Cumulative growth and stress responses to the 2018-2019 drought in a European floodplain forest

Droughts increasingly threaten the world’s forests and their potential to mitigate climate change. In 2018-2019, Central European forests were hit by two consecutive hotter drought years, an unprecedented phenomenon that is likely to occur more frequently with climate change. Here, we examine tree growth and physiological stress responses (increase in carbon isotope composition; Δδ13C) to this consecutive drought based on tree-rings of dominant tree species in a Central European floodplain forest. Tree growth was not reduced for most species in 2018, indicating that water supply in floodplain forests can partly buffer meteorological water deficits. Drought stress responses in 2018 were comparable to former single drought years but the hotter drought in 2018 induced drought legacies in tree growth while former droughts did not. We observed strong decreases in tree growth and increases in Δδ13C across all tree species in 2019, which are likely driven by the cumulative stress both consecutive hotter droughts exerted. Our results show that consecutive hotter droughts pose a novel threat to forests under climate change, even in forest ecosystems with comparably high levels of water supply.


| INTRODUC TI ON
The frequency and intensity of droughts and corresponding surges of forest dieback events around the globe are projected to increase in the 21st century (Allen et al., 2010;IPCC, 2014). This critically endangers the world's forests and the variety of ecosystem services they sustain, such as their potential to act as carbon sink (Anderegg et al., 2020) and as a nature-based solution for climate change mitigation (Griscom et al., 2017). Recent drought events, moreover, belong to a new category, so called 'hotter droughts', where low precipitation coincides with heat waves, which creates a positive feedback loop between soil water depletion through evapotranspiration and increased surface temperatures through reduced cooling by latent heat production (Allen et al., 2015;Buras et al., 2020). In 2018-2019, Central Europe was hit by two consecutive and hotter drought events, a phenomenon unprecedented at least in the last 250 years but likely to occur more frequently with intensifying climate change (Hari et al., 2020). The 2018 hotter drought alone had | 1871 SCHNABEL Et AL. already stronger negative effects on European ecosystems than the formerly severest drought event in 2003  and induced widespread premature leaf senescence and tree mortality (Schuldt et al., 2020). An increasing number of studies has shown that droughts can affect tree growth and hence carbon cycling in forests for years after the actual drought event and that such 'legacy effects' are widespread in forest ecosystems (e.g. Anderegg et al., 2015;Kannenberg et al., 2019;Szejner et al., 2020). The consecutive hotter drought in 2019 may thus have critically amplified drought stress as trees were hit that already had emptied carbon reserves, impaired hydraulic functioning due to embolism and weakened defence systems (Anderegg et al., 2013;Schuldt et al., 2020) and only access to emptied soil water reserves.
Drought effects on forests can be analysed retrospectively through analyses of tree rings, which are an archive of past growing conditions including climate and water availability (Schweingruber, 1996). In dendroecology the annual growth of trees (i.e. the width of tree rings formed each year) is a principal indicator of drought effects, which can be analysed through comparing growth in drought years with mean growth in a reference period, that is, years with 'normal' climate conditions prior to the drought event (with growth reductions indicating drought stress; Lloret et al., 2011;Schwarz et al., 2020). This growth response to drought can be quantified using the growth resistance index of Lloret et al. (2011), which may be an especially suitable approach when rapid impact assessments are needed and no data are available on the post-disturbance period.
Next to growth, the carbon isotope ratio of 13 C to 12 C in woodcalled δ 13 C-is a widely used physiological indicator of a tree's water status and drought stress (Farquhar et al., 1989;Grossiord et al., 2014;Jucker et al., 2017). Under ample water supply and fully open stomata, trees discriminate against the heavier 13 C in favour of the lighter 12 C. However, under water shortage, stomatal conductance is more strongly downregulated than CO 2 assimilation, which induces an increase in δ 13 C in the wood formed during drought (Farquhar et al., 1989;Grossiord et al., 2014). Thus, drought stress can be quantified as increase in wood carbon isotope ratio (Δδ 13 C) between drought and normal years. Hence, growth responses and Δδ 13 C combined provide a powerful tool to quantify drought effects on trees.
Tree species vary greatly in their susceptibility to drought due to physiological and morphological differences. Among other features such as fine-root distribution and their dieback in response to drought (Brunner et al., 2015;Sánchez-Pérez et al., 2008), two key factors that might drive tree species reactions to drought are stomatal control and resistance to cavitation (Choat et al., 2012;Martínez-Vilalta & Garcia-Forner, 2017;McDowell et al., 2008). Stomatal closure in response to water deficits enables plants to avoid critically low water potentials through transpiration losses and thus hydraulic failure but species differ largely in their type of stomatal control (Martínez-Vilalta & Garcia-Forner, 2017;McDowell et al., 2008): Isohydric or water saving species close their stomata fast during water shortage, while anisohydric or water spending species keep their stomata open and continue to transpire (Martínez-Vilalta & Garcia-Forner, 2017). Next to stomatal control, xylem resistance to cavitation is a key determinant of tree responses to drought as embolism decreases water availability, which leads to desiccation and at extreme levels to tree death (Choat et al., 2012). It is conceivable that stomatal control and cavitation resistance interact, as a water spending behaviour necessitates a continued water uptake via roots and, all else being equal, carries an increased risk for xylem cavitation (McDowell et al., 2008). However, whether this translates into water spending species exhibiting generally higher cavitation resistance and vice versa still remains elusive as some studies found indications for such a correlation (Klein, 2014;Martínez-Vilalta & Garcia-Forner, 2017) while others did not (Kröber et al., 2014). We expect water saving species to show earlier growth and Δδ 13 C responses, while water spending species may face high cavitation risks during severe and prolonged drought conditions characterized by very low soil moisture availability. Hence, for understanding and generalizing the effects of consecutive droughts on forests, tree species should be examined that differ in such traits.
The high tree species richness of floodplain forests (Ward et al., 1999) makes them ideally suited for comparative studies of tree species reactions to consecutive droughts as they are one of the few systems where coexisting mature trees spanning an entire gradient of hydraulic behaviours can be found. Floodplain forests rank among the most rapidly disappearing ecosystems due to land conversion and drainage (Leuschner & Ellenberg, 2017;Mikac et al., 2018) and novel climatic conditions-like prolonged droughts-may amplify this trend through changing the hydrological regimes on which these forests depend. For instance, sinking groundwater levels may increase tree growth sensitivity to drought and susceptibility to drought-induced dieback (Mikac et al., 2018;Skiadaresis et al., 2019) and this might bring these forests, which are among the most dynamic, productive and diverse Central European habitats (Kowalska et al., 2020;Tockner & Stanford, 2002), closer to a tipping point.
On the contrary, the higher water availability in floodplain forests may buffer drought effects to a certain extent as trees might have access to groundwater in addition to precipitation-derived moisture (Heklau et al., 2019). Hence, it is conceivable that if drought effects on growth and Δδ 13 C were observed in floodplain trees, other forest ecosystems might experience even stronger effects.
Here, we focus on the effect of the two consecutive drought years 2018-2019 characterized by extremely hot and dry conditions ( Figure 1a,b), as well as their cumulative effects, on tree growth and Δδ 13 C as physiological stress response. To this end, we reconstruct the stress exerted by this unprecedented event and compare it to past (single) drought events based on tree-ring records from the dominant tree species-Quercus robur L. (hereafter oak), Acer pseudoplatanus L. (hereafter maple) and Fraxinus excelsior L. (hereafter ash)-in the Leipzig floodplain forest, one of the few remaining and thus highly protected floodplain forests in Central Europe (BMU & BfN, 2021;Günther-Diringer et al., 2021). We sampled trees in two environmental strata representing topographic differences in distance to groundwater. We expect the results for the hypotheses proposed below to be more pronounced in the drier stratum.
Specifically, we tested the following hypothesis:

| Study site
In this study, we used data collected from a Central European floodplain forest ecosystem located in the northwest of the city of Leipzig, Germany. The Leipzig floodplain forest is one of the few remaining and thus highly protected floodplain forests in Central Europe (BMU & BfN, 2021;Günther-Diringer et al., 2021) and lies in the transition zone between maritime and continental climate characterized by warm summers, with an annual mean temperature of 9.6 °C and an annual precipitation sum of 522 mm ; DWD, Station Leipzig/Halle). Its main rivers Weiße Elster, Luppe, Pleiße and Parthe formed the floodplain landscape, but their course and thus the floodplain forest itself has been strongly influenced by human interventions over the last centuries (Gutte, 2011).
The straightening of rivers as well as dike and canal constructions strongly influenced the hydrological regime of the floodplain forest, which today does not experience regular flooding anymore (Haase & Gläser, 2009). The floodplain soils originated from an accumulation of alluvial sediments, such as gravel, sand and loam, as result of several glacial periods (Haase & Gläser, 2009). These are nowadays covered by an alluvial clay layer with a thickness between 1 and 4 m, rich in nutrients and with a high pH (around 6-7; Gutte, 2011;Haase & Gläser, 2009). The principal soil available to trees is thus a loamy Vega, with partly gleyed conditions, above gravel and sand filled with groundwater.

| Tree species
The contemporary floodplain forest ecosystem can be characterized as Ficario-Ulmetum Knapp ex Medwecka-Kornas 1952 with oak, elm and ash being the dominant tree species (Härdtle et al., 2020). The absence of flooding, however, resulted in an on-going gradual shift to an oak-hornbeam forest (Galium-carpinetum stachyetosum) and allowed other tree species (especially maple), which are intolerant to flooding, to become dominant. Moreover, elm (Ulmus minor) largely disappeared from the tree canopy due to the Dutch elm disease since the 1960s. Nowadays, the dominant tree species of Leipzig's floodplain forest are oak, maple and ash (Haase & Gläser, 2009;Richter et al., 2016), on which we focus in the present study. These F I G U R E 1 Annual standardized water balance of precipitation minus potential evapotranspiration (a, January-December) and mean growing season temperature (b, April-September) per year from 1979 to 2019 in the Leipzig floodplain forest. The water balance was calculated as standardized precipitation evapotranspiration index (SPEI; Vicente-Serrano et al., 2010). Points are coloured according to their value with deeper red indicating increasing drought and temperature. The horizontal line in (a) represents the long-term mean, negative values indicate water deficits and positive values water surpluses. SPEI values below −1 and above 1 can be considered exceptionally dry and wet respectively. See Figures S1 and S2 for additional SPEI lengths, climatic and hydrological variables that we used to identify drought events three species feature contrasting adaptations to drought in terms of stomatal control and cavitation resistance, which allowed us to explore a range of species response strategies to consecutive drought stress. In terms of stomatal control, former studies classified oak (Cocozza et al., 2020;Thomsen et al., 2020) and maple (Köcher et al., 2009;Lemoine et al., 2001;Leuschner et al., 2019) as rather water saving (isohydric), while ash was shown to follow a water spending (anisohydric) strategy (Köcher et al., 2009;Lemoine et al., 2001;Leuschner et al., 2019). Consistent with this view, studies focussing on sap flow measurements on mature trees, which can be considered as proxy for tree transpiration and hence stomatal aperture, reported a significant downregulation of sap flow with decreasing soil water availability for maple but not for ash (Brinkmann et al., 2016;Hölscher et al., 2005;Köcher et al., 2009), indicating water saving versus water spending modes respectively. Importantly, we view stomatal control here as a gradient and not as a dichotomy between water saving or water spending behaviour (but see Martínez-Vilalta We therefore assembled high-resolution sap flow and soil moisture data recorded during the 2018 drought in the Leipzig floodplain forest (Leipzig Canopy Crane facility) to provide a quantitative comparison of species-specific sensitivity to decreasing soil moisture under severe drought (Figure 2a,b;Schnabel et al., 2021a). Consistent with the classification above, oak and maple significantly downregulated sap flow under drought conditions (indicating water saving behaviour) while ash maintained similar sap flow rates (indicating water spending behaviour; Figure 2b). In terms of species resistance to cavitation, we relied on published values of the water potential at which 50% of xylem conductivity is lost due to cavitation (Ψ 50 , Choat et al., 2012), the most common measure of embolism resistance in trees (Choat et al., 2012). This comparison indicates a similar cavitation resistance in oak (−2.8 MPa) and ash (−2.8 MPa) while maple is less resistant (−1.6 MPa).

| Drought year identification
The definition and identification of drought is central to the analysis of drought effects. Here, we define drought as period with water deficits compared to normal conditions, where 'normal' can be quantified as a percentile of the long-term mean of meteorological or hydrological variables (Schwarz et al., 2020;van Loon et al., 2016).

F I G U R E 2
Soil moisture development (a) and sap flux density (J s ) regulation of oak, maple and ash (b) during the 2018 hotter drought in the Leipzig floodplain forest. The observed soil moisture development (a) was used to delineate two periods with contrasting soil moisture conditions, a moist and dry period respectively. During the dry period soil moisture levels approached 0.24 m 3 /m 3 (red horizontal line), the permanent wilting point of vegetation on clay soils (Weil & Brady, 2017). Boxplots (b) show daily maxima in J s during the 2-month period with moist soil (mid-May to mid-July) and during the period with dry soil at later stages during the 2018 drought (mid-July to mid-September). A statistically significant downregulation of J s under dry compared to moist conditions is indicated by asterisks over the respective species' boxplot (***p < .001; **p < .01; *p < .05). See Methods S1 for details on the sap flow and soil moisture measurements and analyses Following suggestions by Schwarz et al. (2020) we selected drought years based on climatic and hydrological information alone without considering tree growth reductions to avoid a biased selection that could for example result in the exclusion of drought years without reduced growth. We used the standardized precipitation evapotranspiration index (SPEI; Vicente-Serrano et al., 2010) and river discharge data to identify drought years. The SPEI is a commonly used drought index (Hari et al., 2020;Schwarz et al., 2020;Skiadaresis et al., 2019) based on the standardized monthly water balance of precipitation minus potential evapotranspiration. It can quantify drought severity according to a droughts intensity and duration and can be calculated at different time scales (e.g. 1-12 months; Vicente-Serrano et al., 2010). Here, we used three different SPEI lengths that represent the climatic water balance of the main vegetation period (SPEI for 3 months, May-July), the full vegetation period (SPEI for 6 months, April-September) and the full year (SPEI for 12 months, January-December) for each year and with a 40-year reference period (1979-2019; Figure S1). SPEI series were calculated with the SPEI package (Beguería & Vicente-Serrano, 2017) in R from monthly precipitation (mm) and potential evapotranspiration (mm) data derived from the weather station located closest to the study sites (DWD Climate Data Center [CDC], Station Leipzig/Halle, ID 2932; see Figure S1 for details).
We classified years with SPEI values ≤−1 as drought years, years with SPEI values ≥1 as particularly wet and years with values between −1 and 1 as 'normal' (McKee et al., 1993). To take into account the hydrological regime of the floodplain forest, which is in addition to local precipitation strongly influenced by its rivers, we compared the SPEI derived classification to river discharge calculated for the same periods as the SPEIs ( Figure S2). We considered only years without particularly high discharge as drought years.

| Tree selection and increment core extraction
We selected trees for extracting wood increment cores from permanent forest research plots of the 'Lebendige Luppe' (living Luppe river) project (Scholz et al., 2018), which cover a gradient in topographic distances to the groundwater level ( Figure S3). The project features three distinct strata of distance to groundwater: dry (>2 m),  Figure S3). Across these plots, we extracted tree-increment cores from at least 40 tree individuals per species (20 trees per stratum) from each of the three dominant tree species oak, maple and ash, amounting to 120 sampled trees. From each tree, we extracted one increment core at a height of 80 cm with a ∅ 5 mm increment corer (Suunto, Sweden) in January-February 2020, that is, in the winter after tree-ring formation of the second consecutive drought year 2019 was completed. Trees with diameters at breast height (dbh) > 20 cm were selected according to their dominance, past management history and health status. Competition for light is a central determinant of tree growth and δ 13 C that might complicate the detection of drought effects (Grossiord et al., 2014).
We therefore sampled only dominant and co-dominant individuals, that is, trees belonging to category 1-2 according to the classification of Kraft (1884), that were no direct competitors and further excluded plots that showed signs of forest management in recent years. We further selected only healthy appearing trees, excluding those ash trees visually affected by 'ash dieback' (Hymenoscyphus fraxineus) and those maple trees visually affected by the 'sooty bark disease' (Cryptostroma corticale). Both fungal pathogens had caused widespread tree damages and diebacks in the Leipzig floodplain forest during the 2018-2019 consecutive drought and especially very few ash trees were completely unaffected (Wirth et al., 2021). We used the classification key of Lenz et al. (2012) for ash dieback infestation and sampled only trees showing no to only little signs of infestation (levels 0-2 of infestation levels 0-5) based on annual infestation records for 4 years prior to sampling. Importantly, our sample is thus representative for the most vital individuals of the entire population. Since the number of trees fulfilling these strict criteria was too low within the plot area, we sampled also oak and maple trees in the direct vicinity of the plots.

| Tree growth analysis
Tree cores were dried at 70°C for at least 3 days and then clamped in wooden alignment strips. For surface preparation, we used a core microtome (WSL, Switzerland; Gärtner & Nievergelt, 2010) to enhance visibility of tree-ring boundaries. with an accuracy of 1/1000 mm. The measured sequences were cross-dated against a species-specific master chronology developed in former works for the same area as well as against each other using COFECHA (Grissino-Mayer, 2001). This allowed us to identify missing rings, which were more often found in maple trees and in the consecutive drought years 2018-2019. Years without growth were included as zero for the respective year. Sequences that could not be dated unequivocally were excluded from further analysis. The final number of trees included for growth analysis was 114 trees, including 40 oak, 32 maple and 42 ash trees from 11 moist and 15 dry plots. Mean series length was 109 years for oak, 79 years for maple and 94 years for ash trees (Table S1).
Tree-ring width provides an integrated record of past growth conditions as influenced by environmental factors including but not limited to climate and shows an inherent decrease in ring width with increasing tree size (Schweingruber, 1996). As we focus here on climatic influences on growth, we removed age-related trends from the raw tree-ring width chronologies via a negative exponential curve (Fritts, 1976), which provided the best compromise between removing long-term age trends and preserving decadal variability in growth using the package dplR (Bunn, 2008;Bunn et al., 2020). We assessed the climatic sensitivity of tree growth through computing bootstrapped Pearson's correlation functions between speciesspecific chronologies and monthly climatic variables ( Figure S4), using the package treeclim (Zang & Biondi, 2015). Species-specific chronologies are shown in Figure S5.  Figure S1) to growth responses calculated with the mean reference period detailed above.
We quantified drought legacies in tree growth as observed growth minus predicted growth as expected based on the climatic water balance in the year after the drought event (Anderegg et al., 2015;Kannenberg et al., 2019). Tree growth in the last 40 years  was predicted using tree-specific regressions between detrended tree-ring width and SPEI12 of December ( Figure S1). We consistently used the 12-month long SPEI of December to capture the climatic water balance of the full year (January-December) for each species to provide estimates of legacy effects that are comparable between species and drought years. Reported drought legacy effects thus quantify the deviation of observed growth from expected growth based on climate in year 1 after single drought years

| Carbon isotope analysis
The stable carbon isotope composition (δ 13 C) in wood of the same cores was measured following tree-ring width measurements. The where δ 13 C (sample) and δ 13 C (standard) are the abundance ratios between 13 C and ¹²C of the given sample and Vienna PeeDee Belemnite international standard (VPDB). Isotope ratios were expressed in δnotation in per mil units (‰). We calculated the increase in δ 13 C from reference to drought years for each individual tree as indicator of a tree's physiological stress response to drought as: (1) Growth response = Dr growth PreDr growth − 1, (2) δ 13 C = δ 13 C(sample) where Dr δ 13 C is the isotope composition in drought year(s) and PreDr δ 13 C the isotope composition in the reference years (see e.g. Grossiord et al., 2014). Positive values of Δδ 13 C thus indicate higher and negative values lower stress during drought year(s) compared to reference years. Drought and reference years used to calculate Δδ 13 C were the same as in the growth response analysis.

| Statistical analysis
We used linear mixed-effects models (LMMs) to understand the effects of consecutive drought years on tree growth and Δδ 13 C in comparison to single drought years (using their mean growth response and Δδ 13 C in all analysis). We were further interested in understanding how these effects were modulated by changes in a tree distance to groundwater. We fitted species-specific LMMs for analysing the growth response and Δδ 13 C with the packages lme4 between drought event and groundwater level, which however disappeared when using non-detrended growth data or a 1-year reference period. Therefore, to report only the most robust relationships, we present all final LMMs with drought event as the only fixed effect. Final LMMs (Table S2) were fit using restricted maximum likelihood estimation (REML) and marginal means and confidence intervals (95%) were predicted with the ggeffects package (Lüdecke, 2018). We used post-hoc pairwise comparisons with adjusted p-values for multiple comparisons (Tukey's honest significant difference) to compare differences between drought events using the emmeans package (Lenth, 2020; Table S3). Model assumptions (normality, independence and homogeneity of variance) were visually checked through examining model residuals and through quantile-quantile plots. Drought legacy effects were analysed using the same modelling procedure (see Table S2). All analyses were conducted in R version 4.1.0 (R Core Team, 2021).
Tree growth, δ 13 C and drought legacy data and analysis scripts are available via the iDiv data repository (Schnabel et al., 2021b).

| RE SULTS
We found pronounced responses to drought stress in terms of tree growth and Δδ 13 C across the examined tree species, with strongest stress responses in the second of two consecutive hotter drought years (2019). The mean growth response to single drought years (2003, 2006 and 2015) before the 2018-2019 consecutive drought ranged around zero for oak and maple, while growth in ash tended to be reduced (Figure 3a-c). This indicates a similar tree growth in single drought years and in climatically 'normal' years for oak and maple but not for ash. Growth of oak and maple even tended to be higher in 2018 compared to normal years (mean growth response above zero). The hotter drought in 2018 did not induce growth responses in oak and maple that differed significantly from single drought years (p > .1 for both species) but ash experienced an overall significantly stronger growth reduction (t = −2.94, p = .004; Figure 3c; Table S2). In 2019, the second consecutive and extreme drought year, the growth reduction in all species was significantly stronger than in single drought years (oak t = −2.00, p = .049; maple t = −2.74, p = .008; ash t = −7.22, p < .001; Figure 3a-c; Table S2) and in comparison to 2018 (Table S3). Observed growth responses were largely insensitive to the type of growth data (raw or detrended) and reference period (1-year or pooled years) used ( Figures S6 and S7). We used species-specific models but provide evidence for significant differences between species in Figure S8.
Distance to groundwater had an overall small influence on the growth response of the examined species (non-significant effect of groundwater level for maple and ash). Only for oak we found indications for a smaller growth response on moist plots in 2019 (significant interaction of drought year and groundwater level, p = .041). Of the three analysed species, ash, followed by maple, showed a high growth sensitivity to drought (especially to SPEI series of summer months indicating summer drought) while oak was the least sensitive ( Figure S4). Moreover, high summer temperatures negatively affected the growth of ash and maple but not of oak ( Figure S4).
We did not find drought legacy effects in tree growth after single drought years, that is, observed tree growth in year 1 after these droughts was not significantly lower than growth predicted based on climate (Figure 4a-c; Table S2). For ash, observed growth even tended to be higher than predicted (Figure 4c). In contrast, the hot-   However, the magnitude of Δδ 13 C increases varied strongly between drought years and species. For oak and maple, Δδ 13 C was not significantly enhanced in 2018 compared to single drought years (p = .85 and p = .79), while ash had significantly higher Δδ 13 C values (t = 2.85, p = .006; Figure 3f; Table S2). Across all species, we found a strong increase in Δδ 13 C in 2019 compared to single drought years (oak t = 3.93, p < .001; maple t = 2.80, p = .007; ash t = 14.80, p < .001; Figure 3d-f; Table S2) and in comparison to 2018 (Table S3). The Δδ 13 C increase was strongest for ash. Distance to groundwater had no significant influence on Δδ 13 C for all exam-

| DISCUSS ION
Using tree growth reductions and increases in Δδ 13 C as indicators of drought stress, we report a strong increase in drought-related stress in the second of two consecutive hotter drought years across all examined species. Drought responses were consistent for both indicators (growth response and Δδ 13 C; Figure 3), but the timing and magnitude of responses were species specific: Oak showed the overall smallest stress response followed by maple with the strongest response in ash. The 2019 drought, although an extreme drought as well, was meteorologically less severe than the preceding drought year 2018 (Figure 1). This and observed drought legacy effects ( Figure 4) indicate that the cumulative drought effect exerted by both years was likely the principal driver of the stress increase in 2019. The 2018 hotter drought was the severest drought so far recorded in Central Europe Hari et al., 2020;Schuldt et al., 2020), but, as predicted, we found physiological stress increases (Δδ 13 C) to be comparable to former single drought years F I G U R E 3 Growth response and increase in the carbon isotope ratio (Δδ 13 C) in wood of oak, maple and ash in drought years. The figure shows the growth response (upper panels) and Δδ 13 C (lower panels) in the consecutive hotter drought years 2018 and 2019 compared to the mean growth response and Δδ 13 C in single drought years (2003, 2006 and 2015). Zero corresponds to a comparable growth and δ 13 C in drought and climatically normal years. Negative growth response values indicate growth reductions while positive Δδ 13 C values indicate stress increases during drought compared to normal years. The growth response and Δδ 13 C were calculated with Equations 1 and 3 respectively. Black points show estimated marginal means and error bars the 95% confidence intervals of linear mixed-effects model fits, with non-overlapping confidence intervals indicating signficant differences. Coloured points show the growth response and Δδ 13 C values per tree and species (oak n = 40, n = 39; maple n = 32, n = 26; ash n = 42, n = 42) and are jittered to enhance visibility. The tree-ring widths have been detrended with a negative exponential function. Statistically significant differences in the growth response and Δδ 13 C between the years 2018 and 2019 compared to single drought years are indicated by asterisks over the respective year (***p < .001; **p < .01; *p < .05) and tree growth to be largely within the range of climatically 'normal' years. Hence, the comparably high water availability in floodplain forests may partly buffer tree stress responses to single but not to consecutive drought years.
Our conclusion that the effects of single drought years were buffered to some extent contrasts with the dramatic drought effects reported across European forests in 2018 that suffered widespread defoliation, xylem hydraulic failure and mortality Schuldt et al., 2020) but is consistent with other floodplain forest studies. For instance, the exceptionally high gross primary production during the warm spring in 2018 was found to compensate for losses later that year due to drought in a Czech floodplain forest (Kowalska et al., 2020). Similarly, tree growth recovered within 2 years after the 1976 drought for all herein analysed tree species, which was attributed to the buffering effect of water availability in floodplain forests (Heklau et al., 2019). Nonetheless, we found physiological stress increases (Δδ 13 C) in 2018 while tree growth in most species did not react. This confirms the view of clearer drought signals in Δδ 13 C compared to tree-ring width, potentially due to tree growth being maintained from carbon reserves even under low soil water availability (Jucker et al., 2017).
This picture changed dramatically in 2019. As hypothesized, we observed the strongest stress responses in the second consecutive drought year. Drought legacy effects (Anderegg et al., 2015) were found to be widespread in forests and to affect tree growth and Δδ 13 C 1-5 years after the actual drought event (Anderegg et al., 2013(Anderegg et al., , 2015Gazol et al., 2020;Kannenberg et al., 2019;Lloret et al., 2011;Szejner et al., 2020). We observed significant drought legacy effects in tree growth after the 2018 hotter drought but not after former single drought years. Hence, in a system where drought legacy effects have not been observed previously, the hotter drought in 2018 was severe enough to induce such legacies. It should be noted that the reference drought years (2003, 2006 and 2015) were themselves considered as some of the severest droughts in Central Europe (Allen et al., 2015;Büntgen et al., 2021). Their comparably low effect thus supports our view of high water availability in floodplain forests partly buffering tree stress responses and simultaneously underlines the unprecedented nature of 2018-2019. Former studies on drought legacy effects examined post-drought periods during which trees were already (partially) recovering (e.g. Gazol et al., 2020). In contrast, we focus here on two consecutive hotter drought years, unprecedented in severity for at least since 250 years (Hari et al., 2020), which left the trees no time to recover. The few studies that studied prolonged droughts, moreover, did not examine the cumulative built-up of drought effects from year-to-year as they used either mean tree growth across drought years or growth in the last year of drought to calculate growth responses to drought (Schwarz et al., 2020). In comparison, the strong reactions we report for 2019 should be mainly attributable to legacy effects of 2018 (see also some early reports of drought legacies in Buras et al. (2020) and Schuldt et al. (2020)). Other changes in the trees' environment like reduced competition for light are unlikely within a single year. In addition, forest management can be excluded as potential cause as we did not sample trees in stands that experienced recent interventions.
Several physiological mechanisms could explain drought legacy effects (Anderegg et al., 2015) (2003, 2006 and 2015). Legacy effects were quantified as observed minus predicted (detrended) tree-ring width based on climate in year 1 after the drought event. Zero corresponds to growth as expected based on climate conditions, while negative values indicate drought legacies in form of lower than expected post-drought growth. Black points show estimated marginal means and error bars the 95% confidence intervals of linear mixedeffects model fits, with non-overlapping confidence intervals indicating signficant differences. Coloured points show legacy effects per tree and species (oak n = 40; maple n = 32; ash n = 42) and are jittered to enhance visibility. Statistically significant differences in legacy effects between 2018 compared to single drought years are indicated by asterisks (***p < .001; **p < .01; *p < .05) et al., 2020). Under consecutive drought, this damage persists, while vulnerability to cavitation may continue to increase under successive drought stress (Anderegg et al., 2013). In the second drought year, less nonstructural carbohydrates (NSC) reserves were likely left for xylem repair, growth and especially for keeping up the trees' defence system, which increases their susceptibility to pests and pathogens (Anderegg et al., 2013;Hartmann & Trumbore, 2016;McDowell et al., 2008;Schuldt et al., 2020). Although we studied only the most vital tree individuals of the population, thus largely excluding disease effects from our sample, the majority of ash trees in the forest were affected to some degree (Wirth et al., 2021). It is therefore not possible, to completely disentangle whether the species intrinsic traits, incipient ash dieback or their interaction caused the strong stress response in this species. Drought induces shifts in carbon allocation in favour of the canopy and root system at the expense of radial growth (Brunner et al., 2015;Kannenberg et al., 2019), for instance to replace fine roots lost during drought (Brunner et al., 2015), which would reduce tree-ring growth and thereby amplify drought legacy effects. Finally, when photosynthesis is insufficient to meet demands, NSC reserves are utilized to maintain autotrophic respiration, growth and tissue repair (Hartmann & Trumbore, 2016;Richardson et al., 2013). This enriches the reserve pool and tissues built from it in 13 C as the isotopically lighter 12 C is turned over faster than 13 C, which may have further contributed to the strong increase in Δδ 13 C in 2019 in addition to fractionation through stomata closure. . The drought legacy effects we found in tree growth therefore likely resulted from both, physiological and abiotic drought legacies. Next to drought duration and intensity, drought timing may influence tree radial growth (Schwarz et al., 2020). We observed variable timings of climatic drought onset, with single drought years being characterized by both spring and summer droughts, 2018 by summer drought (onset in May) and 2019 by drought during spring and summer (onset in February; Figure   S11). Studies examining intra-annual radial growth at high temporal resolution show that maple, ash and oak species continue to grow until August (if not affected by drought; Brinkmann et al., 2016;Dietrich et al., 2018), which, together with our own observation of strong growth-climate correlations in spring and summer months ( Figure S4), points at all species being effected by drought during their growing phase. Nonetheless, drought effects on growth and Δδ 13 C are likely strongest if the timing of drought is such that both early and latewood development are affected (Schwarz et al., 2020).
That the drought in 2019 affected the entire growing season while the drought in 2018 did not, may therefore-in addition to legacy effects of 2018-have contributed to the strong stress responses we report. Our sampling sites cover the whole gradient of groundwater conditions in the examined floodplain forest but interestingly we found only small effects of groundwater level. The reasons remain speculative. Differences in distances to the groundwater level may have been too small to induce strong effects on tree performance or, alternatively, more intense rooting on dry plots may have compensated for lower water availability (Skiadaresis et al., 2019). We did not observe a temporal trend in groundwater levels (neither decrease nor increase) over the study period and decreases in re- The magnitude and timing of drought stress responses were species specific, which may be related to differences in species hydraulic traits. Oak and ash feature similar cavitation resistance but different stomatal control which may explain the stronger drought stress response observed in ash compared to oak. We report a highly signifi-  et al., 2012). Moreover, a water spending strategy necessitates continued water uptake via roots (McDowell et al., 2008), which may be an especially risky strategy on severely dried out clay soils.
Oak and maple showed similar Δδ 13 C responses in all drought years consistent with their similar stomatal control. In contrast, ash showed a stronger response particularly in 2019. On first sight, this may come as a surprise as one may expect lower Δδ 13 C increases (which are related to stomatal closure) in a water spending compared to water saving species. However, potentially high hydraulic damages in ash during the severe 2018 drought would necessitate a high mobilization of NSC reserves for damage repair. As discussed above, this would enrich the reserve pool and tissues built from it in Δδ 13 C and could explain the strong Δδ 13 C increases we observe in the second consecutive drought year 2019. Future studies should directly measure NSC dynamics during drought to confirm these expectations.
The overall intermediate drought reaction of maple, which is often considered drought sensitive (Leuschner et al., 2019), may be related to its higher vulnerability to cavitation and/or its water saving behaviour that may have prevented severe damages to a certain extent.
Moreover, the reaction of maple may also be influenced by its less exposed crown position (maple trees were rather co-dominant) which can reduce irradiance and water pressure deficits (Montgomery et al., 2010). Finally, short-term growth responses to drought need to be contextualized. We found growth of ash and maple to be sensitive to both features of hotter droughts (low water availability and high temperatures) while oak was insensitive to either factor during the last 40 years ( Figure S4). However, this does not mean that oak does not react to drought, but rather that its response is non-linear as highlighted by its unprecedented response to the 2019 drought.
Other traits may have influenced the responses observed but establishing species-specific differences remains challenging. For instance, ash was reported to have fine-and coarse-root biomass concentrated to shallower soil layers than oak in another riparian hardwood forests (Sánchez-Pérez et al., 2008). However, other studies reported rather deep rooting in ash and an intruding ability to plastically shift its water uptake to deeper soil layers (Brinkmann et al., 2019;Meißner et al., 2012). Similarly, fine-root dieback is, just as aboveground leaf shedding, a common tree response to drought (Brunner et al., 2015;Kuster et al., 2013;Meier & Leuschner, 2008). It thus likely contributed to herein reported drought responses but we lack data on species-and site-specific differences to test this hypothesis.
Despite compelling progress in functional trait research (Kattge et al., 2020), assessments of key drought tolerance traits, particularly fine root and stomatal control related ones, thus remain scarce and should be a research priority in future studies including at our study sites.
Finally, here reported drought effects may be influenced through the naturally high tree species richness of floodplain forests (Ward et al., 1999), as diverse tree communities with dissimilar hydraulic traits may outperform species poor communities through complementarity in water use (Sánchez-Pérez et al., 2008;Schnabel et al., 2019).

| CON CLUS ION
The response of forests to the increasing frequency and intensity of droughts (IPCC, 2014) will affect a variety of ecosystem services and will determine if forests act as carbon sink or source in the 21st century. Our retrospective analysis based on tree rings allowed us a robust comparison of the cumulative stress responses observed in the hotter drought years 2018-2019 compared to responses in former severe drought years (2003, 2006 and 2015) on the same tree individuals. Tree stress responses in 2019 were stronger than in any other examined drought year, indicating that consecutive hotter drought years exert a novel stress. Comparisons of living and dead trees affected by drought show that radial growth reductions are widespread before tree mortality and that sudden changes in tree growth often precede mortality caused by tree hydraulic failure (Cailleret et al., 2017;Obladen et al., 2021). Against this background it is important to consider that we found partly buffered tree stress responses, presumably because floodplain trees are fed by groundwater in addition to precipitation, and examined only the most vital tree individuals of the population. Our results thus show a 'best-case scenario' and more severe tree responses, such as widespread tree mortality, could be expected if entire tree populations or other forest ecosystems were examined (see e.g. Buras et al. (2020), Schuldt et al. (2020, Wirth et al. (2021)). Furthermore, it remains unknown how the here observed responses will affect tree recovery after and resilience to (future) drought, but the reported persistence of legacy effects for years (Anderegg et al., 2015) is worrying. Nonetheless, a species like oak that combines a high tolerance to drought and flood (Scharnweber et al., 2013), may remain resilient, underlining its importance for floodplain forests. Consecutive hotter droughts are projected to become more frequent (Hari et al., 2020). Results of this and similar research may contribute towards forecasting tree species and forest responses to this novel climatic phenomenon.

ACK N OWLED G EM ENTS
We thank our colleagues of the 'Lebendige Luppe' (living Luppe river) project for establishing the plot network that allowed us to draw rep-