This research was supported by the PALEOMEX-ISOMEX (MISTRALS) program of the CNRS-INSU. The authors thank the Versaveau family, owners of Villars Cave, Thierry Baritaud for his help in sampling the stalagmite, and two reviewers for their suggestions and comments.

1 - Introduction

1The ever-increasing number of palaeoclimate proxy records deepens our understanding of climate variability in the past on different temporal and spatial scales (Masson-Delmotte et al., 2013). At the same time, field and laboratory studies on the formation processes of climate archive lead to a better comprehension of the climatic and non-climatic information recorded in the proxies, and contribute to a better evaluation of the proxies’ limitations (e.g. Dreybrodt & Scholz 2011; Offermann et al., 2011; Steen-Larsen et al., 2011; Gessler et al., 2013; Genty et al., 2014; Labuhn et al., 2014). However, it remains challenging to compare and combine proxies from different climate archives, as all are afflicted with their own uncertainties related to dating, climate sensitivity, and linearity, and they are limited to certain aspects of variations in climate. A multi-proxy approach can give a broader perspective on past climate change, and may help constrain the causes of variability (Mann, 2002; Guiot et al., 2005; Li et al., 2010).

2Tree rings have been the principal source of information for hemispheric-scale temperature reconstructions of the past millennium (Mann et al., 1999; Esper et al., 2002; Briffa et al., 2004; D’Arrigo et al., 2006). Their advantage lies in the annual resolution and the precise dating. Tree ring proxies can be well replicated and quantified by comparing them directly to meteorological variables from the year of ring formation. Their calibration is based on the statistical relationships with climate (e.g. Fritts & Guiot, 1990). Models for climate reconstruction based on these relationships can be verified using independent meteorological data, i.e. data that have not been used in the calibration, and the variance explained by the model can be quantified (e.g. Briffa et al., 1992). However, tree ring proxies may be biased towards climate conditions of the time of year that has the strongest influence on tree growth. Tree-ring reconstructed changes in summer climate, for example, might not be representative of the evolution of the annual climate (Jones et al., 2003). Furthermore, tree rings, in particular tree ring widths, are limited for reconstructing low-frequency climate variability because of the standardization process and the limited length of individual tree series (Cook et al., 1995; Esper et al., 2004; Moberg et al., 2005).

3Speleothems have the potential to provide a low-frequency signal. However, their dating is less precise compared to the trees. Despite the high analytical precision that is now achieved for U-Th dating (Hoffmann et al., 2009; Cheng et al., 2013), and the possibility to count annual growth layers in some speleothem samples (Genty & Quinif, 1996; Tan et al., 2006; Baker et al., 2008; Shen et al., 2013), slight changes in the growth rate or short hiatuses increase the error in the chronology, making it difficult to compare a stalagmite record to instrumental data or to high-resolution proxy data from tree rings. While adequate samples, which have a high growth rate and a precisely dated lamination, enable a direct calibration by comparison with meteorological data (Genty & Quinif, 1996; Proctor et al., 2000), in many cases the calibration of speleothem proxies relies on an understanding of the cave processes, based on monitoring data, laboratory experiments, and modeling exercises (Spötl et al., 2005; Genty, 2008; Tremaine et al., 2011; Wackerbarth et al., 2012; Treble et al., 2013). Lastly, an environmental signal can be lagged and/or attenuated before it is transmitted to the cave interior (Fairchild & Baker, 2012), whereas tree ring proxies often reflect the environmental signal more directly, i.e. during the current growing season.

4Few previous studies have attempted to directly compare tree rings and speleothems. Berkelhammer et al. (2014) and Trouet et al. (2009) used tree ring and speleothem proxy records from remote regions to study teleconnections and atmospheric circulation indices. Betancourt et al. (2002) compared annual band widths in tree rings and a stalagmite from the same site and found no correspondence, but their approach has been criticized because even if the banding in both archives is annual, the band width in each proxy might not depend on the same influences (Baker & Genty 2003). However, when tree ring and speleothem proxies from the same region respond to the same dominant factors, e.g. droughts, they can show similar patterns (Sinha et al., 2011; Wassenburg et al., 2013). Managave (2014) used a model to investigate to what extent the oxygen isotopic composition of tree ring cellulose and speleothem calcite can be correlated if they have the same source water. The author determined that a correspondence between these proxies is likely when the variation in the δ¹⁸O of precipitation is high compared to the variation induced by the influences on each single proxy, i.e. the cave temperature and equilibrium conditions during precipitation for calcite δ¹⁸O, and relative humidity, leaf temperature and the isotopic composition of atmospheric water vapor for cellulose δ¹⁸O.

5This article explores the potentials and limits of a multi-proxy climate reconstruction based on speleothem and tree ring proxy records from the southwest of France, a region characterized by recurrent drought periods that might prove to be particularly vulnerable to the consequences of global warming (Moisselin et al., 2002; Itier 2008; Lemaire et al., 2010; Levrault et al., 2010). Identifying the patterns of moisture variability in the past may help evaluate the possible future extent of droughts in a changing climate. Both archives have in common that we can measure oxygen isotope ratios in their components: in cellulose from the wood of tree rings (δ¹⁸O_c), and in water that is incorporated in speleothem calcite, the so-called fluid inclusions (δ¹⁸O_fi). Here, we combine a δ¹⁸O_fi record from a speleothem from Villars Cave (Labuhn et al., 2015) and a δ¹⁸O_c chronology from the nearby Angoulême (Labuhn et al., 2016) in order to reconstruct both low and high frequency drought variability during the last 640 years. The oxygen in both proxies originates from the precipitation feeding the soil water reservoir, which subsequently infiltrates into the cave, or is tapped by the trees. We therefore hypothesize that it is possible to use the δ¹⁸O_fi as an independent estimate of the source water δ¹⁸O for the trees. In a first step, we investigate the co-variation of the fluid inclusion and the cellulose δ¹⁸O time series. In a second step, we assume that the δ¹⁸O_fi represents the δ¹⁸O of the tree source water, and, using δ¹⁸O_c and δ¹⁸O_fi, we calculate the isotopic enrichment of the leaf water above the source water, as well as relative humidity (RH), which is the dominant controlling factor of this enrichment. Lastly, we compare this reconstruction of relative humidity with a previous drought reconstruction based only on tree rings (Labuhn et al., 2016).

2 - Study sites and data

6The study sites Villars Cave and Angoulême are situated in the southwest of France at 50 km distance, in similar geological and climatological settings (fig. 1). As both are located at approximately the same altitude (100-175 m a.s.l.) and distance from the coast (~150 km), and no important relief separates them, they experience similar variations in temperature, moisture conditions (Météo-France, 2009), and in the isotopic composition of precipitation (Millot et al., 2010). The bedrock in the region is a Jurassic limestone, and karstic features such as dolines, caves, and surface collapses characterize the landscape.

7The crossdated, annually resolved chronology of tree ring cellulose δ¹⁸O from Angoulême (45°44’N, 0°18’E) has been constructed using living trees and building timbers of oak (Quercus spp.), and covers the period from 1360 to 2004 (Labuhn et al., 2014; 2016). Interannual variations in δ¹⁸O_c at this site are dominated by the atmospheric conditions that influence leaf water enrichment (Labuhn et al., 2014), but the underlying low-frequency variability are likely linked to variations in the source water δ¹⁸O.

8The fluid inclusion δ¹⁸O measurements were acquired from a stalagmite from Villars Cave (45°26’N, 0°47’E; Labuhn et al., 2015). The record covers the last 2,300 years, and the fluid inclusion samples have a resolution of approximately 25 years. The stalagmite was dated by the U-Th and ¹⁴C methods, as well as by laminae counting. Long-term monitoring series of precipitation and cave drip water δ¹⁸O demonstrated that drip water corresponds to a weighted mean of pluri-annual precipitation at this site, without any evaporative enrichment or seasonal bias due to plant transpiration (Genty et al., 2014). An investigation of corresponding modern fluid inclusion samples indicated that the isotopic composition of the infiltrating water was preserved in the fluid inclusions (Labuhn et al., 2015).

9Monthly relative humidity (RH) data was obtained from Météo-France for Cognac (1961-2012, station no. 16089001), which is the longest available RH record from the study area. The standardized precipitation evapotranspiration index (SPEI; Beguería et al., 2010; Vicente-Serrano et al., 2010) was obtained from http://sac.csic.es/​spei/​database.html, and the time series for the Angoulême grid cell (1901-2011) was extracted.

3 - Methodology

3.1 - Comparison of the fluid inclusion and cellulose records

10For the comparison of fluid inclusions and cellulose, the δ¹⁸O_c time series was smoothed by a 25-year running mean. This interval was chosen because it likely corresponds to the time period integrated in a fluid inclusion sample, as deduced from sample size, stalagmite growth rate and the infiltration time of the water from the surface to the cave (Labuhn et al., 2015). Both time series were transformed to z-scores (i.e. the differences of each value and the mean value of the time series, divided by the standard deviation) to make their variations more comparable.

3.2 - Calculation of leaf water enrichment and relative humidity

11In order to calculate the leaf water isotopic composition and RH using the speleothem and tree ring isotope proxies, we suppose that the δ¹⁸O_fi represents the tree source water. The δ¹⁸O_fi values were linearly interpolated in order to obtain a time series of annual resolution. Then, this time series was smoothed using a 25-year running mean. The resulting interpolated and smoothed fluid inclusion δ¹⁸O values were supposed to be the source water δ¹⁸O for the trees each year. The cellulose δ¹⁸O time series, which had been normalized to a mean of zero (as published in Labuhn et al., 2016), was adjusted to the mean δ¹⁸O value of recent cellulose (31 ‰).

12The oxygen isotopic composition of cellulose can be related to the isotopic composition of source water (δ¹⁸O_sw) and leaf water (δ¹⁸O_lw) (Sternberg et al., 1986; Yakir & DeNiro, 1990):

13where ɛ_wc, the fractionation factor between water and carbonyl oxygen, is approximately equal to 27 ‰. Thus, using our δ¹⁸O_c record and the interpolated fluid inclusion time series as an estimate of δ¹⁸O_sw, we can calculate δ¹⁸O_lw.

14The δ¹⁸O_lw is related to relative humidity (RH) as follows (Dongman et al., 1974):

15

16where δ¹⁸O_v is the oxygen isotopic composition of atmospheric water vapor, and ɛ* and ɛ_k are the equilibrium and kinetic fractionation factors. At 20°C, ɛ* is equal to 9.7 ‰ (Horita & Wesolowski, 1994); ɛ_k is equal to 16 ‰ / 21 ‰ / 32 ‰ for turbulent/laminar/static boundary conditions respectively (Burk & Stuiver, 1981). Under normal European summer conditions, the water vapor is, on average, approximately in isotopic equilibrium with soil water (Förstel & Hutzen, 1982). Thus:

17

18Combining Equations (2) and (3) gives the equation to calculate RH:

4 - Results and discussion

4.1 - Comparison of the fluid inclusion and cellulose records

19The δ¹⁸O records in tree ring cellulose from Angoulême and in speleothem fluid inclusions from Villars Cave display some common trends: a decrease from 1500 to 1700, an increase from 1750 to present, and a marked peak around 1720 (fig. 2). However, there is also a period where the two series show opposing trends, between 1360 and 1500. Furthermore, even if the general increasing trend in the most recent period is apparent in both records, the large increase in the cellulose time series from 1750 to 1850 and the subsequent rapid decrease are not seen in the fluid inclusions. The range of δ¹⁸O values in both the fluid inclusions and the smoothed cellulose δ¹⁸O time series is about 2 ‰ (not shown). However, the common peak at 1720 is about twice as large in the fluid inclusions.

20The co-variation in both records can be ascribed to a common source water δ¹⁸O, which is controlled by the average δ¹⁸O of precipitation in the region. Although the δ¹⁸O_c is strongly influenced by leaf water enrichment on an inter-annual scale, the underlying low-frequency component can be attributed to the variability of the source water (Labuhn et al., 2014). Furthermore, enhanced evaporation and transpiration during dry periods might cause an increase of δ¹⁸O in both proxies (e.g. Bar-Matthews et al., 1996; Denniston et al., 1999; Sternberg, 2009; Gessler et al., 2013). Several factors may contribute to the disagreement between cellulose and fluid inclusions:

21(1) - The isotopic composition of the source water can be modified before it becomes preserved in the proxies. The modern calibration indicates that speleothem fluid inclusions from Villars cave correspond to a weighted mean of pluri-annual precipitation, without any significant modification (Genty et al., 2014), but this might not hold true for the past. The δ¹⁸O_c, on the other hand, reflects partly the isotopic composition of the source water, and partly the isotopic composition of enriched leaf water (eq. 1). Their relative contribution to the δ¹⁸O_c has been estimated (Roden et al., 2000; Cernusak et al., 2005), but is likely to vary over the growing season, as well as over longer time periods (Gessler et al., 2009; Offermann et al., 2011);

22(2) - The δ¹⁸O_c time series might not capture well the low frequency variability in precipitation/source water δ¹⁸O. A correction had to be applied for offsets in average δ¹⁸O_c values between the tree cohorts that were combined in this chronology, which might have removed some of the low-frequency trends (Labuhn et al., 2016). Offsets occur because the source water δ¹⁸O can be different for individual trees, due to local soil hydrology or rooting depth (Phillips & Ehleringer, 1995; Tang & Feng, 2001; Labuhn et al., 2014). Moreover, the oak species had not been determined and both Quercus petraea and Q. robur could have contributed to the chronology. To date, however, no study has shown a difference in δ¹⁸O_c between these two species;

23(3) - The resolution of the fluid inclusion record is much lower than the resolution of the tree rings. Therefore, the peaks in the smoothed cellulose time series are not all seen in the fluid inclusions, e.g. around 1540 or 1850;

24(4) - The dating of the stalagmite is less precise than the dating of the tree rings. Although there is generally a good agreement between growth layer counting, ¹⁴C and U-Th dates, the uncertainty of the stalagmite age model is up to 100 years (Labuhn et al., 2015);

25(5) - The seasonality of the proxies is different. The speleothem fluid inclusions represent the mean annual precipitation δ¹⁸O, which is dominated by the δ¹⁸O of winter precipitation because most precipitation falls during the winter months (Genty et al., 2014). Tree ring cellulose δ¹⁸O could be biased towards growing season precipitation. However, it is possible that trees use water stored in the soil from the previous winter, especially during dry summers (Phillips & Ehleringer, 1995; Daux et al., 2011), so their source water δ¹⁸O may still reflect annual precipitation.

4.2 - Calculated leaf water enrichment and relative humidity

4.2.1 - Leaf water enrichment

26The assumption that δ¹⁸O_fi represents the tree source water is supported by the fact that, at Villars Cave, there is no seasonal bias in the isotopic composition of the drip water due to the transpiration of the vegetation (Genty et al., 2014). This indicates that plants take up a mixed water, and not only growing season water, and this water consequently must have the same δ¹⁸O as the drip water. The calculated leaf water is enriched compared to the source water by 13 to 23 ‰, which is in the range of values found for trees (Saurer et al., 1998a,b; Gessler et al., 2013) and other plants (Sheshshayee et al., 2005; Ferrio et al., 2012).

4.2.2 - Observed vs. reconstructed relative humidity

27We compare the reconstructed RH to the observed summer (JJA) RH, as June to August are the months that give the highest significant correlations between RH and δ¹⁸O_c (Labuhn et al., 2014). The inter-annual variability of RH is well captured by the reconstruction, but average reconstructed RH values are too low (fig. 3). Our calculation is likely too simplistic for the following reasons.

28It assumes the kinetic isotope fractionation as water diffuses through stomata and through the boundary layer (ɛ_k in eq. 2) to be constant; but ɛ_k varies with the boundary layer conditions, which in turn depend on wind speed (Dongman et al., 1974; Burk & Stuiver, 1981). The fractionation factor significantly influences the reconstructed RH. Applying the values given by Burk and Stuiver (1981) for the period 1961-2004 yields an average RH of 36 % (ɛ_k = 16), 46 % (ɛ_k = 21), and 60 % (ɛ_k = 32) respectively. The average of the observed RH during this period is 71 %. The calculation using static leaf boundary layer conditions (ɛ_k = 32) therefore yields the best estimate. However, turbulent conditions are more realistic in nature (Gonfiantini, 1986).

29As we do not have an independent temperature reconstruction covering the time period of the proxy records, we also suppose that the temperature, and consequently the temperature-dependent equilibrium isotope fractionation (ɛ_k* in eq. 2), are constant, which is not realistic. However, according to the fractionation factors given by Barbour et al. (2004) for 20°C and 25°C, the reconstructed RH would only vary by 1 % for a 5°C temperature change.

30Furthermore, our calculation ignores the influence of a Péclet effect, which leads to a leaf water that is less enriched than predicted by the equations. If the calculated leaf water enrichment is too high, the reconstructed RH is too low. The Péclet effect could lower the leaf water δ¹⁸O by 2 ‰ (Barbour et al., 2004). This would translate into a 7 % increase in reconstructed RH according to our calculation.

31Likewise, the calculation neglects that trees can take water from different soil depths (Bréda et al., 1995). Superficial soil water is more enriched than deeper soil water, and its isotopic composition is more variable as it is influenced by evaporation and rainfall events (Tang & Feng, 2001). The average summer precipitation δ¹⁸O is 3 ‰ higher than average annual precipitation in the study area (Genty et al., 2014). This difference in the δ¹⁸O of the source water results in a 10 % change in the reconstructed RH.

32Lastly, it is also supposed that the atmospheric water vapor is in isotopic equilibrium with the soil water. Although this might be a good approximation for a growing-season average (Förstel & Hutzen, 1982), the isotopic composition of the vapor varies significantly on shorter time scales (Berkelhammer et al., 2013).

33The average RH is significantly underestimated by the reconstructions, and none of the factors discussed here are sufficient to explain the differences with the observations, unless we suppose unrealistic boundary layer conditions. A more complex model would be needed to adequately represent RH values, e.g. by taking into account the Péclet effect, as well variations in the depth of the water supply, in the δ¹⁸O of atmospheric water vapor, and in temperature. However, even if some of these parameters could be adjusted to present-day observations, they are unknown in the past and are likely to have varied over time.

34Nevertheless, the inter-annual variability in RH is well represented by the reconstruction. Despite the offset, we can compare relative variations in reconstructed SPEI (based on δ¹⁸O_c) and reconstructed RH (based on a combination of δ¹⁸O_c and δ¹⁸O_fi).

4.3 - Drought reconstruction with and without a source water δ¹⁸O estimate

35The observed summer drought index SPEI (JJA) and summer RH (JJA) are highly correlated during the period of observations (1961-2011) and are comparable to represent summer moisture conditions in the study area (fig. 4). The inter-annual variability in the RH reconstruction based on fluid inclusion and cellulose δ¹⁸O and the SPEI reconstruction based only on cellulose δ¹⁸O (Labuhn et al., 2016) is the same as they both depend linearly on the inter-annual variability in the cellulose δ¹⁸O time series. However, the two reconstructions give a different picture of the long-term moisture changes in the past (fig. 5). The marked δ¹⁸O peak at 1720, for example, disappears in the RH reconstruction, because it is ascribed to changes in the source water isotopic composition, and not to increased drought. Moreover, the RH reconstruction indicates drier conditions during the 16^th and 17^th century, and wetter conditions during the 19^th and 20^th century compared to the SPEI reconstruction.

36The model for the tree-ring based drought reconstruction is well-verified with independent data, and correlations between the proxy and the reconstructed drought index are highly significant and stable throughout the 20^th century, all of which gives strong indications for the validity of this reconstruction (Labuhn et al., 2016). However, the low-frequency drought variability may not be well represented in this record. The RH reconstruction supposedly captures better the low frequency. However, it must be interpreted with caution because of the dating uncertainties in the stalagmite and the low resolution of fluid inclusion measurements compared to the tree rings, as well as the simplicity of the model used to calculate RH. Most importantly, a replication of the fluid inclusion record is needed to confirm whether it represents average regional precipitation δ¹⁸O, and therefore can serve as an estimate of the source water of the trees.

5 - Concluding remarks

37These investigations demonstrate the great potential in combining oxygen isotope ratios in speleothem fluid inclusions and tree ring cellulose to reconstruct moisture conditions, and the theoretical approach is established here. The combination of proxies might provide an estimate of past droughts that comprises both the low- and high-frequency variability, combining the strengths of the two climate archives while compensating their weaknesses when used alone. However, further investigation will be necessary to better understand the differences between this multi-proxy reconstruction and the reconstruction based only on tree rings. In particular, the reconstructed low frequency trends in moisture conditions could be supported by a replication of the speleothem fluid inclusion record using another stalagmite from the same region to confirm their utility to estimate the isotopic composition of precipitation and subsequently the trees’ source water. Likewise, a larger sample depth of the tree ring chronology could reduce the uncertainty due to cohort offsets and better capture low frequency climate variability.

38An improved replication of both speleothem and tree ring records could not only increase our confidence in such multi-proxy climate reconstructions, but also reveal spatial patterns of past droughts in France. This is important because drought severity depends on both the duration of a drought and on its spatial extent. In a future research perspective, there is a potential to find other cave sites in France with forests and historic buildings nearby. These might provide additional oxygen isotope data from speleothem fluid inclusions and tree ring cellulose, which can be used to reconstruct drought patterns in the past and to test the validity of the approach demonstrated in this study.