This work was funded by the research projects MAS Dendro-Avenidas (CGL2010-19274; www.dendro-avenidas.es, Spanish Ministry of Economy and Competitiveness), and IDEA-GesPPNN (OAPN 163/2010; Spanish Ministry of Agriculture, Food and Environment). Meteorological data were freely provided by the Spanish Weather Service (AEMET) thanks to a collaboration agreement with the Spanish Geological Survey (IGME; Jose María Pernía). Juan Vázquez (UAM) facilitated contact between the two research groups involved in this study. Jordi Voltas contributed to the experimental design and assisted us in the statistical analysis. We would also like to thank Javier Rodríguez Arévalo and Marifé Díaz Teijeiro for providing isotope data from the Red Española de Vigilancia de Isótopos en la Precipitación (REVIP), which is run by CEDEX in co-operation with AEMET. JPF was supported by the Ramón y Cajal Programme (MEC, Spain). We thank the comments and corrections from the guest editors Stéphane Cordier, Gerardo Benito and David Bridgland, and the reviewers Matthias Saurer and Alberto Viglione, who improved the original manuscript significantly.

1 - Introduction

1Floods are one of the most catastrophic natural hazards in the world, causing thousands of casualties and great economic loss (Pielke Jr. & Downton, 2000; Mitchell, 2003; Barredo, 2009). Integrated strategies for flood-risk mitigation invariably include a detailed flood frequency analysis (FFA) using a statistical approach (Gumbel, 1941; Kidson & Richards, 2005; Merz & Blöschl, 2008). In the case of gauged basins, the FFA uses mathematical models (a combination of a frequency-distribution function and a parameter fitting procedure) on the series of river discharges (e.g. annual series of maximum daily average stream flows), as a single population of data (Bobée et al., 1993). Nevertheless, floods may have different origins and causes, such as heavy convective rainstorms, persistent frontal precipitation, snow melting, dam failure, etc., and thus cannot be considered as repeated events (Merz & Blöschl, 2003; Díez-Herrero et al., 2009). In several areas, flood events combine two or more of these origins during each year or even during the same season. In this regard, flood-frequency analysis has often failed to take into account the division of stream flows according to their origin (Botero & Francés, 2010). For recent flood events it is easy to use national weather forecast services bulletins, which have been available since the early-mid 20^th century. However, for flood events that took place prior to weather forecasts the origin of flood-inducing precipitation events is unknown and discharge events of different origins are mixed together within the series of extreme events. This contradicts one of the main principles of statistics: the separation of populations prior to analysis.

2Tree-ring studies have been used for over fifty years to analyze flood frequency and magnitude. This is termed dendrogeomorphology or dendrohydrology, which are two subdisciplines of dendrochronology. From the pioneers (Sigafoos, 1964) to recent reviews of the state of the art (St. George, 2010; Stoffel & Wilford, 2012; Díez-Herrero et al., 2013a,b), dendrogeomorphological techniques have enabled us to date past flood events, estimate discharge magnitude, calibrate roughness parameters (a distributed calibration of Manning’s n values; Ballesteros et al., 2011) and even compute the velocity and depth of previous flooding from tilted trees (Díez-Herrero et al., 2013a,b).

3Stable and radiogenic isotopes in tree-rings have also been used for several years to reconstruct past environmental conditions (i.e., temperature, precipitation and water availability) and their pattern of change (see e.g. McCarroll & Loader, 2004; Ferrio & Voltas, 2005; Aguilera et al., 2011; Gessler et al., 2014). In particular, the oxygen-isotope composition (δ¹⁸O) of plant tissues reflects the variation in (1) δ¹⁸O in source water, (2) evaporative enrichment of leaf water due to transpiration, and (3) biochemical fractionation during the synthesis of organic matter (Yakir, 1992; Farquhar & Lloyd, 1993). Generally, the δ¹⁸O of source water is strongly dependent on that of rain water (δ¹⁸O_R). δ¹⁸O_R is mainly determined by the temperature of droplet formation, which is higher with higher temperatures (Dansgaard, 1964). Other factors, such as altitude, precipitation intensity and origin, also affect δ¹⁸O_R; thus, particular rain events may show a distinct isotopic signature (Poussart & Schrag, 2005; Díaz et al., 2007; Liu et al., 2008). Although the isotopic signature of organic matter is also affected by leaf-level processes (for further details, see Farquhar & Lloyd, 1993), in the case of cellulose, about 50 % of the leaf-level signal is further exchangeable with xylem water (Sternberg et al., 2006). This exchange is the main driver for the observed relationship between the δ¹⁸O of tree-ring cellulose and δ¹⁸O_R, as it enhances the original source-water signal, softening the effect of leaf-level enrichment (Saurer et al., 1997; Anderson et al., 1998; Barbour et al., 2001; Offermann et al., 2011). Thus, tree-ring δ¹⁸O has the potential to record the isotopic signature of precipitation events, and therefore to characterise precipitation events from different origins.

4However, to date the two approaches (dendrohydrology and stable isotopes) have not been combined. One possible application of this integration might be to improve flood-frequency analysis for hazard-and-risk assessment. In this regard, we hypothesize that, in a similar way to the dating of flood events using dendrogeomorphic techniques based on anatomic evidence, it may be possible to associate past flood events with their meteorological origin by means of tree-ring isotopes. This paper proposes an innovative approach that could improve flood-frequency analysis by incorporating stable isotope data, using a case study to get a first insight into its possibilities and limitations. This year-by-year or even intraseasonal decoding of isotopic information may represent a powerful tool both for the reconstruction of past extreme flood events and in its application to the assessment and prevention of natural risks, by means of the incorporation of historical and palaeoflood data into flood frequency analysis. Nevertheless, since the isotope signal is derived from a complex combination of physiological and environmental signals, we also assume that the extreme-event signal may be masked by long-term responses, mainly associated with the prevailing meteorological conditions and the physiological attributes of different species. Accordingly, we have also assessed the responses of different functional types to average and extreme events, in order to clarify which would be the best candidate for our proposed approach.

2 - Material and methods

2.1 - Conceptual framework

5The theoretical framework for the proposed approach is summarized in fig. 1. Given the existing variability in isotopes from precipitation, floodwaters originating from different sources are expected to have particular isotopic fingerprints that can be used to identify the meteorological origin of the flood. Furthermore, distinct isotope values in precipitation could be subsequently stored in the isotope signature of tree-rings, given the ability of the latter to record a source-water signal. This classification would allow us to separate different populations of flood events and then apply individual frequency analysis to each data sample, thereby improving estimation of return period in comparison with conventional analyses of the entire mixed dataset.

2.2 - Study areas

6The two study areas are located in the central part of the Iberian Peninsula, in the mountain chain called the Central System, an alpine intraplate pop-up 500 km long, extending from central Portugal to the eastern part of Spain, in a WSW-ENE direction. The highest part of the Spanish Central System is called “the Sierra de Gredos”; its eastern area is formed by the Sierra de la Paramera and the Sierra del Valle ranges, where both study areas are located (fig. 2).

7The first area of study was located in the headwaters of the Gaznata River catchment (Tagus River Basin), at the edge of the Gaznatilla gorge (fig. 2A). The substratum is formed by Variscan granites with thick colluvial and weathering mantles. These form fissured and detrital surficial aquifers respectively. The tree vegetation comprises a pine forest (Pinus sylvestris) planted in the mid 20^th century. The altitude ranges between 1,450 and 1,380 m asl; the average annual precipitation is around 630-700 mm; and the mean annual temperature is 9.6ºC. On September 1^st 1999, a convective storm caused heavy rain in this area and a severe flood flash with three fatalities and significant economic losses downstream (Díez-Herrero, 2003).

8The second study area is located in the Venero Claro catchment (also in the Tagus River Basin), at the drainage divide of many of its tributaries or on the riverbanks (fig. 2B). The substratum is also formed by Variscan granites with thick alluvial formations. These form a fissured aquifer and several detrital surficial aquifers respectively. Tree vegetation comprises a mixed forest of pines (Pinus pinaster; Ballesteros et al., 2010a), oaks (Quercus pyrenaica) and riverside species (Alnus glutinosa, Fraxinus angustifolia, Salix sp., Populus sp., etc.; (Ballesteros et al., 2010b). The altitude ranges between 750 and 900 m asl; the average annual precipitation is around 600-700 mm, and the mean annual temperature is 12.6ºC.

2.3 - meteorological data and isotopes in precipitation

9The methodological proposal is applied to a recent flash flood event occurring in Central Spain. For this purpose, a well-known heavy-rain convective event (01-09-1999) was chosen from the recent flood record. The meteorological data series were supplied by the Spanish National Weather Service (AEMET). All the available series of seven rain gauges around the study areas have been analyzed (fig. 3). For the area of Gaznata, we used precipitation and temperature data from the station at Navas del Marqués, whereas for Venero Claro, we used data from El Tiemblo and Ávila, for precipitation and temperature, respectively. In both cases, relative humidity was obtained from the automatic station in Ávila. The isotopes in precipitation data were supplied by the Spanish Network of Isotopes in Precipitation (REVIP, Díaz-Teijeiro et al., 2009), using the data of the three nearest stations (Madrid-Retiro, 1997-2006 period ; Ávila, 2002-2003 period ; Valladolid, 2000-2006 period; see fig. 3). In addition to the target event, we selected two additional years combining markedly different isotopic signatures in the REVIP stations with relatively strong precipitation events in the study area (2003 and 2005).

2.4 - Methods

2.4.1 - Tree sampling and sample preparation

10In the two forested study areas, samples were taken from at least six trees of each one of four species selected, representative of distinct functional types (two conifer species: the widespread temperate pine Pinus sylvestris L. and the Mediterranean pine Pinus pinaster Aiton; and two angiosperms: a sub-Mediterranean oak, Quercus pyrenaica Willd., and a riparian species, Alnus glutinosa (L.) Gaertn.). The sampling of individual trees started with a rigorous, objective selection, based on their position in the divide to ensure that the roots were as far as possible from the water table; to ensure that the water supply was mainly derived from the unsaturated zone of the soil (Mantese et al., 2012 ; Penna et al., 2013).

11Each tree to be sampled was located (with a GPS receiver) and tagged (with a univocal label), logging the characteristics of the tree itself (species, subspecies, dimensions, condition, etc.) and of its immediate environment (substratum, slope, slope orientation, distance from other examples, etc.). Subsequently, two cylindrical cores were extracted in the approximate direction of the shaft radius, using a standard Pressler increment borer (400 mm long x 5.5 mm interior diameter), which allows a representative sample to be obtained of all tree-rings formed between the bark and the interior of the trunk. Normally the cores were extracted at 1.30 m above ground level, except where this was not possible (particularly for A. glutinosa) because of deformations and defects in the trunk or the presence of low branches. In those cases the cores were extracted lower down. All samples were oven-dried at 60°C for 48 h and one core per tree was sanded with sandpapers of progressively finer grain until tree rings were clearly visible. The remaining core of every tree was kept intact for isotope analysis.

2.4.2 - Tree-ring measurement and dating

12The cores selected for dating were scanned and tree-ring widths were measured using the software WinDendroTM (Regent Instruments, Canada), assisted with visual cross-dating under a binocular microscope. The COFECHA program (Holmes, 1983) was used to evaluate the quality of visual cross-dating. The quality of the average chronology for every site-species combination was examined by calculating the mean inter-series correlation (â) and the Expressed Population Signal (EPS) statistics, as described in Cook & Kairiukstis (1990):

13Where N is the number of individuals and R is the mean inter-series correlation. Standard quality threshold is EPS > 0.85.

Fig. 2: Location of the study are and sampling sites.

14Once the wood cores were dated, the tree-ring corresponding to the event year (1999), along with the preceding and subsequent rings (1998-2000), were identified and isolated for isotope analysis. The same procedure was applied to the outlier years recognised in the isotopic record of precipitation (2003, 2005) and the adjacent tree-rings (2002-2006). In all samples, earlywood and latewood were analysed separately. Additionally, given the larger size of tree-rings in oaks, latewood was divided into two subsections, “early-” and “late-” latewood (hereinafter, latewood-1 and latewood-2).

2.4.3 - Analysis of tree-ring isotopes

15Each wood section was processed individually to obtain the oxygen isotope composition (δ¹⁸O) of alpha-cellulose. For α-cellulose purification we followed the method first described by (Brendel et al., 2000), as modified by Gaudinski et al. (2005). Briefly, the extraction involves soaking the wood samples in a 10:1 mixture of acetic (80 %) and nitric (69 %) acid at 120°C for 45 mn in 1.5 ml polypropylene tubes, followed by a series of rinses using ethanol, deionized water and acetone to remove breakdown products and facilitate sample dehydration. A final purification step uses 17 % w/v NaOH and subsequent rinsing steps with water, diluted acetic acid and acetone to increase the purity of the extracted cellulose. The samples are finally left to dry overnight in a desiccator, then placed in an oven for 2 h at 50°C. For oxygen isotope analysis, ca. 0.40 mg of dry α-cellulose was weighed into silver foil capsules and combusted in an elementar PyroCube (Elementar Analysensysteme GmbH, Hanau, Germany) interfaced to a PDZ Europa 20-20 isotope ratio mass spectrometer (Sercon Ltd., Cheshire, UK), at the Stable Isotope Facility of the University of California-Davis (USA). Isotope ratios were expressed as per mill deviations using the delta-notation (δ¹⁸O) relative to VSMOW standards. The accuracy of the analyses (standard deviation of working standards) was 0.25 ‰.

2.4.4 - Statistical analyses

16To obtain a first insight into the different variability components in the stable isotopic signature of tree-rings, data on δ¹⁸O were subjected to analysis of variance using the SAS proc-mix procedure (SAS Institute, Cary, USA). The analysis included species x tree and the residual as covariance parameters, and species, year, section (early- or latewood) and subsection (latewood-1 and latewood-2, nested to section) as fixed factors, together with the corresponding interactions. Product-moment Pearson correlations between δ¹⁸O of tree-rings (early and latewood) and environmental variables (maximum and minimum temperature, total precipitation, δ¹⁸O in precipitation) for the period 1998-2006 were calculated to determine the variables determining the δ¹⁸O. In addition to total precipitation, we also assessed correlations with monthly maximum precipitation in a single event, as an indicator of the presence of strong rain events. Finally, in order to determine the set of attributes that best define the selected event year (1999), a series of Principal Component Analyses (PCA) were performed. The candidate variables included the robust means of δ¹⁸O for each combination of species and tree-ring sections, resulting from ANOVA, together with the growth index derived from the standardised chronology.

3 - Results

3.1 - Quality of tree-ring dating

17Series sensitivity was generally high in both pine species and in oak, but very low in alder (tab. 1). For P. sylvestris, after removing one outlier tree showing poor correlation with the master series, inter-series correlation increased from 0.59 to 0.62, while the expressed population signal (EPS) remained almost the same (0.896 to 0.892), so the latter chronology was selected for subsequent studies. For P. pinaster, EPS was 0.821, very close to the quality standard of EPS > 0.85, so this series was also included in subsequent studies. In A. glutinosa, however, EPS was extremely low, so the resulting chronology was not reliable enough for dendrochronological studies.

3.2 - Sources of variability in δ¹⁸O of tree-rings

3.2.1 - Analysis of variance

18The analysis of variance (tab. 2) showed significant differences among species, together with a significant year x species interaction, indicating that each species shows a distinct response to environmental conditions, and must therefore be considered separately. Both sections and subsections also showed significant effects, indicating that the environmental signature differs among early- and latewood, as well as between latewood-1 and latewood-2 in oak. Consequently, all subsequent analyses were performed separately for each species and tree-ring section (see average values in tab. 3).

3.2.2 - Correlations with isotope values in precipitation

19Correlations with isotopes in precipitation were generally poor and inconsistent. We found a weak correlation between latewood δ¹⁸O for the pines and inter-annual variations in δ¹⁸O_R for the station Madrid-Retiro (nearest station with a long-term record). The strongest correlations were found with previous autumn δ¹⁸O_R (Sept-Nov, r = 0.73, P < 0.10) for P. sylvestris, and between current spring δ¹⁸O_R (March-June, r = 0.87, P < 0.10) and δ¹⁸O of late wood for P. pinaster. In contrast, we did not find any significant correlation between δ¹⁸O and δ¹⁸O_R for the second station with a long-term record (Valladolid). On a monthly basis, the three REVIP stations were significantly correlated (Madrid vs. Valladolid: r = 0.67, P < 0.001; Madrid vs. Ávila: r = 0.65, P < 0.05; Valladolid vs. Ávila: r = 0.82; P < 0.001); however, inter-annual correlations between the long-term stations (Madrid and Valladolid) were not significant, considering neither annual nor seasonal means (r = [-0.35 + 0.44], period 2000-2006). Due to the scarcity of data, it was not possible to assess inter-annual correlations with the nearest station (Ávila), but the trends for the two years recorded (2002-2003) coincided with those observed in Madrid, and were the opposite of those found in Valladolid.

3.2.3 - Correlations with climate variables

20Correlations between δ¹⁸O in early wood and climate variables mainly showed a spring signal, with precipitation, temperature and humidity in pines, but only temperature and humidity in oaks (results summarised in table 4 and figures 4a & b). In oaks, δ¹⁸O in the two latewood sections showed a consistent relationship with spring conditions (mainly minimum temperature and maximum humidity). We also found a positive relationship between oak δ¹⁸O (early- and late-wood) and both precipitation and minimum temperature of current-year autumn (tab. 4, fig. 4c & d). In P. sylvestris, δ¹⁸O in latewood was positively correlated with mean precipitation in previous-year autumn (tab. 4), whereas for Q. pyrenaica a negative relationship was found between early and late wood δ¹⁸O and maximum relative humidity in previous-year autumn (tab. 4).

3.2.4 - Principal Component Analysis

21With the aim to disentangle the signal of the event year (1999) from average meteorological responses, we performed a series of Principal Component Analyses (PCA). We discarded Alnus glutinosa from the analysis, due to the lack of a reliable chronology for this species. We also observed that including the standard tree-ring chronology for each species as input variables did not provide additional information, so we finally adopted a PCA including only isotope data for the two pine species and the oak (fig. 5). The model explained 83.1 % of total variance with the first two components. The first component (PC1) explained 44.8 % of total variance, and was mainly explained by a positive weight for early wood δ¹⁸O, particularly in pines. For the oaks, we also found a positive weight for latewood-1 δ¹⁸O (Qp_L1 in fig. 5). Maximum values for PC1 (i.e. high early wood δ¹⁸O) were found in 2006, whereas similar low values were found in 1998, 2000 and 2004. The second component (PC2) explained 38.3 % of total variance, and was mainly defined by a positive weight for latewood δ¹⁸O in Pinus sylvestris (Ps_LW in fig. 5). A clear outlier year for this component was 1999, showing the lowest values for δ¹⁸O, whereas 2003 showed values slightly higher than average.

4 - Discussion

22For all species, we found that the δ¹⁸O was mainly explained by the environmental conditions during spring. For the two pines, spring precipitation showed a strong negative effect on earlywood δ¹⁸O, while temperature was positively correlated. This suggests an effect of spring water availability, on the one hand, and water demand, on the other, thus mainly related to enrichment effects at the leaf level (Farquhar & Lloyd, 1993), which in turn are associated to the tight stomatal regulation of pines (Barbour et al., 2001; Ferrio & Voltas, 2005; Barnard et al., 2007). This appears to be a common feature among pine species, particularly under Mediterranean conditions (Ferrio & Voltas, 2005; Ferrio et al., 2005; Granados, 2011). However, this was not observed in latewood, which under Mediterranean conditions would be expected to be more strongly influenced by evaporative demand. The reason for that might be that in the area of study inter-annual variability during summer was much smaller than during spring (in all stations, inter-annual standard deviation in winter and spring was, respectively, ca. 4 and 2 fold higher than in summer months). On the other hand, the oaks also showed a negative correlation with spring precipitation, but were more closely correlated with temperature, which is consistent with their reliance on deep water reservoirs, showing less dependence on growing-season precipitation (Aguilera et al., 2011; Michelot et al., 2012; del Castillo et al., 2013; Klein et al., 2013). Interestingly, in oaks not only the earlywood but also the latewood signal was mainly affected by spring conditions, although monthly correlations showed peaks moving from April in earlywood to June in the second half of latewood (data not shown). This is in agreement with the stronger reliance on stored carbohydrates of deciduous angiosperms, as compared to the evergreen conifers, which imply a carry-over signal from spring that can extend throughout the growing season (Damesin et al., 1998; Barbour et al., 2002; Gessler et al., 2009; Michelot et al., 2011; Offermann et al., 2011). On the other hand, previous-year signals and age effects might have partly masked the response of δ¹⁸O in tree rings to current-season conditions (see e.g. Labuhn et al., 2014). In this regard, although the short series studied did not allow assessment of age effects in δ¹⁸O, they are likely to be irrelevant for the pines, but might have introduced an additional source of error in the two species including younger individuals, Q. pyrenaica and A. glutinosa (see tab. 1).

23Interestingly, we also found some correlations between current-year autumn maximum precipitation (as an indicator of heavy rains) and the δ¹⁸O of spring wood (see table 4). Although this relationship is not causal, it may indicate an underlying connection between spring isotopic signature in precipitation and the prevalence of high-intensity rain events, which may deserve further study as a potential predictive tool.

24The PCA confirmed the results from correlation analyses, showing that the main source of variability for δ¹⁸O in tree-rings was related to spring conditions. Thus, the first component was mainly correlated with the early-wood δ¹⁸O in both pines and oaks (fig. 5, Ps_E, Pp_E and Qp_E). For oaks, the first component was also correlated with δ¹⁸O in the first section of latewood (fig. 5, Qp_L1), which was also responsive to spring conditions, as shown in table 4. Considering the different location of P. sylvestris (Gaznata) and the other two species (Venero Claro), this agreement suggest a strong spatial coherence for the spring signal, which further supports a more significant role of temperature as compared to precipitation, since the former shows better spatial correlations than the latter (Ninyerola et al., 2000; del Castillo et al., 2013). Interestingly, the only single variable explaining the second principal component was latewood δ¹⁸O in P. sylvestris, which in turn was mainly influenced by the particularly low value of the event year, 1999. These results are particularly encouraging, since they suggest that the lack of clearer correlations with autumn precipitation values was mainly due to the restricted spatial coverage of the heavy rain event in 1999, which was only weakly reflected in meteorological records. In contrast, the isotope signal in tree-rings showed a clear response of P. sylvestris growing in the area affected by the extreme event (Gaznata), but left unaffected the other two species, which were growing in the Venero Claro area. Although this still requires further assessment with a more extensive dataset, these results confirm our working hypothesis that a clearer signature of extreme events could appear once background trends in response to average climate variables are removed.

25Despite the clear signature of the 1999 event in tree-rings, we did not find clear relationships between δ¹⁸O in tree-rings and in precipitation. However, this could be explained by the erratic spatial distribution of precipitation events, which also explains the weak inter-annual correlations between δ¹⁸O of precipitation for the different meteorological stations included in the study. In particular, convective precipitation, such as the 1999 event studied, are developed in a very localized way, covering only a few tens of kilometres (Díez-Herrero, 2003). Therefore, the isotopic response of these events is restricted to the area where precipitation occurred, and may not have been collected even by the nearest meteorological stations. In particular, within our area of study, the Central System range constitutes a major barrier for the distribution of convective storms, which follow different pathways in the two areas (Ávila-Madrid and Valladolid). Whereas convective nuclei in Madrid and Ávila usually move along the Tajo and Jarama Valleys, in Valladolid they move along the Duero Valley. In contrast, precipitations associated with Atlantic fronts, such as those occurring in winter and spring, are produced in large regions, leaving their isotopic fingerprint over very large areas. Although this could be considered a limitation of the isotopic approach, it also opens an alternative line of study, taking advantage of the ability of tree rings to record mainly precipitation events of frontal origin. In particular, we hypothesize that strong but site-specific signals in the tree-ring record might be associated with local precipitation events of convective origin, whereas very consistent signals across a wide range of sampling sites would indicate a more generalized event, one likely to be of frontal origin. However, this hypothesis can only be tested by incorporating sampling sites in the vicinity of those stations where an extensive record of isotopes in precipitation is available. Following a similar reasoning, it is likely that other kind of floods, e.g. those caused by snow melt, could be also distinguished in the isotopic record, given the particularly depleted isotope signature of snow (Tang & Feng, 2001; Treydte et al., 2006). In this case, we would expect this signal to appear in early wood, and to be particularly strong in evergreen species with early phenology (Treydte et al., 2006), but it might not appear in deciduous species, which are not actively transpiring during snowmelt (see e.g. Treydte et al., 2014).

26Another possible source of error that can mask the results of the correlation is the influence of the amount of precipitation on isotopic composition; this has been discussed in several studies in recent decades (Poussart & Schrag, 2005; Liu et al., 2008). Similarly, very short and intense flood events are likely to have a greater run-off/infiltration ratio (see e.g. Langhans et al., 2010), and thus might not be able to modify the isotope signature of soil water, and subsequently would not be stored in tree rings. On top of that, the response of tree-ring isotopic composition to precipitation amount is not only due to the isotopic signal of precipitation and soil water (Granados, 2011), but has also a strong leaf-level signal that is carried over from the leaf to xylem cellulose (Ferrio & Voltas 2005; Gessler et al., 2009; Gessler et al., 2013). Briefly, higher water availability or humidity would increase stomatal conductance, reducing leaf-water isotopic enrichment during evaporation, and this signal would be transferred to exported sugars, thus potentially contributing to the isotopic signature of wood (Offermann et al., 2011; Gessler et al., 2013; Treydte et al., 2014). However, the contribution of a leaf level signal can be variable with species and even during the season (Offermann et al., 2011; Gessler et al., 2013), and on the other hand leaf-water signature is further complicated by the opposing effects of transpiration flow and evaporative enrichment (Farquhar & Lloyd 1993; Ferrio et al., 2012). In this context, since the exchange of oxygen isotopes during cellulose biosynthesis is variable at the intra-molecular level, novel methods have emerged that will allow leaf and source signals in tree-ring archives to be disentangled in the future (Ellsworth et al., 2013). Meanwhile, the selection of tree species has also a significant influence on the isotopic response, because the different climatic sensitivity of the tree species, as is also demonstrated in the results presented, is in line with previous research (Saurer et al., 1997; Treydte et al., 2007; Battipaglia et al., 2008; Daley et al., 2010).

27Despite the aforementioned methodological limitations, tree-ring oxygen stable-isotopes have a potential utility in flood-hazard assessment. Specifically, this source of data could be used in the improvement of the frequency analysis and in the management and assessment of hazard. In this respect, the most straightforward method for estimating flood frequency is to adjust a cumulative probability function to the recorded annual maximum flows. However, this approach produces flood quantile estimators with a high degree of uncertainty (Botero & Francés, 2010). Oxygen stable isotope data, combined with non-systematic data from the past (e.g. historical and paleo-flood censored data), could increase the information available and, therefore, reduce uncertainty in estimates. The ability of stable isotopes to explore the meteorological origin of precipitation could also help to characterize hazard, which is a process that is highly dynamic due to the seasonal climatic pattern and variability in vulnerability. So, seasonal effects may lead to significant differences between the flood hazard in summer and winter and, at the same time, vulnerability to extreme events can be also variable throughout the year (Merz et al., 2007).

5 - Conclusion

28Although the results are preliminary and based on a limited number of years, our study suggests that the isotope signal in tree-ring cellulose for both pines and oak is dominated by spring conditions, seemingly precluding its use as a proxy for precipitation events occurring during late summer-autumn. However, after removing this “spring signal”, we found a clear and site-specific signal in latewood δ¹⁸O, associated with the heavy rain of autumn 1999. Thus, we cannot discard our working hypothesis that a clearer signature of extreme events could appear once background trends in response to average climate variables are removed. Nevertheless, we did not find a direct relationship between δ¹⁸O in precipitation and tree-rings, most likely due to the strong spatial heterogeneity of precipitation. Therefore, we still lack a direct proof of a causal association between an extreme-event isotopic fingerprint and the observed site-specific signal in tree-ring δ¹⁸O. In this regard, our results encourage further research to test our initial hypothesis more thoroughly, by considering a larger number of events, and using plant material as close as possible to meteorological stations where a record of isotopes in precipitation is available.