Modelling ecohydrological feedbacks in forest and grassland plots under a prolonged drought anomaly in Central Europe 2018–2020

Recent studies have highlighted the importance of understanding ecohydrological drought feedbacks to secure water resources under a changing climate and increasing anthropogenic impacts. In this study, we monitored and modelled feedbacks in the soil–plant‐atmosphere continuum to the European drought summer 2018 and the following 2 years. The physically based, isotope‐aided model EcH2O‐iso was applied to generic vegetation plots (forest and grassland) in the lowland, groundwater‐dominated research catchment Demnitzer Millcreek (NE Germany; 66 km2). We included, inter alia, soil water isotope data in the model calibration and quantified changing “blue” (groundwater recharge) and “green” (evapotranspiration) water fluxes and ages under each land use as the drought progressed. Novel plant xylem isotope data were excluded from calibration but were compared with simulated root uptake signatures in model validation. Results indicated inter‐site differences in the dynamics of soil water storage and fluxes with contrasting water age both during the drought and the subsequent 2 years. Forest vegetation consistently showed a greater moisture stress, more rapid recovery and higher variability in root water uptake depths from a generally younger soil water storage. In contrast, the grassland site, which had more water‐retentive soils, showed higher and older soil water storage and groundwater recharge fluxes. The damped storage and flux dynamics under grassland led to a slower return to younger water ages at depth. Such evidence‐based and quantitative differences in ecohydrological feedbacks to drought stress in contrasting soil‐vegetation units provide important insights into Critical Zone water cycling. This can help inform future progress in the monitoring, modelling and development of climate mitigation strategies in drought‐sensitive lowlands.


| INTRODUCTION
Sustaining water resources and ecosystem services are complex challenges in the context of accelerating land use and climate change in the Anthropocene (Gleeson et al., 2020). The important role of vegetation in regulating terrestrial water fluxes Jasechko et al., 2013) as well as the potential for manipulating land cover for climate change mitigation (Silva & Lambers, 2020) are increasingly recognized. However, the precise ways in which different vegetation communities affect ecohydrological partitioning are poorly understood, with quantitative separation of interception, evaporation and transpiration usually being highly uncertain . Consequently, the ways in which vegetation and ecohydrological partitioning respond to climate change as well as the effects on water availability for root zone storage and groundwater recharge is still a key challenge (Brooks et al., 2015). This makes it difficult to assess the sensitivity and resilience of different land use strategies to climate change (Li, Migliavacca, et al., 2021).
Under climate change, in many areas, droughts are predicted to become more frequent, with multiple impacts on hydrological systems (Mishra & Singh, 2010). The expected increase in drought occurrence in central Europe in the 21st century, underline the need to develop effective mitigation to ensure integrated and sustainable land and water management policies for future climatic conditions . Of course, there are differences in drought-induced reductions of blue (recharging ground and surface water) and green water (transporting moisture back to the atmosphere; Falkenmark and Rockström, 2006) fluxes in near-natural systems across Europe (Orth & Destouni, 2018). Contrasting vegetation communities show varying sensitivity to water scarcity, depending on physiological adaptations and the nature of subsurface water storage (Lobet et al., 2014). Thus, there is potential for mitigating the effects of drought and subsequent "memory effects" through management of local green water fluxes, rather than basing management decisions solely on maintaining the provision of blue water fluxes (Rockström et al., 2009).
One way forward to address existing knowledge gaps is integrating multiple streams of relevant data into the calibration and validation of process-based ecohydrological models (Fatichi et al., 2016;Guswa et al., 2020). Such models facilitate quantitative estimates of blue and green water fluxes from different soil-vegetation systems. This helps to inter-compare between landuses and thus, to understand differences in partitioning under drought conditions and subsequent recovery. To constrain models and reduce uncertainty, multicriteria calibration is invaluable, potentially incorporating high information content on key processes.
The abundances of the heavier stable isotopes, deuterium (δ 2 H) and oxygen-18 (δ 18 O), in the water molecule are particularly useful (Birkel & Soulsby, 2015;Turner & Barnes, 1998) and well-established tracers (Gat & Gonfiantini, 1981) for providing such additional information. Isotopes are natural tracers that reflect phase changes (e.g., evaporative effects) and mixing with storage in different compartments of the Critical Zone. This is the thin, dynamic, life-sustaining skin of the Earth that extends between the atmospheric boundary layer and the bottom of the groundwater. Using isotopes in the calibration and/or validation of ecohydrological models can test whether process-based conceptualisations are "getting the right answers for the right reasons" (Kirchner, 2006). Importantly, such models can also estimate water ages (Sprenger et al., 2019) and provide insight into large scale ecohydrological partitioning Tetzlaff et al., 2015). Water age is an important metric of hydrological function which indexes linkages between mixing, storages and fluxes in landscapes. Observations of water stable isotope dynamics in the subsurface are useful to understand the pathways of water (Li, Sullivan, et al., 2020) and can substantially aid multi-criteria calibration . Recently, the model EcH 2 O (Maneta & Silverman, 2013) has been advanced to EcH 2 O-iso (Kuppel et al., 2018) to quantify the relevant fluxes governing ecohydrological partitioning and to track the isotopic (δ 2 H, δ 18 O) composition and age of water through the model domain. This allows isotopes to be used as both calibration constraints for key processes, as well as a means of validating model performance if sufficient isotope time series are available . The quantification of water ages in green water fluxes helps to assess the resilience of the associated ecohydrological fluxes and ecosystem services, as well as the temporal dimension of feedbacks to climate extremes (Kuppel et al., 2020).
The State of Brandenburg in NE Germany forms part of the Northern European Plain and is a drought-sensitive lowland area surrounding the capital city of Berlin. The region has high societal importance for the provision of several ecosystem services; these include food and timber production, groundwater recharge and contributions to drinking water supplies for over 5 million people. The Demnitzer Millcreek experimental catchment (DMC, 40 km SE of Berlin) was established in 1990 to understand the effects of agricultural pollution on surface water quality (Gelbrecht et al., 2000(Gelbrecht et al., , 2005. Latterly, work has focused on understanding ecohydrological partitioning at the catchment scale , adding spatially distributed monitoring of soil moisture and groundwater, to complement the long-term rainfall and stream flow measurements . Extensive monitoring of isotope dynamics in the catchment started just before the European drought in 2018 and was expanded to more ecohydrological compartments thereafter . The drought of 2018 was followed by a prolonged period of reduced rainfall when most monthly rainfall anomalies were negative and temperatures remained above average. Such conditions are anticipated to become more common in the next decades (Lüttger et al., 2011).
Future climate may result in lower groundwater recharge, stream network disconnections and reduced production of soil organic matter (Fleck et al., 2016).
Here, we aim to build on preliminary work by , to integrate new and extended isotopic data from the subsurface and vegetation into an integrated monitoring and model-based assessment of how prolonged (two subsequent vegetation growing periods) drought affects ecohydrological feedbacks in two contrasting soil-vegetation units. We focused on the time-variant effects of a prolonged period (2018-2020) of predominantly negative rainfall anomaly and its effects on water storage, flux and age dynamics and persistence in the Critical Zone. By including soil water isotopes in model calibration, as well as plant xylem isotopes in the model evaluation, we aimed to further constrain the model application with reduced parametrisation to assess the extent and persistence of the extreme atmospheric conditions. Crucially, this work examined how the system responded in the growing season in the 2 years following the most severe regional drought conditions of the 21st century. To do this, our investigation used the ecohydrological model EcH 2 O-iso to address the following objectives: 1. To quantify the impacts of prolonged drought on ecohydrological fluxes in two common soil-vegetation (forest and grassland) units; 2. To use the water stable isotopes dynamics of soil and vegetation in model calibration and evaluation, respectively; 3. To explore the contrasting time-variant impact of ongoing drought conditions on the storage-age-flux dynamics between sites.
Further, we discuss the implications of our findings on drought and recovery for future sustainable management of water resources and associated ecosystem services in the Demnitzer Millcreek catchment, which is representative for other lowland, mixed land use, groundwater dominant landscapes.

| STUDY SITE
The data used in this study were collected from the DMC, which is located in NE Germany (52 23 0 N, 14 15 0 E; Figure 1). This lowland region experiences a temperate humid warm summer climate (Kottek et al., 2006). Mean air temperature is 9.6 C with a mean annual precipitation of 567 mm/yr (DWD, 2020(DWD, , for the period 2006(DWD, -2015.
Precipitation falls throughout the year, but seasonal differences lead to higher summer precipitation from fewer, high intensity, convective events and lower amounts during more frequent frontal rain in winter.
The DMC lowland landscape (Figure 1(b)) was shaped by the last glaciation (Weichselian), which resulted in generally sandy soils on glacial and fluvial deposits. The catchment is groundwater-dominated and historically had little surface runoff and was characterized by numerous peat fens and freshwater lakes in hollows, but these were drained during a long history of anthropogenic usage (Nützmann et al., 2011). Current land use is dominated by forestry and farming (for more details see Kleine et al., 2020;Smith et al., 2021). The relatively sparsely populated catchment is a setting for recovering wildlife populations including recolonization of beaver (Smith, Tetzlaff, Gelbrecht, et al., 2020), wolf (Vogel, 2014) and even sporadic sighting of elk (Martin, 2014).
For the landscape to maintain its important ecosystem services in this lowland part of Brandenburg, sufficient seasonal precipitation input is needed to retain root zone soil moisture levels that sustain crop and tree growth (Drastig et al., 2011). Further, adequate groundwater recharge is needed to sustain groundwater-surface interactions. However, the low water retention in the dominant sandy soils and high ($90%) proportions of evapotranspiration losses dominate the water balance , resulting in drought sensitivity of the catchment . Additionally, the imprint of vegetation by mediating ecohydrological partitioning results in temporary catchment scale patterns of stream network disconnections during droughts Smith et al., 2021).
In this study, the ecohydrological fluxes in the near-surface Critical Zone were investigated at two plot sites in the western parts of the catchment (Figure 1). As the topography is flat, elevation differences between the sites are negligible. Given the high permeability of the soils, the occurrence of surface ponding of water or surface runoff was not observed during the study period. The plots are characterized by different soil properties and vegetation types. The forested site is dominated by broad-leaved trees (mainly European oak) with one mature Scots Pine in the plot. Other species like maple and elm tree or hazel are present in the immediate vicinity (<10 m). The soil is a sandy freely draining Lamellic Brunic Arenosol (Humic; Table 1). The second, grassland site is characterized by pasture including higher proportions of finer grain sizes in the upper soil relative to the forest and a somewhat more water retentive Eutric Arenosol (Humic, Transportic; Table 1). This site is in close spatial proximity to the forested site ($400 m) as well as the stream ($10 m) and subject to some shading effects . The grassland site is fenced and usual management (with cutting once a year) was simulated within the plot.

| Climatic input data
The model forcing climatic daily input data (see Table 2) for the study period (January 2018-September 2020) and spin-up period (2016 and 2017) were based on long-term weather station data of the German Weather Service (Deutscher Wetterdienst (DWD), 2020) and an automatic weather station (AWS, Environmental Measurement Limited, UK) at Hasenfelde, which was installed in May 2018. This data was further supplemented by global atmospheric reanalysis dataset ERA 5 (Hersbach et al., 2020) for radiation, as well as MODIS (Running et al., 2017) data for 8 day estimates of evapotranspiration and latent heat. We also calculated vapour pressure deficit (VPD) (Allen et al., 1998).

| Plot site installations
Transpiration rates were derived from 12 trees at the forested site using 2-4 Granier-type sensors per tree (Thermal Dissipation Probes, Dynamax Inc., Houston, details in . The time-series was normalized by subtracting the data's mean and dividing by the standard deviation. Volumetric soil moisture content was

| Isotopic sampling
Stable water isotopes in precipitation were sampled daily from July 2018 onwards with a modified ISCO 3700 autosampler (Teledyne ISCO, Lincoln) at the Hasenfelde AWS (Figure 1 bulk soil water were derived using the direct equilibrium method (Wassenaar et al., 2008). The time for equilibration between liquid water and added dry air headspace was $48 h at room temperature (21 C). Quality criteria for measurements were applied to a 2 minute plateau in the standard deviation of water content (< 100 ppm), δ 2 H (< 0.55 ‰) and δ 18 O (< 0.25 ‰). Derived isotopic signatures were corrected for potential gas matrix change   As for the soil isotopes, three replicates were obtained per site and sampled vegetation type. Samples were rapidly taken and immediately placed in sealable glass vials (9.190605, Faust Lab Science GmbH, Klettgau, Germany) and frozen at À20 C upon return to the laboratory. Water from vegetation samples was extracted in January 2020 in the laboratory of the Ecosystem Physiology (University Freiburg) using a cryogenic extraction line routine described in Dubbert et al. (2013Dubbert et al. ( , 2014. The samples were heated to 100 C, the applied extraction pressure was 0.03 Pa and the extraction time approximately 90 min. After extraction, samples were weighed and ovendried for 24 h at 105 C. We excluded samples with bad extraction T A B L E 1 Soil properties at the two plot sites (for more details see Kleine et al., 2020) Allen and Kirchner (2021) to assess the offset between forest plant xylem isotopes as well as soil isotopes and model output on the other. This method is useful for visual comparison between modelled and simulated xylem isotopes to strengthen the confidence in the model results despite methodological uncertainties in xylem isotope sampling from woody plants.
Here, we constrained the reporting of water isotopes to mainly δ 2 H signatures to reduce redundant information content. To assess relative changes between isotope abundances, we utilized the lineconditioned excess (lc-excess; Landwehr & Coplen, 2006) as nonconformity with the local meteoric water line (LMWL): We used an amount weighted least squares regression (Hughes & Crawford, 2012) to calculate the LMWL with precipitation exceeding 1 mm

| Standardized precipitation index (SPI)
The the deviation from normal precipitation amounts in the same periods in the long-term data from the value of 2 (extreme wet) to À2 (extreme dry). The SPI can be calculated for a specific period (here 6 months) and phenomena reflected by the SPI vary with the set period.

| The EcH 2 O-iso model
The spatially distributed ecohydrological model EcH 2 O (Maneta & Silverman, 2013) was extended to EcH 2 O-iso to include tracking of water stable isotopes and water ages by Kuppel et al. (2018). shown with ranges in Appendix A). Based on the soil profiles (Table 1) and these previous simulations, the soil in the model application was discretised into three soil layers (1, Figure 2 and Table 2. We used the model to derive flux and storage quantities during the study period and therefore deliberately excluded a validation time-series. We produced 100 000 parameter sets per site for Monte Carlo calibration using Latin Hypercube Sampling. We then identified the "best" 30 runs by multi-criteria calibration over the study period, covering drought and recovery conditions. We included soil moisture, δ 2 H and lc-excess for the three soil layers, evapotranspiration, latent heat (and additionally sap flow at the forested site; see Table 3) in the calibration. The multi-criteria calibration used the combination of all efficiency criteria of calibration parameters ranked between 0 (worst) and 1 (best) for all runs to identify the best parameter sets. We used the mean absolute error (MAE) and the Nash-Sutcliffe efficiency (NSE, Nash & Sutcliffe, 1970) as efficiency criteria for model calibration.
NSE was exclusively used for volumetric soil moisture in layers 1 and 2, as well as sap flow due to the observed high temporal variability, whereas less dynamic variables were assessed with the MAE (see Table 3). To assess annual differences of the water balance within and between sites, the best 30 model runs were averaged and aggregated over calendar years. As the study period ended in September 2020, the values allow inter-site comparison rather than inter-annual comparisons.
F I G U R E 2 Climate input dataset for model forcing and atmospheric vapour pressure deficit after Allen et al. (1998) T A B L E 3 Model best 30 runs mean and range of efficiency criteria of multicriteria calibration parameters

| Model performance
The site-specific dynamics and damping of volumetric soil moisture with depth were well reproduced for layer 1 and 2 by the model, both in 2018 and the subsequent two summers (Figure 3). The model also hindcast the moisture conditions at the start of 2018 following a wetter autumn and winter. The NSE for soil moisture simulations (Table 3)

| Prolonged drought impacts on ecohydrological fluxes
The 6  The deepest soil layer showed the oldest water ages increasing in age over the study period. Like the forest site, uncertainties in water age F I G U R E 9 Weekly transpiration rate and age (top) with according stacked root water uptake (RWU) percentage from model soil layers 1 (0-15 cm), 2 (15-50 cm) and layer 3 (50-100 cm) of the forest (left) and grass (right) plot site F I G U R E 8 Mean annual water ages of soil storages 1, 2, 3 and transpiration at forest (left) and grass plot (right) estimates increased with soil depth but were less pronounced from few days in layer 1-2 months in layer 3.
Weekly transpiration sources and their ages also differed between the sites (Figure 9). Transpiration fluxes at the forest showed higher mean simulated fluxes (3.4 ± 1.1 mm/week) and higher maximum values of 22.3 ± 4.7 mm/week over the study period. Peak transpiration rates were much higher in 2018 (after a rain event), though overall transpiration amounts were similar to 2019 and only 10% higher than 2020. Infiltration of precipitation events into the low storages of soil layer 1 increased soil moisture and fractions of RWU from layer 1, resulting in younger associated water in transpiration (shown in the lower panel of Figure 9). This led to a mean transpiration age (weighted by transpiration flux) of 162 days. There was no evidence of soil water ages systematically increasing. Soil water in deeper storage only became older during times of very low transpiration fluxes and RWU in winter.
The grassland site showed lower variability in transpiration fluxes during the study period with a lower mean (2.7 ± 0.8 mm/week) and lower RWU dynamics in all 3 soil layers ( Figure 9). This led to the youngest water ages in transpiration at the end of summer (as under forest), but with less pronounced influences of new precipitation on water ages in the soil storages. The RWU was generally from higher and therefore younger, soil storages compared to the forest. Soil water in layer 3, which also feeds groundwater recharge, became older throughout the year as RWU was limited. New precipitation water of larger precipitation events in late summer percolated down the soil profile to mix with the existing older water. However, the con- were 40% lower in 2018 and the effects of reduced groundwater are expected to persist for several years (Kannenberg et al., 2019). Such conditions are expected to occur more often in future .
A strong seasonality in ecohydrological fluxes and water partitioning was apparent under both plots ( Figure 6) and the timevariable response is in accordance with previous studies (Sprenger et al., 2016;Thaw et al., 2021). Both sites showed small negative storage dynamics in their water balance in 2018, with drought effects seemingly mitigated by a wet winter in 2017-2018 (Figures 6 and 7).
Thus, subsurface storage sustained green water demands, with no observed major limitation in forest transpiration due to precipitation input of an summer rain event (10-12.7.2018, 59 mm) that likely mediated the developing soil water deficit. Groundwater recharge primarily occurred at both sites in winter and was reduced in 2019, when subsurface storages were not fully rewetted and many blue water fluxes almost completely ceased in the subsequent growing season.
We also observed a return to shallower RWU under wetter conditions in 2020. Overall, our findings were similar as in Orth and Destouni (2018) in terms of drought impacts being stronger on blue rather than green water fluxes in NE Germany. More precipitation inputs in late 2019 and early 2020 recovered the 6 months SPI and fractions of blue water fluxes increased. However, the effect was transient and at the end of the study period, with the SPI declining again, blue water fractions were reduced at both sites.
Our study underlined the differences in the drainage characteristics of subsurface storage between sites. Besides the differences in soil characteristics (Table 1), soil moisture dynamics were influenced by the higher interception losses  and deeper RWU under forest vegetation. The more retentive grassland soil storage showed higher water content and less variability in RWU depths.
The dominant oak at the forest site can adapt to drought conditions by plasticity in physiological characteristics (e.g., inter-calary veins, leaf size etc.) and therefore shows acclimation properties (Günthardt-Goerg et al., 2013). In addition to the regional indications for higher oak drought resistance (Scharnweber et al., 2011), the rooting depth of the oak forest stand might be deeper than in monocultural conifer stands (Bello et al., 2019) which are common in Brandenburg and contribute greater resilience to changing climate (Pretzsch et al., 2020).
Whilst grassland vegetation can also show physiological drought adaptations (Hanslin et al., 2019) and species-dependent water use strategies (Nippert & Knapp, 2007), we did not find such dynamics in the grassland plot due to limited drought effects on shallow soil water storage. Nevertheless, grassland in the DMC could still provide further potential in drought mitigation strategies (Volaire et al., 2014).
Drought adaptations are dependent on the hydraulic properties of the soil-root system (Lobet et al., 2014) and the more retentive grassland soil provided higher soil moisture in the upper soil profile throughout the study. The observed vegetation strategies of RWU under drought were linked to plant-available soil moisture.
The forest site also showed higher variability in groundwater recharge fluxes which is consistent with the soil water dynamics.
Annual groundwater recharge fractions in the forest water balance were also lower reflecting the higher interception, transpiration and more freely draining soil (Figure 7). The simulated groundwater recharge is consistent with other modelling studies in the region (Douinot et al., 2019;Smith et al., 2021). This flux is especially important in DMC where green water fluxes dominate (Smith, Tetzlaff, Gelbrecht, et al., 2020), but surface water presence and related ecosystem services are dependent on groundwater .
In drought and climate change mitigation efforts, understanding vegetation effects on hydrological functioning (Levia et al., 2020) and cross-scale assessment of the water cycle will be essential to enable the management of future societal demands (Gleeson et al., 2020).

| Dynamics in stable water isotopes under different land use types
During passage through the soil-plant atmosphere continuum, the stable isotopic signature of water is affected by phase changes, flow paths, hydrological connectivity and associated mixing with water stored in the Critical Zone (Kendall and McDonnell, 1998). This has motivated the extensive isotope sampling conducted at DMC. Integrated into hydrological models, water stable isotopes have the potential to assess mixing relationships between fluxes and storages as associated effect on water ages (Birkel & Soulsby, 2015). Over a 14 months period we observed site specific temporal dynamics in bulk soil water isotopic signatures and damping with depth . Dynamics in soil moisture and soil isotope signatures were quite well reproduced by EcH 2 O-iso for the 14 months in the upper two layers. The subsurface sampling strategy of bulk soil isotopes was important to constrain model parameters, providing six datasets per site in the multicriteria calibration strengthening our confidence in the representation of subsurface processes. We assumed lateral water movement was negligible as done by others in less flat landscapes and areas dominated by freely draining soils (McGuire & McDonnell, 2010;Sprenger et al., 2016).
The dry conditions were reflected by evaporatively enriched isotopic signatures (Craig et al., 1963) mainly in the upper soil profile (Sprenger et al., 2017), which were well captured in the modelling, as was the mixing and dampened dynamics with depth. It seemed important to exclude potential misinterpretation from higher organic material in the upper soil (Table 2) and associated effects on isotope measurements . Although vegetation isotopes were not used in model calibration, the simulated isotopes represented well the measured dynamics at both sites ( Figure 5). In the forest, plant xylem isotopes showed much less similarities to upper soil layer isotopes than under grassland (Figure 4).
Recent scientific studies emphasize the need to consider methodological uncertainties in isotope sampling and analysis (Chen et al., 2020;Orlowski et al., 2018). New investigations suggest a potential correction range for the water isotopic signal of woody plant matrix extracts with a mean of $ + 8.1 ‰ in δ 2 H (Allen & Kirchner, 2021;Chen et al., 2020). If we consider these uncertainties, the offset in woody forest vegetation relative to the soil isotopes ( Figure 4) and simulated transpiration isotopes ( Figure 5) might support such an adaptation magnitude at our site. Regardless of the adaptation, the forest vegetation isotope signatures still reflected deeper water sources than at the grassland site which more clearly resembled bulk soil water isotope dynamics in the shallow soil water (Figure 4).
These observed patterns in grassland RWU were supporting the modelled dominance of the upper soil water on transpiration fluxes here and as observed for other lowland sites (Prechsl et al., 2015).

| Drought impact on storage-age-flux dynamics
We assessed water ages in soil-plant storage and fluxes at our two sites as well as their temporal variations under prolonged, exceptional atmospheric conditions to understand interactions between multiple ecohydrological compartments (Dimitrova-Petrova et al., 2020;Evaristo et al., 2019;Sprenger et al., 2019). Soil water ages in the forest site were generally younger and more dynamic. This was explained by the smaller soil water retention capacity and higher "water use" by the forest vegetation relative to grassland (Douinot et al., 2019). The older soil water ages under grassland were supported by the reduced variability in soil moisture and higher water retention. In soil layer 3, water ages still became older during 2020, showing a slower response to rewetting conditions due to higher water content and limited depletion by RWU.
Transpiration ages were younger for the forested site and directly linked to RWU patterns from the younger, more limited soil storage ( Figure 9). The lower clay content in the upper forest soil profile promotes younger water ages (Sprenger et al., 2016). The depth of RWU at the forest site was more dynamic and deeper soil storage was especially important for vegetation under drought conditions to sustain green rather than blue water fluxes (Orth & Destouni, 2018). We simulated that during the growing season, the depth of modelled RWU from the forest shifted downwards as soils dried (Figure 9), underlining the importance of older soil water for temperate forests (Brinkmann et al., 2018). Forest vegetation accessing younger water and being more dynamic in exploiting water sources was also found in other recent research (Thaw et al., 2021). This reflects the higher transpiration potential in summer coupled with low soil potential in layer 1, in combination with the time-invariant root proportion distribution in EcH 2 O-iso, which exponential decreases with depth. The continuously increasing grassland transpiration ages beyond drought conditions also how the root distribution is conceptualized, allowing grassland RWU from older layer 3 storage. It is interesting that transpiration at both sites is depressed in 2020 compared to 2019, despite increasing wetness, which is rather related to lower atmospheric demand than to the (not simulated) adaptation by the vegetation cover.
Water ages derived from integrating extensive isotope data into ecohydrological models can give important insights on temporal aspects of storage and flux vulnerability to drought (Kuppel et al., 2020). Younger water ages of forest soil storages and transpiration highlight the potential vulnerability to drought conditions; and although faster recovery (of the soil storage) occurs, the differences in soil water ages and soil moisture are more vulnerable to negative rainfall anomalies. Further, the grassland site experienced a longer drought legacy in deeper soil water ages after rewetting. The older soil water ages during the drought (forest) and during rewetting (grassland) highlight the importance of transpiration water sources fallen prior to the growing season (Brinkmann et al., 2018).

| Wider implications
The observed differences in blue and green water fluxes emphasize the need for considering spatially discretised mitigation objectives in the DMC and comparable lowland landscapes . It is crucial to further assess such water dynamics and related effects on forest ecosystems under a changing climate (Vido & Nalevanková, 2021). It is also important to further investigate how spatial and temporal patterns of green water use in droughts impact blue water provision (Freire-González et al., 2017) from more soil/vegetation units over subsequent growing seasons. We emphasize that the complex and dynamic vegetation effects on soil properties and vice versa associated with land use management strategies (Silva & Lambers, 2020) will increasingly need to be included in long-term ecohydrological modelling to understand effects on subsurface water storage by sustainable management. Forest water use and reduced drought recovery could also be assessed by expanding more routine monitoring of radial stem growth and sap flow dynamics over a wider range of species (Dang et al., 2019). This is important to evaluate other long-term impacts for example, expected increased mortality in regional forest ecosystems by the 2018 drought and secondary drought impact events (Schuldt et al., 2020). The grassland differed in dynamics, indicated longer drought effects on the subsurface water ages in our study.
Isotope-aided ecohydrological modelling as a process-and evidence-based tool proved invaluable in assessing such drought feedbacks and can be used to help evaluate vegetation-focused drought mitigation strategies . Given, the differences in the persistence of drought effects between sites, multi-year assessments of drought events are required. On the basis of their importance in the DMC, we see potential in further development of the EcH 2 O-iso model (Kuppel et al., 2018). This could include the implementation of more