Isotope‐aided modelling of ecohydrologic fluxes and water ages under mixed land use in Central Europe: The 2018 drought and its recovery

Understanding the interactions of vegetation and soil water under varying hydrological conditions is crucial to aid quantitative assessment of land‐use sustainability for maintaining water supply for humans and plants. Isolating and estimating the volume and ages of water stored within different compartments of the critical zone, and the associated fluxes of evaporation, transpiration, and groundwater recharge, facilitates quantification of these soil–plant‐water interactions and the response of ecohydrological fluxes to wet and dry periods. We used the tracer‐aided ecohydrological model EcH2O‐iso to examine the response of water ages of soil water storage, groundwater recharge, evaporation, and root‐uptake at a mixed land use site, in northeastern Germany during the drought of 2018 and in the following winter months. The approach applied uses a dynamic vegetation routine which constrains water use by ecological mechanisms. Two sites with regionally typical land‐use types were investigated: a forested site with sandy soils and a deep rooting zone and a grassland site, with loamier soils and shallower rooting zone. This results in much younger water ages (<1 year) through the soil profile in the forest compared to the grass, coupled with younger groundwater recharge. The higher water use in the forest resulted in a more pronounced annual cycle of water ages compared to the more consistent water age in the loamier soil of the grasslands. The deeper rooting zone of the forested site also resulted in older root‐uptake water usage relative to soil evaporation, while the grassland site root‐uptake was similar to that of soil evaporation. Besides more dynamic water ages in the forest, replenishment of younger soil waters to soil storage was within 6 months following the drought (cf. >8 months in the grassland). The temporal evaluation of the responsiveness of soil and vegetation interactions in hydrologic extremes such as 2018 is essential to understand changes in hydrological processes and the resilience of the landscape to the longer and more severe summer droughts predicted under future climate change.

groundwater recharge, evaporation, and root-uptake at a mixed land use site, in northeastern Germany during the drought of 2018 and in the following winter months. The approach applied uses a dynamic vegetation routine which constrains water use by ecological mechanisms. Two sites with regionally typical land-use types were investigated: a forested site with sandy soils and a deep rooting zone and a grassland site, with loamier soils and shallower rooting zone. This results in much younger water ages (<1 year) through the soil profile in the forest compared to the grass, coupled with younger groundwater recharge. The higher water use in the forest resulted in a more pronounced annual cycle of water ages compared to the more consistent water age in the loamier soil of the grasslands. The deeper rooting zone of the forested site also resulted in older root-uptake water usage relative to soil evaporation, while the grassland site root-uptake was similar to that of soil evaporation.
Besides more dynamic water ages in the forest, replenishment of younger soil waters to soil storage was within 6 months following the drought (cf. >8 months in the grassland). The temporal evaluation of the responsiveness of soil and vegetation interactions in hydrologic extremes such as 2018 is essential to understand changes in hydrological processes and the resilience of the landscape to the longer and more severe summer droughts predicted under future climate change.

| INTRODUCTION
Many regions worldwide have been under increasing duress from warmer, drier summers, with long-term projections showing likely future exacerbation of these conditions (King & Karoly, 2017;Quesada, Vautard, Yiou, Hirschi, & Seneviratne, 2012). Longer-term climatic projections indicate more frequent droughts, which will negatively impact agricultural productivity with increased annual soil evaporation and transpiration ("green" water fluxes), with a complementary reduction in groundwater recharge ("blue" water fluxes) and annual water availability (Iglesias & Garrote, 2015;Orth & Destouni, 2018). In Europe, with increased occurrence of extreme heatwaves and long-term trends of decreasing soil moisture (Christidis, Jones, & Stott, 2014;Hanel et al., 2018;Schoetter, Cattiaux, & Douville, 2014), it is necessary to quantitatively examine how different species, vegetation communities, and land management strategies affect water partitioning to balance increased crop and timber production with reductions in blue fluxes and while maintaining ecosystem health (Teuling et al., 2010). There is a need to continuously assess the recovery time of different species/communities and soils under recurrent droughts to identify regions with high sustainability risk due to limited availability of green water sources (Ivits, Horion, Fensholt, & Cherlet, 2014;Schwalm et al., 2017). Consequently, future land use and agricultural planning will require an evidence base to inform the prioritization of vegetation management strategies. These strategies are needed to optimize yield under low soil moisture conditions while reducing the financial and ecological burden of abstraction (Berbel et al., 2019), and maintaining a balanced partitioning of green and blue water fluxes. To discern how long-term change will affect the coupling of green water fluxes with landscape storage dynamics, we need to investigate the current response of partitioning and recovery of fluxes and water ages following extreme events.
Examining the partitioning of ecohydrologic fluxes (evaporation and transpiration) and their accompanying water ages requires the use of ecohydrological models to isolate the effects of water availability and plant physiological processes (Newman et al., 2006). Without a sophisticated ecohydrological model or very large in situ datasets, accurate assessment of the partitioning of ecohydrological fluxes temporally and spatially with depth is difficult. While many hydrologic models can estimate evaporation and transpiration, the majority of these models do not consider the interaction of water stress and vegetation phenology (e.g., dynamic allocation to leaf growth) on transpiration. During drought conditions, the non-linearity of plant water usage becomes more important in assessing the role of different vegetation in exacerbating hydrological water stresses (Porporato, Laio, Ridolif, & Rodriguez-Iturbe, 2001).
Besides ecohydrologic models, stable isotopes of water, deuterium (δ 2 H), and oxygen-18 (δ 18 O), and estimated water ages are useful tools to identify the partitioning of ecohydrologic fluxes at both the plot and catchment scale (Rothfuss et al., 2010;Tetzlaff et al., 2015). Incorporating isotope data into ecohydrological models can help model evaluation. For example, isotopic fractionation during evaporation can enrich soil waters and indicate the degree of soil evaporation relative to modelled soil moisture estimations, where over-enrichment indicates an overestimation of soil evaporation; and vice versa (Sprenger, Tetzlaff, & Soulsby, 2017;Sutanto, Wenninger, Coenders-Gerrits, & Uhlenbrook, 2012). Stable isotopes thus hold considerable potential for evaluating whether the modelled mixing mechanisms, flushing rates, and residence times of water in various storage components are consistent with processes operating in reality (Hrachowitz, Savenije, Bogaard, Tetzlaff, & Soulsby, 2013;Rinaldo et al., 2011). The model, EcH 2 O-iso, incorporates the essential ecohydrologic fluxes governing water partitioning; tracks the isotopic composition of water in various storages and associated fluxes, conceptualizes various mixing processes; estimates the associated water ages; and has been applied and verified in a range of environments (Douinot et al., 2019;Kuppel, Tetzlaff, Maneta, & Soulsby, 2018a;Maneta & Silverman, 2013;Smith, Tetzlaff, Laudon, Maneta, & Soulsby, 2019). The evaluation of water ages is essential for both understanding and validating critical zone and atmospheric water sources and water recycling (Evaristo et al., 2019), storage and contamination vulnerability, sustainability of biological growth, and duration of soil chemical reactions (Sprenger et al., 2019). Despite the importance of vegetation for atmospheric recycling, there are limited studies quantifying ages of water used by vegetation (Sprenger et al., 2019), which are likely profoundly influenced by extreme conditions. As such, quantitative evaluation of the water age dynamics of soil water and vegetation water during extreme events is useful in interpreting the recovery (time until the system returns to more normal functional states), resilience (the ability of the system to maintain function during extreme events), and lethal water stress (no vegetation recovery) of an ecosystem. Water age has shown to be correlated with the water deficit caused by droughts and the recovery of "blue" water fluxes and storages in catchments (Stoelzle, Stahl, Morhard, & Weiler, 2014), and is useful in drought analysis with pre-drought water dominating the "blue" water fluxes and smaller rain events during drought used in "green" water fluxes (McGuire, DeWalle, & Gburek, 2002). The correlation of water age to catchment recovery during droughts may be due to storage-related lag times of "blue" water fluxes ("memory" effect) between inputs and outputs (e.g., 4-year recovery suggested Thomas, Lischeid, Steidl, & Dietrich, 2015) and the associated influence of vegetation depending on these storages. However, the "memory" effect of extreme events is often site-specific and can highly depend on prevailing hydroclimate, soil properties (e.g., porosity and permeability) and vegetation species (e.g., rooting depth, water use efficiency, degree of stomatal control, etc.) (e.g., Walter et al., 2011).
The primary aim of this study was to assess the influence of drought and recovery on ecohydrologic fluxes and water ages of a mixed land use site during the 2018 drought in northeast Germany. In many parts of Europe, the drought of 2018 caused dramatically reduced crop yields and in some places forest dieback (Toreti et al., 2019).
The drought provides an opportunity to couple uninterrupted field investigations including soil water stable isotopes (δ 2 H and δ 18 O) with an isotope-aided ecohydrologic model (EcH 2 O-iso) to assess: (a) the dynamics of ecohydrological fluxes of grasslands and forests during and following the drought; and (b) evaluate how water ages change in soil evaporation, transpiration, groundwater recharge and water storage under varying periods of drought and recovery. It was envisaged that quantifying the ecohydrologic partitioning and assessing resilience during and following the drought would enhance understanding of hydrological processes under drought and contribute to an evidence base for informing long-term land-use planning in drought-sensitive regions such as the North European plain where the study is based.

| Study site
The study site is in an agriculturally-dominated catchment, with longterm monitoring for hydrology and stream water chemistry, known as the Demnitzer Millcreek catchment (Smith, Tetzlaff, Gelbrecht, Kleine, & Soulsby, 2020). Situated in northeast Germany, the catchment is generally underlain by freely draining brown-earths and sandy soils over glacial drift containing deeper groundwaters. The catchment experiences a mid-continental climate with relatively low precipitation (557 mm average annual precipitation between 1981 and 2010, DWD, 2019) with higher precipitation during summers due to convective storms (317 mm average rainfall and 74% relative humidity in May-October between 1981 and 2010) and lower, less intense precipitation during winters (240 mm average rainfall and 81% relative humidity in November-April between 1981 and 2010).
The site has two high-resolution soil and vegetation monitoring plots situated in the mid-section of the catchment (Figure 1a). The two high-resolution monitoring plots are within 400 m (Figure 1b), encompassing a grassland site and forest site, with different soil properties but a negligible elevation difference (both sites at 54 m a.s.l.).
F I G U R E 1 (a) Location of the Demnitzer Millcreek catchment (DMC) in Germany, and location of the grassland and forest study sites in the DMC, (b) location of the grassland and forest sites, (c) numbered trees (number in the circle) with sap flow sensor orientation and number (numbered ticks) in the forest site The grassland site is characterized as a sandy loam, while the forest site is a loamy sand (podzolic brown-earth). While both sites show soil texture dominated by larger grain-sizes (>0.063 mm) the clay content in the surface horizons of the forest site ($3%) is around half that at the grassland site (7%) ( Table 1). Soil moisture conditions between the sites reflect the soil textural differences, with more freely draining coarser soils in the forest resulting in generally lower soil moisture.
The drawdown of soil moisture during the summer months is consistent with the higher summer evapotranspiration in the catchment .
The vegetation in the grassland is homogenous (10 m × 10 m fenced area) with relatively new growth of grass (Echinochloa polystachya) over a previously forested and ploughed soil. The grassland site is substantially sheltered by the adjacent forest to the south and east ( Figure 1b) and receives less rainfall than the forest, and lower radiation because of the adjacent forest shading the southerly sun angle, and is protected from the prevailing south and easterly winds. The forest site has a variety of tree species including European oak (Quercus robur), Red oak (Quercus rubra), common hazel (Corylus avellana), and Scots Pine (Pinus sylvestris) with relatively even spacing throughout the forest plot ( Figure 1c). The European oak is the most common tree in the plot, while the Scots Pine is the least common.
2.2 | The EcH 2 O-iso model and model set-up 2.2.1 | EcH 2 O-iso water balance EcH 2 O-iso is an ecohydrologic model based on core modules that integrate water balance, energy balance, carbon fixation and vegetation dynamics (Maneta & Silverman, 2013). The full description of the EcH 2 O model structure and parameterisation is provided in Maneta and Silverman (2013) with a description of the isotope-related structure and parameterization provided in Kuppel et al. (2018a). Here we provide a synopsis of the key modules for the plot sites. In this study, we applied the model in 1-D at the two sites, the forest and grassland monitoring plots, using local data for each site to force, calibrate, and evaluate the model ( Table 2). The water balance component of EcH 2 O-iso is divided into three distinct storages: vegetation storage (canopy storage), ponded water, and soil storage. However, vegetation and ponded water storages are relatively minor, whereas the soil storage controls the primary hydrological functions. Soil storage is divided into three subsurface layers, with the first layer constraining infiltration and evaporation, and the third layer regulating recharge to the groundwater system. Infiltration to layer 1 is based on the Green-Ampt model (Mein & Larson, 1973), and is a function of the Brooks-Corey and air-entry pressure parameters, and the depth of soil layer 1 (top soil). Soil evaporation is constrained to the top soil layer, which assumes that the deepest extent of soil evaporation is the depth of soil layer 1. Vertical water flow between soil layers is based on a gravitational drainage model that permits downward flow when soil moisture exceeds field capacity. Downward flows occur at a rate controlled by the soil moisture excess. Gravitational drainage from layer 3 can leak downward through the bottom boundary of the layer or leave the plot via lateral groundwater outflow.

| EcH 2 O-iso energy balance and vegetation dynamics
The energy balance in EcH 2 O-iso is divided into canopy and surface energy balance components, which are solved sequentially. In the canopy, available radiative energy (net radiation) is partitioned into latent heat for transpiration and interception evaporation, and canopy sensible heat, by iteratively solving for the canopy temperature. Canopy interception latent heat is governed by the maximum canopy storage (CWS max , Table 3), which controls the maximum water stored on the vegetation per leaf area index. The latent heat associated with Soil grain size properties at the grassland and forest sites and the average soil moisture content during the winter (November-April) and summer (May-October) The total transpiration is determined by the sum of the products of root-uptake and root proportion of all soil layers. The rooting zone function (proportion of roots in each soil layer) is an exponential function, and allocates the root distribution by the depth of each of the three soils layers. This formulation allows for a larger proportion of total root-uptake to occur from a soil layer with high water availability but lower rooting density than the other soil layers, albeit with a lower total transpiration rate. Canopy temperature and net radiation are used to estimate the surface energy balance, comprised of soil heat T A B L E 2 Forcing, calibration, and validation datasets with the spatial and temporal resolution of collection, with the location of data acquisition storage, latent heat of soil evaporation, and net radiation from the surface. Soil evaporation is then estimated as the maximum of the estimated latent heat and the total soil available water in layer 1 (the difference in porosity and residual soil moisture). EcH 2 O-iso also comprises a vegetation dynamics module, which allows for the growth and decay of vegetation species. Growth and decay are simulated through the dynamics of leaves, stems, and roots, each of which influences the energy and water balance in the canopy and the soil (Maneta & Silverman, 2013). The growth and decay is regulated by the total estimated gross primary product, which is sensitive to the efficiency of the canopy to utilize water (Q we , Table 3). Leaf growth dynamics are estimated as the difference between: the product of the allocation of net primary product to leaf growth and specific leaf area (SLA , Table 3), and the leaf turnover rate. Leaf turnover rate is governed by the mean turnover rate (LTr), temperature stress (LT r,temp , Table 3) and water stress (LT r,sat , Table 3). The dynamics of the leaf area index (LAI) largely regulates the net radiation reaching the ground and also regulates total stomatal surface that can transpire water, and the total water storage capacity of the canopy, among other effects.
LAI seasonality therefore exerts a strong control on the dynamics of intercepted energy and water, which may especially drive the severity and recovery of droughts (Royer, Breshears, Zou, Cobb, & Kurc, 2010).

| EcH 2 O-iso: Fractionation, water ages, and fractional water ages
Within EcH 2 O-iso, the isotopic and water age module (Kuppel et al., 2018a) mixes and tracks incoming water (age and tracer composition) according to storage and accounting for each water layer. Complete mixing is conducted for each volume (layers 1, 2 and 3, and ponded water) on every time-step using a simplification of the complete mixing assumption (for both isotopes and water ages): where δ new represents the new tracer or water age in storage, δ old is the previous tracer or water age, Q in is the incoming water flux to the layer, δ in is the incoming tracer or water age, and V new is a flux balance of incoming and outgoing water to each water layer

Vegetation parameters
Forest Grassland Stomatal conductance sensitivity factor due to vapour pressure deficit (−) LT r,temp Maximum leaf turnover rate due to cold stress (s −1 ) 1 × 10 −9 -1 × 10 −6 1 × 10 −9 -1 × 10 −5 LT r,sat Maximum leaf turnover rate due to water stress (s −1 ) 1 × 10 −9 -1 × 10 −6 1 × 10 −9 -1 × 10 −5 by any isotopic fractionation using the Craig-Gordon model, adjusting the soil relative humidity based on Lee and Pielke (1992), isotopic turbulent factor based on Mathieu and Bariac (1996), and kinetic diffu-  Table 2) was used to initialize the fractional water ages for calibrated model parameters. Fractional water ages were assessed using the summation of soil water concentrations with moving averages of 7-day and for each 30-day increments up to 1 year to correspond to the isotopic measurement intervals (i.e., 7, 30, 60, 90, etc.). For each time-step, t, a backward window of the desired increment (e.g., 7-day) was used to identify the precipitation events within the window. At time t, the concentrations of each precipitation event within the backward window were summed to estimate the fraction of water in storage or flux (e.g., fraction of water younger than 7 days).

| Model forcing (meteorological) datasets
The data available in the catchment for plot scale modelling is centred on combining existing weather stations operated by Deutscher Wetterdienst (DWD, 2019), a site implemented weather station at Hasenfelde (Figure 1a), and atmospheric reanalysis data (ERA-interim, ERA, 2019). As the Hasenfelde station was a new addition to the catchment, the total precipitation was compared to precipitation collected at Müncheberg, and only minor differences were recorded.
Temperature, relative humidity, and wind speed (measured at 2 m above the surface) were also collected at Hasenfelde on 15-min sampling intervals and compared to the Müncheberg station with good correlation. This allowed the Müncheberg data to be used to complete the time-series at Hasenfelde between January-March 2018 (as the Hasenfelde weather station was installed in March 2018). As short and longwave downward radiation was unavailable at either the DWD station or at Hasenfelde, ERA-interim data were used as forcing data for the model. ERA-interim has been shown to be close to the measured values in the study area (Douinot et al., 2019). The three-hourly data of ERA-interim were interpolated to hourly using the p-chip interpolation method (Fritsch & Carlson, 1980) (Kleine, Tetzlaff, Smith, Wang, & Soulsby, 2020).

| Calibration and validation datasets
Soil samples were analysed for stable isotopes δ 2 H and δ 18 O by equilibrating dry air added to the sealed soil bag over 48 hours and analysing the vapour isotopic composition with Picarro Gas Analyser (L2130-i) (Wassenaar, Hendry, Chostner, & Lis, 2008). Corrections of the isotopic compositions of soil water for organic gas matrix changes were made after isotopic analysis using the method described in Gralher, Herbstritt, Weiler, Wassenaar, and Stumpp (2018).

| Standardized precipitation index (SPI) and lag analysis
To evaluate the stress of limited precipitation on the soils and vegetation, the standardized precipitation index (SPI) was assessed using the daily precipitation from 1951-2018, the total duration of the precipita- Temporal lags in the soil moisture time-series data were estimated using the lag in cross-correlation (Stoica & Moses, 2005). The lag in cross-correlation between soil moisture values in different soil layers is output with the greatest significance (with conditions that p < .05).
2.5 | Model calibration, validation, and evaluation 2.5.1 | Performance metrics The model was evaluated using a variety of performance metrics, used to best evaluate the temporal variability of particular data sets. The primary performance metrics used to assess the calibration and validation of the model were the mean absolute error (MAE), Kling-Gupta efficiency (KGE) (Kling, Fuchs, & Paulin, 2012), and Nash-Sutcliffe efficiency (NSE) (Nash & Sutcliffe, 1970 These methods (or their subsequent components, e.g., r) have been used to compare model simulations of soil moisture to measured soil moisture and model simulations of LAI, LE and ET to MODIS images (Fensholt, Sandholt, & Rasmussen, 2004;Hwang et al., 2008; Trombetta, Iacobellis, Tarantino, & Gentile, 2016).

| Calibration
The model was set up independently for the grassland and forest sites using the fenced (controlled) area of each site to average the mea-  Table 3, posterior parameter ranges shown in Appendix A). Multicriteria calibration was conducted for each stepwise calibration. In both sites, soil moisture in layers 1 and 2 was calibrated using KGE and NSE (based on the variability criteria), while layer 3 was calibrated using MAE because of a lack of variability.
The LAI, LE, transpiration (transpiration in the forest site only) and ET were all calibrated using KGE and NSE. Both calibration steps included 100,000 Latin Hypercube Sampling calibration parameters sets (200,000 total calibration parameter sets). Parameter sets were calibrated by normalizing each performance metric between 1 (best) and 0 (worst). For each simulation, the minimum of all performance metrics was used to rank the simulation performance and retain the 100 "best" simulations.

| Validation
Model validation was conducted between January 1, 2019 and May 31, 2019 following the beginning of the wet-up period (December 2018), including the soil moisture recovery period and the subsequent drawdown in the following spring (April-May 2019). Similar to the calibration period, the validation period was assessed using soil moisture, LAI, LE, and ET; however, transpiration in the forest site was not available for the validation period in the forest. Soil water isotopes (δ 2 H, and lc-excess) were used to informally validate soil water mixing and soil evaporation processes through the whole simulation period (calibration and validation period inclusive). The MAE of the δ 2 H and lc-excess were calculated for each soil layer (no maximum criteria was set for rejection). As there were relatively few soil isotope samples available during the calibration period, the whole data set were separated into each soil layer (layer 1:0-15 cm, layer 2:15-50 cm, and layer 3: >50 cm) and evaluated as validation of the models' longerscale internal simulation dynamics.

| Evaluation of soil and vegetation sensitivity to fluxes and water ages
To test the principal factors in the quantity and responsiveness of storages, fluxes, and water ages, different combinations of soil (sandy loam (grassland) and loamy sand (forest)) and vegetation (grass and forest) were evaluated against the calibrated simulations (referred to as the baseline conditions). From calibration there were 100 "best" parameter sets pertaining to vegetation and soil. New uncalibrated combinations of soil and vegetation parameter sets (e.g., sandy loam and forest) were randomly designated from the calibrated vegetation and soil parameter sets. To remove bias of a single combination of soil and vegetation parameter sets, 100 random combinations of vegetation and soil parameters were selected. One hundred random combinations were chosen for direct comparison to the 100 "best" baseline conditions.
The effects of changing the soil under the grass (sandy loam to loamy sand) and vegetation on sandy loam (grass to forest) was evaluated using the 100 "best" calibrated simulations at the grassland site as baseline conditions. Similarly, the effects of changing the soil under forests (loamy sand to sandy loam) and vegetation on loamy sand (forest to grass) were assessed using the forest site as baseline conditions. The change was assessed as the absolute percent change on each time-step (%Change = where C(t) is the calibrated simulation baseline, S(t) is the simulation with changed soil (or vegetation) parameters, and ts is the number of time-steps. With parameters for each soil and vegetation type, the significance of the change from the baseline in storage, flux or water age was assessed using the Wilcox-rank sum test, which does not assume a normal distribution (Mann & Whitney, 1947  The general sensitivity of storages and fluxes is temporally insensitive to the drought or recovery periods (Appendix C). The ET and transpiration were consistent throughout the simulation, showing highest sensitivity to vegetation type, while soil moisture in layers 2 and 3 were consistently most sensitive to the soil type (Table C1).
Soil evaporation, groundwater recharge, and soil layer 1 exhibited some temporal differences in the sensitivity, though changes in the sensitivity were inconsistent between the sites with the sensitivity against the forest baseline conditions showing the greatest temporal differences (Table C1).

| Influence of drought and rewetting on soil water isotopes and ages
Simulated soil water isotopes reflect the dynamics of the measured bulk soil water isotopes in each soil layer, despite not being used in calibration (Figure 5b-d, f-h). In the forest, δ 2 H simulations captured the higher dynamics of the measured δ 2 H in layer 1 (Figure 5b,  (Figure 5h, measured and simulated temporal standard deviation, 3.8 and 0.8‰, respectively); however, the relatively low simulated variability and higher measurement variability suggests that some of the measured variability may be due to spatial differences.
Simulated soil water ages for each soil layer were distinctly different for both the forest and grassland sites. Soil water in layer 1 in the forest was the youngest and most dynamic water on average ( Figure 5a) but exhibited slightly greater changes in the mean age during the drought compared to the long-term average or recovery period (Table 4). The decrease in water age in layer 1 in the forest during the drought was likely due to the low soil moisture accentuating the influence of new precipitation inputs (age of 0 days). The water ages in the grassland were consistently older than the forest, as well F I G U R E 4 Calibration (light grey) and validation (dark grey) of the latent heat, green water fluxes (evapotranspiration, and soil evaporation) and blue water fluxes (groundwater recharge) in the forest site (a-d) and grassland site (e-h), respectively as less variable throughout the year (Figure 5e). Water ages in layer 1 were almost twice as old as the forest (Table 4), with more comparable water ages in layer 2. However, layer 2 in the grassland showed limited temporal variability (Figure 5e).
Throughout the year, soil water at the forest site in layer 1 was dominated by water younger than 30 days (Figure 6a, Table 5), with a notable and seemingly paradoxical increase in the younger fraction of 30 and 60 day water during the drought (Table 5). However, the fractional water ages in layer 1 are very flashy in response to new water inputs, with high peaks of fractional water age younger than 7 days following precipitation events. Despite this, the contribution of water younger than 30 and 60 days during the drought does not return to pre-drought fractions until the post-recovery period (Table 5). Unlike layer 1, during the drought layers 2 and 3 in the forest showed decreasing fractions of water in the younger than 30, 60 and 180-day fractions (Figure 6b,c and Table 5). Layer 3 in the forest additionally showed decreasing fractions of 365-day water during the recovery period, likely lagged from the drought as younger waters percolated to depth (Table 5). In both layers, the fractions of 30, 60, and 180-day water returned to pre-drought conditions either in the recovery or post-recovery period. With soil evaporation occurring only from layer 1, the fractional water ages of evaporation in the forest are the same as to the fractional ages in layer 1 (Figure 6a). Rooting depths spanning all three soil layers in the forest results in a unique temporal change in the fractional water used in transpiration. In the driest periods of the drought, transpiration has a higher fraction of old water ( Figure 6d,e,g older than 180 days) than after the drought (Table 5). In the post-recovery period, (recovery period has limited transpiration), transpiration again is fed much more by younger water fractions (Table 5).
The grassland site has a substantially higher fraction of older water in layer 1 compared to the forest site (Figure 6a  water younger than 60 days increase in the grassland site layer 1 (0.59-0.76, Table 5); however, the fractions of water younger than 60 days did not decrease to pre-drought conditions within the simulation period (0.68 following post-recovery). Similar to the forest site, young water fractions in layer 2 and 3 in the grassland decreased during the drought, for 30, 60, and 180-day fractions in layer 2, and 180-day fractions in layer 3 (Figure 6f,g). These fractions rebounded during the recovery period. However, decreases in the water age fraction of F I G U R E 6 Forest and grassland site fractional water ages of water in storage younger than a certain age (7-day, and 30-day increments up to 1 year). Water older than 1 year was aggregated into a single fraction, water >365 days. Fractional water ages are shown for layer 1 (a and e), layer 2 (b and f), layer 3 (c and g), and transpiration (   365 days in layers 2 and 3 were delayed until the recovery period, with only the fraction in layer 2 rebounding by the end of the post-recovery period. With soil evaporation occurring only from layer 1, the fractional water of soil evaporation in the grassland site was equivalent to that in layer 1. Shallower rooting depths in the grassland resulted in higher young water fractions in transpiration throughout the year and were similar to the age fraction in layer 1. However, during the drought, the transpiration fraction of water younger than 30 days decreased rather than increased, more closely reflecting layer 2 than layer 1 (Figure 6h).
The sensitivity of the model water age simulations to changes in soil and vegetation show temporal variability from during the drought to the recovery following the drought. At both sites, the change in vegetation was the most significant (p < .01, symbol b) for estimating the transpiration water age, with larger influences in the transpiration water age during the recovery period. During the drought, layer 1 water age was most influenced by the vegetation type (Figure 7b,e; values of green boxplots) rather than soil type, with the soil type influencing the layer 1 water age most significantly during the recovery period (Figure 7c,f). The higher change in layer 1 water age due to vegetation was not consistent with the change in soil moisture in layer 1 during the drought (more significant change due to soil type, Appendix C). During the recovery period, water ages in layer 2 were most affected by the soil type (Figure 7c,f). However, during the drought, water age in layer 2 was influenced most by soil type in the grassland site (e.g., changing sandy loam to loamy sand under a grassland), while at the forest site the water age in layer 2 was most significantly influenced by vegetation (e.g., changing forest to grassland over a sandy soil). In all periods, the soil layer 3 water age was most affected by the soil type (Figure 7). The high hydraulic conductivity and lower soil tension in the loamy sand (Dingman, 2002) aids rapid and free-drainage in the forest, compounding with a slower mean increase in near-surface soils wetness (layer 1) relative to the grassland after the drought, but a more rapid response of soil moisture in deeper layers (Figure 2e). There are limitations of EcH 2 O-iso that fail to capture all of the moisture dynamics and deficits of deeper soil layers, which is most likely due to the lessrobust, but computationally-efficient Green-Ampt soil moisture routing (Rao, Raghuwanshi, & Singh, 2006), with gravitational drainage to lower soil layers (Kuppel et al., 2018b), which may under-estimate water movement through more sandy soils (Ogden & Saghafian, 1997). However, the relatively small estimated error in simulations of uncalibrated soil isotopes provides credible evidence that differences between modelled and measured soil responses at depth are not seriously in error. Furthermore, the crucial differences in the variability of soil moisture between the two sites with depth were captured.
The evaluation of the "blue" and "green" fluxes during the drought suggests further implications of the drought that are not observed solely with the soil moisture dynamics. Both simulations and measurements of transpiration in the forest suggest higher water use for the forest following rainfall events (transpiration simulation, Appendix B), with highly variable (moisture-dependent) transpiration, and low soil evaporation (Barbeta & Peñuelas, 2016;Limousin et al., 2009). The high simulated lc-excess in the forest (Figure 5b) suggests some under-estimation of soil evaporation during the summer, likely following rainfall events when soils were wetter. Some of the under-estimation in soil evaporation may be due to the simulated rapid response in transpiration to precipitation during the drought, which was higher than the sap flow variability. The potential over-estimation of transpiration response following the summer rainfall may be due to a "stress imprint" on vegetation impacting the response of vegetation to water availability (Walter, Jentsch, Beierkuhnlein, & Kreyling, 2013). Under a more extreme event, such as a hypothetical situation where the July 2018 precipitation event had not occurred, more severe water stress conditions for vegetation may result in vegetation mortality (e.g., Allen et al., 2010). While there was some under-estimation of total ET in the forest compared to the MODIS ET data, the MODIS ET data may have bias during dry periods due to the estimation method using vapour pressure deficit as a proxy for water availability for transpiration. The partitioning of the "green" fluxes into transpiration and soil evaporation was more complicated in the grassland due to limitations in measurements of grass transpiration and the use of large-scale MODIS data for calibration. However, the similarities between simulated and observed lc-excess in soil water suggest the model accurately simulated the high estimated soil evaporation fraction of ET. In addition, the high fraction of soil evaporation in ET in the shel-

| Resilience and recovery of soil water ages and ecohydrologic fluxes
The water age dynamics of the forest and grassland sites were comparable to previous studies with younger water in more freely draining soil and older water in more retentive soils (e.g., Tetzlaff, Birkel, Dick, Geris, and Soulsby [2014] and Stumpp, Maloszewski, Stichler, and Fank [2009]); while simultaneously revealing the influence of the drought and recovery on the vulnerability of the vegetation and soils to water limiting conditions. Surprisingly, the shallow soil water age (layer 1, 0-15 cm) did not greatly change progressing through predrought, drought, and recovery at either site ( Figure 5), which highlights the differences in the reaction of water age and soil moisture to drought. The fractional water ages suggest that the lack of change in mean water age in shallow soils is complex. The mean age estimation is driven by a low soil moisture content which causes a bimodal distribution of young precipitation (increasing the 30-day water fraction, Table 5) and older (with a decreasing volume) stored soil water. With very young water ages in layer 1, it is clear that replenishment by precipitation is a dominant driver in setting water ages. While the moisture content in the shallow soils during the drought is likely driven by the soil properties at each site (Appendix C), the water age is controlled more by the vegetation than the soil type (Figure 7). This suggests that the type of vegetation has implications for the ability of each site to recover to initial conditions, as older water ages imply longer retention times and slower replenishment of the deeper storages sustaining "blue" water fluxes. Layer 2 at both sites experienced limited percolation and increasing water ages during the drought, with relatively limited change in soil moisture, thus highlighting the vulnerability of sub-soil storage in droughts. In the sandier forest soils, the root-uptake during the drought significantly affected the water ages in layer 2 (Figure 7), and similar to layer 1, has implications for the recovery time of storages sustaining "blue" water fluxes. Similarly, the marked change in root-uptake water ages during the drought suggests that the drought resilience of the plot was approaching a limit. In the loamier grassland site, the much older water present (Table 5) despite the relatively limited change in the mean water age (Table 4) exemplifies the limited percolation of younger water during the drought.
Both sites showed a notable lag of 365-day water in layer 2, with higher old water proportions during the recovery period compared to the drought. This lag implies that while the water content appears to have recovered, the system is still under stress from the drought.
While the complete mixing assumption used in EcH 2 O-iso (Kuppel et al., 2018a) may influence the water ages estimated, the inclusion of incomplete mixing would likely result in greater young water fractions in deeper soils during the wetter recovery period (Kim et al., 2016) and would length the whole system recovery.
Since soil evaporation is sustained by the moisture in the upper soil layer, the effect of evaporation on water ages during the drought is relatively minor. Transpired water ages reveal more distinct differences between the sites. With a notable decrease in transpiration, the water ages of transpiration in the forest during the drought revealed changes in primary water sources from layer 1 to layers 2 and 3 due to physiological changes in the vegetation resulting from extreme water stress in layer 1. With negligible transpiration during the recovery period, the start of the 2019 growing season showed transpiration in the forest rebounding to show a dominance of near-surface (layer 1) water. Despite the relatively consistent transpiration rate in the grassland site, the water ages did not deviate greatly from the water ages in layer 1 suggesting the grass did not experience water stress as severe as the forest. Only a small proportion of transpired water in the grassland originated from layer 2, as evident from the older (>365-day) water (Figure 6h). While the effect of rooting depths on different species transpiration ages has been observed in other regions (Douinot et al., 2019;Kuppel et al., 2018a;Sprenger et al., 2018) the water ages uniquely reveal the vulnerability of the forest vegetation to temporal variation in available waters during drought conditions. These estimates of transpiration age are likely to be conservative as recent work has shown that internal storage and cycling of water by trees may involve additional residence times of several months (Knighton et al., 2020;Köcher, Horna, Beckmeyer, & Leuschner, 2012;Urban, Čermák, & Ceulemans, 2014).
Seeking to provide land and water managers with a quantitative evidence base for building resilience in the face of increased probability of more frequent and intense droughts is the fundamental motivation for integrating empirical measurements and ecohydrologic modelling techniques. A different combination of vegetation and soil could likely reveal wider ranges of resilience in this region, and there is an urgent need for monitoring in contrasting soil-vegetation units and over longer periods (Kundzewicz et al., 2009). Further exploration of either modelling or data-driven analysis could also be used to aid upscaling and developing such understanding to larger management areas (Vereecken, Pachepsky, Bogena, & Montzka, 2019).

| CONCLUSION
With the projected increases in the frequency of extreme climate conditions, it is crucial to discern the long-term sustainability of land and water management in drought-sensitive landscapes. This is highly dependent on the partitioning of precipitation into plant available water and water recharging aquifers and sustaining river flows (Orth & Destouni, 2018). Quantifying this partitioning of water, and projecting how it is likely to change, and assessing the resilience of different vegetation and soil systems are during extreme events are all urgent scientific needs. At the grassland and forest plot sites in northeast Germany examined in this study, green water fluxes of evaporation and transpiration were dominant in the summer drought of 2018, with only limited blue water fluxes to groundwater recharge following rainfall events.
The loamier soils in the grassland aided in maintaining higher soil moisture and slower response to drought than the sandier soils in the forest.
Despite the recovery of soil moisture in the grassland following the drought, fractional water ages in the deeper soil layers suggest that the grassland had not fully recovered from the drought more than half a year after the end of the drought. The sandier forested site had more variable soil moisture, high plant water usage, and stronger young water influence. In the sandier soils, vegetation had a much larger influence on the water age in storage during the drought. Similar to the grassland, fractional water ages indicate a longer recovery period than the soil moisture, though recovery of water ages is less than the grassland (less than half a year). Recharge to the groundwater system under the forest only occurred during the wetter winter period when soil moisture was replenished which coincided with the period of smallest young water fractions and highlights the potential vulnerability of long-term recharge in the forest under future climate scenarios. Utilizing sophisticated tracer-aided ecohydrological models and consideration of water ages can contribute to the evidence base on the hydrologic recovery of "blue" and "green" water fluxes and storage which can inform sustainable land and water management strategies in the future. In this regard, quantifying water use and age dynamics under drought conditions is particularly insightful. Moreover, the research approach adopted in this study provides the potential for further exploration and upscaling of ecohydrological partitioning and resilience to larger catchment scales.
would also like to acknowledge David Dubbert for the isotope analysis. Lastly, the authors thank the University of Aberdeen for the use of the High-Performance Cluster (HPC), which was used for all model runs. The authors thank the issue editor (Paulo Benettin), Stefanie Lutz and two anonymous reviewers for their constructive comments through the review process which has aided in improving the manuscript.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request. Chris Soulsby https://orcid.org/0000-0001-6910-2118