity ratios < 0.5 and have not yielded resolvable U-Th ages. We measured U and Th isotopes on a sample from the top of the actively forming stalagmite in order to assess a reliable correction factor. The results show a 238 U concentration of 18.3 ± 0.1 ng g −1 and a 232 Th concentration of 12.2 ± 0.1 ng g −1 . The measured 230 Th in the top sample is assumed to be entirely of detrital origin and the apparent age of 8.3 ka a result of initial thorium contamination. The 230 Th/ 232 Th activity ratio of this sample is 0.6 ± 0.05, which indicates detrital activity ratios for 230 Th/ 232 Th, 234 U/ 232 Th and 238 U/ 232 Th of 0.6 ± 0.05, if we assume the detritus to be in secular equilibrium. We note that this factor is well within the range of the bulk earth value of 0.8 ± 0.4 (Wedepohl, 1995). We therefore use the value of 0.6 with a conservative uncertainty of 50 % to correct for initial Th.
The growth model of stalagmite POM2 (Fig. 5) was generated using the StalAge algorithm of Scholz and Hoffmann (2011).

Cave monitoring results
Monitoring data show a stable average temperature of 8.2 ± 0.6 • C at the stalagmite site. Relative humidity is also stable around 94 ± 2.5 % during the year, especially at sampling sites POM2 (where stalagmite POM2 was sampled) and POM B situated deeper inside the cave (Table 2).
Isotope measurements of drip waters at POM2 site show rather consistent values for both δ 18 O (−10.57 ± 0.04 ‰) and δD (−70.58 ± 0.20 ‰), during the autumn-winter months. This may indicate an efficient mixing of waters in the aquifer, without capturing any individual rain events.
Analysis of calcite farmed on glass plates also revealed relatively constant values with mean δ 13 C of −10.30 ± 0.8 ‰ and δ 18 O of −7.91 ± 0.2 ‰ for both POM2 and an adjacent stalagmite, POM X (Table 3).
We constructed a local drip water line for Ascunsȃ Cave (Fig. 6) using δ 18 O and δD values of drip waters from all three sampling sites. Compared to the global (GMWL) and Mediterranean (MMWL) meteoric water lines, the Ascunsȃ groundwater line (AGWL) is defined as δD = 6.9 × δ 18 O + 2 and plots above the GMWL. This suggests a mixture of Atlantic and Mediterranean moisture. The slope of the local water line is slightly lower than that of the GMWL and MMWL, respectively. This may be an indication of slight enrichment due to evaporation from the soil above the cave.
The resulting value of −8.5 ± 0.1 ‰ is slightly below the average of −7.9 ‰ measured on calcite farmed at POM2 and POM X sites. The 0.6 ‰ offset from generally predicted values could indicate some kinetic fractionation. However, the average drip water value derived from point-sampling in time may not exactly represent the annual amount-weighted average rainfall and its isotopic composition. As such, calcite precipitation may still be considered to have taken place close to equilibrium during the monitored period. The offset is somewhat larger when the calibration of Kim and O'Neill (1997) or Day and Henderson (2011) are used (−9.2 ‰ and −9.6 ‰, respectively). In this study, we base our calculations on the empirical equation of Tremaine et al. (2011), which appears to better characterise conditions in Ascunsȃ Cave.

The δ 18 O record in Romanian speleothems
The most prominent feature of the δ 18 O isotopic profile of stalagmite POM2 (Fig. 7) is a trend towards higher values during the middle Holocene. This trend shows some similarity to the stacked eastern Mediterranean lacustrine δ 18 O record (Roberts et al., 2011). In order to place the Ascunsȃ Cave δ 18 O trend across the mid-Holocene into a regional context, we compare our record with δ 18 O profiles of stalagmites from Poleva Cave (Constantin et al., 2007), Urşilor Cave (Onac et al., 2002) and V11 Craig, 1961) and Mediterranean meteoric water lines (MMWL; Gat and Carmi, 1970), as well as Ascunsȃ Cave groundwater line.
longitude at low altitudes (Fig. 8a). The general trends of measured values agree to a variable degree with modelled Rayleigh distillation of Atlantic air masses at low altitude (McDermott et al., 2011). A somewhat more negative measured δ 18 O overall at Ascunsȃ Cave ( Fig. 8b) could reflect the higher altitude of the cave and higher rainout on route leading to increasingly lower δ 18 O in the moisture arriving at the cave site, but the observed trend may not be explained completely by this model. Poleva Cave shows slightly higher calcite δ 18 O values after 4 ka (−7.6 ‰) in comparison to > 6 ka (−8.3 ‰ ) and Constantin et al. (2007) interpret this increase as a general warming trend. At Urşilor Cave (NW Romania), late Holocene δ 18 O values are slightly higher (by 0.2 ‰) than during the middle Holocene and Onac et al. (2002) suggests that an apparent lack of variability at this site reflects relatively stable climate conditions. Well-known rapid climate change events (Mayewski et al., 2004) such as the 8.2 ka event are not always clearly expressed in Romanian speleothem δ 18 O records (Onac et al., 2002;Tȃmaş et al., 2005;Constantin et al., 2007). This absence in the speleothem δ 18 O record is in contrast to reports from the marine records from the eastern Mediterranean (Rohling et al., 2002), and lacustrine records from the wider Balkan region (Feurdean et al., 2008, Pross et al., 2009). An exception is a common negative excursion occurring at ∼ 3.2 ka that is recorded in the δ 18 O time series of Ascunsȃ and Poleva caves in the Southern Carpathians. This century-long cold event has been identified in marine records  from the eastern Mediterranean (Rohling et al., 2002) and in lacustrine records from the Balkan region (Feurdean et al., 2008).

Constraining regional temperature change in the speleothem δ 18 O record with independent temperature reconstructions
Stable oxygen isotopes in individual speleothems are potentially influenced by local effects, such as cave hydrology and cave ventilation, which may obscure the regional climate signal (Tremaine et al., 2011;Riechelmann et al., 2013). Here, we employ coeval data recorded in more than one cave to account for such potential biases. We specifically address the three features reported above: (1) the general mid-Holocene trend by comparing the isotopic difference between 2000 years averaged time intervals between the early and late Holocene, from 8-6 ka and 4-2 ka, respectively (Fig. 8a); (2) the absence of an unambiguous 8.2 ka event in isotopic speleothem records from Romania, and (3) the nature of a clear isotope excursion ∼ 3.2 ka in two Southern Carpathian speleothems. The principal controls of oxygen isotope fractionation during speleothem calcite precipitation are temperature in the cave and isotopic composition of drip water (e.g. Day and Henderson, 2011;Tremaine et al., 2011). Both aspects directly respond to changes of annual average air temperature above the cave. In addition, drip water δ 18 O may also record variations in hydrologic climate characteristics, such as rainfall seasonality, evaporation, moisture sources and rainout efficiency along the path (McDermott, 2004;Fairchild et al., 2006;Lachniet, 2009). The two principal contributors to change -temperature and climate hydrology -may be separated as follows: assuming that (1)   1000 ln α = 16.1(10 3 T −1 ) − 24.6. (1) For the δ 18 O-temperature relationship in rainwater we use the empirical global mid-latitude relationship suggested by Rozanski et al. (1993):  several Romanian speleothem records in order to identify the likelihood of additional changes in other climate-hydrology related parameters (e.g. rainfall seasonality, evaporation from the soil, or variable moisture sources and pathways). If only temperature variability contributed to the observed difference between two time intervals of interest, δ 18 O c_spel will plot inside the expected pollen temperatureconstrained range of values (range of δ 18 O c_ptc ) calculated from the pollen input data TANN, MTCO and MTWA. If rainfall seasonality changes along with temperature, δ 18 O c_spel may still fall inside the range of δ 18 O c_ptc -if changes in rainfall seasonality are relatively small -or outside, depending on the combination of changing temperature seasonality and rainfall seasonality. Example calculations can be easily performed (not shown) and they demonstrate that, for example, in the case of warming an additional change towards a more pronounced rainfall seasonality with a higher proportion of summer rain would result in δ 18 O c_spel falling above the range of δ 18 O c_ptc . If rainfall seasonality changed towards more winter rain under the same warming scenario, δ 18 O c_spel would fall below the range of δ 18 O c_spel from the range of δ 18 O c_ptc . A cooling scenario instead of a warming one would produce the same relationship between δ 18 O c_spel and δ 18 O c_ptc , but the sign of both δ 18 O would be negative. A significant change in rainfall seasonality in addition to a change in any other climatehydrology parameter -such as evaporation from the soil or moisture source and pathways, for example the relative contribution of Atlantic and Mediterranean moisture -would further add to the distance of δ 18 O c_spel from the range of δ 18 O c_ptc .

Temperature and hydrology-related changes in speleothem δ 18 O records from Romania and the Mediterranean basin
We analyse the broad δ 18 O change across the time interval 6-4 ka as outlined above for the Romanian stalagmites and also for a selection of southern European records (Fig. 9). The pollen-based temperature reconstruction by Davis et al. (2003) divides Europe into six main regions: northwestern (NW), northeastern (NE), central-western (CW), centraleastern (CE), southwestern (SW) and southeastern (SE). The boundary between central and southern zones is 45 • N, the boundary between western and eastern zones is 15 • E. This places the Alps and much of northern Italy inside the CW zone and divides Romania between the CE and SE zone along the Southern Carpathians. Across the last 8000 years the CW shows a slight winter warming, the CE zone shows only little change, the SW zone shows a 2 • C warming trend for both seasons, and the SE zone shows a 1 • C warming during summer (Davis et al., 2003). The specific regionally aver- aged pollen data sets used to calculate isotopic variability in different cave records are summarised in Table 4 and calculation details are given in Table 5. Potential shortcomings of the chosen pollen zones are also discussed in the Appendix. For Ascunsȃ Cave, which is inside the SE pollen zone of Davis et al. (2003), speleothem δ 18 O 6−4 ka is 0.69 ‰, whereas values expected from the pollen-based temperature reconstruction are between 0.16 ‰ (summer) and −0.05 ‰ (winter) (Fig. 9). This implies that across the middle Holocene transition speleothem δ 18 O values at the cave site became higher than expected if only temperature change occurred. As δ 18 O increases in the majority of observed speleothems from the eastern Mediterranean domain across the 6-4 ka interval beyond the temperature-controlled amount (Fig. 9), there must have been a common large-scale change of climate-hydrology. This may include a combination of change in rainfall seasonality with any other hydrological factor, such as local evaporation, a change in the proportion of Atlantic vs. Mediterranean moisture source (Rozanski et al., 1993), or a change in the isotopic composition of the two vapour sources.
To rule out local climate effects, we compare our speleothem record with other isotope records from Poleva (Constantin et al., 2006) and Urşilor (Onac et al., 2002) caves. Considering that the Davis et al. (2003) CE pollen zone is not well represented around 45 • N latitude in Romania, we use the local temperature reconstruction derived from the pollen record of Steregoiu (Feurdean et al., 2008) for Urşilor Cave. Figure 9 shows that measured δ 18 O 6−4 ka at Poleva is similar to that at Ascunsȃ and falls well outside the pollen temperature-constrained range of change, whereas Urşilor falls within the constrained range close to the summer temperature value. Even if the Urşilor record is completed with nearby V-11 data to complement the 8-6 ka BP interval, δ 18 O 6−4 ka remains fully explained by δ 18 O c_ptc (see Fig. 9). This is despite the fact that due to the higher altitude of V-11 cave, its δ 18 O data are most likely skewed www.clim-past.net/10/1363/2014/ Clim. Past, 10, 1363-1380, 2014  Figure 9. Comparison of isotopic changes in stalagmites across different longitudes in Europe in the interval 6-4 ka, δ 18 O c_spel , with pollen temperature-constrained range of δ 18 O c_pct (summer and winter) calculated by using data from Davis et al. (2003) and Feurdean et al. (2008). For Urşilor Cave, two options are shown: (I) δ 18 O 6−4 ka based on the Urşilor data alone, (II) δ 18 O 6−4 ka based on the combination of Urşilor and V-11 isotope data.
towards lower values in comparison to Urşilor (V-11 is at 1254 m, Urşilor is at 486 m). Altogether, the data suggest that the 6-4 ka transition at Ascunsȃ and Poleva in the Southern Carpathians marks a significant hydrologic change in southern Romania that is not obvious in the Apuseni Mountains of NW Romania. A similar gradient in the expression of Holocene climate change can also be found in other parts of Europe at these latitudes (see review in Magny et al., 2013).
To verify a suspected large-scale gradient in the nature of climate change across Southern Europe, we calculated similar pollen temperature-constrained δ 18 O values for several cave records around the Mediterranean, across the same 6-4 ka transition. These records are: Grotte de Clamouse, France (McDermott et al., 1999), in combination with the Davis et al. (2003) SW European temperature time series; Buca della Renella, Italy (Drysdale et al., 2006) and Span-nagel Cave, Austria (Vollweiler et al., 2006), each in combination with the Davis et al. (2003) CW European temperature time series; and finally Sofular Cave, Turkey (Fleitmann et al., 2009), Soreq Cave, Israel (Bar-Matthews et al., 2003 and Jeita Cave, Lebanon (Verheyden et al., 2008), each in combination with the Davis et al. (2003) SE European temperature time series. The pollen reconstructions show rising temperatures throughout the year for the SW zone (Davis et al., 2003), whereas in the CW zone, increasing winter temperatures offset decreasing summer temperatures. The results are also shown in Fig. 9. It is apparent that isotope values from the western Mediterranean (Clamouse) and CW Europe south of the Alpine divide (Renella and Spannagel) show a change in δ 18 O that is explained almost entirely by pollen-constrained temperature change. Only at the Renella site, could a small hydrologic influence be argued for, as the observed change is close to the summer end of the range (Fig. 9). This likely represents a shift towards the present-day rainfall domination by the winter season (Scholz et al., 2012). On the other hand, sites in Turkey (Sofular), Israel (Soreq), Lebanon (Jeita) and southern Romania (Poleva and Ascunsȃ) that are potentially influenced by the eastern Mediterranean plot well above the temperature-constrained change. Thus, it appears that across the middle Holocene transition, the entire eastern Mediterranean domain including the Southern Carpathians underwent a significant moisture-balance change in addition to some temperature change, whereas in the western Mediterranean and the southern Alps such change in climate-hydrology is not apparent.
In principle, three aspects may have contributed to the observed hydrologic change in the east: (1) a significant change in rainfall seasonality -in this case a higher proportion of summer rain, (2) a change in evaporation from the soil, (3) a change in proportion of moisture source or a different isotopic composition of the moisture sources. Higher δ 18 O values in the speleothems after the change could be the result of more rain falling in the warm summer months (higher δ 18 O w ) during the late Holocene, but this scenario is unlikely because large-scale subsidence over the eastern Mediterranean region due to the Asian monsoon most likely strengthened (Staubwasser and Weiss, 2006). As a result, there is vir- Past, 10, 1363-1380, 2014 www.clim-past.net/10/1363/2014/ V. Drȃguşin et al.: Constraining Holocene hydrological changes in the Carpathian-Balkan region 1373 Figure 10. Comparison of isotopic changes in speleothems ( δ 18 O c_spel ) for the 8.2 ka event (left) and 3.2 ka event (right) in stalagmites from Ascunsȃ and V11 caves with pollen temperatureconstrained range of δ 18 O c_pct (summer and winter) calculated by using data from Steregoiu peat bog (Feurdean et al., 2008 andLake Maliq (Bordon et al., 2009). tually no rainfall during summer in the Levant, and rainfall is significantly reduced over SE Europe. Precipitation is at a minimum in August in SW Romania (Fig. 2). Subsidence and accompanying low humidity over the eastern Mediterranean may, however, have increased evaporation from the soil, which would be in agreement with rising summer temperatures in SE Europe across the Holocene (Davis et al., 2003). Evaporation in the soil and epikarst drives drip water δ 18 O towards more positive values, resulting in higher δ 18 O in speleothem calcite (Bar-Matthews et al., 1996;Fairchild et al., 2006). A measurable effect of evaporation at Ascunsȃ may be implied from the slope of the local drip water line (see Sect. 3.2). Alternatively, the proportion of summer rain could have been higher just because winter rainfall decreased. McDermott et al. (2011) suggested lower rainout efficiency during winter along a west-east transect across central Europe. However, higher winter temperatures were only observed in the western Mediterranean region (Davis et al., 2003, SW Europe quadrant). This would increase the temperature gradient between SW Romania and the source of cyclones in the Gulf of Genoa -which is inside the SW Europe quadrant -possibly increasing rainout efficiency in South Europe. An interesting analysis of synoptic-scale mid-tropospheric (500 mbar) pressure distribution of exceptionally wet winters in the Levant points to a possible physical mechanism causing the climate changes discussed above (Enzel et al. 2004). Such years coincide with a large-scale negative pressure anomaly centred over Asia Minor and the Middle East that encompasses all the eastern Mediterranean speleothem sites discussed above. Scenarios (1) enhanced summer evaporation and (2) reduced winter rainfall notably do not exclude each other.
The third factor that may have added to the observed increase in speleothem δ 18 O is a change in the isotopic composition of the Mediterranean mixed layer. A steady increase in δ 18 O by approximately 0.7 ‰ was derived from δ 18 O in tests of surface dwelling foraminifera and an independent sea surface temperature estimate for the Aegean Sea between 8 and 4 ka BP, which is uncorrelated with the abundance of cold water species (Marino et al., 2009). In the Adriatic Sea, an increase by 0.5 ‰ was recorded in δ 18 O of foraminifera at the same time (Siani et al., 2010). As such, a combination of warmer summer temperatures, enhanced evaporation from the soil, and higher δ 18 O of the eastern Mediterranean moisture source are currently the favoured explanation for the observed increase in δ 18 O of speleothems across the mid-Holocene from the eastern Mediterranean domain.

The δ 13 C record
Interpretation of speleothem δ 13 C data (Fig. 7) is generally hampered by a host of local factors such as changes in soil CO 2 production and content, closed versus open system dissolution of carbonates in the soil/epikarst system, residence time and mixing of waters along the pathway to the drip point, or solution degassing (Hendy, 1971;Bar-Matthews et al., 1996;Fairchild et al., 2006).
Percolating water degassing could be greater during certain periods at POM2 sampling site, as the CO 2 content of the cave's atmosphere drops from ∼ 1800 ppm in November-December to ∼ 1000 ppm in April-May. This seasonal variation in the CO 2 content of cave air is likely the combined result of soil CO 2 productivity and cave ventilation (Spötl et al., 2005, Kowalczk andFroelich, 2010;Frisia et al., 2011;Tremaine et al., 2011;Riechelmann et al., 2013).
At the POM A site, which is the shallowest and closest to the entrance, the CO 2 concentration reaches a minimum value of 760 ppm, well above values of outside air (between 200 and 310 ppm). The two deeper sites, POM2 and POM B, show even less ventilation, with minimal values of 960 ppm. This indicates that cave ventilation is moderate (although continuous) at Ascunsȃ Cave.
Supposing that the cave ventilation regime remained unchanged during the middle Holocene, cave air CO 2 was probably controlled mostly by soil/vegetation dynamics. If so, higher speleothem δ 13 C values are indicative of reduced CO 2 input from the soil and/or prior precipitated calcite (Fairchild et al., 2000). These two processes could be the result of drier conditions being established across the mid-Holocene and might have been responsible for producing the upward trend observed in δ 13 C values between 6 and 4 ka BP.

The 8.2 ka and 3.2 ka events
The 8.2 ka climate change event (Alley et al., 1997;Rohling and Pälike, 2005) is one of the most prominent events of environmental change in the Holocene. Pollen assemblages from northern Romania (Feurdean et al., 2008), Macedonia (Bordon et al., 2009) and Greece (Pross et al., 2009), testate amoebae from northern Romania (Schnitchen et al., 2006), speleothem carbon stable isotopes from Israel (Bar-V. Drȃguşin et al.: Constraining Holocene hydrological changes in the Carpathian-Balkan region Matthews et al., 2000) and marine faunal composition from the Aegean Sea (Rohling et al., 2002) document a decrease in winter temperature and precipitation, while summer conditions remained rather stable. Similar to other Romanian stalagmites (Tȃmaş et al., 2005;Constantin et al., 2007), the POM2 δ 18 O and δ 13 C records do not show significant variability across the 8.2 ka event. The only indication of changing environmental conditions is that the growth rate was six times higher during this event compared to the rest of the Holocene in the Ascunsȃ Cave. Figure 10 shows a comparison of calculated δ 18 O c_ptc measured δ 18 O c_spel values for the 8.2 and 3.2 ka events. For the 8.2 ka event, δ 18 O c_ptc is calculated as the difference between a 500-year interval succeeding the event (8.1-7.6 ka) and the event itself (8.3-8.1 ka). Here we used pollen-based temperature reconstructions as follows: for V11 Cave, those from the Steregoiu peat bog in northern Romania (Feurdean et al., 2008), and for Ascunsȃ Cave, the Lake Maliq in Macedonia (Bordon et al., 2009).
The shift in isotopic values after the 8.2 ka event at V-11 and Ascunsȃ Cave is well explained by the pollen-based temperature rise (Feurdean et al., 2008;Bordon et al., 2009). The pollen data suggest a significant warming in winter and very slight cooling in summer after the 8.2 ka cold event. Lake Maliq suggests a strong warming in winter and a lesser warming in summer after the event. δ 18 O c_spel generally agrees with calculated δ 18 O c_pct Nevertheless, as δ 18 O c_spel from Ascunsȃ falls closer to the summer end of δ 18 O c_ptc , a climate-related hydrological influence on the speleothem data could be argued for. A δ 18 O c_spel close to the low-δ 18 O summer end of the calculated range of δ 18 O c_ptc may be interpreted as a slight increase in the proportion of winter rain after the 8.2 ka event. Lower winter rainfall during the 8.2 ka event is in agreement with precipitation reconstruction from Lake Maliq and Steregoiu (Bordon et al., 2009;Feurdean et al., 2008), but the speleothem data would only support a small reduction of winter rain during the 8.2 ka event. However, a high speleothem precipitation rate during the 8.2 ka event may suggest a high infiltration rate and more rain. The apparent high infiltration rate of drip water could mean that instead of a reduction in winter rain, summer rainfall increased during the 8.2 ka event.
The 0.4 ‰ increase in δ 18 O values in the Ascunsȃ record between the average of periods 3.2-2.9 ka and 2.9-2.5 ka defines the 3.2 ka event. Both Ascunsȃ and Poleva (Constantin et al., 2007) show a comparable signal, although with a small difference in chronology. This difference in timing is still within chronological uncertainty. The magnitude of rising δ 18 O values in both speleothems after the event is outside the pollen-defined expected range of δ 18 O c_ptc due to temperature change. Thus, the event likely reflects a change also related to climate-hydrology. The interval coincides with events documented in both archaeological and palaeoclimate records around the eastern Mediterranean (see review in Drake, 2012) that may have led to cultural demise at the end of the late Bronze Age. Associated with this cold event is a decrease of the Aegean Sea winter surface temperatures, as documented by Rohling et al. (2002). Changes in climate-hydrology are reflected by drought in Cyprus and Syria (Kaniewski et al., 2010(Kaniewski et al., , 2013, and a drop of the Dead Sea level (Migowski et al., 2006). In northern Romania, testate amoebae data also indicate a dry phase between 3.4 and 3.0 ka (Schnitchen et al., 2006).
The calculated range of δ 18 O c_ptc (2.9-2.5 ka to 3.2-2.9 ka) using pollen reconstructed temperatures from Lake Maliq (Bordon et al., 2009) is from −0.13 ‰ for the cold season (reflecting a slight cooling after the event) to 0.27 ‰ for summer season (reflecting a warming). The δ 18 O c_spel from Ascunsȃ Cave is 0.4‰ . This offset (0.13‰ with respect to the upper margin of δ 18 O c_ptc ) indicates that changes related to climate-hydrology contributed significantly to the speleothem δ 18 O signal. In this case the isotopic evolution out of the 3.2 ka event suggests an increase in summer rainfall or a reduction in winter rain after the 3.2 ka event. The event itself would consequently be characterised by either lower summer rainfall or higher winter rainfall. Higher winter rainfall is not in agreement with widespread drought in the winter rain-dominated eastern Mediterranean. The Mediterranean plays an important role as winter moisture source for southwestern Romania (Bojariu and Paliu, 2001). Consequently, a change in moisture source composition of Mediterranean-derived rainfall could potentially have contributed, but is not observable in Mediterranean foraminifera records (Rohling et al., 2002;Siani et al., 2010). It seems that pollen data from Lake Maliq and marine records from the Mediterranean are not in agreement concerning the timing of temperature change, perhaps due to chronological uncertainty. We note that significant winter cooling is apparent in the abundance record of planktic foraminifera species in the Aegean Sea (Rohling et al., 2002), whereas the Lake Maliq pollen record shows a winter cooling centred between 2.9 and 2.8 ka. Aligning Lake Maliq with the marine chronology would change the sign of δ 18 O c_spel and δ 18 O c_pct , and the interpretation would be a winter drought for the 3.2 ka event. As such, a change in climate-hydrology is evident for the 3.2 ka event, but cannot currently be defined given chronological uncertainty. However, there is clear evidence for coincidence of a climate event which includes a hydrologic component with a time interval of significant cultural change.

Conclusions
The stable isotope record of Ascunsȃ Cave in southern Romania was used to identify centennial to millennial-scale climate change during the Holocene in this area. By using a novel approach to discriminate between the effects of temperature and hydrology on speleothem δ 18 O values, we demonstrate that this shift not only reflects rising regional temperatures as documented by pollen assemblages (Davis et al., 2003), but also a combination of influences related to changes in climate-hydrology. The approach presented in this study relies on using pollen-based temperature reconstructions to constrain temperature-driven isotopic changes of speleothem calcite. This method considers isotope fractionation occurring during water vapour condensation and calcite precipitation. A constrained range of temperaturedriven isotopic changes between winter and summer is obtained, and the deviation of measured data from this range suggests that additional climate-hydrology factors contributed to isotopic variability over the studied period. Between 6 and 4 ka, δ 18 O gradually shifted towards higher values in a variety of speleothems across the Mediterranean. Using this approach we find that the middle Holocene enrichment in SW Romania was 0.5-0.6 ‰ greater than values attributable to rising temperatures. A similar situation persists throughout the eastern Mediterranean domain. In the Atlantic-dominated western Romania, as in all parts of the western Mediterranean domain analysed here, changes in δ 18 O of speleothems largely reflect temperature variations. This reveals a different climate response between the two regions. In the eastern Mediterranean, a combination of higher summer temperatures, enhanced evaporation from the soil and a higher δ 18 O of the Mediterranean surface water may plausibly explain the observed evolution of δ 18 O in speleothems across the mid-Holocene.
We also analysed two events of rapid climate change, at 8.2 and 3.2 ka. At Ascunsȃ Cave, the 8.2 ka event is characterised by a growth rate six times higher than during the rest of the Holocene. The low isotopic variability during the 8.2 ka event seems to reflect mostly temperature variations, but hydrologic conditions such as relatively more summer rainfall at Ascunsȃ and V11 caves cannot be ruled out. During the 3.2 ka event, both temperature and rainfall seasonality appear to have changed in southern Romania. Data from Ascunsȃ Cave suggest a significant contribution of changing climate-hydrology, but its nature cannot currently be resolved due to chronological uncertainty in the pollen database. The calculation of the temperature-related part of an observed change in speleothem calcite δ 18 O from pollen-based temperature reconstructions relies on two basic assumptions: (1) the cave temperature reflects the annual average surface air temperature and only fluctuates very little around that value (i.e. temperatures in the Ascunsȃ Cave chamber from which the POM2 stalagmite was collected vary by 0.6 • C over the year); (2) the coldest and warmest month reasonably define the range of temperature-controlled oxygen isotope fractionation during rainfall. This cannot be currently tested for the Ascunsȃ Cave site due to lack of a continuously recording of stable isotopes in precipitation. However, this assumption is based on information from a large number of European station recordings (Rozanski et al., 1993). In the following, we only consider temperature-related aspects contributing to an observed isotopic change, δ 18 O. This has the advantage that accuracy of δ 18 O is determined only by the much less relevant difference of slopes between various water-calcite calibration functions rather than the considerable offsets. For example, the difference in δ 18 O for a temperature change from 10 to 20 • C that would result from using the calibration functions by Kim et al. (2007) and Tremaine et al. (2011), respectively, is only 0.22 ‰, whereas the difference in absolute δ 18 O values is 0.75 ‰ for 10 • C and 0.97 ‰ for 20 • C between the two calibration functions.
Generally, δ 18 O in calcite (δ 18 O c ) is determined by the calcification temperature and the isotopic composition of ambient water (Epstein et al., 1953), the latter reflecting rain formation temperature among other hydrologic factors (Rozanski et al., 1993). Consequently, the temperaturerelated δ 18 O c can be broken down into δ 18 O ct -the contribution due to changing calcification temperature in the cave -and δ 18 O w -the contribution due to changing rainfall temperature. The relative change of δ 18 O between two time intervals, t1 and t2, in a speleothem is Likewise, the change in pollen temperature anomaly (Davis et al., 2003) is The empirically determined fractionation factor α for oxygen isotopes between water and calcite in speleothems is defined as (Tremaine et al., 2011) 1000 ln α = 16.1(10 3 T −1 ) − 24.6. (A4) The fractionation factor is related to measured values for δ 18 O c and δ 18 O w by (e.g. Sharp, 2007) 1000 1000.
To constrain the range of temperature-related variability of δ 18 O w we use the empirical relationship of 0.58 ‰ (δ 18 O) / • C for mid-latitudes (Rozanski et al., 1993): with summer (MTWA) and winter (MTCO) temperatures, respectively, from the Davis et al. (2003) pollenbased reconstructions, respectively, for TA p-seasonal . Inserting Eqs. (A8) and (A9) into Eq. (A1) yields the pollen temperature-constrained range of change in speleothem calcite ( δ 18 O c_ptc−winter and δ 18 O c_ptc−summer ), after conversion to the VPDB scale used for calcite δ 18 O. This interval constitutes the part of the observed change in the speleothem ( δ 18 O c_spel ) that may be explained by temperature variability between two defined time intervals (Fig. 8a). Calculations for all cave sites discussed in the main text are summarised in Table 5.