Evidence of thermophilisation and elevation-dependent warming during the Last Interglacial in the Italian Alps

Thermophilisation is the response of plants communities in mountainous areas to increasing temperatures, causing an upward migration of warm-adapted (thermophilic) species and consequently, the timberline. This greening, associated with warming, causes enhanced evapotranspiration that leads to intensification of the hydrological cycle, which is recorded by hydroclimate-sensitive archives, such as stalagmites and flowstones formed in caves. Understanding how hydroclimate manifests at high altitudes is important for predicting future water resources of many regions of Europe that rely on glaciers and snow accumulation. Using proxy data from three coeval speleothems (stalagmites and flowstone) from the Italian Alps, we reconstructed both the ecosystem and hydrological setting during the Last Interglacial (LIG); a warm period that may provide an analogue to a near-future climate scenario. Our speleothem proxy data, including calcite fabrics and the stable isotopes of calcite and fluid inclusions, indicate a +4.3 ± 1.6 °C temperature anomaly at ~2000 m a.s.l. for the peak LIG, with respect to present-day values (1961–1990). This anomaly is significantly higher than any low-altitude reconstructions for the LIG in Europe, implying elevation-dependent warming during the LIG. The enhanced warming at high altitudes must be accounted for when considering future climate adaption strategies in sensitive mountainous regions.

be accomplished by extracting proxy data from archives of past interglacials that were warmer than the late Holocene and, specifically, the Anthropocene.
Speleothems, defined as secondary mineral deposits formed in caves 5 , identify one of the most accurate, precisely datable, continental archives of climate and environmental proxy data. Their formation, morphology, mineralogy, structure and chemical composition depend on the cave setting, ventilation dynamics and on the type of vegetation above the cave [6][7][8] . In the European Alps, speleothem formation is mostly limited to caves located below the timberline, where efficient soil turnover in mixed conifer-deciduous forests promotes elevated biogenic pCO 2 that combines with rainwater to produce weak carbonic acid that dissolves the carbonate host rock 8 . Subsequent re-deposition of calcite is promoted by degassing of the dripwaters in the cave atmosphere, whereby the critical threshold in the calcite saturation state (SI CC ) of the film of fluid wetting a speleothem is >0.1 and preferably >0.5 (refs 9,10 ). Presence of fossil speleothems in high-altitude caves where significant sparitic calcite deposits are not currently forming, suggests that an upward shift in the timberline, associated with thermophilisation, must have occurred in the past. This implies that warmer than present-day temperatures characterised the region prior to the late Holocene. The impact of a warmer-than-current climate on water resources and vegetation can be elucidated through a multi-proxy record approach reconstructing past climate and environmental processes. Here, we present multi-proxy records from three coeval speleothems that formed during the LIG in a subalpine cave located in the Italian Alps to yield the temperature of the cave catchment area during the LIG. Our reconstructed temperatures are then compared with local-to regional-scale modern and LIG temperature estimates to gain an understanding of ecosystem and hydrological responses to warm interglacial climates in sensitive, mountainous regions.
Specimens. Three inactive speleothems were sampled from two locations in CB Cave, characterised by morphologies and fabrics that stem from diverse discharge and parent-water compositions, ensuring the full spectrum of isotope and trace-element characteristics (from equilibrium to kinetically influenced) is accounted for. A 70 mm-thick flowstone (CB25) and a 185 mm-tall stalagmite (CB47) were sampled from the "Scrigno" chamber, with a ceiling height of 0.6 m, located ~50 m below the surface (1930 m a.s.l.). Flowstone CB39 (33 mm-thick) ). The locations of the caves Grotta di Ernesto (ER) and Pozzo di Val dal Parol (VP) are also shown, as they provide modern-day analogues for CB Cave during the peak and late LIG, respectively. The maps were generated using the software GRASS GIS version 6.1 (https://grass.osgeo.org/) with topographic data taken from the National Aeronautics and Space Administration's (NASA) Shuttle Radar Topography Mission (SRTM), available from the U.S. Geological Survey (https://lta.cr.usgs.gov/SRTM).  Fig. S4). CB25 was sampled close to its presumed discharge point and, given its fabric, we expect formation under relatively high flow rates, with minimum disequilibrium isotope fractionation. The cone-shaped stalagmite CB47 consists of porous, open columnar calcite, suggesting relatively fast and constant drip rates, likely resulting in negligible disequilibrium isotope fractionation during formation. In contrast, the dendritic fabric that characterises its upper 35 mm suggests a variable drip rate regime, at a seasonal or annual scale 7 . In flowstone CB39, the lack of intercrystalline porosity within its compact columnar calcite fabric layered with micrite and elongated columnar calcite with lateral overgrowths suggests increased discharge and inclusion of impurities 10 (Supplementary Discussion and Fig. S3).

Speleothem fluid inclusions (FI).
Stable isotope values from tiny amounts of dripwater trapped within the host calcite as fluid inclusions (FI) from the time of speleothem formation (δ 18 O FI and δD FI ) were analysed to gain insight on past temperatures. Comparing the δ 18 O FI values with those of the host-calcite (δ 18 O C ) using calcite-water geothermometer equations allows assessment of speleothem formation temperatures (Table 1 and  Supplementary Table S11). For this, the δ 18 O C values must be in isotopic equilibrium with the δ 18 O FI of the coevally trapped inclusions 17 . Soaking blocks of CB speleothems in water of different isotopic composition indicated that there was negligible fluid exchange through the interconnected pore space (see Supplementary Discussion), which signifies that FI data accurately reflect the properties of the dripwater.
Trace elements. Both Mg and Sr concentrations are significantly lower in the CB47 stalagmite than in the CB39 flowstone. CB39 Mg concentrations (mean 3283 ± 418 ppm) exhibit considerable variability, with a maximum at ~124-123 ka. CB39 Sr concentrations (mean 44 ± 7 ppm) rise rather steadily until reaching a maximum at ~123 ka. Average Mg and Sr concentrations in a representative sector of CB47 stalagmite are 856 ± 171 ppm and 28 ± 5 ppm, respectively.

Discussion
In caves formed below a well-developed soil cover, often the case for low-altitude sites in temperate climate settings, CO 2 is produced constantly in the soil, hence, it not a limiting factor for carbonate dissolution. However, the strongly changeable water availability found in such regions thus contributes the environmental variable that dominates carbonate geochemistry. Consequently, rainfall amount is often a key parameter encoded in low-altitude speleothem proxy data 16,18,19 . By contrast, at high-altitude sites, speleothem formation is limited by the soil efficiency and climate parameters that influence soil CO 2 production. Soil efficiency is a crucial factor that gives rise to adequate CO 2 production, resulting in an acidic solution capable of dissolving enough host rock to overcome the threshold in SI CC required to precipitate speleothem carbonate once the infiltration waters reach the cave atmosphere. This SI CC threshold typically corresponds to a local altitudinal (or latitudinal) band that coincides with the speleothem limit 9 , which defines a change from a warm and highly vegetated speleothem-forming environment to more hostile conditions where sparitic speleothem development is unlikely. Therefore, the presence of sparitic speleothems, which were fed by waters discharged from pure carbonate host rocks, where present dripwater SI CC values lie below zero (e.g., at high-altitude or high-latitude sites), likely identifies periods in the past when higher temperatures enhanced soil productivity and dripwater SI CC above the threshold required for sparitic speleothem formation. Enhanced soil productivity is thought to be associated with thermophilisation and high-altitude greening, which in turn, favours sparitic speleothem development. Below, we investigate whether the LIG speleothem formation in CB Cave was prompted by warmer temperatures than modern and Holocene conditions.
Geochemical data extracted from speleothems are widely used to reconstruct past climate and environmental conditions. δ 13 C values of speleothem calcite generally reflect soil efficiency and vegetation type and density above the cave. The δ 13 C values of CB25 and CB47 are significantly lower than modern values, suggesting enhanced soil microbial activity and root respiration at the time of their formation with respect to current conditions, arguing for significant LIG greening above the cave. This is consistent with the notion that speleothem formation itself required that infiltration waters had higher SI CC than today, as a result of an upward shift of the sparitic speleothem limit. It is, therefore, reasonable to infer an upward shift of the LIG timberline to an altitude well above the elevation of the cave's catchment area.
Scientific REpoRTs | (2018) 8:2680 | DOI:10.1038/s41598-018-21027-3 CB39 δ 13 C values are considerably higher than those of CB25 and CB47. Significantly higher Mg and Sr concentrations in CB39 than in CB47 suggest enhanced water-rock interactions (WRIs) along the longer flowpath to CB39's deeper location. Furthermore, the marked correspondence between CB39's Mg concentration and δ 13 C time series (Fig. 2h) provides tangible evidence that the δ 13 C values were strongly modulated by WRI. High concentrations of dissolved carbonates in CB39's feeding water, due to more extensive WRIs, caused an elevated SI CC , permitting carbonate deposition, and may have been responsible for fabric differences and an extended growth period of CB39 with respect to CB25 and CB47. Such WRIs would mean that a large component of the dissolved inorganic carbon was derived from the host limestone, rather than the soil (as is the case for CB25 and CB47) (e.g., 20 ). Therefore, CB39's δ 13 C values are not directly related to soil productivity and environmental conditions above the cave. However, the initiation of CB39 formation (Fig. 2) likely corresponded with the onset of thermophilisation, confirmed by the increase in the percentage of woody taxa following Termination II as seen in pollen records from Lago Grande di Monticchio in southern Italy 21 (Fig. 2d). The following period of stability in woody taxa coincided with the onset of CB47 formation, which was delayed until soil and vegetation above the cave became fully established, creating the conditions required for soil-respiration-instigated speleothem growth.
Speleothem δ 18 O values are often used to constrain the hydrological and environmental conditions during formation, in addition to the moisture source location. The lowest CB39 δ 18 O values are recorded at ~127-126 ka, in analogy to the composite δ 18 O record of Soreq Cave 14 , Israel (Fig. 2b), pointing to a regional hydroclimate response to increasing solar insolation 22 (Fig. 2a) following Termination II, which also coincided with the onset of CB47 growth ( Fig. 2f and g). Sudden decreases in δ 18 O values recorded in speleothems SCH-5 from Schneckenloch Cave in Austria 23 (Fig. 2c), CC5 from Corchia Cave in Italy 24 (Fig. 2e) and CB47 (Fig. 2f) at ~125 ka coincided with the onset of CB25 growth, which may have been related to an increase of rainfall, activating CB Cave flowstone development simultaneously to Atlantic storm tracks hitting the western coast of the Italian Peninsula and the Austrian Alps. Increases in CB39 δ 13 C, Mg and Sr records at 124-123 ka (Fig. 2h,i), concomitant with strong fluctuations in δ 18 O values and decreased growth rate of SCH-5 ( Fig. 2c) 23 , in addition to cessation of CB47 growth, likely indicate a drying that caused enhanced WRIs in the dripwater feeding CB39. Cessation of CB25 growth at ~121 ka is reasonably explained by a decrease in soil pCO 2 that resulted in percolation water SI CC inadequate for speleothem formation. The speleothem limit and corresponding timberline had, therefore, moved below the altitude of the CB Cave catchment. However, enhanced WRIs of the percolation water feeding CB39 still permitted its formation until the end of the LIG.
The possibility of LIG thermophilisation at the high-altitude CB Cave site has been tested by constraining the temperatures of the LIG using isotope values from speleothem FIs and their enclosing calcite, using calcitewater oxygen isotope geothermometer equations. Such equations are derived from both experimental 25,26 and field-based methods 11,27,28 . However, due to inherent disequilibrium isotope fractionation that occurs during speleothem formation, equations characterising near-perfect isotopic equilibrium do not successfully predict speleothem formation temperatures 19,29 . Using cave-derived calcite and water oxygen isotope pairs yields the most accurate speleothem formation temperature reconstructions because the empiric equation includes the intrinsic in-cave fractionation 11,27 . Here, the equation of Johnston et al. 11 Fig. S4). Furthermore, δ 18 O FI values have been shown to change after FI entrapment due to diagenetic processes, including effects of non-classical crystallisation, while the isotope composition of the δD FI values remained unaltered 30 . For this reason, in this study, δ 18 O FI values were calculated using the δD FI values and the modern relationship between δ 18 O and δD in meteoric waters derived from the local meteoric water line constructed using data from Mt. Paganella (Paganella-MWL) (Table 1, Fig. 3). By using the δ 18 O C and the δD-derived δ 18 O FI values as inputs for the geothermometer equation, the resulting speleothem formation temperature estimates are 7.9 ± 1.9 °C for CB47 (126-124 ka) and 3.6 ± 1.5 °C for CB25 (125-122 ka), with the propagation of uncertainties detailed in Supplementary Table S11. When compared with modern Scrigno chamber temperatures (3.45 ± 0.05 °C, Supplementary Fig. S12), the temperature anomalies estimated from FIs are +4.5 ± 1.9 °C at 126-124 ka and +0.2 ± 1.5 °C at 125-122 ka with respect to present-day cave temperatures (Table 2).
Temperature estimates derived from FI data were then compared with temperature reconstructions based on geochemical and petrographic data from the CB speleothems. Speleothem geochemistry and calcite fabrics respond to surface air temperature, cave temperature, drip rate and dripwater SI CC . By using available datasets from 11 caves in the region 9 and comparing these with modern and Holocene speleothems from the same caves 11,31 , the temperature and conditions under which geochemical signals are incorporated and fabrics form in the speleothem calcite were reconstructed. The relationship between infiltration temperature and δ 13 C values for modern speleothems in Trentino can be used to estimate the formation temperatures of LIG sparitic speleothems from their δ 13 C values. Since δ 13 C values of CB39 were strongly affected by WRIs, the calculation was performed for CB25 and CB47 that encode a vegetation signal, which can be related to temperature. To minimise the inclusion of values potentially affected by strong disequilibrium fractionation associated with initiation or cessation of speleothem growth, only the δ 13 C values from the middle sections of the speleothems were used. Using the least fractionated modern speleothem δ 13 C values from Johnston et al. 11 and MAT inf data from Borsato et al. 9 , the relationship MAT inf (°C) = −1.68·δ 13 C-7.57 (R 2 = 0.96) was obtained for sparitic speleothems that formed near isotopic equilibrium (Supplementary Material S14). MAT inf estimates were obtained for the LIG at CB Cave by applying this relationship to the mean of δ 13 C values selected from the least isotopically fractionated middle section of the Scrigno speleothem δ 13 C records. This yielded a mean δ 13 C value of −8.7 ± 0.6‰ (126.0-125.3 ka) for CB47, corresponding to a peak LIG MAT inf of 7.1 ± 1.3 °C and, thus, a peak LIG temperature anomaly of + 4.2 ± 1.4 °C, with respect to modern conditions (Supplementary Material S13). The late LIG, represented by CB25, has a mean δ 13 C value of −7.8 ± 0.7‰ (123.8-121.9 ka), and yielded a late LIG MAT inf of 5.5 ± 1.3 °C, hence an anomaly of + 2.7 ± 1.4 °C, with respect to the modern MAT inf (Table 2).
Alternatively, in a novel approach, we exploit the relationship between SI CC of speleothem-forming water and the coexisting speleothem fabric. Extensive monitoring in the study region provides strong evidence that speleothem fabric is controlled by SI CC 7,31 , though a link with the vegetation above the cave, connected with soil and cave air pCO 2 , which is ultimately related to temperature 8,9 . Active columnar fabrics in the region are diagnostic of low carbonate saturation, with SI CC in the range of 0.15-0.35, which also suggests monomer by monomer attachment growth 31 . Whereas, dripwater SI CC for dendritic fabrics is 0.2-0.4, with a mean SI CC that is higher than dripwaters   (Table 2, Supplementary Material S13). The late LIG temperature estimated using the reconstructed-SI CC is slightly higher than the other reconstructions provided here, albeit within the uncertainties, likely reflecting the wide range of possible SI CC values that form columnar calcite. Encouragingly, the peak LIG value gained from the reconstructed-SI CC is remarkably consistent with the temperatures derived from both the FI and δ 13 C methods. In summary, temperature anomaly estimates based on different physical and geochemical data from CB speleothems (Table 2) yield a peak LIG temperature anomaly of +4.3 ± 1.6 °C and a late LIG anomaly of +2.3 ± 1.4 °C (+1.4 ± 1.5 °C excluding the reconstructed-SI CC value), with respect to modern temperatures (1961-1990Fig. 4) at this subalpine site. The good agreement of the independent temperature estimates from three different proxy data grants confidence in our temperature reconstructions and has tested the validity of the fabric-SI CC -temperature relationship. The positive temperature anomalies, which are significant with respect to their uncertainties (propagation of uncertainties detailed in Supplementary Table S13), demonstrate that the LIG MAT inf were higher than present-day. In fact, the mere occurence of sparitic speleothems at the altitude of CB Cave already attests to higher-than-present LIG temperatures, and associated better-developed soil and vegetation cover than today.
The magnitude of the peak LIG temperature anomaly calculated here (+4.3 ± 1.6 °C) is within the uncertainty of the +8 ± 4 °C anomaly reconstructed from central Greenland ice cores 32 and suggests amplification of warming at both high altitudes and latitudes. Over the 4.4 kyrs that encompass the temperature reconstructions from both FIs and δ 13 C values (126.3-121.9 ka) in the CB speleothems, temperature decreased at a rate of 0.5 °C kyr −1 . This compares with a cooling rate of 0.75 °C kyr −1 estimated over the same period of time from temperature reconstructions from Greenland ice cores 32 . In contrast, the global rate of cooling over the same period has been reconstructed at 0.11 °C kyr −1 (ref. 33 ), implying that, in addition to the magnitude, the rate of temperature change at high altitudes and latitudes was amplified with respect to the global mean.
A LIG temperature increase of +4.3 ± 1.6 °C, associated with thermophilisation, is surmised to have caused an upward shift of the speleothem limit to an altitude of ~2500 m a.s.l. and, accordingly, the catchment area of CB Cave at the peak LIG (7.2 ± 1.6 °C) would have been colonised by a mixed deciduous forest typical  ) and dripwater SI CC in the region (MAT inf (°C) = 4.36 + 9.95·SI CC ; R 2 = 0.93) 9 . Present-day CB Cave dripwater data (2.8 ± 0.5 °C; black diamond) lie below the local speleothem limit (at a temperature of 4.4 °C, corresponding to a SI CC = 0, current altitude ~1660 m a.s.l.) 9 where their negative SI CC values indicate that formation of sparitic speleothem calcite is unlikely. The MAT inf reconstructed for the peak LIG (from CB47) with a SI CC of 0.28 ± 0.18 based on calcite fabrics (orange diamond; 7.2 ± 1.6 °C) lies well above the speleothem limit. The late LIG temperature reconstruction (from CB25; blue diamond) with a SI CC of 0.25 ± 0.14 and a MAT inf of 5.1 ± 1.4 °C also lies above the speleothem limit.
Scientific REpoRTs | (2018) 8:2680 | DOI:10.1038/s41598-018-21027-3 of the current upper montane zone 8 . This association is now observed in the catchment of Ernesto Cave (ER; Fig. 1), located at 1180 m a.s.l. with a MAT inf of 7.1 °C and a mean dripwater SI CC of 0.22 ± 0.12 (ref. 9 ), which contains Holocene stalagmites with columnar, microcrystalline and dendritic fabrics, similar to CB47 (ref. 34 ). The temperature of 5.1 ± 1.4 °C calculated from CB25 for the late LIG suggests similar temperatures to those currently experienced in other caves in Trentino, such as "Pozzo di Val del Parol" (VP), located at 1585 m a.s.l. with a MAT inf of 4.8 °C and a dripwater SI CC of 0.12 ± 0.20 (ref. 9 ). The VP Cave has active speleothems, including conical stalagmites and drapery stalactites that form below a catchment covered by pastures with scattered conifer trees. Here, the major driving force of modern speleothem development is the vegetation assemblages and associated pCO 2 of the percolating water that increases the SI CC 8 . The increase in LIG temperatures most likely caused significant thermophilisation above CB Cave, thus permitting crystalline speleothems to form.
To test the hypothesis of elevation-dependent warming (EDW) during the LIG, the peak LIG temperature anomaly estimated for the CB site was compared with temperature reconstructions from lower altitude sites. Modelled central European summer temperatures for the LIG maximum are 1-2 °C higher than present 35 . Pollen records in Iberian Margin sediment core MD04-2845 further indicate that the warmest period of the LIG reached temperatures ~2 °C higher than present 36 . Turney and Jones 37 reconstructed global temperatures 1.5 °C higher than today from a compilation of marine and terrestrial (including ice) proxy records. These temperature anomaly reconstructions are noticeably lower than our peak LIG subalpine speleothem-based estimate of +4.3 ± 1.6 °C. This discrepancy implies that high altitudes were more sensitive to warming than low-altitudes and, thus, argues for EDW during the LIG. Considering the LIG as an analogue for future warming, as expected in a scenario of increased greenhouse gas forcing, applying the low-altitude LIG temperature anomaly data to an alpine region without making adjustments for topography and associated EDW would result in an underestimation of the predicted temperature anomaly. In turn, this would cause an underestimation of the extent of thermophilisation and its impact on the local ecosystems and hydrological cycle, extending to downstream water resources. Using low-altitude temperature estimates to formulate adaptive strategies for mountainous regions is therefore inappropriate. More quantitative temperature reconstructions from well-dated proxy archives at both high and low altitudes in mountainous regions are, therefore, urgently needed. EDW can be caused by a number of mechanisms including: (i) albedo from snow, ice and vegetation changes, (ii) water vapour and radiative fluxes, (iii) clouds and (iv) aerosols 38 . Our results suggest that the vegetation belts shifted upwards (thermophilisation) at the CB Cave site during the LIG period. This afforestation reduced surface albedo due to increased radiation absorption by plants, enhancing warming in the alpine zone that became more vegetated. As temperatures increased and thermophilisation proceeded, evaporation and evapotranspiration also increased 4 , causing enhanced humidity that has a positive-feedback on warming as water vapour absorbs and emits long-wave radiation. Although a global factor, its non-linear sensitivity means that areas with lower initial humidity, such as cold high-altitude sites, are strongly affected by humidity changes. Therefore, through the positive water vapour feedback, high elevation sites incur enhanced warming relative to areas of high humidity, such as lower altitudes and tropical regions. Today, the temperature gradient (lapse rate) between the valley bottom (200 m a.s.l.; 12.6 °C) and the mountain top (Mt. Paganella, 2125 m a.s.l.; 1.7 °C) is −0.57 °C 100 −1 m (ref. 39 ). If a ~2 °C warming is expected at the valley bottom 35,37 and +4.3 ± 1.6 °C at 1930 m a.s.l. (present study), the calculated peak LIG altitudinal temperature gradient of −0.4 °C 100 −1 m implies a reduced rate of cooling with increasing elevation. This suggests that, similarly to observations of current warming at high altitudes [40][41][42] , EDW likely occurred during the LIG.
Forward-modelling of elevation gradients in the alpine region for the period 2070-2099, against the reference period 1961-1990, indicates a rise of 3.5-5.0 °C for an altitude similar to CB Cave. Winter precipitation is expected to increase in the southern watershed of the Alps while summer precipitation should decrease slightly 43 . Furthermore, maximum changes in albedo are estimated in the altitude band of the CB Cave site with a reduction of 80 snow days per year. This scenario can be compared with the CB Cave reconstruction for the peak LIG with a +4.3 ± 1.6 °C temperature rise. Our data suggest that higher temperatures resulted in high precipitation amounts that provided the water necessary for the formation of flowstones. Data from CB Cave also imply a longer vegetation period, resulting from a reduction in the duration of the snow cover that enhanced greening and soil productivity, which are also beneficial for speleothem formation at high altitudes. The similarity of forward-modelling observations and our LIG reconstruction suggests that by the end of the 21 st century, conditions in the southern watershed of the European Alps will likely to be similar to those experienced during the peak of the LIG.

Methods
U/Th dating. Fifteen samples from CB25, CB39 and CB47 were analysed for U and Th concentrations at two different laboratories (Supplementary Table S5). At the University of Melbourne, Australia, sub-samples were sawn from visibly clean calcite with vertical dimensions of ~1.5 mm. Chemical separation, measurements of U and Th isotope ratios and multi-collector inductively coupled plasma mass spectrometer (MC-ICP-MS; Nu-instruments Nu Plasma) analysis followed the methods described in Hellstrom 44 . At the University of Minnesota, U.S.A., sub-samples were hand-drilled along growth laminae (width ~1 mm). Chemical and MC-ICP-MS (Finnigan Neptune) analytical methods followed those described in Shen et al. 45 . Decay constants of Cheng et al. 46 were used. Corrected 230 Th ages assume the initial 230 Th/ 232 Th atomic ratio of 4.4 ± 2.2 × 10 −6 , which refer to a material at secular equilibrium with the bulk earth 232 Th/ 238 U value of 3.8. The age datum is defined as 1950 AD and dating uncertainties are reported as ±2 standard errors. The age models of the three speleothems ( Supplementary Fig. S6) were developed using the StalAge algorithm 47 , with age uncertainties quoted at the 95% confidence limit. Fabric analysis. Uncoated, polished thin sections were observed under plane (PPL) and cross-polarised light (XPL) on a Zeiss Axioskop optical microscope. Fabrics, fabric types, fabric coding and microstratigraphic logging followed the conceptual framework proposed in Frisia 10 , which is based on models of fabric development.

Stable isotope analysis.
Oxygen and carbon stable isotope values were obtained from calcite powders.
Flowstones were sampled perpendicular to the visible lamina, whereas stalagmite CB47 was drilled as close as possible to its growth axis but avoiding areas of apparent recrystallization (Supplementary Fig. S3). Micro-milling at increments of 150 µm was used to obtain samples from flowstone CB39, which shows a slow vertical extension rate (4.5 mm kyr −1 ). For CB25 and CB47, whose ages suggest a faster vertical extension rate (18 mm kyr −1 and 57 mm kyr −1 , respectively), sub-samples were obtained by hand-drilling at increments of 1 mm. Stable isotope values of C and O were measured using a Thermo Fisher Delta Plus XL mass spectrometer at the University of Innsbruck, Austria. Long-term 2σ reproducibility for δ 13 C and δ 18 O values is ± 0.06‰ and ± 0.08‰, respectively. All results are given in per mil notation with respect to the Vienna Pee Dee belemnite reference material.
Fluid inclusion analysis. FI isotope data were obtained for CB25 and CB47 only, as the CB39 fabric did not yield sufficient water for analysis. Blocks of calcite ~0.5-1 cm 3 , the minimum volume required to obtain >0.2 μl of water from our samples, were crushed in a custom-built crushing device, in-line with a Delta V Advantage isotope ratio mass spectrometer (IRMS; Thermo Fisher Scientific) at the University of Innsbruck, Austria. Crushed samples were heated at 120 °C in a chamber flushed with helium carrier gas that transported the evolved water vapour to a cryo-focusing cell (−150 °C). The frozen water was subsequently flash-heated to 300 °C and the resulting vapour transferred in a single pulse into the Thermal Combustion/Elemental Analyser (Thermo Fisher Scientific) where it reacted at 1400 °C with glassy carbon, producing H 2 and CO. The evolved gases were separated in a gas chromatography column and then admitted to the IRMS, where measurements of δD and δ 18 O were carried out 48 . Between each measurement, the line was conditioned with 0.4 μl of in-house reference water. Samples were taken for CB25 at 10-18 mm, 23-27 mm, 38-43 mm and 61-66 mm and for CB47 at 26-42 mm, 53-67 mm, 112-129 mm and 141-152 mm, where distances are calculated from the top of the speleothem (DFT; Supplementary Table S11). Several repeats of each sample were carried out on sub-samples positioned along the same growth layers and the reported uncertainties refer to one standard deviation of these repeat measurements. Results are given in per mil notation relative to Vienna Standard Mean Ocean Water (‰ VSMOW).
Trace element analysis. Mg/Ca and Sr/Ca ratios were measured at the Research School of Earth Sciences, Australian National University, using laser ablation inductively coupled mass spectrometry 49 . Analyses were carried out as a continuous 33 mm-long transect across the full thickness of sample CB39 and a 36 mm-long transect of the base of the stalagmite CB47. Analyses were conducted with a 200 × 20 µm laser mask image, calibrated with NIST 610 standard and repeated at a lateral offset of 300-400 µm. The repeat track served to identify outliers caused by photomechanical ablation artifacts, which were then removed from the data. Elemental ratios were converted into elemental concentrations using a standard concentration of Ca in calcite of 400,000 ppm (40 wt%). Data were smoothed using a 9-point running average.