Introduction

Pedunculate oak (Quercus robur L.) is a broadly distributed tree species in the European deciduous temperate forests (Čehulić et al. 2019). In general, this tree species can thrive under very heterogeneous soil conditions (Vastag et al. 2020): both high and low water supplies are tolerated due to its specific isohydric strategy (i.e., downregulation of sap flow to avoid water loss during high transpiration rates) and a deep-penetrating root system (Bose et al. 2021; Thomsen et al. 2020). Nevertheless, pedunculate oak is mostly known as a flood-tolerant tree, it thrives on fertile and moist soil conditions, and constitutes a late successional species in valleys and floodplains (Ducousso and Bordacs 2004). Concerning the wood anatomical classification, pedunculate oak forms ring-porous wood: within each tree ring, large earlywood vessels are followed by remarkably smaller latewood vessels, organized in a flame-like pattern (Bräuning et al. 2016). As a riparian species, it can cope with regular flooding events, surviving the resulting anoxic soil conditions. When these events occur, they may affect the typical wood anatomical features, forming what is known as a “flood ring” (Copini et al. 2016). In flood rings, the type and the degree of vessel alteration depends on the timing of the flooding. In general, the most frequently observed effect is the reduction in size of the earlywood vessels after a spring flooding (St. George and Nielsen 2003). While several studies have been conducted highlighting the short-term effect of flooding events at different timings (Yanosky 1983; Astrade and Bégin 1997; St. George and Nielsen 2003; Ballesteros-Cánovas et al. 2015; Tumajer und Treml 2016), the effects of long-term rewetting on pedunculate oak have not yet been systematically explored.

In view of climate change predictions, the importance of peatland ecosystems has gained significant recognition. Peatlands are not only vital reservoirs of biodiversity but also play a crucial role in regulating soil water content, and serve as established carbon sinks (Joosten et al. 2017). However, due to human activities such as drainage for agricultural purposes, these important ecosystems have been significantly degraded. To counter the adverse effects of peatland drainage, rewetting policies have been implemented with the aim of restoring the fundamental functions of these ecosystems (Liu und Lennartz 2019). The restoration of peatlands aims to recreate suitable conditions for diverse plant and animal species to thrive, and mitigate the release of stored carbon by promoting the reestablishment of peat accumulation, thereby ultimately transforming drained peatlands back into effective carbon sinks. While the concept of rewetting shows promise for peatland restoration, uncertainties remain regarding the extent and time frame required to fully recover the original status of these ecosystems (Kreyling et al. 2021). Factors such as the initial degradation level, the specific characteristics of each peatland, and the success of vegetation recolonization can influence the effectiveness and duration of the rewetting process. The success of restoration efforts in peatland forests relies significantly on the tree species’ ability to acclimate to elevated water tables (Anadon-Rosell et al. 2022).

In this paper, we investigated wood anatomical changes in surviving pedunculate oaks after the flooding and the consequent permanent rewetting of a formerly drained peatland: the Anklamer Stadtbruch. This area, located in NE Germany facing the Szczecin lagoon, is characterized by a peatland complex that was drained until November 1995, when a severe storm broke the dyke that separated the site from the lagoon. A rewetting process was initiated from that moment on, with a further relevant intervention in autumn 2004, when an artificial canal was opened and the pum** station in place was removed. This strategy was implemented to promote peatland restoration through a better connection between the lagoon and the peatland, and to counter the negative effect of the summer drought occurred in 2003. Consequently, first the water table raised to the ground level right after the flooding (1995), and second, records obtained close to the sampling area reported the water level to lay permanently above the ground level after the artificial canal opening (2004). Since the rewetting took place in two different phases, we investigated the response in three separate time spans: before the dyke breakage (pre), after dyke breakage (post-1), and after the canal opening (post-2).

In particular, we elaborated on the following questions:

  • Q.I—Analysis of the mean trend at stand level: How do the three different hydrologic regimes influence the wood anatomical traits over time?

  • Q.II—Analysis of the variability at stand level: How does trait distribution vary in response to changes in the hydrologic regime?

  • Q.III—Analysis of the individual tree variability: How do single trees react to the three different hydrologic regimes? Do traits show a synchronized response among trees?

  • Q.IV—Did “flood ring” characteristics emerge in reaction to the flooding event?

These questions are addressed by comparing anatomical and hydraulic data from pedunculate oak trees growing on drained and rewetted peatlands. Our findings support predictions concerning changes in vessel diameter attributed to the vessel widening mechanism. We discuss trees’ adaptability to shifting hydrological conditions and the sensitivity of oak growth to local site conditions.

Material and methods

Site description

The study area is located in “Anklamer Stadtbruch”, a large peatland formation of approximately 2000 ha, and situated 10 km east of the city of Anklam (NE Germany). Directly facing the Szczecin lagoon, this area is located at the edge of the lower Peene river valley, and was originally constituted by a complex of three different mire types: a central bog, fed only by rainwater; a percolation mire, fed by percolating groundwater from the moraines; and an outer coastal transgression mire, periodically flooded by the Baltic Sea (Timmermann et al. 2008).

The Swedish cadaster in 1695 reported the area as forest pasture with loose oak stands. For centuries, the land use was oriented to pasture, timber extraction, and peat cutting. Intensive drainage, peat extraction interventions, and agricultural use transformed this environment leading to severe peatland degradation and a large-scale subsidence of usable land below sea level. At the edges of the peatland, oak can still be found today due to the historical management of the area for wood provision. To allow for a more intensive use of the meadows as pasture and grassland, a dyke (yellow line in Fig. 1) was built in 1932/1933 to prevent flooding events (Schulz 2005). Furthermore, as part of the complex amelioration interventions, in the 70 s, the drainage of the area was enhanced to reduce the effect of flooding and to increase grassland yields (Gremer et al. 2000). Drainage measures were in place until November 1995, when a severe storm damaged the dyke and caused the inundation of the whole area. Since the intended use was no longer viable, the area was left to natural water dynamics to allow the regeneration of the existing peatland through rewetting. After the rewetting process started in 1996, an intervention was successively performed in autumn 2004 with the opening of a 10 m-wide artificial canal that regulates the inflow of the water from the Szczecin lagoon, together with the removal of the pum** station in place (Schulz 2005). The aim was to restore high water table levels after the hot summer of 2003 (pers. Comm. From Kai Pauling). Another major event experienced by this stand is the drought occurred in 2018, when especially the north-eastern part of Europe underwent a strong summer heatwave (Salomón et al. 2022).

Fig. 1
figure 1

Map of the sampling site in the Anklamer Stadtbruch area, NE of Germany. Green dot in the upper right square is where the sampled trees are located. Dotted lines identify the flooding basins. The basins are connected via a net of channels that regulate the water influx. For this reason, basins are highlighted with different colors according to the sequence in which they receive water during flooding (from dark blue the first to receive water, to white the last and the least amount of water received). The yellow solid line identifies the old dyke, while the red solid line identifies the new dyke. Map retrieved and adapted from Timmermann et al. (2008)

Full size image

Although we did not have direct measurements of the water table level in the exact area where the trees were sampled, we were able to collect several sources to explain the site hydrologic dynamics. An excerpt from the Ornithological Newsletter for Mecklenburg-Western Pomerania (Eichstädt and Eichstädt 2015) reported the water level measurements from the Rosenhagen dam. This second dyke was built in 1999 after the first dyke collapsed to protect the inhabited area of Rosenhagen, and it is located at the margin of the Anklamer Stadtbruch, opposite to the lagoon (Fig. 1). According to the data collected, the intervention on the artificial canal was successful, since from 2004 onwards, data exhibit a switch of the water level toward positive values, with relatively high values between 2009 and 2015 (Online Resource 1, panel b).

Another indicator of the water level in the area was derived through the tasseled cap wetness index (TCW), which besides other parameters such as land cover change, vegetation structure, and forest disturbances (** and Sader 2005), showed to be sensitive to soil and plant moisture (Crist and Cicone 1984). The data related to the flooded area in Anklamer Stadtbruch were obtained via the Google Earth engine, and revealed that the soil moisture substantially increased after 1996, reaching values permanently above zero from 1998 onwards (Online Resource 1, panel c). TCW values higher than zero have been associated with water bodies (Li et al 2016). In addition, the flooding event in 1995 is supported by the water level data of the Szczecin lagoon, which in that year reports the highest water level. These data were obtained from the measuring station located in Karnin (Online Resource 1, panel a). The extensive flooding in 1995, the permanently flooded condition starting from 1998, and the pump removal to open the artificial canal in 2004 are known events reported in the list of relevant episodes annotated in (Schulz 2005), for which a translation was provided in the supplementary material (Online Resource 2).

In light of the two substantial changes in the hydrological regime, we could summarize the site conditions in three different time periods: before the dyke damage (pre: 1977–1995) characterized by controlled soil conditions; after the dyke damage (post-1: 1996–2003) when the flooding occurred and the soil moisture variability was high; and after the opening of the artificial canal and the water pump demolition (post-2: 2004–2019) when the water table level was recorded to be permanently above zero even in the inner side of the Anklamer Stadtbruch area, where the new dyke was built.

For our investigation, we selected an oak stand (Quercus robur (L.)) of circa 3 ha (Fig. 1), object of a previous research (Scharnweber et al. 2014). As described there, the stand was subjected to a high coppice management until it was abandoned in 1996. Beside pedunculate oak, the main species present are Alnus glutinosa (L.), Betula pubescens (Ehrh.), and some Pinus sylvestris (L.), while the understory is characterized by Frangula alnus (L.). A large part of the area has been affected by intense tree dieback after the flooding event and the onset of the rewetting dynamic. Pictures of the area retrieved by different sources are visible in the supplementary material (Online Resource 8). At the time of sampling, although trees are largely dead, some oak specimens survived and were able to cope with the hydrological changes in the area. The trees selected for this study are located in the basin that is characterized by the highest relative elevation, 10–20 cm higher than the other basins in the area (Timmermann et al. 2008), and according to the channel distribution, the last to receive water when flooding occur (Fig. 1). Moreover, single trees’ elevational difference ranges between 0 and 20 cm.

The climate characterizing the region is temperate. Precipitation is evenly distributed over the year. Mean annual temperature and precipitation sums for the time span covered by the wood anatomical traits time series (1980–2019) are, respectively, 9.2 °C and 577 mm (National Meteorological Service of Germany DWD, 2023).

Sample collection and processing

Between Winter 2019/2020 and Spring 2020, we sampled 12 trees, collecting 2 cores per tree at breast height with a 5 mm increment borer. The trees chosen showed a fully vital crown, indicating that they coped well with the rewetting dynamics over time. For every tree, GPS position, diameter at breast height (DBH), and height measurements were collected. These data are available at Online Resource 3, a table reporting the information collected in the field.

One core per tree was used to produce tree ring width time series. In the laboratory, cores were mounted on a wooden support, sanded to highlight tree ring visibility, and then scanned with an A3 optical flat-bed scanner with a resolution of 2400 dpi. Tree ring width was then measured with Coorecorder, and crossdated with cDendro 9.3 version (Cybis Electronic 2013). Tree age of the 12 scanned cores is reported in the field data table (Online Resource 3). The second core was used for wood anatomical analyses. For this, cores were cut into 3–4 cm pieces and processed with a Leica RM 2245 rotary microtome to produce thin sections of 12–14 µm thickness. These were then stained with a 1:1 Safranin and Astra Blue solution to highlight the lignin and cellulose content, and to improve contrast. Thin sections were eventually mounted on a glass slide and sealed with Euparal mounting media and dried on drying benches at 65–70 °C for 48 h.

Images of the slides were obtained through a Zeiss Axio Scan.Z1 slide scanner (Carl Zeiss AG, Germany) with a 20× lens and a resolution of 2.267574 pixel/µm. Image analysis was carried out via CARROT (Cell And Ring RecOgnition Tool), an open source software suited for wood anatomical research, available at https://github.com/alexander-g/CARROT (release nr 2). CARROT is an evolution of the tool presented in Resente et al. (2021), and it works on both the detection and the computation of wood anatomical features. The models operating the recognition of such features employ Mask R-CNN architectures, taking advantage of artificial intelligence, and specifically deep learning efficiency, to perform the task. These models have been trained on conifers and three angiosperm species, namely Quercus robur, Fagus sylvatica, and Alnus glutinosa, to incorporate the three categories of vessels distribution (ring porous, semi-diffuse porous, and diffuse porous). The features that are computed by the software are tree ring width, tree ring area, and tree lumen area of vessels and tracheids. In the context of pedunculate oak anatomy, the advance provided by the employment of deep learning allowed us to train the model on the large earlywood vessels as well as on the smaller latewood vessels. Therefore, the oak model is able to distinguish vessels from the other cell types that constitute pedunculate oak anatomy with a fairly high rate of success (97%) (Resente et al. 2021).

Wood anatomical traits chronologies

For each image, CARROT worked on earlywood and latewood areas separately, to produce distinguished chronologies. The software provides images where the detected vessels are highlighted in different colors according to the ring, and it provides a datasheet reporting the coordinates and the lumen area of these vessels. By means of these two outputs, it was possible to visually establish the overall performance of the tool, and to filter out suspected outliers. As suggested by Kniesel et al. (2015), we analyzed the vessels that carried the strongest common signal between the cores. For this operation, we used the EPS (expressed population signal), that allowed us to select the quantile of the vessel distribution that was the most representative of the whole chronology, respectively, 0.87 for earlywood and 0.77 for latewood. With this operation, we extracted information for the most affected vessels from the environmental changes in the water regime and partially overcame individual variability (Rita et al. 2022).

From the raw lumen area of the vessels, we derived the following parameters: theoretical hydraulic diameter (Dh, Eq. 1) calculated on the quantile previously selected, and specific hydraulic conductivity (Ks, Eq. 2).

$$Dh={\left(\frac{\sum {D}^{4}}{N}\right)}^\frac{1}{4}$$
(1)
$$Ks=\frac{\pi \rho }{128\eta A}\sum_{1}^{n}{d}_{t}^{4}$$
(2)

Dh is a trait related to the conductivity weighted on all the vessels in a specific ring, that assigns the correct importance considering the size through the Hagen–Poiseuille law. Tree Ks, instead, reflects conductance efficiency within each tree ring and weights vessel lumen area according to the tree ring area. For this reason, we distinguished earlywood and latewood area and computed Ks separately. Moreover, for both earlywood and latewood, we calculated width (EW-W, LW-W), in addition to vessel density (EW-VD, LW-VD) and vessel number (EW-VN, LW-VN).

We accounted for the vessel widening effect from the pith to the bark due to increasing tree height and crown size (Lechthaler et al. 2019) and detrended all the calculated traits with a cubic smoothing spline function with a 50% frequency and a cutoff of 30 years. For this operation, we used the detrend function of the dplR Package (Bunn 2008) that calculates the ratio between the observed and fitted values for each annual tree ring to obtain detrended series (Cook and Kairiukstis, 1990). The function then averages the detrended series employing a bi-weight robust mean to build the mean anatomical chronologies.

Statistical analyses

The analysis on the trait chronologies was performed according to the three periods previously mentioned (pre, post-1, post-2). Chronologies start in 1977 where the sample depth equals to half of the total number of samples we collected. This threshold was established on the base of the quality of the slides that were obtained from the cores, to ensure that for the years included, quantitative wood anatomical parameters were also available. Tree ring width chronologies built from the scans, instead, were used to crossdate the tree ring width obtained from the slides, and range over the time period 1948–2019 (Online Resource 4). To ensure clarity in the analyses, and following the classification suggested by Anadon-Rosell (2022), we distinguished two categories of traits—growth anatomical traits, which are represented by tree ring width (TRW), width of the earlywood and the latewood (EW-W and LW-W, respectively), and vessel number (VN); and hydraulic traits, which include vessel density (VD), theoretical hydraulic diameter (Dh), and specific hydraulic conductivity (Ks). Except TRW, all the traits have been calculated separately for EW and LW.

To investigate the general long-term trend in the traits and understand how they reacted to the rewetting process, we employed a piecewise linear mixed effects model, as implemented in the package lme4 (Bates et al. 2015). We fitted the model on the raw data of pre, post-1, and post-2, with the following formula:

lmer(single anatomical data ~ year * timeframe + (1 | individual tree)

where “single anatomical data” stands for one data point per year per tree belonging to the respective trait, with a random effect on the intercept for each individual tree. The interaction term with the dummy variable “timeframe” allows to fit a different linear regressions for each time frame, with a single call to the lmer() function. To confirm the changes in trend among the three time frames, we tested for the significance of the slope differences with the emtrends function from the emmeans package (Lenth 2023). P values were adjusted for multiple testing via Benjamini–Yekutieli method.

In addition, considering the relevance of the data spread over the fitted models, we explored the variability of the wood anatomical traits over time on a population level. We applied an ANOVA on the residuals from the piecewise model excluding the random effects (which is equivalent to the Levene’s test for homogeneity of variances), and tested for differences among timeframes (Warnes et al. 2022) with a Tukey HSD post hoc test. A further analysis tested for the variability within the trait time series. We investigated the coherence among single trees for each trait with the running mean inter-series correlation index (running rbar) from the dplR Package (Bunn 2008). The “rwi.stats.running” function was applied with a window length of 11 years, and a window overlap of 10, to have a 1-year shift between each value and enough years to build a reliable rbar value. Rbar was calculated on detrended data.

Finally, we used the pointer years analysis on the detrended data, to understand the short-term effect of the two events on each wood anatomical trait. The methodology employed was the extreme values of chronology method (zChron) from the pointRes package (van der Maaten-Theunissen et al. 2015). All analyses were performed in R version 4.3.0 (R Core Team 2023). To compensate for the low sample replication, pointer years analysis with the same parameter settings was performed on the pedunculate oak time series analyzed in Scharnweber et al. (2014). The study was carried out in the same area and aimed at understanding the ecological differences in dead, damaged, and healthy trees consequent to the dyke breakage. Each category was represented, respectively, by 47, 40, and 46 samples. Pointer years analysis was performed on the full time range of the three groups (1908–2010, 1908–2013, 1920–2013), and the results were used as a reference to validate the pointer years found in the overlap** period between the two studies.

Climate correlation analysis was performed on all the wood anatomical and hydraulic traits, to test if the water table level of the lagoon, precipitation, or temperature had an effect in sha** the traits response. We ran the analysis on the three periods (pre, post-1, post-2) employing the dcc function from the dplR Package (Bunn 2008). Climate data were retrieved form the DWD (National Meteorological Service of Germany) that holds records of the local climate station in Anklam. Results of the climate correlations can be found in the supplementary materials (Online Resource 8).

Results

The thorough identification of vessels was possible, thanks to CARROT and its specific outputs (Resente et al. 2021). The result of the visual assessment performed on the images reported that only 0.22% of the vessels were incorrectly identified in the earlywood area, being too small to be identified as vessel cells, while 3.86% of the vessels were removed from the latewood. The error in this second category was mainly generated by parenchyma cells which in shape, size, and appearance look very similar to latewood vessels. Overall, the percentage of correctly identified vessels was 95.92%.

Long-term effects of the hydraulic regime shifts on the stand

Analysis of the mean trend of the stand: piecewise model analysis on the raw data

Growth anatomical traits showed one common slope pattern. Specifically, TRW, EW-W, and EW-VN showed a significant increasing trend after the dyke broke (P < 0.05 for TRW, and p < 0.001 for EW-W and EW-VN), and again significant between before and after the opening of the artificial canal, with the same P values as the previous comparison (Fig. 2). Concerning EW-W and LW-W, the slopes of the fitted models in the different time frames are similar to TRW, but the data distribution within the periods led to certain differences in the significance (Fig. 2). For LW-W, no comparison appeared to be significant. Regarding LW-VN, post-2 differed significantly from post-1 (P < 0.05) showing that while at the beginning, VN was increasing (post-1), afterward the trait slightly decreased (post-2) (Fig. 2).

Fig. 2
figure 2

First and third row: piecewise linear model applied to the three time frames of the growth anatomical traits, width (W), vessel number (VN) for earlywood (EW) and latewood (LW), and tree ring width (TRW). Only fixed effects are shown. Second and fourth row: variance analysis of the residuals calculated from the model (excluding random effects) in the three time frames of the corresponding traits. Grey dots correspond to data points, asterisks over brackets indicate the statistical significance of the t test for equality of the slopes. (N.s. not significant, *P < 0.05, **P < 0.01, ***P < 0.001)

Full size image

Within the hydraulic traits, two different patterns emerged. The first was shown by VD, and it differed according to the ring area investigated (Fig. 3). EW-VD decreased over the whole time frame, with a particularly negative trend in post-1 with respect to post-2 (significance P < 0.05). The slope of the pre vs post-2 period is also significant (P < 0.01). On the other hand, LW-VD slightly increased in pre, settled over the post-1 period, and slightly decreased again in post-2, with the only significant comparison being pre vs post-2 (P < 0.01).

Fig. 3
figure 3

First and third row: piecewise linear model applied to the three time frames of the hydraulic anatomical traits, vessel density (VD), hydraulic diameter (Dh), and specific conductivity (Ks), for earlywood (EW) and latewood (LW). Only fixed effects are shown. Second and fourth row: variance analysis of the residuals calculated from the model (excluding random effects) calculated from the model in the three time frames of the corresponding traits. Grey dots correspond to data points, asterisks over brackets indicate the statistical significance of the t test for equality of the slopes. (N.s. not significant, *P < 0.05, **P < 0.01, ***P < 0.001)

Full size image

Dh and Ks showed a similar trend, and within both traits, EW and LW were also similar (Fig. 3). Specifically, Dh mean trend was not affected by the dyke breakage neither in the EW nor in the LW, and displayed an increasing trend until post-2 started. In post-2, the slope showed a strong negative behavior, with an abrupt decrease of the Dh for both EW and LW. Similarly, Ks slope slowly increased up to the dyke breakage, settled with a slight decreasing trend in post-1, and abruptly decreased in post-2, with the highest significant comparison being pre vs post-2 (P < 0.001).

Analysis of the variability of the stand: variance analysis on the residuals of the piecewise model

In the previous analysis, we found that the growth anatomical traits (Fig. 2) were the most similar ones in terms of pattern of the fitted model. Generally, the non-significant slope comparisons resulted in significant differences in variance between the time periods.

Specifically, the variance of TRW data before the dyke broke (pre) was very low, but since the dyke broke and the start of the rewetting process, it became significantly higher than pre (P < 0.001), and no significant difference was shown between post-1 and post-2 (Fig. 2). EW-W followed the same pattern. Its variance abruptly increased after the dyke breakage, and difference between pre vs post-1 showed significant result (P < 0.01), and even more between pre vs post-2 (P < 0.001). Regarding LW-W, the variance value was the highest in post-2, and it resulted significantly different from post-1 (P < 0.01) and from pre (P < 0.001). Significant difference was also found between pre and post-1 (P < 0.001). The variability of VN was strongest for the EW. Here, in particular, variability is the highest in post-1 rather than post-2 with minimum non-significant difference between the two periods, but highly significant when these two periods are compared to pre (P < 0.001). LW-VN variability, on the other side, steadily increased with significant difference between pre and post-1 (P < 0.05) and higher significant difference between pre and post-2 (P < 0.001).

With regard to vessel density, the results for EW were the only ones showing significant comparisons among the time frames. In EW, the values decreased significantly from pre to post-1 (P < 0.001) and from pre to post-2 (P < 0.001) with no significant difference between post-1 and post-2 (Fig. 3).

Dh and Ks showed a different behavior than the other traits in the variance analysis (Fig. 3). EW-Dh variance was the lowest in post-1 with respect to pre (P < 0.001), and no difference was found between pre and post-2. As for LW-Dh, pre and post-1 showed even variability and no significant difference, while post-2 differed from post-1 (P < 0.05), and from pre (P < 0.01). With respect to the Ks traits, we found no significant variance comparisons between the time frames in EW and LW.

Analysis of the variability of the individual trees: running rbar

Within the variability analysis, we tested if single trees reactions were synchronous for each trait (Online Resource 5). TRW is the trait that showed the highest coherence among individual time series, this can be appreciated both in the raw time series (Fig. 4a), as much as in the detrended time series (Fig. 4b). The rbar calculated for this trait is represented by the black curve (Fig. 4b) plotted against the detrended ring width time series, which reports the total rbar value for the end year of a time window of 11 years. For this reason, the lowest value we observe in year 2007 is the result of the total rbar calculation from the 1997–2007 period. The highest rbar value is reached in 2018 (0.75) which incorporates the time series behavior of 2008–2018. Rbar analysis also underlined how, over the last ~ 15 years, the common signal between the individuals was much higher than in the period before (Fig. 4b).

Fig. 4
figure 4

TRW raw time series (a) compared to TRW detrended time series (b). Running mean inter-series correlation index (r bar) analysis on the detrended tree ring chronology (b), calculated over a time window of 11 year and a shift of 10 years. Total rbar values are represented by the black curve (left axis). Single raw time series appear in pale colors (right axis). Grey vertical lines to highlight 1996 (flooding event) and 2004 (opening of the artificial canal and removal of the pum** station in autumn 2004)

Full size image

Short- and long-term effect: is 1996 a “flood ring”?

The comparison carried out to validate pointer years on TRW time series revealed that the events identified in the current dataset can be considered reliable despite the low sample size. These events showed high correspondence in nature, number, and significance to the pointer years identified in the broader time series of the three groups of samples—dead, healthy, and damaged—retrieved from Scharnweber et al. (2014) (Online Resource 6). The similarity is particularly high with the damaged and the healthy pointer years chronologies.

In relation to the current study, the pointer years analysis revealed that trees were strongly affected by the flooding event in November 1995, showing a clear response in the following year (Fig. 5). 1996 resulted in a significant negative pointer year, respectively, for TRW, EW-W, EW-Dh, LW-W, LW-VN. All the parameters recorded negative pointer years with different intensities. EW-VD and EW-VN also showed a high Z-score for 1996, but they did not prove to be significant.

Fig. 5
figure 5

Heatmap of the pointer years detected on the detrended time series of all the investigated traits. The methodology employs the extreme values of chronology method (zChron). Stars indicate statistically significant z values. Red marked years correspond to 1996, year following the flooding event; 2004, year of the artificial canal opening; and 2018, year of extreme drought

Full size image

The second intervention in the hydrological regime (artificial canal 2004) led to wood anatomical reactions at different levels, albeit not directly affecting the current year. 2004 does not correspond to any pointer year in the trait chronologies. However, the group of healthy trees sampled in the previous study, characterized by a bigger sample size, recorded 2004 as a significant positive pointer year. What is revealed by the heatmap of the current study is that the number of pointer years in post-2 increased over time (Fig. 5). Within this period, while in the earlywood, hydraulic traits were the most affected ones; in the latewood, traits seemed to present more pointer years and of higher intensity on both growth anatomical and hydraulic traits. 2009, 2012, 2015, 2018 pointer years consistency over the traits appear to be relevant events, despite the uncertainties of the ecological explanations.

In addition, this analysis allowed to appreciate the effect of the flooding on the traits’ anatomy compared to the effect of the severe drought that occurred in 2018, which resulted in negative pointer years for Dh, W, and VN, and a positive deviation for VD (in both EW and LW) (Fig. 5). In both cases (1996 and 2018), the pointer years of different traits, regardless of the significance, showed the same nature. Among the significant pointer year values, the traits that reacted most similarly are TRW, EW-Dh, LW-W, and LW-VN. Another significant year in the pointer years chronology is 2009, which is an extremely positive event, mainly for LW values (Dh and LW-W) and TRW. In this year, the data retrieved from the Polder Rosenhagen Dam report the highest water level registered in that time series.

Discussion

Analysis of the mean trend of the stand

Among the investigated traits, we identified three different response patterns which can be summarized by the following groups: growth anatomical traits (TRW, EW-W, LW-W, EW-VN, and LW-VN), and two groups belonging to the hydraulic traits, one corresponded to VD, and the other to Dh and Ks. Surprisingly, despite our assumption that the flooding event would have had a large effect on the population mean trend of the traits analyzed (Q.I), our results revealed that only the growth anatomical traits were significantly affected, and among these only TRW, EW-W, and EW-VN. The common response of these traits after the dyke broke was a general sudden increase in post-1, especially true for the EW traits. Subsequently, the opening of the artificial canal in 2004 initiated what was considered the second part of the rewetting process (post-2). Despite the local records covering only a short time span, it is possible to see that this distinct event did not produce an immediate effect, but allowed the water table level to rise in those areas that were not directly influenced by the flooding, resulting in a permanent shift to positive values after 2004 (Online Resource 1, panel b). Conceivably, the last active intervention in 2004 did not add more water to the already flooded areas, as TCW confirms that some stands were already permanently flooded since 1998 (Online Resource 1, panel C). This event rather resulted in a higher water level in those area that received water only from the channel system (visible in Fig. 1), due to the different elevation of the stands and the heterogeneous soil conditions. It is likely that the water level measured in the second dyke (Rosenhagen dam), which is located in the inner side of the area and far from the lagoon, stood permanently above zero only after 2004 in view of the abovementioned reasons (Online Resource 1, panel B).

The analysis on the mean trend of the hydraulic diameter (Q.I) highlighted the sudden reduction displayed by post-2, that was possibly the consequence of a hydraulic adaptation. This phenomenon—called vessel widening—links the vessel size to the length of the path from the root system to the top of the tree (Anfodillo et al., 2013), and it is considered an adaptive mechanism that allows to limit the increase in hydraulic resistance occurring with the vertical development of the plant (Lechtaler et al. 2019). As such, the rise of the water table level, which corresponded to a shorter distance for the water to reach the upper part of the crown, potentially resulted in the development of smaller vessels and a reduced hydraulic diameter. In addition, this interpretation is supported by the higher number of vessels and wider tree rings compared to the previous periods, which suggested that trees reacted to provide a hydraulic system potentially functional to maintain the photosynthetic activity of the crown. A possible connection to both anatomical responses (smaller Dh and increase in VN) are hormones such as auxin and ethylene, as their increase is known to produce similar results (Copini et al., 2016; Jughans et al., 2004). To generalize, the hydraulic and growth traits development adapted in response to the new particular site conditions, co** with the rewetting and the change in hydrologic regime. This response highlights how vessel widening is an adaptive mechanism that contributes to the hydraulic structure of the plants, as described in Olson et al. (2021).

In relation to the decline of the specific hydraulic conductivity, an impaired capacity of transporting water along the stem is closely linked to a decline in trees’ health (Tulik 2014), but in this case, it rather suggested an adaptive strategy to the new condition, since tree ring width in post-2 is overall higher and trees did not undergo flooding as an abrupt event. As a support to this, pedunculate oak seedlings were found to be particularly adapted to flood resistance forming adventitious roots and a high density of hypertrophied lenticels (Parelle et al. 2006), despite the roots dieback caused by the anoxic conditions.

Analysis of the variability of the stand

Concerning the hydraulic traits, beside the growth traits, our results show the importance of analyzing singular time series rather than mean trends, as shown by Rita et al. (2022). In fact, variance analysis, as a measure of the inter-individual variation within the population, is the distribution property that was affected the most by the flooding and the initial phase of the rewetting (Q.II). In this specific period, the variability of the growth anatomical traits significantly increased, linked to a stronger inter-individual variation, while the variability of earlywood hydraulic diameter experienced a significant reduction, showing remarkable inter-individual uniformity. A possible explanation for these opposite reactions could be found in the complex conditions of the area, which is characterized by different stands with differences in elevations (from − 25 cm to + 50 cm), and a convoluted network of channels (Fig. 1). Moreover, the waterlogging following the dyke break was an abrupt event that took place in an already disturbed ecosystem, which underwent several human interventions. All these co-occurring factors possibly allowed for the microsite effect to become more relevant and differently affect the growth of individual trees (e.g., TRW in Fig. 4a), especially in this first adaptation period. In fact, the importance of microsite effects for oak responses to flooding was already reported by Gricar et al. (2013). Nonetheless, since the whole area was affected by the flooding event, the hydraulic traits responded rather uniformly among the surviving trees, as shown by EW-Dh (Fig. 3). Possibly, this peculiar response in the hydraulic parameters was a consequences of the rise of the water table level that might trigger similar responses in the hormonal component of the trees. Despite the general need of the plants for more water consequent to the dry conditions characterizing the previous period, this dynamic could similarly affect the first rows of vessels produced during the growing season.

Analysis of the variability of the single trees

Raw TRW time series supported the previous assumption on the rise of the water table level and the definition of the three hydrologic regimes selected (Fig. 4a). An initial dry phase (pre) showed low variability and a low synchronization. This was followed by a transition phase (post-1) with very high variability, but low synchronization; and eventually, an adaptation phase (post-2) with raw TRW time series of individual trees being highly variable in time and strongly synchronized (Q.III). To further elaborate on this, as previously stated, oaks in the selected stand had a higher relative elevation (10–20 cm) with respect to the rest of the site (Fig. 1). On the one side, this could be one of the explaining factors of these trees survival so far, as it is well documented in Scharnweber et al. (2014). On the other side, this could explain the low synchronicity in pre and post-1 in relation to TRW: trees reacted similarly only in post-2, in response to the expression of a strong limiting factor represented by the change in water regime (Roibu et al. 2020). Moreover, the high variability and the general increase in TRW connected to post-1 might be dependent on two main factors, a higher availability of water in a formerly controlled and dry environment, and the release effect cause by the death of the neighboring trees, that fostered a competitive release dynamic of the stand.

Short- and long-term effect of the rewetting

Pointer years analysis was used to address traits behavior in relation to specific events (Q.IV). Results for the TRW time series were validated with existing chronologies from the previous study conducted in the area (Scharnweber et al. 2014) to avoid the identification of false pointer years. This comparison was performed as the low sample size could give misleading results, leading to a higher number of detected pointer years. Nevertheless, the analysis revealed that the pointer years identified are reliable, as they match in nature and number with the pattern of the other three TRW chronologies (dead–damaged–healthy), and especially with the pointer years found for the damaged trees. This correspondence could hint at some sort of difficulties in co** with the rewetting. Results on the current trait chronologies revealed how the trees recorded the dyke brakeage with the characteristics of the “flood year” in 1996, affecting mainly the earlywood traits (St. George et al. 2002; Copini et al. 2016; Ballesteros et al. 2010). The delay of 1 year with respect to the event is explained by the fact that the storm that broke the dyke and the consequent flooding happened in November 1995, when the growing season was already over. In contrast with another study (Wertz et al. 2013), we were able to identify the effect of the flooding event on the following year (1996), though this had happened during winter dormancy. This finding is in line with results showed in Sample and Babst (2020), where growth decrease was found in the subsequent year after autumn flooding. Pointer years analysis did not show significant results for the second event, the opening of the artificial canal, and the removal of the pum** station in 2004, nor in the following year. However, pointer years analysis on the healthy TRW chronology from Scharnweber (2014) recorded this event as a significant positive pointer year. Despite in the current chronologies, 2004 was not significant, the last period of the trait analyses (post-2) is characterized by consequent significant pointer years, both positive and negative. This could potentially support the previous conclusion for which we assumed that the permanently high water table level, which is at the same time variable due to the opening of the artificial canal, enhanced trees sensitivity to local microsite conditions. Moreover, since traits chronologies covered the drought that occurred in 2018, we could observe how known opposite events (drought and flooding) possibly led to a similar response in the nature of the pointer years. We could not verify to which extent this drought event was experienced by this peculiar peatland site, but a similar effect was already shown in the longer time frame by Scharnweber et. al (2014) where both regime shifts (drainage after the dyke construction and rewetting after the dyke damage) resulted in a positive growth reaction in the surviving oaks. In general, although the drought effects concentrated on the latewood and flooding effect on the earlywood, we would advise for a careful interpretation of the wood anatomical traits, especially to infer climatic information. In this regards, questions arose in relation to the drought event that happened in 2019 as it was shown to have impacted trees in a stronger extent than in 2018 (Schuldt 2020, Schnabel 2022). A possible explanation resides in the lagoon’s influence on the site, as its rising level might have mitigated the effect of this second drought event.

Wood anatomical and hydraulic traits were also tested for correlations with climate to understand what was the role of the water level of the lagoon, temperature, and precipitations in sha** the traits response to the new hydrologic conditions. As these trees are located in a highly disturbed environment, and tree response is highly dependent on the local site condition, we looked at the results with caution (Online Resource 7). Climate correlation supported the interpretation that the post-1 period was the most sensitive to the overall environmental conditions, as it represents the transition phase from a very controlled environment (pre), to a permanently flooded environment (post-2). Beside a negative correlation for May precipitation, and a positive correlation for winter temperature in post-1, which showed somewhat consistent significance through the analyzed traits, the other correlations were either weak or mostly non-significant. Moreover, as demonstrated in previous dendroecological studies (Anadon-Rosell et al. 2022; Sáenz-Romero et al. 2017; Mérian et al. 2011), the anatomical traits, and in turn TRW, showed no clear pattern when tested toward several climatic variables, nor to the water level of the adjacent lagoon; hence, it could be further implied that probably the high local soil water content represented the main limiting factor for the area.

The practical implications for the intervention operated in this area are difficult to summarize as the whole dynamic occurred in a highly disturbed and diverse environment, therefore leading to a complex setting of variables involved. While on one side, more than the 50% of the trees died in the forested areas, according to the vegetation zonation conducted in 2004 (Timmermann et al., 2008), leading to the clear conclusion that the forest could not survive the rewetting, on the other side, trees that survived seemed to benefit from the rise in the water availability. The difference in the survival rate was already investigated in Scharnweber et al. (2014) where relative elevation of single trees played a crucial role in determining the future of the plant, together with a slower growth rate before the flooding. The EW-W of these trees was significantly higher than the one of damaged and dead trees, which could possibly connect to the elevation explanation, indicating that in general they were less affected by the rise in water level. All these findings suggested that the rewetting measures on forested peatland will permanently affect the stand, but microsite condition, together with the release effect cause by the death of the neighboring trees, could still promote the survival of the individuals.

CARROT

As a novel advancement, a fully functioning AI software has been employed to perform the complete quantitative wood anatomical study. CARROT, which is an upgrade of a previously published software (Resente et al., 2021), proved to be reliable in both the detection and the calculation of the vessels and tree ring parameters, that were later used to derive the commonly employed wood anatomical traits. The error analysis carried out by Resente et al. (2021) supports our findings, where the smallest fraction of the whole vessel distribution is the most affected category by detection inaccuracy. However, our results proved the very high performance of the software.

Such performance level was achieved, thanks to the initial training of the oak model, and in this regard, it is worth mentioning the importance of this stage. The images included in the current dataset did not differ much in appearance from the ones used for training the model. On the one side, the quality of the sections was on average acceptable, and on the other side, the acquisition process via the Zeiss Axio Scan.Z1 slide scanner (Carl Zeiss AG, Germany) provided clear and focused images. These characteristics, which are as well shared by the images in the training dataset, allowed us to achieve the abovementioned high-quality results. The features, and especially the variety of the features that characterize the images used for training the algorithms, determine the accuracy with which new images are interpreted. The more the training dataset entails a large variability of images, the more the algorithm will be able to process different datasets. As this is an acknowledged issue in visual machine learning and image processing, CARROT was provided with the possibility of retraining the existing models. This additional setting is still dependent on the capabilities of the initial model, and therefore on the initial training dataset. However, as technology is advancing in every field of wood anatomy, from the slicing technique to the acquisition process, we expect an increase in quality and a more homogeneous appearance of the wood anatomical images.

Conclusion

Generally, in the first phase of the rewetting, we could infer that flooding triggered (i) variability and a rapid increase in the growth anatomical traits, where trees could locally benefit or be hindered by a higher amount of water, and (ii) homogeneity in the hydraulic structure that, beside the microsite effect, reacted similarly to the new soil conditions.

When the rewetting process was enhanced by the canal opening interventions, local water table level emerged as the main limiting factor in the area. Decrease in the hydraulic diameter and specific hydraulic conductivity, associated with a general increase in vessel number and ring width, suggested the importance of the vessel widening mechanism for trees ability to cope with the new hydrologic regime. The synchronized pattern of the individual ring width series supported this assumption and suggested that oak became more sensitive to local site conditions.

The focus on the short-term effect of the flooding event showed that wood anatomical traits reacted as expected from the literature, beside highlighting similarities in response to two contrasting events: the flooding in 1995 and the drought in 2018, despite the wet site conditions. For this reason, we recommend particular care in the inference of climatic events strictly from wood anatomical parameters. Giving precise information on policies implementation pose a challenge due to the highly varied nature of the environment, resulting in a complex interplay of factors. Sustainable forest management aimed at restoring peatlands should carefully plan rewetting interventions considering the importance of the microsite effect for the existing vegetation.