To obtain QWA and δ¹³C data, we resampled the Scots pine increment cores taken in 2016 at one of the five study sites presented in Buras et al. (2018), namely the site ‘Maisenlach’ which is located south of the city of Schwabach in Franconia, Germany (lon: 11.029°, lat: 49.305°, WGS 84). Among the 5 sites, we specifically selected this site due to its south-facing forest edge and thus strongest micro-climatic effects at the edge (Chen et al., 1993; Matlack, 1993; Young and Mitchell, 1994) as well as the largest sample size of available increment cores (see Buras et al., 2018). Two other stands from Buras et al. (2018) also featured south-facing edges, but the site ‘Neustadt an der Aisch’ represents a small forest patch with overlapping effects from different forest edges, while the site ‘Hessdorf’ was characterized by a fragmented forest-edge which may cause additional modifications to the forest microclimate.

At the time of sampling the ‘Maisenlach’ forest stand was dominated by Scots pine with a mean age of 102 ± 17 years. After the extreme drought of 2015, a remarkable tree canopy browning and dieback of individual trees at the forest edge was observed in spring 2016, which lead to the investigations presented in Buras et al. (2018). Considering tree dimensions, diameter at breast height and remotely sensed canopy area statistically belonged to the same distribution (p > 0.05), while tree height was significantly (p < 0.001) lower at the forest edge (see Figure 2 in Buras et al., 2018). Indexed ring-width series from the ‘Maisenlach’ featured a high growth synchronicity as indicated by an average Gleichläufigkeit (glk, see Eckstein and Bauch, 1969; Buras and Wilmking, 2015) of 0.67, a mean inter-series correlation (rbar) of 0.47, and an expressed population signal of 0.98 (EPS, see Wigley et al., 1984; Buras, 2017). The forest soil was characterized as a stagnic Gleysol.

To identify the most-suitable tree individuals for our purpose, we utilized the initially obtained 46 individual ring-width series consisting of 8 dead edge trees, 18 live edge trees, and 20 live interior trees. The ring-width series were detrended into ring-width indices (RWI) using an autoregressive model (known as pre-whitening) to emphasize the high-frequency growth variations representative of the climate signal in ring-width variations. Subsequently, we applied a Principal Component Gradient Analysis over the common overlap period 1942–2011 (PCGA, Buras et al., 2016). PCGA allows for identifying ecological gradients within populations of tree-ring samples that are expressed in subtle variations in RWI series (Buras et al., 2016; Rehschuh et al., 2017). This population-inherent gradient was used to identify each five specimens of each of the three categories (dead vs. live at the edge vs. interior trees) within the population gradient. Here, we aimed at selecting specimens which featured extreme PCGA-ranks along the population-inherent gradient. This was done to particularly emphasize on growth patterns representative of the three different categories. However, since some of the samples could not be used for QWA assessments due to suboptimal wood quality of the increment cores (e.g., wood decay and/or cracks), we slightly modified the statistical selection by selecting alternative individuals along the population gradient. Nevertheless, the eventually selected samples represent statistically significantly differing growth patterns between edge and interior trees as indicated by a pairwise Wilcoxon rank-sum test (see Supplementary Figure S1). To confirm that these subsamples represent the corresponding tree categories, we computed inter-series correlations and glk between subsample chronologies and their corresponding counterpart including all trees from a specific tree category. Since all correlations between subsample and counterpart chronologies were larger than 0.88 (p < 0.001) and glk values were larger than 0.82 (p < 0.001), we considered the subsample chronologies representative of the three tree categories. Furthermore, we tested whether diameters at breast height of the three subsample categories belonged to the same population using a pairwise Wilcoxon rank-sum test which returned non-significant (p > 0.05) results for all three pairwise comparisons. For each of these altogether 15 trees, we further utilized (1) detrended RWI (see above), (2) basal-area increments (BAI) based on ring-width and measured diameters at breast height as a measure of secondary growth (bai.out function in the ‘dplR’ R-package), (3) various QWA traits (see section 2.2) at intra-annual resolution as well as (4) δ¹³C values (see section 2.3) at annual resolution to address the above mentioned four hypotheses H1–H4.

The cell anatomical measurements were performed on cross-section images from microslides obtained from the outmost 20 annual rings of each selected radial stem core using standard protocols (Gärtner and Schweingruber, 2013; von Arx et al., 2016). Resin was extracted using a Soxhlet apparatus for 24 h. After that, cores were split into 2–3 cm long segments, embedded in paraffin, and cut in 12-μm-thick transverse sections using a rotary microtome (Leica RM2245, Leica Biosystems, Nussloch, Germany). Sections were stained with a 1:1 Safranin-Astrablue solution and permanently mounted on glass slides with Euparal (Carl Roth, Germany). For each slide, digital images were taken using a slide scanner (Axio Scan Z1, Zeiss, Germany). ROXAS software (von Arx and Carrer, 2014) combined with Image-Pro Plus (Media Cybernetics, Rockville, MD, USA) was used to semi-automatically identify tracheids from the anatomical images (von Arx et al., 2016; Prendin et al., 2017). Tracheid identification and tree-ring dating was additionally quality-checked by an experienced operator. Cell lumen area and cell wall thickness of each tracheid, together with its positional information within the dated tree ring were eventually obtained.

The following xylem anatomical traits were subsequently calculated from the initial cell measurements and integrated at the ring level for early- and latewood separately. The earlywood-latewood boundary was defined by applying the common threshold of a Mork value of 1 (Denne, 1989). Although the ROXAS output provides a multitude of wood anatomy traits, we focused on the mean cell-wall thickness (average of radial and tangential walls) of the earlywood and latewood, the anatomically derived mean ring density, i.e., a proxy for annually resolved wood density on a relative scale that is calculated as the ratio between cell wall and whole-cell area and then multiplied by 1.5 g/cm³ (Björklund et al., 2017, 2019), the earlywood mean hydraulic diameter Dh according to Tyree and Zimmermann (2002), and the early- and latewood cell-wall reinforcement (t/b)² of the entire ring according to Hacke et al. (2001), where t is the double-cell wall thickness and b is the lumen diameter of a given tracheid measured perpendicularly to the double-cell wall, thus expressing the implosion safety that is strongly correlated with hydraulic safety or P₅₀ (Hacke et al., 2001). We are well aware that some of these metrics are not independent from each other (e.g., cell-wall thickness and mean ring density, Dh and cell-wall reinforcement). Nevertheless, we decided for evaluating all of them, since their combined interpretation allows for a more comprehensive picture of possible wood-anatomical differences between the investigated groups and in context of our hypotheses. Eventually, the time series were truncated to the period 1998 to 2015 to assure that each of the three categories was represented by at least three individuals in a given year (individual years from individual samples were excluded due to micro-cracks).

For δ¹³C analysis of cellulose, tree rings of the last 15 years (2001–2015) of dead and live edge and living interior trees (n = 5, respectively) were analyzed. Approximately 15 mg of chopped material per tree ring was extracted in Teflon filter bags (Fiber Filter Bags F57, Ankom Technology, Macedon, NY, USA) using 5% NaOH at 60°C for 2 h to remove hemicellulose (Galiano et al., 2017); this procedure was carried out twice. In the following, samples were soaked in 7% acidified NaClO₂ (using 96% acetic acid) at 60°C for 20–40 h to fully remove resin and lignin. The extraction solution was renewed every 10 h. Lastly, Teflon filter bags were oven-dried for 12 h at 60°C and samples removed. Samples were then soaked in deionized water for several hours and homogenized using an Ultrasonic transducer (Laumer et al., 2009). Samples were finally freeze-dried for 36 h.

To determine the isotopic composition, ca. 1 mg of homogenized cellulose material was weighed into tin capsules. Subsequently, samples were combusted to CO₂ in an elemental analyzer (EA1110 CHN; Carlo Erba, Milan, Italy) linked to an isotope ratio mass spectrometer (Delta XL; Thermo Scientific, Bremen, Germany). Laboratory and international standards with known δ¹³C were used for calibration, resulting in a precision of 0.2‰.

The ¹³C/¹²C ratios are expressed in δ notation (‰) relative to the international standard Vienna Pee Dee Belemnite (VPDB; R_Standard = 0.0111802):

Since we were only interested in relative difference of δ¹³C among the different tree groups investigated, δ¹³C of wood cellulose was not corrected for atmospheric δ¹³C to account for the increase in ¹³C-depleted CO₂ emissions with industrialization (Loader et al., 2007), since a corresponding correction would not change the relative differences due to a common correction procedure. Altogether, the sampled δ¹³C thus represent the annual isotopic signal integrated over a whole growing season that result from the transport of photo-assimilates from the canopy to the sinks (in case of our samples secondary wood formation) via the phloem. We therefore consider them as proxies of assimilation rates, intrinsic water-use efficiency and stomatal conductance at the canopy scale (Klein et al., 2013).

To assess whether the sample selection represents statistically differing population samples (and thus confirm representativity of the groups presented in Buras et al., 2018), we re-ran a PCGA on the subset of the 15 RWI time series. Subsequently, we computed group-wise master-chronologies, i.e., for each parameter and group we computed the corresponding annual mean value to represent group-specific variations of tree-ring parameters. These chronologies were then compared against each other in context of our formulated hypotheses. Since the data were partly non-normally distributed (as tested for using Shapiro–Wilk normality test), we consistently applied non-parametric tests, i.e., Wilcoxon rank-sum test and Spearman’s rank correlation.

To obtain a general overview on differences in growth patterns (RWI) and growth levels (BAI), we firstly visually compared the RWI- and BAI- master chronologies of the three different categories. Moreover, we for each of the three categories calculated glk, rbar, and EPS from individual series as a measure of growth synchronicity and statistically compared glk and inter-series correlations among the three categories using a pairwise Wilcoxon rank sum test.

In terms of addressing H1, we assessed the averaged δ¹³C time series (i.e., category-specific chronologies) for significant differences between the three groups using a paired (by year), pairwise Wilcoxon rank-sum test. Here, a significant result indicates systematically different yearly values in the compared chronologies. In case H1 were true, forest interior trees should feature significantly lower δ¹³C values compared to forest-edge trees, with dead forest edge trees featuring highest δ¹³C values, i.e., lowest photosynthetic discrimination against ¹³C.

To address H2, we correlated δ¹³C time series with the standardized precipitation evapotranspiration index (SPEI, Vicente-Serrano et al., 2009) as retrieved from the corresponding grid-cell of CRU monthly temperature and precipitation data (version 4.06; Harris et al., 2020). The SPEI represents the locally standardized climatic water balance (P-PET; Thornthwaite, 1948) which was shown to indicate plants sensitivity to water supply during the growing season (e.g., Buras et al., 2018). To represent water availability over the full growth period, we integrated SPEI over 9 months representing January–September SPEI for each year. We included the water balance in late winter/ early spring (i.e., January, February, March) since this better reflects soil water availability at the onset of growth. The January–September SPEI time series was finally correlated with each of the three δ¹³C chronologies. These correlations allow for directly addressing H2, which refers to the coupling of tree-ring δ¹³C with plant water availability as reflected by January–September SPEI. That is, in case H2 were true, the obtained correlations should be highest for dead edge trees, followed by live edge trees and live interior trees, reflecting a higher frequency of reduced stomatal conductance induced by low soil water potentials at the forest edge and in particular for dead edge trees.

Regarding H3, we statistically compared category-specific chronologies of earlywood and latewood cell-wall thickness as well as mean ring density using paired (by year) pairwise Wilcoxon rank-sum tests. In case H3 were true, all three traits should feature lower values at the forest edge in comparison to the forest interior, potentially reflecting a lower availability of photo-assimilates for xylem formation.

Finally, to address H4 we statistically compared category-specific chronologies of the mean hydraulic diameter as well as the cell-wall reinforcement as proxies for tracheid-implosion safety using paired (by year) pairwise Wilcoxon rank-sum tests. In case H4 were true, dead trees should feature statistically higher mean hydraulic diameters and lower cell-wall reinforcement compared to live trees.

All analyses were conducted in ‘R’ (version 4.2.2; R Core Team, 2022) extended for the packages ‘SPEI’ (Beguería and Vicente-Serrano, 2013), ‘dplR’ (Bunn, 2008), and ‘dendRolAB’ (Buras, 2022).

Regarding H1, we found significantly higher δ¹³C values for forest-edge trees but no significant differences between live and dead edge trees. As initially hypothesized, we interpret this finding as a systematically earlier reduction of stomatal conductance of forest-edge trees compared to forest-interior trees under dry conditions. Photosynthetic carbon isotope fractionation and thus the resulting δ¹³C in the assimilates is in a first approximation proportional to the relationship between the leaf internal (C_i) and the ambient CO₂ (C_a) concentration (Farquhar et al., 1982), which is in turn determined by the assimilation rate on the one hand and by stomatal conductance on the other hand (Gessler et al., 2018). Any increase in δ¹³C might thus be due to a higher assimilation rate or a lower stomatal conductance since they both decrease C_i and thus photosynthetic carbon isotope fractionation. The observed reduction of tree growth at the forest edge since approx. 1990 (Figure 1) as well as the lower cell-wall thickness (Figure 3) suggests that the higher δ¹³C values of edge trees are an indication of reduced stomatal conductance and not of higher assimilation rates. It may be argued, that edge trees are comparably less shaded resulting in a higher assimilation rate. However, Brandes et al. (2006) showed that there is no shading effect on δ¹³C of assimilates in Scots pine. Moreover, the study area of Franconia, Germany, has experienced an increasingly drier climate since 1990 (Buras et al., 2018; Supplementary Figure S4), which coincides with a lower growth of forest-edge trees compared to forest-interior trees since then (Figure 1B) and supports the assumption of a drought-induced reduction of stomatal conductance as the primary reason for the observed differences in δ¹³C (Figure 2A). Namely, due to the more frequent reduction of stomatal conductance, forest-edge trees featured a relatively lower carbon assimilation, which likely resulted in the observed lower growth values since 1990. Another possible explanation for reduced growth rates at the forest edge is a possibly enhanced respiration at the forest edge. However, since absolute growth at the forest edge declined in the 1990ies (Figure 1B), i.e., under an increasingly drier climate in the region (Buras et al., 2018), it seems more likely that the relatively isohydric Scots pine specimens studied here, reduced stomatal conductance at the cost of photosynthesis. This interpretation is supported by previously published research. For instance, Pellizzari et al. (2016) reported a higher δ¹³C of declining Scots pine individuals when compared to neighboring non-declining individuals. Another study reported a rapid decrease of δ¹³C after irrigating Scots pine in a dry inner-alpine valley which coincided with increased growth rates (Eilmann et al., 2010). In our interpretation, it is yet important to mention that differences in rooting depth and thus soil-water access may affect the observed differences in δ¹³C of edge and interior trees. Unfortunately, corresponding data was not available wherefore this hypothetical influencing factor remains subject to further investigation. We want to stress, that δ¹³C is not a direct measure of whole-tree carbon fluxes at the time of wood formation (e.g., due to the mobilization of non-structural carbohydrates) wherefore there might be additional factors (e.g., carry-over effects) determining the variability of δ¹³C time-series. Yet, since our results are in line with observations for Scots pine in other investigations (Eilmann et al., 2010; Pellizzari et al., 2016) and given the observed positive relationship between δ¹³C and January–September SPEI, it seems likely that stomatal conductivity and intrinsic water-use efficiency are to a high degree reflected by our measurements.

A potentially more frequent reduction of stomatal conductance at the forest-edge compared to the forest-interior as suggested by our results is likely caused by the different micro-climatic conditions. First of all, the forest edge generally features a micro-climate more representative of open land, which has been reported to be warmer and drier particularly under drought (Chen et al., 1993; Bonan, 2008; De Frenne et al., 2021; Meeussen et al., 2021). Secondly, forest-edge trees feature a higher leaf-area index due to more leaves at lower branches compared to the forest interior despite a similar canopy area (Sherich et al., 2007; Buras et al., 2018) potentially causing higher water losses via transpiration. In other words, the higher leaf-area index due to more favorable light conditions at the forest edge results in an unsupplied water demand of the canopy under drought, which can be interpreted as structural overshoot (Jump et al., 2017). Under drought, these two effects in combination result in a faster soil water depletion and thus more negative soil water potentials in comparison to the forest interior. Relatively isohydric tree species – such as the here investigated Scots pine – reduce stomatal conductance under increasingly negative water potentials (Klein, 2014). In conclusion, the physiological drought-response of edge trees did not only result in relatively higher δ¹³C values but also reduced secondary growth in comparison to forest interior trees (see also relative temporal changes of BAI in Figure 1). The previously shown higher SPEI-sensitivity of RWI derived from forest-edge Scots pines in comparison to the forest interior (Buras et al., 2018) supports this interpretation. That is, the higher SPEI-sensitivity of forest-edge trees reflects a stronger coupling to soil-water conditions compared to the forest interior, which is in line with the observed ongoing growth decline at the forest edge since the 1990ies under an increasingly drier climate. Further support to this interpretation comes from Buras et al. (2018), who reported significantly lower latewood-earlywood ratios at the forest edge, potentially indicating resource shortage in the typically warm and dry later part of the growing season. The higher sensitivity of δ¹³C to January–September SPEI in dead trees provides further support to this interpretation and supports H2. That is, the tree-ring δ¹³C variations were strongly negatively correlated to January–September SPEI in dead edge trees but to a lesser extent in live edge trees. Combined with the higher δ¹³C values at the forest edge this suggests particularly reduced ¹³C discrimination in dry years. Again, this can be interpreted as a faster drying of the forest-edge microclimate in dry years, causing an earlier reduction of stomatal conductance and therefore lower discrimination against ¹³C. The observed much stronger link between δ¹³C and water availability in dead edge trees (Spearman’s rho = 0.72) compared to live edge trees (rho = 0.38) possibly indicates that these trees were more constrained by the dry forest-edge microclimate.

Possible factors causing a drier micro-climate for dead trees in comparison to live trees at the forest edge refer to differences in soil properties, leaf-area index, and tree exposure to macro-climate. Previous studies have shown that plant available water capacity modulates the coupling of tree growth with soil-water availability (e.g., Lévesque et al., 2013; Rehschuh et al., 2017). Theoretically, small-scale variations of soil texture – governing plant available water capacity – may to some degree explain the observed individual differences between the investigated trees concerning detrimental effects that likely were caused by a higher exposure to drought. Moreover, belowground competition for water resources may cause trees’ differing drought-sensitivity. That is, trees suffering a stronger belowground competition, e.g., by the roots of understory shrubs or neighboring trees are more likely to experience water shortage in dry years (Giuggiola et al., 2016, 2018). Besides soil conditions, exposure of individual trees to the macro-climate can also lead to a higher drought-induced mortality risk. During drought, trees whose crowns are more exposed to the macro-climate suffer from higher canopy temperatures, higher turbulence, and in combination a higher vapor pressure deficit (Grote et al., 2016). Another important factor is leaf-area index, which tends to be larger for exposed trees, since lower branches receive more sunlight and thus retain a higher amount of foliage (Sherich et al., 2007). If assuming similar stomatal conductance under well-watered conditions early in the growing season, water-loss via transpiration is higher in exposed trees, resulting in earlier drought stress compared to less exposed individuals. The observation of significantly lower glk and inter-series correlation of forest-edge trees compared to the forest interior indicates more heterogeneous tree-responses to environmental conditions at the forest edge. Unfortunately, we neither recorded understory vegetation nor did we analyze soil samples or tree exposure to macro-climate to investigate these possible explanations, which is why these hypothetical explanations should be taken into consideration in future investigations. We can also not exclude genetic differences among live and dead individuals, which are also known to potentially affect mortality patterns (Hentschel et al., 2014).

The observed lower cell-wall thickness and consequently mean ring density of edge trees supports H3. This finding likely indicates a lower availability of photo-assimilates for xylem formation during the period of investigation (i.e., at least since 1998) and is in line with the interpretation of photosynthetic downregulation in response to the reduction of stomatal conductance during drought as a dominant physiological agent of the observed forest-edge effects (H1 and H2). Our interpretation is in line with a study from Central Switzerland, in which the authors reported a temporarily reduced earlywood cell-wall thickness for Scots pine individuals that had experienced an abrupt reduction in water availability due to the drainage of an adjacent water-channel (Eilmann et al., 2009). In their study, the authors also interpreted reduced cell-wall thickness as a reduced carbon investment under drought. Further support to this interpretation comes from a drought-related study on Pinus nigra in Northern Italy (Petrucco et al., 2017), a drought experiment on Pinus edulis in New Mexico (Guérin et al., 2020), and a study focusing on drought-induced mortality of Pinus canariensis (López et al., 2021). It is important to stress, that trees growing under relatively drier conditions typically adjust to lower stem-water potentials by growing stronger tracheids with thicker cell walls (Hacke and Sperry, 2001) which is in contrast to our finding. We see this discrepancy between our findings and existing theory as further support to our interpretation, namely that the observed thinner cell walls at the forest edge reflect a drought-induced lower availability of photo-assimilates as also reported in other studies dealing with drought effects in pine species (Eilmann et al., 2009; Guérin et al., 2020; López et al., 2021).

Under this framework, the fact that earlywood cell-wall thickness was systematically and significantly lower for dead compared to live edge trees may further support the micro-environment hypothesis formulated in context of δ¹³C and H2. That is, a soil texture causing a lower water-holding capacity and increased water competition, of dead trees may also result in a reduced cell-wall thickness compared to neighboring trees, given a reduced production of photo-assimilates. Yet, this hypothetical interpretation needs further support by additional investigations based on a higher replication of dead and live individuals and a minute investigation of individual trees’ macro-climatic exposure and soil conditions.

Finally, our results support H4. That is, edge trees and in particular dead trees featured a systematically and significantly higher earlywood mean hydraulic diameter. In combination with the observed lower earlywood cell-wall thickness, this results in a lower cell-wall reinforcement and thus hydraulic safety in edge vs. interior trees. Again, this finding is in line with Eilmann et al. (2009) who reported a significantly higher lumen diameter (and thus mean hydraulic diameter) of Scots pine after the cessation of irrigation from which they concluded a higher implosion risk. Similar observations were made for Pinus edulis, which featured a lower tracheid-implosion safety in a drought experiment (Guérin et al., 2020). The observed larger mean hydraulic diameter may be interpreted as a morphological adaption to resource scarcity under drought-prone conditions at the forest edge. That is, comparably less photo-assimilates are invested to maintain hydraulic conductivity high enough to supply crowns with water rather than spending more photo-assimilates to build more cells with a higher cell-implosion safety (Sperry, 2003; Prendin et al., 2018; Guérin et al., 2020). The observation, that differences between the investigated tree categories was most prominent in the latewood supports this interpretation. Namely, latewood is typically formed in summer and thus at a time when photo-assimilation frequently is reduced due to drought-stress due to a high atmospheric water demand. As a consequence, edge trees which may experience an amplified summer drought may feature thinner cell walls and larger lumen diameters to favor water transport. In addition, larger canopy volumes resulting from higher light availability and reduced competition at the forest edge might need to be supplied with relatively more water than the canopy of interior trees (Sherich et al., 2007), potentially amplifying this hydraulic adjustment, which however, comes at the cost of an increased risk of implosion (Hacke et al., 2001). In fact, Hacke et al. (2001) have shown a clear relationship between cell-wall reinforcement and P₅₀, i.e., the xylem water potential at which 50% loss of conductivity occurs due to cavitation. Thus, with a lower cell-wall reinforcement, a less negative water potential is needed for 50% cavitation. Given this negative relationship between cell-wall reinforcement and cavitation risk, we interpret the significantly lower cell-wall reinforcement at the forest edge as a higher probability of cavitation, which under an increasingly drier climate may result in vitality decline and tree dieback.

The observed larger hydraulic diameters at the forest edge may theoretically also result from larger tree heights. At breast height taller trees typically feature larger hydraulic diameters compared to smaller trees (Anfodillo et al., 2006), which also translates into time series (Carrer et al., 2015). However, as shown in Buras et al. (2018) forest edge trees were in fact significantly shorter compared to interior trees (see Figure 2F in Buras et al., 2018), possibly due to a lower competition for light at the forest edge and thus a lower necessity to outcompete neighboring trees via apical growth or due to mechanical acclimation to a higher wind exposure at the forest edge (Chehab et al., 2009; Moulia et al., 2015). Given the positive link between hydraulic diameter and tree height, the finding of higher hydraulic diameters at the forest edge – where trees were shorter – even more points toward a hydraulic adjustment due to differing micro-climatic conditions.

Besides the comparison among groups over the whole investigated period, we want to stress the abrupt decline of several anatomical parameters for dying trees since the year 2012. In particular, latewood cell-wall thickness, mean ring density, earlywood hydraulic diameter and cell-wall reinforcement of dying trees abruptly declined in 2012 (Figures 3A,C,D,E), whereas earlywood cell-wall thickness declined in 2013. At the study site, the year of 2012 was characterized by a moderate drought (January–September SPEI = −1.3, see Supplementary Figure 4) which ranks eleventh among all local droughts since 1901. Moreover 2012 was preceded by the relatively dry year of 2011 (January–September SPEI = −1.1, see Supplementary Figure 4), potentially amplifying soil drought in case the soil water storage was not replenished in the winter 2011/2012. The comparably strong drought effects of 2012 are likely also reflected in the comparably high δ¹³C values for all trees (Figure 2A). In context of how droughts evolve in Central Europe (e.g., Buras et al., 2020), the observed lag of earlywood cell wall thickness decline seems plausible. Because drought usually develops during the growing season and peaks in July–August, a time when earlywood formation is largely completed, the effects of drought on reduced photo-assimilate availability in 2012 did not strongly affect earlywood cell wall thickness. In contrast, the observed decline in earlywood hydraulic diameter already in 2012 may be considered an anatomical adjustment to the moderately dry conditions resulting from 2011. That is, low soil water potentials in spring 2012 hampered turgor-driven tracheid enlargement as has been shown for other sites before (Cabon et al., 2020). This possibly also explains the abrupt decline of earlywood hydraulic diameter of all edge trees in 2014 which was characterized by dry conditions (January–September SPEI = −1.1, see Supplementary Figure 4) as is reflected in higher δ¹³C values in all trees. Therefore, the large decrease in earlywood hydraulic diameter for all edge trees in 2014 could be due to accumulated drought effects related to the plant available soil water over the period 2011–2014. It is important to stress, that this interpretation does not contradict our interpretation of an anatomical adjustment of the xylem to dry conditions on the longer term as formulated in section 4.2 and context of H4. That is, on the one hand the frequent, drought-induced reduction of stomatal conductance probably caused a shortage in photo-assimilates which in turn led to a more cost-efficient wood anatomy featuring thinner cell walls (reduced carbon investment) with wider lumina (more efficient sap-flow). On the other hand, overly dry conditions early in the growing season of 2012 and 2014 caused by soil-moisture related carry-over effects from the, respectively, previous year hampered cell enlargement due to a relatively low cell turgor. In conclusion, the observations of abruptly declining wood anatomical traits point toward potential early-warning indicators of tree dieback (Pellizzari et al., 2016). Moreover, our results suggest that carbon depletion goes hand in hand with an increase in hydraulic vulnerability, a pattern often seen in drought-induced tree mortality (Adams et al., 2017).

It is important to stress, that we cannot rule out effects of forest management on our results due to a lack of information on local management history. In fact, BAI time-series suggest a thinning-induced growth release of the forest interior around the year 2006 (Figure 1B). As a consequence, competition for water and light would be reduced. Yet, a possible thinning would likely also reduce competition of forest-edge trees, which however showed no growth release in that period, possibly due to detrimental drought effects counteracting a potential thinning effect. Independent of possible management activities around 2006, the time series under investigation showed systematic differences between edge and interior trees before 2006 and featured no abrupt shift of the investigated parameters around 2006. Thus, even if management had taken place, we believe related effects on the studied parameters to be statistically negligible. Yet, the weak negative trend in δ¹³C observed for interior trees (Figure 2A) might result from a slightly improved plant-water availability after 2006 due to a lower stand density.